e1000: test link state conclusively
[pandora-kernel.git] / drivers / net / e1000 / e1000_main.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2006 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include "e1000.h"
30 #include <net/ip6_checksum.h>
31
32 char e1000_driver_name[] = "e1000";
33 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
34 #define DRV_VERSION "7.3.21-k5-NAPI"
35 const char e1000_driver_version[] = DRV_VERSION;
36 static const char e1000_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
37
38 /* e1000_pci_tbl - PCI Device ID Table
39  *
40  * Last entry must be all 0s
41  *
42  * Macro expands to...
43  *   {PCI_DEVICE(PCI_VENDOR_ID_INTEL, device_id)}
44  */
45 static struct pci_device_id e1000_pci_tbl[] = {
46         INTEL_E1000_ETHERNET_DEVICE(0x1000),
47         INTEL_E1000_ETHERNET_DEVICE(0x1001),
48         INTEL_E1000_ETHERNET_DEVICE(0x1004),
49         INTEL_E1000_ETHERNET_DEVICE(0x1008),
50         INTEL_E1000_ETHERNET_DEVICE(0x1009),
51         INTEL_E1000_ETHERNET_DEVICE(0x100C),
52         INTEL_E1000_ETHERNET_DEVICE(0x100D),
53         INTEL_E1000_ETHERNET_DEVICE(0x100E),
54         INTEL_E1000_ETHERNET_DEVICE(0x100F),
55         INTEL_E1000_ETHERNET_DEVICE(0x1010),
56         INTEL_E1000_ETHERNET_DEVICE(0x1011),
57         INTEL_E1000_ETHERNET_DEVICE(0x1012),
58         INTEL_E1000_ETHERNET_DEVICE(0x1013),
59         INTEL_E1000_ETHERNET_DEVICE(0x1014),
60         INTEL_E1000_ETHERNET_DEVICE(0x1015),
61         INTEL_E1000_ETHERNET_DEVICE(0x1016),
62         INTEL_E1000_ETHERNET_DEVICE(0x1017),
63         INTEL_E1000_ETHERNET_DEVICE(0x1018),
64         INTEL_E1000_ETHERNET_DEVICE(0x1019),
65         INTEL_E1000_ETHERNET_DEVICE(0x101A),
66         INTEL_E1000_ETHERNET_DEVICE(0x101D),
67         INTEL_E1000_ETHERNET_DEVICE(0x101E),
68         INTEL_E1000_ETHERNET_DEVICE(0x1026),
69         INTEL_E1000_ETHERNET_DEVICE(0x1027),
70         INTEL_E1000_ETHERNET_DEVICE(0x1028),
71         INTEL_E1000_ETHERNET_DEVICE(0x1075),
72         INTEL_E1000_ETHERNET_DEVICE(0x1076),
73         INTEL_E1000_ETHERNET_DEVICE(0x1077),
74         INTEL_E1000_ETHERNET_DEVICE(0x1078),
75         INTEL_E1000_ETHERNET_DEVICE(0x1079),
76         INTEL_E1000_ETHERNET_DEVICE(0x107A),
77         INTEL_E1000_ETHERNET_DEVICE(0x107B),
78         INTEL_E1000_ETHERNET_DEVICE(0x107C),
79         INTEL_E1000_ETHERNET_DEVICE(0x108A),
80         INTEL_E1000_ETHERNET_DEVICE(0x1099),
81         INTEL_E1000_ETHERNET_DEVICE(0x10B5),
82         /* required last entry */
83         {0,}
84 };
85
86 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
87
88 int e1000_up(struct e1000_adapter *adapter);
89 void e1000_down(struct e1000_adapter *adapter);
90 void e1000_reinit_locked(struct e1000_adapter *adapter);
91 void e1000_reset(struct e1000_adapter *adapter);
92 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx);
93 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
94 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
95 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
96 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
97 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
98                              struct e1000_tx_ring *txdr);
99 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
100                              struct e1000_rx_ring *rxdr);
101 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
102                              struct e1000_tx_ring *tx_ring);
103 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
104                              struct e1000_rx_ring *rx_ring);
105 void e1000_update_stats(struct e1000_adapter *adapter);
106
107 static int e1000_init_module(void);
108 static void e1000_exit_module(void);
109 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
110 static void __devexit e1000_remove(struct pci_dev *pdev);
111 static int e1000_alloc_queues(struct e1000_adapter *adapter);
112 static int e1000_sw_init(struct e1000_adapter *adapter);
113 static int e1000_open(struct net_device *netdev);
114 static int e1000_close(struct net_device *netdev);
115 static void e1000_configure_tx(struct e1000_adapter *adapter);
116 static void e1000_configure_rx(struct e1000_adapter *adapter);
117 static void e1000_setup_rctl(struct e1000_adapter *adapter);
118 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
119 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
120 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
121                                 struct e1000_tx_ring *tx_ring);
122 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
123                                 struct e1000_rx_ring *rx_ring);
124 static void e1000_set_rx_mode(struct net_device *netdev);
125 static void e1000_update_phy_info(unsigned long data);
126 static void e1000_watchdog(unsigned long data);
127 static void e1000_82547_tx_fifo_stall(unsigned long data);
128 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
129                                     struct net_device *netdev);
130 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
131 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
132 static int e1000_set_mac(struct net_device *netdev, void *p);
133 static irqreturn_t e1000_intr(int irq, void *data);
134 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
135                                struct e1000_tx_ring *tx_ring);
136 static int e1000_clean(struct napi_struct *napi, int budget);
137 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
138                                struct e1000_rx_ring *rx_ring,
139                                int *work_done, int work_to_do);
140 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
141                                      struct e1000_rx_ring *rx_ring,
142                                      int *work_done, int work_to_do);
143 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
144                                    struct e1000_rx_ring *rx_ring,
145                                    int cleaned_count);
146 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
147                                          struct e1000_rx_ring *rx_ring,
148                                          int cleaned_count);
149 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
150 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
151                            int cmd);
152 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
153 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
154 static void e1000_tx_timeout(struct net_device *dev);
155 static void e1000_reset_task(struct work_struct *work);
156 static void e1000_smartspeed(struct e1000_adapter *adapter);
157 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
158                                        struct sk_buff *skb);
159
160 static void e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp);
161 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid);
162 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid);
163 static void e1000_restore_vlan(struct e1000_adapter *adapter);
164
165 #ifdef CONFIG_PM
166 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
167 static int e1000_resume(struct pci_dev *pdev);
168 #endif
169 static void e1000_shutdown(struct pci_dev *pdev);
170
171 #ifdef CONFIG_NET_POLL_CONTROLLER
172 /* for netdump / net console */
173 static void e1000_netpoll (struct net_device *netdev);
174 #endif
175
176 #define COPYBREAK_DEFAULT 256
177 static unsigned int copybreak __read_mostly = COPYBREAK_DEFAULT;
178 module_param(copybreak, uint, 0644);
179 MODULE_PARM_DESC(copybreak,
180         "Maximum size of packet that is copied to a new buffer on receive");
181
182 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
183                      pci_channel_state_t state);
184 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
185 static void e1000_io_resume(struct pci_dev *pdev);
186
187 static struct pci_error_handlers e1000_err_handler = {
188         .error_detected = e1000_io_error_detected,
189         .slot_reset = e1000_io_slot_reset,
190         .resume = e1000_io_resume,
191 };
192
193 static struct pci_driver e1000_driver = {
194         .name     = e1000_driver_name,
195         .id_table = e1000_pci_tbl,
196         .probe    = e1000_probe,
197         .remove   = __devexit_p(e1000_remove),
198 #ifdef CONFIG_PM
199         /* Power Managment Hooks */
200         .suspend  = e1000_suspend,
201         .resume   = e1000_resume,
202 #endif
203         .shutdown = e1000_shutdown,
204         .err_handler = &e1000_err_handler
205 };
206
207 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
208 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
209 MODULE_LICENSE("GPL");
210 MODULE_VERSION(DRV_VERSION);
211
212 static int debug = NETIF_MSG_DRV | NETIF_MSG_PROBE;
213 module_param(debug, int, 0);
214 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
215
216 /**
217  * e1000_init_module - Driver Registration Routine
218  *
219  * e1000_init_module is the first routine called when the driver is
220  * loaded. All it does is register with the PCI subsystem.
221  **/
222
223 static int __init e1000_init_module(void)
224 {
225         int ret;
226         printk(KERN_INFO "%s - version %s\n",
227                e1000_driver_string, e1000_driver_version);
228
229         printk(KERN_INFO "%s\n", e1000_copyright);
230
231         ret = pci_register_driver(&e1000_driver);
232         if (copybreak != COPYBREAK_DEFAULT) {
233                 if (copybreak == 0)
234                         printk(KERN_INFO "e1000: copybreak disabled\n");
235                 else
236                         printk(KERN_INFO "e1000: copybreak enabled for "
237                                "packets <= %u bytes\n", copybreak);
238         }
239         return ret;
240 }
241
242 module_init(e1000_init_module);
243
244 /**
245  * e1000_exit_module - Driver Exit Cleanup Routine
246  *
247  * e1000_exit_module is called just before the driver is removed
248  * from memory.
249  **/
250
251 static void __exit e1000_exit_module(void)
252 {
253         pci_unregister_driver(&e1000_driver);
254 }
255
256 module_exit(e1000_exit_module);
257
258 static int e1000_request_irq(struct e1000_adapter *adapter)
259 {
260         struct net_device *netdev = adapter->netdev;
261         irq_handler_t handler = e1000_intr;
262         int irq_flags = IRQF_SHARED;
263         int err;
264
265         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
266                           netdev);
267         if (err) {
268                 DPRINTK(PROBE, ERR,
269                         "Unable to allocate interrupt Error: %d\n", err);
270         }
271
272         return err;
273 }
274
275 static void e1000_free_irq(struct e1000_adapter *adapter)
276 {
277         struct net_device *netdev = adapter->netdev;
278
279         free_irq(adapter->pdev->irq, netdev);
280 }
281
282 /**
283  * e1000_irq_disable - Mask off interrupt generation on the NIC
284  * @adapter: board private structure
285  **/
286
287 static void e1000_irq_disable(struct e1000_adapter *adapter)
288 {
289         struct e1000_hw *hw = &adapter->hw;
290
291         ew32(IMC, ~0);
292         E1000_WRITE_FLUSH();
293         synchronize_irq(adapter->pdev->irq);
294 }
295
296 /**
297  * e1000_irq_enable - Enable default interrupt generation settings
298  * @adapter: board private structure
299  **/
300
301 static void e1000_irq_enable(struct e1000_adapter *adapter)
302 {
303         struct e1000_hw *hw = &adapter->hw;
304
305         ew32(IMS, IMS_ENABLE_MASK);
306         E1000_WRITE_FLUSH();
307 }
308
309 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
310 {
311         struct e1000_hw *hw = &adapter->hw;
312         struct net_device *netdev = adapter->netdev;
313         u16 vid = hw->mng_cookie.vlan_id;
314         u16 old_vid = adapter->mng_vlan_id;
315         if (adapter->vlgrp) {
316                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
317                         if (hw->mng_cookie.status &
318                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
319                                 e1000_vlan_rx_add_vid(netdev, vid);
320                                 adapter->mng_vlan_id = vid;
321                         } else
322                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
323
324                         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
325                                         (vid != old_vid) &&
326                             !vlan_group_get_device(adapter->vlgrp, old_vid))
327                                 e1000_vlan_rx_kill_vid(netdev, old_vid);
328                 } else
329                         adapter->mng_vlan_id = vid;
330         }
331 }
332
333 static void e1000_init_manageability(struct e1000_adapter *adapter)
334 {
335         struct e1000_hw *hw = &adapter->hw;
336
337         if (adapter->en_mng_pt) {
338                 u32 manc = er32(MANC);
339
340                 /* disable hardware interception of ARP */
341                 manc &= ~(E1000_MANC_ARP_EN);
342
343                 ew32(MANC, manc);
344         }
345 }
346
347 static void e1000_release_manageability(struct e1000_adapter *adapter)
348 {
349         struct e1000_hw *hw = &adapter->hw;
350
351         if (adapter->en_mng_pt) {
352                 u32 manc = er32(MANC);
353
354                 /* re-enable hardware interception of ARP */
355                 manc |= E1000_MANC_ARP_EN;
356
357                 ew32(MANC, manc);
358         }
359 }
360
361 /**
362  * e1000_configure - configure the hardware for RX and TX
363  * @adapter = private board structure
364  **/
365 static void e1000_configure(struct e1000_adapter *adapter)
366 {
367         struct net_device *netdev = adapter->netdev;
368         int i;
369
370         e1000_set_rx_mode(netdev);
371
372         e1000_restore_vlan(adapter);
373         e1000_init_manageability(adapter);
374
375         e1000_configure_tx(adapter);
376         e1000_setup_rctl(adapter);
377         e1000_configure_rx(adapter);
378         /* call E1000_DESC_UNUSED which always leaves
379          * at least 1 descriptor unused to make sure
380          * next_to_use != next_to_clean */
381         for (i = 0; i < adapter->num_rx_queues; i++) {
382                 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
383                 adapter->alloc_rx_buf(adapter, ring,
384                                       E1000_DESC_UNUSED(ring));
385         }
386
387         adapter->tx_queue_len = netdev->tx_queue_len;
388 }
389
390 int e1000_up(struct e1000_adapter *adapter)
391 {
392         struct e1000_hw *hw = &adapter->hw;
393
394         /* hardware has been reset, we need to reload some things */
395         e1000_configure(adapter);
396
397         clear_bit(__E1000_DOWN, &adapter->flags);
398
399         napi_enable(&adapter->napi);
400
401         e1000_irq_enable(adapter);
402
403         netif_wake_queue(adapter->netdev);
404
405         /* fire a link change interrupt to start the watchdog */
406         ew32(ICS, E1000_ICS_LSC);
407         return 0;
408 }
409
410 /**
411  * e1000_power_up_phy - restore link in case the phy was powered down
412  * @adapter: address of board private structure
413  *
414  * The phy may be powered down to save power and turn off link when the
415  * driver is unloaded and wake on lan is not enabled (among others)
416  * *** this routine MUST be followed by a call to e1000_reset ***
417  *
418  **/
419
420 void e1000_power_up_phy(struct e1000_adapter *adapter)
421 {
422         struct e1000_hw *hw = &adapter->hw;
423         u16 mii_reg = 0;
424
425         /* Just clear the power down bit to wake the phy back up */
426         if (hw->media_type == e1000_media_type_copper) {
427                 /* according to the manual, the phy will retain its
428                  * settings across a power-down/up cycle */
429                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
430                 mii_reg &= ~MII_CR_POWER_DOWN;
431                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
432         }
433 }
434
435 static void e1000_power_down_phy(struct e1000_adapter *adapter)
436 {
437         struct e1000_hw *hw = &adapter->hw;
438
439         /* Power down the PHY so no link is implied when interface is down *
440          * The PHY cannot be powered down if any of the following is true *
441          * (a) WoL is enabled
442          * (b) AMT is active
443          * (c) SoL/IDER session is active */
444         if (!adapter->wol && hw->mac_type >= e1000_82540 &&
445            hw->media_type == e1000_media_type_copper) {
446                 u16 mii_reg = 0;
447
448                 switch (hw->mac_type) {
449                 case e1000_82540:
450                 case e1000_82545:
451                 case e1000_82545_rev_3:
452                 case e1000_82546:
453                 case e1000_82546_rev_3:
454                 case e1000_82541:
455                 case e1000_82541_rev_2:
456                 case e1000_82547:
457                 case e1000_82547_rev_2:
458                         if (er32(MANC) & E1000_MANC_SMBUS_EN)
459                                 goto out;
460                         break;
461                 default:
462                         goto out;
463                 }
464                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
465                 mii_reg |= MII_CR_POWER_DOWN;
466                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
467                 mdelay(1);
468         }
469 out:
470         return;
471 }
472
473 void e1000_down(struct e1000_adapter *adapter)
474 {
475         struct e1000_hw *hw = &adapter->hw;
476         struct net_device *netdev = adapter->netdev;
477         u32 rctl, tctl;
478
479         /* signal that we're down so the interrupt handler does not
480          * reschedule our watchdog timer */
481         set_bit(__E1000_DOWN, &adapter->flags);
482
483         /* disable receives in the hardware */
484         rctl = er32(RCTL);
485         ew32(RCTL, rctl & ~E1000_RCTL_EN);
486         /* flush and sleep below */
487
488         netif_tx_disable(netdev);
489
490         /* disable transmits in the hardware */
491         tctl = er32(TCTL);
492         tctl &= ~E1000_TCTL_EN;
493         ew32(TCTL, tctl);
494         /* flush both disables and wait for them to finish */
495         E1000_WRITE_FLUSH();
496         msleep(10);
497
498         napi_disable(&adapter->napi);
499
500         e1000_irq_disable(adapter);
501
502         del_timer_sync(&adapter->tx_fifo_stall_timer);
503         del_timer_sync(&adapter->watchdog_timer);
504         del_timer_sync(&adapter->phy_info_timer);
505
506         netdev->tx_queue_len = adapter->tx_queue_len;
507         adapter->link_speed = 0;
508         adapter->link_duplex = 0;
509         netif_carrier_off(netdev);
510
511         e1000_reset(adapter);
512         e1000_clean_all_tx_rings(adapter);
513         e1000_clean_all_rx_rings(adapter);
514 }
515
516 void e1000_reinit_locked(struct e1000_adapter *adapter)
517 {
518         WARN_ON(in_interrupt());
519         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
520                 msleep(1);
521         e1000_down(adapter);
522         e1000_up(adapter);
523         clear_bit(__E1000_RESETTING, &adapter->flags);
524 }
525
526 void e1000_reset(struct e1000_adapter *adapter)
527 {
528         struct e1000_hw *hw = &adapter->hw;
529         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
530         bool legacy_pba_adjust = false;
531         u16 hwm;
532
533         /* Repartition Pba for greater than 9k mtu
534          * To take effect CTRL.RST is required.
535          */
536
537         switch (hw->mac_type) {
538         case e1000_82542_rev2_0:
539         case e1000_82542_rev2_1:
540         case e1000_82543:
541         case e1000_82544:
542         case e1000_82540:
543         case e1000_82541:
544         case e1000_82541_rev_2:
545                 legacy_pba_adjust = true;
546                 pba = E1000_PBA_48K;
547                 break;
548         case e1000_82545:
549         case e1000_82545_rev_3:
550         case e1000_82546:
551         case e1000_82546_rev_3:
552                 pba = E1000_PBA_48K;
553                 break;
554         case e1000_82547:
555         case e1000_82547_rev_2:
556                 legacy_pba_adjust = true;
557                 pba = E1000_PBA_30K;
558                 break;
559         case e1000_undefined:
560         case e1000_num_macs:
561                 break;
562         }
563
564         if (legacy_pba_adjust) {
565                 if (hw->max_frame_size > E1000_RXBUFFER_8192)
566                         pba -= 8; /* allocate more FIFO for Tx */
567
568                 if (hw->mac_type == e1000_82547) {
569                         adapter->tx_fifo_head = 0;
570                         adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
571                         adapter->tx_fifo_size =
572                                 (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
573                         atomic_set(&adapter->tx_fifo_stall, 0);
574                 }
575         } else if (hw->max_frame_size >  ETH_FRAME_LEN + ETH_FCS_LEN) {
576                 /* adjust PBA for jumbo frames */
577                 ew32(PBA, pba);
578
579                 /* To maintain wire speed transmits, the Tx FIFO should be
580                  * large enough to accommodate two full transmit packets,
581                  * rounded up to the next 1KB and expressed in KB.  Likewise,
582                  * the Rx FIFO should be large enough to accommodate at least
583                  * one full receive packet and is similarly rounded up and
584                  * expressed in KB. */
585                 pba = er32(PBA);
586                 /* upper 16 bits has Tx packet buffer allocation size in KB */
587                 tx_space = pba >> 16;
588                 /* lower 16 bits has Rx packet buffer allocation size in KB */
589                 pba &= 0xffff;
590                 /*
591                  * the tx fifo also stores 16 bytes of information about the tx
592                  * but don't include ethernet FCS because hardware appends it
593                  */
594                 min_tx_space = (hw->max_frame_size +
595                                 sizeof(struct e1000_tx_desc) -
596                                 ETH_FCS_LEN) * 2;
597                 min_tx_space = ALIGN(min_tx_space, 1024);
598                 min_tx_space >>= 10;
599                 /* software strips receive CRC, so leave room for it */
600                 min_rx_space = hw->max_frame_size;
601                 min_rx_space = ALIGN(min_rx_space, 1024);
602                 min_rx_space >>= 10;
603
604                 /* If current Tx allocation is less than the min Tx FIFO size,
605                  * and the min Tx FIFO size is less than the current Rx FIFO
606                  * allocation, take space away from current Rx allocation */
607                 if (tx_space < min_tx_space &&
608                     ((min_tx_space - tx_space) < pba)) {
609                         pba = pba - (min_tx_space - tx_space);
610
611                         /* PCI/PCIx hardware has PBA alignment constraints */
612                         switch (hw->mac_type) {
613                         case e1000_82545 ... e1000_82546_rev_3:
614                                 pba &= ~(E1000_PBA_8K - 1);
615                                 break;
616                         default:
617                                 break;
618                         }
619
620                         /* if short on rx space, rx wins and must trump tx
621                          * adjustment or use Early Receive if available */
622                         if (pba < min_rx_space)
623                                 pba = min_rx_space;
624                 }
625         }
626
627         ew32(PBA, pba);
628
629         /*
630          * flow control settings:
631          * The high water mark must be low enough to fit one full frame
632          * (or the size used for early receive) above it in the Rx FIFO.
633          * Set it to the lower of:
634          * - 90% of the Rx FIFO size, and
635          * - the full Rx FIFO size minus the early receive size (for parts
636          *   with ERT support assuming ERT set to E1000_ERT_2048), or
637          * - the full Rx FIFO size minus one full frame
638          */
639         hwm = min(((pba << 10) * 9 / 10),
640                   ((pba << 10) - hw->max_frame_size));
641
642         hw->fc_high_water = hwm & 0xFFF8;       /* 8-byte granularity */
643         hw->fc_low_water = hw->fc_high_water - 8;
644         hw->fc_pause_time = E1000_FC_PAUSE_TIME;
645         hw->fc_send_xon = 1;
646         hw->fc = hw->original_fc;
647
648         /* Allow time for pending master requests to run */
649         e1000_reset_hw(hw);
650         if (hw->mac_type >= e1000_82544)
651                 ew32(WUC, 0);
652
653         if (e1000_init_hw(hw))
654                 DPRINTK(PROBE, ERR, "Hardware Error\n");
655         e1000_update_mng_vlan(adapter);
656
657         /* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
658         if (hw->mac_type >= e1000_82544 &&
659             hw->autoneg == 1 &&
660             hw->autoneg_advertised == ADVERTISE_1000_FULL) {
661                 u32 ctrl = er32(CTRL);
662                 /* clear phy power management bit if we are in gig only mode,
663                  * which if enabled will attempt negotiation to 100Mb, which
664                  * can cause a loss of link at power off or driver unload */
665                 ctrl &= ~E1000_CTRL_SWDPIN3;
666                 ew32(CTRL, ctrl);
667         }
668
669         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
670         ew32(VET, ETHERNET_IEEE_VLAN_TYPE);
671
672         e1000_reset_adaptive(hw);
673         e1000_phy_get_info(hw, &adapter->phy_info);
674
675         e1000_release_manageability(adapter);
676 }
677
678 /**
679  *  Dump the eeprom for users having checksum issues
680  **/
681 static void e1000_dump_eeprom(struct e1000_adapter *adapter)
682 {
683         struct net_device *netdev = adapter->netdev;
684         struct ethtool_eeprom eeprom;
685         const struct ethtool_ops *ops = netdev->ethtool_ops;
686         u8 *data;
687         int i;
688         u16 csum_old, csum_new = 0;
689
690         eeprom.len = ops->get_eeprom_len(netdev);
691         eeprom.offset = 0;
692
693         data = kmalloc(eeprom.len, GFP_KERNEL);
694         if (!data) {
695                 printk(KERN_ERR "Unable to allocate memory to dump EEPROM"
696                        " data\n");
697                 return;
698         }
699
700         ops->get_eeprom(netdev, &eeprom, data);
701
702         csum_old = (data[EEPROM_CHECKSUM_REG * 2]) +
703                    (data[EEPROM_CHECKSUM_REG * 2 + 1] << 8);
704         for (i = 0; i < EEPROM_CHECKSUM_REG * 2; i += 2)
705                 csum_new += data[i] + (data[i + 1] << 8);
706         csum_new = EEPROM_SUM - csum_new;
707
708         printk(KERN_ERR "/*********************/\n");
709         printk(KERN_ERR "Current EEPROM Checksum : 0x%04x\n", csum_old);
710         printk(KERN_ERR "Calculated              : 0x%04x\n", csum_new);
711
712         printk(KERN_ERR "Offset    Values\n");
713         printk(KERN_ERR "========  ======\n");
714         print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 16, 1, data, 128, 0);
715
716         printk(KERN_ERR "Include this output when contacting your support "
717                "provider.\n");
718         printk(KERN_ERR "This is not a software error! Something bad "
719                "happened to your hardware or\n");
720         printk(KERN_ERR "EEPROM image. Ignoring this "
721                "problem could result in further problems,\n");
722         printk(KERN_ERR "possibly loss of data, corruption or system hangs!\n");
723         printk(KERN_ERR "The MAC Address will be reset to 00:00:00:00:00:00, "
724                "which is invalid\n");
725         printk(KERN_ERR "and requires you to set the proper MAC "
726                "address manually before continuing\n");
727         printk(KERN_ERR "to enable this network device.\n");
728         printk(KERN_ERR "Please inspect the EEPROM dump and report the issue "
729                "to your hardware vendor\n");
730         printk(KERN_ERR "or Intel Customer Support.\n");
731         printk(KERN_ERR "/*********************/\n");
732
733         kfree(data);
734 }
735
736 /**
737  * e1000_is_need_ioport - determine if an adapter needs ioport resources or not
738  * @pdev: PCI device information struct
739  *
740  * Return true if an adapter needs ioport resources
741  **/
742 static int e1000_is_need_ioport(struct pci_dev *pdev)
743 {
744         switch (pdev->device) {
745         case E1000_DEV_ID_82540EM:
746         case E1000_DEV_ID_82540EM_LOM:
747         case E1000_DEV_ID_82540EP:
748         case E1000_DEV_ID_82540EP_LOM:
749         case E1000_DEV_ID_82540EP_LP:
750         case E1000_DEV_ID_82541EI:
751         case E1000_DEV_ID_82541EI_MOBILE:
752         case E1000_DEV_ID_82541ER:
753         case E1000_DEV_ID_82541ER_LOM:
754         case E1000_DEV_ID_82541GI:
755         case E1000_DEV_ID_82541GI_LF:
756         case E1000_DEV_ID_82541GI_MOBILE:
757         case E1000_DEV_ID_82544EI_COPPER:
758         case E1000_DEV_ID_82544EI_FIBER:
759         case E1000_DEV_ID_82544GC_COPPER:
760         case E1000_DEV_ID_82544GC_LOM:
761         case E1000_DEV_ID_82545EM_COPPER:
762         case E1000_DEV_ID_82545EM_FIBER:
763         case E1000_DEV_ID_82546EB_COPPER:
764         case E1000_DEV_ID_82546EB_FIBER:
765         case E1000_DEV_ID_82546EB_QUAD_COPPER:
766                 return true;
767         default:
768                 return false;
769         }
770 }
771
772 static const struct net_device_ops e1000_netdev_ops = {
773         .ndo_open               = e1000_open,
774         .ndo_stop               = e1000_close,
775         .ndo_start_xmit         = e1000_xmit_frame,
776         .ndo_get_stats          = e1000_get_stats,
777         .ndo_set_rx_mode        = e1000_set_rx_mode,
778         .ndo_set_mac_address    = e1000_set_mac,
779         .ndo_tx_timeout         = e1000_tx_timeout,
780         .ndo_change_mtu         = e1000_change_mtu,
781         .ndo_do_ioctl           = e1000_ioctl,
782         .ndo_validate_addr      = eth_validate_addr,
783
784         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
785         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
786         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
787 #ifdef CONFIG_NET_POLL_CONTROLLER
788         .ndo_poll_controller    = e1000_netpoll,
789 #endif
790 };
791
792 /**
793  * e1000_probe - Device Initialization Routine
794  * @pdev: PCI device information struct
795  * @ent: entry in e1000_pci_tbl
796  *
797  * Returns 0 on success, negative on failure
798  *
799  * e1000_probe initializes an adapter identified by a pci_dev structure.
800  * The OS initialization, configuring of the adapter private structure,
801  * and a hardware reset occur.
802  **/
803 static int __devinit e1000_probe(struct pci_dev *pdev,
804                                  const struct pci_device_id *ent)
805 {
806         struct net_device *netdev;
807         struct e1000_adapter *adapter;
808         struct e1000_hw *hw;
809
810         static int cards_found = 0;
811         static int global_quad_port_a = 0; /* global ksp3 port a indication */
812         int i, err, pci_using_dac;
813         u16 eeprom_data = 0;
814         u16 eeprom_apme_mask = E1000_EEPROM_APME;
815         int bars, need_ioport;
816
817         /* do not allocate ioport bars when not needed */
818         need_ioport = e1000_is_need_ioport(pdev);
819         if (need_ioport) {
820                 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
821                 err = pci_enable_device(pdev);
822         } else {
823                 bars = pci_select_bars(pdev, IORESOURCE_MEM);
824                 err = pci_enable_device_mem(pdev);
825         }
826         if (err)
827                 return err;
828
829         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) &&
830             !pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64))) {
831                 pci_using_dac = 1;
832         } else {
833                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
834                 if (err) {
835                         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
836                         if (err) {
837                                 E1000_ERR("No usable DMA configuration, "
838                                           "aborting\n");
839                                 goto err_dma;
840                         }
841                 }
842                 pci_using_dac = 0;
843         }
844
845         err = pci_request_selected_regions(pdev, bars, e1000_driver_name);
846         if (err)
847                 goto err_pci_reg;
848
849         pci_set_master(pdev);
850
851         err = -ENOMEM;
852         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
853         if (!netdev)
854                 goto err_alloc_etherdev;
855
856         SET_NETDEV_DEV(netdev, &pdev->dev);
857
858         pci_set_drvdata(pdev, netdev);
859         adapter = netdev_priv(netdev);
860         adapter->netdev = netdev;
861         adapter->pdev = pdev;
862         adapter->msg_enable = (1 << debug) - 1;
863         adapter->bars = bars;
864         adapter->need_ioport = need_ioport;
865
866         hw = &adapter->hw;
867         hw->back = adapter;
868
869         err = -EIO;
870         hw->hw_addr = pci_ioremap_bar(pdev, BAR_0);
871         if (!hw->hw_addr)
872                 goto err_ioremap;
873
874         if (adapter->need_ioport) {
875                 for (i = BAR_1; i <= BAR_5; i++) {
876                         if (pci_resource_len(pdev, i) == 0)
877                                 continue;
878                         if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
879                                 hw->io_base = pci_resource_start(pdev, i);
880                                 break;
881                         }
882                 }
883         }
884
885         netdev->netdev_ops = &e1000_netdev_ops;
886         e1000_set_ethtool_ops(netdev);
887         netdev->watchdog_timeo = 5 * HZ;
888         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
889
890         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
891
892         adapter->bd_number = cards_found;
893
894         /* setup the private structure */
895
896         err = e1000_sw_init(adapter);
897         if (err)
898                 goto err_sw_init;
899
900         err = -EIO;
901
902         if (hw->mac_type >= e1000_82543) {
903                 netdev->features = NETIF_F_SG |
904                                    NETIF_F_HW_CSUM |
905                                    NETIF_F_HW_VLAN_TX |
906                                    NETIF_F_HW_VLAN_RX |
907                                    NETIF_F_HW_VLAN_FILTER;
908         }
909
910         if ((hw->mac_type >= e1000_82544) &&
911            (hw->mac_type != e1000_82547))
912                 netdev->features |= NETIF_F_TSO;
913
914         if (pci_using_dac)
915                 netdev->features |= NETIF_F_HIGHDMA;
916
917         netdev->vlan_features |= NETIF_F_TSO;
918         netdev->vlan_features |= NETIF_F_HW_CSUM;
919         netdev->vlan_features |= NETIF_F_SG;
920
921         adapter->en_mng_pt = e1000_enable_mng_pass_thru(hw);
922
923         /* initialize eeprom parameters */
924         if (e1000_init_eeprom_params(hw)) {
925                 E1000_ERR("EEPROM initialization failed\n");
926                 goto err_eeprom;
927         }
928
929         /* before reading the EEPROM, reset the controller to
930          * put the device in a known good starting state */
931
932         e1000_reset_hw(hw);
933
934         /* make sure the EEPROM is good */
935         if (e1000_validate_eeprom_checksum(hw) < 0) {
936                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
937                 e1000_dump_eeprom(adapter);
938                 /*
939                  * set MAC address to all zeroes to invalidate and temporary
940                  * disable this device for the user. This blocks regular
941                  * traffic while still permitting ethtool ioctls from reaching
942                  * the hardware as well as allowing the user to run the
943                  * interface after manually setting a hw addr using
944                  * `ip set address`
945                  */
946                 memset(hw->mac_addr, 0, netdev->addr_len);
947         } else {
948                 /* copy the MAC address out of the EEPROM */
949                 if (e1000_read_mac_addr(hw))
950                         DPRINTK(PROBE, ERR, "EEPROM Read Error\n");
951         }
952         /* don't block initalization here due to bad MAC address */
953         memcpy(netdev->dev_addr, hw->mac_addr, netdev->addr_len);
954         memcpy(netdev->perm_addr, hw->mac_addr, netdev->addr_len);
955
956         if (!is_valid_ether_addr(netdev->perm_addr))
957                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
958
959         e1000_get_bus_info(hw);
960
961         init_timer(&adapter->tx_fifo_stall_timer);
962         adapter->tx_fifo_stall_timer.function = &e1000_82547_tx_fifo_stall;
963         adapter->tx_fifo_stall_timer.data = (unsigned long)adapter;
964
965         init_timer(&adapter->watchdog_timer);
966         adapter->watchdog_timer.function = &e1000_watchdog;
967         adapter->watchdog_timer.data = (unsigned long) adapter;
968
969         init_timer(&adapter->phy_info_timer);
970         adapter->phy_info_timer.function = &e1000_update_phy_info;
971         adapter->phy_info_timer.data = (unsigned long)adapter;
972
973         INIT_WORK(&adapter->reset_task, e1000_reset_task);
974
975         e1000_check_options(adapter);
976
977         /* Initial Wake on LAN setting
978          * If APM wake is enabled in the EEPROM,
979          * enable the ACPI Magic Packet filter
980          */
981
982         switch (hw->mac_type) {
983         case e1000_82542_rev2_0:
984         case e1000_82542_rev2_1:
985         case e1000_82543:
986                 break;
987         case e1000_82544:
988                 e1000_read_eeprom(hw,
989                         EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
990                 eeprom_apme_mask = E1000_EEPROM_82544_APM;
991                 break;
992         case e1000_82546:
993         case e1000_82546_rev_3:
994                 if (er32(STATUS) & E1000_STATUS_FUNC_1){
995                         e1000_read_eeprom(hw,
996                                 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
997                         break;
998                 }
999                 /* Fall Through */
1000         default:
1001                 e1000_read_eeprom(hw,
1002                         EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1003                 break;
1004         }
1005         if (eeprom_data & eeprom_apme_mask)
1006                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1007
1008         /* now that we have the eeprom settings, apply the special cases
1009          * where the eeprom may be wrong or the board simply won't support
1010          * wake on lan on a particular port */
1011         switch (pdev->device) {
1012         case E1000_DEV_ID_82546GB_PCIE:
1013                 adapter->eeprom_wol = 0;
1014                 break;
1015         case E1000_DEV_ID_82546EB_FIBER:
1016         case E1000_DEV_ID_82546GB_FIBER:
1017                 /* Wake events only supported on port A for dual fiber
1018                  * regardless of eeprom setting */
1019                 if (er32(STATUS) & E1000_STATUS_FUNC_1)
1020                         adapter->eeprom_wol = 0;
1021                 break;
1022         case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
1023                 /* if quad port adapter, disable WoL on all but port A */
1024                 if (global_quad_port_a != 0)
1025                         adapter->eeprom_wol = 0;
1026                 else
1027                         adapter->quad_port_a = 1;
1028                 /* Reset for multiple quad port adapters */
1029                 if (++global_quad_port_a == 4)
1030                         global_quad_port_a = 0;
1031                 break;
1032         }
1033
1034         /* initialize the wol settings based on the eeprom settings */
1035         adapter->wol = adapter->eeprom_wol;
1036         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1037
1038         /* print bus type/speed/width info */
1039         DPRINTK(PROBE, INFO, "(PCI%s:%s:%s) ",
1040                 ((hw->bus_type == e1000_bus_type_pcix) ? "-X" : ""),
1041                 ((hw->bus_speed == e1000_bus_speed_133) ? "133MHz" :
1042                  (hw->bus_speed == e1000_bus_speed_120) ? "120MHz" :
1043                  (hw->bus_speed == e1000_bus_speed_100) ? "100MHz" :
1044                  (hw->bus_speed == e1000_bus_speed_66) ? "66MHz" : "33MHz"),
1045                 ((hw->bus_width == e1000_bus_width_64) ? "64-bit" : "32-bit"));
1046
1047         printk("%pM\n", netdev->dev_addr);
1048
1049         /* reset the hardware with the new settings */
1050         e1000_reset(adapter);
1051
1052         strcpy(netdev->name, "eth%d");
1053         err = register_netdev(netdev);
1054         if (err)
1055                 goto err_register;
1056
1057         /* carrier off reporting is important to ethtool even BEFORE open */
1058         netif_carrier_off(netdev);
1059
1060         DPRINTK(PROBE, INFO, "Intel(R) PRO/1000 Network Connection\n");
1061
1062         cards_found++;
1063         return 0;
1064
1065 err_register:
1066 err_eeprom:
1067         e1000_phy_hw_reset(hw);
1068
1069         if (hw->flash_address)
1070                 iounmap(hw->flash_address);
1071         kfree(adapter->tx_ring);
1072         kfree(adapter->rx_ring);
1073 err_sw_init:
1074         iounmap(hw->hw_addr);
1075 err_ioremap:
1076         free_netdev(netdev);
1077 err_alloc_etherdev:
1078         pci_release_selected_regions(pdev, bars);
1079 err_pci_reg:
1080 err_dma:
1081         pci_disable_device(pdev);
1082         return err;
1083 }
1084
1085 /**
1086  * e1000_remove - Device Removal Routine
1087  * @pdev: PCI device information struct
1088  *
1089  * e1000_remove is called by the PCI subsystem to alert the driver
1090  * that it should release a PCI device.  The could be caused by a
1091  * Hot-Plug event, or because the driver is going to be removed from
1092  * memory.
1093  **/
1094
1095 static void __devexit e1000_remove(struct pci_dev *pdev)
1096 {
1097         struct net_device *netdev = pci_get_drvdata(pdev);
1098         struct e1000_adapter *adapter = netdev_priv(netdev);
1099         struct e1000_hw *hw = &adapter->hw;
1100
1101         set_bit(__E1000_DOWN, &adapter->flags);
1102         del_timer_sync(&adapter->tx_fifo_stall_timer);
1103         del_timer_sync(&adapter->watchdog_timer);
1104         del_timer_sync(&adapter->phy_info_timer);
1105
1106         cancel_work_sync(&adapter->reset_task);
1107
1108         e1000_release_manageability(adapter);
1109
1110         unregister_netdev(netdev);
1111
1112         e1000_phy_hw_reset(hw);
1113
1114         kfree(adapter->tx_ring);
1115         kfree(adapter->rx_ring);
1116
1117         iounmap(hw->hw_addr);
1118         if (hw->flash_address)
1119                 iounmap(hw->flash_address);
1120         pci_release_selected_regions(pdev, adapter->bars);
1121
1122         free_netdev(netdev);
1123
1124         pci_disable_device(pdev);
1125 }
1126
1127 /**
1128  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1129  * @adapter: board private structure to initialize
1130  *
1131  * e1000_sw_init initializes the Adapter private data structure.
1132  * Fields are initialized based on PCI device information and
1133  * OS network device settings (MTU size).
1134  **/
1135
1136 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
1137 {
1138         struct e1000_hw *hw = &adapter->hw;
1139         struct net_device *netdev = adapter->netdev;
1140         struct pci_dev *pdev = adapter->pdev;
1141
1142         /* PCI config space info */
1143
1144         hw->vendor_id = pdev->vendor;
1145         hw->device_id = pdev->device;
1146         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1147         hw->subsystem_id = pdev->subsystem_device;
1148         hw->revision_id = pdev->revision;
1149
1150         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
1151
1152         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1153         hw->max_frame_size = netdev->mtu +
1154                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
1155         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
1156
1157         /* identify the MAC */
1158
1159         if (e1000_set_mac_type(hw)) {
1160                 DPRINTK(PROBE, ERR, "Unknown MAC Type\n");
1161                 return -EIO;
1162         }
1163
1164         switch (hw->mac_type) {
1165         default:
1166                 break;
1167         case e1000_82541:
1168         case e1000_82547:
1169         case e1000_82541_rev_2:
1170         case e1000_82547_rev_2:
1171                 hw->phy_init_script = 1;
1172                 break;
1173         }
1174
1175         e1000_set_media_type(hw);
1176
1177         hw->wait_autoneg_complete = false;
1178         hw->tbi_compatibility_en = true;
1179         hw->adaptive_ifs = true;
1180
1181         /* Copper options */
1182
1183         if (hw->media_type == e1000_media_type_copper) {
1184                 hw->mdix = AUTO_ALL_MODES;
1185                 hw->disable_polarity_correction = false;
1186                 hw->master_slave = E1000_MASTER_SLAVE;
1187         }
1188
1189         adapter->num_tx_queues = 1;
1190         adapter->num_rx_queues = 1;
1191
1192         if (e1000_alloc_queues(adapter)) {
1193                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
1194                 return -ENOMEM;
1195         }
1196
1197         /* Explicitly disable IRQ since the NIC can be in any state. */
1198         e1000_irq_disable(adapter);
1199
1200         spin_lock_init(&adapter->stats_lock);
1201
1202         set_bit(__E1000_DOWN, &adapter->flags);
1203
1204         return 0;
1205 }
1206
1207 /**
1208  * e1000_alloc_queues - Allocate memory for all rings
1209  * @adapter: board private structure to initialize
1210  *
1211  * We allocate one ring per queue at run-time since we don't know the
1212  * number of queues at compile-time.
1213  **/
1214
1215 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1216 {
1217         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1218                                    sizeof(struct e1000_tx_ring), GFP_KERNEL);
1219         if (!adapter->tx_ring)
1220                 return -ENOMEM;
1221
1222         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1223                                    sizeof(struct e1000_rx_ring), GFP_KERNEL);
1224         if (!adapter->rx_ring) {
1225                 kfree(adapter->tx_ring);
1226                 return -ENOMEM;
1227         }
1228
1229         return E1000_SUCCESS;
1230 }
1231
1232 /**
1233  * e1000_open - Called when a network interface is made active
1234  * @netdev: network interface device structure
1235  *
1236  * Returns 0 on success, negative value on failure
1237  *
1238  * The open entry point is called when a network interface is made
1239  * active by the system (IFF_UP).  At this point all resources needed
1240  * for transmit and receive operations are allocated, the interrupt
1241  * handler is registered with the OS, the watchdog timer is started,
1242  * and the stack is notified that the interface is ready.
1243  **/
1244
1245 static int e1000_open(struct net_device *netdev)
1246 {
1247         struct e1000_adapter *adapter = netdev_priv(netdev);
1248         struct e1000_hw *hw = &adapter->hw;
1249         int err;
1250
1251         /* disallow open during test */
1252         if (test_bit(__E1000_TESTING, &adapter->flags))
1253                 return -EBUSY;
1254
1255         netif_carrier_off(netdev);
1256
1257         /* allocate transmit descriptors */
1258         err = e1000_setup_all_tx_resources(adapter);
1259         if (err)
1260                 goto err_setup_tx;
1261
1262         /* allocate receive descriptors */
1263         err = e1000_setup_all_rx_resources(adapter);
1264         if (err)
1265                 goto err_setup_rx;
1266
1267         e1000_power_up_phy(adapter);
1268
1269         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1270         if ((hw->mng_cookie.status &
1271                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1272                 e1000_update_mng_vlan(adapter);
1273         }
1274
1275         /* before we allocate an interrupt, we must be ready to handle it.
1276          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1277          * as soon as we call pci_request_irq, so we have to setup our
1278          * clean_rx handler before we do so.  */
1279         e1000_configure(adapter);
1280
1281         err = e1000_request_irq(adapter);
1282         if (err)
1283                 goto err_req_irq;
1284
1285         /* From here on the code is the same as e1000_up() */
1286         clear_bit(__E1000_DOWN, &adapter->flags);
1287
1288         napi_enable(&adapter->napi);
1289
1290         e1000_irq_enable(adapter);
1291
1292         netif_start_queue(netdev);
1293
1294         /* fire a link status change interrupt to start the watchdog */
1295         ew32(ICS, E1000_ICS_LSC);
1296
1297         return E1000_SUCCESS;
1298
1299 err_req_irq:
1300         e1000_power_down_phy(adapter);
1301         e1000_free_all_rx_resources(adapter);
1302 err_setup_rx:
1303         e1000_free_all_tx_resources(adapter);
1304 err_setup_tx:
1305         e1000_reset(adapter);
1306
1307         return err;
1308 }
1309
1310 /**
1311  * e1000_close - Disables a network interface
1312  * @netdev: network interface device structure
1313  *
1314  * Returns 0, this is not allowed to fail
1315  *
1316  * The close entry point is called when an interface is de-activated
1317  * by the OS.  The hardware is still under the drivers control, but
1318  * needs to be disabled.  A global MAC reset is issued to stop the
1319  * hardware, and all transmit and receive resources are freed.
1320  **/
1321
1322 static int e1000_close(struct net_device *netdev)
1323 {
1324         struct e1000_adapter *adapter = netdev_priv(netdev);
1325         struct e1000_hw *hw = &adapter->hw;
1326
1327         WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1328         e1000_down(adapter);
1329         e1000_power_down_phy(adapter);
1330         e1000_free_irq(adapter);
1331
1332         e1000_free_all_tx_resources(adapter);
1333         e1000_free_all_rx_resources(adapter);
1334
1335         /* kill manageability vlan ID if supported, but not if a vlan with
1336          * the same ID is registered on the host OS (let 8021q kill it) */
1337         if ((hw->mng_cookie.status &
1338                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1339              !(adapter->vlgrp &&
1340                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id))) {
1341                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1342         }
1343
1344         return 0;
1345 }
1346
1347 /**
1348  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1349  * @adapter: address of board private structure
1350  * @start: address of beginning of memory
1351  * @len: length of memory
1352  **/
1353 static bool e1000_check_64k_bound(struct e1000_adapter *adapter, void *start,
1354                                   unsigned long len)
1355 {
1356         struct e1000_hw *hw = &adapter->hw;
1357         unsigned long begin = (unsigned long)start;
1358         unsigned long end = begin + len;
1359
1360         /* First rev 82545 and 82546 need to not allow any memory
1361          * write location to cross 64k boundary due to errata 23 */
1362         if (hw->mac_type == e1000_82545 ||
1363             hw->mac_type == e1000_82546) {
1364                 return ((begin ^ (end - 1)) >> 16) != 0 ? false : true;
1365         }
1366
1367         return true;
1368 }
1369
1370 /**
1371  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1372  * @adapter: board private structure
1373  * @txdr:    tx descriptor ring (for a specific queue) to setup
1374  *
1375  * Return 0 on success, negative on failure
1376  **/
1377
1378 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
1379                                     struct e1000_tx_ring *txdr)
1380 {
1381         struct pci_dev *pdev = adapter->pdev;
1382         int size;
1383
1384         size = sizeof(struct e1000_buffer) * txdr->count;
1385         txdr->buffer_info = vmalloc(size);
1386         if (!txdr->buffer_info) {
1387                 DPRINTK(PROBE, ERR,
1388                 "Unable to allocate memory for the transmit descriptor ring\n");
1389                 return -ENOMEM;
1390         }
1391         memset(txdr->buffer_info, 0, size);
1392
1393         /* round up to nearest 4K */
1394
1395         txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1396         txdr->size = ALIGN(txdr->size, 4096);
1397
1398         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1399         if (!txdr->desc) {
1400 setup_tx_desc_die:
1401                 vfree(txdr->buffer_info);
1402                 DPRINTK(PROBE, ERR,
1403                 "Unable to allocate memory for the transmit descriptor ring\n");
1404                 return -ENOMEM;
1405         }
1406
1407         /* Fix for errata 23, can't cross 64kB boundary */
1408         if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1409                 void *olddesc = txdr->desc;
1410                 dma_addr_t olddma = txdr->dma;
1411                 DPRINTK(TX_ERR, ERR, "txdr align check failed: %u bytes "
1412                                      "at %p\n", txdr->size, txdr->desc);
1413                 /* Try again, without freeing the previous */
1414                 txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1415                 /* Failed allocation, critical failure */
1416                 if (!txdr->desc) {
1417                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1418                         goto setup_tx_desc_die;
1419                 }
1420
1421                 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1422                         /* give up */
1423                         pci_free_consistent(pdev, txdr->size, txdr->desc,
1424                                             txdr->dma);
1425                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1426                         DPRINTK(PROBE, ERR,
1427                                 "Unable to allocate aligned memory "
1428                                 "for the transmit descriptor ring\n");
1429                         vfree(txdr->buffer_info);
1430                         return -ENOMEM;
1431                 } else {
1432                         /* Free old allocation, new allocation was successful */
1433                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1434                 }
1435         }
1436         memset(txdr->desc, 0, txdr->size);
1437
1438         txdr->next_to_use = 0;
1439         txdr->next_to_clean = 0;
1440
1441         return 0;
1442 }
1443
1444 /**
1445  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1446  *                                (Descriptors) for all queues
1447  * @adapter: board private structure
1448  *
1449  * Return 0 on success, negative on failure
1450  **/
1451
1452 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1453 {
1454         int i, err = 0;
1455
1456         for (i = 0; i < adapter->num_tx_queues; i++) {
1457                 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1458                 if (err) {
1459                         DPRINTK(PROBE, ERR,
1460                                 "Allocation for Tx Queue %u failed\n", i);
1461                         for (i-- ; i >= 0; i--)
1462                                 e1000_free_tx_resources(adapter,
1463                                                         &adapter->tx_ring[i]);
1464                         break;
1465                 }
1466         }
1467
1468         return err;
1469 }
1470
1471 /**
1472  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1473  * @adapter: board private structure
1474  *
1475  * Configure the Tx unit of the MAC after a reset.
1476  **/
1477
1478 static void e1000_configure_tx(struct e1000_adapter *adapter)
1479 {
1480         u64 tdba;
1481         struct e1000_hw *hw = &adapter->hw;
1482         u32 tdlen, tctl, tipg;
1483         u32 ipgr1, ipgr2;
1484
1485         /* Setup the HW Tx Head and Tail descriptor pointers */
1486
1487         switch (adapter->num_tx_queues) {
1488         case 1:
1489         default:
1490                 tdba = adapter->tx_ring[0].dma;
1491                 tdlen = adapter->tx_ring[0].count *
1492                         sizeof(struct e1000_tx_desc);
1493                 ew32(TDLEN, tdlen);
1494                 ew32(TDBAH, (tdba >> 32));
1495                 ew32(TDBAL, (tdba & 0x00000000ffffffffULL));
1496                 ew32(TDT, 0);
1497                 ew32(TDH, 0);
1498                 adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ? E1000_TDH : E1000_82542_TDH);
1499                 adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ? E1000_TDT : E1000_82542_TDT);
1500                 break;
1501         }
1502
1503         /* Set the default values for the Tx Inter Packet Gap timer */
1504         if ((hw->media_type == e1000_media_type_fiber ||
1505              hw->media_type == e1000_media_type_internal_serdes))
1506                 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1507         else
1508                 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1509
1510         switch (hw->mac_type) {
1511         case e1000_82542_rev2_0:
1512         case e1000_82542_rev2_1:
1513                 tipg = DEFAULT_82542_TIPG_IPGT;
1514                 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1515                 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1516                 break;
1517         default:
1518                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1519                 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1520                 break;
1521         }
1522         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1523         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1524         ew32(TIPG, tipg);
1525
1526         /* Set the Tx Interrupt Delay register */
1527
1528         ew32(TIDV, adapter->tx_int_delay);
1529         if (hw->mac_type >= e1000_82540)
1530                 ew32(TADV, adapter->tx_abs_int_delay);
1531
1532         /* Program the Transmit Control Register */
1533
1534         tctl = er32(TCTL);
1535         tctl &= ~E1000_TCTL_CT;
1536         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1537                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1538
1539         e1000_config_collision_dist(hw);
1540
1541         /* Setup Transmit Descriptor Settings for eop descriptor */
1542         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1543
1544         /* only set IDE if we are delaying interrupts using the timers */
1545         if (adapter->tx_int_delay)
1546                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1547
1548         if (hw->mac_type < e1000_82543)
1549                 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1550         else
1551                 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1552
1553         /* Cache if we're 82544 running in PCI-X because we'll
1554          * need this to apply a workaround later in the send path. */
1555         if (hw->mac_type == e1000_82544 &&
1556             hw->bus_type == e1000_bus_type_pcix)
1557                 adapter->pcix_82544 = 1;
1558
1559         ew32(TCTL, tctl);
1560
1561 }
1562
1563 /**
1564  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1565  * @adapter: board private structure
1566  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1567  *
1568  * Returns 0 on success, negative on failure
1569  **/
1570
1571 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
1572                                     struct e1000_rx_ring *rxdr)
1573 {
1574         struct pci_dev *pdev = adapter->pdev;
1575         int size, desc_len;
1576
1577         size = sizeof(struct e1000_buffer) * rxdr->count;
1578         rxdr->buffer_info = vmalloc(size);
1579         if (!rxdr->buffer_info) {
1580                 DPRINTK(PROBE, ERR,
1581                 "Unable to allocate memory for the receive descriptor ring\n");
1582                 return -ENOMEM;
1583         }
1584         memset(rxdr->buffer_info, 0, size);
1585
1586         desc_len = sizeof(struct e1000_rx_desc);
1587
1588         /* Round up to nearest 4K */
1589
1590         rxdr->size = rxdr->count * desc_len;
1591         rxdr->size = ALIGN(rxdr->size, 4096);
1592
1593         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1594
1595         if (!rxdr->desc) {
1596                 DPRINTK(PROBE, ERR,
1597                 "Unable to allocate memory for the receive descriptor ring\n");
1598 setup_rx_desc_die:
1599                 vfree(rxdr->buffer_info);
1600                 return -ENOMEM;
1601         }
1602
1603         /* Fix for errata 23, can't cross 64kB boundary */
1604         if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1605                 void *olddesc = rxdr->desc;
1606                 dma_addr_t olddma = rxdr->dma;
1607                 DPRINTK(RX_ERR, ERR, "rxdr align check failed: %u bytes "
1608                                      "at %p\n", rxdr->size, rxdr->desc);
1609                 /* Try again, without freeing the previous */
1610                 rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1611                 /* Failed allocation, critical failure */
1612                 if (!rxdr->desc) {
1613                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1614                         DPRINTK(PROBE, ERR,
1615                                 "Unable to allocate memory "
1616                                 "for the receive descriptor ring\n");
1617                         goto setup_rx_desc_die;
1618                 }
1619
1620                 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1621                         /* give up */
1622                         pci_free_consistent(pdev, rxdr->size, rxdr->desc,
1623                                             rxdr->dma);
1624                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1625                         DPRINTK(PROBE, ERR,
1626                                 "Unable to allocate aligned memory "
1627                                 "for the receive descriptor ring\n");
1628                         goto setup_rx_desc_die;
1629                 } else {
1630                         /* Free old allocation, new allocation was successful */
1631                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1632                 }
1633         }
1634         memset(rxdr->desc, 0, rxdr->size);
1635
1636         rxdr->next_to_clean = 0;
1637         rxdr->next_to_use = 0;
1638         rxdr->rx_skb_top = NULL;
1639
1640         return 0;
1641 }
1642
1643 /**
1644  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1645  *                                (Descriptors) for all queues
1646  * @adapter: board private structure
1647  *
1648  * Return 0 on success, negative on failure
1649  **/
1650
1651 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1652 {
1653         int i, err = 0;
1654
1655         for (i = 0; i < adapter->num_rx_queues; i++) {
1656                 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1657                 if (err) {
1658                         DPRINTK(PROBE, ERR,
1659                                 "Allocation for Rx Queue %u failed\n", i);
1660                         for (i-- ; i >= 0; i--)
1661                                 e1000_free_rx_resources(adapter,
1662                                                         &adapter->rx_ring[i]);
1663                         break;
1664                 }
1665         }
1666
1667         return err;
1668 }
1669
1670 /**
1671  * e1000_setup_rctl - configure the receive control registers
1672  * @adapter: Board private structure
1673  **/
1674 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1675 {
1676         struct e1000_hw *hw = &adapter->hw;
1677         u32 rctl;
1678
1679         rctl = er32(RCTL);
1680
1681         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1682
1683         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1684                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1685                 (hw->mc_filter_type << E1000_RCTL_MO_SHIFT);
1686
1687         if (hw->tbi_compatibility_on == 1)
1688                 rctl |= E1000_RCTL_SBP;
1689         else
1690                 rctl &= ~E1000_RCTL_SBP;
1691
1692         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1693                 rctl &= ~E1000_RCTL_LPE;
1694         else
1695                 rctl |= E1000_RCTL_LPE;
1696
1697         /* Setup buffer sizes */
1698         rctl &= ~E1000_RCTL_SZ_4096;
1699         rctl |= E1000_RCTL_BSEX;
1700         switch (adapter->rx_buffer_len) {
1701                 case E1000_RXBUFFER_256:
1702                         rctl |= E1000_RCTL_SZ_256;
1703                         rctl &= ~E1000_RCTL_BSEX;
1704                         break;
1705                 case E1000_RXBUFFER_512:
1706                         rctl |= E1000_RCTL_SZ_512;
1707                         rctl &= ~E1000_RCTL_BSEX;
1708                         break;
1709                 case E1000_RXBUFFER_1024:
1710                         rctl |= E1000_RCTL_SZ_1024;
1711                         rctl &= ~E1000_RCTL_BSEX;
1712                         break;
1713                 case E1000_RXBUFFER_2048:
1714                 default:
1715                         rctl |= E1000_RCTL_SZ_2048;
1716                         rctl &= ~E1000_RCTL_BSEX;
1717                         break;
1718                 case E1000_RXBUFFER_4096:
1719                         rctl |= E1000_RCTL_SZ_4096;
1720                         break;
1721                 case E1000_RXBUFFER_8192:
1722                         rctl |= E1000_RCTL_SZ_8192;
1723                         break;
1724                 case E1000_RXBUFFER_16384:
1725                         rctl |= E1000_RCTL_SZ_16384;
1726                         break;
1727         }
1728
1729         ew32(RCTL, rctl);
1730 }
1731
1732 /**
1733  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1734  * @adapter: board private structure
1735  *
1736  * Configure the Rx unit of the MAC after a reset.
1737  **/
1738
1739 static void e1000_configure_rx(struct e1000_adapter *adapter)
1740 {
1741         u64 rdba;
1742         struct e1000_hw *hw = &adapter->hw;
1743         u32 rdlen, rctl, rxcsum;
1744
1745         if (adapter->netdev->mtu > ETH_DATA_LEN) {
1746                 rdlen = adapter->rx_ring[0].count *
1747                         sizeof(struct e1000_rx_desc);
1748                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
1749                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
1750         } else {
1751                 rdlen = adapter->rx_ring[0].count *
1752                         sizeof(struct e1000_rx_desc);
1753                 adapter->clean_rx = e1000_clean_rx_irq;
1754                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1755         }
1756
1757         /* disable receives while setting up the descriptors */
1758         rctl = er32(RCTL);
1759         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1760
1761         /* set the Receive Delay Timer Register */
1762         ew32(RDTR, adapter->rx_int_delay);
1763
1764         if (hw->mac_type >= e1000_82540) {
1765                 ew32(RADV, adapter->rx_abs_int_delay);
1766                 if (adapter->itr_setting != 0)
1767                         ew32(ITR, 1000000000 / (adapter->itr * 256));
1768         }
1769
1770         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1771          * the Base and Length of the Rx Descriptor Ring */
1772         switch (adapter->num_rx_queues) {
1773         case 1:
1774         default:
1775                 rdba = adapter->rx_ring[0].dma;
1776                 ew32(RDLEN, rdlen);
1777                 ew32(RDBAH, (rdba >> 32));
1778                 ew32(RDBAL, (rdba & 0x00000000ffffffffULL));
1779                 ew32(RDT, 0);
1780                 ew32(RDH, 0);
1781                 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ? E1000_RDH : E1000_82542_RDH);
1782                 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ? E1000_RDT : E1000_82542_RDT);
1783                 break;
1784         }
1785
1786         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
1787         if (hw->mac_type >= e1000_82543) {
1788                 rxcsum = er32(RXCSUM);
1789                 if (adapter->rx_csum)
1790                         rxcsum |= E1000_RXCSUM_TUOFL;
1791                 else
1792                         /* don't need to clear IPPCSE as it defaults to 0 */
1793                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1794                 ew32(RXCSUM, rxcsum);
1795         }
1796
1797         /* Enable Receives */
1798         ew32(RCTL, rctl);
1799 }
1800
1801 /**
1802  * e1000_free_tx_resources - Free Tx Resources per Queue
1803  * @adapter: board private structure
1804  * @tx_ring: Tx descriptor ring for a specific queue
1805  *
1806  * Free all transmit software resources
1807  **/
1808
1809 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
1810                                     struct e1000_tx_ring *tx_ring)
1811 {
1812         struct pci_dev *pdev = adapter->pdev;
1813
1814         e1000_clean_tx_ring(adapter, tx_ring);
1815
1816         vfree(tx_ring->buffer_info);
1817         tx_ring->buffer_info = NULL;
1818
1819         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
1820
1821         tx_ring->desc = NULL;
1822 }
1823
1824 /**
1825  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1826  * @adapter: board private structure
1827  *
1828  * Free all transmit software resources
1829  **/
1830
1831 void e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1832 {
1833         int i;
1834
1835         for (i = 0; i < adapter->num_tx_queues; i++)
1836                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1837 }
1838
1839 static void e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1840                                              struct e1000_buffer *buffer_info)
1841 {
1842         buffer_info->dma = 0;
1843         if (buffer_info->skb) {
1844                 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
1845                               DMA_TO_DEVICE);
1846                 dev_kfree_skb_any(buffer_info->skb);
1847                 buffer_info->skb = NULL;
1848         }
1849         buffer_info->time_stamp = 0;
1850         /* buffer_info must be completely set up in the transmit path */
1851 }
1852
1853 /**
1854  * e1000_clean_tx_ring - Free Tx Buffers
1855  * @adapter: board private structure
1856  * @tx_ring: ring to be cleaned
1857  **/
1858
1859 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
1860                                 struct e1000_tx_ring *tx_ring)
1861 {
1862         struct e1000_hw *hw = &adapter->hw;
1863         struct e1000_buffer *buffer_info;
1864         unsigned long size;
1865         unsigned int i;
1866
1867         /* Free all the Tx ring sk_buffs */
1868
1869         for (i = 0; i < tx_ring->count; i++) {
1870                 buffer_info = &tx_ring->buffer_info[i];
1871                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
1872         }
1873
1874         size = sizeof(struct e1000_buffer) * tx_ring->count;
1875         memset(tx_ring->buffer_info, 0, size);
1876
1877         /* Zero out the descriptor ring */
1878
1879         memset(tx_ring->desc, 0, tx_ring->size);
1880
1881         tx_ring->next_to_use = 0;
1882         tx_ring->next_to_clean = 0;
1883         tx_ring->last_tx_tso = 0;
1884
1885         writel(0, hw->hw_addr + tx_ring->tdh);
1886         writel(0, hw->hw_addr + tx_ring->tdt);
1887 }
1888
1889 /**
1890  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
1891  * @adapter: board private structure
1892  **/
1893
1894 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
1895 {
1896         int i;
1897
1898         for (i = 0; i < adapter->num_tx_queues; i++)
1899                 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1900 }
1901
1902 /**
1903  * e1000_free_rx_resources - Free Rx Resources
1904  * @adapter: board private structure
1905  * @rx_ring: ring to clean the resources from
1906  *
1907  * Free all receive software resources
1908  **/
1909
1910 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
1911                                     struct e1000_rx_ring *rx_ring)
1912 {
1913         struct pci_dev *pdev = adapter->pdev;
1914
1915         e1000_clean_rx_ring(adapter, rx_ring);
1916
1917         vfree(rx_ring->buffer_info);
1918         rx_ring->buffer_info = NULL;
1919
1920         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
1921
1922         rx_ring->desc = NULL;
1923 }
1924
1925 /**
1926  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
1927  * @adapter: board private structure
1928  *
1929  * Free all receive software resources
1930  **/
1931
1932 void e1000_free_all_rx_resources(struct e1000_adapter *adapter)
1933 {
1934         int i;
1935
1936         for (i = 0; i < adapter->num_rx_queues; i++)
1937                 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
1938 }
1939
1940 /**
1941  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1942  * @adapter: board private structure
1943  * @rx_ring: ring to free buffers from
1944  **/
1945
1946 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
1947                                 struct e1000_rx_ring *rx_ring)
1948 {
1949         struct e1000_hw *hw = &adapter->hw;
1950         struct e1000_buffer *buffer_info;
1951         struct pci_dev *pdev = adapter->pdev;
1952         unsigned long size;
1953         unsigned int i;
1954
1955         /* Free all the Rx ring sk_buffs */
1956         for (i = 0; i < rx_ring->count; i++) {
1957                 buffer_info = &rx_ring->buffer_info[i];
1958                 if (buffer_info->dma &&
1959                     adapter->clean_rx == e1000_clean_rx_irq) {
1960                         pci_unmap_single(pdev, buffer_info->dma,
1961                                          buffer_info->length,
1962                                          PCI_DMA_FROMDEVICE);
1963                 } else if (buffer_info->dma &&
1964                            adapter->clean_rx == e1000_clean_jumbo_rx_irq) {
1965                         pci_unmap_page(pdev, buffer_info->dma,
1966                                        buffer_info->length,
1967                                        PCI_DMA_FROMDEVICE);
1968                 }
1969
1970                 buffer_info->dma = 0;
1971                 if (buffer_info->page) {
1972                         put_page(buffer_info->page);
1973                         buffer_info->page = NULL;
1974                 }
1975                 if (buffer_info->skb) {
1976                         dev_kfree_skb(buffer_info->skb);
1977                         buffer_info->skb = NULL;
1978                 }
1979         }
1980
1981         /* there also may be some cached data from a chained receive */
1982         if (rx_ring->rx_skb_top) {
1983                 dev_kfree_skb(rx_ring->rx_skb_top);
1984                 rx_ring->rx_skb_top = NULL;
1985         }
1986
1987         size = sizeof(struct e1000_buffer) * rx_ring->count;
1988         memset(rx_ring->buffer_info, 0, size);
1989
1990         /* Zero out the descriptor ring */
1991         memset(rx_ring->desc, 0, rx_ring->size);
1992
1993         rx_ring->next_to_clean = 0;
1994         rx_ring->next_to_use = 0;
1995
1996         writel(0, hw->hw_addr + rx_ring->rdh);
1997         writel(0, hw->hw_addr + rx_ring->rdt);
1998 }
1999
2000 /**
2001  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2002  * @adapter: board private structure
2003  **/
2004
2005 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2006 {
2007         int i;
2008
2009         for (i = 0; i < adapter->num_rx_queues; i++)
2010                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2011 }
2012
2013 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2014  * and memory write and invalidate disabled for certain operations
2015  */
2016 static void e1000_enter_82542_rst(struct e1000_adapter *adapter)
2017 {
2018         struct e1000_hw *hw = &adapter->hw;
2019         struct net_device *netdev = adapter->netdev;
2020         u32 rctl;
2021
2022         e1000_pci_clear_mwi(hw);
2023
2024         rctl = er32(RCTL);
2025         rctl |= E1000_RCTL_RST;
2026         ew32(RCTL, rctl);
2027         E1000_WRITE_FLUSH();
2028         mdelay(5);
2029
2030         if (netif_running(netdev))
2031                 e1000_clean_all_rx_rings(adapter);
2032 }
2033
2034 static void e1000_leave_82542_rst(struct e1000_adapter *adapter)
2035 {
2036         struct e1000_hw *hw = &adapter->hw;
2037         struct net_device *netdev = adapter->netdev;
2038         u32 rctl;
2039
2040         rctl = er32(RCTL);
2041         rctl &= ~E1000_RCTL_RST;
2042         ew32(RCTL, rctl);
2043         E1000_WRITE_FLUSH();
2044         mdelay(5);
2045
2046         if (hw->pci_cmd_word & PCI_COMMAND_INVALIDATE)
2047                 e1000_pci_set_mwi(hw);
2048
2049         if (netif_running(netdev)) {
2050                 /* No need to loop, because 82542 supports only 1 queue */
2051                 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2052                 e1000_configure_rx(adapter);
2053                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2054         }
2055 }
2056
2057 /**
2058  * e1000_set_mac - Change the Ethernet Address of the NIC
2059  * @netdev: network interface device structure
2060  * @p: pointer to an address structure
2061  *
2062  * Returns 0 on success, negative on failure
2063  **/
2064
2065 static int e1000_set_mac(struct net_device *netdev, void *p)
2066 {
2067         struct e1000_adapter *adapter = netdev_priv(netdev);
2068         struct e1000_hw *hw = &adapter->hw;
2069         struct sockaddr *addr = p;
2070
2071         if (!is_valid_ether_addr(addr->sa_data))
2072                 return -EADDRNOTAVAIL;
2073
2074         /* 82542 2.0 needs to be in reset to write receive address registers */
2075
2076         if (hw->mac_type == e1000_82542_rev2_0)
2077                 e1000_enter_82542_rst(adapter);
2078
2079         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2080         memcpy(hw->mac_addr, addr->sa_data, netdev->addr_len);
2081
2082         e1000_rar_set(hw, hw->mac_addr, 0);
2083
2084         if (hw->mac_type == e1000_82542_rev2_0)
2085                 e1000_leave_82542_rst(adapter);
2086
2087         return 0;
2088 }
2089
2090 /**
2091  * e1000_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2092  * @netdev: network interface device structure
2093  *
2094  * The set_rx_mode entry point is called whenever the unicast or multicast
2095  * address lists or the network interface flags are updated. This routine is
2096  * responsible for configuring the hardware for proper unicast, multicast,
2097  * promiscuous mode, and all-multi behavior.
2098  **/
2099
2100 static void e1000_set_rx_mode(struct net_device *netdev)
2101 {
2102         struct e1000_adapter *adapter = netdev_priv(netdev);
2103         struct e1000_hw *hw = &adapter->hw;
2104         struct netdev_hw_addr *ha;
2105         bool use_uc = false;
2106         struct dev_addr_list *mc_ptr;
2107         u32 rctl;
2108         u32 hash_value;
2109         int i, rar_entries = E1000_RAR_ENTRIES;
2110         int mta_reg_count = E1000_NUM_MTA_REGISTERS;
2111         u32 *mcarray = kcalloc(mta_reg_count, sizeof(u32), GFP_ATOMIC);
2112
2113         if (!mcarray) {
2114                 DPRINTK(PROBE, ERR, "memory allocation failed\n");
2115                 return;
2116         }
2117
2118         /* Check for Promiscuous and All Multicast modes */
2119
2120         rctl = er32(RCTL);
2121
2122         if (netdev->flags & IFF_PROMISC) {
2123                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2124                 rctl &= ~E1000_RCTL_VFE;
2125         } else {
2126                 if (netdev->flags & IFF_ALLMULTI)
2127                         rctl |= E1000_RCTL_MPE;
2128                 else
2129                         rctl &= ~E1000_RCTL_MPE;
2130                 /* Enable VLAN filter if there is a VLAN */
2131                 if (adapter->vlgrp)
2132                         rctl |= E1000_RCTL_VFE;
2133         }
2134
2135         if (netdev->uc.count > rar_entries - 1) {
2136                 rctl |= E1000_RCTL_UPE;
2137         } else if (!(netdev->flags & IFF_PROMISC)) {
2138                 rctl &= ~E1000_RCTL_UPE;
2139                 use_uc = true;
2140         }
2141
2142         ew32(RCTL, rctl);
2143
2144         /* 82542 2.0 needs to be in reset to write receive address registers */
2145
2146         if (hw->mac_type == e1000_82542_rev2_0)
2147                 e1000_enter_82542_rst(adapter);
2148
2149         /* load the first 14 addresses into the exact filters 1-14. Unicast
2150          * addresses take precedence to avoid disabling unicast filtering
2151          * when possible.
2152          *
2153          * RAR 0 is used for the station MAC adddress
2154          * if there are not 14 addresses, go ahead and clear the filters
2155          */
2156         i = 1;
2157         if (use_uc)
2158                 list_for_each_entry(ha, &netdev->uc.list, list) {
2159                         if (i == rar_entries)
2160                                 break;
2161                         e1000_rar_set(hw, ha->addr, i++);
2162                 }
2163
2164         WARN_ON(i == rar_entries);
2165
2166         mc_ptr = netdev->mc_list;
2167
2168         for (; i < rar_entries; i++) {
2169                 if (mc_ptr) {
2170                         e1000_rar_set(hw, mc_ptr->da_addr, i);
2171                         mc_ptr = mc_ptr->next;
2172                 } else {
2173                         E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2174                         E1000_WRITE_FLUSH();
2175                         E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2176                         E1000_WRITE_FLUSH();
2177                 }
2178         }
2179
2180         /* load any remaining addresses into the hash table */
2181
2182         for (; mc_ptr; mc_ptr = mc_ptr->next) {
2183                 u32 hash_reg, hash_bit, mta;
2184                 hash_value = e1000_hash_mc_addr(hw, mc_ptr->da_addr);
2185                 hash_reg = (hash_value >> 5) & 0x7F;
2186                 hash_bit = hash_value & 0x1F;
2187                 mta = (1 << hash_bit);
2188                 mcarray[hash_reg] |= mta;
2189         }
2190
2191         /* write the hash table completely, write from bottom to avoid
2192          * both stupid write combining chipsets, and flushing each write */
2193         for (i = mta_reg_count - 1; i >= 0 ; i--) {
2194                 /*
2195                  * If we are on an 82544 has an errata where writing odd
2196                  * offsets overwrites the previous even offset, but writing
2197                  * backwards over the range solves the issue by always
2198                  * writing the odd offset first
2199                  */
2200                 E1000_WRITE_REG_ARRAY(hw, MTA, i, mcarray[i]);
2201         }
2202         E1000_WRITE_FLUSH();
2203
2204         if (hw->mac_type == e1000_82542_rev2_0)
2205                 e1000_leave_82542_rst(adapter);
2206
2207         kfree(mcarray);
2208 }
2209
2210 /* Need to wait a few seconds after link up to get diagnostic information from
2211  * the phy */
2212
2213 static void e1000_update_phy_info(unsigned long data)
2214 {
2215         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2216         struct e1000_hw *hw = &adapter->hw;
2217         e1000_phy_get_info(hw, &adapter->phy_info);
2218 }
2219
2220 /**
2221  * e1000_82547_tx_fifo_stall - Timer Call-back
2222  * @data: pointer to adapter cast into an unsigned long
2223  **/
2224
2225 static void e1000_82547_tx_fifo_stall(unsigned long data)
2226 {
2227         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2228         struct e1000_hw *hw = &adapter->hw;
2229         struct net_device *netdev = adapter->netdev;
2230         u32 tctl;
2231
2232         if (atomic_read(&adapter->tx_fifo_stall)) {
2233                 if ((er32(TDT) == er32(TDH)) &&
2234                    (er32(TDFT) == er32(TDFH)) &&
2235                    (er32(TDFTS) == er32(TDFHS))) {
2236                         tctl = er32(TCTL);
2237                         ew32(TCTL, tctl & ~E1000_TCTL_EN);
2238                         ew32(TDFT, adapter->tx_head_addr);
2239                         ew32(TDFH, adapter->tx_head_addr);
2240                         ew32(TDFTS, adapter->tx_head_addr);
2241                         ew32(TDFHS, adapter->tx_head_addr);
2242                         ew32(TCTL, tctl);
2243                         E1000_WRITE_FLUSH();
2244
2245                         adapter->tx_fifo_head = 0;
2246                         atomic_set(&adapter->tx_fifo_stall, 0);
2247                         netif_wake_queue(netdev);
2248                 } else if (!test_bit(__E1000_DOWN, &adapter->flags)) {
2249                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2250                 }
2251         }
2252 }
2253
2254 static bool e1000_has_link(struct e1000_adapter *adapter)
2255 {
2256         struct e1000_hw *hw = &adapter->hw;
2257         bool link_active = false;
2258         s32 ret_val = 0;
2259
2260         /* get_link_status is set on LSC (link status) interrupt or
2261          * rx sequence error interrupt.  get_link_status will stay
2262          * false until the e1000_check_for_link establishes link
2263          * for copper adapters ONLY
2264          */
2265         switch (hw->media_type) {
2266         case e1000_media_type_copper:
2267                 if (hw->get_link_status) {
2268                         ret_val = e1000_check_for_link(hw);
2269                         link_active = !hw->get_link_status;
2270                 } else {
2271                         link_active = true;
2272                 }
2273                 break;
2274         case e1000_media_type_fiber:
2275                 ret_val = e1000_check_for_link(hw);
2276                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2277                 break;
2278         case e1000_media_type_internal_serdes:
2279                 ret_val = e1000_check_for_link(hw);
2280                 link_active = hw->serdes_has_link;
2281                 break;
2282         default:
2283                 break;
2284         }
2285
2286         return link_active;
2287 }
2288
2289 /**
2290  * e1000_watchdog - Timer Call-back
2291  * @data: pointer to adapter cast into an unsigned long
2292  **/
2293 static void e1000_watchdog(unsigned long data)
2294 {
2295         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2296         struct e1000_hw *hw = &adapter->hw;
2297         struct net_device *netdev = adapter->netdev;
2298         struct e1000_tx_ring *txdr = adapter->tx_ring;
2299         u32 link, tctl;
2300
2301         link = e1000_has_link(adapter);
2302         if ((netif_carrier_ok(netdev)) && link)
2303                 goto link_up;
2304
2305         if (link) {
2306                 if (!netif_carrier_ok(netdev)) {
2307                         u32 ctrl;
2308                         bool txb2b = true;
2309                         /* update snapshot of PHY registers on LSC */
2310                         e1000_get_speed_and_duplex(hw,
2311                                                    &adapter->link_speed,
2312                                                    &adapter->link_duplex);
2313
2314                         ctrl = er32(CTRL);
2315                         printk(KERN_INFO "e1000: %s NIC Link is Up %d Mbps %s, "
2316                                "Flow Control: %s\n",
2317                                netdev->name,
2318                                adapter->link_speed,
2319                                adapter->link_duplex == FULL_DUPLEX ?
2320                                 "Full Duplex" : "Half Duplex",
2321                                 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2322                                 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2323                                 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2324                                 E1000_CTRL_TFCE) ? "TX" : "None" )));
2325
2326                         /* tweak tx_queue_len according to speed/duplex
2327                          * and adjust the timeout factor */
2328                         netdev->tx_queue_len = adapter->tx_queue_len;
2329                         adapter->tx_timeout_factor = 1;
2330                         switch (adapter->link_speed) {
2331                         case SPEED_10:
2332                                 txb2b = false;
2333                                 netdev->tx_queue_len = 10;
2334                                 adapter->tx_timeout_factor = 16;
2335                                 break;
2336                         case SPEED_100:
2337                                 txb2b = false;
2338                                 netdev->tx_queue_len = 100;
2339                                 /* maybe add some timeout factor ? */
2340                                 break;
2341                         }
2342
2343                         /* enable transmits in the hardware */
2344                         tctl = er32(TCTL);
2345                         tctl |= E1000_TCTL_EN;
2346                         ew32(TCTL, tctl);
2347
2348                         netif_carrier_on(netdev);
2349                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2350                                 mod_timer(&adapter->phy_info_timer,
2351                                           round_jiffies(jiffies + 2 * HZ));
2352                         adapter->smartspeed = 0;
2353                 }
2354         } else {
2355                 if (netif_carrier_ok(netdev)) {
2356                         adapter->link_speed = 0;
2357                         adapter->link_duplex = 0;
2358                         printk(KERN_INFO "e1000: %s NIC Link is Down\n",
2359                                netdev->name);
2360                         netif_carrier_off(netdev);
2361
2362                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2363                                 mod_timer(&adapter->phy_info_timer,
2364                                           round_jiffies(jiffies + 2 * HZ));
2365                 }
2366
2367                 e1000_smartspeed(adapter);
2368         }
2369
2370 link_up:
2371         e1000_update_stats(adapter);
2372
2373         hw->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2374         adapter->tpt_old = adapter->stats.tpt;
2375         hw->collision_delta = adapter->stats.colc - adapter->colc_old;
2376         adapter->colc_old = adapter->stats.colc;
2377
2378         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2379         adapter->gorcl_old = adapter->stats.gorcl;
2380         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2381         adapter->gotcl_old = adapter->stats.gotcl;
2382
2383         e1000_update_adaptive(hw);
2384
2385         if (!netif_carrier_ok(netdev)) {
2386                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2387                         /* We've lost link, so the controller stops DMA,
2388                          * but we've got queued Tx work that's never going
2389                          * to get done, so reset controller to flush Tx.
2390                          * (Do the reset outside of interrupt context). */
2391                         adapter->tx_timeout_count++;
2392                         schedule_work(&adapter->reset_task);
2393                         /* return immediately since reset is imminent */
2394                         return;
2395                 }
2396         }
2397
2398         /* Cause software interrupt to ensure rx ring is cleaned */
2399         ew32(ICS, E1000_ICS_RXDMT0);
2400
2401         /* Force detection of hung controller every watchdog period */
2402         adapter->detect_tx_hung = true;
2403
2404         /* Reset the timer */
2405         if (!test_bit(__E1000_DOWN, &adapter->flags))
2406                 mod_timer(&adapter->watchdog_timer,
2407                           round_jiffies(jiffies + 2 * HZ));
2408 }
2409
2410 enum latency_range {
2411         lowest_latency = 0,
2412         low_latency = 1,
2413         bulk_latency = 2,
2414         latency_invalid = 255
2415 };
2416
2417 /**
2418  * e1000_update_itr - update the dynamic ITR value based on statistics
2419  *      Stores a new ITR value based on packets and byte
2420  *      counts during the last interrupt.  The advantage of per interrupt
2421  *      computation is faster updates and more accurate ITR for the current
2422  *      traffic pattern.  Constants in this function were computed
2423  *      based on theoretical maximum wire speed and thresholds were set based
2424  *      on testing data as well as attempting to minimize response time
2425  *      while increasing bulk throughput.
2426  *      this functionality is controlled by the InterruptThrottleRate module
2427  *      parameter (see e1000_param.c)
2428  * @adapter: pointer to adapter
2429  * @itr_setting: current adapter->itr
2430  * @packets: the number of packets during this measurement interval
2431  * @bytes: the number of bytes during this measurement interval
2432  **/
2433 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2434                                      u16 itr_setting, int packets, int bytes)
2435 {
2436         unsigned int retval = itr_setting;
2437         struct e1000_hw *hw = &adapter->hw;
2438
2439         if (unlikely(hw->mac_type < e1000_82540))
2440                 goto update_itr_done;
2441
2442         if (packets == 0)
2443                 goto update_itr_done;
2444
2445         switch (itr_setting) {
2446         case lowest_latency:
2447                 /* jumbo frames get bulk treatment*/
2448                 if (bytes/packets > 8000)
2449                         retval = bulk_latency;
2450                 else if ((packets < 5) && (bytes > 512))
2451                         retval = low_latency;
2452                 break;
2453         case low_latency:  /* 50 usec aka 20000 ints/s */
2454                 if (bytes > 10000) {
2455                         /* jumbo frames need bulk latency setting */
2456                         if (bytes/packets > 8000)
2457                                 retval = bulk_latency;
2458                         else if ((packets < 10) || ((bytes/packets) > 1200))
2459                                 retval = bulk_latency;
2460                         else if ((packets > 35))
2461                                 retval = lowest_latency;
2462                 } else if (bytes/packets > 2000)
2463                         retval = bulk_latency;
2464                 else if (packets <= 2 && bytes < 512)
2465                         retval = lowest_latency;
2466                 break;
2467         case bulk_latency: /* 250 usec aka 4000 ints/s */
2468                 if (bytes > 25000) {
2469                         if (packets > 35)
2470                                 retval = low_latency;
2471                 } else if (bytes < 6000) {
2472                         retval = low_latency;
2473                 }
2474                 break;
2475         }
2476
2477 update_itr_done:
2478         return retval;
2479 }
2480
2481 static void e1000_set_itr(struct e1000_adapter *adapter)
2482 {
2483         struct e1000_hw *hw = &adapter->hw;
2484         u16 current_itr;
2485         u32 new_itr = adapter->itr;
2486
2487         if (unlikely(hw->mac_type < e1000_82540))
2488                 return;
2489
2490         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2491         if (unlikely(adapter->link_speed != SPEED_1000)) {
2492                 current_itr = 0;
2493                 new_itr = 4000;
2494                 goto set_itr_now;
2495         }
2496
2497         adapter->tx_itr = e1000_update_itr(adapter,
2498                                     adapter->tx_itr,
2499                                     adapter->total_tx_packets,
2500                                     adapter->total_tx_bytes);
2501         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2502         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2503                 adapter->tx_itr = low_latency;
2504
2505         adapter->rx_itr = e1000_update_itr(adapter,
2506                                     adapter->rx_itr,
2507                                     adapter->total_rx_packets,
2508                                     adapter->total_rx_bytes);
2509         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2510         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2511                 adapter->rx_itr = low_latency;
2512
2513         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2514
2515         switch (current_itr) {
2516         /* counts and packets in update_itr are dependent on these numbers */
2517         case lowest_latency:
2518                 new_itr = 70000;
2519                 break;
2520         case low_latency:
2521                 new_itr = 20000; /* aka hwitr = ~200 */
2522                 break;
2523         case bulk_latency:
2524                 new_itr = 4000;
2525                 break;
2526         default:
2527                 break;
2528         }
2529
2530 set_itr_now:
2531         if (new_itr != adapter->itr) {
2532                 /* this attempts to bias the interrupt rate towards Bulk
2533                  * by adding intermediate steps when interrupt rate is
2534                  * increasing */
2535                 new_itr = new_itr > adapter->itr ?
2536                              min(adapter->itr + (new_itr >> 2), new_itr) :
2537                              new_itr;
2538                 adapter->itr = new_itr;
2539                 ew32(ITR, 1000000000 / (new_itr * 256));
2540         }
2541
2542         return;
2543 }
2544
2545 #define E1000_TX_FLAGS_CSUM             0x00000001
2546 #define E1000_TX_FLAGS_VLAN             0x00000002
2547 #define E1000_TX_FLAGS_TSO              0x00000004
2548 #define E1000_TX_FLAGS_IPV4             0x00000008
2549 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2550 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2551
2552 static int e1000_tso(struct e1000_adapter *adapter,
2553                      struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2554 {
2555         struct e1000_context_desc *context_desc;
2556         struct e1000_buffer *buffer_info;
2557         unsigned int i;
2558         u32 cmd_length = 0;
2559         u16 ipcse = 0, tucse, mss;
2560         u8 ipcss, ipcso, tucss, tucso, hdr_len;
2561         int err;
2562
2563         if (skb_is_gso(skb)) {
2564                 if (skb_header_cloned(skb)) {
2565                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2566                         if (err)
2567                                 return err;
2568                 }
2569
2570                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2571                 mss = skb_shinfo(skb)->gso_size;
2572                 if (skb->protocol == htons(ETH_P_IP)) {
2573                         struct iphdr *iph = ip_hdr(skb);
2574                         iph->tot_len = 0;
2575                         iph->check = 0;
2576                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2577                                                                  iph->daddr, 0,
2578                                                                  IPPROTO_TCP,
2579                                                                  0);
2580                         cmd_length = E1000_TXD_CMD_IP;
2581                         ipcse = skb_transport_offset(skb) - 1;
2582                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2583                         ipv6_hdr(skb)->payload_len = 0;
2584                         tcp_hdr(skb)->check =
2585                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2586                                                  &ipv6_hdr(skb)->daddr,
2587                                                  0, IPPROTO_TCP, 0);
2588                         ipcse = 0;
2589                 }
2590                 ipcss = skb_network_offset(skb);
2591                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2592                 tucss = skb_transport_offset(skb);
2593                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2594                 tucse = 0;
2595
2596                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2597                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2598
2599                 i = tx_ring->next_to_use;
2600                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2601                 buffer_info = &tx_ring->buffer_info[i];
2602
2603                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2604                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2605                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2606                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2607                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2608                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2609                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2610                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2611                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2612
2613                 buffer_info->time_stamp = jiffies;
2614                 buffer_info->next_to_watch = i;
2615
2616                 if (++i == tx_ring->count) i = 0;
2617                 tx_ring->next_to_use = i;
2618
2619                 return true;
2620         }
2621         return false;
2622 }
2623
2624 static bool e1000_tx_csum(struct e1000_adapter *adapter,
2625                           struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2626 {
2627         struct e1000_context_desc *context_desc;
2628         struct e1000_buffer *buffer_info;
2629         unsigned int i;
2630         u8 css;
2631         u32 cmd_len = E1000_TXD_CMD_DEXT;
2632
2633         if (skb->ip_summed != CHECKSUM_PARTIAL)
2634                 return false;
2635
2636         switch (skb->protocol) {
2637         case cpu_to_be16(ETH_P_IP):
2638                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2639                         cmd_len |= E1000_TXD_CMD_TCP;
2640                 break;
2641         case cpu_to_be16(ETH_P_IPV6):
2642                 /* XXX not handling all IPV6 headers */
2643                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2644                         cmd_len |= E1000_TXD_CMD_TCP;
2645                 break;
2646         default:
2647                 if (unlikely(net_ratelimit()))
2648                         DPRINTK(DRV, WARNING,
2649                                 "checksum_partial proto=%x!\n", skb->protocol);
2650                 break;
2651         }
2652
2653         css = skb_transport_offset(skb);
2654
2655         i = tx_ring->next_to_use;
2656         buffer_info = &tx_ring->buffer_info[i];
2657         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2658
2659         context_desc->lower_setup.ip_config = 0;
2660         context_desc->upper_setup.tcp_fields.tucss = css;
2661         context_desc->upper_setup.tcp_fields.tucso =
2662                 css + skb->csum_offset;
2663         context_desc->upper_setup.tcp_fields.tucse = 0;
2664         context_desc->tcp_seg_setup.data = 0;
2665         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
2666
2667         buffer_info->time_stamp = jiffies;
2668         buffer_info->next_to_watch = i;
2669
2670         if (unlikely(++i == tx_ring->count)) i = 0;
2671         tx_ring->next_to_use = i;
2672
2673         return true;
2674 }
2675
2676 #define E1000_MAX_TXD_PWR       12
2677 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2678
2679 static int e1000_tx_map(struct e1000_adapter *adapter,
2680                         struct e1000_tx_ring *tx_ring,
2681                         struct sk_buff *skb, unsigned int first,
2682                         unsigned int max_per_txd, unsigned int nr_frags,
2683                         unsigned int mss)
2684 {
2685         struct e1000_hw *hw = &adapter->hw;
2686         struct e1000_buffer *buffer_info;
2687         unsigned int len = skb_headlen(skb);
2688         unsigned int offset, size, count = 0, i;
2689         unsigned int f;
2690         dma_addr_t *map;
2691
2692         i = tx_ring->next_to_use;
2693
2694         if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
2695                 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
2696                 return 0;
2697         }
2698
2699         map = skb_shinfo(skb)->dma_maps;
2700         offset = 0;
2701
2702         while (len) {
2703                 buffer_info = &tx_ring->buffer_info[i];
2704                 size = min(len, max_per_txd);
2705                 /* Workaround for Controller erratum --
2706                  * descriptor for non-tso packet in a linear SKB that follows a
2707                  * tso gets written back prematurely before the data is fully
2708                  * DMA'd to the controller */
2709                 if (!skb->data_len && tx_ring->last_tx_tso &&
2710                     !skb_is_gso(skb)) {
2711                         tx_ring->last_tx_tso = 0;
2712                         size -= 4;
2713                 }
2714
2715                 /* Workaround for premature desc write-backs
2716                  * in TSO mode.  Append 4-byte sentinel desc */
2717                 if (unlikely(mss && !nr_frags && size == len && size > 8))
2718                         size -= 4;
2719                 /* work-around for errata 10 and it applies
2720                  * to all controllers in PCI-X mode
2721                  * The fix is to make sure that the first descriptor of a
2722                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2723                  */
2724                 if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
2725                                 (size > 2015) && count == 0))
2726                         size = 2015;
2727
2728                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
2729                  * terminating buffers within evenly-aligned dwords. */
2730                 if (unlikely(adapter->pcix_82544 &&
2731                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2732                    size > 4))
2733                         size -= 4;
2734
2735                 buffer_info->length = size;
2736                 buffer_info->dma = skb_shinfo(skb)->dma_head + offset;
2737                 buffer_info->time_stamp = jiffies;
2738                 buffer_info->next_to_watch = i;
2739
2740                 len -= size;
2741                 offset += size;
2742                 count++;
2743                 if (len) {
2744                         i++;
2745                         if (unlikely(i == tx_ring->count))
2746                                 i = 0;
2747                 }
2748         }
2749
2750         for (f = 0; f < nr_frags; f++) {
2751                 struct skb_frag_struct *frag;
2752
2753                 frag = &skb_shinfo(skb)->frags[f];
2754                 len = frag->size;
2755                 offset = 0;
2756
2757                 while (len) {
2758                         i++;
2759                         if (unlikely(i == tx_ring->count))
2760                                 i = 0;
2761
2762                         buffer_info = &tx_ring->buffer_info[i];
2763                         size = min(len, max_per_txd);
2764                         /* Workaround for premature desc write-backs
2765                          * in TSO mode.  Append 4-byte sentinel desc */
2766                         if (unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
2767                                 size -= 4;
2768                         /* Workaround for potential 82544 hang in PCI-X.
2769                          * Avoid terminating buffers within evenly-aligned
2770                          * dwords. */
2771                         if (unlikely(adapter->pcix_82544 &&
2772                            !((unsigned long)(frag->page+offset+size-1) & 4) &&
2773                            size > 4))
2774                                 size -= 4;
2775
2776                         buffer_info->length = size;
2777                         buffer_info->dma = map[f] + offset;
2778                         buffer_info->time_stamp = jiffies;
2779                         buffer_info->next_to_watch = i;
2780
2781                         len -= size;
2782                         offset += size;
2783                         count++;
2784                 }
2785         }
2786
2787         tx_ring->buffer_info[i].skb = skb;
2788         tx_ring->buffer_info[first].next_to_watch = i;
2789
2790         return count;
2791 }
2792
2793 static void e1000_tx_queue(struct e1000_adapter *adapter,
2794                            struct e1000_tx_ring *tx_ring, int tx_flags,
2795                            int count)
2796 {
2797         struct e1000_hw *hw = &adapter->hw;
2798         struct e1000_tx_desc *tx_desc = NULL;
2799         struct e1000_buffer *buffer_info;
2800         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2801         unsigned int i;
2802
2803         if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
2804                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2805                              E1000_TXD_CMD_TSE;
2806                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2807
2808                 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
2809                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2810         }
2811
2812         if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
2813                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
2814                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2815         }
2816
2817         if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
2818                 txd_lower |= E1000_TXD_CMD_VLE;
2819                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
2820         }
2821
2822         i = tx_ring->next_to_use;
2823
2824         while (count--) {
2825                 buffer_info = &tx_ring->buffer_info[i];
2826                 tx_desc = E1000_TX_DESC(*tx_ring, i);
2827                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
2828                 tx_desc->lower.data =
2829                         cpu_to_le32(txd_lower | buffer_info->length);
2830                 tx_desc->upper.data = cpu_to_le32(txd_upper);
2831                 if (unlikely(++i == tx_ring->count)) i = 0;
2832         }
2833
2834         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
2835
2836         /* Force memory writes to complete before letting h/w
2837          * know there are new descriptors to fetch.  (Only
2838          * applicable for weak-ordered memory model archs,
2839          * such as IA-64). */
2840         wmb();
2841
2842         tx_ring->next_to_use = i;
2843         writel(i, hw->hw_addr + tx_ring->tdt);
2844         /* we need this if more than one processor can write to our tail
2845          * at a time, it syncronizes IO on IA64/Altix systems */
2846         mmiowb();
2847 }
2848
2849 /**
2850  * 82547 workaround to avoid controller hang in half-duplex environment.
2851  * The workaround is to avoid queuing a large packet that would span
2852  * the internal Tx FIFO ring boundary by notifying the stack to resend
2853  * the packet at a later time.  This gives the Tx FIFO an opportunity to
2854  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
2855  * to the beginning of the Tx FIFO.
2856  **/
2857
2858 #define E1000_FIFO_HDR                  0x10
2859 #define E1000_82547_PAD_LEN             0x3E0
2860
2861 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
2862                                        struct sk_buff *skb)
2863 {
2864         u32 fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
2865         u32 skb_fifo_len = skb->len + E1000_FIFO_HDR;
2866
2867         skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
2868
2869         if (adapter->link_duplex != HALF_DUPLEX)
2870                 goto no_fifo_stall_required;
2871
2872         if (atomic_read(&adapter->tx_fifo_stall))
2873                 return 1;
2874
2875         if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
2876                 atomic_set(&adapter->tx_fifo_stall, 1);
2877                 return 1;
2878         }
2879
2880 no_fifo_stall_required:
2881         adapter->tx_fifo_head += skb_fifo_len;
2882         if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
2883                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
2884         return 0;
2885 }
2886
2887 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
2888 {
2889         struct e1000_adapter *adapter = netdev_priv(netdev);
2890         struct e1000_tx_ring *tx_ring = adapter->tx_ring;
2891
2892         netif_stop_queue(netdev);
2893         /* Herbert's original patch had:
2894          *  smp_mb__after_netif_stop_queue();
2895          * but since that doesn't exist yet, just open code it. */
2896         smp_mb();
2897
2898         /* We need to check again in a case another CPU has just
2899          * made room available. */
2900         if (likely(E1000_DESC_UNUSED(tx_ring) < size))
2901                 return -EBUSY;
2902
2903         /* A reprieve! */
2904         netif_start_queue(netdev);
2905         ++adapter->restart_queue;
2906         return 0;
2907 }
2908
2909 static int e1000_maybe_stop_tx(struct net_device *netdev,
2910                                struct e1000_tx_ring *tx_ring, int size)
2911 {
2912         if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
2913                 return 0;
2914         return __e1000_maybe_stop_tx(netdev, size);
2915 }
2916
2917 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
2918 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
2919                                     struct net_device *netdev)
2920 {
2921         struct e1000_adapter *adapter = netdev_priv(netdev);
2922         struct e1000_hw *hw = &adapter->hw;
2923         struct e1000_tx_ring *tx_ring;
2924         unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
2925         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
2926         unsigned int tx_flags = 0;
2927         unsigned int len = skb->len - skb->data_len;
2928         unsigned int nr_frags;
2929         unsigned int mss;
2930         int count = 0;
2931         int tso;
2932         unsigned int f;
2933
2934         /* This goes back to the question of how to logically map a tx queue
2935          * to a flow.  Right now, performance is impacted slightly negatively
2936          * if using multiple tx queues.  If the stack breaks away from a
2937          * single qdisc implementation, we can look at this again. */
2938         tx_ring = adapter->tx_ring;
2939
2940         if (unlikely(skb->len <= 0)) {
2941                 dev_kfree_skb_any(skb);
2942                 return NETDEV_TX_OK;
2943         }
2944
2945         mss = skb_shinfo(skb)->gso_size;
2946         /* The controller does a simple calculation to
2947          * make sure there is enough room in the FIFO before
2948          * initiating the DMA for each buffer.  The calc is:
2949          * 4 = ceil(buffer len/mss).  To make sure we don't
2950          * overrun the FIFO, adjust the max buffer len if mss
2951          * drops. */
2952         if (mss) {
2953                 u8 hdr_len;
2954                 max_per_txd = min(mss << 2, max_per_txd);
2955                 max_txd_pwr = fls(max_per_txd) - 1;
2956
2957                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2958                 if (skb->data_len && hdr_len == len) {
2959                         switch (hw->mac_type) {
2960                                 unsigned int pull_size;
2961                         case e1000_82544:
2962                                 /* Make sure we have room to chop off 4 bytes,
2963                                  * and that the end alignment will work out to
2964                                  * this hardware's requirements
2965                                  * NOTE: this is a TSO only workaround
2966                                  * if end byte alignment not correct move us
2967                                  * into the next dword */
2968                                 if ((unsigned long)(skb_tail_pointer(skb) - 1) & 4)
2969                                         break;
2970                                 /* fall through */
2971                                 pull_size = min((unsigned int)4, skb->data_len);
2972                                 if (!__pskb_pull_tail(skb, pull_size)) {
2973                                         DPRINTK(DRV, ERR,
2974                                                 "__pskb_pull_tail failed.\n");
2975                                         dev_kfree_skb_any(skb);
2976                                         return NETDEV_TX_OK;
2977                                 }
2978                                 len = skb->len - skb->data_len;
2979                                 break;
2980                         default:
2981                                 /* do nothing */
2982                                 break;
2983                         }
2984                 }
2985         }
2986
2987         /* reserve a descriptor for the offload context */
2988         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
2989                 count++;
2990         count++;
2991
2992         /* Controller Erratum workaround */
2993         if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
2994                 count++;
2995
2996         count += TXD_USE_COUNT(len, max_txd_pwr);
2997
2998         if (adapter->pcix_82544)
2999                 count++;
3000
3001         /* work-around for errata 10 and it applies to all controllers
3002          * in PCI-X mode, so add one more descriptor to the count
3003          */
3004         if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
3005                         (len > 2015)))
3006                 count++;
3007
3008         nr_frags = skb_shinfo(skb)->nr_frags;
3009         for (f = 0; f < nr_frags; f++)
3010                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3011                                        max_txd_pwr);
3012         if (adapter->pcix_82544)
3013                 count += nr_frags;
3014
3015         /* need: count + 2 desc gap to keep tail from touching
3016          * head, otherwise try next time */
3017         if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2)))
3018                 return NETDEV_TX_BUSY;
3019
3020         if (unlikely(hw->mac_type == e1000_82547)) {
3021                 if (unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
3022                         netif_stop_queue(netdev);
3023                         if (!test_bit(__E1000_DOWN, &adapter->flags))
3024                                 mod_timer(&adapter->tx_fifo_stall_timer,
3025                                           jiffies + 1);
3026                         return NETDEV_TX_BUSY;
3027                 }
3028         }
3029
3030         if (unlikely(adapter->vlgrp && vlan_tx_tag_present(skb))) {
3031                 tx_flags |= E1000_TX_FLAGS_VLAN;
3032                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3033         }
3034
3035         first = tx_ring->next_to_use;
3036
3037         tso = e1000_tso(adapter, tx_ring, skb);
3038         if (tso < 0) {
3039                 dev_kfree_skb_any(skb);
3040                 return NETDEV_TX_OK;
3041         }
3042
3043         if (likely(tso)) {
3044                 tx_ring->last_tx_tso = 1;
3045                 tx_flags |= E1000_TX_FLAGS_TSO;
3046         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3047                 tx_flags |= E1000_TX_FLAGS_CSUM;
3048
3049         if (likely(skb->protocol == htons(ETH_P_IP)))
3050                 tx_flags |= E1000_TX_FLAGS_IPV4;
3051
3052         count = e1000_tx_map(adapter, tx_ring, skb, first, max_per_txd,
3053                              nr_frags, mss);
3054
3055         if (count) {
3056                 e1000_tx_queue(adapter, tx_ring, tx_flags, count);
3057                 /* Make sure there is space in the ring for the next send. */
3058                 e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3059
3060         } else {
3061                 dev_kfree_skb_any(skb);
3062                 tx_ring->buffer_info[first].time_stamp = 0;
3063                 tx_ring->next_to_use = first;
3064         }
3065
3066         return NETDEV_TX_OK;
3067 }
3068
3069 /**
3070  * e1000_tx_timeout - Respond to a Tx Hang
3071  * @netdev: network interface device structure
3072  **/
3073
3074 static void e1000_tx_timeout(struct net_device *netdev)
3075 {
3076         struct e1000_adapter *adapter = netdev_priv(netdev);
3077
3078         /* Do the reset outside of interrupt context */
3079         adapter->tx_timeout_count++;
3080         schedule_work(&adapter->reset_task);
3081 }
3082
3083 static void e1000_reset_task(struct work_struct *work)
3084 {
3085         struct e1000_adapter *adapter =
3086                 container_of(work, struct e1000_adapter, reset_task);
3087
3088         e1000_reinit_locked(adapter);
3089 }
3090
3091 /**
3092  * e1000_get_stats - Get System Network Statistics
3093  * @netdev: network interface device structure
3094  *
3095  * Returns the address of the device statistics structure.
3096  * The statistics are actually updated from the timer callback.
3097  **/
3098
3099 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3100 {
3101         struct e1000_adapter *adapter = netdev_priv(netdev);
3102
3103         /* only return the current stats */
3104         return &adapter->net_stats;
3105 }
3106
3107 /**
3108  * e1000_change_mtu - Change the Maximum Transfer Unit
3109  * @netdev: network interface device structure
3110  * @new_mtu: new value for maximum frame size
3111  *
3112  * Returns 0 on success, negative on failure
3113  **/
3114
3115 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3116 {
3117         struct e1000_adapter *adapter = netdev_priv(netdev);
3118         struct e1000_hw *hw = &adapter->hw;
3119         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3120
3121         if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3122             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3123                 DPRINTK(PROBE, ERR, "Invalid MTU setting\n");
3124                 return -EINVAL;
3125         }
3126
3127         /* Adapter-specific max frame size limits. */
3128         switch (hw->mac_type) {
3129         case e1000_undefined ... e1000_82542_rev2_1:
3130                 if (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)) {
3131                         DPRINTK(PROBE, ERR, "Jumbo Frames not supported.\n");
3132                         return -EINVAL;
3133                 }
3134                 break;
3135         default:
3136                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3137                 break;
3138         }
3139
3140         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3141          * means we reserve 2 more, this pushes us to allocate from the next
3142          * larger slab size.
3143          * i.e. RXBUFFER_2048 --> size-4096 slab
3144          *  however with the new *_jumbo_rx* routines, jumbo receives will use
3145          *  fragmented skbs */
3146
3147         if (max_frame <= E1000_RXBUFFER_256)
3148                 adapter->rx_buffer_len = E1000_RXBUFFER_256;
3149         else if (max_frame <= E1000_RXBUFFER_512)
3150                 adapter->rx_buffer_len = E1000_RXBUFFER_512;
3151         else if (max_frame <= E1000_RXBUFFER_1024)
3152                 adapter->rx_buffer_len = E1000_RXBUFFER_1024;
3153         else if (max_frame <= E1000_RXBUFFER_2048)
3154                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3155         else
3156 #if (PAGE_SIZE >= E1000_RXBUFFER_16384)
3157                 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3158 #elif (PAGE_SIZE >= E1000_RXBUFFER_4096)
3159                 adapter->rx_buffer_len = PAGE_SIZE;
3160 #endif
3161
3162         /* adjust allocation if LPE protects us, and we aren't using SBP */
3163         if (!hw->tbi_compatibility_on &&
3164             ((max_frame == (ETH_FRAME_LEN + ETH_FCS_LEN)) ||
3165              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3166                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3167
3168         netdev->mtu = new_mtu;
3169         hw->max_frame_size = max_frame;
3170
3171         if (netif_running(netdev))
3172                 e1000_reinit_locked(adapter);
3173
3174         return 0;
3175 }
3176
3177 /**
3178  * e1000_update_stats - Update the board statistics counters
3179  * @adapter: board private structure
3180  **/
3181
3182 void e1000_update_stats(struct e1000_adapter *adapter)
3183 {
3184         struct e1000_hw *hw = &adapter->hw;
3185         struct pci_dev *pdev = adapter->pdev;
3186         unsigned long flags;
3187         u16 phy_tmp;
3188
3189 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3190
3191         /*
3192          * Prevent stats update while adapter is being reset, or if the pci
3193          * connection is down.
3194          */
3195         if (adapter->link_speed == 0)
3196                 return;
3197         if (pci_channel_offline(pdev))
3198                 return;
3199
3200         spin_lock_irqsave(&adapter->stats_lock, flags);
3201
3202         /* these counters are modified from e1000_tbi_adjust_stats,
3203          * called from the interrupt context, so they must only
3204          * be written while holding adapter->stats_lock
3205          */
3206
3207         adapter->stats.crcerrs += er32(CRCERRS);
3208         adapter->stats.gprc += er32(GPRC);
3209         adapter->stats.gorcl += er32(GORCL);
3210         adapter->stats.gorch += er32(GORCH);
3211         adapter->stats.bprc += er32(BPRC);
3212         adapter->stats.mprc += er32(MPRC);
3213         adapter->stats.roc += er32(ROC);
3214
3215         adapter->stats.prc64 += er32(PRC64);
3216         adapter->stats.prc127 += er32(PRC127);
3217         adapter->stats.prc255 += er32(PRC255);
3218         adapter->stats.prc511 += er32(PRC511);
3219         adapter->stats.prc1023 += er32(PRC1023);
3220         adapter->stats.prc1522 += er32(PRC1522);
3221
3222         adapter->stats.symerrs += er32(SYMERRS);
3223         adapter->stats.mpc += er32(MPC);
3224         adapter->stats.scc += er32(SCC);
3225         adapter->stats.ecol += er32(ECOL);
3226         adapter->stats.mcc += er32(MCC);
3227         adapter->stats.latecol += er32(LATECOL);
3228         adapter->stats.dc += er32(DC);
3229         adapter->stats.sec += er32(SEC);
3230         adapter->stats.rlec += er32(RLEC);
3231         adapter->stats.xonrxc += er32(XONRXC);
3232         adapter->stats.xontxc += er32(XONTXC);
3233         adapter->stats.xoffrxc += er32(XOFFRXC);
3234         adapter->stats.xofftxc += er32(XOFFTXC);
3235         adapter->stats.fcruc += er32(FCRUC);
3236         adapter->stats.gptc += er32(GPTC);
3237         adapter->stats.gotcl += er32(GOTCL);
3238         adapter->stats.gotch += er32(GOTCH);
3239         adapter->stats.rnbc += er32(RNBC);
3240         adapter->stats.ruc += er32(RUC);
3241         adapter->stats.rfc += er32(RFC);
3242         adapter->stats.rjc += er32(RJC);
3243         adapter->stats.torl += er32(TORL);
3244         adapter->stats.torh += er32(TORH);
3245         adapter->stats.totl += er32(TOTL);
3246         adapter->stats.toth += er32(TOTH);
3247         adapter->stats.tpr += er32(TPR);
3248
3249         adapter->stats.ptc64 += er32(PTC64);
3250         adapter->stats.ptc127 += er32(PTC127);
3251         adapter->stats.ptc255 += er32(PTC255);
3252         adapter->stats.ptc511 += er32(PTC511);
3253         adapter->stats.ptc1023 += er32(PTC1023);
3254         adapter->stats.ptc1522 += er32(PTC1522);
3255
3256         adapter->stats.mptc += er32(MPTC);
3257         adapter->stats.bptc += er32(BPTC);
3258
3259         /* used for adaptive IFS */
3260
3261         hw->tx_packet_delta = er32(TPT);
3262         adapter->stats.tpt += hw->tx_packet_delta;
3263         hw->collision_delta = er32(COLC);
3264         adapter->stats.colc += hw->collision_delta;
3265
3266         if (hw->mac_type >= e1000_82543) {
3267                 adapter->stats.algnerrc += er32(ALGNERRC);
3268                 adapter->stats.rxerrc += er32(RXERRC);
3269                 adapter->stats.tncrs += er32(TNCRS);
3270                 adapter->stats.cexterr += er32(CEXTERR);
3271                 adapter->stats.tsctc += er32(TSCTC);
3272                 adapter->stats.tsctfc += er32(TSCTFC);
3273         }
3274
3275         /* Fill out the OS statistics structure */
3276         adapter->net_stats.multicast = adapter->stats.mprc;
3277         adapter->net_stats.collisions = adapter->stats.colc;
3278
3279         /* Rx Errors */
3280
3281         /* RLEC on some newer hardware can be incorrect so build
3282         * our own version based on RUC and ROC */
3283         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3284                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3285                 adapter->stats.ruc + adapter->stats.roc +
3286                 adapter->stats.cexterr;
3287         adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3288         adapter->net_stats.rx_length_errors = adapter->stats.rlerrc;
3289         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3290         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3291         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3292
3293         /* Tx Errors */
3294         adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3295         adapter->net_stats.tx_errors = adapter->stats.txerrc;
3296         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3297         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3298         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3299         if (hw->bad_tx_carr_stats_fd &&
3300             adapter->link_duplex == FULL_DUPLEX) {
3301                 adapter->net_stats.tx_carrier_errors = 0;
3302                 adapter->stats.tncrs = 0;
3303         }
3304
3305         /* Tx Dropped needs to be maintained elsewhere */
3306
3307         /* Phy Stats */
3308         if (hw->media_type == e1000_media_type_copper) {
3309                 if ((adapter->link_speed == SPEED_1000) &&
3310                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3311                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3312                         adapter->phy_stats.idle_errors += phy_tmp;
3313                 }
3314
3315                 if ((hw->mac_type <= e1000_82546) &&
3316                    (hw->phy_type == e1000_phy_m88) &&
3317                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3318                         adapter->phy_stats.receive_errors += phy_tmp;
3319         }
3320
3321         /* Management Stats */
3322         if (hw->has_smbus) {
3323                 adapter->stats.mgptc += er32(MGTPTC);
3324                 adapter->stats.mgprc += er32(MGTPRC);
3325                 adapter->stats.mgpdc += er32(MGTPDC);
3326         }
3327
3328         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3329 }
3330
3331 /**
3332  * e1000_intr - Interrupt Handler
3333  * @irq: interrupt number
3334  * @data: pointer to a network interface device structure
3335  **/
3336
3337 static irqreturn_t e1000_intr(int irq, void *data)
3338 {
3339         struct net_device *netdev = data;
3340         struct e1000_adapter *adapter = netdev_priv(netdev);
3341         struct e1000_hw *hw = &adapter->hw;
3342         u32 icr = er32(ICR);
3343
3344         if (unlikely((!icr) || test_bit(__E1000_DOWN, &adapter->flags)))
3345                 return IRQ_NONE;  /* Not our interrupt */
3346
3347         if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3348                 hw->get_link_status = 1;
3349                 /* guard against interrupt when we're going down */
3350                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3351                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3352         }
3353
3354         /* disable interrupts, without the synchronize_irq bit */
3355         ew32(IMC, ~0);
3356         E1000_WRITE_FLUSH();
3357
3358         if (likely(napi_schedule_prep(&adapter->napi))) {
3359                 adapter->total_tx_bytes = 0;
3360                 adapter->total_tx_packets = 0;
3361                 adapter->total_rx_bytes = 0;
3362                 adapter->total_rx_packets = 0;
3363                 __napi_schedule(&adapter->napi);
3364         } else {
3365                 /* this really should not happen! if it does it is basically a
3366                  * bug, but not a hard error, so enable ints and continue */
3367                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3368                         e1000_irq_enable(adapter);
3369         }
3370
3371         return IRQ_HANDLED;
3372 }
3373
3374 /**
3375  * e1000_clean - NAPI Rx polling callback
3376  * @adapter: board private structure
3377  **/
3378 static int e1000_clean(struct napi_struct *napi, int budget)
3379 {
3380         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
3381         struct net_device *poll_dev = adapter->netdev;
3382         int tx_cleaned = 0, work_done = 0;
3383
3384         adapter = netdev_priv(poll_dev);
3385
3386         tx_cleaned = e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);
3387
3388         adapter->clean_rx(adapter, &adapter->rx_ring[0],
3389                           &work_done, budget);
3390
3391         if (!tx_cleaned)
3392                 work_done = budget;
3393
3394         /* If budget not fully consumed, exit the polling mode */
3395         if (work_done < budget) {
3396                 if (likely(adapter->itr_setting & 3))
3397                         e1000_set_itr(adapter);
3398                 napi_complete(napi);
3399                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3400                         e1000_irq_enable(adapter);
3401         }
3402
3403         return work_done;
3404 }
3405
3406 /**
3407  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3408  * @adapter: board private structure
3409  **/
3410 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
3411                                struct e1000_tx_ring *tx_ring)
3412 {
3413         struct e1000_hw *hw = &adapter->hw;
3414         struct net_device *netdev = adapter->netdev;
3415         struct e1000_tx_desc *tx_desc, *eop_desc;
3416         struct e1000_buffer *buffer_info;
3417         unsigned int i, eop;
3418         unsigned int count = 0;
3419         unsigned int total_tx_bytes=0, total_tx_packets=0;
3420
3421         i = tx_ring->next_to_clean;
3422         eop = tx_ring->buffer_info[i].next_to_watch;
3423         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3424
3425         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
3426                (count < tx_ring->count)) {
3427                 bool cleaned = false;
3428                 for ( ; !cleaned; count++) {
3429                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3430                         buffer_info = &tx_ring->buffer_info[i];
3431                         cleaned = (i == eop);
3432
3433                         if (cleaned) {
3434                                 struct sk_buff *skb = buffer_info->skb;
3435                                 unsigned int segs, bytecount;
3436                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3437                                 /* multiply data chunks by size of headers */
3438                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3439                                             skb->len;
3440                                 total_tx_packets += segs;
3441                                 total_tx_bytes += bytecount;
3442                         }
3443                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3444                         tx_desc->upper.data = 0;
3445
3446                         if (unlikely(++i == tx_ring->count)) i = 0;
3447                 }
3448
3449                 eop = tx_ring->buffer_info[i].next_to_watch;
3450                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3451         }
3452
3453         tx_ring->next_to_clean = i;
3454
3455 #define TX_WAKE_THRESHOLD 32
3456         if (unlikely(count && netif_carrier_ok(netdev) &&
3457                      E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
3458                 /* Make sure that anybody stopping the queue after this
3459                  * sees the new next_to_clean.
3460                  */
3461                 smp_mb();
3462                 if (netif_queue_stopped(netdev)) {
3463                         netif_wake_queue(netdev);
3464                         ++adapter->restart_queue;
3465                 }
3466         }
3467
3468         if (adapter->detect_tx_hung) {
3469                 /* Detect a transmit hang in hardware, this serializes the
3470                  * check with the clearing of time_stamp and movement of i */
3471                 adapter->detect_tx_hung = false;
3472                 if (tx_ring->buffer_info[i].time_stamp &&
3473                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
3474                                (adapter->tx_timeout_factor * HZ))
3475                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
3476
3477                         /* detected Tx unit hang */
3478                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
3479                                         "  Tx Queue             <%lu>\n"
3480                                         "  TDH                  <%x>\n"
3481                                         "  TDT                  <%x>\n"
3482                                         "  next_to_use          <%x>\n"
3483                                         "  next_to_clean        <%x>\n"
3484                                         "buffer_info[next_to_clean]\n"
3485                                         "  time_stamp           <%lx>\n"
3486                                         "  next_to_watch        <%x>\n"
3487                                         "  jiffies              <%lx>\n"
3488                                         "  next_to_watch.status <%x>\n",
3489                                 (unsigned long)((tx_ring - adapter->tx_ring) /
3490                                         sizeof(struct e1000_tx_ring)),
3491                                 readl(hw->hw_addr + tx_ring->tdh),
3492                                 readl(hw->hw_addr + tx_ring->tdt),
3493                                 tx_ring->next_to_use,
3494                                 tx_ring->next_to_clean,
3495                                 tx_ring->buffer_info[i].time_stamp,
3496                                 eop,
3497                                 jiffies,
3498                                 eop_desc->upper.fields.status);
3499                         netif_stop_queue(netdev);
3500                 }
3501         }
3502         adapter->total_tx_bytes += total_tx_bytes;
3503         adapter->total_tx_packets += total_tx_packets;
3504         adapter->net_stats.tx_bytes += total_tx_bytes;
3505         adapter->net_stats.tx_packets += total_tx_packets;
3506         return (count < tx_ring->count);
3507 }
3508
3509 /**
3510  * e1000_rx_checksum - Receive Checksum Offload for 82543
3511  * @adapter:     board private structure
3512  * @status_err:  receive descriptor status and error fields
3513  * @csum:        receive descriptor csum field
3514  * @sk_buff:     socket buffer with received data
3515  **/
3516
3517 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
3518                               u32 csum, struct sk_buff *skb)
3519 {
3520         struct e1000_hw *hw = &adapter->hw;
3521         u16 status = (u16)status_err;
3522         u8 errors = (u8)(status_err >> 24);
3523         skb->ip_summed = CHECKSUM_NONE;
3524
3525         /* 82543 or newer only */
3526         if (unlikely(hw->mac_type < e1000_82543)) return;
3527         /* Ignore Checksum bit is set */
3528         if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
3529         /* TCP/UDP checksum error bit is set */
3530         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
3531                 /* let the stack verify checksum errors */
3532                 adapter->hw_csum_err++;
3533                 return;
3534         }
3535         /* TCP/UDP Checksum has not been calculated */
3536         if (!(status & E1000_RXD_STAT_TCPCS))
3537                 return;
3538
3539         /* It must be a TCP or UDP packet with a valid checksum */
3540         if (likely(status & E1000_RXD_STAT_TCPCS)) {
3541                 /* TCP checksum is good */
3542                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3543         }
3544         adapter->hw_csum_good++;
3545 }
3546
3547 /**
3548  * e1000_consume_page - helper function
3549  **/
3550 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
3551                                u16 length)
3552 {
3553         bi->page = NULL;
3554         skb->len += length;
3555         skb->data_len += length;
3556         skb->truesize += length;
3557 }
3558
3559 /**
3560  * e1000_receive_skb - helper function to handle rx indications
3561  * @adapter: board private structure
3562  * @status: descriptor status field as written by hardware
3563  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3564  * @skb: pointer to sk_buff to be indicated to stack
3565  */
3566 static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
3567                               __le16 vlan, struct sk_buff *skb)
3568 {
3569         if (unlikely(adapter->vlgrp && (status & E1000_RXD_STAT_VP))) {
3570                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3571                                          le16_to_cpu(vlan) &
3572                                          E1000_RXD_SPC_VLAN_MASK);
3573         } else {
3574                 netif_receive_skb(skb);
3575         }
3576 }
3577
3578 /**
3579  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
3580  * @adapter: board private structure
3581  * @rx_ring: ring to clean
3582  * @work_done: amount of napi work completed this call
3583  * @work_to_do: max amount of work allowed for this call to do
3584  *
3585  * the return value indicates whether actual cleaning was done, there
3586  * is no guarantee that everything was cleaned
3587  */
3588 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
3589                                      struct e1000_rx_ring *rx_ring,
3590                                      int *work_done, int work_to_do)
3591 {
3592         struct e1000_hw *hw = &adapter->hw;
3593         struct net_device *netdev = adapter->netdev;
3594         struct pci_dev *pdev = adapter->pdev;
3595         struct e1000_rx_desc *rx_desc, *next_rxd;
3596         struct e1000_buffer *buffer_info, *next_buffer;
3597         unsigned long irq_flags;
3598         u32 length;
3599         unsigned int i;
3600         int cleaned_count = 0;
3601         bool cleaned = false;
3602         unsigned int total_rx_bytes=0, total_rx_packets=0;
3603
3604         i = rx_ring->next_to_clean;
3605         rx_desc = E1000_RX_DESC(*rx_ring, i);
3606         buffer_info = &rx_ring->buffer_info[i];
3607
3608         while (rx_desc->status & E1000_RXD_STAT_DD) {
3609                 struct sk_buff *skb;
3610                 u8 status;
3611
3612                 if (*work_done >= work_to_do)
3613                         break;
3614                 (*work_done)++;
3615
3616                 status = rx_desc->status;
3617                 skb = buffer_info->skb;
3618                 buffer_info->skb = NULL;
3619
3620                 if (++i == rx_ring->count) i = 0;
3621                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3622                 prefetch(next_rxd);
3623
3624                 next_buffer = &rx_ring->buffer_info[i];
3625
3626                 cleaned = true;
3627                 cleaned_count++;
3628                 pci_unmap_page(pdev, buffer_info->dma, buffer_info->length,
3629                                PCI_DMA_FROMDEVICE);
3630                 buffer_info->dma = 0;
3631
3632                 length = le16_to_cpu(rx_desc->length);
3633
3634                 /* errors is only valid for DD + EOP descriptors */
3635                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
3636                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
3637                         u8 last_byte = *(skb->data + length - 1);
3638                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3639                                        last_byte)) {
3640                                 spin_lock_irqsave(&adapter->stats_lock,
3641                                                   irq_flags);
3642                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3643                                                        length, skb->data);
3644                                 spin_unlock_irqrestore(&adapter->stats_lock,
3645                                                        irq_flags);
3646                                 length--;
3647                         } else {
3648                                 /* recycle both page and skb */
3649                                 buffer_info->skb = skb;
3650                                 /* an error means any chain goes out the window
3651                                  * too */
3652                                 if (rx_ring->rx_skb_top)
3653                                         dev_kfree_skb(rx_ring->rx_skb_top);
3654                                 rx_ring->rx_skb_top = NULL;
3655                                 goto next_desc;
3656                         }
3657                 }
3658
3659 #define rxtop rx_ring->rx_skb_top
3660                 if (!(status & E1000_RXD_STAT_EOP)) {
3661                         /* this descriptor is only the beginning (or middle) */
3662                         if (!rxtop) {
3663                                 /* this is the beginning of a chain */
3664                                 rxtop = skb;
3665                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
3666                                                    0, length);
3667                         } else {
3668                                 /* this is the middle of a chain */
3669                                 skb_fill_page_desc(rxtop,
3670                                     skb_shinfo(rxtop)->nr_frags,
3671                                     buffer_info->page, 0, length);
3672                                 /* re-use the skb, only consumed the page */
3673                                 buffer_info->skb = skb;
3674                         }
3675                         e1000_consume_page(buffer_info, rxtop, length);
3676                         goto next_desc;
3677                 } else {
3678                         if (rxtop) {
3679                                 /* end of the chain */
3680                                 skb_fill_page_desc(rxtop,
3681                                     skb_shinfo(rxtop)->nr_frags,
3682                                     buffer_info->page, 0, length);
3683                                 /* re-use the current skb, we only consumed the
3684                                  * page */
3685                                 buffer_info->skb = skb;
3686                                 skb = rxtop;
3687                                 rxtop = NULL;
3688                                 e1000_consume_page(buffer_info, skb, length);
3689                         } else {
3690                                 /* no chain, got EOP, this buf is the packet
3691                                  * copybreak to save the put_page/alloc_page */
3692                                 if (length <= copybreak &&
3693                                     skb_tailroom(skb) >= length) {
3694                                         u8 *vaddr;
3695                                         vaddr = kmap_atomic(buffer_info->page,
3696                                                             KM_SKB_DATA_SOFTIRQ);
3697                                         memcpy(skb_tail_pointer(skb), vaddr, length);
3698                                         kunmap_atomic(vaddr,
3699                                                       KM_SKB_DATA_SOFTIRQ);
3700                                         /* re-use the page, so don't erase
3701                                          * buffer_info->page */
3702                                         skb_put(skb, length);
3703                                 } else {
3704                                         skb_fill_page_desc(skb, 0,
3705                                                            buffer_info->page, 0,
3706                                                            length);
3707                                         e1000_consume_page(buffer_info, skb,
3708                                                            length);
3709                                 }
3710                         }
3711                 }
3712
3713                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
3714                 e1000_rx_checksum(adapter,
3715                                   (u32)(status) |
3716                                   ((u32)(rx_desc->errors) << 24),
3717                                   le16_to_cpu(rx_desc->csum), skb);
3718
3719                 pskb_trim(skb, skb->len - 4);
3720
3721                 /* probably a little skewed due to removing CRC */
3722                 total_rx_bytes += skb->len;
3723                 total_rx_packets++;
3724
3725                 /* eth type trans needs skb->data to point to something */
3726                 if (!pskb_may_pull(skb, ETH_HLEN)) {
3727                         DPRINTK(DRV, ERR, "pskb_may_pull failed.\n");
3728                         dev_kfree_skb(skb);
3729                         goto next_desc;
3730                 }
3731
3732                 skb->protocol = eth_type_trans(skb, netdev);
3733
3734                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3735
3736 next_desc:
3737                 rx_desc->status = 0;
3738
3739                 /* return some buffers to hardware, one at a time is too slow */
3740                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3741                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3742                         cleaned_count = 0;
3743                 }
3744
3745                 /* use prefetched values */
3746                 rx_desc = next_rxd;
3747                 buffer_info = next_buffer;
3748         }
3749         rx_ring->next_to_clean = i;
3750
3751         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3752         if (cleaned_count)
3753                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3754
3755         adapter->total_rx_packets += total_rx_packets;
3756         adapter->total_rx_bytes += total_rx_bytes;
3757         adapter->net_stats.rx_bytes += total_rx_bytes;
3758         adapter->net_stats.rx_packets += total_rx_packets;
3759         return cleaned;
3760 }
3761
3762 /**
3763  * e1000_clean_rx_irq - Send received data up the network stack; legacy
3764  * @adapter: board private structure
3765  * @rx_ring: ring to clean
3766  * @work_done: amount of napi work completed this call
3767  * @work_to_do: max amount of work allowed for this call to do
3768  */
3769 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
3770                                struct e1000_rx_ring *rx_ring,
3771                                int *work_done, int work_to_do)
3772 {
3773         struct e1000_hw *hw = &adapter->hw;
3774         struct net_device *netdev = adapter->netdev;
3775         struct pci_dev *pdev = adapter->pdev;
3776         struct e1000_rx_desc *rx_desc, *next_rxd;
3777         struct e1000_buffer *buffer_info, *next_buffer;
3778         unsigned long flags;
3779         u32 length;
3780         unsigned int i;
3781         int cleaned_count = 0;
3782         bool cleaned = false;
3783         unsigned int total_rx_bytes=0, total_rx_packets=0;
3784
3785         i = rx_ring->next_to_clean;
3786         rx_desc = E1000_RX_DESC(*rx_ring, i);
3787         buffer_info = &rx_ring->buffer_info[i];
3788
3789         while (rx_desc->status & E1000_RXD_STAT_DD) {
3790                 struct sk_buff *skb;
3791                 u8 status;
3792
3793                 if (*work_done >= work_to_do)
3794                         break;
3795                 (*work_done)++;
3796
3797                 status = rx_desc->status;
3798                 skb = buffer_info->skb;
3799                 buffer_info->skb = NULL;
3800
3801                 prefetch(skb->data - NET_IP_ALIGN);
3802
3803                 if (++i == rx_ring->count) i = 0;
3804                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3805                 prefetch(next_rxd);
3806
3807                 next_buffer = &rx_ring->buffer_info[i];
3808
3809                 cleaned = true;
3810                 cleaned_count++;
3811                 pci_unmap_single(pdev, buffer_info->dma, buffer_info->length,
3812                                  PCI_DMA_FROMDEVICE);
3813                 buffer_info->dma = 0;
3814
3815                 length = le16_to_cpu(rx_desc->length);
3816                 /* !EOP means multiple descriptors were used to store a single
3817                  * packet, also make sure the frame isn't just CRC only */
3818                 if (unlikely(!(status & E1000_RXD_STAT_EOP) || (length <= 4))) {
3819                         /* All receives must fit into a single buffer */
3820                         E1000_DBG("%s: Receive packet consumed multiple"
3821                                   " buffers\n", netdev->name);
3822                         /* recycle */
3823                         buffer_info->skb = skb;
3824                         goto next_desc;
3825                 }
3826
3827                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
3828                         u8 last_byte = *(skb->data + length - 1);
3829                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3830                                        last_byte)) {
3831                                 spin_lock_irqsave(&adapter->stats_lock, flags);
3832                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3833                                                        length, skb->data);
3834                                 spin_unlock_irqrestore(&adapter->stats_lock,
3835                                                        flags);
3836                                 length--;
3837                         } else {
3838                                 /* recycle */
3839                                 buffer_info->skb = skb;
3840                                 goto next_desc;
3841                         }
3842                 }
3843
3844                 /* adjust length to remove Ethernet CRC, this must be
3845                  * done after the TBI_ACCEPT workaround above */
3846                 length -= 4;
3847
3848                 /* probably a little skewed due to removing CRC */
3849                 total_rx_bytes += length;
3850                 total_rx_packets++;
3851
3852                 /* code added for copybreak, this should improve
3853                  * performance for small packets with large amounts
3854                  * of reassembly being done in the stack */
3855                 if (length < copybreak) {
3856                         struct sk_buff *new_skb =
3857                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
3858                         if (new_skb) {
3859                                 skb_reserve(new_skb, NET_IP_ALIGN);
3860                                 skb_copy_to_linear_data_offset(new_skb,
3861                                                                -NET_IP_ALIGN,
3862                                                                (skb->data -
3863                                                                 NET_IP_ALIGN),
3864                                                                (length +
3865                                                                 NET_IP_ALIGN));
3866                                 /* save the skb in buffer_info as good */
3867                                 buffer_info->skb = skb;
3868                                 skb = new_skb;
3869                         }
3870                         /* else just continue with the old one */
3871                 }
3872                 /* end copybreak code */
3873                 skb_put(skb, length);
3874
3875                 /* Receive Checksum Offload */
3876                 e1000_rx_checksum(adapter,
3877                                   (u32)(status) |
3878                                   ((u32)(rx_desc->errors) << 24),
3879                                   le16_to_cpu(rx_desc->csum), skb);
3880
3881                 skb->protocol = eth_type_trans(skb, netdev);
3882
3883                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3884
3885 next_desc:
3886                 rx_desc->status = 0;
3887
3888                 /* return some buffers to hardware, one at a time is too slow */
3889                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3890                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3891                         cleaned_count = 0;
3892                 }
3893
3894                 /* use prefetched values */
3895                 rx_desc = next_rxd;
3896                 buffer_info = next_buffer;
3897         }
3898         rx_ring->next_to_clean = i;
3899
3900         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3901         if (cleaned_count)
3902                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3903
3904         adapter->total_rx_packets += total_rx_packets;
3905         adapter->total_rx_bytes += total_rx_bytes;
3906         adapter->net_stats.rx_bytes += total_rx_bytes;
3907         adapter->net_stats.rx_packets += total_rx_packets;
3908         return cleaned;
3909 }
3910
3911 /**
3912  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
3913  * @adapter: address of board private structure
3914  * @rx_ring: pointer to receive ring structure
3915  * @cleaned_count: number of buffers to allocate this pass
3916  **/
3917
3918 static void
3919 e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
3920                              struct e1000_rx_ring *rx_ring, int cleaned_count)
3921 {
3922         struct net_device *netdev = adapter->netdev;
3923         struct pci_dev *pdev = adapter->pdev;
3924         struct e1000_rx_desc *rx_desc;
3925         struct e1000_buffer *buffer_info;
3926         struct sk_buff *skb;
3927         unsigned int i;
3928         unsigned int bufsz = 256 -
3929                              16 /*for skb_reserve */ -
3930                              NET_IP_ALIGN;
3931
3932         i = rx_ring->next_to_use;
3933         buffer_info = &rx_ring->buffer_info[i];
3934
3935         while (cleaned_count--) {
3936                 skb = buffer_info->skb;
3937                 if (skb) {
3938                         skb_trim(skb, 0);
3939                         goto check_page;
3940                 }
3941
3942                 skb = netdev_alloc_skb(netdev, bufsz);
3943                 if (unlikely(!skb)) {
3944                         /* Better luck next round */
3945                         adapter->alloc_rx_buff_failed++;
3946                         break;
3947                 }
3948
3949                 /* Fix for errata 23, can't cross 64kB boundary */
3950                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3951                         struct sk_buff *oldskb = skb;
3952                         DPRINTK(PROBE, ERR, "skb align check failed: %u bytes "
3953                                              "at %p\n", bufsz, skb->data);
3954                         /* Try again, without freeing the previous */
3955                         skb = netdev_alloc_skb(netdev, bufsz);
3956                         /* Failed allocation, critical failure */
3957                         if (!skb) {
3958                                 dev_kfree_skb(oldskb);
3959                                 adapter->alloc_rx_buff_failed++;
3960                                 break;
3961                         }
3962
3963                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3964                                 /* give up */
3965                                 dev_kfree_skb(skb);
3966                                 dev_kfree_skb(oldskb);
3967                                 break; /* while (cleaned_count--) */
3968                         }
3969
3970                         /* Use new allocation */
3971                         dev_kfree_skb(oldskb);
3972                 }
3973                 /* Make buffer alignment 2 beyond a 16 byte boundary
3974                  * this will result in a 16 byte aligned IP header after
3975                  * the 14 byte MAC header is removed
3976                  */
3977                 skb_reserve(skb, NET_IP_ALIGN);
3978
3979                 buffer_info->skb = skb;
3980                 buffer_info->length = adapter->rx_buffer_len;
3981 check_page:
3982                 /* allocate a new page if necessary */
3983                 if (!buffer_info->page) {
3984                         buffer_info->page = alloc_page(GFP_ATOMIC);
3985                         if (unlikely(!buffer_info->page)) {
3986                                 adapter->alloc_rx_buff_failed++;
3987                                 break;
3988                         }
3989                 }
3990
3991                 if (!buffer_info->dma)
3992                         buffer_info->dma = pci_map_page(pdev,
3993                                                         buffer_info->page, 0,
3994                                                         buffer_info->length,
3995                                                         PCI_DMA_FROMDEVICE);
3996
3997                 rx_desc = E1000_RX_DESC(*rx_ring, i);
3998                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3999
4000                 if (unlikely(++i == rx_ring->count))
4001                         i = 0;
4002                 buffer_info = &rx_ring->buffer_info[i];
4003         }
4004
4005         if (likely(rx_ring->next_to_use != i)) {
4006                 rx_ring->next_to_use = i;
4007                 if (unlikely(i-- == 0))
4008                         i = (rx_ring->count - 1);
4009
4010                 /* Force memory writes to complete before letting h/w
4011                  * know there are new descriptors to fetch.  (Only
4012                  * applicable for weak-ordered memory model archs,
4013                  * such as IA-64). */
4014                 wmb();
4015                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4016         }
4017 }
4018
4019 /**
4020  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4021  * @adapter: address of board private structure
4022  **/
4023
4024 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4025                                    struct e1000_rx_ring *rx_ring,
4026                                    int cleaned_count)
4027 {
4028         struct e1000_hw *hw = &adapter->hw;
4029         struct net_device *netdev = adapter->netdev;
4030         struct pci_dev *pdev = adapter->pdev;
4031         struct e1000_rx_desc *rx_desc;
4032         struct e1000_buffer *buffer_info;
4033         struct sk_buff *skb;
4034         unsigned int i;
4035         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
4036
4037         i = rx_ring->next_to_use;
4038         buffer_info = &rx_ring->buffer_info[i];
4039
4040         while (cleaned_count--) {
4041                 skb = buffer_info->skb;
4042                 if (skb) {
4043                         skb_trim(skb, 0);
4044                         goto map_skb;
4045                 }
4046
4047                 skb = netdev_alloc_skb(netdev, bufsz);
4048                 if (unlikely(!skb)) {
4049                         /* Better luck next round */
4050                         adapter->alloc_rx_buff_failed++;
4051                         break;
4052                 }
4053
4054                 /* Fix for errata 23, can't cross 64kB boundary */
4055                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4056                         struct sk_buff *oldskb = skb;
4057                         DPRINTK(RX_ERR, ERR, "skb align check failed: %u bytes "
4058                                              "at %p\n", bufsz, skb->data);
4059                         /* Try again, without freeing the previous */
4060                         skb = netdev_alloc_skb(netdev, bufsz);
4061                         /* Failed allocation, critical failure */
4062                         if (!skb) {
4063                                 dev_kfree_skb(oldskb);
4064                                 adapter->alloc_rx_buff_failed++;
4065                                 break;
4066                         }
4067
4068                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4069                                 /* give up */
4070                                 dev_kfree_skb(skb);
4071                                 dev_kfree_skb(oldskb);
4072                                 adapter->alloc_rx_buff_failed++;
4073                                 break; /* while !buffer_info->skb */
4074                         }
4075
4076                         /* Use new allocation */
4077                         dev_kfree_skb(oldskb);
4078                 }
4079                 /* Make buffer alignment 2 beyond a 16 byte boundary
4080                  * this will result in a 16 byte aligned IP header after
4081                  * the 14 byte MAC header is removed
4082                  */
4083                 skb_reserve(skb, NET_IP_ALIGN);
4084
4085                 buffer_info->skb = skb;
4086                 buffer_info->length = adapter->rx_buffer_len;
4087 map_skb:
4088                 buffer_info->dma = pci_map_single(pdev,
4089                                                   skb->data,
4090                                                   buffer_info->length,
4091                                                   PCI_DMA_FROMDEVICE);
4092
4093                 /*
4094                  * XXX if it was allocated cleanly it will never map to a
4095                  * boundary crossing
4096                  */
4097
4098                 /* Fix for errata 23, can't cross 64kB boundary */
4099                 if (!e1000_check_64k_bound(adapter,
4100                                         (void *)(unsigned long)buffer_info->dma,
4101                                         adapter->rx_buffer_len)) {
4102                         DPRINTK(RX_ERR, ERR,
4103                                 "dma align check failed: %u bytes at %p\n",
4104                                 adapter->rx_buffer_len,
4105                                 (void *)(unsigned long)buffer_info->dma);
4106                         dev_kfree_skb(skb);
4107                         buffer_info->skb = NULL;
4108
4109                         pci_unmap_single(pdev, buffer_info->dma,
4110                                          adapter->rx_buffer_len,
4111                                          PCI_DMA_FROMDEVICE);
4112                         buffer_info->dma = 0;
4113
4114                         adapter->alloc_rx_buff_failed++;
4115                         break; /* while !buffer_info->skb */
4116                 }
4117                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4118                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4119
4120                 if (unlikely(++i == rx_ring->count))
4121                         i = 0;
4122                 buffer_info = &rx_ring->buffer_info[i];
4123         }
4124
4125         if (likely(rx_ring->next_to_use != i)) {
4126                 rx_ring->next_to_use = i;
4127                 if (unlikely(i-- == 0))
4128                         i = (rx_ring->count - 1);
4129
4130                 /* Force memory writes to complete before letting h/w
4131                  * know there are new descriptors to fetch.  (Only
4132                  * applicable for weak-ordered memory model archs,
4133                  * such as IA-64). */
4134                 wmb();
4135                 writel(i, hw->hw_addr + rx_ring->rdt);
4136         }
4137 }
4138
4139 /**
4140  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4141  * @adapter:
4142  **/
4143
4144 static void e1000_smartspeed(struct e1000_adapter *adapter)
4145 {
4146         struct e1000_hw *hw = &adapter->hw;
4147         u16 phy_status;
4148         u16 phy_ctrl;
4149
4150         if ((hw->phy_type != e1000_phy_igp) || !hw->autoneg ||
4151            !(hw->autoneg_advertised & ADVERTISE_1000_FULL))
4152                 return;
4153
4154         if (adapter->smartspeed == 0) {
4155                 /* If Master/Slave config fault is asserted twice,
4156                  * we assume back-to-back */
4157                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4158                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4159                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4160                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4161                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4162                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4163                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4164                         e1000_write_phy_reg(hw, PHY_1000T_CTRL,
4165                                             phy_ctrl);
4166                         adapter->smartspeed++;
4167                         if (!e1000_phy_setup_autoneg(hw) &&
4168                            !e1000_read_phy_reg(hw, PHY_CTRL,
4169                                                &phy_ctrl)) {
4170                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4171                                              MII_CR_RESTART_AUTO_NEG);
4172                                 e1000_write_phy_reg(hw, PHY_CTRL,
4173                                                     phy_ctrl);
4174                         }
4175                 }
4176                 return;
4177         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4178                 /* If still no link, perhaps using 2/3 pair cable */
4179                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4180                 phy_ctrl |= CR_1000T_MS_ENABLE;
4181                 e1000_write_phy_reg(hw, PHY_1000T_CTRL, phy_ctrl);
4182                 if (!e1000_phy_setup_autoneg(hw) &&
4183                    !e1000_read_phy_reg(hw, PHY_CTRL, &phy_ctrl)) {
4184                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4185                                      MII_CR_RESTART_AUTO_NEG);
4186                         e1000_write_phy_reg(hw, PHY_CTRL, phy_ctrl);
4187                 }
4188         }
4189         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4190         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4191                 adapter->smartspeed = 0;
4192 }
4193
4194 /**
4195  * e1000_ioctl -
4196  * @netdev:
4197  * @ifreq:
4198  * @cmd:
4199  **/
4200
4201 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4202 {
4203         switch (cmd) {
4204         case SIOCGMIIPHY:
4205         case SIOCGMIIREG:
4206         case SIOCSMIIREG:
4207                 return e1000_mii_ioctl(netdev, ifr, cmd);
4208         default:
4209                 return -EOPNOTSUPP;
4210         }
4211 }
4212
4213 /**
4214  * e1000_mii_ioctl -
4215  * @netdev:
4216  * @ifreq:
4217  * @cmd:
4218  **/
4219
4220 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4221                            int cmd)
4222 {
4223         struct e1000_adapter *adapter = netdev_priv(netdev);
4224         struct e1000_hw *hw = &adapter->hw;
4225         struct mii_ioctl_data *data = if_mii(ifr);
4226         int retval;
4227         u16 mii_reg;
4228         u16 spddplx;
4229         unsigned long flags;
4230
4231         if (hw->media_type != e1000_media_type_copper)
4232                 return -EOPNOTSUPP;
4233
4234         switch (cmd) {
4235         case SIOCGMIIPHY:
4236                 data->phy_id = hw->phy_addr;
4237                 break;
4238         case SIOCGMIIREG:
4239                 spin_lock_irqsave(&adapter->stats_lock, flags);
4240                 if (e1000_read_phy_reg(hw, data->reg_num & 0x1F,
4241                                    &data->val_out)) {
4242                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4243                         return -EIO;
4244                 }
4245                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4246                 break;
4247         case SIOCSMIIREG:
4248                 if (data->reg_num & ~(0x1F))
4249                         return -EFAULT;
4250                 mii_reg = data->val_in;
4251                 spin_lock_irqsave(&adapter->stats_lock, flags);
4252                 if (e1000_write_phy_reg(hw, data->reg_num,
4253                                         mii_reg)) {
4254                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4255                         return -EIO;
4256                 }
4257                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4258                 if (hw->media_type == e1000_media_type_copper) {
4259                         switch (data->reg_num) {
4260                         case PHY_CTRL:
4261                                 if (mii_reg & MII_CR_POWER_DOWN)
4262                                         break;
4263                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4264                                         hw->autoneg = 1;
4265                                         hw->autoneg_advertised = 0x2F;
4266                                 } else {
4267                                         if (mii_reg & 0x40)
4268                                                 spddplx = SPEED_1000;
4269                                         else if (mii_reg & 0x2000)
4270                                                 spddplx = SPEED_100;
4271                                         else
4272                                                 spddplx = SPEED_10;
4273                                         spddplx += (mii_reg & 0x100)
4274                                                    ? DUPLEX_FULL :
4275                                                    DUPLEX_HALF;
4276                                         retval = e1000_set_spd_dplx(adapter,
4277                                                                     spddplx);
4278                                         if (retval)
4279                                                 return retval;
4280                                 }
4281                                 if (netif_running(adapter->netdev))
4282                                         e1000_reinit_locked(adapter);
4283                                 else
4284                                         e1000_reset(adapter);
4285                                 break;
4286                         case M88E1000_PHY_SPEC_CTRL:
4287                         case M88E1000_EXT_PHY_SPEC_CTRL:
4288                                 if (e1000_phy_reset(hw))
4289                                         return -EIO;
4290                                 break;
4291                         }
4292                 } else {
4293                         switch (data->reg_num) {
4294                         case PHY_CTRL:
4295                                 if (mii_reg & MII_CR_POWER_DOWN)
4296                                         break;
4297                                 if (netif_running(adapter->netdev))
4298                                         e1000_reinit_locked(adapter);
4299                                 else
4300                                         e1000_reset(adapter);
4301                                 break;
4302                         }
4303                 }
4304                 break;
4305         default:
4306                 return -EOPNOTSUPP;
4307         }
4308         return E1000_SUCCESS;
4309 }
4310
4311 void e1000_pci_set_mwi(struct e1000_hw *hw)
4312 {
4313         struct e1000_adapter *adapter = hw->back;
4314         int ret_val = pci_set_mwi(adapter->pdev);
4315
4316         if (ret_val)
4317                 DPRINTK(PROBE, ERR, "Error in setting MWI\n");
4318 }
4319
4320 void e1000_pci_clear_mwi(struct e1000_hw *hw)
4321 {
4322         struct e1000_adapter *adapter = hw->back;
4323
4324         pci_clear_mwi(adapter->pdev);
4325 }
4326
4327 int e1000_pcix_get_mmrbc(struct e1000_hw *hw)
4328 {
4329         struct e1000_adapter *adapter = hw->back;
4330         return pcix_get_mmrbc(adapter->pdev);
4331 }
4332
4333 void e1000_pcix_set_mmrbc(struct e1000_hw *hw, int mmrbc)
4334 {
4335         struct e1000_adapter *adapter = hw->back;
4336         pcix_set_mmrbc(adapter->pdev, mmrbc);
4337 }
4338
4339 s32 e1000_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
4340 {
4341     struct e1000_adapter *adapter = hw->back;
4342     u16 cap_offset;
4343
4344     cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
4345     if (!cap_offset)
4346         return -E1000_ERR_CONFIG;
4347
4348     pci_read_config_word(adapter->pdev, cap_offset + reg, value);
4349
4350     return E1000_SUCCESS;
4351 }
4352
4353 void e1000_io_write(struct e1000_hw *hw, unsigned long port, u32 value)
4354 {
4355         outl(value, port);
4356 }
4357
4358 static void e1000_vlan_rx_register(struct net_device *netdev,
4359                                    struct vlan_group *grp)
4360 {
4361         struct e1000_adapter *adapter = netdev_priv(netdev);
4362         struct e1000_hw *hw = &adapter->hw;
4363         u32 ctrl, rctl;
4364
4365         if (!test_bit(__E1000_DOWN, &adapter->flags))
4366                 e1000_irq_disable(adapter);
4367         adapter->vlgrp = grp;
4368
4369         if (grp) {
4370                 /* enable VLAN tag insert/strip */
4371                 ctrl = er32(CTRL);
4372                 ctrl |= E1000_CTRL_VME;
4373                 ew32(CTRL, ctrl);
4374
4375                 /* enable VLAN receive filtering */
4376                 rctl = er32(RCTL);
4377                 rctl &= ~E1000_RCTL_CFIEN;
4378                 if (!(netdev->flags & IFF_PROMISC))
4379                         rctl |= E1000_RCTL_VFE;
4380                 ew32(RCTL, rctl);
4381                 e1000_update_mng_vlan(adapter);
4382         } else {
4383                 /* disable VLAN tag insert/strip */
4384                 ctrl = er32(CTRL);
4385                 ctrl &= ~E1000_CTRL_VME;
4386                 ew32(CTRL, ctrl);
4387
4388                 /* disable VLAN receive filtering */
4389                 rctl = er32(RCTL);
4390                 rctl &= ~E1000_RCTL_VFE;
4391                 ew32(RCTL, rctl);
4392
4393                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
4394                         e1000_vlan_rx_kill_vid(netdev,
4395                                                adapter->mng_vlan_id);
4396                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4397                 }
4398         }
4399
4400         if (!test_bit(__E1000_DOWN, &adapter->flags))
4401                 e1000_irq_enable(adapter);
4402 }
4403
4404 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4405 {
4406         struct e1000_adapter *adapter = netdev_priv(netdev);
4407         struct e1000_hw *hw = &adapter->hw;
4408         u32 vfta, index;
4409
4410         if ((hw->mng_cookie.status &
4411              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4412             (vid == adapter->mng_vlan_id))
4413                 return;
4414         /* add VID to filter table */
4415         index = (vid >> 5) & 0x7F;
4416         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4417         vfta |= (1 << (vid & 0x1F));
4418         e1000_write_vfta(hw, index, vfta);
4419 }
4420
4421 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
4422 {
4423         struct e1000_adapter *adapter = netdev_priv(netdev);
4424         struct e1000_hw *hw = &adapter->hw;
4425         u32 vfta, index;
4426
4427         if (!test_bit(__E1000_DOWN, &adapter->flags))
4428                 e1000_irq_disable(adapter);
4429         vlan_group_set_device(adapter->vlgrp, vid, NULL);
4430         if (!test_bit(__E1000_DOWN, &adapter->flags))
4431                 e1000_irq_enable(adapter);
4432
4433         /* remove VID from filter table */
4434         index = (vid >> 5) & 0x7F;
4435         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4436         vfta &= ~(1 << (vid & 0x1F));
4437         e1000_write_vfta(hw, index, vfta);
4438 }
4439
4440 static void e1000_restore_vlan(struct e1000_adapter *adapter)
4441 {
4442         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
4443
4444         if (adapter->vlgrp) {
4445                 u16 vid;
4446                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
4447                         if (!vlan_group_get_device(adapter->vlgrp, vid))
4448                                 continue;
4449                         e1000_vlan_rx_add_vid(adapter->netdev, vid);
4450                 }
4451         }
4452 }
4453
4454 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx)
4455 {
4456         struct e1000_hw *hw = &adapter->hw;
4457
4458         hw->autoneg = 0;
4459
4460         /* Fiber NICs only allow 1000 gbps Full duplex */
4461         if ((hw->media_type == e1000_media_type_fiber) &&
4462                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
4463                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
4464                 return -EINVAL;
4465         }
4466
4467         switch (spddplx) {
4468         case SPEED_10 + DUPLEX_HALF:
4469                 hw->forced_speed_duplex = e1000_10_half;
4470                 break;
4471         case SPEED_10 + DUPLEX_FULL:
4472                 hw->forced_speed_duplex = e1000_10_full;
4473                 break;
4474         case SPEED_100 + DUPLEX_HALF:
4475                 hw->forced_speed_duplex = e1000_100_half;
4476                 break;
4477         case SPEED_100 + DUPLEX_FULL:
4478                 hw->forced_speed_duplex = e1000_100_full;
4479                 break;
4480         case SPEED_1000 + DUPLEX_FULL:
4481                 hw->autoneg = 1;
4482                 hw->autoneg_advertised = ADVERTISE_1000_FULL;
4483                 break;
4484         case SPEED_1000 + DUPLEX_HALF: /* not supported */
4485         default:
4486                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
4487                 return -EINVAL;
4488         }
4489         return 0;
4490 }
4491
4492 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4493 {
4494         struct net_device *netdev = pci_get_drvdata(pdev);
4495         struct e1000_adapter *adapter = netdev_priv(netdev);
4496         struct e1000_hw *hw = &adapter->hw;
4497         u32 ctrl, ctrl_ext, rctl, status;
4498         u32 wufc = adapter->wol;
4499 #ifdef CONFIG_PM
4500         int retval = 0;
4501 #endif
4502
4503         netif_device_detach(netdev);
4504
4505         if (netif_running(netdev)) {
4506                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
4507                 e1000_down(adapter);
4508         }
4509
4510 #ifdef CONFIG_PM
4511         retval = pci_save_state(pdev);
4512         if (retval)
4513                 return retval;
4514 #endif
4515
4516         status = er32(STATUS);
4517         if (status & E1000_STATUS_LU)
4518                 wufc &= ~E1000_WUFC_LNKC;
4519
4520         if (wufc) {
4521                 e1000_setup_rctl(adapter);
4522                 e1000_set_rx_mode(netdev);
4523
4524                 /* turn on all-multi mode if wake on multicast is enabled */
4525                 if (wufc & E1000_WUFC_MC) {
4526                         rctl = er32(RCTL);
4527                         rctl |= E1000_RCTL_MPE;
4528                         ew32(RCTL, rctl);
4529                 }
4530
4531                 if (hw->mac_type >= e1000_82540) {
4532                         ctrl = er32(CTRL);
4533                         /* advertise wake from D3Cold */
4534                         #define E1000_CTRL_ADVD3WUC 0x00100000
4535                         /* phy power management enable */
4536                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4537                         ctrl |= E1000_CTRL_ADVD3WUC |
4538                                 E1000_CTRL_EN_PHY_PWR_MGMT;
4539                         ew32(CTRL, ctrl);
4540                 }
4541
4542                 if (hw->media_type == e1000_media_type_fiber ||
4543                     hw->media_type == e1000_media_type_internal_serdes) {
4544                         /* keep the laser running in D3 */
4545                         ctrl_ext = er32(CTRL_EXT);
4546                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4547                         ew32(CTRL_EXT, ctrl_ext);
4548                 }
4549
4550                 ew32(WUC, E1000_WUC_PME_EN);
4551                 ew32(WUFC, wufc);
4552         } else {
4553                 ew32(WUC, 0);
4554                 ew32(WUFC, 0);
4555         }
4556
4557         e1000_release_manageability(adapter);
4558
4559         *enable_wake = !!wufc;
4560
4561         /* make sure adapter isn't asleep if manageability is enabled */
4562         if (adapter->en_mng_pt)
4563                 *enable_wake = true;
4564
4565         if (netif_running(netdev))
4566                 e1000_free_irq(adapter);
4567
4568         pci_disable_device(pdev);
4569
4570         return 0;
4571 }
4572
4573 #ifdef CONFIG_PM
4574 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4575 {
4576         int retval;
4577         bool wake;
4578
4579         retval = __e1000_shutdown(pdev, &wake);
4580         if (retval)
4581                 return retval;
4582
4583         if (wake) {
4584                 pci_prepare_to_sleep(pdev);
4585         } else {
4586                 pci_wake_from_d3(pdev, false);
4587                 pci_set_power_state(pdev, PCI_D3hot);
4588         }
4589
4590         return 0;
4591 }
4592
4593 static int e1000_resume(struct pci_dev *pdev)
4594 {
4595         struct net_device *netdev = pci_get_drvdata(pdev);
4596         struct e1000_adapter *adapter = netdev_priv(netdev);
4597         struct e1000_hw *hw = &adapter->hw;
4598         u32 err;
4599
4600         pci_set_power_state(pdev, PCI_D0);
4601         pci_restore_state(pdev);
4602
4603         if (adapter->need_ioport)
4604                 err = pci_enable_device(pdev);
4605         else
4606                 err = pci_enable_device_mem(pdev);
4607         if (err) {
4608                 printk(KERN_ERR "e1000: Cannot enable PCI device from suspend\n");
4609                 return err;
4610         }
4611         pci_set_master(pdev);
4612
4613         pci_enable_wake(pdev, PCI_D3hot, 0);
4614         pci_enable_wake(pdev, PCI_D3cold, 0);
4615
4616         if (netif_running(netdev)) {
4617                 err = e1000_request_irq(adapter);
4618                 if (err)
4619                         return err;
4620         }
4621
4622         e1000_power_up_phy(adapter);
4623         e1000_reset(adapter);
4624         ew32(WUS, ~0);
4625
4626         e1000_init_manageability(adapter);
4627
4628         if (netif_running(netdev))
4629                 e1000_up(adapter);
4630
4631         netif_device_attach(netdev);
4632
4633         return 0;
4634 }
4635 #endif
4636
4637 static void e1000_shutdown(struct pci_dev *pdev)
4638 {
4639         bool wake;
4640
4641         __e1000_shutdown(pdev, &wake);
4642
4643         if (system_state == SYSTEM_POWER_OFF) {
4644                 pci_wake_from_d3(pdev, wake);
4645                 pci_set_power_state(pdev, PCI_D3hot);
4646         }
4647 }
4648
4649 #ifdef CONFIG_NET_POLL_CONTROLLER
4650 /*
4651  * Polling 'interrupt' - used by things like netconsole to send skbs
4652  * without having to re-enable interrupts. It's not called while
4653  * the interrupt routine is executing.
4654  */
4655 static void e1000_netpoll(struct net_device *netdev)
4656 {
4657         struct e1000_adapter *adapter = netdev_priv(netdev);
4658
4659         disable_irq(adapter->pdev->irq);
4660         e1000_intr(adapter->pdev->irq, netdev);
4661         enable_irq(adapter->pdev->irq);
4662 }
4663 #endif
4664
4665 /**
4666  * e1000_io_error_detected - called when PCI error is detected
4667  * @pdev: Pointer to PCI device
4668  * @state: The current pci conneection state
4669  *
4670  * This function is called after a PCI bus error affecting
4671  * this device has been detected.
4672  */
4673 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4674                                                 pci_channel_state_t state)
4675 {
4676         struct net_device *netdev = pci_get_drvdata(pdev);
4677         struct e1000_adapter *adapter = netdev_priv(netdev);
4678
4679         netif_device_detach(netdev);
4680
4681         if (state == pci_channel_io_perm_failure)
4682                 return PCI_ERS_RESULT_DISCONNECT;
4683
4684         if (netif_running(netdev))
4685                 e1000_down(adapter);
4686         pci_disable_device(pdev);
4687
4688         /* Request a slot slot reset. */
4689         return PCI_ERS_RESULT_NEED_RESET;
4690 }
4691
4692 /**
4693  * e1000_io_slot_reset - called after the pci bus has been reset.
4694  * @pdev: Pointer to PCI device
4695  *
4696  * Restart the card from scratch, as if from a cold-boot. Implementation
4697  * resembles the first-half of the e1000_resume routine.
4698  */
4699 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4700 {
4701         struct net_device *netdev = pci_get_drvdata(pdev);
4702         struct e1000_adapter *adapter = netdev_priv(netdev);
4703         struct e1000_hw *hw = &adapter->hw;
4704         int err;
4705
4706         if (adapter->need_ioport)
4707                 err = pci_enable_device(pdev);
4708         else
4709                 err = pci_enable_device_mem(pdev);
4710         if (err) {
4711                 printk(KERN_ERR "e1000: Cannot re-enable PCI device after reset.\n");
4712                 return PCI_ERS_RESULT_DISCONNECT;
4713         }
4714         pci_set_master(pdev);
4715
4716         pci_enable_wake(pdev, PCI_D3hot, 0);
4717         pci_enable_wake(pdev, PCI_D3cold, 0);
4718
4719         e1000_reset(adapter);
4720         ew32(WUS, ~0);
4721
4722         return PCI_ERS_RESULT_RECOVERED;
4723 }
4724
4725 /**
4726  * e1000_io_resume - called when traffic can start flowing again.
4727  * @pdev: Pointer to PCI device
4728  *
4729  * This callback is called when the error recovery driver tells us that
4730  * its OK to resume normal operation. Implementation resembles the
4731  * second-half of the e1000_resume routine.
4732  */
4733 static void e1000_io_resume(struct pci_dev *pdev)
4734 {
4735         struct net_device *netdev = pci_get_drvdata(pdev);
4736         struct e1000_adapter *adapter = netdev_priv(netdev);
4737
4738         e1000_init_manageability(adapter);
4739
4740         if (netif_running(netdev)) {
4741                 if (e1000_up(adapter)) {
4742                         printk("e1000: can't bring device back up after reset\n");
4743                         return;
4744                 }
4745         }
4746
4747         netif_device_attach(netdev);
4748 }
4749
4750 /* e1000_main.c */