Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[safe/jmp/linux-2.6] / drivers / net / e1000 / e1000_main.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2006 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include "e1000.h"
30 #include <net/ip6_checksum.h>
31
32 char e1000_driver_name[] = "e1000";
33 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
34 #define DRV_VERSION "7.3.21-k5-NAPI"
35 const char e1000_driver_version[] = DRV_VERSION;
36 static const char e1000_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
37
38 /* e1000_pci_tbl - PCI Device ID Table
39  *
40  * Last entry must be all 0s
41  *
42  * Macro expands to...
43  *   {PCI_DEVICE(PCI_VENDOR_ID_INTEL, device_id)}
44  */
45 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
46         INTEL_E1000_ETHERNET_DEVICE(0x1000),
47         INTEL_E1000_ETHERNET_DEVICE(0x1001),
48         INTEL_E1000_ETHERNET_DEVICE(0x1004),
49         INTEL_E1000_ETHERNET_DEVICE(0x1008),
50         INTEL_E1000_ETHERNET_DEVICE(0x1009),
51         INTEL_E1000_ETHERNET_DEVICE(0x100C),
52         INTEL_E1000_ETHERNET_DEVICE(0x100D),
53         INTEL_E1000_ETHERNET_DEVICE(0x100E),
54         INTEL_E1000_ETHERNET_DEVICE(0x100F),
55         INTEL_E1000_ETHERNET_DEVICE(0x1010),
56         INTEL_E1000_ETHERNET_DEVICE(0x1011),
57         INTEL_E1000_ETHERNET_DEVICE(0x1012),
58         INTEL_E1000_ETHERNET_DEVICE(0x1013),
59         INTEL_E1000_ETHERNET_DEVICE(0x1014),
60         INTEL_E1000_ETHERNET_DEVICE(0x1015),
61         INTEL_E1000_ETHERNET_DEVICE(0x1016),
62         INTEL_E1000_ETHERNET_DEVICE(0x1017),
63         INTEL_E1000_ETHERNET_DEVICE(0x1018),
64         INTEL_E1000_ETHERNET_DEVICE(0x1019),
65         INTEL_E1000_ETHERNET_DEVICE(0x101A),
66         INTEL_E1000_ETHERNET_DEVICE(0x101D),
67         INTEL_E1000_ETHERNET_DEVICE(0x101E),
68         INTEL_E1000_ETHERNET_DEVICE(0x1026),
69         INTEL_E1000_ETHERNET_DEVICE(0x1027),
70         INTEL_E1000_ETHERNET_DEVICE(0x1028),
71         INTEL_E1000_ETHERNET_DEVICE(0x1075),
72         INTEL_E1000_ETHERNET_DEVICE(0x1076),
73         INTEL_E1000_ETHERNET_DEVICE(0x1077),
74         INTEL_E1000_ETHERNET_DEVICE(0x1078),
75         INTEL_E1000_ETHERNET_DEVICE(0x1079),
76         INTEL_E1000_ETHERNET_DEVICE(0x107A),
77         INTEL_E1000_ETHERNET_DEVICE(0x107B),
78         INTEL_E1000_ETHERNET_DEVICE(0x107C),
79         INTEL_E1000_ETHERNET_DEVICE(0x108A),
80         INTEL_E1000_ETHERNET_DEVICE(0x1099),
81         INTEL_E1000_ETHERNET_DEVICE(0x10B5),
82         /* required last entry */
83         {0,}
84 };
85
86 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
87
88 int e1000_up(struct e1000_adapter *adapter);
89 void e1000_down(struct e1000_adapter *adapter);
90 void e1000_reinit_locked(struct e1000_adapter *adapter);
91 void e1000_reset(struct e1000_adapter *adapter);
92 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx);
93 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
94 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
95 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
96 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
97 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
98                              struct e1000_tx_ring *txdr);
99 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
100                              struct e1000_rx_ring *rxdr);
101 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
102                              struct e1000_tx_ring *tx_ring);
103 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
104                              struct e1000_rx_ring *rx_ring);
105 void e1000_update_stats(struct e1000_adapter *adapter);
106
107 static int e1000_init_module(void);
108 static void e1000_exit_module(void);
109 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
110 static void __devexit e1000_remove(struct pci_dev *pdev);
111 static int e1000_alloc_queues(struct e1000_adapter *adapter);
112 static int e1000_sw_init(struct e1000_adapter *adapter);
113 static int e1000_open(struct net_device *netdev);
114 static int e1000_close(struct net_device *netdev);
115 static void e1000_configure_tx(struct e1000_adapter *adapter);
116 static void e1000_configure_rx(struct e1000_adapter *adapter);
117 static void e1000_setup_rctl(struct e1000_adapter *adapter);
118 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
119 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
120 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
121                                 struct e1000_tx_ring *tx_ring);
122 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
123                                 struct e1000_rx_ring *rx_ring);
124 static void e1000_set_rx_mode(struct net_device *netdev);
125 static void e1000_update_phy_info(unsigned long data);
126 static void e1000_watchdog(unsigned long data);
127 static void e1000_82547_tx_fifo_stall(unsigned long data);
128 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
129                                     struct net_device *netdev);
130 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
131 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
132 static int e1000_set_mac(struct net_device *netdev, void *p);
133 static irqreturn_t e1000_intr(int irq, void *data);
134 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
135                                struct e1000_tx_ring *tx_ring);
136 static int e1000_clean(struct napi_struct *napi, int budget);
137 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
138                                struct e1000_rx_ring *rx_ring,
139                                int *work_done, int work_to_do);
140 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
141                                      struct e1000_rx_ring *rx_ring,
142                                      int *work_done, int work_to_do);
143 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
144                                    struct e1000_rx_ring *rx_ring,
145                                    int cleaned_count);
146 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
147                                          struct e1000_rx_ring *rx_ring,
148                                          int cleaned_count);
149 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
150 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
151                            int cmd);
152 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
153 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
154 static void e1000_tx_timeout(struct net_device *dev);
155 static void e1000_reset_task(struct work_struct *work);
156 static void e1000_smartspeed(struct e1000_adapter *adapter);
157 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
158                                        struct sk_buff *skb);
159
160 static void e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp);
161 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid);
162 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid);
163 static void e1000_restore_vlan(struct e1000_adapter *adapter);
164
165 #ifdef CONFIG_PM
166 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
167 static int e1000_resume(struct pci_dev *pdev);
168 #endif
169 static void e1000_shutdown(struct pci_dev *pdev);
170
171 #ifdef CONFIG_NET_POLL_CONTROLLER
172 /* for netdump / net console */
173 static void e1000_netpoll (struct net_device *netdev);
174 #endif
175
176 #define COPYBREAK_DEFAULT 256
177 static unsigned int copybreak __read_mostly = COPYBREAK_DEFAULT;
178 module_param(copybreak, uint, 0644);
179 MODULE_PARM_DESC(copybreak,
180         "Maximum size of packet that is copied to a new buffer on receive");
181
182 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
183                      pci_channel_state_t state);
184 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
185 static void e1000_io_resume(struct pci_dev *pdev);
186
187 static struct pci_error_handlers e1000_err_handler = {
188         .error_detected = e1000_io_error_detected,
189         .slot_reset = e1000_io_slot_reset,
190         .resume = e1000_io_resume,
191 };
192
193 static struct pci_driver e1000_driver = {
194         .name     = e1000_driver_name,
195         .id_table = e1000_pci_tbl,
196         .probe    = e1000_probe,
197         .remove   = __devexit_p(e1000_remove),
198 #ifdef CONFIG_PM
199         /* Power Managment Hooks */
200         .suspend  = e1000_suspend,
201         .resume   = e1000_resume,
202 #endif
203         .shutdown = e1000_shutdown,
204         .err_handler = &e1000_err_handler
205 };
206
207 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
208 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
209 MODULE_LICENSE("GPL");
210 MODULE_VERSION(DRV_VERSION);
211
212 static int debug = NETIF_MSG_DRV | NETIF_MSG_PROBE;
213 module_param(debug, int, 0);
214 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
215
216 /**
217  * e1000_init_module - Driver Registration Routine
218  *
219  * e1000_init_module is the first routine called when the driver is
220  * loaded. All it does is register with the PCI subsystem.
221  **/
222
223 static int __init e1000_init_module(void)
224 {
225         int ret;
226         printk(KERN_INFO "%s - version %s\n",
227                e1000_driver_string, e1000_driver_version);
228
229         printk(KERN_INFO "%s\n", e1000_copyright);
230
231         ret = pci_register_driver(&e1000_driver);
232         if (copybreak != COPYBREAK_DEFAULT) {
233                 if (copybreak == 0)
234                         printk(KERN_INFO "e1000: copybreak disabled\n");
235                 else
236                         printk(KERN_INFO "e1000: copybreak enabled for "
237                                "packets <= %u bytes\n", copybreak);
238         }
239         return ret;
240 }
241
242 module_init(e1000_init_module);
243
244 /**
245  * e1000_exit_module - Driver Exit Cleanup Routine
246  *
247  * e1000_exit_module is called just before the driver is removed
248  * from memory.
249  **/
250
251 static void __exit e1000_exit_module(void)
252 {
253         pci_unregister_driver(&e1000_driver);
254 }
255
256 module_exit(e1000_exit_module);
257
258 static int e1000_request_irq(struct e1000_adapter *adapter)
259 {
260         struct net_device *netdev = adapter->netdev;
261         irq_handler_t handler = e1000_intr;
262         int irq_flags = IRQF_SHARED;
263         int err;
264
265         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
266                           netdev);
267         if (err) {
268                 DPRINTK(PROBE, ERR,
269                         "Unable to allocate interrupt Error: %d\n", err);
270         }
271
272         return err;
273 }
274
275 static void e1000_free_irq(struct e1000_adapter *adapter)
276 {
277         struct net_device *netdev = adapter->netdev;
278
279         free_irq(adapter->pdev->irq, netdev);
280 }
281
282 /**
283  * e1000_irq_disable - Mask off interrupt generation on the NIC
284  * @adapter: board private structure
285  **/
286
287 static void e1000_irq_disable(struct e1000_adapter *adapter)
288 {
289         struct e1000_hw *hw = &adapter->hw;
290
291         ew32(IMC, ~0);
292         E1000_WRITE_FLUSH();
293         synchronize_irq(adapter->pdev->irq);
294 }
295
296 /**
297  * e1000_irq_enable - Enable default interrupt generation settings
298  * @adapter: board private structure
299  **/
300
301 static void e1000_irq_enable(struct e1000_adapter *adapter)
302 {
303         struct e1000_hw *hw = &adapter->hw;
304
305         ew32(IMS, IMS_ENABLE_MASK);
306         E1000_WRITE_FLUSH();
307 }
308
309 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
310 {
311         struct e1000_hw *hw = &adapter->hw;
312         struct net_device *netdev = adapter->netdev;
313         u16 vid = hw->mng_cookie.vlan_id;
314         u16 old_vid = adapter->mng_vlan_id;
315         if (adapter->vlgrp) {
316                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
317                         if (hw->mng_cookie.status &
318                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
319                                 e1000_vlan_rx_add_vid(netdev, vid);
320                                 adapter->mng_vlan_id = vid;
321                         } else
322                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
323
324                         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
325                                         (vid != old_vid) &&
326                             !vlan_group_get_device(adapter->vlgrp, old_vid))
327                                 e1000_vlan_rx_kill_vid(netdev, old_vid);
328                 } else
329                         adapter->mng_vlan_id = vid;
330         }
331 }
332
333 static void e1000_init_manageability(struct e1000_adapter *adapter)
334 {
335         struct e1000_hw *hw = &adapter->hw;
336
337         if (adapter->en_mng_pt) {
338                 u32 manc = er32(MANC);
339
340                 /* disable hardware interception of ARP */
341                 manc &= ~(E1000_MANC_ARP_EN);
342
343                 ew32(MANC, manc);
344         }
345 }
346
347 static void e1000_release_manageability(struct e1000_adapter *adapter)
348 {
349         struct e1000_hw *hw = &adapter->hw;
350
351         if (adapter->en_mng_pt) {
352                 u32 manc = er32(MANC);
353
354                 /* re-enable hardware interception of ARP */
355                 manc |= E1000_MANC_ARP_EN;
356
357                 ew32(MANC, manc);
358         }
359 }
360
361 /**
362  * e1000_configure - configure the hardware for RX and TX
363  * @adapter = private board structure
364  **/
365 static void e1000_configure(struct e1000_adapter *adapter)
366 {
367         struct net_device *netdev = adapter->netdev;
368         int i;
369
370         e1000_set_rx_mode(netdev);
371
372         e1000_restore_vlan(adapter);
373         e1000_init_manageability(adapter);
374
375         e1000_configure_tx(adapter);
376         e1000_setup_rctl(adapter);
377         e1000_configure_rx(adapter);
378         /* call E1000_DESC_UNUSED which always leaves
379          * at least 1 descriptor unused to make sure
380          * next_to_use != next_to_clean */
381         for (i = 0; i < adapter->num_rx_queues; i++) {
382                 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
383                 adapter->alloc_rx_buf(adapter, ring,
384                                       E1000_DESC_UNUSED(ring));
385         }
386
387         adapter->tx_queue_len = netdev->tx_queue_len;
388 }
389
390 int e1000_up(struct e1000_adapter *adapter)
391 {
392         struct e1000_hw *hw = &adapter->hw;
393
394         /* hardware has been reset, we need to reload some things */
395         e1000_configure(adapter);
396
397         clear_bit(__E1000_DOWN, &adapter->flags);
398
399         napi_enable(&adapter->napi);
400
401         e1000_irq_enable(adapter);
402
403         netif_wake_queue(adapter->netdev);
404
405         /* fire a link change interrupt to start the watchdog */
406         ew32(ICS, E1000_ICS_LSC);
407         return 0;
408 }
409
410 /**
411  * e1000_power_up_phy - restore link in case the phy was powered down
412  * @adapter: address of board private structure
413  *
414  * The phy may be powered down to save power and turn off link when the
415  * driver is unloaded and wake on lan is not enabled (among others)
416  * *** this routine MUST be followed by a call to e1000_reset ***
417  *
418  **/
419
420 void e1000_power_up_phy(struct e1000_adapter *adapter)
421 {
422         struct e1000_hw *hw = &adapter->hw;
423         u16 mii_reg = 0;
424
425         /* Just clear the power down bit to wake the phy back up */
426         if (hw->media_type == e1000_media_type_copper) {
427                 /* according to the manual, the phy will retain its
428                  * settings across a power-down/up cycle */
429                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
430                 mii_reg &= ~MII_CR_POWER_DOWN;
431                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
432         }
433 }
434
435 static void e1000_power_down_phy(struct e1000_adapter *adapter)
436 {
437         struct e1000_hw *hw = &adapter->hw;
438
439         /* Power down the PHY so no link is implied when interface is down *
440          * The PHY cannot be powered down if any of the following is true *
441          * (a) WoL is enabled
442          * (b) AMT is active
443          * (c) SoL/IDER session is active */
444         if (!adapter->wol && hw->mac_type >= e1000_82540 &&
445            hw->media_type == e1000_media_type_copper) {
446                 u16 mii_reg = 0;
447
448                 switch (hw->mac_type) {
449                 case e1000_82540:
450                 case e1000_82545:
451                 case e1000_82545_rev_3:
452                 case e1000_82546:
453                 case e1000_82546_rev_3:
454                 case e1000_82541:
455                 case e1000_82541_rev_2:
456                 case e1000_82547:
457                 case e1000_82547_rev_2:
458                         if (er32(MANC) & E1000_MANC_SMBUS_EN)
459                                 goto out;
460                         break;
461                 default:
462                         goto out;
463                 }
464                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
465                 mii_reg |= MII_CR_POWER_DOWN;
466                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
467                 mdelay(1);
468         }
469 out:
470         return;
471 }
472
473 void e1000_down(struct e1000_adapter *adapter)
474 {
475         struct e1000_hw *hw = &adapter->hw;
476         struct net_device *netdev = adapter->netdev;
477         u32 rctl, tctl;
478
479         /* signal that we're down so the interrupt handler does not
480          * reschedule our watchdog timer */
481         set_bit(__E1000_DOWN, &adapter->flags);
482
483         /* disable receives in the hardware */
484         rctl = er32(RCTL);
485         ew32(RCTL, rctl & ~E1000_RCTL_EN);
486         /* flush and sleep below */
487
488         netif_tx_disable(netdev);
489
490         /* disable transmits in the hardware */
491         tctl = er32(TCTL);
492         tctl &= ~E1000_TCTL_EN;
493         ew32(TCTL, tctl);
494         /* flush both disables and wait for them to finish */
495         E1000_WRITE_FLUSH();
496         msleep(10);
497
498         napi_disable(&adapter->napi);
499
500         e1000_irq_disable(adapter);
501
502         del_timer_sync(&adapter->tx_fifo_stall_timer);
503         del_timer_sync(&adapter->watchdog_timer);
504         del_timer_sync(&adapter->phy_info_timer);
505
506         netdev->tx_queue_len = adapter->tx_queue_len;
507         adapter->link_speed = 0;
508         adapter->link_duplex = 0;
509         netif_carrier_off(netdev);
510
511         e1000_reset(adapter);
512         e1000_clean_all_tx_rings(adapter);
513         e1000_clean_all_rx_rings(adapter);
514 }
515
516 void e1000_reinit_locked(struct e1000_adapter *adapter)
517 {
518         WARN_ON(in_interrupt());
519         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
520                 msleep(1);
521         e1000_down(adapter);
522         e1000_up(adapter);
523         clear_bit(__E1000_RESETTING, &adapter->flags);
524 }
525
526 void e1000_reset(struct e1000_adapter *adapter)
527 {
528         struct e1000_hw *hw = &adapter->hw;
529         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
530         bool legacy_pba_adjust = false;
531         u16 hwm;
532
533         /* Repartition Pba for greater than 9k mtu
534          * To take effect CTRL.RST is required.
535          */
536
537         switch (hw->mac_type) {
538         case e1000_82542_rev2_0:
539         case e1000_82542_rev2_1:
540         case e1000_82543:
541         case e1000_82544:
542         case e1000_82540:
543         case e1000_82541:
544         case e1000_82541_rev_2:
545                 legacy_pba_adjust = true;
546                 pba = E1000_PBA_48K;
547                 break;
548         case e1000_82545:
549         case e1000_82545_rev_3:
550         case e1000_82546:
551         case e1000_82546_rev_3:
552                 pba = E1000_PBA_48K;
553                 break;
554         case e1000_82547:
555         case e1000_82547_rev_2:
556                 legacy_pba_adjust = true;
557                 pba = E1000_PBA_30K;
558                 break;
559         case e1000_undefined:
560         case e1000_num_macs:
561                 break;
562         }
563
564         if (legacy_pba_adjust) {
565                 if (hw->max_frame_size > E1000_RXBUFFER_8192)
566                         pba -= 8; /* allocate more FIFO for Tx */
567
568                 if (hw->mac_type == e1000_82547) {
569                         adapter->tx_fifo_head = 0;
570                         adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
571                         adapter->tx_fifo_size =
572                                 (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
573                         atomic_set(&adapter->tx_fifo_stall, 0);
574                 }
575         } else if (hw->max_frame_size >  ETH_FRAME_LEN + ETH_FCS_LEN) {
576                 /* adjust PBA for jumbo frames */
577                 ew32(PBA, pba);
578
579                 /* To maintain wire speed transmits, the Tx FIFO should be
580                  * large enough to accommodate two full transmit packets,
581                  * rounded up to the next 1KB and expressed in KB.  Likewise,
582                  * the Rx FIFO should be large enough to accommodate at least
583                  * one full receive packet and is similarly rounded up and
584                  * expressed in KB. */
585                 pba = er32(PBA);
586                 /* upper 16 bits has Tx packet buffer allocation size in KB */
587                 tx_space = pba >> 16;
588                 /* lower 16 bits has Rx packet buffer allocation size in KB */
589                 pba &= 0xffff;
590                 /*
591                  * the tx fifo also stores 16 bytes of information about the tx
592                  * but don't include ethernet FCS because hardware appends it
593                  */
594                 min_tx_space = (hw->max_frame_size +
595                                 sizeof(struct e1000_tx_desc) -
596                                 ETH_FCS_LEN) * 2;
597                 min_tx_space = ALIGN(min_tx_space, 1024);
598                 min_tx_space >>= 10;
599                 /* software strips receive CRC, so leave room for it */
600                 min_rx_space = hw->max_frame_size;
601                 min_rx_space = ALIGN(min_rx_space, 1024);
602                 min_rx_space >>= 10;
603
604                 /* If current Tx allocation is less than the min Tx FIFO size,
605                  * and the min Tx FIFO size is less than the current Rx FIFO
606                  * allocation, take space away from current Rx allocation */
607                 if (tx_space < min_tx_space &&
608                     ((min_tx_space - tx_space) < pba)) {
609                         pba = pba - (min_tx_space - tx_space);
610
611                         /* PCI/PCIx hardware has PBA alignment constraints */
612                         switch (hw->mac_type) {
613                         case e1000_82545 ... e1000_82546_rev_3:
614                                 pba &= ~(E1000_PBA_8K - 1);
615                                 break;
616                         default:
617                                 break;
618                         }
619
620                         /* if short on rx space, rx wins and must trump tx
621                          * adjustment or use Early Receive if available */
622                         if (pba < min_rx_space)
623                                 pba = min_rx_space;
624                 }
625         }
626
627         ew32(PBA, pba);
628
629         /*
630          * flow control settings:
631          * The high water mark must be low enough to fit one full frame
632          * (or the size used for early receive) above it in the Rx FIFO.
633          * Set it to the lower of:
634          * - 90% of the Rx FIFO size, and
635          * - the full Rx FIFO size minus the early receive size (for parts
636          *   with ERT support assuming ERT set to E1000_ERT_2048), or
637          * - the full Rx FIFO size minus one full frame
638          */
639         hwm = min(((pba << 10) * 9 / 10),
640                   ((pba << 10) - hw->max_frame_size));
641
642         hw->fc_high_water = hwm & 0xFFF8;       /* 8-byte granularity */
643         hw->fc_low_water = hw->fc_high_water - 8;
644         hw->fc_pause_time = E1000_FC_PAUSE_TIME;
645         hw->fc_send_xon = 1;
646         hw->fc = hw->original_fc;
647
648         /* Allow time for pending master requests to run */
649         e1000_reset_hw(hw);
650         if (hw->mac_type >= e1000_82544)
651                 ew32(WUC, 0);
652
653         if (e1000_init_hw(hw))
654                 DPRINTK(PROBE, ERR, "Hardware Error\n");
655         e1000_update_mng_vlan(adapter);
656
657         /* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
658         if (hw->mac_type >= e1000_82544 &&
659             hw->autoneg == 1 &&
660             hw->autoneg_advertised == ADVERTISE_1000_FULL) {
661                 u32 ctrl = er32(CTRL);
662                 /* clear phy power management bit if we are in gig only mode,
663                  * which if enabled will attempt negotiation to 100Mb, which
664                  * can cause a loss of link at power off or driver unload */
665                 ctrl &= ~E1000_CTRL_SWDPIN3;
666                 ew32(CTRL, ctrl);
667         }
668
669         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
670         ew32(VET, ETHERNET_IEEE_VLAN_TYPE);
671
672         e1000_reset_adaptive(hw);
673         e1000_phy_get_info(hw, &adapter->phy_info);
674
675         e1000_release_manageability(adapter);
676 }
677
678 /**
679  *  Dump the eeprom for users having checksum issues
680  **/
681 static void e1000_dump_eeprom(struct e1000_adapter *adapter)
682 {
683         struct net_device *netdev = adapter->netdev;
684         struct ethtool_eeprom eeprom;
685         const struct ethtool_ops *ops = netdev->ethtool_ops;
686         u8 *data;
687         int i;
688         u16 csum_old, csum_new = 0;
689
690         eeprom.len = ops->get_eeprom_len(netdev);
691         eeprom.offset = 0;
692
693         data = kmalloc(eeprom.len, GFP_KERNEL);
694         if (!data) {
695                 printk(KERN_ERR "Unable to allocate memory to dump EEPROM"
696                        " data\n");
697                 return;
698         }
699
700         ops->get_eeprom(netdev, &eeprom, data);
701
702         csum_old = (data[EEPROM_CHECKSUM_REG * 2]) +
703                    (data[EEPROM_CHECKSUM_REG * 2 + 1] << 8);
704         for (i = 0; i < EEPROM_CHECKSUM_REG * 2; i += 2)
705                 csum_new += data[i] + (data[i + 1] << 8);
706         csum_new = EEPROM_SUM - csum_new;
707
708         printk(KERN_ERR "/*********************/\n");
709         printk(KERN_ERR "Current EEPROM Checksum : 0x%04x\n", csum_old);
710         printk(KERN_ERR "Calculated              : 0x%04x\n", csum_new);
711
712         printk(KERN_ERR "Offset    Values\n");
713         printk(KERN_ERR "========  ======\n");
714         print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 16, 1, data, 128, 0);
715
716         printk(KERN_ERR "Include this output when contacting your support "
717                "provider.\n");
718         printk(KERN_ERR "This is not a software error! Something bad "
719                "happened to your hardware or\n");
720         printk(KERN_ERR "EEPROM image. Ignoring this "
721                "problem could result in further problems,\n");
722         printk(KERN_ERR "possibly loss of data, corruption or system hangs!\n");
723         printk(KERN_ERR "The MAC Address will be reset to 00:00:00:00:00:00, "
724                "which is invalid\n");
725         printk(KERN_ERR "and requires you to set the proper MAC "
726                "address manually before continuing\n");
727         printk(KERN_ERR "to enable this network device.\n");
728         printk(KERN_ERR "Please inspect the EEPROM dump and report the issue "
729                "to your hardware vendor\n");
730         printk(KERN_ERR "or Intel Customer Support.\n");
731         printk(KERN_ERR "/*********************/\n");
732
733         kfree(data);
734 }
735
736 /**
737  * e1000_is_need_ioport - determine if an adapter needs ioport resources or not
738  * @pdev: PCI device information struct
739  *
740  * Return true if an adapter needs ioport resources
741  **/
742 static int e1000_is_need_ioport(struct pci_dev *pdev)
743 {
744         switch (pdev->device) {
745         case E1000_DEV_ID_82540EM:
746         case E1000_DEV_ID_82540EM_LOM:
747         case E1000_DEV_ID_82540EP:
748         case E1000_DEV_ID_82540EP_LOM:
749         case E1000_DEV_ID_82540EP_LP:
750         case E1000_DEV_ID_82541EI:
751         case E1000_DEV_ID_82541EI_MOBILE:
752         case E1000_DEV_ID_82541ER:
753         case E1000_DEV_ID_82541ER_LOM:
754         case E1000_DEV_ID_82541GI:
755         case E1000_DEV_ID_82541GI_LF:
756         case E1000_DEV_ID_82541GI_MOBILE:
757         case E1000_DEV_ID_82544EI_COPPER:
758         case E1000_DEV_ID_82544EI_FIBER:
759         case E1000_DEV_ID_82544GC_COPPER:
760         case E1000_DEV_ID_82544GC_LOM:
761         case E1000_DEV_ID_82545EM_COPPER:
762         case E1000_DEV_ID_82545EM_FIBER:
763         case E1000_DEV_ID_82546EB_COPPER:
764         case E1000_DEV_ID_82546EB_FIBER:
765         case E1000_DEV_ID_82546EB_QUAD_COPPER:
766                 return true;
767         default:
768                 return false;
769         }
770 }
771
772 static const struct net_device_ops e1000_netdev_ops = {
773         .ndo_open               = e1000_open,
774         .ndo_stop               = e1000_close,
775         .ndo_start_xmit         = e1000_xmit_frame,
776         .ndo_get_stats          = e1000_get_stats,
777         .ndo_set_rx_mode        = e1000_set_rx_mode,
778         .ndo_set_mac_address    = e1000_set_mac,
779         .ndo_tx_timeout         = e1000_tx_timeout,
780         .ndo_change_mtu         = e1000_change_mtu,
781         .ndo_do_ioctl           = e1000_ioctl,
782         .ndo_validate_addr      = eth_validate_addr,
783
784         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
785         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
786         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
787 #ifdef CONFIG_NET_POLL_CONTROLLER
788         .ndo_poll_controller    = e1000_netpoll,
789 #endif
790 };
791
792 /**
793  * e1000_probe - Device Initialization Routine
794  * @pdev: PCI device information struct
795  * @ent: entry in e1000_pci_tbl
796  *
797  * Returns 0 on success, negative on failure
798  *
799  * e1000_probe initializes an adapter identified by a pci_dev structure.
800  * The OS initialization, configuring of the adapter private structure,
801  * and a hardware reset occur.
802  **/
803 static int __devinit e1000_probe(struct pci_dev *pdev,
804                                  const struct pci_device_id *ent)
805 {
806         struct net_device *netdev;
807         struct e1000_adapter *adapter;
808         struct e1000_hw *hw;
809
810         static int cards_found = 0;
811         static int global_quad_port_a = 0; /* global ksp3 port a indication */
812         int i, err, pci_using_dac;
813         u16 eeprom_data = 0;
814         u16 eeprom_apme_mask = E1000_EEPROM_APME;
815         int bars, need_ioport;
816
817         /* do not allocate ioport bars when not needed */
818         need_ioport = e1000_is_need_ioport(pdev);
819         if (need_ioport) {
820                 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
821                 err = pci_enable_device(pdev);
822         } else {
823                 bars = pci_select_bars(pdev, IORESOURCE_MEM);
824                 err = pci_enable_device_mem(pdev);
825         }
826         if (err)
827                 return err;
828
829         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) &&
830             !pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64))) {
831                 pci_using_dac = 1;
832         } else {
833                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
834                 if (err) {
835                         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
836                         if (err) {
837                                 E1000_ERR("No usable DMA configuration, "
838                                           "aborting\n");
839                                 goto err_dma;
840                         }
841                 }
842                 pci_using_dac = 0;
843         }
844
845         err = pci_request_selected_regions(pdev, bars, e1000_driver_name);
846         if (err)
847                 goto err_pci_reg;
848
849         pci_set_master(pdev);
850
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_2048:
1702                 default:
1703                         rctl |= E1000_RCTL_SZ_2048;
1704                         rctl &= ~E1000_RCTL_BSEX;
1705                         break;
1706                 case E1000_RXBUFFER_4096:
1707                         rctl |= E1000_RCTL_SZ_4096;
1708                         break;
1709                 case E1000_RXBUFFER_8192:
1710                         rctl |= E1000_RCTL_SZ_8192;
1711                         break;
1712                 case E1000_RXBUFFER_16384:
1713                         rctl |= E1000_RCTL_SZ_16384;
1714                         break;
1715         }
1716
1717         ew32(RCTL, rctl);
1718 }
1719
1720 /**
1721  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1722  * @adapter: board private structure
1723  *
1724  * Configure the Rx unit of the MAC after a reset.
1725  **/
1726
1727 static void e1000_configure_rx(struct e1000_adapter *adapter)
1728 {
1729         u64 rdba;
1730         struct e1000_hw *hw = &adapter->hw;
1731         u32 rdlen, rctl, rxcsum;
1732
1733         if (adapter->netdev->mtu > ETH_DATA_LEN) {
1734                 rdlen = adapter->rx_ring[0].count *
1735                         sizeof(struct e1000_rx_desc);
1736                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
1737                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
1738         } else {
1739                 rdlen = adapter->rx_ring[0].count *
1740                         sizeof(struct e1000_rx_desc);
1741                 adapter->clean_rx = e1000_clean_rx_irq;
1742                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1743         }
1744
1745         /* disable receives while setting up the descriptors */
1746         rctl = er32(RCTL);
1747         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1748
1749         /* set the Receive Delay Timer Register */
1750         ew32(RDTR, adapter->rx_int_delay);
1751
1752         if (hw->mac_type >= e1000_82540) {
1753                 ew32(RADV, adapter->rx_abs_int_delay);
1754                 if (adapter->itr_setting != 0)
1755                         ew32(ITR, 1000000000 / (adapter->itr * 256));
1756         }
1757
1758         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1759          * the Base and Length of the Rx Descriptor Ring */
1760         switch (adapter->num_rx_queues) {
1761         case 1:
1762         default:
1763                 rdba = adapter->rx_ring[0].dma;
1764                 ew32(RDLEN, rdlen);
1765                 ew32(RDBAH, (rdba >> 32));
1766                 ew32(RDBAL, (rdba & 0x00000000ffffffffULL));
1767                 ew32(RDT, 0);
1768                 ew32(RDH, 0);
1769                 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ? E1000_RDH : E1000_82542_RDH);
1770                 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ? E1000_RDT : E1000_82542_RDT);
1771                 break;
1772         }
1773
1774         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
1775         if (hw->mac_type >= e1000_82543) {
1776                 rxcsum = er32(RXCSUM);
1777                 if (adapter->rx_csum)
1778                         rxcsum |= E1000_RXCSUM_TUOFL;
1779                 else
1780                         /* don't need to clear IPPCSE as it defaults to 0 */
1781                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1782                 ew32(RXCSUM, rxcsum);
1783         }
1784
1785         /* Enable Receives */
1786         ew32(RCTL, rctl);
1787 }
1788
1789 /**
1790  * e1000_free_tx_resources - Free Tx Resources per Queue
1791  * @adapter: board private structure
1792  * @tx_ring: Tx descriptor ring for a specific queue
1793  *
1794  * Free all transmit software resources
1795  **/
1796
1797 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
1798                                     struct e1000_tx_ring *tx_ring)
1799 {
1800         struct pci_dev *pdev = adapter->pdev;
1801
1802         e1000_clean_tx_ring(adapter, tx_ring);
1803
1804         vfree(tx_ring->buffer_info);
1805         tx_ring->buffer_info = NULL;
1806
1807         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
1808
1809         tx_ring->desc = NULL;
1810 }
1811
1812 /**
1813  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1814  * @adapter: board private structure
1815  *
1816  * Free all transmit software resources
1817  **/
1818
1819 void e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1820 {
1821         int i;
1822
1823         for (i = 0; i < adapter->num_tx_queues; i++)
1824                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1825 }
1826
1827 static void e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1828                                              struct e1000_buffer *buffer_info)
1829 {
1830         if (buffer_info->dma) {
1831                 if (buffer_info->mapped_as_page)
1832                         pci_unmap_page(adapter->pdev, buffer_info->dma,
1833                                        buffer_info->length, PCI_DMA_TODEVICE);
1834                 else
1835                         pci_unmap_single(adapter->pdev, buffer_info->dma,
1836                                          buffer_info->length,
1837                                          PCI_DMA_TODEVICE);
1838                 buffer_info->dma = 0;
1839         }
1840         if (buffer_info->skb) {
1841                 dev_kfree_skb_any(buffer_info->skb);
1842                 buffer_info->skb = NULL;
1843         }
1844         buffer_info->time_stamp = 0;
1845         /* buffer_info must be completely set up in the transmit path */
1846 }
1847
1848 /**
1849  * e1000_clean_tx_ring - Free Tx Buffers
1850  * @adapter: board private structure
1851  * @tx_ring: ring to be cleaned
1852  **/
1853
1854 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
1855                                 struct e1000_tx_ring *tx_ring)
1856 {
1857         struct e1000_hw *hw = &adapter->hw;
1858         struct e1000_buffer *buffer_info;
1859         unsigned long size;
1860         unsigned int i;
1861
1862         /* Free all the Tx ring sk_buffs */
1863
1864         for (i = 0; i < tx_ring->count; i++) {
1865                 buffer_info = &tx_ring->buffer_info[i];
1866                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
1867         }
1868
1869         size = sizeof(struct e1000_buffer) * tx_ring->count;
1870         memset(tx_ring->buffer_info, 0, size);
1871
1872         /* Zero out the descriptor ring */
1873
1874         memset(tx_ring->desc, 0, tx_ring->size);
1875
1876         tx_ring->next_to_use = 0;
1877         tx_ring->next_to_clean = 0;
1878         tx_ring->last_tx_tso = 0;
1879
1880         writel(0, hw->hw_addr + tx_ring->tdh);
1881         writel(0, hw->hw_addr + tx_ring->tdt);
1882 }
1883
1884 /**
1885  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
1886  * @adapter: board private structure
1887  **/
1888
1889 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
1890 {
1891         int i;
1892
1893         for (i = 0; i < adapter->num_tx_queues; i++)
1894                 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1895 }
1896
1897 /**
1898  * e1000_free_rx_resources - Free Rx Resources
1899  * @adapter: board private structure
1900  * @rx_ring: ring to clean the resources from
1901  *
1902  * Free all receive software resources
1903  **/
1904
1905 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
1906                                     struct e1000_rx_ring *rx_ring)
1907 {
1908         struct pci_dev *pdev = adapter->pdev;
1909
1910         e1000_clean_rx_ring(adapter, rx_ring);
1911
1912         vfree(rx_ring->buffer_info);
1913         rx_ring->buffer_info = NULL;
1914
1915         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
1916
1917         rx_ring->desc = NULL;
1918 }
1919
1920 /**
1921  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
1922  * @adapter: board private structure
1923  *
1924  * Free all receive software resources
1925  **/
1926
1927 void e1000_free_all_rx_resources(struct e1000_adapter *adapter)
1928 {
1929         int i;
1930
1931         for (i = 0; i < adapter->num_rx_queues; i++)
1932                 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
1933 }
1934
1935 /**
1936  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1937  * @adapter: board private structure
1938  * @rx_ring: ring to free buffers from
1939  **/
1940
1941 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
1942                                 struct e1000_rx_ring *rx_ring)
1943 {
1944         struct e1000_hw *hw = &adapter->hw;
1945         struct e1000_buffer *buffer_info;
1946         struct pci_dev *pdev = adapter->pdev;
1947         unsigned long size;
1948         unsigned int i;
1949
1950         /* Free all the Rx ring sk_buffs */
1951         for (i = 0; i < rx_ring->count; i++) {
1952                 buffer_info = &rx_ring->buffer_info[i];
1953                 if (buffer_info->dma &&
1954                     adapter->clean_rx == e1000_clean_rx_irq) {
1955                         pci_unmap_single(pdev, buffer_info->dma,
1956                                          buffer_info->length,
1957                                          PCI_DMA_FROMDEVICE);
1958                 } else if (buffer_info->dma &&
1959                            adapter->clean_rx == e1000_clean_jumbo_rx_irq) {
1960                         pci_unmap_page(pdev, buffer_info->dma,
1961                                        buffer_info->length,
1962                                        PCI_DMA_FROMDEVICE);
1963                 }
1964
1965                 buffer_info->dma = 0;
1966                 if (buffer_info->page) {
1967                         put_page(buffer_info->page);
1968                         buffer_info->page = NULL;
1969                 }
1970                 if (buffer_info->skb) {
1971                         dev_kfree_skb(buffer_info->skb);
1972                         buffer_info->skb = NULL;
1973                 }
1974         }
1975
1976         /* there also may be some cached data from a chained receive */
1977         if (rx_ring->rx_skb_top) {
1978                 dev_kfree_skb(rx_ring->rx_skb_top);
1979                 rx_ring->rx_skb_top = NULL;
1980         }
1981
1982         size = sizeof(struct e1000_buffer) * rx_ring->count;
1983         memset(rx_ring->buffer_info, 0, size);
1984
1985         /* Zero out the descriptor ring */
1986         memset(rx_ring->desc, 0, rx_ring->size);
1987
1988         rx_ring->next_to_clean = 0;
1989         rx_ring->next_to_use = 0;
1990
1991         writel(0, hw->hw_addr + rx_ring->rdh);
1992         writel(0, hw->hw_addr + rx_ring->rdt);
1993 }
1994
1995 /**
1996  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
1997  * @adapter: board private structure
1998  **/
1999
2000 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2001 {
2002         int i;
2003
2004         for (i = 0; i < adapter->num_rx_queues; i++)
2005                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2006 }
2007
2008 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2009  * and memory write and invalidate disabled for certain operations
2010  */
2011 static void e1000_enter_82542_rst(struct e1000_adapter *adapter)
2012 {
2013         struct e1000_hw *hw = &adapter->hw;
2014         struct net_device *netdev = adapter->netdev;
2015         u32 rctl;
2016
2017         e1000_pci_clear_mwi(hw);
2018
2019         rctl = er32(RCTL);
2020         rctl |= E1000_RCTL_RST;
2021         ew32(RCTL, rctl);
2022         E1000_WRITE_FLUSH();
2023         mdelay(5);
2024
2025         if (netif_running(netdev))
2026                 e1000_clean_all_rx_rings(adapter);
2027 }
2028
2029 static void e1000_leave_82542_rst(struct e1000_adapter *adapter)
2030 {
2031         struct e1000_hw *hw = &adapter->hw;
2032         struct net_device *netdev = adapter->netdev;
2033         u32 rctl;
2034
2035         rctl = er32(RCTL);
2036         rctl &= ~E1000_RCTL_RST;
2037         ew32(RCTL, rctl);
2038         E1000_WRITE_FLUSH();
2039         mdelay(5);
2040
2041         if (hw->pci_cmd_word & PCI_COMMAND_INVALIDATE)
2042                 e1000_pci_set_mwi(hw);
2043
2044         if (netif_running(netdev)) {
2045                 /* No need to loop, because 82542 supports only 1 queue */
2046                 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2047                 e1000_configure_rx(adapter);
2048                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2049         }
2050 }
2051
2052 /**
2053  * e1000_set_mac - Change the Ethernet Address of the NIC
2054  * @netdev: network interface device structure
2055  * @p: pointer to an address structure
2056  *
2057  * Returns 0 on success, negative on failure
2058  **/
2059
2060 static int e1000_set_mac(struct net_device *netdev, void *p)
2061 {
2062         struct e1000_adapter *adapter = netdev_priv(netdev);
2063         struct e1000_hw *hw = &adapter->hw;
2064         struct sockaddr *addr = p;
2065
2066         if (!is_valid_ether_addr(addr->sa_data))
2067                 return -EADDRNOTAVAIL;
2068
2069         /* 82542 2.0 needs to be in reset to write receive address registers */
2070
2071         if (hw->mac_type == e1000_82542_rev2_0)
2072                 e1000_enter_82542_rst(adapter);
2073
2074         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2075         memcpy(hw->mac_addr, addr->sa_data, netdev->addr_len);
2076
2077         e1000_rar_set(hw, hw->mac_addr, 0);
2078
2079         if (hw->mac_type == e1000_82542_rev2_0)
2080                 e1000_leave_82542_rst(adapter);
2081
2082         return 0;
2083 }
2084
2085 /**
2086  * e1000_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2087  * @netdev: network interface device structure
2088  *
2089  * The set_rx_mode entry point is called whenever the unicast or multicast
2090  * address lists or the network interface flags are updated. This routine is
2091  * responsible for configuring the hardware for proper unicast, multicast,
2092  * promiscuous mode, and all-multi behavior.
2093  **/
2094
2095 static void e1000_set_rx_mode(struct net_device *netdev)
2096 {
2097         struct e1000_adapter *adapter = netdev_priv(netdev);
2098         struct e1000_hw *hw = &adapter->hw;
2099         struct netdev_hw_addr *ha;
2100         bool use_uc = false;
2101         struct dev_addr_list *mc_ptr;
2102         u32 rctl;
2103         u32 hash_value;
2104         int i, rar_entries = E1000_RAR_ENTRIES;
2105         int mta_reg_count = E1000_NUM_MTA_REGISTERS;
2106         u32 *mcarray = kcalloc(mta_reg_count, sizeof(u32), GFP_ATOMIC);
2107
2108         if (!mcarray) {
2109                 DPRINTK(PROBE, ERR, "memory allocation failed\n");
2110                 return;
2111         }
2112
2113         /* Check for Promiscuous and All Multicast modes */
2114
2115         rctl = er32(RCTL);
2116
2117         if (netdev->flags & IFF_PROMISC) {
2118                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2119                 rctl &= ~E1000_RCTL_VFE;
2120         } else {
2121                 if (netdev->flags & IFF_ALLMULTI)
2122                         rctl |= E1000_RCTL_MPE;
2123                 else
2124                         rctl &= ~E1000_RCTL_MPE;
2125                 /* Enable VLAN filter if there is a VLAN */
2126                 if (adapter->vlgrp)
2127                         rctl |= E1000_RCTL_VFE;
2128         }
2129
2130         if (netdev_uc_count(netdev) > rar_entries - 1) {
2131                 rctl |= E1000_RCTL_UPE;
2132         } else if (!(netdev->flags & IFF_PROMISC)) {
2133                 rctl &= ~E1000_RCTL_UPE;
2134                 use_uc = true;
2135         }
2136
2137         ew32(RCTL, rctl);
2138
2139         /* 82542 2.0 needs to be in reset to write receive address registers */
2140
2141         if (hw->mac_type == e1000_82542_rev2_0)
2142                 e1000_enter_82542_rst(adapter);
2143
2144         /* load the first 14 addresses into the exact filters 1-14. Unicast
2145          * addresses take precedence to avoid disabling unicast filtering
2146          * when possible.
2147          *
2148          * RAR 0 is used for the station MAC adddress
2149          * if there are not 14 addresses, go ahead and clear the filters
2150          */
2151         i = 1;
2152         if (use_uc)
2153                 netdev_for_each_uc_addr(ha, netdev) {
2154                         if (i == rar_entries)
2155                                 break;
2156                         e1000_rar_set(hw, ha->addr, i++);
2157                 }
2158
2159         WARN_ON(i == rar_entries);
2160
2161         mc_ptr = netdev->mc_list;
2162
2163         for (; i < rar_entries; i++) {
2164                 if (mc_ptr) {
2165                         e1000_rar_set(hw, mc_ptr->da_addr, i);
2166                         mc_ptr = mc_ptr->next;
2167                 } else {
2168                         E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2169                         E1000_WRITE_FLUSH();
2170                         E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2171                         E1000_WRITE_FLUSH();
2172                 }
2173         }
2174
2175         /* load any remaining addresses into the hash table */
2176
2177         for (; mc_ptr; mc_ptr = mc_ptr->next) {
2178                 u32 hash_reg, hash_bit, mta;
2179                 hash_value = e1000_hash_mc_addr(hw, mc_ptr->da_addr);
2180                 hash_reg = (hash_value >> 5) & 0x7F;
2181                 hash_bit = hash_value & 0x1F;
2182                 mta = (1 << hash_bit);
2183                 mcarray[hash_reg] |= mta;
2184         }
2185
2186         /* write the hash table completely, write from bottom to avoid
2187          * both stupid write combining chipsets, and flushing each write */
2188         for (i = mta_reg_count - 1; i >= 0 ; i--) {
2189                 /*
2190                  * If we are on an 82544 has an errata where writing odd
2191                  * offsets overwrites the previous even offset, but writing
2192                  * backwards over the range solves the issue by always
2193                  * writing the odd offset first
2194                  */
2195                 E1000_WRITE_REG_ARRAY(hw, MTA, i, mcarray[i]);
2196         }
2197         E1000_WRITE_FLUSH();
2198
2199         if (hw->mac_type == e1000_82542_rev2_0)
2200                 e1000_leave_82542_rst(adapter);
2201
2202         kfree(mcarray);
2203 }
2204
2205 /* Need to wait a few seconds after link up to get diagnostic information from
2206  * the phy */
2207
2208 static void e1000_update_phy_info(unsigned long data)
2209 {
2210         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2211         struct e1000_hw *hw = &adapter->hw;
2212         e1000_phy_get_info(hw, &adapter->phy_info);
2213 }
2214
2215 /**
2216  * e1000_82547_tx_fifo_stall - Timer Call-back
2217  * @data: pointer to adapter cast into an unsigned long
2218  **/
2219
2220 static void e1000_82547_tx_fifo_stall(unsigned long data)
2221 {
2222         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2223         struct e1000_hw *hw = &adapter->hw;
2224         struct net_device *netdev = adapter->netdev;
2225         u32 tctl;
2226
2227         if (atomic_read(&adapter->tx_fifo_stall)) {
2228                 if ((er32(TDT) == er32(TDH)) &&
2229                    (er32(TDFT) == er32(TDFH)) &&
2230                    (er32(TDFTS) == er32(TDFHS))) {
2231                         tctl = er32(TCTL);
2232                         ew32(TCTL, tctl & ~E1000_TCTL_EN);
2233                         ew32(TDFT, adapter->tx_head_addr);
2234                         ew32(TDFH, adapter->tx_head_addr);
2235                         ew32(TDFTS, adapter->tx_head_addr);
2236                         ew32(TDFHS, adapter->tx_head_addr);
2237                         ew32(TCTL, tctl);
2238                         E1000_WRITE_FLUSH();
2239
2240                         adapter->tx_fifo_head = 0;
2241                         atomic_set(&adapter->tx_fifo_stall, 0);
2242                         netif_wake_queue(netdev);
2243                 } else if (!test_bit(__E1000_DOWN, &adapter->flags)) {
2244                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2245                 }
2246         }
2247 }
2248
2249 static bool e1000_has_link(struct e1000_adapter *adapter)
2250 {
2251         struct e1000_hw *hw = &adapter->hw;
2252         bool link_active = false;
2253
2254         /* get_link_status is set on LSC (link status) interrupt or
2255          * rx sequence error interrupt.  get_link_status will stay
2256          * false until the e1000_check_for_link establishes link
2257          * for copper adapters ONLY
2258          */
2259         switch (hw->media_type) {
2260         case e1000_media_type_copper:
2261                 if (hw->get_link_status) {
2262                         e1000_check_for_link(hw);
2263                         link_active = !hw->get_link_status;
2264                 } else {
2265                         link_active = true;
2266                 }
2267                 break;
2268         case e1000_media_type_fiber:
2269                 e1000_check_for_link(hw);
2270                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2271                 break;
2272         case e1000_media_type_internal_serdes:
2273                 e1000_check_for_link(hw);
2274                 link_active = hw->serdes_has_link;
2275                 break;
2276         default:
2277                 break;
2278         }
2279
2280         return link_active;
2281 }
2282
2283 /**
2284  * e1000_watchdog - Timer Call-back
2285  * @data: pointer to adapter cast into an unsigned long
2286  **/
2287 static void e1000_watchdog(unsigned long data)
2288 {
2289         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2290         struct e1000_hw *hw = &adapter->hw;
2291         struct net_device *netdev = adapter->netdev;
2292         struct e1000_tx_ring *txdr = adapter->tx_ring;
2293         u32 link, tctl;
2294
2295         link = e1000_has_link(adapter);
2296         if ((netif_carrier_ok(netdev)) && link)
2297                 goto link_up;
2298
2299         if (link) {
2300                 if (!netif_carrier_ok(netdev)) {
2301                         u32 ctrl;
2302                         bool txb2b = true;
2303                         /* update snapshot of PHY registers on LSC */
2304                         e1000_get_speed_and_duplex(hw,
2305                                                    &adapter->link_speed,
2306                                                    &adapter->link_duplex);
2307
2308                         ctrl = er32(CTRL);
2309                         printk(KERN_INFO "e1000: %s NIC Link is Up %d Mbps %s, "
2310                                "Flow Control: %s\n",
2311                                netdev->name,
2312                                adapter->link_speed,
2313                                adapter->link_duplex == FULL_DUPLEX ?
2314                                 "Full Duplex" : "Half Duplex",
2315                                 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2316                                 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2317                                 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2318                                 E1000_CTRL_TFCE) ? "TX" : "None" )));
2319
2320                         /* tweak tx_queue_len according to speed/duplex
2321                          * and adjust the timeout factor */
2322                         netdev->tx_queue_len = adapter->tx_queue_len;
2323                         adapter->tx_timeout_factor = 1;
2324                         switch (adapter->link_speed) {
2325                         case SPEED_10:
2326                                 txb2b = false;
2327                                 netdev->tx_queue_len = 10;
2328                                 adapter->tx_timeout_factor = 16;
2329                                 break;
2330                         case SPEED_100:
2331                                 txb2b = false;
2332                                 netdev->tx_queue_len = 100;
2333                                 /* maybe add some timeout factor ? */
2334                                 break;
2335                         }
2336
2337                         /* enable transmits in the hardware */
2338                         tctl = er32(TCTL);
2339                         tctl |= E1000_TCTL_EN;
2340                         ew32(TCTL, tctl);
2341
2342                         netif_carrier_on(netdev);
2343                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2344                                 mod_timer(&adapter->phy_info_timer,
2345                                           round_jiffies(jiffies + 2 * HZ));
2346                         adapter->smartspeed = 0;
2347                 }
2348         } else {
2349                 if (netif_carrier_ok(netdev)) {
2350                         adapter->link_speed = 0;
2351                         adapter->link_duplex = 0;
2352                         printk(KERN_INFO "e1000: %s NIC Link is Down\n",
2353                                netdev->name);
2354                         netif_carrier_off(netdev);
2355
2356                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2357                                 mod_timer(&adapter->phy_info_timer,
2358                                           round_jiffies(jiffies + 2 * HZ));
2359                 }
2360
2361                 e1000_smartspeed(adapter);
2362         }
2363
2364 link_up:
2365         e1000_update_stats(adapter);
2366
2367         hw->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2368         adapter->tpt_old = adapter->stats.tpt;
2369         hw->collision_delta = adapter->stats.colc - adapter->colc_old;
2370         adapter->colc_old = adapter->stats.colc;
2371
2372         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2373         adapter->gorcl_old = adapter->stats.gorcl;
2374         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2375         adapter->gotcl_old = adapter->stats.gotcl;
2376
2377         e1000_update_adaptive(hw);
2378
2379         if (!netif_carrier_ok(netdev)) {
2380                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2381                         /* We've lost link, so the controller stops DMA,
2382                          * but we've got queued Tx work that's never going
2383                          * to get done, so reset controller to flush Tx.
2384                          * (Do the reset outside of interrupt context). */
2385                         adapter->tx_timeout_count++;
2386                         schedule_work(&adapter->reset_task);
2387                         /* return immediately since reset is imminent */
2388                         return;
2389                 }
2390         }
2391
2392         /* Cause software interrupt to ensure rx ring is cleaned */
2393         ew32(ICS, E1000_ICS_RXDMT0);
2394
2395         /* Force detection of hung controller every watchdog period */
2396         adapter->detect_tx_hung = true;
2397
2398         /* Reset the timer */
2399         if (!test_bit(__E1000_DOWN, &adapter->flags))
2400                 mod_timer(&adapter->watchdog_timer,
2401                           round_jiffies(jiffies + 2 * HZ));
2402 }
2403
2404 enum latency_range {
2405         lowest_latency = 0,
2406         low_latency = 1,
2407         bulk_latency = 2,
2408         latency_invalid = 255
2409 };
2410
2411 /**
2412  * e1000_update_itr - update the dynamic ITR value based on statistics
2413  * @adapter: pointer to adapter
2414  * @itr_setting: current adapter->itr
2415  * @packets: the number of packets during this measurement interval
2416  * @bytes: the number of bytes during this measurement interval
2417  *
2418  *      Stores a new ITR value based on packets and byte
2419  *      counts during the last interrupt.  The advantage of per interrupt
2420  *      computation is faster updates and more accurate ITR for the current
2421  *      traffic pattern.  Constants in this function were computed
2422  *      based on theoretical maximum wire speed and thresholds were set based
2423  *      on testing data as well as attempting to minimize response time
2424  *      while increasing bulk throughput.
2425  *      this functionality is controlled by the InterruptThrottleRate module
2426  *      parameter (see e1000_param.c)
2427  **/
2428 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2429                                      u16 itr_setting, int packets, int bytes)
2430 {
2431         unsigned int retval = itr_setting;
2432         struct e1000_hw *hw = &adapter->hw;
2433
2434         if (unlikely(hw->mac_type < e1000_82540))
2435                 goto update_itr_done;
2436
2437         if (packets == 0)
2438                 goto update_itr_done;
2439
2440         switch (itr_setting) {
2441         case lowest_latency:
2442                 /* jumbo frames get bulk treatment*/
2443                 if (bytes/packets > 8000)
2444                         retval = bulk_latency;
2445                 else if ((packets < 5) && (bytes > 512))
2446                         retval = low_latency;
2447                 break;
2448         case low_latency:  /* 50 usec aka 20000 ints/s */
2449                 if (bytes > 10000) {
2450                         /* jumbo frames need bulk latency setting */
2451                         if (bytes/packets > 8000)
2452                                 retval = bulk_latency;
2453                         else if ((packets < 10) || ((bytes/packets) > 1200))
2454                                 retval = bulk_latency;
2455                         else if ((packets > 35))
2456                                 retval = lowest_latency;
2457                 } else if (bytes/packets > 2000)
2458                         retval = bulk_latency;
2459                 else if (packets <= 2 && bytes < 512)
2460                         retval = lowest_latency;
2461                 break;
2462         case bulk_latency: /* 250 usec aka 4000 ints/s */
2463                 if (bytes > 25000) {
2464                         if (packets > 35)
2465                                 retval = low_latency;
2466                 } else if (bytes < 6000) {
2467                         retval = low_latency;
2468                 }
2469                 break;
2470         }
2471
2472 update_itr_done:
2473         return retval;
2474 }
2475
2476 static void e1000_set_itr(struct e1000_adapter *adapter)
2477 {
2478         struct e1000_hw *hw = &adapter->hw;
2479         u16 current_itr;
2480         u32 new_itr = adapter->itr;
2481
2482         if (unlikely(hw->mac_type < e1000_82540))
2483                 return;
2484
2485         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2486         if (unlikely(adapter->link_speed != SPEED_1000)) {
2487                 current_itr = 0;
2488                 new_itr = 4000;
2489                 goto set_itr_now;
2490         }
2491
2492         adapter->tx_itr = e1000_update_itr(adapter,
2493                                     adapter->tx_itr,
2494                                     adapter->total_tx_packets,
2495                                     adapter->total_tx_bytes);
2496         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2497         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2498                 adapter->tx_itr = low_latency;
2499
2500         adapter->rx_itr = e1000_update_itr(adapter,
2501                                     adapter->rx_itr,
2502                                     adapter->total_rx_packets,
2503                                     adapter->total_rx_bytes);
2504         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2505         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2506                 adapter->rx_itr = low_latency;
2507
2508         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2509
2510         switch (current_itr) {
2511         /* counts and packets in update_itr are dependent on these numbers */
2512         case lowest_latency:
2513                 new_itr = 70000;
2514                 break;
2515         case low_latency:
2516                 new_itr = 20000; /* aka hwitr = ~200 */
2517                 break;
2518         case bulk_latency:
2519                 new_itr = 4000;
2520                 break;
2521         default:
2522                 break;
2523         }
2524
2525 set_itr_now:
2526         if (new_itr != adapter->itr) {
2527                 /* this attempts to bias the interrupt rate towards Bulk
2528                  * by adding intermediate steps when interrupt rate is
2529                  * increasing */
2530                 new_itr = new_itr > adapter->itr ?
2531                              min(adapter->itr + (new_itr >> 2), new_itr) :
2532                              new_itr;
2533                 adapter->itr = new_itr;
2534                 ew32(ITR, 1000000000 / (new_itr * 256));
2535         }
2536
2537         return;
2538 }
2539
2540 #define E1000_TX_FLAGS_CSUM             0x00000001
2541 #define E1000_TX_FLAGS_VLAN             0x00000002
2542 #define E1000_TX_FLAGS_TSO              0x00000004
2543 #define E1000_TX_FLAGS_IPV4             0x00000008
2544 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2545 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2546
2547 static int e1000_tso(struct e1000_adapter *adapter,
2548                      struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2549 {
2550         struct e1000_context_desc *context_desc;
2551         struct e1000_buffer *buffer_info;
2552         unsigned int i;
2553         u32 cmd_length = 0;
2554         u16 ipcse = 0, tucse, mss;
2555         u8 ipcss, ipcso, tucss, tucso, hdr_len;
2556         int err;
2557
2558         if (skb_is_gso(skb)) {
2559                 if (skb_header_cloned(skb)) {
2560                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2561                         if (err)
2562                                 return err;
2563                 }
2564
2565                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2566                 mss = skb_shinfo(skb)->gso_size;
2567                 if (skb->protocol == htons(ETH_P_IP)) {
2568                         struct iphdr *iph = ip_hdr(skb);
2569                         iph->tot_len = 0;
2570                         iph->check = 0;
2571                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2572                                                                  iph->daddr, 0,
2573                                                                  IPPROTO_TCP,
2574                                                                  0);
2575                         cmd_length = E1000_TXD_CMD_IP;
2576                         ipcse = skb_transport_offset(skb) - 1;
2577                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2578                         ipv6_hdr(skb)->payload_len = 0;
2579                         tcp_hdr(skb)->check =
2580                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2581                                                  &ipv6_hdr(skb)->daddr,
2582                                                  0, IPPROTO_TCP, 0);
2583                         ipcse = 0;
2584                 }
2585                 ipcss = skb_network_offset(skb);
2586                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2587                 tucss = skb_transport_offset(skb);
2588                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2589                 tucse = 0;
2590
2591                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2592                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2593
2594                 i = tx_ring->next_to_use;
2595                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2596                 buffer_info = &tx_ring->buffer_info[i];
2597
2598                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2599                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2600                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2601                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2602                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2603                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2604                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2605                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2606                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2607
2608                 buffer_info->time_stamp = jiffies;
2609                 buffer_info->next_to_watch = i;
2610
2611                 if (++i == tx_ring->count) i = 0;
2612                 tx_ring->next_to_use = i;
2613
2614                 return true;
2615         }
2616         return false;
2617 }
2618
2619 static bool e1000_tx_csum(struct e1000_adapter *adapter,
2620                           struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2621 {
2622         struct e1000_context_desc *context_desc;
2623         struct e1000_buffer *buffer_info;
2624         unsigned int i;
2625         u8 css;
2626         u32 cmd_len = E1000_TXD_CMD_DEXT;
2627
2628         if (skb->ip_summed != CHECKSUM_PARTIAL)
2629                 return false;
2630
2631         switch (skb->protocol) {
2632         case cpu_to_be16(ETH_P_IP):
2633                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2634                         cmd_len |= E1000_TXD_CMD_TCP;
2635                 break;
2636         case cpu_to_be16(ETH_P_IPV6):
2637                 /* XXX not handling all IPV6 headers */
2638                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2639                         cmd_len |= E1000_TXD_CMD_TCP;
2640                 break;
2641         default:
2642                 if (unlikely(net_ratelimit()))
2643                         DPRINTK(DRV, WARNING,
2644                                 "checksum_partial proto=%x!\n", skb->protocol);
2645                 break;
2646         }
2647
2648         css = skb_transport_offset(skb);
2649
2650         i = tx_ring->next_to_use;
2651         buffer_info = &tx_ring->buffer_info[i];
2652         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2653
2654         context_desc->lower_setup.ip_config = 0;
2655         context_desc->upper_setup.tcp_fields.tucss = css;
2656         context_desc->upper_setup.tcp_fields.tucso =
2657                 css + skb->csum_offset;
2658         context_desc->upper_setup.tcp_fields.tucse = 0;
2659         context_desc->tcp_seg_setup.data = 0;
2660         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
2661
2662         buffer_info->time_stamp = jiffies;
2663         buffer_info->next_to_watch = i;
2664
2665         if (unlikely(++i == tx_ring->count)) i = 0;
2666         tx_ring->next_to_use = i;
2667
2668         return true;
2669 }
2670
2671 #define E1000_MAX_TXD_PWR       12
2672 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2673
2674 static int e1000_tx_map(struct e1000_adapter *adapter,
2675                         struct e1000_tx_ring *tx_ring,
2676                         struct sk_buff *skb, unsigned int first,
2677                         unsigned int max_per_txd, unsigned int nr_frags,
2678                         unsigned int mss)
2679 {
2680         struct e1000_hw *hw = &adapter->hw;
2681         struct pci_dev *pdev = adapter->pdev;
2682         struct e1000_buffer *buffer_info;
2683         unsigned int len = skb_headlen(skb);
2684         unsigned int offset = 0, size, count = 0, i;
2685         unsigned int f;
2686
2687         i = tx_ring->next_to_use;
2688
2689         while (len) {
2690                 buffer_info = &tx_ring->buffer_info[i];
2691                 size = min(len, max_per_txd);
2692                 /* Workaround for Controller erratum --
2693                  * descriptor for non-tso packet in a linear SKB that follows a
2694                  * tso gets written back prematurely before the data is fully
2695                  * DMA'd to the controller */
2696                 if (!skb->data_len && tx_ring->last_tx_tso &&
2697                     !skb_is_gso(skb)) {
2698                         tx_ring->last_tx_tso = 0;
2699                         size -= 4;
2700                 }
2701
2702                 /* Workaround for premature desc write-backs
2703                  * in TSO mode.  Append 4-byte sentinel desc */
2704                 if (unlikely(mss && !nr_frags && size == len && size > 8))
2705                         size -= 4;
2706                 /* work-around for errata 10 and it applies
2707                  * to all controllers in PCI-X mode
2708                  * The fix is to make sure that the first descriptor of a
2709                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2710                  */
2711                 if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
2712                                 (size > 2015) && count == 0))
2713                         size = 2015;
2714
2715                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
2716                  * terminating buffers within evenly-aligned dwords. */
2717                 if (unlikely(adapter->pcix_82544 &&
2718                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2719                    size > 4))
2720                         size -= 4;
2721
2722                 buffer_info->length = size;
2723                 /* set time_stamp *before* dma to help avoid a possible race */
2724                 buffer_info->time_stamp = jiffies;
2725                 buffer_info->mapped_as_page = false;
2726                 buffer_info->dma = pci_map_single(pdev, skb->data + offset,
2727                                                   size, PCI_DMA_TODEVICE);
2728                 if (pci_dma_mapping_error(pdev, buffer_info->dma))
2729                         goto dma_error;
2730                 buffer_info->next_to_watch = i;
2731
2732                 len -= size;
2733                 offset += size;
2734                 count++;
2735                 if (len) {
2736                         i++;
2737                         if (unlikely(i == tx_ring->count))
2738                                 i = 0;
2739                 }
2740         }
2741
2742         for (f = 0; f < nr_frags; f++) {
2743                 struct skb_frag_struct *frag;
2744
2745                 frag = &skb_shinfo(skb)->frags[f];
2746                 len = frag->size;
2747                 offset = frag->page_offset;
2748
2749                 while (len) {
2750                         i++;
2751                         if (unlikely(i == tx_ring->count))
2752                                 i = 0;
2753
2754                         buffer_info = &tx_ring->buffer_info[i];
2755                         size = min(len, max_per_txd);
2756                         /* Workaround for premature desc write-backs
2757                          * in TSO mode.  Append 4-byte sentinel desc */
2758                         if (unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
2759                                 size -= 4;
2760                         /* Workaround for potential 82544 hang in PCI-X.
2761                          * Avoid terminating buffers within evenly-aligned
2762                          * dwords. */
2763                         if (unlikely(adapter->pcix_82544 &&
2764                             !((unsigned long)(page_to_phys(frag->page) + offset
2765                                               + size - 1) & 4) &&
2766                             size > 4))
2767                                 size -= 4;
2768
2769                         buffer_info->length = size;
2770                         buffer_info->time_stamp = jiffies;
2771                         buffer_info->mapped_as_page = true;
2772                         buffer_info->dma = pci_map_page(pdev, frag->page,
2773                                                         offset, size,
2774                                                         PCI_DMA_TODEVICE);
2775                         if (pci_dma_mapping_error(pdev, buffer_info->dma))
2776                                 goto dma_error;
2777                         buffer_info->next_to_watch = i;
2778
2779                         len -= size;
2780                         offset += size;
2781                         count++;
2782                 }
2783         }
2784
2785         tx_ring->buffer_info[i].skb = skb;
2786         tx_ring->buffer_info[first].next_to_watch = i;
2787
2788         return count;
2789
2790 dma_error:
2791         dev_err(&pdev->dev, "TX DMA map failed\n");
2792         buffer_info->dma = 0;
2793         if (count)
2794                 count--;
2795
2796         while (count--) {
2797                 if (i==0)
2798                         i += tx_ring->count;
2799                 i--;
2800                 buffer_info = &tx_ring->buffer_info[i];
2801                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2802         }
2803
2804         return 0;
2805 }
2806
2807 static void e1000_tx_queue(struct e1000_adapter *adapter,
2808                            struct e1000_tx_ring *tx_ring, int tx_flags,
2809                            int count)
2810 {
2811         struct e1000_hw *hw = &adapter->hw;
2812         struct e1000_tx_desc *tx_desc = NULL;
2813         struct e1000_buffer *buffer_info;
2814         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2815         unsigned int i;
2816
2817         if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
2818                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2819                              E1000_TXD_CMD_TSE;
2820                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2821
2822                 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
2823                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2824         }
2825
2826         if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
2827                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
2828                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2829         }
2830
2831         if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
2832                 txd_lower |= E1000_TXD_CMD_VLE;
2833                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
2834         }
2835
2836         i = tx_ring->next_to_use;
2837
2838         while (count--) {
2839                 buffer_info = &tx_ring->buffer_info[i];
2840                 tx_desc = E1000_TX_DESC(*tx_ring, i);
2841                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
2842                 tx_desc->lower.data =
2843                         cpu_to_le32(txd_lower | buffer_info->length);
2844                 tx_desc->upper.data = cpu_to_le32(txd_upper);
2845                 if (unlikely(++i == tx_ring->count)) i = 0;
2846         }
2847
2848         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
2849
2850         /* Force memory writes to complete before letting h/w
2851          * know there are new descriptors to fetch.  (Only
2852          * applicable for weak-ordered memory model archs,
2853          * such as IA-64). */
2854         wmb();
2855
2856         tx_ring->next_to_use = i;
2857         writel(i, hw->hw_addr + tx_ring->tdt);
2858         /* we need this if more than one processor can write to our tail
2859          * at a time, it syncronizes IO on IA64/Altix systems */
2860         mmiowb();
2861 }
2862
2863 /**
2864  * 82547 workaround to avoid controller hang in half-duplex environment.
2865  * The workaround is to avoid queuing a large packet that would span
2866  * the internal Tx FIFO ring boundary by notifying the stack to resend
2867  * the packet at a later time.  This gives the Tx FIFO an opportunity to
2868  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
2869  * to the beginning of the Tx FIFO.
2870  **/
2871
2872 #define E1000_FIFO_HDR                  0x10
2873 #define E1000_82547_PAD_LEN             0x3E0
2874
2875 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
2876                                        struct sk_buff *skb)
2877 {
2878         u32 fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
2879         u32 skb_fifo_len = skb->len + E1000_FIFO_HDR;
2880
2881         skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
2882
2883         if (adapter->link_duplex != HALF_DUPLEX)
2884                 goto no_fifo_stall_required;
2885
2886         if (atomic_read(&adapter->tx_fifo_stall))
2887                 return 1;
2888
2889         if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
2890                 atomic_set(&adapter->tx_fifo_stall, 1);
2891                 return 1;
2892         }
2893
2894 no_fifo_stall_required:
2895         adapter->tx_fifo_head += skb_fifo_len;
2896         if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
2897                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
2898         return 0;
2899 }
2900
2901 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
2902 {
2903         struct e1000_adapter *adapter = netdev_priv(netdev);
2904         struct e1000_tx_ring *tx_ring = adapter->tx_ring;
2905
2906         netif_stop_queue(netdev);
2907         /* Herbert's original patch had:
2908          *  smp_mb__after_netif_stop_queue();
2909          * but since that doesn't exist yet, just open code it. */
2910         smp_mb();
2911
2912         /* We need to check again in a case another CPU has just
2913          * made room available. */
2914         if (likely(E1000_DESC_UNUSED(tx_ring) < size))
2915                 return -EBUSY;
2916
2917         /* A reprieve! */
2918         netif_start_queue(netdev);
2919         ++adapter->restart_queue;
2920         return 0;
2921 }
2922
2923 static int e1000_maybe_stop_tx(struct net_device *netdev,
2924                                struct e1000_tx_ring *tx_ring, int size)
2925 {
2926         if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
2927                 return 0;
2928         return __e1000_maybe_stop_tx(netdev, size);
2929 }
2930
2931 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
2932 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
2933                                     struct net_device *netdev)
2934 {
2935         struct e1000_adapter *adapter = netdev_priv(netdev);
2936         struct e1000_hw *hw = &adapter->hw;
2937         struct e1000_tx_ring *tx_ring;
2938         unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
2939         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
2940         unsigned int tx_flags = 0;
2941         unsigned int len = skb->len - skb->data_len;
2942         unsigned int nr_frags;
2943         unsigned int mss;
2944         int count = 0;
2945         int tso;
2946         unsigned int f;
2947
2948         /* This goes back to the question of how to logically map a tx queue
2949          * to a flow.  Right now, performance is impacted slightly negatively
2950          * if using multiple tx queues.  If the stack breaks away from a
2951          * single qdisc implementation, we can look at this again. */
2952         tx_ring = adapter->tx_ring;
2953
2954         if (unlikely(skb->len <= 0)) {
2955                 dev_kfree_skb_any(skb);
2956                 return NETDEV_TX_OK;
2957         }
2958
2959         mss = skb_shinfo(skb)->gso_size;
2960         /* The controller does a simple calculation to
2961          * make sure there is enough room in the FIFO before
2962          * initiating the DMA for each buffer.  The calc is:
2963          * 4 = ceil(buffer len/mss).  To make sure we don't
2964          * overrun the FIFO, adjust the max buffer len if mss
2965          * drops. */
2966         if (mss) {
2967                 u8 hdr_len;
2968                 max_per_txd = min(mss << 2, max_per_txd);
2969                 max_txd_pwr = fls(max_per_txd) - 1;
2970
2971                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2972                 if (skb->data_len && hdr_len == len) {
2973                         switch (hw->mac_type) {
2974                                 unsigned int pull_size;
2975                         case e1000_82544:
2976                                 /* Make sure we have room to chop off 4 bytes,
2977                                  * and that the end alignment will work out to
2978                                  * this hardware's requirements
2979                                  * NOTE: this is a TSO only workaround
2980                                  * if end byte alignment not correct move us
2981                                  * into the next dword */
2982                                 if ((unsigned long)(skb_tail_pointer(skb) - 1) & 4)
2983                                         break;
2984                                 /* fall through */
2985                                 pull_size = min((unsigned int)4, skb->data_len);
2986                                 if (!__pskb_pull_tail(skb, pull_size)) {
2987                                         DPRINTK(DRV, ERR,
2988                                                 "__pskb_pull_tail failed.\n");
2989                                         dev_kfree_skb_any(skb);
2990                                         return NETDEV_TX_OK;
2991                                 }
2992                                 len = skb->len - skb->data_len;
2993                                 break;
2994                         default:
2995                                 /* do nothing */
2996                                 break;
2997                         }
2998                 }
2999         }
3000
3001         /* reserve a descriptor for the offload context */
3002         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3003                 count++;
3004         count++;
3005
3006         /* Controller Erratum workaround */
3007         if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3008                 count++;
3009
3010         count += TXD_USE_COUNT(len, max_txd_pwr);
3011
3012         if (adapter->pcix_82544)
3013                 count++;
3014
3015         /* work-around for errata 10 and it applies to all controllers
3016          * in PCI-X mode, so add one more descriptor to the count
3017          */
3018         if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
3019                         (len > 2015)))
3020                 count++;
3021
3022         nr_frags = skb_shinfo(skb)->nr_frags;
3023         for (f = 0; f < nr_frags; f++)
3024                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3025                                        max_txd_pwr);
3026         if (adapter->pcix_82544)
3027                 count += nr_frags;
3028
3029         /* need: count + 2 desc gap to keep tail from touching
3030          * head, otherwise try next time */
3031         if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2)))
3032                 return NETDEV_TX_BUSY;
3033
3034         if (unlikely(hw->mac_type == e1000_82547)) {
3035                 if (unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
3036                         netif_stop_queue(netdev);
3037                         if (!test_bit(__E1000_DOWN, &adapter->flags))
3038                                 mod_timer(&adapter->tx_fifo_stall_timer,
3039                                           jiffies + 1);
3040                         return NETDEV_TX_BUSY;
3041                 }
3042         }
3043
3044         if (unlikely(adapter->vlgrp && vlan_tx_tag_present(skb))) {
3045                 tx_flags |= E1000_TX_FLAGS_VLAN;
3046                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3047         }
3048
3049         first = tx_ring->next_to_use;
3050
3051         tso = e1000_tso(adapter, tx_ring, skb);
3052         if (tso < 0) {
3053                 dev_kfree_skb_any(skb);
3054                 return NETDEV_TX_OK;
3055         }
3056
3057         if (likely(tso)) {
3058                 if (likely(hw->mac_type != e1000_82544))
3059                         tx_ring->last_tx_tso = 1;
3060                 tx_flags |= E1000_TX_FLAGS_TSO;
3061         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3062                 tx_flags |= E1000_TX_FLAGS_CSUM;
3063
3064         if (likely(skb->protocol == htons(ETH_P_IP)))
3065                 tx_flags |= E1000_TX_FLAGS_IPV4;
3066
3067         count = e1000_tx_map(adapter, tx_ring, skb, first, max_per_txd,
3068                              nr_frags, mss);
3069
3070         if (count) {
3071                 e1000_tx_queue(adapter, tx_ring, tx_flags, count);
3072                 /* Make sure there is space in the ring for the next send. */
3073                 e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3074
3075         } else {
3076                 dev_kfree_skb_any(skb);
3077                 tx_ring->buffer_info[first].time_stamp = 0;
3078                 tx_ring->next_to_use = first;
3079         }
3080
3081         return NETDEV_TX_OK;
3082 }
3083
3084 /**
3085  * e1000_tx_timeout - Respond to a Tx Hang
3086  * @netdev: network interface device structure
3087  **/
3088
3089 static void e1000_tx_timeout(struct net_device *netdev)
3090 {
3091         struct e1000_adapter *adapter = netdev_priv(netdev);
3092
3093         /* Do the reset outside of interrupt context */
3094         adapter->tx_timeout_count++;
3095         schedule_work(&adapter->reset_task);
3096 }
3097
3098 static void e1000_reset_task(struct work_struct *work)
3099 {
3100         struct e1000_adapter *adapter =
3101                 container_of(work, struct e1000_adapter, reset_task);
3102
3103         e1000_reinit_locked(adapter);
3104 }
3105
3106 /**
3107  * e1000_get_stats - Get System Network Statistics
3108  * @netdev: network interface device structure
3109  *
3110  * Returns the address of the device statistics structure.
3111  * The statistics are actually updated from the timer callback.
3112  **/
3113
3114 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3115 {
3116         /* only return the current stats */
3117         return &netdev->stats;
3118 }
3119
3120 /**
3121  * e1000_change_mtu - Change the Maximum Transfer Unit
3122  * @netdev: network interface device structure
3123  * @new_mtu: new value for maximum frame size
3124  *
3125  * Returns 0 on success, negative on failure
3126  **/
3127
3128 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3129 {
3130         struct e1000_adapter *adapter = netdev_priv(netdev);
3131         struct e1000_hw *hw = &adapter->hw;
3132         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3133
3134         if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3135             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3136                 DPRINTK(PROBE, ERR, "Invalid MTU setting\n");
3137                 return -EINVAL;
3138         }
3139
3140         /* Adapter-specific max frame size limits. */
3141         switch (hw->mac_type) {
3142         case e1000_undefined ... e1000_82542_rev2_1:
3143                 if (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)) {
3144                         DPRINTK(PROBE, ERR, "Jumbo Frames not supported.\n");
3145                         return -EINVAL;
3146                 }
3147                 break;
3148         default:
3149                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3150                 break;
3151         }
3152
3153         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
3154                 msleep(1);
3155         /* e1000_down has a dependency on max_frame_size */
3156         hw->max_frame_size = max_frame;
3157         if (netif_running(netdev))
3158                 e1000_down(adapter);
3159
3160         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3161          * means we reserve 2 more, this pushes us to allocate from the next
3162          * larger slab size.
3163          * i.e. RXBUFFER_2048 --> size-4096 slab
3164          *  however with the new *_jumbo_rx* routines, jumbo receives will use
3165          *  fragmented skbs */
3166
3167         if (max_frame <= E1000_RXBUFFER_2048)
3168                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3169         else
3170 #if (PAGE_SIZE >= E1000_RXBUFFER_16384)
3171                 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3172 #elif (PAGE_SIZE >= E1000_RXBUFFER_4096)
3173                 adapter->rx_buffer_len = PAGE_SIZE;
3174 #endif
3175
3176         /* adjust allocation if LPE protects us, and we aren't using SBP */
3177         if (!hw->tbi_compatibility_on &&
3178             ((max_frame == (ETH_FRAME_LEN + ETH_FCS_LEN)) ||
3179              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3180                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3181
3182         printk(KERN_INFO "e1000: %s changing MTU from %d to %d\n",
3183                netdev->name, netdev->mtu, new_mtu);
3184         netdev->mtu = new_mtu;
3185
3186         if (netif_running(netdev))
3187                 e1000_up(adapter);
3188         else
3189                 e1000_reset(adapter);
3190
3191         clear_bit(__E1000_RESETTING, &adapter->flags);
3192
3193         return 0;
3194 }
3195
3196 /**
3197  * e1000_update_stats - Update the board statistics counters
3198  * @adapter: board private structure
3199  **/
3200
3201 void e1000_update_stats(struct e1000_adapter *adapter)
3202 {
3203         struct net_device *netdev = adapter->netdev;
3204         struct e1000_hw *hw = &adapter->hw;
3205         struct pci_dev *pdev = adapter->pdev;
3206         unsigned long flags;
3207         u16 phy_tmp;
3208
3209 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3210
3211         /*
3212          * Prevent stats update while adapter is being reset, or if the pci
3213          * connection is down.
3214          */
3215         if (adapter->link_speed == 0)
3216                 return;
3217         if (pci_channel_offline(pdev))
3218                 return;
3219
3220         spin_lock_irqsave(&adapter->stats_lock, flags);
3221
3222         /* these counters are modified from e1000_tbi_adjust_stats,
3223          * called from the interrupt context, so they must only
3224          * be written while holding adapter->stats_lock
3225          */
3226
3227         adapter->stats.crcerrs += er32(CRCERRS);
3228         adapter->stats.gprc += er32(GPRC);
3229         adapter->stats.gorcl += er32(GORCL);
3230         adapter->stats.gorch += er32(GORCH);
3231         adapter->stats.bprc += er32(BPRC);
3232         adapter->stats.mprc += er32(MPRC);
3233         adapter->stats.roc += er32(ROC);
3234
3235         adapter->stats.prc64 += er32(PRC64);
3236         adapter->stats.prc127 += er32(PRC127);
3237         adapter->stats.prc255 += er32(PRC255);
3238         adapter->stats.prc511 += er32(PRC511);
3239         adapter->stats.prc1023 += er32(PRC1023);
3240         adapter->stats.prc1522 += er32(PRC1522);
3241
3242         adapter->stats.symerrs += er32(SYMERRS);
3243         adapter->stats.mpc += er32(MPC);
3244         adapter->stats.scc += er32(SCC);
3245         adapter->stats.ecol += er32(ECOL);
3246         adapter->stats.mcc += er32(MCC);
3247         adapter->stats.latecol += er32(LATECOL);
3248         adapter->stats.dc += er32(DC);
3249         adapter->stats.sec += er32(SEC);
3250         adapter->stats.rlec += er32(RLEC);
3251         adapter->stats.xonrxc += er32(XONRXC);
3252         adapter->stats.xontxc += er32(XONTXC);
3253         adapter->stats.xoffrxc += er32(XOFFRXC);
3254         adapter->stats.xofftxc += er32(XOFFTXC);
3255         adapter->stats.fcruc += er32(FCRUC);
3256         adapter->stats.gptc += er32(GPTC);
3257         adapter->stats.gotcl += er32(GOTCL);
3258         adapter->stats.gotch += er32(GOTCH);
3259         adapter->stats.rnbc += er32(RNBC);
3260         adapter->stats.ruc += er32(RUC);
3261         adapter->stats.rfc += er32(RFC);
3262         adapter->stats.rjc += er32(RJC);
3263         adapter->stats.torl += er32(TORL);
3264         adapter->stats.torh += er32(TORH);
3265         adapter->stats.totl += er32(TOTL);
3266         adapter->stats.toth += er32(TOTH);
3267         adapter->stats.tpr += er32(TPR);
3268
3269         adapter->stats.ptc64 += er32(PTC64);
3270         adapter->stats.ptc127 += er32(PTC127);
3271         adapter->stats.ptc255 += er32(PTC255);
3272         adapter->stats.ptc511 += er32(PTC511);
3273         adapter->stats.ptc1023 += er32(PTC1023);
3274         adapter->stats.ptc1522 += er32(PTC1522);
3275
3276         adapter->stats.mptc += er32(MPTC);
3277         adapter->stats.bptc += er32(BPTC);
3278
3279         /* used for adaptive IFS */
3280
3281         hw->tx_packet_delta = er32(TPT);
3282         adapter->stats.tpt += hw->tx_packet_delta;
3283         hw->collision_delta = er32(COLC);
3284         adapter->stats.colc += hw->collision_delta;
3285
3286         if (hw->mac_type >= e1000_82543) {
3287                 adapter->stats.algnerrc += er32(ALGNERRC);
3288                 adapter->stats.rxerrc += er32(RXERRC);
3289                 adapter->stats.tncrs += er32(TNCRS);
3290                 adapter->stats.cexterr += er32(CEXTERR);
3291                 adapter->stats.tsctc += er32(TSCTC);
3292                 adapter->stats.tsctfc += er32(TSCTFC);
3293         }
3294
3295         /* Fill out the OS statistics structure */
3296         netdev->stats.multicast = adapter->stats.mprc;
3297         netdev->stats.collisions = adapter->stats.colc;
3298
3299         /* Rx Errors */
3300
3301         /* RLEC on some newer hardware can be incorrect so build
3302         * our own version based on RUC and ROC */
3303         netdev->stats.rx_errors = adapter->stats.rxerrc +
3304                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3305                 adapter->stats.ruc + adapter->stats.roc +
3306                 adapter->stats.cexterr;
3307         adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3308         netdev->stats.rx_length_errors = adapter->stats.rlerrc;
3309         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3310         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3311         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3312
3313         /* Tx Errors */
3314         adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3315         netdev->stats.tx_errors = adapter->stats.txerrc;
3316         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3317         netdev->stats.tx_window_errors = adapter->stats.latecol;
3318         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3319         if (hw->bad_tx_carr_stats_fd &&
3320             adapter->link_duplex == FULL_DUPLEX) {
3321                 netdev->stats.tx_carrier_errors = 0;
3322                 adapter->stats.tncrs = 0;
3323         }
3324
3325         /* Tx Dropped needs to be maintained elsewhere */
3326
3327         /* Phy Stats */
3328         if (hw->media_type == e1000_media_type_copper) {
3329                 if ((adapter->link_speed == SPEED_1000) &&
3330                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3331                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3332                         adapter->phy_stats.idle_errors += phy_tmp;
3333                 }
3334
3335                 if ((hw->mac_type <= e1000_82546) &&
3336                    (hw->phy_type == e1000_phy_m88) &&
3337                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3338                         adapter->phy_stats.receive_errors += phy_tmp;
3339         }
3340
3341         /* Management Stats */
3342         if (hw->has_smbus) {
3343                 adapter->stats.mgptc += er32(MGTPTC);
3344                 adapter->stats.mgprc += er32(MGTPRC);
3345                 adapter->stats.mgpdc += er32(MGTPDC);
3346         }
3347
3348         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3349 }
3350
3351 /**
3352  * e1000_intr - Interrupt Handler
3353  * @irq: interrupt number
3354  * @data: pointer to a network interface device structure
3355  **/
3356
3357 static irqreturn_t e1000_intr(int irq, void *data)
3358 {
3359         struct net_device *netdev = data;
3360         struct e1000_adapter *adapter = netdev_priv(netdev);
3361         struct e1000_hw *hw = &adapter->hw;
3362         u32 icr = er32(ICR);
3363
3364         if (unlikely((!icr) || test_bit(__E1000_DOWN, &adapter->flags)))
3365                 return IRQ_NONE;  /* Not our interrupt */
3366
3367         if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3368                 hw->get_link_status = 1;
3369                 /* guard against interrupt when we're going down */
3370                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3371                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3372         }
3373
3374         /* disable interrupts, without the synchronize_irq bit */
3375         ew32(IMC, ~0);
3376         E1000_WRITE_FLUSH();
3377
3378         if (likely(napi_schedule_prep(&adapter->napi))) {
3379                 adapter->total_tx_bytes = 0;
3380                 adapter->total_tx_packets = 0;
3381                 adapter->total_rx_bytes = 0;
3382                 adapter->total_rx_packets = 0;
3383                 __napi_schedule(&adapter->napi);
3384         } else {
3385                 /* this really should not happen! if it does it is basically a
3386                  * bug, but not a hard error, so enable ints and continue */
3387                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3388                         e1000_irq_enable(adapter);
3389         }
3390
3391         return IRQ_HANDLED;
3392 }
3393
3394 /**
3395  * e1000_clean - NAPI Rx polling callback
3396  * @adapter: board private structure
3397  **/
3398 static int e1000_clean(struct napi_struct *napi, int budget)
3399 {
3400         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
3401         int tx_clean_complete = 0, work_done = 0;
3402
3403         tx_clean_complete = e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);
3404
3405         adapter->clean_rx(adapter, &adapter->rx_ring[0], &work_done, budget);
3406
3407         if (!tx_clean_complete)
3408                 work_done = budget;
3409
3410         /* If budget not fully consumed, exit the polling mode */
3411         if (work_done < budget) {
3412                 if (likely(adapter->itr_setting & 3))
3413                         e1000_set_itr(adapter);
3414                 napi_complete(napi);
3415                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3416                         e1000_irq_enable(adapter);
3417         }
3418
3419         return work_done;
3420 }
3421
3422 /**
3423  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3424  * @adapter: board private structure
3425  **/
3426 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
3427                                struct e1000_tx_ring *tx_ring)
3428 {
3429         struct e1000_hw *hw = &adapter->hw;
3430         struct net_device *netdev = adapter->netdev;
3431         struct e1000_tx_desc *tx_desc, *eop_desc;
3432         struct e1000_buffer *buffer_info;
3433         unsigned int i, eop;
3434         unsigned int count = 0;
3435         unsigned int total_tx_bytes=0, total_tx_packets=0;
3436
3437         i = tx_ring->next_to_clean;
3438         eop = tx_ring->buffer_info[i].next_to_watch;
3439         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3440
3441         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
3442                (count < tx_ring->count)) {
3443                 bool cleaned = false;
3444                 for ( ; !cleaned; count++) {
3445                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3446                         buffer_info = &tx_ring->buffer_info[i];
3447                         cleaned = (i == eop);
3448
3449                         if (cleaned) {
3450                                 struct sk_buff *skb = buffer_info->skb;
3451                                 unsigned int segs, bytecount;
3452                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3453                                 /* multiply data chunks by size of headers */
3454                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3455                                             skb->len;
3456                                 total_tx_packets += segs;
3457                                 total_tx_bytes += bytecount;
3458                         }
3459                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3460                         tx_desc->upper.data = 0;
3461
3462                         if (unlikely(++i == tx_ring->count)) i = 0;
3463                 }
3464
3465                 eop = tx_ring->buffer_info[i].next_to_watch;
3466                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3467         }
3468
3469         tx_ring->next_to_clean = i;
3470
3471 #define TX_WAKE_THRESHOLD 32
3472         if (unlikely(count && netif_carrier_ok(netdev) &&
3473                      E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
3474                 /* Make sure that anybody stopping the queue after this
3475                  * sees the new next_to_clean.
3476                  */
3477                 smp_mb();
3478
3479                 if (netif_queue_stopped(netdev) &&
3480                     !(test_bit(__E1000_DOWN, &adapter->flags))) {
3481                         netif_wake_queue(netdev);
3482                         ++adapter->restart_queue;
3483                 }
3484         }
3485
3486         if (adapter->detect_tx_hung) {
3487                 /* Detect a transmit hang in hardware, this serializes the
3488                  * check with the clearing of time_stamp and movement of i */
3489                 adapter->detect_tx_hung = false;
3490                 if (tx_ring->buffer_info[eop].time_stamp &&
3491                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
3492                                (adapter->tx_timeout_factor * HZ)) &&
3493                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
3494
3495                         /* detected Tx unit hang */
3496                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
3497                                         "  Tx Queue             <%lu>\n"
3498                                         "  TDH                  <%x>\n"
3499                                         "  TDT                  <%x>\n"
3500                                         "  next_to_use          <%x>\n"
3501                                         "  next_to_clean        <%x>\n"
3502                                         "buffer_info[next_to_clean]\n"
3503                                         "  time_stamp           <%lx>\n"
3504                                         "  next_to_watch        <%x>\n"
3505                                         "  jiffies              <%lx>\n"
3506                                         "  next_to_watch.status <%x>\n",
3507                                 (unsigned long)((tx_ring - adapter->tx_ring) /
3508                                         sizeof(struct e1000_tx_ring)),
3509                                 readl(hw->hw_addr + tx_ring->tdh),
3510                                 readl(hw->hw_addr + tx_ring->tdt),
3511                                 tx_ring->next_to_use,
3512                                 tx_ring->next_to_clean,
3513                                 tx_ring->buffer_info[eop].time_stamp,
3514                                 eop,
3515                                 jiffies,
3516                                 eop_desc->upper.fields.status);
3517                         netif_stop_queue(netdev);
3518                 }
3519         }
3520         adapter->total_tx_bytes += total_tx_bytes;
3521         adapter->total_tx_packets += total_tx_packets;
3522         netdev->stats.tx_bytes += total_tx_bytes;
3523         netdev->stats.tx_packets += total_tx_packets;
3524         return (count < tx_ring->count);
3525 }
3526
3527 /**
3528  * e1000_rx_checksum - Receive Checksum Offload for 82543
3529  * @adapter:     board private structure
3530  * @status_err:  receive descriptor status and error fields
3531  * @csum:        receive descriptor csum field
3532  * @sk_buff:     socket buffer with received data
3533  **/
3534
3535 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
3536                               u32 csum, struct sk_buff *skb)
3537 {
3538         struct e1000_hw *hw = &adapter->hw;
3539         u16 status = (u16)status_err;
3540         u8 errors = (u8)(status_err >> 24);
3541         skb->ip_summed = CHECKSUM_NONE;
3542
3543         /* 82543 or newer only */
3544         if (unlikely(hw->mac_type < e1000_82543)) return;
3545         /* Ignore Checksum bit is set */
3546         if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
3547         /* TCP/UDP checksum error bit is set */
3548         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
3549                 /* let the stack verify checksum errors */
3550                 adapter->hw_csum_err++;
3551                 return;
3552         }
3553         /* TCP/UDP Checksum has not been calculated */
3554         if (!(status & E1000_RXD_STAT_TCPCS))
3555                 return;
3556
3557         /* It must be a TCP or UDP packet with a valid checksum */
3558         if (likely(status & E1000_RXD_STAT_TCPCS)) {
3559                 /* TCP checksum is good */
3560                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3561         }
3562         adapter->hw_csum_good++;
3563 }
3564
3565 /**
3566  * e1000_consume_page - helper function
3567  **/
3568 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
3569                                u16 length)
3570 {
3571         bi->page = NULL;
3572         skb->len += length;
3573         skb->data_len += length;
3574         skb->truesize += length;
3575 }
3576
3577 /**
3578  * e1000_receive_skb - helper function to handle rx indications
3579  * @adapter: board private structure
3580  * @status: descriptor status field as written by hardware
3581  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3582  * @skb: pointer to sk_buff to be indicated to stack
3583  */
3584 static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
3585                               __le16 vlan, struct sk_buff *skb)
3586 {
3587         if (unlikely(adapter->vlgrp && (status & E1000_RXD_STAT_VP))) {
3588                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3589                                          le16_to_cpu(vlan) &
3590                                          E1000_RXD_SPC_VLAN_MASK);
3591         } else {
3592                 netif_receive_skb(skb);
3593         }
3594 }
3595
3596 /**
3597  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
3598  * @adapter: board private structure
3599  * @rx_ring: ring to clean
3600  * @work_done: amount of napi work completed this call
3601  * @work_to_do: max amount of work allowed for this call to do
3602  *
3603  * the return value indicates whether actual cleaning was done, there
3604  * is no guarantee that everything was cleaned
3605  */
3606 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
3607                                      struct e1000_rx_ring *rx_ring,
3608                                      int *work_done, int work_to_do)
3609 {
3610         struct e1000_hw *hw = &adapter->hw;
3611         struct net_device *netdev = adapter->netdev;
3612         struct pci_dev *pdev = adapter->pdev;
3613         struct e1000_rx_desc *rx_desc, *next_rxd;
3614         struct e1000_buffer *buffer_info, *next_buffer;
3615         unsigned long irq_flags;
3616         u32 length;
3617         unsigned int i;
3618         int cleaned_count = 0;
3619         bool cleaned = false;
3620         unsigned int total_rx_bytes=0, total_rx_packets=0;
3621
3622         i = rx_ring->next_to_clean;
3623         rx_desc = E1000_RX_DESC(*rx_ring, i);
3624         buffer_info = &rx_ring->buffer_info[i];
3625
3626         while (rx_desc->status & E1000_RXD_STAT_DD) {
3627                 struct sk_buff *skb;
3628                 u8 status;
3629
3630                 if (*work_done >= work_to_do)
3631                         break;
3632                 (*work_done)++;
3633
3634                 status = rx_desc->status;
3635                 skb = buffer_info->skb;
3636                 buffer_info->skb = NULL;
3637
3638                 if (++i == rx_ring->count) i = 0;
3639                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3640                 prefetch(next_rxd);
3641
3642                 next_buffer = &rx_ring->buffer_info[i];
3643
3644                 cleaned = true;
3645                 cleaned_count++;
3646                 pci_unmap_page(pdev, buffer_info->dma, buffer_info->length,
3647                                PCI_DMA_FROMDEVICE);
3648                 buffer_info->dma = 0;
3649
3650                 length = le16_to_cpu(rx_desc->length);
3651
3652                 /* errors is only valid for DD + EOP descriptors */
3653                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
3654                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
3655                         u8 last_byte = *(skb->data + length - 1);
3656                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3657                                        last_byte)) {
3658                                 spin_lock_irqsave(&adapter->stats_lock,
3659                                                   irq_flags);
3660                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3661                                                        length, skb->data);
3662                                 spin_unlock_irqrestore(&adapter->stats_lock,
3663                                                        irq_flags);
3664                                 length--;
3665                         } else {
3666                                 /* recycle both page and skb */
3667                                 buffer_info->skb = skb;
3668                                 /* an error means any chain goes out the window
3669                                  * too */
3670                                 if (rx_ring->rx_skb_top)
3671                                         dev_kfree_skb(rx_ring->rx_skb_top);
3672                                 rx_ring->rx_skb_top = NULL;
3673                                 goto next_desc;
3674                         }
3675                 }
3676
3677 #define rxtop rx_ring->rx_skb_top
3678                 if (!(status & E1000_RXD_STAT_EOP)) {
3679                         /* this descriptor is only the beginning (or middle) */
3680                         if (!rxtop) {
3681                                 /* this is the beginning of a chain */
3682                                 rxtop = skb;
3683                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
3684                                                    0, length);
3685                         } else {
3686                                 /* this is the middle of a chain */
3687                                 skb_fill_page_desc(rxtop,
3688                                     skb_shinfo(rxtop)->nr_frags,
3689                                     buffer_info->page, 0, length);
3690                                 /* re-use the skb, only consumed the page */
3691                                 buffer_info->skb = skb;
3692                         }
3693                         e1000_consume_page(buffer_info, rxtop, length);
3694                         goto next_desc;
3695                 } else {
3696                         if (rxtop) {
3697                                 /* end of the chain */
3698                                 skb_fill_page_desc(rxtop,
3699                                     skb_shinfo(rxtop)->nr_frags,
3700                                     buffer_info->page, 0, length);
3701                                 /* re-use the current skb, we only consumed the
3702                                  * page */
3703                                 buffer_info->skb = skb;
3704                                 skb = rxtop;
3705                                 rxtop = NULL;
3706                                 e1000_consume_page(buffer_info, skb, length);
3707                         } else {
3708                                 /* no chain, got EOP, this buf is the packet
3709                                  * copybreak to save the put_page/alloc_page */
3710                                 if (length <= copybreak &&
3711                                     skb_tailroom(skb) >= length) {
3712                                         u8 *vaddr;
3713                                         vaddr = kmap_atomic(buffer_info->page,
3714                                                             KM_SKB_DATA_SOFTIRQ);
3715                                         memcpy(skb_tail_pointer(skb), vaddr, length);
3716                                         kunmap_atomic(vaddr,
3717                                                       KM_SKB_DATA_SOFTIRQ);
3718                                         /* re-use the page, so don't erase
3719                                          * buffer_info->page */
3720                                         skb_put(skb, length);
3721                                 } else {
3722                                         skb_fill_page_desc(skb, 0,
3723                                                            buffer_info->page, 0,
3724                                                            length);
3725                                         e1000_consume_page(buffer_info, skb,
3726                                                            length);
3727                                 }
3728                         }
3729                 }
3730
3731                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
3732                 e1000_rx_checksum(adapter,
3733                                   (u32)(status) |
3734                                   ((u32)(rx_desc->errors) << 24),
3735                                   le16_to_cpu(rx_desc->csum), skb);
3736
3737                 pskb_trim(skb, skb->len - 4);
3738
3739                 /* probably a little skewed due to removing CRC */
3740                 total_rx_bytes += skb->len;
3741                 total_rx_packets++;
3742
3743                 /* eth type trans needs skb->data to point to something */
3744                 if (!pskb_may_pull(skb, ETH_HLEN)) {
3745                         DPRINTK(DRV, ERR, "pskb_may_pull failed.\n");
3746                         dev_kfree_skb(skb);
3747                         goto next_desc;
3748                 }
3749
3750                 skb->protocol = eth_type_trans(skb, netdev);
3751
3752                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3753
3754 next_desc:
3755                 rx_desc->status = 0;
3756
3757                 /* return some buffers to hardware, one at a time is too slow */
3758                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3759                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3760                         cleaned_count = 0;
3761                 }
3762
3763                 /* use prefetched values */
3764                 rx_desc = next_rxd;
3765                 buffer_info = next_buffer;
3766         }
3767         rx_ring->next_to_clean = i;
3768
3769         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3770         if (cleaned_count)
3771                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3772
3773         adapter->total_rx_packets += total_rx_packets;
3774         adapter->total_rx_bytes += total_rx_bytes;
3775         netdev->stats.rx_bytes += total_rx_bytes;
3776         netdev->stats.rx_packets += total_rx_packets;
3777         return cleaned;
3778 }
3779
3780 /**
3781  * e1000_clean_rx_irq - Send received data up the network stack; legacy
3782  * @adapter: board private structure
3783  * @rx_ring: ring to clean
3784  * @work_done: amount of napi work completed this call
3785  * @work_to_do: max amount of work allowed for this call to do
3786  */
3787 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
3788                                struct e1000_rx_ring *rx_ring,
3789                                int *work_done, int work_to_do)
3790 {
3791         struct e1000_hw *hw = &adapter->hw;
3792         struct net_device *netdev = adapter->netdev;
3793         struct pci_dev *pdev = adapter->pdev;
3794         struct e1000_rx_desc *rx_desc, *next_rxd;
3795         struct e1000_buffer *buffer_info, *next_buffer;
3796         unsigned long flags;
3797         u32 length;
3798         unsigned int i;
3799         int cleaned_count = 0;
3800         bool cleaned = false;
3801         unsigned int total_rx_bytes=0, total_rx_packets=0;
3802
3803         i = rx_ring->next_to_clean;
3804         rx_desc = E1000_RX_DESC(*rx_ring, i);
3805         buffer_info = &rx_ring->buffer_info[i];
3806
3807         while (rx_desc->status & E1000_RXD_STAT_DD) {
3808                 struct sk_buff *skb;
3809                 u8 status;
3810
3811                 if (*work_done >= work_to_do)
3812                         break;
3813                 (*work_done)++;
3814
3815                 status = rx_desc->status;
3816                 skb = buffer_info->skb;
3817                 buffer_info->skb = NULL;
3818
3819                 prefetch(skb->data - NET_IP_ALIGN);
3820
3821                 if (++i == rx_ring->count) i = 0;
3822                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3823                 prefetch(next_rxd);
3824
3825                 next_buffer = &rx_ring->buffer_info[i];
3826
3827                 cleaned = true;
3828                 cleaned_count++;
3829                 pci_unmap_single(pdev, buffer_info->dma, buffer_info->length,
3830                                  PCI_DMA_FROMDEVICE);
3831                 buffer_info->dma = 0;
3832
3833                 length = le16_to_cpu(rx_desc->length);
3834                 /* !EOP means multiple descriptors were used to store a single
3835                  * packet, if thats the case we need to toss it.  In fact, we
3836                  * to toss every packet with the EOP bit clear and the next
3837                  * frame that _does_ have the EOP bit set, as it is by
3838                  * definition only a frame fragment
3839                  */
3840                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
3841                         adapter->discarding = true;
3842
3843                 if (adapter->discarding) {
3844                         /* All receives must fit into a single buffer */
3845                         E1000_DBG("%s: Receive packet consumed multiple"
3846                                   " buffers\n", netdev->name);
3847                         /* recycle */
3848                         buffer_info->skb = skb;
3849                         if (status & E1000_RXD_STAT_EOP)
3850                                 adapter->discarding = false;
3851                         goto next_desc;
3852                 }
3853
3854                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
3855                         u8 last_byte = *(skb->data + length - 1);
3856                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3857                                        last_byte)) {
3858                                 spin_lock_irqsave(&adapter->stats_lock, flags);
3859                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3860                                                        length, skb->data);
3861                                 spin_unlock_irqrestore(&adapter->stats_lock,
3862                                                        flags);
3863                                 length--;
3864                         } else {
3865                                 /* recycle */
3866                                 buffer_info->skb = skb;
3867                                 goto next_desc;
3868                         }
3869                 }
3870
3871                 /* adjust length to remove Ethernet CRC, this must be
3872                  * done after the TBI_ACCEPT workaround above */
3873                 length -= 4;
3874
3875                 /* probably a little skewed due to removing CRC */
3876                 total_rx_bytes += length;
3877                 total_rx_packets++;
3878
3879                 /* code added for copybreak, this should improve
3880                  * performance for small packets with large amounts
3881                  * of reassembly being done in the stack */
3882                 if (length < copybreak) {
3883                         struct sk_buff *new_skb =
3884                             netdev_alloc_skb_ip_align(netdev, length);
3885                         if (new_skb) {
3886                                 skb_copy_to_linear_data_offset(new_skb,
3887                                                                -NET_IP_ALIGN,
3888                                                                (skb->data -
3889                                                                 NET_IP_ALIGN),
3890                                                                (length +
3891                                                                 NET_IP_ALIGN));
3892                                 /* save the skb in buffer_info as good */
3893                                 buffer_info->skb = skb;
3894                                 skb = new_skb;
3895                         }
3896                         /* else just continue with the old one */
3897                 }
3898                 /* end copybreak code */
3899                 skb_put(skb, length);
3900
3901                 /* Receive Checksum Offload */
3902                 e1000_rx_checksum(adapter,
3903                                   (u32)(status) |
3904                                   ((u32)(rx_desc->errors) << 24),
3905                                   le16_to_cpu(rx_desc->csum), skb);
3906
3907                 skb->protocol = eth_type_trans(skb, netdev);
3908
3909                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3910
3911 next_desc:
3912                 rx_desc->status = 0;
3913
3914                 /* return some buffers to hardware, one at a time is too slow */
3915                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3916                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3917                         cleaned_count = 0;
3918                 }
3919
3920                 /* use prefetched values */
3921                 rx_desc = next_rxd;
3922                 buffer_info = next_buffer;
3923         }
3924         rx_ring->next_to_clean = i;
3925
3926         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3927         if (cleaned_count)
3928                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3929
3930         adapter->total_rx_packets += total_rx_packets;
3931         adapter->total_rx_bytes += total_rx_bytes;
3932         netdev->stats.rx_bytes += total_rx_bytes;
3933         netdev->stats.rx_packets += total_rx_packets;
3934         return cleaned;
3935 }
3936
3937 /**
3938  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
3939  * @adapter: address of board private structure
3940  * @rx_ring: pointer to receive ring structure
3941  * @cleaned_count: number of buffers to allocate this pass
3942  **/
3943
3944 static void
3945 e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
3946                              struct e1000_rx_ring *rx_ring, int cleaned_count)
3947 {
3948         struct net_device *netdev = adapter->netdev;
3949         struct pci_dev *pdev = adapter->pdev;
3950         struct e1000_rx_desc *rx_desc;
3951         struct e1000_buffer *buffer_info;
3952         struct sk_buff *skb;
3953         unsigned int i;
3954         unsigned int bufsz = 256 - 16 /*for skb_reserve */ ;
3955
3956         i = rx_ring->next_to_use;
3957         buffer_info = &rx_ring->buffer_info[i];
3958
3959         while (cleaned_count--) {
3960                 skb = buffer_info->skb;
3961                 if (skb) {
3962                         skb_trim(skb, 0);
3963                         goto check_page;
3964                 }
3965
3966                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
3967                 if (unlikely(!skb)) {
3968                         /* Better luck next round */
3969                         adapter->alloc_rx_buff_failed++;
3970                         break;
3971                 }
3972
3973                 /* Fix for errata 23, can't cross 64kB boundary */
3974                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3975                         struct sk_buff *oldskb = skb;
3976                         DPRINTK(PROBE, ERR, "skb align check failed: %u bytes "
3977                                              "at %p\n", bufsz, skb->data);
3978                         /* Try again, without freeing the previous */
3979                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
3980                         /* Failed allocation, critical failure */
3981                         if (!skb) {
3982                                 dev_kfree_skb(oldskb);
3983                                 adapter->alloc_rx_buff_failed++;
3984                                 break;
3985                         }
3986
3987                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3988                                 /* give up */
3989                                 dev_kfree_skb(skb);
3990                                 dev_kfree_skb(oldskb);
3991                                 break; /* while (cleaned_count--) */
3992                         }
3993
3994                         /* Use new allocation */
3995                         dev_kfree_skb(oldskb);
3996                 }
3997                 buffer_info->skb = skb;
3998                 buffer_info->length = adapter->rx_buffer_len;
3999 check_page:
4000                 /* allocate a new page if necessary */
4001                 if (!buffer_info->page) {
4002                         buffer_info->page = alloc_page(GFP_ATOMIC);
4003                         if (unlikely(!buffer_info->page)) {
4004                                 adapter->alloc_rx_buff_failed++;
4005                                 break;
4006                         }
4007                 }
4008
4009                 if (!buffer_info->dma)
4010                         buffer_info->dma = pci_map_page(pdev,
4011                                                         buffer_info->page, 0,
4012                                                         buffer_info->length,
4013                                                         PCI_DMA_FROMDEVICE);
4014
4015                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4016                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4017
4018                 if (unlikely(++i == rx_ring->count))
4019                         i = 0;
4020                 buffer_info = &rx_ring->buffer_info[i];
4021         }
4022
4023         if (likely(rx_ring->next_to_use != i)) {
4024                 rx_ring->next_to_use = i;
4025                 if (unlikely(i-- == 0))
4026                         i = (rx_ring->count - 1);
4027
4028                 /* Force memory writes to complete before letting h/w
4029                  * know there are new descriptors to fetch.  (Only
4030                  * applicable for weak-ordered memory model archs,
4031                  * such as IA-64). */
4032                 wmb();
4033                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4034         }
4035 }
4036
4037 /**
4038  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4039  * @adapter: address of board private structure
4040  **/
4041
4042 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4043                                    struct e1000_rx_ring *rx_ring,
4044                                    int cleaned_count)
4045 {
4046         struct e1000_hw *hw = &adapter->hw;
4047         struct net_device *netdev = adapter->netdev;
4048         struct pci_dev *pdev = adapter->pdev;
4049         struct e1000_rx_desc *rx_desc;
4050         struct e1000_buffer *buffer_info;
4051         struct sk_buff *skb;
4052         unsigned int i;
4053         unsigned int bufsz = adapter->rx_buffer_len;
4054
4055         i = rx_ring->next_to_use;
4056         buffer_info = &rx_ring->buffer_info[i];
4057
4058         while (cleaned_count--) {
4059                 skb = buffer_info->skb;
4060                 if (skb) {
4061                         skb_trim(skb, 0);
4062                         goto map_skb;
4063                 }
4064
4065                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4066                 if (unlikely(!skb)) {
4067                         /* Better luck next round */
4068                         adapter->alloc_rx_buff_failed++;
4069                         break;
4070                 }
4071
4072                 /* Fix for errata 23, can't cross 64kB boundary */
4073                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4074                         struct sk_buff *oldskb = skb;
4075                         DPRINTK(RX_ERR, ERR, "skb align check failed: %u bytes "
4076                                              "at %p\n", bufsz, skb->data);
4077                         /* Try again, without freeing the previous */
4078                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4079                         /* Failed allocation, critical failure */
4080                         if (!skb) {
4081                                 dev_kfree_skb(oldskb);
4082                                 adapter->alloc_rx_buff_failed++;
4083                                 break;
4084                         }
4085
4086                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4087                                 /* give up */
4088                                 dev_kfree_skb(skb);
4089                                 dev_kfree_skb(oldskb);
4090                                 adapter->alloc_rx_buff_failed++;
4091                                 break; /* while !buffer_info->skb */
4092                         }
4093
4094                         /* Use new allocation */
4095                         dev_kfree_skb(oldskb);
4096                 }
4097                 buffer_info->skb = skb;
4098                 buffer_info->length = adapter->rx_buffer_len;
4099 map_skb:
4100                 buffer_info->dma = pci_map_single(pdev,
4101                                                   skb->data,
4102                                                   buffer_info->length,
4103                                                   PCI_DMA_FROMDEVICE);
4104
4105                 /*
4106                  * XXX if it was allocated cleanly it will never map to a
4107                  * boundary crossing
4108                  */
4109
4110                 /* Fix for errata 23, can't cross 64kB boundary */
4111                 if (!e1000_check_64k_bound(adapter,
4112                                         (void *)(unsigned long)buffer_info->dma,
4113                                         adapter->rx_buffer_len)) {
4114                         DPRINTK(RX_ERR, ERR,
4115                                 "dma align check failed: %u bytes at %p\n",
4116                                 adapter->rx_buffer_len,
4117                                 (void *)(unsigned long)buffer_info->dma);
4118                         dev_kfree_skb(skb);
4119                         buffer_info->skb = NULL;
4120
4121                         pci_unmap_single(pdev, buffer_info->dma,
4122                                          adapter->rx_buffer_len,
4123                                          PCI_DMA_FROMDEVICE);
4124                         buffer_info->dma = 0;
4125
4126                         adapter->alloc_rx_buff_failed++;
4127                         break; /* while !buffer_info->skb */
4128                 }
4129                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4130                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4131
4132                 if (unlikely(++i == rx_ring->count))
4133                         i = 0;
4134                 buffer_info = &rx_ring->buffer_info[i];
4135         }
4136
4137         if (likely(rx_ring->next_to_use != i)) {
4138                 rx_ring->next_to_use = i;
4139                 if (unlikely(i-- == 0))
4140                         i = (rx_ring->count - 1);
4141
4142                 /* Force memory writes to complete before letting h/w
4143                  * know there are new descriptors to fetch.  (Only
4144                  * applicable for weak-ordered memory model archs,
4145                  * such as IA-64). */
4146                 wmb();
4147                 writel(i, hw->hw_addr + rx_ring->rdt);
4148         }
4149 }
4150
4151 /**
4152  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4153  * @adapter:
4154  **/
4155
4156 static void e1000_smartspeed(struct e1000_adapter *adapter)
4157 {
4158         struct e1000_hw *hw = &adapter->hw;
4159         u16 phy_status;
4160         u16 phy_ctrl;
4161
4162         if ((hw->phy_type != e1000_phy_igp) || !hw->autoneg ||
4163            !(hw->autoneg_advertised & ADVERTISE_1000_FULL))
4164                 return;
4165
4166         if (adapter->smartspeed == 0) {
4167                 /* If Master/Slave config fault is asserted twice,
4168                  * we assume back-to-back */
4169                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4170                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4171                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4172                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4173                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4174                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4175                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4176                         e1000_write_phy_reg(hw, PHY_1000T_CTRL,
4177                                             phy_ctrl);
4178                         adapter->smartspeed++;
4179                         if (!e1000_phy_setup_autoneg(hw) &&
4180                            !e1000_read_phy_reg(hw, PHY_CTRL,
4181                                                &phy_ctrl)) {
4182                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4183                                              MII_CR_RESTART_AUTO_NEG);
4184                                 e1000_write_phy_reg(hw, PHY_CTRL,
4185                                                     phy_ctrl);
4186                         }
4187                 }
4188                 return;
4189         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4190                 /* If still no link, perhaps using 2/3 pair cable */
4191                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4192                 phy_ctrl |= CR_1000T_MS_ENABLE;
4193                 e1000_write_phy_reg(hw, PHY_1000T_CTRL, phy_ctrl);
4194                 if (!e1000_phy_setup_autoneg(hw) &&
4195                    !e1000_read_phy_reg(hw, PHY_CTRL, &phy_ctrl)) {
4196                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4197                                      MII_CR_RESTART_AUTO_NEG);
4198                         e1000_write_phy_reg(hw, PHY_CTRL, phy_ctrl);
4199                 }
4200         }
4201         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4202         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4203                 adapter->smartspeed = 0;
4204 }
4205
4206 /**
4207  * e1000_ioctl -
4208  * @netdev:
4209  * @ifreq:
4210  * @cmd:
4211  **/
4212
4213 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4214 {
4215         switch (cmd) {
4216         case SIOCGMIIPHY:
4217         case SIOCGMIIREG:
4218         case SIOCSMIIREG:
4219                 return e1000_mii_ioctl(netdev, ifr, cmd);
4220         default:
4221                 return -EOPNOTSUPP;
4222         }
4223 }
4224
4225 /**
4226  * e1000_mii_ioctl -
4227  * @netdev:
4228  * @ifreq:
4229  * @cmd:
4230  **/
4231
4232 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4233                            int cmd)
4234 {
4235         struct e1000_adapter *adapter = netdev_priv(netdev);
4236         struct e1000_hw *hw = &adapter->hw;
4237         struct mii_ioctl_data *data = if_mii(ifr);
4238         int retval;
4239         u16 mii_reg;
4240         u16 spddplx;
4241         unsigned long flags;
4242
4243         if (hw->media_type != e1000_media_type_copper)
4244                 return -EOPNOTSUPP;
4245
4246         switch (cmd) {
4247         case SIOCGMIIPHY:
4248                 data->phy_id = hw->phy_addr;
4249                 break;
4250         case SIOCGMIIREG:
4251                 spin_lock_irqsave(&adapter->stats_lock, flags);
4252                 if (e1000_read_phy_reg(hw, data->reg_num & 0x1F,
4253                                    &data->val_out)) {
4254                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4255                         return -EIO;
4256                 }
4257                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4258                 break;
4259         case SIOCSMIIREG:
4260                 if (data->reg_num & ~(0x1F))
4261                         return -EFAULT;
4262                 mii_reg = data->val_in;
4263                 spin_lock_irqsave(&adapter->stats_lock, flags);
4264                 if (e1000_write_phy_reg(hw, data->reg_num,
4265                                         mii_reg)) {
4266                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4267                         return -EIO;
4268                 }
4269                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4270                 if (hw->media_type == e1000_media_type_copper) {
4271                         switch (data->reg_num) {
4272                         case PHY_CTRL:
4273                                 if (mii_reg & MII_CR_POWER_DOWN)
4274                                         break;
4275                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4276                                         hw->autoneg = 1;
4277                                         hw->autoneg_advertised = 0x2F;
4278                                 } else {
4279                                         if (mii_reg & 0x40)
4280                                                 spddplx = SPEED_1000;
4281                                         else if (mii_reg & 0x2000)
4282                                                 spddplx = SPEED_100;
4283                                         else
4284                                                 spddplx = SPEED_10;
4285                                         spddplx += (mii_reg & 0x100)
4286                                                    ? DUPLEX_FULL :
4287                                                    DUPLEX_HALF;
4288                                         retval = e1000_set_spd_dplx(adapter,
4289                                                                     spddplx);
4290                                         if (retval)
4291                                                 return retval;
4292                                 }
4293                                 if (netif_running(adapter->netdev))
4294                                         e1000_reinit_locked(adapter);
4295                                 else
4296                                         e1000_reset(adapter);
4297                                 break;
4298                         case M88E1000_PHY_SPEC_CTRL:
4299                         case M88E1000_EXT_PHY_SPEC_CTRL:
4300                                 if (e1000_phy_reset(hw))
4301                                         return -EIO;
4302                                 break;
4303                         }
4304                 } else {
4305                         switch (data->reg_num) {
4306                         case PHY_CTRL:
4307                                 if (mii_reg & MII_CR_POWER_DOWN)
4308                                         break;
4309                                 if (netif_running(adapter->netdev))
4310                                         e1000_reinit_locked(adapter);
4311                                 else
4312                                         e1000_reset(adapter);
4313                                 break;
4314                         }
4315                 }
4316                 break;
4317         default:
4318                 return -EOPNOTSUPP;
4319         }
4320         return E1000_SUCCESS;
4321 }
4322
4323 void e1000_pci_set_mwi(struct e1000_hw *hw)
4324 {
4325         struct e1000_adapter *adapter = hw->back;
4326         int ret_val = pci_set_mwi(adapter->pdev);
4327
4328         if (ret_val)
4329                 DPRINTK(PROBE, ERR, "Error in setting MWI\n");
4330 }
4331
4332 void e1000_pci_clear_mwi(struct e1000_hw *hw)
4333 {
4334         struct e1000_adapter *adapter = hw->back;
4335
4336         pci_clear_mwi(adapter->pdev);
4337 }
4338
4339 int e1000_pcix_get_mmrbc(struct e1000_hw *hw)
4340 {
4341         struct e1000_adapter *adapter = hw->back;
4342         return pcix_get_mmrbc(adapter->pdev);
4343 }
4344
4345 void e1000_pcix_set_mmrbc(struct e1000_hw *hw, int mmrbc)
4346 {
4347         struct e1000_adapter *adapter = hw->back;
4348         pcix_set_mmrbc(adapter->pdev, mmrbc);
4349 }
4350
4351 void e1000_io_write(struct e1000_hw *hw, unsigned long port, u32 value)
4352 {
4353         outl(value, port);
4354 }
4355
4356 static void e1000_vlan_rx_register(struct net_device *netdev,
4357                                    struct vlan_group *grp)
4358 {
4359         struct e1000_adapter *adapter = netdev_priv(netdev);
4360         struct e1000_hw *hw = &adapter->hw;
4361         u32 ctrl, rctl;
4362
4363         if (!test_bit(__E1000_DOWN, &adapter->flags))
4364                 e1000_irq_disable(adapter);
4365         adapter->vlgrp = grp;
4366
4367         if (grp) {
4368                 /* enable VLAN tag insert/strip */
4369                 ctrl = er32(CTRL);
4370                 ctrl |= E1000_CTRL_VME;
4371                 ew32(CTRL, ctrl);
4372
4373                 /* enable VLAN receive filtering */
4374                 rctl = er32(RCTL);
4375                 rctl &= ~E1000_RCTL_CFIEN;
4376                 if (!(netdev->flags & IFF_PROMISC))
4377                         rctl |= E1000_RCTL_VFE;
4378                 ew32(RCTL, rctl);
4379                 e1000_update_mng_vlan(adapter);
4380         } else {
4381                 /* disable VLAN tag insert/strip */
4382                 ctrl = er32(CTRL);
4383                 ctrl &= ~E1000_CTRL_VME;
4384                 ew32(CTRL, ctrl);
4385
4386                 /* disable VLAN receive filtering */
4387                 rctl = er32(RCTL);
4388                 rctl &= ~E1000_RCTL_VFE;
4389                 ew32(RCTL, rctl);
4390
4391                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
4392                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4393                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4394                 }
4395         }
4396
4397         if (!test_bit(__E1000_DOWN, &adapter->flags))
4398                 e1000_irq_enable(adapter);
4399 }
4400
4401 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4402 {
4403         struct e1000_adapter *adapter = netdev_priv(netdev);
4404         struct e1000_hw *hw = &adapter->hw;
4405         u32 vfta, index;
4406
4407         if ((hw->mng_cookie.status &
4408              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4409             (vid == adapter->mng_vlan_id))
4410                 return;
4411         /* add VID to filter table */
4412         index = (vid >> 5) & 0x7F;
4413         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4414         vfta |= (1 << (vid & 0x1F));
4415         e1000_write_vfta(hw, index, vfta);
4416 }
4417
4418 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
4419 {
4420         struct e1000_adapter *adapter = netdev_priv(netdev);
4421         struct e1000_hw *hw = &adapter->hw;
4422         u32 vfta, index;
4423
4424         if (!test_bit(__E1000_DOWN, &adapter->flags))
4425                 e1000_irq_disable(adapter);
4426         vlan_group_set_device(adapter->vlgrp, vid, NULL);
4427         if (!test_bit(__E1000_DOWN, &adapter->flags))
4428                 e1000_irq_enable(adapter);
4429
4430         /* remove VID from filter table */
4431         index = (vid >> 5) & 0x7F;
4432         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4433         vfta &= ~(1 << (vid & 0x1F));
4434         e1000_write_vfta(hw, index, vfta);
4435 }
4436
4437 static void e1000_restore_vlan(struct e1000_adapter *adapter)
4438 {
4439         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
4440
4441         if (adapter->vlgrp) {
4442                 u16 vid;
4443                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
4444                         if (!vlan_group_get_device(adapter->vlgrp, vid))
4445                                 continue;
4446                         e1000_vlan_rx_add_vid(adapter->netdev, vid);
4447                 }
4448         }
4449 }
4450
4451 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx)
4452 {
4453         struct e1000_hw *hw = &adapter->hw;
4454
4455         hw->autoneg = 0;
4456
4457         /* Fiber NICs only allow 1000 gbps Full duplex */
4458         if ((hw->media_type == e1000_media_type_fiber) &&
4459                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
4460                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
4461                 return -EINVAL;
4462         }
4463
4464         switch (spddplx) {
4465         case SPEED_10 + DUPLEX_HALF:
4466                 hw->forced_speed_duplex = e1000_10_half;
4467                 break;
4468         case SPEED_10 + DUPLEX_FULL:
4469                 hw->forced_speed_duplex = e1000_10_full;
4470                 break;
4471         case SPEED_100 + DUPLEX_HALF:
4472                 hw->forced_speed_duplex = e1000_100_half;
4473                 break;
4474         case SPEED_100 + DUPLEX_FULL:
4475                 hw->forced_speed_duplex = e1000_100_full;
4476                 break;
4477         case SPEED_1000 + DUPLEX_FULL:
4478                 hw->autoneg = 1;
4479                 hw->autoneg_advertised = ADVERTISE_1000_FULL;
4480                 break;
4481         case SPEED_1000 + DUPLEX_HALF: /* not supported */
4482         default:
4483                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
4484                 return -EINVAL;
4485         }
4486         return 0;
4487 }
4488
4489 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4490 {
4491         struct net_device *netdev = pci_get_drvdata(pdev);
4492         struct e1000_adapter *adapter = netdev_priv(netdev);
4493         struct e1000_hw *hw = &adapter->hw;
4494         u32 ctrl, ctrl_ext, rctl, status;
4495         u32 wufc = adapter->wol;
4496 #ifdef CONFIG_PM
4497         int retval = 0;
4498 #endif
4499
4500         netif_device_detach(netdev);
4501
4502         if (netif_running(netdev)) {
4503                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
4504                 e1000_down(adapter);
4505         }
4506
4507 #ifdef CONFIG_PM
4508         retval = pci_save_state(pdev);
4509         if (retval)
4510                 return retval;
4511 #endif
4512
4513         status = er32(STATUS);
4514         if (status & E1000_STATUS_LU)
4515                 wufc &= ~E1000_WUFC_LNKC;
4516
4517         if (wufc) {
4518                 e1000_setup_rctl(adapter);
4519                 e1000_set_rx_mode(netdev);
4520
4521                 /* turn on all-multi mode if wake on multicast is enabled */
4522                 if (wufc & E1000_WUFC_MC) {
4523                         rctl = er32(RCTL);
4524                         rctl |= E1000_RCTL_MPE;
4525                         ew32(RCTL, rctl);
4526                 }
4527
4528                 if (hw->mac_type >= e1000_82540) {
4529                         ctrl = er32(CTRL);
4530                         /* advertise wake from D3Cold */
4531                         #define E1000_CTRL_ADVD3WUC 0x00100000
4532                         /* phy power management enable */
4533                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4534                         ctrl |= E1000_CTRL_ADVD3WUC |
4535                                 E1000_CTRL_EN_PHY_PWR_MGMT;
4536                         ew32(CTRL, ctrl);
4537                 }
4538
4539                 if (hw->media_type == e1000_media_type_fiber ||
4540                     hw->media_type == e1000_media_type_internal_serdes) {
4541                         /* keep the laser running in D3 */
4542                         ctrl_ext = er32(CTRL_EXT);
4543                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4544                         ew32(CTRL_EXT, ctrl_ext);
4545                 }
4546
4547                 ew32(WUC, E1000_WUC_PME_EN);
4548                 ew32(WUFC, wufc);
4549         } else {
4550                 ew32(WUC, 0);
4551                 ew32(WUFC, 0);
4552         }
4553
4554         e1000_release_manageability(adapter);
4555
4556         *enable_wake = !!wufc;
4557
4558         /* make sure adapter isn't asleep if manageability is enabled */
4559         if (adapter->en_mng_pt)
4560                 *enable_wake = true;
4561
4562         if (netif_running(netdev))
4563                 e1000_free_irq(adapter);
4564
4565         pci_disable_device(pdev);
4566
4567         return 0;
4568 }
4569
4570 #ifdef CONFIG_PM
4571 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4572 {
4573         int retval;
4574         bool wake;
4575
4576         retval = __e1000_shutdown(pdev, &wake);
4577         if (retval)
4578                 return retval;
4579
4580         if (wake) {
4581                 pci_prepare_to_sleep(pdev);
4582         } else {
4583                 pci_wake_from_d3(pdev, false);
4584                 pci_set_power_state(pdev, PCI_D3hot);
4585         }
4586
4587         return 0;
4588 }
4589
4590 static int e1000_resume(struct pci_dev *pdev)
4591 {
4592         struct net_device *netdev = pci_get_drvdata(pdev);
4593         struct e1000_adapter *adapter = netdev_priv(netdev);
4594         struct e1000_hw *hw = &adapter->hw;
4595         u32 err;
4596
4597         pci_set_power_state(pdev, PCI_D0);
4598         pci_restore_state(pdev);
4599
4600         if (adapter->need_ioport)
4601                 err = pci_enable_device(pdev);
4602         else
4603                 err = pci_enable_device_mem(pdev);
4604         if (err) {
4605                 printk(KERN_ERR "e1000: Cannot enable PCI device from suspend\n");
4606                 return err;
4607         }
4608         pci_set_master(pdev);
4609
4610         pci_enable_wake(pdev, PCI_D3hot, 0);
4611         pci_enable_wake(pdev, PCI_D3cold, 0);
4612
4613         if (netif_running(netdev)) {
4614                 err = e1000_request_irq(adapter);
4615                 if (err)
4616                         return err;
4617         }
4618
4619         e1000_power_up_phy(adapter);
4620         e1000_reset(adapter);
4621         ew32(WUS, ~0);
4622
4623         e1000_init_manageability(adapter);
4624
4625         if (netif_running(netdev))
4626                 e1000_up(adapter);
4627
4628         netif_device_attach(netdev);
4629
4630         return 0;
4631 }
4632 #endif
4633
4634 static void e1000_shutdown(struct pci_dev *pdev)
4635 {
4636         bool wake;
4637
4638         __e1000_shutdown(pdev, &wake);
4639
4640         if (system_state == SYSTEM_POWER_OFF) {
4641                 pci_wake_from_d3(pdev, wake);
4642                 pci_set_power_state(pdev, PCI_D3hot);
4643         }
4644 }
4645
4646 #ifdef CONFIG_NET_POLL_CONTROLLER
4647 /*
4648  * Polling 'interrupt' - used by things like netconsole to send skbs
4649  * without having to re-enable interrupts. It's not called while
4650  * the interrupt routine is executing.
4651  */
4652 static void e1000_netpoll(struct net_device *netdev)
4653 {
4654         struct e1000_adapter *adapter = netdev_priv(netdev);
4655
4656         disable_irq(adapter->pdev->irq);
4657         e1000_intr(adapter->pdev->irq, netdev);
4658         enable_irq(adapter->pdev->irq);
4659 }
4660 #endif
4661
4662 /**
4663  * e1000_io_error_detected - called when PCI error is detected
4664  * @pdev: Pointer to PCI device
4665  * @state: The current pci connection state
4666  *
4667  * This function is called after a PCI bus error affecting
4668  * this device has been detected.
4669  */
4670 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4671                                                 pci_channel_state_t state)
4672 {
4673         struct net_device *netdev = pci_get_drvdata(pdev);
4674         struct e1000_adapter *adapter = netdev_priv(netdev);
4675
4676         netif_device_detach(netdev);
4677
4678         if (state == pci_channel_io_perm_failure)
4679                 return PCI_ERS_RESULT_DISCONNECT;
4680
4681         if (netif_running(netdev))
4682                 e1000_down(adapter);
4683         pci_disable_device(pdev);
4684
4685         /* Request a slot slot reset. */
4686         return PCI_ERS_RESULT_NEED_RESET;
4687 }
4688
4689 /**
4690  * e1000_io_slot_reset - called after the pci bus has been reset.
4691  * @pdev: Pointer to PCI device
4692  *
4693  * Restart the card from scratch, as if from a cold-boot. Implementation
4694  * resembles the first-half of the e1000_resume routine.
4695  */
4696 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4697 {
4698         struct net_device *netdev = pci_get_drvdata(pdev);
4699         struct e1000_adapter *adapter = netdev_priv(netdev);
4700         struct e1000_hw *hw = &adapter->hw;
4701         int err;
4702
4703         if (adapter->need_ioport)
4704                 err = pci_enable_device(pdev);
4705         else
4706                 err = pci_enable_device_mem(pdev);
4707         if (err) {
4708                 printk(KERN_ERR "e1000: Cannot re-enable PCI device after reset.\n");
4709                 return PCI_ERS_RESULT_DISCONNECT;
4710         }
4711         pci_set_master(pdev);
4712
4713         pci_enable_wake(pdev, PCI_D3hot, 0);
4714         pci_enable_wake(pdev, PCI_D3cold, 0);
4715
4716         e1000_reset(adapter);
4717         ew32(WUS, ~0);
4718
4719         return PCI_ERS_RESULT_RECOVERED;
4720 }
4721
4722 /**
4723  * e1000_io_resume - called when traffic can start flowing again.
4724  * @pdev: Pointer to PCI device
4725  *
4726  * This callback is called when the error recovery driver tells us that
4727  * its OK to resume normal operation. Implementation resembles the
4728  * second-half of the e1000_resume routine.
4729  */
4730 static void e1000_io_resume(struct pci_dev *pdev)
4731 {
4732         struct net_device *netdev = pci_get_drvdata(pdev);
4733         struct e1000_adapter *adapter = netdev_priv(netdev);
4734
4735         e1000_init_manageability(adapter);
4736
4737         if (netif_running(netdev)) {
4738                 if (e1000_up(adapter)) {
4739                         printk("e1000: can't bring device back up after reset\n");
4740                         return;
4741                 }
4742         }
4743
4744         netif_device_attach(netdev);
4745 }
4746
4747 /* e1000_main.c */