4d6677e4f9f26933e823c6cb66ffe1627da6329e
[linux-2.6-block.git] / drivers / net / e1000 / e1000_main.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2006 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include "e1000.h"
30 #include <net/ip6_checksum.h>
31
32 char e1000_driver_name[] = "e1000";
33 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
34 #define DRV_VERSION "7.3.21-k5-NAPI"
35 const char e1000_driver_version[] = DRV_VERSION;
36 static const char e1000_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
37
38 /* e1000_pci_tbl - PCI Device ID Table
39  *
40  * Last entry must be all 0s
41  *
42  * Macro expands to...
43  *   {PCI_DEVICE(PCI_VENDOR_ID_INTEL, device_id)}
44  */
45 static struct pci_device_id e1000_pci_tbl[] = {
46         INTEL_E1000_ETHERNET_DEVICE(0x1000),
47         INTEL_E1000_ETHERNET_DEVICE(0x1001),
48         INTEL_E1000_ETHERNET_DEVICE(0x1004),
49         INTEL_E1000_ETHERNET_DEVICE(0x1008),
50         INTEL_E1000_ETHERNET_DEVICE(0x1009),
51         INTEL_E1000_ETHERNET_DEVICE(0x100C),
52         INTEL_E1000_ETHERNET_DEVICE(0x100D),
53         INTEL_E1000_ETHERNET_DEVICE(0x100E),
54         INTEL_E1000_ETHERNET_DEVICE(0x100F),
55         INTEL_E1000_ETHERNET_DEVICE(0x1010),
56         INTEL_E1000_ETHERNET_DEVICE(0x1011),
57         INTEL_E1000_ETHERNET_DEVICE(0x1012),
58         INTEL_E1000_ETHERNET_DEVICE(0x1013),
59         INTEL_E1000_ETHERNET_DEVICE(0x1014),
60         INTEL_E1000_ETHERNET_DEVICE(0x1015),
61         INTEL_E1000_ETHERNET_DEVICE(0x1016),
62         INTEL_E1000_ETHERNET_DEVICE(0x1017),
63         INTEL_E1000_ETHERNET_DEVICE(0x1018),
64         INTEL_E1000_ETHERNET_DEVICE(0x1019),
65         INTEL_E1000_ETHERNET_DEVICE(0x101A),
66         INTEL_E1000_ETHERNET_DEVICE(0x101D),
67         INTEL_E1000_ETHERNET_DEVICE(0x101E),
68         INTEL_E1000_ETHERNET_DEVICE(0x1026),
69         INTEL_E1000_ETHERNET_DEVICE(0x1027),
70         INTEL_E1000_ETHERNET_DEVICE(0x1028),
71         INTEL_E1000_ETHERNET_DEVICE(0x1075),
72         INTEL_E1000_ETHERNET_DEVICE(0x1076),
73         INTEL_E1000_ETHERNET_DEVICE(0x1077),
74         INTEL_E1000_ETHERNET_DEVICE(0x1078),
75         INTEL_E1000_ETHERNET_DEVICE(0x1079),
76         INTEL_E1000_ETHERNET_DEVICE(0x107A),
77         INTEL_E1000_ETHERNET_DEVICE(0x107B),
78         INTEL_E1000_ETHERNET_DEVICE(0x107C),
79         INTEL_E1000_ETHERNET_DEVICE(0x108A),
80         INTEL_E1000_ETHERNET_DEVICE(0x1099),
81         INTEL_E1000_ETHERNET_DEVICE(0x10B5),
82         /* required last entry */
83         {0,}
84 };
85
86 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
87
88 int e1000_up(struct e1000_adapter *adapter);
89 void e1000_down(struct e1000_adapter *adapter);
90 void e1000_reinit_locked(struct e1000_adapter *adapter);
91 void e1000_reset(struct e1000_adapter *adapter);
92 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx);
93 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
94 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
95 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
96 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
97 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
98                              struct e1000_tx_ring *txdr);
99 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
100                              struct e1000_rx_ring *rxdr);
101 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
102                              struct e1000_tx_ring *tx_ring);
103 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
104                              struct e1000_rx_ring *rx_ring);
105 void e1000_update_stats(struct e1000_adapter *adapter);
106
107 static int e1000_init_module(void);
108 static void e1000_exit_module(void);
109 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
110 static void __devexit e1000_remove(struct pci_dev *pdev);
111 static int e1000_alloc_queues(struct e1000_adapter *adapter);
112 static int e1000_sw_init(struct e1000_adapter *adapter);
113 static int e1000_open(struct net_device *netdev);
114 static int e1000_close(struct net_device *netdev);
115 static void e1000_configure_tx(struct e1000_adapter *adapter);
116 static void e1000_configure_rx(struct e1000_adapter *adapter);
117 static void e1000_setup_rctl(struct e1000_adapter *adapter);
118 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
119 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
120 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
121                                 struct e1000_tx_ring *tx_ring);
122 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
123                                 struct e1000_rx_ring *rx_ring);
124 static void e1000_set_rx_mode(struct net_device *netdev);
125 static void e1000_update_phy_info(unsigned long data);
126 static void e1000_watchdog(unsigned long data);
127 static void e1000_82547_tx_fifo_stall(unsigned long data);
128 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
129                                     struct net_device *netdev);
130 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
131 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
132 static int e1000_set_mac(struct net_device *netdev, void *p);
133 static irqreturn_t e1000_intr(int irq, void *data);
134 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
135                                struct e1000_tx_ring *tx_ring);
136 static int e1000_clean(struct napi_struct *napi, int budget);
137 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
138                                struct e1000_rx_ring *rx_ring,
139                                int *work_done, int work_to_do);
140 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
141                                      struct e1000_rx_ring *rx_ring,
142                                      int *work_done, int work_to_do);
143 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
144                                    struct e1000_rx_ring *rx_ring,
145                                    int cleaned_count);
146 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
147                                          struct e1000_rx_ring *rx_ring,
148                                          int cleaned_count);
149 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
150 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
151                            int cmd);
152 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
153 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
154 static void e1000_tx_timeout(struct net_device *dev);
155 static void e1000_reset_task(struct work_struct *work);
156 static void e1000_smartspeed(struct e1000_adapter *adapter);
157 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
158                                        struct sk_buff *skb);
159
160 static void e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp);
161 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid);
162 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid);
163 static void e1000_restore_vlan(struct e1000_adapter *adapter);
164
165 #ifdef CONFIG_PM
166 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
167 static int e1000_resume(struct pci_dev *pdev);
168 #endif
169 static void e1000_shutdown(struct pci_dev *pdev);
170
171 #ifdef CONFIG_NET_POLL_CONTROLLER
172 /* for netdump / net console */
173 static void e1000_netpoll (struct net_device *netdev);
174 #endif
175
176 #define COPYBREAK_DEFAULT 256
177 static unsigned int copybreak __read_mostly = COPYBREAK_DEFAULT;
178 module_param(copybreak, uint, 0644);
179 MODULE_PARM_DESC(copybreak,
180         "Maximum size of packet that is copied to a new buffer on receive");
181
182 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
183                      pci_channel_state_t state);
184 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
185 static void e1000_io_resume(struct pci_dev *pdev);
186
187 static struct pci_error_handlers e1000_err_handler = {
188         .error_detected = e1000_io_error_detected,
189         .slot_reset = e1000_io_slot_reset,
190         .resume = e1000_io_resume,
191 };
192
193 static struct pci_driver e1000_driver = {
194         .name     = e1000_driver_name,
195         .id_table = e1000_pci_tbl,
196         .probe    = e1000_probe,
197         .remove   = __devexit_p(e1000_remove),
198 #ifdef CONFIG_PM
199         /* Power Managment Hooks */
200         .suspend  = e1000_suspend,
201         .resume   = e1000_resume,
202 #endif
203         .shutdown = e1000_shutdown,
204         .err_handler = &e1000_err_handler
205 };
206
207 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
208 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
209 MODULE_LICENSE("GPL");
210 MODULE_VERSION(DRV_VERSION);
211
212 static int debug = NETIF_MSG_DRV | NETIF_MSG_PROBE;
213 module_param(debug, int, 0);
214 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
215
216 /**
217  * e1000_init_module - Driver Registration Routine
218  *
219  * e1000_init_module is the first routine called when the driver is
220  * loaded. All it does is register with the PCI subsystem.
221  **/
222
223 static int __init e1000_init_module(void)
224 {
225         int ret;
226         printk(KERN_INFO "%s - version %s\n",
227                e1000_driver_string, e1000_driver_version);
228
229         printk(KERN_INFO "%s\n", e1000_copyright);
230
231         ret = pci_register_driver(&e1000_driver);
232         if (copybreak != COPYBREAK_DEFAULT) {
233                 if (copybreak == 0)
234                         printk(KERN_INFO "e1000: copybreak disabled\n");
235                 else
236                         printk(KERN_INFO "e1000: copybreak enabled for "
237                                "packets <= %u bytes\n", copybreak);
238         }
239         return ret;
240 }
241
242 module_init(e1000_init_module);
243
244 /**
245  * e1000_exit_module - Driver Exit Cleanup Routine
246  *
247  * e1000_exit_module is called just before the driver is removed
248  * from memory.
249  **/
250
251 static void __exit e1000_exit_module(void)
252 {
253         pci_unregister_driver(&e1000_driver);
254 }
255
256 module_exit(e1000_exit_module);
257
258 static int e1000_request_irq(struct e1000_adapter *adapter)
259 {
260         struct net_device *netdev = adapter->netdev;
261         irq_handler_t handler = e1000_intr;
262         int irq_flags = IRQF_SHARED;
263         int err;
264
265         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
266                           netdev);
267         if (err) {
268                 DPRINTK(PROBE, ERR,
269                         "Unable to allocate interrupt Error: %d\n", err);
270         }
271
272         return err;
273 }
274
275 static void e1000_free_irq(struct e1000_adapter *adapter)
276 {
277         struct net_device *netdev = adapter->netdev;
278
279         free_irq(adapter->pdev->irq, netdev);
280 }
281
282 /**
283  * e1000_irq_disable - Mask off interrupt generation on the NIC
284  * @adapter: board private structure
285  **/
286
287 static void e1000_irq_disable(struct e1000_adapter *adapter)
288 {
289         struct e1000_hw *hw = &adapter->hw;
290
291         ew32(IMC, ~0);
292         E1000_WRITE_FLUSH();
293         synchronize_irq(adapter->pdev->irq);
294 }
295
296 /**
297  * e1000_irq_enable - Enable default interrupt generation settings
298  * @adapter: board private structure
299  **/
300
301 static void e1000_irq_enable(struct e1000_adapter *adapter)
302 {
303         struct e1000_hw *hw = &adapter->hw;
304
305         ew32(IMS, IMS_ENABLE_MASK);
306         E1000_WRITE_FLUSH();
307 }
308
309 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
310 {
311         struct e1000_hw *hw = &adapter->hw;
312         struct net_device *netdev = adapter->netdev;
313         u16 vid = hw->mng_cookie.vlan_id;
314         u16 old_vid = adapter->mng_vlan_id;
315         if (adapter->vlgrp) {
316                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
317                         if (hw->mng_cookie.status &
318                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
319                                 e1000_vlan_rx_add_vid(netdev, vid);
320                                 adapter->mng_vlan_id = vid;
321                         } else
322                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
323
324                         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
325                                         (vid != old_vid) &&
326                             !vlan_group_get_device(adapter->vlgrp, old_vid))
327                                 e1000_vlan_rx_kill_vid(netdev, old_vid);
328                 } else
329                         adapter->mng_vlan_id = vid;
330         }
331 }
332
333 static void e1000_init_manageability(struct e1000_adapter *adapter)
334 {
335         struct e1000_hw *hw = &adapter->hw;
336
337         if (adapter->en_mng_pt) {
338                 u32 manc = er32(MANC);
339
340                 /* disable hardware interception of ARP */
341                 manc &= ~(E1000_MANC_ARP_EN);
342
343                 ew32(MANC, manc);
344         }
345 }
346
347 static void e1000_release_manageability(struct e1000_adapter *adapter)
348 {
349         struct e1000_hw *hw = &adapter->hw;
350
351         if (adapter->en_mng_pt) {
352                 u32 manc = er32(MANC);
353
354                 /* re-enable hardware interception of ARP */
355                 manc |= E1000_MANC_ARP_EN;
356
357                 ew32(MANC, manc);
358         }
359 }
360
361 /**
362  * e1000_configure - configure the hardware for RX and TX
363  * @adapter = private board structure
364  **/
365 static void e1000_configure(struct e1000_adapter *adapter)
366 {
367         struct net_device *netdev = adapter->netdev;
368         int i;
369
370         e1000_set_rx_mode(netdev);
371
372         e1000_restore_vlan(adapter);
373         e1000_init_manageability(adapter);
374
375         e1000_configure_tx(adapter);
376         e1000_setup_rctl(adapter);
377         e1000_configure_rx(adapter);
378         /* call E1000_DESC_UNUSED which always leaves
379          * at least 1 descriptor unused to make sure
380          * next_to_use != next_to_clean */
381         for (i = 0; i < adapter->num_rx_queues; i++) {
382                 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
383                 adapter->alloc_rx_buf(adapter, ring,
384                                       E1000_DESC_UNUSED(ring));
385         }
386
387         adapter->tx_queue_len = netdev->tx_queue_len;
388 }
389
390 int e1000_up(struct e1000_adapter *adapter)
391 {
392         struct e1000_hw *hw = &adapter->hw;
393
394         /* hardware has been reset, we need to reload some things */
395         e1000_configure(adapter);
396
397         clear_bit(__E1000_DOWN, &adapter->flags);
398
399         napi_enable(&adapter->napi);
400
401         e1000_irq_enable(adapter);
402
403         netif_wake_queue(adapter->netdev);
404
405         /* fire a link change interrupt to start the watchdog */
406         ew32(ICS, E1000_ICS_LSC);
407         return 0;
408 }
409
410 /**
411  * e1000_power_up_phy - restore link in case the phy was powered down
412  * @adapter: address of board private structure
413  *
414  * The phy may be powered down to save power and turn off link when the
415  * driver is unloaded and wake on lan is not enabled (among others)
416  * *** this routine MUST be followed by a call to e1000_reset ***
417  *
418  **/
419
420 void e1000_power_up_phy(struct e1000_adapter *adapter)
421 {
422         struct e1000_hw *hw = &adapter->hw;
423         u16 mii_reg = 0;
424
425         /* Just clear the power down bit to wake the phy back up */
426         if (hw->media_type == e1000_media_type_copper) {
427                 /* according to the manual, the phy will retain its
428                  * settings across a power-down/up cycle */
429                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
430                 mii_reg &= ~MII_CR_POWER_DOWN;
431                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
432         }
433 }
434
435 static void e1000_power_down_phy(struct e1000_adapter *adapter)
436 {
437         struct e1000_hw *hw = &adapter->hw;
438
439         /* Power down the PHY so no link is implied when interface is down *
440          * The PHY cannot be powered down if any of the following is true *
441          * (a) WoL is enabled
442          * (b) AMT is active
443          * (c) SoL/IDER session is active */
444         if (!adapter->wol && hw->mac_type >= e1000_82540 &&
445            hw->media_type == e1000_media_type_copper) {
446                 u16 mii_reg = 0;
447
448                 switch (hw->mac_type) {
449                 case e1000_82540:
450                 case e1000_82545:
451                 case e1000_82545_rev_3:
452                 case e1000_82546:
453                 case e1000_82546_rev_3:
454                 case e1000_82541:
455                 case e1000_82541_rev_2:
456                 case e1000_82547:
457                 case e1000_82547_rev_2:
458                         if (er32(MANC) & E1000_MANC_SMBUS_EN)
459                                 goto out;
460                         break;
461                 default:
462                         goto out;
463                 }
464                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
465                 mii_reg |= MII_CR_POWER_DOWN;
466                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
467                 mdelay(1);
468         }
469 out:
470         return;
471 }
472
473 void e1000_down(struct e1000_adapter *adapter)
474 {
475         struct e1000_hw *hw = &adapter->hw;
476         struct net_device *netdev = adapter->netdev;
477         u32 rctl, tctl;
478
479         /* signal that we're down so the interrupt handler does not
480          * reschedule our watchdog timer */
481         set_bit(__E1000_DOWN, &adapter->flags);
482
483         /* disable receives in the hardware */
484         rctl = er32(RCTL);
485         ew32(RCTL, rctl & ~E1000_RCTL_EN);
486         /* flush and sleep below */
487
488         netif_tx_disable(netdev);
489
490         /* disable transmits in the hardware */
491         tctl = er32(TCTL);
492         tctl &= ~E1000_TCTL_EN;
493         ew32(TCTL, tctl);
494         /* flush both disables and wait for them to finish */
495         E1000_WRITE_FLUSH();
496         msleep(10);
497
498         napi_disable(&adapter->napi);
499
500         e1000_irq_disable(adapter);
501
502         del_timer_sync(&adapter->tx_fifo_stall_timer);
503         del_timer_sync(&adapter->watchdog_timer);
504         del_timer_sync(&adapter->phy_info_timer);
505
506         netdev->tx_queue_len = adapter->tx_queue_len;
507         adapter->link_speed = 0;
508         adapter->link_duplex = 0;
509         netif_carrier_off(netdev);
510
511         e1000_reset(adapter);
512         e1000_clean_all_tx_rings(adapter);
513         e1000_clean_all_rx_rings(adapter);
514 }
515
516 void e1000_reinit_locked(struct e1000_adapter *adapter)
517 {
518         WARN_ON(in_interrupt());
519         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
520                 msleep(1);
521         e1000_down(adapter);
522         e1000_up(adapter);
523         clear_bit(__E1000_RESETTING, &adapter->flags);
524 }
525
526 void e1000_reset(struct e1000_adapter *adapter)
527 {
528         struct e1000_hw *hw = &adapter->hw;
529         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
530         bool legacy_pba_adjust = false;
531         u16 hwm;
532
533         /* Repartition Pba for greater than 9k mtu
534          * To take effect CTRL.RST is required.
535          */
536
537         switch (hw->mac_type) {
538         case e1000_82542_rev2_0:
539         case e1000_82542_rev2_1:
540         case e1000_82543:
541         case e1000_82544:
542         case e1000_82540:
543         case e1000_82541:
544         case e1000_82541_rev_2:
545                 legacy_pba_adjust = true;
546                 pba = E1000_PBA_48K;
547                 break;
548         case e1000_82545:
549         case e1000_82545_rev_3:
550         case e1000_82546:
551         case e1000_82546_rev_3:
552                 pba = E1000_PBA_48K;
553                 break;
554         case e1000_82547:
555         case e1000_82547_rev_2:
556                 legacy_pba_adjust = true;
557                 pba = E1000_PBA_30K;
558                 break;
559         case e1000_undefined:
560         case e1000_num_macs:
561                 break;
562         }
563
564         if (legacy_pba_adjust) {
565                 if (hw->max_frame_size > E1000_RXBUFFER_8192)
566                         pba -= 8; /* allocate more FIFO for Tx */
567
568                 if (hw->mac_type == e1000_82547) {
569                         adapter->tx_fifo_head = 0;
570                         adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
571                         adapter->tx_fifo_size =
572                                 (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
573                         atomic_set(&adapter->tx_fifo_stall, 0);
574                 }
575         } else if (hw->max_frame_size >  ETH_FRAME_LEN + ETH_FCS_LEN) {
576                 /* adjust PBA for jumbo frames */
577                 ew32(PBA, pba);
578
579                 /* To maintain wire speed transmits, the Tx FIFO should be
580                  * large enough to accommodate two full transmit packets,
581                  * rounded up to the next 1KB and expressed in KB.  Likewise,
582                  * the Rx FIFO should be large enough to accommodate at least
583                  * one full receive packet and is similarly rounded up and
584                  * expressed in KB. */
585                 pba = er32(PBA);
586                 /* upper 16 bits has Tx packet buffer allocation size in KB */
587                 tx_space = pba >> 16;
588                 /* lower 16 bits has Rx packet buffer allocation size in KB */
589                 pba &= 0xffff;
590                 /*
591                  * the tx fifo also stores 16 bytes of information about the tx
592                  * but don't include ethernet FCS because hardware appends it
593                  */
594                 min_tx_space = (hw->max_frame_size +
595                                 sizeof(struct e1000_tx_desc) -
596                                 ETH_FCS_LEN) * 2;
597                 min_tx_space = ALIGN(min_tx_space, 1024);
598                 min_tx_space >>= 10;
599                 /* software strips receive CRC, so leave room for it */
600                 min_rx_space = hw->max_frame_size;
601                 min_rx_space = ALIGN(min_rx_space, 1024);
602                 min_rx_space >>= 10;
603
604                 /* If current Tx allocation is less than the min Tx FIFO size,
605                  * and the min Tx FIFO size is less than the current Rx FIFO
606                  * allocation, take space away from current Rx allocation */
607                 if (tx_space < min_tx_space &&
608                     ((min_tx_space - tx_space) < pba)) {
609                         pba = pba - (min_tx_space - tx_space);
610
611                         /* PCI/PCIx hardware has PBA alignment constraints */
612                         switch (hw->mac_type) {
613                         case e1000_82545 ... e1000_82546_rev_3:
614                                 pba &= ~(E1000_PBA_8K - 1);
615                                 break;
616                         default:
617                                 break;
618                         }
619
620                         /* if short on rx space, rx wins and must trump tx
621                          * adjustment or use Early Receive if available */
622                         if (pba < min_rx_space)
623                                 pba = min_rx_space;
624                 }
625         }
626
627         ew32(PBA, pba);
628
629         /*
630          * flow control settings:
631          * The high water mark must be low enough to fit one full frame
632          * (or the size used for early receive) above it in the Rx FIFO.
633          * Set it to the lower of:
634          * - 90% of the Rx FIFO size, and
635          * - the full Rx FIFO size minus the early receive size (for parts
636          *   with ERT support assuming ERT set to E1000_ERT_2048), or
637          * - the full Rx FIFO size minus one full frame
638          */
639         hwm = min(((pba << 10) * 9 / 10),
640                   ((pba << 10) - hw->max_frame_size));
641
642         hw->fc_high_water = hwm & 0xFFF8;       /* 8-byte granularity */
643         hw->fc_low_water = hw->fc_high_water - 8;
644         hw->fc_pause_time = E1000_FC_PAUSE_TIME;
645         hw->fc_send_xon = 1;
646         hw->fc = hw->original_fc;
647
648         /* Allow time for pending master requests to run */
649         e1000_reset_hw(hw);
650         if (hw->mac_type >= e1000_82544)
651                 ew32(WUC, 0);
652
653         if (e1000_init_hw(hw))
654                 DPRINTK(PROBE, ERR, "Hardware Error\n");
655         e1000_update_mng_vlan(adapter);
656
657         /* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
658         if (hw->mac_type >= e1000_82544 &&
659             hw->autoneg == 1 &&
660             hw->autoneg_advertised == ADVERTISE_1000_FULL) {
661                 u32 ctrl = er32(CTRL);
662                 /* clear phy power management bit if we are in gig only mode,
663                  * which if enabled will attempt negotiation to 100Mb, which
664                  * can cause a loss of link at power off or driver unload */
665                 ctrl &= ~E1000_CTRL_SWDPIN3;
666                 ew32(CTRL, ctrl);
667         }
668
669         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
670         ew32(VET, ETHERNET_IEEE_VLAN_TYPE);
671
672         e1000_reset_adaptive(hw);
673         e1000_phy_get_info(hw, &adapter->phy_info);
674
675         e1000_release_manageability(adapter);
676 }
677
678 /**
679  *  Dump the eeprom for users having checksum issues
680  **/
681 static void e1000_dump_eeprom(struct e1000_adapter *adapter)
682 {
683         struct net_device *netdev = adapter->netdev;
684         struct ethtool_eeprom eeprom;
685         const struct ethtool_ops *ops = netdev->ethtool_ops;
686         u8 *data;
687         int i;
688         u16 csum_old, csum_new = 0;
689
690         eeprom.len = ops->get_eeprom_len(netdev);
691         eeprom.offset = 0;
692
693         data = kmalloc(eeprom.len, GFP_KERNEL);
694         if (!data) {
695                 printk(KERN_ERR "Unable to allocate memory to dump EEPROM"
696                        " data\n");
697                 return;
698         }
699
700         ops->get_eeprom(netdev, &eeprom, data);
701
702         csum_old = (data[EEPROM_CHECKSUM_REG * 2]) +
703                    (data[EEPROM_CHECKSUM_REG * 2 + 1] << 8);
704         for (i = 0; i < EEPROM_CHECKSUM_REG * 2; i += 2)
705                 csum_new += data[i] + (data[i + 1] << 8);
706         csum_new = EEPROM_SUM - csum_new;
707
708         printk(KERN_ERR "/*********************/\n");
709         printk(KERN_ERR "Current EEPROM Checksum : 0x%04x\n", csum_old);
710         printk(KERN_ERR "Calculated              : 0x%04x\n", csum_new);
711
712         printk(KERN_ERR "Offset    Values\n");
713         printk(KERN_ERR "========  ======\n");
714         print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 16, 1, data, 128, 0);
715
716         printk(KERN_ERR "Include this output when contacting your support "
717                "provider.\n");
718         printk(KERN_ERR "This is not a software error! Something bad "
719                "happened to your hardware or\n");
720         printk(KERN_ERR "EEPROM image. Ignoring this "
721                "problem could result in further problems,\n");
722         printk(KERN_ERR "possibly loss of data, corruption or system hangs!\n");
723         printk(KERN_ERR "The MAC Address will be reset to 00:00:00:00:00:00, "
724                "which is invalid\n");
725         printk(KERN_ERR "and requires you to set the proper MAC "
726                "address manually before continuing\n");
727         printk(KERN_ERR "to enable this network device.\n");
728         printk(KERN_ERR "Please inspect the EEPROM dump and report the issue "
729                "to your hardware vendor\n");
730         printk(KERN_ERR "or Intel Customer Support.\n");
731         printk(KERN_ERR "/*********************/\n");
732
733         kfree(data);
734 }
735
736 /**
737  * e1000_is_need_ioport - determine if an adapter needs ioport resources or not
738  * @pdev: PCI device information struct
739  *
740  * Return true if an adapter needs ioport resources
741  **/
742 static int e1000_is_need_ioport(struct pci_dev *pdev)
743 {
744         switch (pdev->device) {
745         case E1000_DEV_ID_82540EM:
746         case E1000_DEV_ID_82540EM_LOM:
747         case E1000_DEV_ID_82540EP:
748         case E1000_DEV_ID_82540EP_LOM:
749         case E1000_DEV_ID_82540EP_LP:
750         case E1000_DEV_ID_82541EI:
751         case E1000_DEV_ID_82541EI_MOBILE:
752         case E1000_DEV_ID_82541ER:
753         case E1000_DEV_ID_82541ER_LOM:
754         case E1000_DEV_ID_82541GI:
755         case E1000_DEV_ID_82541GI_LF:
756         case E1000_DEV_ID_82541GI_MOBILE:
757         case E1000_DEV_ID_82544EI_COPPER:
758         case E1000_DEV_ID_82544EI_FIBER:
759         case E1000_DEV_ID_82544GC_COPPER:
760         case E1000_DEV_ID_82544GC_LOM:
761         case E1000_DEV_ID_82545EM_COPPER:
762         case E1000_DEV_ID_82545EM_FIBER:
763         case E1000_DEV_ID_82546EB_COPPER:
764         case E1000_DEV_ID_82546EB_FIBER:
765         case E1000_DEV_ID_82546EB_QUAD_COPPER:
766                 return true;
767         default:
768                 return false;
769         }
770 }
771
772 static const struct net_device_ops e1000_netdev_ops = {
773         .ndo_open               = e1000_open,
774         .ndo_stop               = e1000_close,
775         .ndo_start_xmit         = e1000_xmit_frame,
776         .ndo_get_stats          = e1000_get_stats,
777         .ndo_set_rx_mode        = e1000_set_rx_mode,
778         .ndo_set_mac_address    = e1000_set_mac,
779         .ndo_tx_timeout         = e1000_tx_timeout,
780         .ndo_change_mtu         = e1000_change_mtu,
781         .ndo_do_ioctl           = e1000_ioctl,
782         .ndo_validate_addr      = eth_validate_addr,
783
784         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
785         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
786         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
787 #ifdef CONFIG_NET_POLL_CONTROLLER
788         .ndo_poll_controller    = e1000_netpoll,
789 #endif
790 };
791
792 /**
793  * e1000_probe - Device Initialization Routine
794  * @pdev: PCI device information struct
795  * @ent: entry in e1000_pci_tbl
796  *
797  * Returns 0 on success, negative on failure
798  *
799  * e1000_probe initializes an adapter identified by a pci_dev structure.
800  * The OS initialization, configuring of the adapter private structure,
801  * and a hardware reset occur.
802  **/
803 static int __devinit e1000_probe(struct pci_dev *pdev,
804                                  const struct pci_device_id *ent)
805 {
806         struct net_device *netdev;
807         struct e1000_adapter *adapter;
808         struct e1000_hw *hw;
809
810         static int cards_found = 0;
811         static int global_quad_port_a = 0; /* global ksp3 port a indication */
812         int i, err, pci_using_dac;
813         u16 eeprom_data = 0;
814         u16 eeprom_apme_mask = E1000_EEPROM_APME;
815         int bars, need_ioport;
816
817         /* do not allocate ioport bars when not needed */
818         need_ioport = e1000_is_need_ioport(pdev);
819         if (need_ioport) {
820                 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
821                 err = pci_enable_device(pdev);
822         } else {
823                 bars = pci_select_bars(pdev, IORESOURCE_MEM);
824                 err = pci_enable_device_mem(pdev);
825         }
826         if (err)
827                 return err;
828
829         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) &&
830             !pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64))) {
831                 pci_using_dac = 1;
832         } else {
833                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
834                 if (err) {
835                         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
836                         if (err) {
837                                 E1000_ERR("No usable DMA configuration, "
838                                           "aborting\n");
839                                 goto err_dma;
840                         }
841                 }
842                 pci_using_dac = 0;
843         }
844
845         err = pci_request_selected_regions(pdev, bars, e1000_driver_name);
846         if (err)
847                 goto err_pci_reg;
848
849         pci_set_master(pdev);
850
851         err = -ENOMEM;
852         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
853         if (!netdev)
854                 goto err_alloc_etherdev;
855
856         SET_NETDEV_DEV(netdev, &pdev->dev);
857
858         pci_set_drvdata(pdev, netdev);
859         adapter = netdev_priv(netdev);
860         adapter->netdev = netdev;
861         adapter->pdev = pdev;
862         adapter->msg_enable = (1 << debug) - 1;
863         adapter->bars = bars;
864         adapter->need_ioport = need_ioport;
865
866         hw = &adapter->hw;
867         hw->back = adapter;
868
869         err = -EIO;
870         hw->hw_addr = pci_ioremap_bar(pdev, BAR_0);
871         if (!hw->hw_addr)
872                 goto err_ioremap;
873
874         if (adapter->need_ioport) {
875                 for (i = BAR_1; i <= BAR_5; i++) {
876                         if (pci_resource_len(pdev, i) == 0)
877                                 continue;
878                         if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
879                                 hw->io_base = pci_resource_start(pdev, i);
880                                 break;
881                         }
882                 }
883         }
884
885         netdev->netdev_ops = &e1000_netdev_ops;
886         e1000_set_ethtool_ops(netdev);
887         netdev->watchdog_timeo = 5 * HZ;
888         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
889
890         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
891
892         adapter->bd_number = cards_found;
893
894         /* setup the private structure */
895
896         err = e1000_sw_init(adapter);
897         if (err)
898                 goto err_sw_init;
899
900         err = -EIO;
901
902         if (hw->mac_type >= e1000_82543) {
903                 netdev->features = NETIF_F_SG |
904                                    NETIF_F_HW_CSUM |
905                                    NETIF_F_HW_VLAN_TX |
906                                    NETIF_F_HW_VLAN_RX |
907                                    NETIF_F_HW_VLAN_FILTER;
908         }
909
910         if ((hw->mac_type >= e1000_82544) &&
911            (hw->mac_type != e1000_82547))
912                 netdev->features |= NETIF_F_TSO;
913
914         if (pci_using_dac)
915                 netdev->features |= NETIF_F_HIGHDMA;
916
917         netdev->vlan_features |= NETIF_F_TSO;
918         netdev->vlan_features |= NETIF_F_HW_CSUM;
919         netdev->vlan_features |= NETIF_F_SG;
920
921         adapter->en_mng_pt = e1000_enable_mng_pass_thru(hw);
922
923         /* initialize eeprom parameters */
924         if (e1000_init_eeprom_params(hw)) {
925                 E1000_ERR("EEPROM initialization failed\n");
926                 goto err_eeprom;
927         }
928
929         /* before reading the EEPROM, reset the controller to
930          * put the device in a known good starting state */
931
932         e1000_reset_hw(hw);
933
934         /* make sure the EEPROM is good */
935         if (e1000_validate_eeprom_checksum(hw) < 0) {
936                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
937                 e1000_dump_eeprom(adapter);
938                 /*
939                  * set MAC address to all zeroes to invalidate and temporary
940                  * disable this device for the user. This blocks regular
941                  * traffic while still permitting ethtool ioctls from reaching
942                  * the hardware as well as allowing the user to run the
943                  * interface after manually setting a hw addr using
944                  * `ip set address`
945                  */
946                 memset(hw->mac_addr, 0, netdev->addr_len);
947         } else {
948                 /* copy the MAC address out of the EEPROM */
949                 if (e1000_read_mac_addr(hw))
950                         DPRINTK(PROBE, ERR, "EEPROM Read Error\n");
951         }
952         /* don't block initalization here due to bad MAC address */
953         memcpy(netdev->dev_addr, hw->mac_addr, netdev->addr_len);
954         memcpy(netdev->perm_addr, hw->mac_addr, netdev->addr_len);
955
956         if (!is_valid_ether_addr(netdev->perm_addr))
957                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
958
959         e1000_get_bus_info(hw);
960
961         init_timer(&adapter->tx_fifo_stall_timer);
962         adapter->tx_fifo_stall_timer.function = &e1000_82547_tx_fifo_stall;
963         adapter->tx_fifo_stall_timer.data = (unsigned long)adapter;
964
965         init_timer(&adapter->watchdog_timer);
966         adapter->watchdog_timer.function = &e1000_watchdog;
967         adapter->watchdog_timer.data = (unsigned long) adapter;
968
969         init_timer(&adapter->phy_info_timer);
970         adapter->phy_info_timer.function = &e1000_update_phy_info;
971         adapter->phy_info_timer.data = (unsigned long)adapter;
972
973         INIT_WORK(&adapter->reset_task, e1000_reset_task);
974
975         e1000_check_options(adapter);
976
977         /* Initial Wake on LAN setting
978          * If APM wake is enabled in the EEPROM,
979          * enable the ACPI Magic Packet filter
980          */
981
982         switch (hw->mac_type) {
983         case e1000_82542_rev2_0:
984         case e1000_82542_rev2_1:
985         case e1000_82543:
986                 break;
987         case e1000_82544:
988                 e1000_read_eeprom(hw,
989                         EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
990                 eeprom_apme_mask = E1000_EEPROM_82544_APM;
991                 break;
992         case e1000_82546:
993         case e1000_82546_rev_3:
994                 if (er32(STATUS) & E1000_STATUS_FUNC_1){
995                         e1000_read_eeprom(hw,
996                                 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
997                         break;
998                 }
999                 /* Fall Through */
1000         default:
1001                 e1000_read_eeprom(hw,
1002                         EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1003                 break;
1004         }
1005         if (eeprom_data & eeprom_apme_mask)
1006                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1007
1008         /* now that we have the eeprom settings, apply the special cases
1009          * where the eeprom may be wrong or the board simply won't support
1010          * wake on lan on a particular port */
1011         switch (pdev->device) {
1012         case E1000_DEV_ID_82546GB_PCIE:
1013                 adapter->eeprom_wol = 0;
1014                 break;
1015         case E1000_DEV_ID_82546EB_FIBER:
1016         case E1000_DEV_ID_82546GB_FIBER:
1017                 /* Wake events only supported on port A for dual fiber
1018                  * regardless of eeprom setting */
1019                 if (er32(STATUS) & E1000_STATUS_FUNC_1)
1020                         adapter->eeprom_wol = 0;
1021                 break;
1022         case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
1023                 /* if quad port adapter, disable WoL on all but port A */
1024                 if (global_quad_port_a != 0)
1025                         adapter->eeprom_wol = 0;
1026                 else
1027                         adapter->quad_port_a = 1;
1028                 /* Reset for multiple quad port adapters */
1029                 if (++global_quad_port_a == 4)
1030                         global_quad_port_a = 0;
1031                 break;
1032         }
1033
1034         /* initialize the wol settings based on the eeprom settings */
1035         adapter->wol = adapter->eeprom_wol;
1036         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1037
1038         /* print bus type/speed/width info */
1039         DPRINTK(PROBE, INFO, "(PCI%s:%s:%s) ",
1040                 ((hw->bus_type == e1000_bus_type_pcix) ? "-X" : ""),
1041                 ((hw->bus_speed == e1000_bus_speed_133) ? "133MHz" :
1042                  (hw->bus_speed == e1000_bus_speed_120) ? "120MHz" :
1043                  (hw->bus_speed == e1000_bus_speed_100) ? "100MHz" :
1044                  (hw->bus_speed == e1000_bus_speed_66) ? "66MHz" : "33MHz"),
1045                 ((hw->bus_width == e1000_bus_width_64) ? "64-bit" : "32-bit"));
1046
1047         printk("%pM\n", netdev->dev_addr);
1048
1049         /* reset the hardware with the new settings */
1050         e1000_reset(adapter);
1051
1052         strcpy(netdev->name, "eth%d");
1053         err = register_netdev(netdev);
1054         if (err)
1055                 goto err_register;
1056
1057         /* carrier off reporting is important to ethtool even BEFORE open */
1058         netif_carrier_off(netdev);
1059
1060         DPRINTK(PROBE, INFO, "Intel(R) PRO/1000 Network Connection\n");
1061
1062         cards_found++;
1063         return 0;
1064
1065 err_register:
1066 err_eeprom:
1067         e1000_phy_hw_reset(hw);
1068
1069         if (hw->flash_address)
1070                 iounmap(hw->flash_address);
1071         kfree(adapter->tx_ring);
1072         kfree(adapter->rx_ring);
1073 err_sw_init:
1074         iounmap(hw->hw_addr);
1075 err_ioremap:
1076         free_netdev(netdev);
1077 err_alloc_etherdev:
1078         pci_release_selected_regions(pdev, bars);
1079 err_pci_reg:
1080 err_dma:
1081         pci_disable_device(pdev);
1082         return err;
1083 }
1084
1085 /**
1086  * e1000_remove - Device Removal Routine
1087  * @pdev: PCI device information struct
1088  *
1089  * e1000_remove is called by the PCI subsystem to alert the driver
1090  * that it should release a PCI device.  The could be caused by a
1091  * Hot-Plug event, or because the driver is going to be removed from
1092  * memory.
1093  **/
1094
1095 static void __devexit e1000_remove(struct pci_dev *pdev)
1096 {
1097         struct net_device *netdev = pci_get_drvdata(pdev);
1098         struct e1000_adapter *adapter = netdev_priv(netdev);
1099         struct e1000_hw *hw = &adapter->hw;
1100
1101         cancel_work_sync(&adapter->reset_task);
1102
1103         e1000_release_manageability(adapter);
1104
1105         unregister_netdev(netdev);
1106
1107         e1000_phy_hw_reset(hw);
1108
1109         kfree(adapter->tx_ring);
1110         kfree(adapter->rx_ring);
1111
1112         iounmap(hw->hw_addr);
1113         if (hw->flash_address)
1114                 iounmap(hw->flash_address);
1115         pci_release_selected_regions(pdev, adapter->bars);
1116
1117         free_netdev(netdev);
1118
1119         pci_disable_device(pdev);
1120 }
1121
1122 /**
1123  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1124  * @adapter: board private structure to initialize
1125  *
1126  * e1000_sw_init initializes the Adapter private data structure.
1127  * Fields are initialized based on PCI device information and
1128  * OS network device settings (MTU size).
1129  **/
1130
1131 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
1132 {
1133         struct e1000_hw *hw = &adapter->hw;
1134         struct net_device *netdev = adapter->netdev;
1135         struct pci_dev *pdev = adapter->pdev;
1136
1137         /* PCI config space info */
1138
1139         hw->vendor_id = pdev->vendor;
1140         hw->device_id = pdev->device;
1141         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1142         hw->subsystem_id = pdev->subsystem_device;
1143         hw->revision_id = pdev->revision;
1144
1145         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
1146
1147         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1148         hw->max_frame_size = netdev->mtu +
1149                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
1150         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
1151
1152         /* identify the MAC */
1153
1154         if (e1000_set_mac_type(hw)) {
1155                 DPRINTK(PROBE, ERR, "Unknown MAC Type\n");
1156                 return -EIO;
1157         }
1158
1159         switch (hw->mac_type) {
1160         default:
1161                 break;
1162         case e1000_82541:
1163         case e1000_82547:
1164         case e1000_82541_rev_2:
1165         case e1000_82547_rev_2:
1166                 hw->phy_init_script = 1;
1167                 break;
1168         }
1169
1170         e1000_set_media_type(hw);
1171
1172         hw->wait_autoneg_complete = false;
1173         hw->tbi_compatibility_en = true;
1174         hw->adaptive_ifs = true;
1175
1176         /* Copper options */
1177
1178         if (hw->media_type == e1000_media_type_copper) {
1179                 hw->mdix = AUTO_ALL_MODES;
1180                 hw->disable_polarity_correction = false;
1181                 hw->master_slave = E1000_MASTER_SLAVE;
1182         }
1183
1184         adapter->num_tx_queues = 1;
1185         adapter->num_rx_queues = 1;
1186
1187         if (e1000_alloc_queues(adapter)) {
1188                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
1189                 return -ENOMEM;
1190         }
1191
1192         /* Explicitly disable IRQ since the NIC can be in any state. */
1193         e1000_irq_disable(adapter);
1194
1195         spin_lock_init(&adapter->stats_lock);
1196
1197         set_bit(__E1000_DOWN, &adapter->flags);
1198
1199         return 0;
1200 }
1201
1202 /**
1203  * e1000_alloc_queues - Allocate memory for all rings
1204  * @adapter: board private structure to initialize
1205  *
1206  * We allocate one ring per queue at run-time since we don't know the
1207  * number of queues at compile-time.
1208  **/
1209
1210 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1211 {
1212         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1213                                    sizeof(struct e1000_tx_ring), GFP_KERNEL);
1214         if (!adapter->tx_ring)
1215                 return -ENOMEM;
1216
1217         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1218                                    sizeof(struct e1000_rx_ring), GFP_KERNEL);
1219         if (!adapter->rx_ring) {
1220                 kfree(adapter->tx_ring);
1221                 return -ENOMEM;
1222         }
1223
1224         return E1000_SUCCESS;
1225 }
1226
1227 /**
1228  * e1000_open - Called when a network interface is made active
1229  * @netdev: network interface device structure
1230  *
1231  * Returns 0 on success, negative value on failure
1232  *
1233  * The open entry point is called when a network interface is made
1234  * active by the system (IFF_UP).  At this point all resources needed
1235  * for transmit and receive operations are allocated, the interrupt
1236  * handler is registered with the OS, the watchdog timer is started,
1237  * and the stack is notified that the interface is ready.
1238  **/
1239
1240 static int e1000_open(struct net_device *netdev)
1241 {
1242         struct e1000_adapter *adapter = netdev_priv(netdev);
1243         struct e1000_hw *hw = &adapter->hw;
1244         int err;
1245
1246         /* disallow open during test */
1247         if (test_bit(__E1000_TESTING, &adapter->flags))
1248                 return -EBUSY;
1249
1250         netif_carrier_off(netdev);
1251
1252         /* allocate transmit descriptors */
1253         err = e1000_setup_all_tx_resources(adapter);
1254         if (err)
1255                 goto err_setup_tx;
1256
1257         /* allocate receive descriptors */
1258         err = e1000_setup_all_rx_resources(adapter);
1259         if (err)
1260                 goto err_setup_rx;
1261
1262         e1000_power_up_phy(adapter);
1263
1264         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1265         if ((hw->mng_cookie.status &
1266                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1267                 e1000_update_mng_vlan(adapter);
1268         }
1269
1270         /* before we allocate an interrupt, we must be ready to handle it.
1271          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1272          * as soon as we call pci_request_irq, so we have to setup our
1273          * clean_rx handler before we do so.  */
1274         e1000_configure(adapter);
1275
1276         err = e1000_request_irq(adapter);
1277         if (err)
1278                 goto err_req_irq;
1279
1280         /* From here on the code is the same as e1000_up() */
1281         clear_bit(__E1000_DOWN, &adapter->flags);
1282
1283         napi_enable(&adapter->napi);
1284
1285         e1000_irq_enable(adapter);
1286
1287         netif_start_queue(netdev);
1288
1289         /* fire a link status change interrupt to start the watchdog */
1290         ew32(ICS, E1000_ICS_LSC);
1291
1292         return E1000_SUCCESS;
1293
1294 err_req_irq:
1295         e1000_power_down_phy(adapter);
1296         e1000_free_all_rx_resources(adapter);
1297 err_setup_rx:
1298         e1000_free_all_tx_resources(adapter);
1299 err_setup_tx:
1300         e1000_reset(adapter);
1301
1302         return err;
1303 }
1304
1305 /**
1306  * e1000_close - Disables a network interface
1307  * @netdev: network interface device structure
1308  *
1309  * Returns 0, this is not allowed to fail
1310  *
1311  * The close entry point is called when an interface is de-activated
1312  * by the OS.  The hardware is still under the drivers control, but
1313  * needs to be disabled.  A global MAC reset is issued to stop the
1314  * hardware, and all transmit and receive resources are freed.
1315  **/
1316
1317 static int e1000_close(struct net_device *netdev)
1318 {
1319         struct e1000_adapter *adapter = netdev_priv(netdev);
1320         struct e1000_hw *hw = &adapter->hw;
1321
1322         WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1323         e1000_down(adapter);
1324         e1000_power_down_phy(adapter);
1325         e1000_free_irq(adapter);
1326
1327         e1000_free_all_tx_resources(adapter);
1328         e1000_free_all_rx_resources(adapter);
1329
1330         /* kill manageability vlan ID if supported, but not if a vlan with
1331          * the same ID is registered on the host OS (let 8021q kill it) */
1332         if ((hw->mng_cookie.status &
1333                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1334              !(adapter->vlgrp &&
1335                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id))) {
1336                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1337         }
1338
1339         return 0;
1340 }
1341
1342 /**
1343  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1344  * @adapter: address of board private structure
1345  * @start: address of beginning of memory
1346  * @len: length of memory
1347  **/
1348 static bool e1000_check_64k_bound(struct e1000_adapter *adapter, void *start,
1349                                   unsigned long len)
1350 {
1351         struct e1000_hw *hw = &adapter->hw;
1352         unsigned long begin = (unsigned long)start;
1353         unsigned long end = begin + len;
1354
1355         /* First rev 82545 and 82546 need to not allow any memory
1356          * write location to cross 64k boundary due to errata 23 */
1357         if (hw->mac_type == e1000_82545 ||
1358             hw->mac_type == e1000_82546) {
1359                 return ((begin ^ (end - 1)) >> 16) != 0 ? false : true;
1360         }
1361
1362         return true;
1363 }
1364
1365 /**
1366  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1367  * @adapter: board private structure
1368  * @txdr:    tx descriptor ring (for a specific queue) to setup
1369  *
1370  * Return 0 on success, negative on failure
1371  **/
1372
1373 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
1374                                     struct e1000_tx_ring *txdr)
1375 {
1376         struct pci_dev *pdev = adapter->pdev;
1377         int size;
1378
1379         size = sizeof(struct e1000_buffer) * txdr->count;
1380         txdr->buffer_info = vmalloc(size);
1381         if (!txdr->buffer_info) {
1382                 DPRINTK(PROBE, ERR,
1383                 "Unable to allocate memory for the transmit descriptor ring\n");
1384                 return -ENOMEM;
1385         }
1386         memset(txdr->buffer_info, 0, size);
1387
1388         /* round up to nearest 4K */
1389
1390         txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1391         txdr->size = ALIGN(txdr->size, 4096);
1392
1393         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1394         if (!txdr->desc) {
1395 setup_tx_desc_die:
1396                 vfree(txdr->buffer_info);
1397                 DPRINTK(PROBE, ERR,
1398                 "Unable to allocate memory for the transmit descriptor ring\n");
1399                 return -ENOMEM;
1400         }
1401
1402         /* Fix for errata 23, can't cross 64kB boundary */
1403         if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1404                 void *olddesc = txdr->desc;
1405                 dma_addr_t olddma = txdr->dma;
1406                 DPRINTK(TX_ERR, ERR, "txdr align check failed: %u bytes "
1407                                      "at %p\n", txdr->size, txdr->desc);
1408                 /* Try again, without freeing the previous */
1409                 txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1410                 /* Failed allocation, critical failure */
1411                 if (!txdr->desc) {
1412                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1413                         goto setup_tx_desc_die;
1414                 }
1415
1416                 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1417                         /* give up */
1418                         pci_free_consistent(pdev, txdr->size, txdr->desc,
1419                                             txdr->dma);
1420                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1421                         DPRINTK(PROBE, ERR,
1422                                 "Unable to allocate aligned memory "
1423                                 "for the transmit descriptor ring\n");
1424                         vfree(txdr->buffer_info);
1425                         return -ENOMEM;
1426                 } else {
1427                         /* Free old allocation, new allocation was successful */
1428                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1429                 }
1430         }
1431         memset(txdr->desc, 0, txdr->size);
1432
1433         txdr->next_to_use = 0;
1434         txdr->next_to_clean = 0;
1435
1436         return 0;
1437 }
1438
1439 /**
1440  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1441  *                                (Descriptors) for all queues
1442  * @adapter: board private structure
1443  *
1444  * Return 0 on success, negative on failure
1445  **/
1446
1447 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1448 {
1449         int i, err = 0;
1450
1451         for (i = 0; i < adapter->num_tx_queues; i++) {
1452                 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1453                 if (err) {
1454                         DPRINTK(PROBE, ERR,
1455                                 "Allocation for Tx Queue %u failed\n", i);
1456                         for (i-- ; i >= 0; i--)
1457                                 e1000_free_tx_resources(adapter,
1458                                                         &adapter->tx_ring[i]);
1459                         break;
1460                 }
1461         }
1462
1463         return err;
1464 }
1465
1466 /**
1467  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1468  * @adapter: board private structure
1469  *
1470  * Configure the Tx unit of the MAC after a reset.
1471  **/
1472
1473 static void e1000_configure_tx(struct e1000_adapter *adapter)
1474 {
1475         u64 tdba;
1476         struct e1000_hw *hw = &adapter->hw;
1477         u32 tdlen, tctl, tipg;
1478         u32 ipgr1, ipgr2;
1479
1480         /* Setup the HW Tx Head and Tail descriptor pointers */
1481
1482         switch (adapter->num_tx_queues) {
1483         case 1:
1484         default:
1485                 tdba = adapter->tx_ring[0].dma;
1486                 tdlen = adapter->tx_ring[0].count *
1487                         sizeof(struct e1000_tx_desc);
1488                 ew32(TDLEN, tdlen);
1489                 ew32(TDBAH, (tdba >> 32));
1490                 ew32(TDBAL, (tdba & 0x00000000ffffffffULL));
1491                 ew32(TDT, 0);
1492                 ew32(TDH, 0);
1493                 adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ? E1000_TDH : E1000_82542_TDH);
1494                 adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ? E1000_TDT : E1000_82542_TDT);
1495                 break;
1496         }
1497
1498         /* Set the default values for the Tx Inter Packet Gap timer */
1499         if ((hw->media_type == e1000_media_type_fiber ||
1500              hw->media_type == e1000_media_type_internal_serdes))
1501                 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1502         else
1503                 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1504
1505         switch (hw->mac_type) {
1506         case e1000_82542_rev2_0:
1507         case e1000_82542_rev2_1:
1508                 tipg = DEFAULT_82542_TIPG_IPGT;
1509                 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1510                 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1511                 break;
1512         default:
1513                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1514                 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1515                 break;
1516         }
1517         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1518         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1519         ew32(TIPG, tipg);
1520
1521         /* Set the Tx Interrupt Delay register */
1522
1523         ew32(TIDV, adapter->tx_int_delay);
1524         if (hw->mac_type >= e1000_82540)
1525                 ew32(TADV, adapter->tx_abs_int_delay);
1526
1527         /* Program the Transmit Control Register */
1528
1529         tctl = er32(TCTL);
1530         tctl &= ~E1000_TCTL_CT;
1531         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1532                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1533
1534         e1000_config_collision_dist(hw);
1535
1536         /* Setup Transmit Descriptor Settings for eop descriptor */
1537         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1538
1539         /* only set IDE if we are delaying interrupts using the timers */
1540         if (adapter->tx_int_delay)
1541                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1542
1543         if (hw->mac_type < e1000_82543)
1544                 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1545         else
1546                 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1547
1548         /* Cache if we're 82544 running in PCI-X because we'll
1549          * need this to apply a workaround later in the send path. */
1550         if (hw->mac_type == e1000_82544 &&
1551             hw->bus_type == e1000_bus_type_pcix)
1552                 adapter->pcix_82544 = 1;
1553
1554         ew32(TCTL, tctl);
1555
1556 }
1557
1558 /**
1559  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1560  * @adapter: board private structure
1561  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1562  *
1563  * Returns 0 on success, negative on failure
1564  **/
1565
1566 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
1567                                     struct e1000_rx_ring *rxdr)
1568 {
1569         struct pci_dev *pdev = adapter->pdev;
1570         int size, desc_len;
1571
1572         size = sizeof(struct e1000_buffer) * rxdr->count;
1573         rxdr->buffer_info = vmalloc(size);
1574         if (!rxdr->buffer_info) {
1575                 DPRINTK(PROBE, ERR,
1576                 "Unable to allocate memory for the receive descriptor ring\n");
1577                 return -ENOMEM;
1578         }
1579         memset(rxdr->buffer_info, 0, size);
1580
1581         desc_len = sizeof(struct e1000_rx_desc);
1582
1583         /* Round up to nearest 4K */
1584
1585         rxdr->size = rxdr->count * desc_len;
1586         rxdr->size = ALIGN(rxdr->size, 4096);
1587
1588         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1589
1590         if (!rxdr->desc) {
1591                 DPRINTK(PROBE, ERR,
1592                 "Unable to allocate memory for the receive descriptor ring\n");
1593 setup_rx_desc_die:
1594                 vfree(rxdr->buffer_info);
1595                 return -ENOMEM;
1596         }
1597
1598         /* Fix for errata 23, can't cross 64kB boundary */
1599         if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1600                 void *olddesc = rxdr->desc;
1601                 dma_addr_t olddma = rxdr->dma;
1602                 DPRINTK(RX_ERR, ERR, "rxdr align check failed: %u bytes "
1603                                      "at %p\n", rxdr->size, rxdr->desc);
1604                 /* Try again, without freeing the previous */
1605                 rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1606                 /* Failed allocation, critical failure */
1607                 if (!rxdr->desc) {
1608                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1609                         DPRINTK(PROBE, ERR,
1610                                 "Unable to allocate memory "
1611                                 "for the receive descriptor ring\n");
1612                         goto setup_rx_desc_die;
1613                 }
1614
1615                 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1616                         /* give up */
1617                         pci_free_consistent(pdev, rxdr->size, rxdr->desc,
1618                                             rxdr->dma);
1619                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1620                         DPRINTK(PROBE, ERR,
1621                                 "Unable to allocate aligned memory "
1622                                 "for the receive descriptor ring\n");
1623                         goto setup_rx_desc_die;
1624                 } else {
1625                         /* Free old allocation, new allocation was successful */
1626                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1627                 }
1628         }
1629         memset(rxdr->desc, 0, rxdr->size);
1630
1631         rxdr->next_to_clean = 0;
1632         rxdr->next_to_use = 0;
1633         rxdr->rx_skb_top = NULL;
1634
1635         return 0;
1636 }
1637
1638 /**
1639  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1640  *                                (Descriptors) for all queues
1641  * @adapter: board private structure
1642  *
1643  * Return 0 on success, negative on failure
1644  **/
1645
1646 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1647 {
1648         int i, err = 0;
1649
1650         for (i = 0; i < adapter->num_rx_queues; i++) {
1651                 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1652                 if (err) {
1653                         DPRINTK(PROBE, ERR,
1654                                 "Allocation for Rx Queue %u failed\n", i);
1655                         for (i-- ; i >= 0; i--)
1656                                 e1000_free_rx_resources(adapter,
1657                                                         &adapter->rx_ring[i]);
1658                         break;
1659                 }
1660         }
1661
1662         return err;
1663 }
1664
1665 /**
1666  * e1000_setup_rctl - configure the receive control registers
1667  * @adapter: Board private structure
1668  **/
1669 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1670 {
1671         struct e1000_hw *hw = &adapter->hw;
1672         u32 rctl;
1673
1674         rctl = er32(RCTL);
1675
1676         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1677
1678         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1679                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1680                 (hw->mc_filter_type << E1000_RCTL_MO_SHIFT);
1681
1682         if (hw->tbi_compatibility_on == 1)
1683                 rctl |= E1000_RCTL_SBP;
1684         else
1685                 rctl &= ~E1000_RCTL_SBP;
1686
1687         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1688                 rctl &= ~E1000_RCTL_LPE;
1689         else
1690                 rctl |= E1000_RCTL_LPE;
1691
1692         /* Setup buffer sizes */
1693         rctl &= ~E1000_RCTL_SZ_4096;
1694         rctl |= E1000_RCTL_BSEX;
1695         switch (adapter->rx_buffer_len) {
1696                 case E1000_RXBUFFER_256:
1697                         rctl |= E1000_RCTL_SZ_256;
1698                         rctl &= ~E1000_RCTL_BSEX;
1699                         break;
1700                 case E1000_RXBUFFER_512:
1701                         rctl |= E1000_RCTL_SZ_512;
1702                         rctl &= ~E1000_RCTL_BSEX;
1703                         break;
1704                 case E1000_RXBUFFER_1024:
1705                         rctl |= E1000_RCTL_SZ_1024;
1706                         rctl &= ~E1000_RCTL_BSEX;
1707                         break;
1708                 case E1000_RXBUFFER_2048:
1709                 default:
1710                         rctl |= E1000_RCTL_SZ_2048;
1711                         rctl &= ~E1000_RCTL_BSEX;
1712                         break;
1713                 case E1000_RXBUFFER_4096:
1714                         rctl |= E1000_RCTL_SZ_4096;
1715                         break;
1716                 case E1000_RXBUFFER_8192:
1717                         rctl |= E1000_RCTL_SZ_8192;
1718                         break;
1719                 case E1000_RXBUFFER_16384:
1720                         rctl |= E1000_RCTL_SZ_16384;
1721                         break;
1722         }
1723
1724         ew32(RCTL, rctl);
1725 }
1726
1727 /**
1728  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1729  * @adapter: board private structure
1730  *
1731  * Configure the Rx unit of the MAC after a reset.
1732  **/
1733
1734 static void e1000_configure_rx(struct e1000_adapter *adapter)
1735 {
1736         u64 rdba;
1737         struct e1000_hw *hw = &adapter->hw;
1738         u32 rdlen, rctl, rxcsum;
1739
1740         if (adapter->netdev->mtu > ETH_DATA_LEN) {
1741                 rdlen = adapter->rx_ring[0].count *
1742                         sizeof(struct e1000_rx_desc);
1743                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
1744                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
1745         } else {
1746                 rdlen = adapter->rx_ring[0].count *
1747                         sizeof(struct e1000_rx_desc);
1748                 adapter->clean_rx = e1000_clean_rx_irq;
1749                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1750         }
1751
1752         /* disable receives while setting up the descriptors */
1753         rctl = er32(RCTL);
1754         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1755
1756         /* set the Receive Delay Timer Register */
1757         ew32(RDTR, adapter->rx_int_delay);
1758
1759         if (hw->mac_type >= e1000_82540) {
1760                 ew32(RADV, adapter->rx_abs_int_delay);
1761                 if (adapter->itr_setting != 0)
1762                         ew32(ITR, 1000000000 / (adapter->itr * 256));
1763         }
1764
1765         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1766          * the Base and Length of the Rx Descriptor Ring */
1767         switch (adapter->num_rx_queues) {
1768         case 1:
1769         default:
1770                 rdba = adapter->rx_ring[0].dma;
1771                 ew32(RDLEN, rdlen);
1772                 ew32(RDBAH, (rdba >> 32));
1773                 ew32(RDBAL, (rdba & 0x00000000ffffffffULL));
1774                 ew32(RDT, 0);
1775                 ew32(RDH, 0);
1776                 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ? E1000_RDH : E1000_82542_RDH);
1777                 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ? E1000_RDT : E1000_82542_RDT);
1778                 break;
1779         }
1780
1781         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
1782         if (hw->mac_type >= e1000_82543) {
1783                 rxcsum = er32(RXCSUM);
1784                 if (adapter->rx_csum)
1785                         rxcsum |= E1000_RXCSUM_TUOFL;
1786                 else
1787                         /* don't need to clear IPPCSE as it defaults to 0 */
1788                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1789                 ew32(RXCSUM, rxcsum);
1790         }
1791
1792         /* Enable Receives */
1793         ew32(RCTL, rctl);
1794 }
1795
1796 /**
1797  * e1000_free_tx_resources - Free Tx Resources per Queue
1798  * @adapter: board private structure
1799  * @tx_ring: Tx descriptor ring for a specific queue
1800  *
1801  * Free all transmit software resources
1802  **/
1803
1804 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
1805                                     struct e1000_tx_ring *tx_ring)
1806 {
1807         struct pci_dev *pdev = adapter->pdev;
1808
1809         e1000_clean_tx_ring(adapter, tx_ring);
1810
1811         vfree(tx_ring->buffer_info);
1812         tx_ring->buffer_info = NULL;
1813
1814         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
1815
1816         tx_ring->desc = NULL;
1817 }
1818
1819 /**
1820  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1821  * @adapter: board private structure
1822  *
1823  * Free all transmit software resources
1824  **/
1825
1826 void e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1827 {
1828         int i;
1829
1830         for (i = 0; i < adapter->num_tx_queues; i++)
1831                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1832 }
1833
1834 static void e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1835                                              struct e1000_buffer *buffer_info)
1836 {
1837         buffer_info->dma = 0;
1838         if (buffer_info->skb) {
1839                 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
1840                               DMA_TO_DEVICE);
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 > 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                 list_for_each_entry(ha, &netdev->uc.list, list) {
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 {
2244                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2245                 }
2246         }
2247 }
2248
2249 /**
2250  * e1000_watchdog - Timer Call-back
2251  * @data: pointer to adapter cast into an unsigned long
2252  **/
2253 static void e1000_watchdog(unsigned long data)
2254 {
2255         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2256         struct e1000_hw *hw = &adapter->hw;
2257         struct net_device *netdev = adapter->netdev;
2258         struct e1000_tx_ring *txdr = adapter->tx_ring;
2259         u32 link, tctl;
2260
2261         e1000_check_for_link(hw);
2262
2263         if ((hw->media_type == e1000_media_type_internal_serdes) &&
2264            !(er32(TXCW) & E1000_TXCW_ANE))
2265                 link = !hw->serdes_link_down;
2266         else
2267                 link = er32(STATUS) & E1000_STATUS_LU;
2268
2269         if (link) {
2270                 if (!netif_carrier_ok(netdev)) {
2271                         u32 ctrl;
2272                         bool txb2b = true;
2273                         e1000_get_speed_and_duplex(hw,
2274                                                    &adapter->link_speed,
2275                                                    &adapter->link_duplex);
2276
2277                         ctrl = er32(CTRL);
2278                         printk(KERN_INFO "e1000: %s NIC Link is Up %d Mbps %s, "
2279                                "Flow Control: %s\n",
2280                                netdev->name,
2281                                adapter->link_speed,
2282                                adapter->link_duplex == FULL_DUPLEX ?
2283                                 "Full Duplex" : "Half Duplex",
2284                                 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2285                                 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2286                                 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2287                                 E1000_CTRL_TFCE) ? "TX" : "None" )));
2288
2289                         /* tweak tx_queue_len according to speed/duplex
2290                          * and adjust the timeout factor */
2291                         netdev->tx_queue_len = adapter->tx_queue_len;
2292                         adapter->tx_timeout_factor = 1;
2293                         switch (adapter->link_speed) {
2294                         case SPEED_10:
2295                                 txb2b = false;
2296                                 netdev->tx_queue_len = 10;
2297                                 adapter->tx_timeout_factor = 8;
2298                                 break;
2299                         case SPEED_100:
2300                                 txb2b = false;
2301                                 netdev->tx_queue_len = 100;
2302                                 /* maybe add some timeout factor ? */
2303                                 break;
2304                         }
2305
2306                         /* enable transmits in the hardware */
2307                         tctl = er32(TCTL);
2308                         tctl |= E1000_TCTL_EN;
2309                         ew32(TCTL, tctl);
2310
2311                         netif_carrier_on(netdev);
2312                         mod_timer(&adapter->phy_info_timer,
2313                                   round_jiffies(jiffies + 2 * HZ));
2314                         adapter->smartspeed = 0;
2315                 }
2316         } else {
2317                 if (netif_carrier_ok(netdev)) {
2318                         adapter->link_speed = 0;
2319                         adapter->link_duplex = 0;
2320                         printk(KERN_INFO "e1000: %s NIC Link is Down\n",
2321                                netdev->name);
2322                         netif_carrier_off(netdev);
2323                         mod_timer(&adapter->phy_info_timer,
2324                                   round_jiffies(jiffies + 2 * HZ));
2325                 }
2326
2327                 e1000_smartspeed(adapter);
2328         }
2329
2330         e1000_update_stats(adapter);
2331
2332         hw->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2333         adapter->tpt_old = adapter->stats.tpt;
2334         hw->collision_delta = adapter->stats.colc - adapter->colc_old;
2335         adapter->colc_old = adapter->stats.colc;
2336
2337         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2338         adapter->gorcl_old = adapter->stats.gorcl;
2339         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2340         adapter->gotcl_old = adapter->stats.gotcl;
2341
2342         e1000_update_adaptive(hw);
2343
2344         if (!netif_carrier_ok(netdev)) {
2345                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2346                         /* We've lost link, so the controller stops DMA,
2347                          * but we've got queued Tx work that's never going
2348                          * to get done, so reset controller to flush Tx.
2349                          * (Do the reset outside of interrupt context). */
2350                         adapter->tx_timeout_count++;
2351                         schedule_work(&adapter->reset_task);
2352                         /* return immediately since reset is imminent */
2353                         return;
2354                 }
2355         }
2356
2357         /* Cause software interrupt to ensure rx ring is cleaned */
2358         ew32(ICS, E1000_ICS_RXDMT0);
2359
2360         /* Force detection of hung controller every watchdog period */
2361         adapter->detect_tx_hung = true;
2362
2363         /* Reset the timer */
2364         mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
2365 }
2366
2367 enum latency_range {
2368         lowest_latency = 0,
2369         low_latency = 1,
2370         bulk_latency = 2,
2371         latency_invalid = 255
2372 };
2373
2374 /**
2375  * e1000_update_itr - update the dynamic ITR value based on statistics
2376  *      Stores a new ITR value based on packets and byte
2377  *      counts during the last interrupt.  The advantage of per interrupt
2378  *      computation is faster updates and more accurate ITR for the current
2379  *      traffic pattern.  Constants in this function were computed
2380  *      based on theoretical maximum wire speed and thresholds were set based
2381  *      on testing data as well as attempting to minimize response time
2382  *      while increasing bulk throughput.
2383  *      this functionality is controlled by the InterruptThrottleRate module
2384  *      parameter (see e1000_param.c)
2385  * @adapter: pointer to adapter
2386  * @itr_setting: current adapter->itr
2387  * @packets: the number of packets during this measurement interval
2388  * @bytes: the number of bytes during this measurement interval
2389  **/
2390 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2391                                      u16 itr_setting, int packets, int bytes)
2392 {
2393         unsigned int retval = itr_setting;
2394         struct e1000_hw *hw = &adapter->hw;
2395
2396         if (unlikely(hw->mac_type < e1000_82540))
2397                 goto update_itr_done;
2398
2399         if (packets == 0)
2400                 goto update_itr_done;
2401
2402         switch (itr_setting) {
2403         case lowest_latency:
2404                 /* jumbo frames get bulk treatment*/
2405                 if (bytes/packets > 8000)
2406                         retval = bulk_latency;
2407                 else if ((packets < 5) && (bytes > 512))
2408                         retval = low_latency;
2409                 break;
2410         case low_latency:  /* 50 usec aka 20000 ints/s */
2411                 if (bytes > 10000) {
2412                         /* jumbo frames need bulk latency setting */
2413                         if (bytes/packets > 8000)
2414                                 retval = bulk_latency;
2415                         else if ((packets < 10) || ((bytes/packets) > 1200))
2416                                 retval = bulk_latency;
2417                         else if ((packets > 35))
2418                                 retval = lowest_latency;
2419                 } else if (bytes/packets > 2000)
2420                         retval = bulk_latency;
2421                 else if (packets <= 2 && bytes < 512)
2422                         retval = lowest_latency;
2423                 break;
2424         case bulk_latency: /* 250 usec aka 4000 ints/s */
2425                 if (bytes > 25000) {
2426                         if (packets > 35)
2427                                 retval = low_latency;
2428                 } else if (bytes < 6000) {
2429                         retval = low_latency;
2430                 }
2431                 break;
2432         }
2433
2434 update_itr_done:
2435         return retval;
2436 }
2437
2438 static void e1000_set_itr(struct e1000_adapter *adapter)
2439 {
2440         struct e1000_hw *hw = &adapter->hw;
2441         u16 current_itr;
2442         u32 new_itr = adapter->itr;
2443
2444         if (unlikely(hw->mac_type < e1000_82540))
2445                 return;
2446
2447         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2448         if (unlikely(adapter->link_speed != SPEED_1000)) {
2449                 current_itr = 0;
2450                 new_itr = 4000;
2451                 goto set_itr_now;
2452         }
2453
2454         adapter->tx_itr = e1000_update_itr(adapter,
2455                                     adapter->tx_itr,
2456                                     adapter->total_tx_packets,
2457                                     adapter->total_tx_bytes);
2458         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2459         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2460                 adapter->tx_itr = low_latency;
2461
2462         adapter->rx_itr = e1000_update_itr(adapter,
2463                                     adapter->rx_itr,
2464                                     adapter->total_rx_packets,
2465                                     adapter->total_rx_bytes);
2466         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2467         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2468                 adapter->rx_itr = low_latency;
2469
2470         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2471
2472         switch (current_itr) {
2473         /* counts and packets in update_itr are dependent on these numbers */
2474         case lowest_latency:
2475                 new_itr = 70000;
2476                 break;
2477         case low_latency:
2478                 new_itr = 20000; /* aka hwitr = ~200 */
2479                 break;
2480         case bulk_latency:
2481                 new_itr = 4000;
2482                 break;
2483         default:
2484                 break;
2485         }
2486
2487 set_itr_now:
2488         if (new_itr != adapter->itr) {
2489                 /* this attempts to bias the interrupt rate towards Bulk
2490                  * by adding intermediate steps when interrupt rate is
2491                  * increasing */
2492                 new_itr = new_itr > adapter->itr ?
2493                              min(adapter->itr + (new_itr >> 2), new_itr) :
2494                              new_itr;
2495                 adapter->itr = new_itr;
2496                 ew32(ITR, 1000000000 / (new_itr * 256));
2497         }
2498
2499         return;
2500 }
2501
2502 #define E1000_TX_FLAGS_CSUM             0x00000001
2503 #define E1000_TX_FLAGS_VLAN             0x00000002
2504 #define E1000_TX_FLAGS_TSO              0x00000004
2505 #define E1000_TX_FLAGS_IPV4             0x00000008
2506 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2507 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2508
2509 static int e1000_tso(struct e1000_adapter *adapter,
2510                      struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2511 {
2512         struct e1000_context_desc *context_desc;
2513         struct e1000_buffer *buffer_info;
2514         unsigned int i;
2515         u32 cmd_length = 0;
2516         u16 ipcse = 0, tucse, mss;
2517         u8 ipcss, ipcso, tucss, tucso, hdr_len;
2518         int err;
2519
2520         if (skb_is_gso(skb)) {
2521                 if (skb_header_cloned(skb)) {
2522                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2523                         if (err)
2524                                 return err;
2525                 }
2526
2527                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2528                 mss = skb_shinfo(skb)->gso_size;
2529                 if (skb->protocol == htons(ETH_P_IP)) {
2530                         struct iphdr *iph = ip_hdr(skb);
2531                         iph->tot_len = 0;
2532                         iph->check = 0;
2533                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2534                                                                  iph->daddr, 0,
2535                                                                  IPPROTO_TCP,
2536                                                                  0);
2537                         cmd_length = E1000_TXD_CMD_IP;
2538                         ipcse = skb_transport_offset(skb) - 1;
2539                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2540                         ipv6_hdr(skb)->payload_len = 0;
2541                         tcp_hdr(skb)->check =
2542                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2543                                                  &ipv6_hdr(skb)->daddr,
2544                                                  0, IPPROTO_TCP, 0);
2545                         ipcse = 0;
2546                 }
2547                 ipcss = skb_network_offset(skb);
2548                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2549                 tucss = skb_transport_offset(skb);
2550                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2551                 tucse = 0;
2552
2553                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2554                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2555
2556                 i = tx_ring->next_to_use;
2557                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2558                 buffer_info = &tx_ring->buffer_info[i];
2559
2560                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2561                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2562                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2563                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2564                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2565                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2566                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2567                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2568                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2569
2570                 buffer_info->time_stamp = jiffies;
2571                 buffer_info->next_to_watch = i;
2572
2573                 if (++i == tx_ring->count) i = 0;
2574                 tx_ring->next_to_use = i;
2575
2576                 return true;
2577         }
2578         return false;
2579 }
2580
2581 static bool e1000_tx_csum(struct e1000_adapter *adapter,
2582                           struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2583 {
2584         struct e1000_context_desc *context_desc;
2585         struct e1000_buffer *buffer_info;
2586         unsigned int i;
2587         u8 css;
2588         u32 cmd_len = E1000_TXD_CMD_DEXT;
2589
2590         if (skb->ip_summed != CHECKSUM_PARTIAL)
2591                 return false;
2592
2593         switch (skb->protocol) {
2594         case cpu_to_be16(ETH_P_IP):
2595                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2596                         cmd_len |= E1000_TXD_CMD_TCP;
2597                 break;
2598         case cpu_to_be16(ETH_P_IPV6):
2599                 /* XXX not handling all IPV6 headers */
2600                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2601                         cmd_len |= E1000_TXD_CMD_TCP;
2602                 break;
2603         default:
2604                 if (unlikely(net_ratelimit()))
2605                         DPRINTK(DRV, WARNING,
2606                                 "checksum_partial proto=%x!\n", skb->protocol);
2607                 break;
2608         }
2609
2610         css = skb_transport_offset(skb);
2611
2612         i = tx_ring->next_to_use;
2613         buffer_info = &tx_ring->buffer_info[i];
2614         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2615
2616         context_desc->lower_setup.ip_config = 0;
2617         context_desc->upper_setup.tcp_fields.tucss = css;
2618         context_desc->upper_setup.tcp_fields.tucso =
2619                 css + skb->csum_offset;
2620         context_desc->upper_setup.tcp_fields.tucse = 0;
2621         context_desc->tcp_seg_setup.data = 0;
2622         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
2623
2624         buffer_info->time_stamp = jiffies;
2625         buffer_info->next_to_watch = i;
2626
2627         if (unlikely(++i == tx_ring->count)) i = 0;
2628         tx_ring->next_to_use = i;
2629
2630         return true;
2631 }
2632
2633 #define E1000_MAX_TXD_PWR       12
2634 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2635
2636 static int e1000_tx_map(struct e1000_adapter *adapter,
2637                         struct e1000_tx_ring *tx_ring,
2638                         struct sk_buff *skb, unsigned int first,
2639                         unsigned int max_per_txd, unsigned int nr_frags,
2640                         unsigned int mss)
2641 {
2642         struct e1000_hw *hw = &adapter->hw;
2643         struct e1000_buffer *buffer_info;
2644         unsigned int len = skb_headlen(skb);
2645         unsigned int offset, size, count = 0, i;
2646         unsigned int f;
2647         dma_addr_t *map;
2648
2649         i = tx_ring->next_to_use;
2650
2651         if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
2652                 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
2653                 return 0;
2654         }
2655
2656         map = skb_shinfo(skb)->dma_maps;
2657         offset = 0;
2658
2659         while (len) {
2660                 buffer_info = &tx_ring->buffer_info[i];
2661                 size = min(len, max_per_txd);
2662                 /* Workaround for Controller erratum --
2663                  * descriptor for non-tso packet in a linear SKB that follows a
2664                  * tso gets written back prematurely before the data is fully
2665                  * DMA'd to the controller */
2666                 if (!skb->data_len && tx_ring->last_tx_tso &&
2667                     !skb_is_gso(skb)) {
2668                         tx_ring->last_tx_tso = 0;
2669                         size -= 4;
2670                 }
2671
2672                 /* Workaround for premature desc write-backs
2673                  * in TSO mode.  Append 4-byte sentinel desc */
2674                 if (unlikely(mss && !nr_frags && size == len && size > 8))
2675                         size -= 4;
2676                 /* work-around for errata 10 and it applies
2677                  * to all controllers in PCI-X mode
2678                  * The fix is to make sure that the first descriptor of a
2679                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2680                  */
2681                 if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
2682                                 (size > 2015) && count == 0))
2683                         size = 2015;
2684
2685                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
2686                  * terminating buffers within evenly-aligned dwords. */
2687                 if (unlikely(adapter->pcix_82544 &&
2688                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2689                    size > 4))
2690                         size -= 4;
2691
2692                 buffer_info->length = size;
2693                 buffer_info->dma = skb_shinfo(skb)->dma_head + offset;
2694                 buffer_info->time_stamp = jiffies;
2695                 buffer_info->next_to_watch = i;
2696
2697                 len -= size;
2698                 offset += size;
2699                 count++;
2700                 if (len) {
2701                         i++;
2702                         if (unlikely(i == tx_ring->count))
2703                                 i = 0;
2704                 }
2705         }
2706
2707         for (f = 0; f < nr_frags; f++) {
2708                 struct skb_frag_struct *frag;
2709
2710                 frag = &skb_shinfo(skb)->frags[f];
2711                 len = frag->size;
2712                 offset = 0;
2713
2714                 while (len) {
2715                         i++;
2716                         if (unlikely(i == tx_ring->count))
2717                                 i = 0;
2718
2719                         buffer_info = &tx_ring->buffer_info[i];
2720                         size = min(len, max_per_txd);
2721                         /* Workaround for premature desc write-backs
2722                          * in TSO mode.  Append 4-byte sentinel desc */
2723                         if (unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
2724                                 size -= 4;
2725                         /* Workaround for potential 82544 hang in PCI-X.
2726                          * Avoid terminating buffers within evenly-aligned
2727                          * dwords. */
2728                         if (unlikely(adapter->pcix_82544 &&
2729                            !((unsigned long)(frag->page+offset+size-1) & 4) &&
2730                            size > 4))
2731                                 size -= 4;
2732
2733                         buffer_info->length = size;
2734                         buffer_info->dma = map[f] + offset;
2735                         buffer_info->time_stamp = jiffies;
2736                         buffer_info->next_to_watch = i;
2737
2738                         len -= size;
2739                         offset += size;
2740                         count++;
2741                 }
2742         }
2743
2744         tx_ring->buffer_info[i].skb = skb;
2745         tx_ring->buffer_info[first].next_to_watch = i;
2746
2747         return count;
2748 }
2749
2750 static void e1000_tx_queue(struct e1000_adapter *adapter,
2751                            struct e1000_tx_ring *tx_ring, int tx_flags,
2752                            int count)
2753 {
2754         struct e1000_hw *hw = &adapter->hw;
2755         struct e1000_tx_desc *tx_desc = NULL;
2756         struct e1000_buffer *buffer_info;
2757         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2758         unsigned int i;
2759
2760         if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
2761                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2762                              E1000_TXD_CMD_TSE;
2763                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2764
2765                 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
2766                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2767         }
2768
2769         if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
2770                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
2771                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2772         }
2773
2774         if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
2775                 txd_lower |= E1000_TXD_CMD_VLE;
2776                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
2777         }
2778
2779         i = tx_ring->next_to_use;
2780
2781         while (count--) {
2782                 buffer_info = &tx_ring->buffer_info[i];
2783                 tx_desc = E1000_TX_DESC(*tx_ring, i);
2784                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
2785                 tx_desc->lower.data =
2786                         cpu_to_le32(txd_lower | buffer_info->length);
2787                 tx_desc->upper.data = cpu_to_le32(txd_upper);
2788                 if (unlikely(++i == tx_ring->count)) i = 0;
2789         }
2790
2791         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
2792
2793         /* Force memory writes to complete before letting h/w
2794          * know there are new descriptors to fetch.  (Only
2795          * applicable for weak-ordered memory model archs,
2796          * such as IA-64). */
2797         wmb();
2798
2799         tx_ring->next_to_use = i;
2800         writel(i, hw->hw_addr + tx_ring->tdt);
2801         /* we need this if more than one processor can write to our tail
2802          * at a time, it syncronizes IO on IA64/Altix systems */
2803         mmiowb();
2804 }
2805
2806 /**
2807  * 82547 workaround to avoid controller hang in half-duplex environment.
2808  * The workaround is to avoid queuing a large packet that would span
2809  * the internal Tx FIFO ring boundary by notifying the stack to resend
2810  * the packet at a later time.  This gives the Tx FIFO an opportunity to
2811  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
2812  * to the beginning of the Tx FIFO.
2813  **/
2814
2815 #define E1000_FIFO_HDR                  0x10
2816 #define E1000_82547_PAD_LEN             0x3E0
2817
2818 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
2819                                        struct sk_buff *skb)
2820 {
2821         u32 fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
2822         u32 skb_fifo_len = skb->len + E1000_FIFO_HDR;
2823
2824         skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
2825
2826         if (adapter->link_duplex != HALF_DUPLEX)
2827                 goto no_fifo_stall_required;
2828
2829         if (atomic_read(&adapter->tx_fifo_stall))
2830                 return 1;
2831
2832         if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
2833                 atomic_set(&adapter->tx_fifo_stall, 1);
2834                 return 1;
2835         }
2836
2837 no_fifo_stall_required:
2838         adapter->tx_fifo_head += skb_fifo_len;
2839         if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
2840                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
2841         return 0;
2842 }
2843
2844 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
2845 {
2846         struct e1000_adapter *adapter = netdev_priv(netdev);
2847         struct e1000_tx_ring *tx_ring = adapter->tx_ring;
2848
2849         netif_stop_queue(netdev);
2850         /* Herbert's original patch had:
2851          *  smp_mb__after_netif_stop_queue();
2852          * but since that doesn't exist yet, just open code it. */
2853         smp_mb();
2854
2855         /* We need to check again in a case another CPU has just
2856          * made room available. */
2857         if (likely(E1000_DESC_UNUSED(tx_ring) < size))
2858                 return -EBUSY;
2859
2860         /* A reprieve! */
2861         netif_start_queue(netdev);
2862         ++adapter->restart_queue;
2863         return 0;
2864 }
2865
2866 static int e1000_maybe_stop_tx(struct net_device *netdev,
2867                                struct e1000_tx_ring *tx_ring, int size)
2868 {
2869         if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
2870                 return 0;
2871         return __e1000_maybe_stop_tx(netdev, size);
2872 }
2873
2874 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
2875 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
2876                                     struct net_device *netdev)
2877 {
2878         struct e1000_adapter *adapter = netdev_priv(netdev);
2879         struct e1000_hw *hw = &adapter->hw;
2880         struct e1000_tx_ring *tx_ring;
2881         unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
2882         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
2883         unsigned int tx_flags = 0;
2884         unsigned int len = skb->len - skb->data_len;
2885         unsigned int nr_frags;
2886         unsigned int mss;
2887         int count = 0;
2888         int tso;
2889         unsigned int f;
2890
2891         /* This goes back to the question of how to logically map a tx queue
2892          * to a flow.  Right now, performance is impacted slightly negatively
2893          * if using multiple tx queues.  If the stack breaks away from a
2894          * single qdisc implementation, we can look at this again. */
2895         tx_ring = adapter->tx_ring;
2896
2897         if (unlikely(skb->len <= 0)) {
2898                 dev_kfree_skb_any(skb);
2899                 return NETDEV_TX_OK;
2900         }
2901
2902         mss = skb_shinfo(skb)->gso_size;
2903         /* The controller does a simple calculation to
2904          * make sure there is enough room in the FIFO before
2905          * initiating the DMA for each buffer.  The calc is:
2906          * 4 = ceil(buffer len/mss).  To make sure we don't
2907          * overrun the FIFO, adjust the max buffer len if mss
2908          * drops. */
2909         if (mss) {
2910                 u8 hdr_len;
2911                 max_per_txd = min(mss << 2, max_per_txd);
2912                 max_txd_pwr = fls(max_per_txd) - 1;
2913
2914                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2915                 if (skb->data_len && hdr_len == len) {
2916                         switch (hw->mac_type) {
2917                                 unsigned int pull_size;
2918                         case e1000_82544:
2919                                 /* Make sure we have room to chop off 4 bytes,
2920                                  * and that the end alignment will work out to
2921                                  * this hardware's requirements
2922                                  * NOTE: this is a TSO only workaround
2923                                  * if end byte alignment not correct move us
2924                                  * into the next dword */
2925                                 if ((unsigned long)(skb_tail_pointer(skb) - 1) & 4)
2926                                         break;
2927                                 /* fall through */
2928                                 pull_size = min((unsigned int)4, skb->data_len);
2929                                 if (!__pskb_pull_tail(skb, pull_size)) {
2930                                         DPRINTK(DRV, ERR,
2931                                                 "__pskb_pull_tail failed.\n");
2932                                         dev_kfree_skb_any(skb);
2933                                         return NETDEV_TX_OK;
2934                                 }
2935                                 len = skb->len - skb->data_len;
2936                                 break;
2937                         default:
2938                                 /* do nothing */
2939                                 break;
2940                         }
2941                 }
2942         }
2943
2944         /* reserve a descriptor for the offload context */
2945         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
2946                 count++;
2947         count++;
2948
2949         /* Controller Erratum workaround */
2950         if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
2951                 count++;
2952
2953         count += TXD_USE_COUNT(len, max_txd_pwr);
2954
2955         if (adapter->pcix_82544)
2956                 count++;
2957
2958         /* work-around for errata 10 and it applies to all controllers
2959          * in PCI-X mode, so add one more descriptor to the count
2960          */
2961         if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
2962                         (len > 2015)))
2963                 count++;
2964
2965         nr_frags = skb_shinfo(skb)->nr_frags;
2966         for (f = 0; f < nr_frags; f++)
2967                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
2968                                        max_txd_pwr);
2969         if (adapter->pcix_82544)
2970                 count += nr_frags;
2971
2972         /* need: count + 2 desc gap to keep tail from touching
2973          * head, otherwise try next time */
2974         if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2)))
2975                 return NETDEV_TX_BUSY;
2976
2977         if (unlikely(hw->mac_type == e1000_82547)) {
2978                 if (unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
2979                         netif_stop_queue(netdev);
2980                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2981                         return NETDEV_TX_BUSY;
2982                 }
2983         }
2984
2985         if (unlikely(adapter->vlgrp && vlan_tx_tag_present(skb))) {
2986                 tx_flags |= E1000_TX_FLAGS_VLAN;
2987                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
2988         }
2989
2990         first = tx_ring->next_to_use;
2991
2992         tso = e1000_tso(adapter, tx_ring, skb);
2993         if (tso < 0) {
2994                 dev_kfree_skb_any(skb);
2995                 return NETDEV_TX_OK;
2996         }
2997
2998         if (likely(tso)) {
2999                 tx_ring->last_tx_tso = 1;
3000                 tx_flags |= E1000_TX_FLAGS_TSO;
3001         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3002                 tx_flags |= E1000_TX_FLAGS_CSUM;
3003
3004         if (likely(skb->protocol == htons(ETH_P_IP)))
3005                 tx_flags |= E1000_TX_FLAGS_IPV4;
3006
3007         count = e1000_tx_map(adapter, tx_ring, skb, first, max_per_txd,
3008                              nr_frags, mss);
3009
3010         if (count) {
3011                 e1000_tx_queue(adapter, tx_ring, tx_flags, count);
3012                 /* Make sure there is space in the ring for the next send. */
3013                 e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3014
3015         } else {
3016                 dev_kfree_skb_any(skb);
3017                 tx_ring->buffer_info[first].time_stamp = 0;
3018                 tx_ring->next_to_use = first;
3019         }
3020
3021         return NETDEV_TX_OK;
3022 }
3023
3024 /**
3025  * e1000_tx_timeout - Respond to a Tx Hang
3026  * @netdev: network interface device structure
3027  **/
3028
3029 static void e1000_tx_timeout(struct net_device *netdev)
3030 {
3031         struct e1000_adapter *adapter = netdev_priv(netdev);
3032
3033         /* Do the reset outside of interrupt context */
3034         adapter->tx_timeout_count++;
3035         schedule_work(&adapter->reset_task);
3036 }
3037
3038 static void e1000_reset_task(struct work_struct *work)
3039 {
3040         struct e1000_adapter *adapter =
3041                 container_of(work, struct e1000_adapter, reset_task);
3042
3043         e1000_reinit_locked(adapter);
3044 }
3045
3046 /**
3047  * e1000_get_stats - Get System Network Statistics
3048  * @netdev: network interface device structure
3049  *
3050  * Returns the address of the device statistics structure.
3051  * The statistics are actually updated from the timer callback.
3052  **/
3053
3054 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3055 {
3056         struct e1000_adapter *adapter = netdev_priv(netdev);
3057
3058         /* only return the current stats */
3059         return &adapter->net_stats;
3060 }
3061
3062 /**
3063  * e1000_change_mtu - Change the Maximum Transfer Unit
3064  * @netdev: network interface device structure
3065  * @new_mtu: new value for maximum frame size
3066  *
3067  * Returns 0 on success, negative on failure
3068  **/
3069
3070 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3071 {
3072         struct e1000_adapter *adapter = netdev_priv(netdev);
3073         struct e1000_hw *hw = &adapter->hw;
3074         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3075
3076         if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3077             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3078                 DPRINTK(PROBE, ERR, "Invalid MTU setting\n");
3079                 return -EINVAL;
3080         }
3081
3082         /* Adapter-specific max frame size limits. */
3083         switch (hw->mac_type) {
3084         case e1000_undefined ... e1000_82542_rev2_1:
3085                 if (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)) {
3086                         DPRINTK(PROBE, ERR, "Jumbo Frames not supported.\n");
3087                         return -EINVAL;
3088                 }
3089                 break;
3090         default:
3091                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3092                 break;
3093         }
3094
3095         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3096          * means we reserve 2 more, this pushes us to allocate from the next
3097          * larger slab size.
3098          * i.e. RXBUFFER_2048 --> size-4096 slab
3099          *  however with the new *_jumbo_rx* routines, jumbo receives will use
3100          *  fragmented skbs */
3101
3102         if (max_frame <= E1000_RXBUFFER_256)
3103                 adapter->rx_buffer_len = E1000_RXBUFFER_256;
3104         else if (max_frame <= E1000_RXBUFFER_512)
3105                 adapter->rx_buffer_len = E1000_RXBUFFER_512;
3106         else if (max_frame <= E1000_RXBUFFER_1024)
3107                 adapter->rx_buffer_len = E1000_RXBUFFER_1024;
3108         else if (max_frame <= E1000_RXBUFFER_2048)
3109                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3110         else
3111 #if (PAGE_SIZE >= E1000_RXBUFFER_16384)
3112                 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3113 #elif (PAGE_SIZE >= E1000_RXBUFFER_4096)
3114                 adapter->rx_buffer_len = PAGE_SIZE;
3115 #endif
3116
3117         /* adjust allocation if LPE protects us, and we aren't using SBP */
3118         if (!hw->tbi_compatibility_on &&
3119             ((max_frame == (ETH_FRAME_LEN + ETH_FCS_LEN)) ||
3120              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3121                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3122
3123         netdev->mtu = new_mtu;
3124         hw->max_frame_size = max_frame;
3125
3126         if (netif_running(netdev))
3127                 e1000_reinit_locked(adapter);
3128
3129         return 0;
3130 }
3131
3132 /**
3133  * e1000_update_stats - Update the board statistics counters
3134  * @adapter: board private structure
3135  **/
3136
3137 void e1000_update_stats(struct e1000_adapter *adapter)
3138 {
3139         struct e1000_hw *hw = &adapter->hw;
3140         struct pci_dev *pdev = adapter->pdev;
3141         unsigned long flags;
3142         u16 phy_tmp;
3143
3144 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3145
3146         /*
3147          * Prevent stats update while adapter is being reset, or if the pci
3148          * connection is down.
3149          */
3150         if (adapter->link_speed == 0)
3151                 return;
3152         if (pci_channel_offline(pdev))
3153                 return;
3154
3155         spin_lock_irqsave(&adapter->stats_lock, flags);
3156
3157         /* these counters are modified from e1000_tbi_adjust_stats,
3158          * called from the interrupt context, so they must only
3159          * be written while holding adapter->stats_lock
3160          */
3161
3162         adapter->stats.crcerrs += er32(CRCERRS);
3163         adapter->stats.gprc += er32(GPRC);
3164         adapter->stats.gorcl += er32(GORCL);
3165         adapter->stats.gorch += er32(GORCH);
3166         adapter->stats.bprc += er32(BPRC);
3167         adapter->stats.mprc += er32(MPRC);
3168         adapter->stats.roc += er32(ROC);
3169
3170         adapter->stats.prc64 += er32(PRC64);
3171         adapter->stats.prc127 += er32(PRC127);
3172         adapter->stats.prc255 += er32(PRC255);
3173         adapter->stats.prc511 += er32(PRC511);
3174         adapter->stats.prc1023 += er32(PRC1023);
3175         adapter->stats.prc1522 += er32(PRC1522);
3176
3177         adapter->stats.symerrs += er32(SYMERRS);
3178         adapter->stats.mpc += er32(MPC);
3179         adapter->stats.scc += er32(SCC);
3180         adapter->stats.ecol += er32(ECOL);
3181         adapter->stats.mcc += er32(MCC);
3182         adapter->stats.latecol += er32(LATECOL);
3183         adapter->stats.dc += er32(DC);
3184         adapter->stats.sec += er32(SEC);
3185         adapter->stats.rlec += er32(RLEC);
3186         adapter->stats.xonrxc += er32(XONRXC);
3187         adapter->stats.xontxc += er32(XONTXC);
3188         adapter->stats.xoffrxc += er32(XOFFRXC);
3189         adapter->stats.xofftxc += er32(XOFFTXC);
3190         adapter->stats.fcruc += er32(FCRUC);
3191         adapter->stats.gptc += er32(GPTC);
3192         adapter->stats.gotcl += er32(GOTCL);
3193         adapter->stats.gotch += er32(GOTCH);
3194         adapter->stats.rnbc += er32(RNBC);
3195         adapter->stats.ruc += er32(RUC);
3196         adapter->stats.rfc += er32(RFC);
3197         adapter->stats.rjc += er32(RJC);
3198         adapter->stats.torl += er32(TORL);
3199         adapter->stats.torh += er32(TORH);
3200         adapter->stats.totl += er32(TOTL);
3201         adapter->stats.toth += er32(TOTH);
3202         adapter->stats.tpr += er32(TPR);
3203
3204         adapter->stats.ptc64 += er32(PTC64);
3205         adapter->stats.ptc127 += er32(PTC127);
3206         adapter->stats.ptc255 += er32(PTC255);
3207         adapter->stats.ptc511 += er32(PTC511);
3208         adapter->stats.ptc1023 += er32(PTC1023);
3209         adapter->stats.ptc1522 += er32(PTC1522);
3210
3211         adapter->stats.mptc += er32(MPTC);
3212         adapter->stats.bptc += er32(BPTC);
3213
3214         /* used for adaptive IFS */
3215
3216         hw->tx_packet_delta = er32(TPT);
3217         adapter->stats.tpt += hw->tx_packet_delta;
3218         hw->collision_delta = er32(COLC);
3219         adapter->stats.colc += hw->collision_delta;
3220
3221         if (hw->mac_type >= e1000_82543) {
3222                 adapter->stats.algnerrc += er32(ALGNERRC);
3223                 adapter->stats.rxerrc += er32(RXERRC);
3224                 adapter->stats.tncrs += er32(TNCRS);
3225                 adapter->stats.cexterr += er32(CEXTERR);
3226                 adapter->stats.tsctc += er32(TSCTC);
3227                 adapter->stats.tsctfc += er32(TSCTFC);
3228         }
3229
3230         /* Fill out the OS statistics structure */
3231         adapter->net_stats.multicast = adapter->stats.mprc;
3232         adapter->net_stats.collisions = adapter->stats.colc;
3233
3234         /* Rx Errors */
3235
3236         /* RLEC on some newer hardware can be incorrect so build
3237         * our own version based on RUC and ROC */
3238         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3239                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3240                 adapter->stats.ruc + adapter->stats.roc +
3241                 adapter->stats.cexterr;
3242         adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3243         adapter->net_stats.rx_length_errors = adapter->stats.rlerrc;
3244         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3245         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3246         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3247
3248         /* Tx Errors */
3249         adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3250         adapter->net_stats.tx_errors = adapter->stats.txerrc;
3251         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3252         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3253         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3254         if (hw->bad_tx_carr_stats_fd &&
3255             adapter->link_duplex == FULL_DUPLEX) {
3256                 adapter->net_stats.tx_carrier_errors = 0;
3257                 adapter->stats.tncrs = 0;
3258         }
3259
3260         /* Tx Dropped needs to be maintained elsewhere */
3261
3262         /* Phy Stats */
3263         if (hw->media_type == e1000_media_type_copper) {
3264                 if ((adapter->link_speed == SPEED_1000) &&
3265                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3266                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3267                         adapter->phy_stats.idle_errors += phy_tmp;
3268                 }
3269
3270                 if ((hw->mac_type <= e1000_82546) &&
3271                    (hw->phy_type == e1000_phy_m88) &&
3272                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3273                         adapter->phy_stats.receive_errors += phy_tmp;
3274         }
3275
3276         /* Management Stats */
3277         if (hw->has_smbus) {
3278                 adapter->stats.mgptc += er32(MGTPTC);
3279                 adapter->stats.mgprc += er32(MGTPRC);
3280                 adapter->stats.mgpdc += er32(MGTPDC);
3281         }
3282
3283         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3284 }
3285
3286 /**
3287  * e1000_intr - Interrupt Handler
3288  * @irq: interrupt number
3289  * @data: pointer to a network interface device structure
3290  **/
3291
3292 static irqreturn_t e1000_intr(int irq, void *data)
3293 {
3294         struct net_device *netdev = data;
3295         struct e1000_adapter *adapter = netdev_priv(netdev);
3296         struct e1000_hw *hw = &adapter->hw;
3297         u32 icr = er32(ICR);
3298
3299         if (unlikely((!icr) || test_bit(__E1000_DOWN, &adapter->flags)))
3300                 return IRQ_NONE;  /* Not our interrupt */
3301
3302         if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3303                 hw->get_link_status = 1;
3304                 /* guard against interrupt when we're going down */
3305                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3306                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3307         }
3308
3309         /* disable interrupts, without the synchronize_irq bit */
3310         ew32(IMC, ~0);
3311         E1000_WRITE_FLUSH();
3312
3313         if (likely(napi_schedule_prep(&adapter->napi))) {
3314                 adapter->total_tx_bytes = 0;
3315                 adapter->total_tx_packets = 0;
3316                 adapter->total_rx_bytes = 0;
3317                 adapter->total_rx_packets = 0;
3318                 __napi_schedule(&adapter->napi);
3319         } else {
3320                 /* this really should not happen! if it does it is basically a
3321                  * bug, but not a hard error, so enable ints and continue */
3322                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3323                         e1000_irq_enable(adapter);
3324         }
3325
3326         return IRQ_HANDLED;
3327 }
3328
3329 /**
3330  * e1000_clean - NAPI Rx polling callback
3331  * @adapter: board private structure
3332  **/
3333 static int e1000_clean(struct napi_struct *napi, int budget)
3334 {
3335         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
3336         struct net_device *poll_dev = adapter->netdev;
3337         int tx_cleaned = 0, work_done = 0;
3338
3339         adapter = netdev_priv(poll_dev);
3340
3341         tx_cleaned = e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);
3342
3343         adapter->clean_rx(adapter, &adapter->rx_ring[0],
3344                           &work_done, budget);
3345
3346         if (!tx_cleaned)
3347                 work_done = budget;
3348
3349         /* If budget not fully consumed, exit the polling mode */
3350         if (work_done < budget) {
3351                 if (likely(adapter->itr_setting & 3))
3352                         e1000_set_itr(adapter);
3353                 napi_complete(napi);
3354                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3355                         e1000_irq_enable(adapter);
3356         }
3357
3358         return work_done;
3359 }
3360
3361 /**
3362  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3363  * @adapter: board private structure
3364  **/
3365 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
3366                                struct e1000_tx_ring *tx_ring)
3367 {
3368         struct e1000_hw *hw = &adapter->hw;
3369         struct net_device *netdev = adapter->netdev;
3370         struct e1000_tx_desc *tx_desc, *eop_desc;
3371         struct e1000_buffer *buffer_info;
3372         unsigned int i, eop;
3373         unsigned int count = 0;
3374         unsigned int total_tx_bytes=0, total_tx_packets=0;
3375
3376         i = tx_ring->next_to_clean;
3377         eop = tx_ring->buffer_info[i].next_to_watch;
3378         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3379
3380         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
3381                (count < tx_ring->count)) {
3382                 bool cleaned = false;
3383                 for ( ; !cleaned; count++) {
3384                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3385                         buffer_info = &tx_ring->buffer_info[i];
3386                         cleaned = (i == eop);
3387
3388                         if (cleaned) {
3389                                 struct sk_buff *skb = buffer_info->skb;
3390                                 unsigned int segs, bytecount;
3391                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3392                                 /* multiply data chunks by size of headers */
3393                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3394                                             skb->len;
3395                                 total_tx_packets += segs;
3396                                 total_tx_bytes += bytecount;
3397                         }
3398                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3399                         tx_desc->upper.data = 0;
3400
3401                         if (unlikely(++i == tx_ring->count)) i = 0;
3402                 }
3403
3404                 eop = tx_ring->buffer_info[i].next_to_watch;
3405                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3406         }
3407
3408         tx_ring->next_to_clean = i;
3409
3410 #define TX_WAKE_THRESHOLD 32
3411         if (unlikely(count && netif_carrier_ok(netdev) &&
3412                      E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
3413                 /* Make sure that anybody stopping the queue after this
3414                  * sees the new next_to_clean.
3415                  */
3416                 smp_mb();
3417                 if (netif_queue_stopped(netdev)) {
3418                         netif_wake_queue(netdev);
3419                         ++adapter->restart_queue;
3420                 }
3421         }
3422
3423         if (adapter->detect_tx_hung) {
3424                 /* Detect a transmit hang in hardware, this serializes the
3425                  * check with the clearing of time_stamp and movement of i */
3426                 adapter->detect_tx_hung = false;
3427                 if (tx_ring->buffer_info[i].time_stamp &&
3428                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
3429                                (adapter->tx_timeout_factor * HZ))
3430                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
3431
3432                         /* detected Tx unit hang */
3433                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
3434                                         "  Tx Queue             <%lu>\n"
3435                                         "  TDH                  <%x>\n"
3436                                         "  TDT                  <%x>\n"
3437                                         "  next_to_use          <%x>\n"
3438                                         "  next_to_clean        <%x>\n"
3439                                         "buffer_info[next_to_clean]\n"
3440                                         "  time_stamp           <%lx>\n"
3441                                         "  next_to_watch        <%x>\n"
3442                                         "  jiffies              <%lx>\n"
3443                                         "  next_to_watch.status <%x>\n",
3444                                 (unsigned long)((tx_ring - adapter->tx_ring) /
3445                                         sizeof(struct e1000_tx_ring)),
3446                                 readl(hw->hw_addr + tx_ring->tdh),
3447                                 readl(hw->hw_addr + tx_ring->tdt),
3448                                 tx_ring->next_to_use,
3449                                 tx_ring->next_to_clean,
3450                                 tx_ring->buffer_info[i].time_stamp,
3451                                 eop,
3452                                 jiffies,
3453                                 eop_desc->upper.fields.status);
3454                         netif_stop_queue(netdev);
3455                 }
3456         }
3457         adapter->total_tx_bytes += total_tx_bytes;
3458         adapter->total_tx_packets += total_tx_packets;
3459         adapter->net_stats.tx_bytes += total_tx_bytes;
3460         adapter->net_stats.tx_packets += total_tx_packets;
3461         return (count < tx_ring->count);
3462 }
3463
3464 /**
3465  * e1000_rx_checksum - Receive Checksum Offload for 82543
3466  * @adapter:     board private structure
3467  * @status_err:  receive descriptor status and error fields
3468  * @csum:        receive descriptor csum field
3469  * @sk_buff:     socket buffer with received data
3470  **/
3471
3472 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
3473                               u32 csum, struct sk_buff *skb)
3474 {
3475         struct e1000_hw *hw = &adapter->hw;
3476         u16 status = (u16)status_err;
3477         u8 errors = (u8)(status_err >> 24);
3478         skb->ip_summed = CHECKSUM_NONE;
3479
3480         /* 82543 or newer only */
3481         if (unlikely(hw->mac_type < e1000_82543)) return;
3482         /* Ignore Checksum bit is set */
3483         if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
3484         /* TCP/UDP checksum error bit is set */
3485         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
3486                 /* let the stack verify checksum errors */
3487                 adapter->hw_csum_err++;
3488                 return;
3489         }
3490         /* TCP/UDP Checksum has not been calculated */
3491         if (!(status & E1000_RXD_STAT_TCPCS))
3492                 return;
3493
3494         /* It must be a TCP or UDP packet with a valid checksum */
3495         if (likely(status & E1000_RXD_STAT_TCPCS)) {
3496                 /* TCP checksum is good */
3497                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3498         }
3499         adapter->hw_csum_good++;
3500 }
3501
3502 /**
3503  * e1000_consume_page - helper function
3504  **/
3505 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
3506                                u16 length)
3507 {
3508         bi->page = NULL;
3509         skb->len += length;
3510         skb->data_len += length;
3511         skb->truesize += length;
3512 }
3513
3514 /**
3515  * e1000_receive_skb - helper function to handle rx indications
3516  * @adapter: board private structure
3517  * @status: descriptor status field as written by hardware
3518  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3519  * @skb: pointer to sk_buff to be indicated to stack
3520  */
3521 static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
3522                               __le16 vlan, struct sk_buff *skb)
3523 {
3524         if (unlikely(adapter->vlgrp && (status & E1000_RXD_STAT_VP))) {
3525                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3526                                          le16_to_cpu(vlan) &
3527                                          E1000_RXD_SPC_VLAN_MASK);
3528         } else {
3529                 netif_receive_skb(skb);
3530         }
3531 }
3532
3533 /**
3534  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
3535  * @adapter: board private structure
3536  * @rx_ring: ring to clean
3537  * @work_done: amount of napi work completed this call
3538  * @work_to_do: max amount of work allowed for this call to do
3539  *
3540  * the return value indicates whether actual cleaning was done, there
3541  * is no guarantee that everything was cleaned
3542  */
3543 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
3544                                      struct e1000_rx_ring *rx_ring,
3545                                      int *work_done, int work_to_do)
3546 {
3547         struct e1000_hw *hw = &adapter->hw;
3548         struct net_device *netdev = adapter->netdev;
3549         struct pci_dev *pdev = adapter->pdev;
3550         struct e1000_rx_desc *rx_desc, *next_rxd;
3551         struct e1000_buffer *buffer_info, *next_buffer;
3552         unsigned long irq_flags;
3553         u32 length;
3554         unsigned int i;
3555         int cleaned_count = 0;
3556         bool cleaned = false;
3557         unsigned int total_rx_bytes=0, total_rx_packets=0;
3558
3559         i = rx_ring->next_to_clean;
3560         rx_desc = E1000_RX_DESC(*rx_ring, i);
3561         buffer_info = &rx_ring->buffer_info[i];
3562
3563         while (rx_desc->status & E1000_RXD_STAT_DD) {
3564                 struct sk_buff *skb;
3565                 u8 status;
3566
3567                 if (*work_done >= work_to_do)
3568                         break;
3569                 (*work_done)++;
3570
3571                 status = rx_desc->status;
3572                 skb = buffer_info->skb;
3573                 buffer_info->skb = NULL;
3574
3575                 if (++i == rx_ring->count) i = 0;
3576                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3577                 prefetch(next_rxd);
3578
3579                 next_buffer = &rx_ring->buffer_info[i];
3580
3581                 cleaned = true;
3582                 cleaned_count++;
3583                 pci_unmap_page(pdev, buffer_info->dma, buffer_info->length,
3584                                PCI_DMA_FROMDEVICE);
3585                 buffer_info->dma = 0;
3586
3587                 length = le16_to_cpu(rx_desc->length);
3588
3589                 /* errors is only valid for DD + EOP descriptors */
3590                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
3591                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
3592                         u8 last_byte = *(skb->data + length - 1);
3593                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3594                                        last_byte)) {
3595                                 spin_lock_irqsave(&adapter->stats_lock,
3596                                                   irq_flags);
3597                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3598                                                        length, skb->data);
3599                                 spin_unlock_irqrestore(&adapter->stats_lock,
3600                                                        irq_flags);
3601                                 length--;
3602                         } else {
3603                                 /* recycle both page and skb */
3604                                 buffer_info->skb = skb;
3605                                 /* an error means any chain goes out the window
3606                                  * too */
3607                                 if (rx_ring->rx_skb_top)
3608                                         dev_kfree_skb(rx_ring->rx_skb_top);
3609                                 rx_ring->rx_skb_top = NULL;
3610                                 goto next_desc;
3611                         }
3612                 }
3613
3614 #define rxtop rx_ring->rx_skb_top
3615                 if (!(status & E1000_RXD_STAT_EOP)) {
3616                         /* this descriptor is only the beginning (or middle) */
3617                         if (!rxtop) {
3618                                 /* this is the beginning of a chain */
3619                                 rxtop = skb;
3620                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
3621                                                    0, length);
3622                         } else {
3623                                 /* this is the middle of a chain */
3624                                 skb_fill_page_desc(rxtop,
3625                                     skb_shinfo(rxtop)->nr_frags,
3626                                     buffer_info->page, 0, length);
3627                                 /* re-use the skb, only consumed the page */
3628                                 buffer_info->skb = skb;
3629                         }
3630                         e1000_consume_page(buffer_info, rxtop, length);
3631                         goto next_desc;
3632                 } else {
3633                         if (rxtop) {
3634                                 /* end of the chain */
3635                                 skb_fill_page_desc(rxtop,
3636                                     skb_shinfo(rxtop)->nr_frags,
3637                                     buffer_info->page, 0, length);
3638                                 /* re-use the current skb, we only consumed the
3639                                  * page */
3640                                 buffer_info->skb = skb;
3641                                 skb = rxtop;
3642                                 rxtop = NULL;
3643                                 e1000_consume_page(buffer_info, skb, length);
3644                         } else {
3645                                 /* no chain, got EOP, this buf is the packet
3646                                  * copybreak to save the put_page/alloc_page */
3647                                 if (length <= copybreak &&
3648                                     skb_tailroom(skb) >= length) {
3649                                         u8 *vaddr;
3650                                         vaddr = kmap_atomic(buffer_info->page,
3651                                                             KM_SKB_DATA_SOFTIRQ);
3652                                         memcpy(skb_tail_pointer(skb), vaddr, length);
3653                                         kunmap_atomic(vaddr,
3654                                                       KM_SKB_DATA_SOFTIRQ);
3655                                         /* re-use the page, so don't erase
3656                                          * buffer_info->page */
3657                                         skb_put(skb, length);
3658                                 } else {
3659                                         skb_fill_page_desc(skb, 0,
3660                                                            buffer_info->page, 0,
3661                                                            length);
3662                                         e1000_consume_page(buffer_info, skb,
3663                                                            length);
3664                                 }
3665                         }
3666                 }
3667
3668                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
3669                 e1000_rx_checksum(adapter,
3670                                   (u32)(status) |
3671                                   ((u32)(rx_desc->errors) << 24),
3672                                   le16_to_cpu(rx_desc->csum), skb);
3673
3674                 pskb_trim(skb, skb->len - 4);
3675
3676                 /* probably a little skewed due to removing CRC */
3677                 total_rx_bytes += skb->len;
3678                 total_rx_packets++;
3679
3680                 /* eth type trans needs skb->data to point to something */
3681                 if (!pskb_may_pull(skb, ETH_HLEN)) {
3682                         DPRINTK(DRV, ERR, "pskb_may_pull failed.\n");
3683                         dev_kfree_skb(skb);
3684                         goto next_desc;
3685                 }
3686
3687                 skb->protocol = eth_type_trans(skb, netdev);
3688
3689                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3690
3691 next_desc:
3692                 rx_desc->status = 0;
3693
3694                 /* return some buffers to hardware, one at a time is too slow */
3695                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3696                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3697                         cleaned_count = 0;
3698                 }
3699
3700                 /* use prefetched values */
3701                 rx_desc = next_rxd;
3702                 buffer_info = next_buffer;
3703         }
3704         rx_ring->next_to_clean = i;
3705
3706         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3707         if (cleaned_count)
3708                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3709
3710         adapter->total_rx_packets += total_rx_packets;
3711         adapter->total_rx_bytes += total_rx_bytes;
3712         adapter->net_stats.rx_bytes += total_rx_bytes;
3713         adapter->net_stats.rx_packets += total_rx_packets;
3714         return cleaned;
3715 }
3716
3717 /**
3718  * e1000_clean_rx_irq - Send received data up the network stack; legacy
3719  * @adapter: board private structure
3720  * @rx_ring: ring to clean
3721  * @work_done: amount of napi work completed this call
3722  * @work_to_do: max amount of work allowed for this call to do
3723  */
3724 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
3725                                struct e1000_rx_ring *rx_ring,
3726                                int *work_done, int work_to_do)
3727 {
3728         struct e1000_hw *hw = &adapter->hw;
3729         struct net_device *netdev = adapter->netdev;
3730         struct pci_dev *pdev = adapter->pdev;
3731         struct e1000_rx_desc *rx_desc, *next_rxd;
3732         struct e1000_buffer *buffer_info, *next_buffer;
3733         unsigned long flags;
3734         u32 length;
3735         unsigned int i;
3736         int cleaned_count = 0;
3737         bool cleaned = false;
3738         unsigned int total_rx_bytes=0, total_rx_packets=0;
3739
3740         i = rx_ring->next_to_clean;
3741         rx_desc = E1000_RX_DESC(*rx_ring, i);
3742         buffer_info = &rx_ring->buffer_info[i];
3743
3744         while (rx_desc->status & E1000_RXD_STAT_DD) {
3745                 struct sk_buff *skb;
3746                 u8 status;
3747
3748                 if (*work_done >= work_to_do)
3749                         break;
3750                 (*work_done)++;
3751
3752                 status = rx_desc->status;
3753                 skb = buffer_info->skb;
3754                 buffer_info->skb = NULL;
3755
3756                 prefetch(skb->data - NET_IP_ALIGN);
3757
3758                 if (++i == rx_ring->count) i = 0;
3759                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3760                 prefetch(next_rxd);
3761
3762                 next_buffer = &rx_ring->buffer_info[i];
3763
3764                 cleaned = true;
3765                 cleaned_count++;
3766                 pci_unmap_single(pdev, buffer_info->dma, buffer_info->length,
3767                                  PCI_DMA_FROMDEVICE);
3768                 buffer_info->dma = 0;
3769
3770                 length = le16_to_cpu(rx_desc->length);
3771                 /* !EOP means multiple descriptors were used to store a single
3772                  * packet, also make sure the frame isn't just CRC only */
3773                 if (unlikely(!(status & E1000_RXD_STAT_EOP) || (length <= 4))) {
3774                         /* All receives must fit into a single buffer */
3775                         E1000_DBG("%s: Receive packet consumed multiple"
3776                                   " buffers\n", netdev->name);
3777                         /* recycle */
3778                         buffer_info->skb = skb;
3779                         goto next_desc;
3780                 }
3781
3782                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
3783                         u8 last_byte = *(skb->data + length - 1);
3784                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3785                                        last_byte)) {
3786                                 spin_lock_irqsave(&adapter->stats_lock, flags);
3787                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3788                                                        length, skb->data);
3789                                 spin_unlock_irqrestore(&adapter->stats_lock,
3790                                                        flags);
3791                                 length--;
3792                         } else {
3793                                 /* recycle */
3794                                 buffer_info->skb = skb;
3795                                 goto next_desc;
3796                         }
3797                 }
3798
3799                 /* adjust length to remove Ethernet CRC, this must be
3800                  * done after the TBI_ACCEPT workaround above */
3801                 length -= 4;
3802
3803                 /* probably a little skewed due to removing CRC */
3804                 total_rx_bytes += length;
3805                 total_rx_packets++;
3806
3807                 /* code added for copybreak, this should improve
3808                  * performance for small packets with large amounts
3809                  * of reassembly being done in the stack */
3810                 if (length < copybreak) {
3811                         struct sk_buff *new_skb =
3812                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
3813                         if (new_skb) {
3814                                 skb_reserve(new_skb, NET_IP_ALIGN);
3815                                 skb_copy_to_linear_data_offset(new_skb,
3816                                                                -NET_IP_ALIGN,
3817                                                                (skb->data -
3818                                                                 NET_IP_ALIGN),
3819                                                                (length +
3820                                                                 NET_IP_ALIGN));
3821                                 /* save the skb in buffer_info as good */
3822                                 buffer_info->skb = skb;
3823                                 skb = new_skb;
3824                         }
3825                         /* else just continue with the old one */
3826                 }
3827                 /* end copybreak code */
3828                 skb_put(skb, length);
3829
3830                 /* Receive Checksum Offload */
3831                 e1000_rx_checksum(adapter,
3832                                   (u32)(status) |
3833                                   ((u32)(rx_desc->errors) << 24),
3834                                   le16_to_cpu(rx_desc->csum), skb);
3835
3836                 skb->protocol = eth_type_trans(skb, netdev);
3837
3838                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3839
3840 next_desc:
3841                 rx_desc->status = 0;
3842
3843                 /* return some buffers to hardware, one at a time is too slow */
3844                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3845                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3846                         cleaned_count = 0;
3847                 }
3848
3849                 /* use prefetched values */
3850                 rx_desc = next_rxd;
3851                 buffer_info = next_buffer;
3852         }
3853         rx_ring->next_to_clean = i;
3854
3855         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3856         if (cleaned_count)
3857                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3858
3859         adapter->total_rx_packets += total_rx_packets;
3860         adapter->total_rx_bytes += total_rx_bytes;
3861         adapter->net_stats.rx_bytes += total_rx_bytes;
3862         adapter->net_stats.rx_packets += total_rx_packets;
3863         return cleaned;
3864 }
3865
3866 /**
3867  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
3868  * @adapter: address of board private structure
3869  * @rx_ring: pointer to receive ring structure
3870  * @cleaned_count: number of buffers to allocate this pass
3871  **/
3872
3873 static void
3874 e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
3875                              struct e1000_rx_ring *rx_ring, int cleaned_count)
3876 {
3877         struct net_device *netdev = adapter->netdev;
3878         struct pci_dev *pdev = adapter->pdev;
3879         struct e1000_rx_desc *rx_desc;
3880         struct e1000_buffer *buffer_info;
3881         struct sk_buff *skb;
3882         unsigned int i;
3883         unsigned int bufsz = 256 -
3884                              16 /*for skb_reserve */ -
3885                              NET_IP_ALIGN;
3886
3887         i = rx_ring->next_to_use;
3888         buffer_info = &rx_ring->buffer_info[i];
3889
3890         while (cleaned_count--) {
3891                 skb = buffer_info->skb;
3892                 if (skb) {
3893                         skb_trim(skb, 0);
3894                         goto check_page;
3895                 }
3896
3897                 skb = netdev_alloc_skb(netdev, bufsz);
3898                 if (unlikely(!skb)) {
3899                         /* Better luck next round */
3900                         adapter->alloc_rx_buff_failed++;
3901                         break;
3902                 }
3903
3904                 /* Fix for errata 23, can't cross 64kB boundary */
3905                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3906                         struct sk_buff *oldskb = skb;
3907                         DPRINTK(PROBE, ERR, "skb align check failed: %u bytes "
3908                                              "at %p\n", bufsz, skb->data);
3909                         /* Try again, without freeing the previous */
3910                         skb = netdev_alloc_skb(netdev, bufsz);
3911                         /* Failed allocation, critical failure */
3912                         if (!skb) {
3913                                 dev_kfree_skb(oldskb);
3914                                 adapter->alloc_rx_buff_failed++;
3915                                 break;
3916                         }
3917
3918                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3919                                 /* give up */
3920                                 dev_kfree_skb(skb);
3921                                 dev_kfree_skb(oldskb);
3922                                 break; /* while (cleaned_count--) */
3923                         }
3924
3925                         /* Use new allocation */
3926                         dev_kfree_skb(oldskb);
3927                 }
3928                 /* Make buffer alignment 2 beyond a 16 byte boundary
3929                  * this will result in a 16 byte aligned IP header after
3930                  * the 14 byte MAC header is removed
3931                  */
3932                 skb_reserve(skb, NET_IP_ALIGN);
3933
3934                 buffer_info->skb = skb;
3935                 buffer_info->length = adapter->rx_buffer_len;
3936 check_page:
3937                 /* allocate a new page if necessary */
3938                 if (!buffer_info->page) {
3939                         buffer_info->page = alloc_page(GFP_ATOMIC);
3940                         if (unlikely(!buffer_info->page)) {
3941                                 adapter->alloc_rx_buff_failed++;
3942                                 break;
3943                         }
3944                 }
3945
3946                 if (!buffer_info->dma)
3947                         buffer_info->dma = pci_map_page(pdev,
3948                                                         buffer_info->page, 0,
3949                                                         buffer_info->length,
3950                                                         PCI_DMA_FROMDEVICE);
3951
3952                 rx_desc = E1000_RX_DESC(*rx_ring, i);
3953                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3954
3955                 if (unlikely(++i == rx_ring->count))
3956                         i = 0;
3957                 buffer_info = &rx_ring->buffer_info[i];
3958         }
3959
3960         if (likely(rx_ring->next_to_use != i)) {
3961                 rx_ring->next_to_use = i;
3962                 if (unlikely(i-- == 0))
3963                         i = (rx_ring->count - 1);
3964
3965                 /* Force memory writes to complete before letting h/w
3966                  * know there are new descriptors to fetch.  (Only
3967                  * applicable for weak-ordered memory model archs,
3968                  * such as IA-64). */
3969                 wmb();
3970                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
3971         }
3972 }
3973
3974 /**
3975  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
3976  * @adapter: address of board private structure
3977  **/
3978
3979 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
3980                                    struct e1000_rx_ring *rx_ring,
3981                                    int cleaned_count)
3982 {
3983         struct e1000_hw *hw = &adapter->hw;
3984         struct net_device *netdev = adapter->netdev;
3985         struct pci_dev *pdev = adapter->pdev;
3986         struct e1000_rx_desc *rx_desc;
3987         struct e1000_buffer *buffer_info;
3988         struct sk_buff *skb;
3989         unsigned int i;
3990         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
3991
3992         i = rx_ring->next_to_use;
3993         buffer_info = &rx_ring->buffer_info[i];
3994
3995         while (cleaned_count--) {
3996                 skb = buffer_info->skb;
3997                 if (skb) {
3998                         skb_trim(skb, 0);
3999                         goto map_skb;
4000                 }
4001
4002                 skb = netdev_alloc_skb(netdev, bufsz);
4003                 if (unlikely(!skb)) {
4004                         /* Better luck next round */
4005                         adapter->alloc_rx_buff_failed++;
4006                         break;
4007                 }
4008
4009                 /* Fix for errata 23, can't cross 64kB boundary */
4010                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4011                         struct sk_buff *oldskb = skb;
4012                         DPRINTK(RX_ERR, ERR, "skb align check failed: %u bytes "
4013                                              "at %p\n", bufsz, skb->data);
4014                         /* Try again, without freeing the previous */
4015                         skb = netdev_alloc_skb(netdev, bufsz);
4016                         /* Failed allocation, critical failure */
4017                         if (!skb) {
4018                                 dev_kfree_skb(oldskb);
4019                                 adapter->alloc_rx_buff_failed++;
4020                                 break;
4021                         }
4022
4023                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4024                                 /* give up */
4025                                 dev_kfree_skb(skb);
4026                                 dev_kfree_skb(oldskb);
4027                                 adapter->alloc_rx_buff_failed++;
4028                                 break; /* while !buffer_info->skb */
4029                         }
4030
4031                         /* Use new allocation */
4032                         dev_kfree_skb(oldskb);
4033                 }
4034                 /* Make buffer alignment 2 beyond a 16 byte boundary
4035                  * this will result in a 16 byte aligned IP header after
4036                  * the 14 byte MAC header is removed
4037                  */
4038                 skb_reserve(skb, NET_IP_ALIGN);
4039
4040                 buffer_info->skb = skb;
4041                 buffer_info->length = adapter->rx_buffer_len;
4042 map_skb:
4043                 buffer_info->dma = pci_map_single(pdev,
4044                                                   skb->data,
4045                                                   buffer_info->length,
4046                                                   PCI_DMA_FROMDEVICE);
4047
4048                 /*
4049                  * XXX if it was allocated cleanly it will never map to a
4050                  * boundary crossing
4051                  */
4052
4053                 /* Fix for errata 23, can't cross 64kB boundary */
4054                 if (!e1000_check_64k_bound(adapter,
4055                                         (void *)(unsigned long)buffer_info->dma,
4056                                         adapter->rx_buffer_len)) {
4057                         DPRINTK(RX_ERR, ERR,
4058                                 "dma align check failed: %u bytes at %p\n",
4059                                 adapter->rx_buffer_len,
4060                                 (void *)(unsigned long)buffer_info->dma);
4061                         dev_kfree_skb(skb);
4062                         buffer_info->skb = NULL;
4063
4064                         pci_unmap_single(pdev, buffer_info->dma,
4065                                          adapter->rx_buffer_len,
4066                                          PCI_DMA_FROMDEVICE);
4067                         buffer_info->dma = 0;
4068
4069                         adapter->alloc_rx_buff_failed++;
4070                         break; /* while !buffer_info->skb */
4071                 }
4072                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4073                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4074
4075                 if (unlikely(++i == rx_ring->count))
4076                         i = 0;
4077                 buffer_info = &rx_ring->buffer_info[i];
4078         }
4079
4080         if (likely(rx_ring->next_to_use != i)) {
4081                 rx_ring->next_to_use = i;
4082                 if (unlikely(i-- == 0))
4083                         i = (rx_ring->count - 1);
4084
4085                 /* Force memory writes to complete before letting h/w
4086                  * know there are new descriptors to fetch.  (Only
4087                  * applicable for weak-ordered memory model archs,
4088                  * such as IA-64). */
4089                 wmb();
4090                 writel(i, hw->hw_addr + rx_ring->rdt);
4091         }
4092 }
4093
4094 /**
4095  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4096  * @adapter:
4097  **/
4098
4099 static void e1000_smartspeed(struct e1000_adapter *adapter)
4100 {
4101         struct e1000_hw *hw = &adapter->hw;
4102         u16 phy_status;
4103         u16 phy_ctrl;
4104
4105         if ((hw->phy_type != e1000_phy_igp) || !hw->autoneg ||
4106            !(hw->autoneg_advertised & ADVERTISE_1000_FULL))
4107                 return;
4108
4109         if (adapter->smartspeed == 0) {
4110                 /* If Master/Slave config fault is asserted twice,
4111                  * we assume back-to-back */
4112                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4113                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4114                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4115                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4116                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4117                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4118                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4119                         e1000_write_phy_reg(hw, PHY_1000T_CTRL,
4120                                             phy_ctrl);
4121                         adapter->smartspeed++;
4122                         if (!e1000_phy_setup_autoneg(hw) &&
4123                            !e1000_read_phy_reg(hw, PHY_CTRL,
4124                                                &phy_ctrl)) {
4125                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4126                                              MII_CR_RESTART_AUTO_NEG);
4127                                 e1000_write_phy_reg(hw, PHY_CTRL,
4128                                                     phy_ctrl);
4129                         }
4130                 }
4131                 return;
4132         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4133                 /* If still no link, perhaps using 2/3 pair cable */
4134                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4135                 phy_ctrl |= CR_1000T_MS_ENABLE;
4136                 e1000_write_phy_reg(hw, PHY_1000T_CTRL, phy_ctrl);
4137                 if (!e1000_phy_setup_autoneg(hw) &&
4138                    !e1000_read_phy_reg(hw, PHY_CTRL, &phy_ctrl)) {
4139                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4140                                      MII_CR_RESTART_AUTO_NEG);
4141                         e1000_write_phy_reg(hw, PHY_CTRL, phy_ctrl);
4142                 }
4143         }
4144         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4145         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4146                 adapter->smartspeed = 0;
4147 }
4148
4149 /**
4150  * e1000_ioctl -
4151  * @netdev:
4152  * @ifreq:
4153  * @cmd:
4154  **/
4155
4156 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4157 {
4158         switch (cmd) {
4159         case SIOCGMIIPHY:
4160         case SIOCGMIIREG:
4161         case SIOCSMIIREG:
4162                 return e1000_mii_ioctl(netdev, ifr, cmd);
4163         default:
4164                 return -EOPNOTSUPP;
4165         }
4166 }
4167
4168 /**
4169  * e1000_mii_ioctl -
4170  * @netdev:
4171  * @ifreq:
4172  * @cmd:
4173  **/
4174
4175 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4176                            int cmd)
4177 {
4178         struct e1000_adapter *adapter = netdev_priv(netdev);
4179         struct e1000_hw *hw = &adapter->hw;
4180         struct mii_ioctl_data *data = if_mii(ifr);
4181         int retval;
4182         u16 mii_reg;
4183         u16 spddplx;
4184         unsigned long flags;
4185
4186         if (hw->media_type != e1000_media_type_copper)
4187                 return -EOPNOTSUPP;
4188
4189         switch (cmd) {
4190         case SIOCGMIIPHY:
4191                 data->phy_id = hw->phy_addr;
4192                 break;
4193         case SIOCGMIIREG:
4194                 spin_lock_irqsave(&adapter->stats_lock, flags);
4195                 if (e1000_read_phy_reg(hw, data->reg_num & 0x1F,
4196                                    &data->val_out)) {
4197                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4198                         return -EIO;
4199                 }
4200                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4201                 break;
4202         case SIOCSMIIREG:
4203                 if (data->reg_num & ~(0x1F))
4204                         return -EFAULT;
4205                 mii_reg = data->val_in;
4206                 spin_lock_irqsave(&adapter->stats_lock, flags);
4207                 if (e1000_write_phy_reg(hw, data->reg_num,
4208                                         mii_reg)) {
4209                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4210                         return -EIO;
4211                 }
4212                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4213                 if (hw->media_type == e1000_media_type_copper) {
4214                         switch (data->reg_num) {
4215                         case PHY_CTRL:
4216                                 if (mii_reg & MII_CR_POWER_DOWN)
4217                                         break;
4218                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4219                                         hw->autoneg = 1;
4220                                         hw->autoneg_advertised = 0x2F;
4221                                 } else {
4222                                         if (mii_reg & 0x40)
4223                                                 spddplx = SPEED_1000;
4224                                         else if (mii_reg & 0x2000)
4225                                                 spddplx = SPEED_100;
4226                                         else
4227                                                 spddplx = SPEED_10;
4228                                         spddplx += (mii_reg & 0x100)
4229                                                    ? DUPLEX_FULL :
4230                                                    DUPLEX_HALF;
4231                                         retval = e1000_set_spd_dplx(adapter,
4232                                                                     spddplx);
4233                                         if (retval)
4234                                                 return retval;
4235                                 }
4236                                 if (netif_running(adapter->netdev))
4237                                         e1000_reinit_locked(adapter);
4238                                 else
4239                                         e1000_reset(adapter);
4240                                 break;
4241                         case M88E1000_PHY_SPEC_CTRL:
4242                         case M88E1000_EXT_PHY_SPEC_CTRL:
4243                                 if (e1000_phy_reset(hw))
4244                                         return -EIO;
4245                                 break;
4246                         }
4247                 } else {
4248                         switch (data->reg_num) {
4249                         case PHY_CTRL:
4250                                 if (mii_reg & MII_CR_POWER_DOWN)
4251                                         break;
4252                                 if (netif_running(adapter->netdev))
4253                                         e1000_reinit_locked(adapter);
4254                                 else
4255                                         e1000_reset(adapter);
4256                                 break;
4257                         }
4258                 }
4259                 break;
4260         default:
4261                 return -EOPNOTSUPP;
4262         }
4263         return E1000_SUCCESS;
4264 }
4265
4266 void e1000_pci_set_mwi(struct e1000_hw *hw)
4267 {
4268         struct e1000_adapter *adapter = hw->back;
4269         int ret_val = pci_set_mwi(adapter->pdev);
4270
4271         if (ret_val)
4272                 DPRINTK(PROBE, ERR, "Error in setting MWI\n");
4273 }
4274
4275 void e1000_pci_clear_mwi(struct e1000_hw *hw)
4276 {
4277         struct e1000_adapter *adapter = hw->back;
4278
4279         pci_clear_mwi(adapter->pdev);
4280 }
4281
4282 int e1000_pcix_get_mmrbc(struct e1000_hw *hw)
4283 {
4284         struct e1000_adapter *adapter = hw->back;
4285         return pcix_get_mmrbc(adapter->pdev);
4286 }
4287
4288 void e1000_pcix_set_mmrbc(struct e1000_hw *hw, int mmrbc)
4289 {
4290         struct e1000_adapter *adapter = hw->back;
4291         pcix_set_mmrbc(adapter->pdev, mmrbc);
4292 }
4293
4294 s32 e1000_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
4295 {
4296     struct e1000_adapter *adapter = hw->back;
4297     u16 cap_offset;
4298
4299     cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
4300     if (!cap_offset)
4301         return -E1000_ERR_CONFIG;
4302
4303     pci_read_config_word(adapter->pdev, cap_offset + reg, value);
4304
4305     return E1000_SUCCESS;
4306 }
4307
4308 void e1000_io_write(struct e1000_hw *hw, unsigned long port, u32 value)
4309 {
4310         outl(value, port);
4311 }
4312
4313 static void e1000_vlan_rx_register(struct net_device *netdev,
4314                                    struct vlan_group *grp)
4315 {
4316         struct e1000_adapter *adapter = netdev_priv(netdev);
4317         struct e1000_hw *hw = &adapter->hw;
4318         u32 ctrl, rctl;
4319
4320         if (!test_bit(__E1000_DOWN, &adapter->flags))
4321                 e1000_irq_disable(adapter);
4322         adapter->vlgrp = grp;
4323
4324         if (grp) {
4325                 /* enable VLAN tag insert/strip */
4326                 ctrl = er32(CTRL);
4327                 ctrl |= E1000_CTRL_VME;
4328                 ew32(CTRL, ctrl);
4329
4330                 /* enable VLAN receive filtering */
4331                 rctl = er32(RCTL);
4332                 rctl &= ~E1000_RCTL_CFIEN;
4333                 if (!(netdev->flags & IFF_PROMISC))
4334                         rctl |= E1000_RCTL_VFE;
4335                 ew32(RCTL, rctl);
4336                 e1000_update_mng_vlan(adapter);
4337         } else {
4338                 /* disable VLAN tag insert/strip */
4339                 ctrl = er32(CTRL);
4340                 ctrl &= ~E1000_CTRL_VME;
4341                 ew32(CTRL, ctrl);
4342
4343                 /* disable VLAN receive filtering */
4344                 rctl = er32(RCTL);
4345                 rctl &= ~E1000_RCTL_VFE;
4346                 ew32(RCTL, rctl);
4347
4348                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
4349                         e1000_vlan_rx_kill_vid(netdev,
4350                                                adapter->mng_vlan_id);
4351                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4352                 }
4353         }
4354
4355         if (!test_bit(__E1000_DOWN, &adapter->flags))
4356                 e1000_irq_enable(adapter);
4357 }
4358
4359 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4360 {
4361         struct e1000_adapter *adapter = netdev_priv(netdev);
4362         struct e1000_hw *hw = &adapter->hw;
4363         u32 vfta, index;
4364
4365         if ((hw->mng_cookie.status &
4366              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4367             (vid == adapter->mng_vlan_id))
4368                 return;
4369         /* add VID to filter table */
4370         index = (vid >> 5) & 0x7F;
4371         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4372         vfta |= (1 << (vid & 0x1F));
4373         e1000_write_vfta(hw, index, vfta);
4374 }
4375
4376 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
4377 {
4378         struct e1000_adapter *adapter = netdev_priv(netdev);
4379         struct e1000_hw *hw = &adapter->hw;
4380         u32 vfta, index;
4381
4382         if (!test_bit(__E1000_DOWN, &adapter->flags))
4383                 e1000_irq_disable(adapter);
4384         vlan_group_set_device(adapter->vlgrp, vid, NULL);
4385         if (!test_bit(__E1000_DOWN, &adapter->flags))
4386                 e1000_irq_enable(adapter);
4387
4388         /* remove VID from filter table */
4389         index = (vid >> 5) & 0x7F;
4390         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4391         vfta &= ~(1 << (vid & 0x1F));
4392         e1000_write_vfta(hw, index, vfta);
4393 }
4394
4395 static void e1000_restore_vlan(struct e1000_adapter *adapter)
4396 {
4397         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
4398
4399         if (adapter->vlgrp) {
4400                 u16 vid;
4401                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
4402                         if (!vlan_group_get_device(adapter->vlgrp, vid))
4403                                 continue;
4404                         e1000_vlan_rx_add_vid(adapter->netdev, vid);
4405                 }
4406         }
4407 }
4408
4409 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx)
4410 {
4411         struct e1000_hw *hw = &adapter->hw;
4412
4413         hw->autoneg = 0;
4414
4415         /* Fiber NICs only allow 1000 gbps Full duplex */
4416         if ((hw->media_type == e1000_media_type_fiber) &&
4417                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
4418                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
4419                 return -EINVAL;
4420         }
4421
4422         switch (spddplx) {
4423         case SPEED_10 + DUPLEX_HALF:
4424                 hw->forced_speed_duplex = e1000_10_half;
4425                 break;
4426         case SPEED_10 + DUPLEX_FULL:
4427                 hw->forced_speed_duplex = e1000_10_full;
4428                 break;
4429         case SPEED_100 + DUPLEX_HALF:
4430                 hw->forced_speed_duplex = e1000_100_half;
4431                 break;
4432         case SPEED_100 + DUPLEX_FULL:
4433                 hw->forced_speed_duplex = e1000_100_full;
4434                 break;
4435         case SPEED_1000 + DUPLEX_FULL:
4436                 hw->autoneg = 1;
4437                 hw->autoneg_advertised = ADVERTISE_1000_FULL;
4438                 break;
4439         case SPEED_1000 + DUPLEX_HALF: /* not supported */
4440         default:
4441                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
4442                 return -EINVAL;
4443         }
4444         return 0;
4445 }
4446
4447 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4448 {
4449         struct net_device *netdev = pci_get_drvdata(pdev);
4450         struct e1000_adapter *adapter = netdev_priv(netdev);
4451         struct e1000_hw *hw = &adapter->hw;
4452         u32 ctrl, ctrl_ext, rctl, status;
4453         u32 wufc = adapter->wol;
4454 #ifdef CONFIG_PM
4455         int retval = 0;
4456 #endif
4457
4458         netif_device_detach(netdev);
4459
4460         if (netif_running(netdev)) {
4461                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
4462                 e1000_down(adapter);
4463         }
4464
4465 #ifdef CONFIG_PM
4466         retval = pci_save_state(pdev);
4467         if (retval)
4468                 return retval;
4469 #endif
4470
4471         status = er32(STATUS);
4472         if (status & E1000_STATUS_LU)
4473                 wufc &= ~E1000_WUFC_LNKC;
4474
4475         if (wufc) {
4476                 e1000_setup_rctl(adapter);
4477                 e1000_set_rx_mode(netdev);
4478
4479                 /* turn on all-multi mode if wake on multicast is enabled */
4480                 if (wufc & E1000_WUFC_MC) {
4481                         rctl = er32(RCTL);
4482                         rctl |= E1000_RCTL_MPE;
4483                         ew32(RCTL, rctl);
4484                 }
4485
4486                 if (hw->mac_type >= e1000_82540) {
4487                         ctrl = er32(CTRL);
4488                         /* advertise wake from D3Cold */
4489                         #define E1000_CTRL_ADVD3WUC 0x00100000
4490                         /* phy power management enable */
4491                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4492                         ctrl |= E1000_CTRL_ADVD3WUC |
4493                                 E1000_CTRL_EN_PHY_PWR_MGMT;
4494                         ew32(CTRL, ctrl);
4495                 }
4496
4497                 if (hw->media_type == e1000_media_type_fiber ||
4498                     hw->media_type == e1000_media_type_internal_serdes) {
4499                         /* keep the laser running in D3 */
4500                         ctrl_ext = er32(CTRL_EXT);
4501                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4502                         ew32(CTRL_EXT, ctrl_ext);
4503                 }
4504
4505                 ew32(WUC, E1000_WUC_PME_EN);
4506                 ew32(WUFC, wufc);
4507         } else {
4508                 ew32(WUC, 0);
4509                 ew32(WUFC, 0);
4510         }
4511
4512         e1000_release_manageability(adapter);
4513
4514         *enable_wake = !!wufc;
4515
4516         /* make sure adapter isn't asleep if manageability is enabled */
4517         if (adapter->en_mng_pt)
4518                 *enable_wake = true;
4519
4520         if (netif_running(netdev))
4521                 e1000_free_irq(adapter);
4522
4523         pci_disable_device(pdev);
4524
4525         return 0;
4526 }
4527
4528 #ifdef CONFIG_PM
4529 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4530 {
4531         int retval;
4532         bool wake;
4533
4534         retval = __e1000_shutdown(pdev, &wake);
4535         if (retval)
4536                 return retval;
4537
4538         if (wake) {
4539                 pci_prepare_to_sleep(pdev);
4540         } else {
4541                 pci_wake_from_d3(pdev, false);
4542                 pci_set_power_state(pdev, PCI_D3hot);
4543         }
4544
4545         return 0;
4546 }
4547
4548 static int e1000_resume(struct pci_dev *pdev)
4549 {
4550         struct net_device *netdev = pci_get_drvdata(pdev);
4551         struct e1000_adapter *adapter = netdev_priv(netdev);
4552         struct e1000_hw *hw = &adapter->hw;
4553         u32 err;
4554
4555         pci_set_power_state(pdev, PCI_D0);
4556         pci_restore_state(pdev);
4557
4558         if (adapter->need_ioport)
4559                 err = pci_enable_device(pdev);
4560         else
4561                 err = pci_enable_device_mem(pdev);
4562         if (err) {
4563                 printk(KERN_ERR "e1000: Cannot enable PCI device from suspend\n");
4564                 return err;
4565         }
4566         pci_set_master(pdev);
4567
4568         pci_enable_wake(pdev, PCI_D3hot, 0);
4569         pci_enable_wake(pdev, PCI_D3cold, 0);
4570
4571         if (netif_running(netdev)) {
4572                 err = e1000_request_irq(adapter);
4573                 if (err)
4574                         return err;
4575         }
4576
4577         e1000_power_up_phy(adapter);
4578         e1000_reset(adapter);
4579         ew32(WUS, ~0);
4580
4581         e1000_init_manageability(adapter);
4582
4583         if (netif_running(netdev))
4584                 e1000_up(adapter);
4585
4586         netif_device_attach(netdev);
4587
4588         return 0;
4589 }
4590 #endif
4591
4592 static void e1000_shutdown(struct pci_dev *pdev)
4593 {
4594         bool wake;
4595
4596         __e1000_shutdown(pdev, &wake);
4597
4598         if (system_state == SYSTEM_POWER_OFF) {
4599                 pci_wake_from_d3(pdev, wake);
4600                 pci_set_power_state(pdev, PCI_D3hot);
4601         }
4602 }
4603
4604 #ifdef CONFIG_NET_POLL_CONTROLLER
4605 /*
4606  * Polling 'interrupt' - used by things like netconsole to send skbs
4607  * without having to re-enable interrupts. It's not called while
4608  * the interrupt routine is executing.
4609  */
4610 static void e1000_netpoll(struct net_device *netdev)
4611 {
4612         struct e1000_adapter *adapter = netdev_priv(netdev);
4613
4614         disable_irq(adapter->pdev->irq);
4615         e1000_intr(adapter->pdev->irq, netdev);
4616         enable_irq(adapter->pdev->irq);
4617 }
4618 #endif
4619
4620 /**
4621  * e1000_io_error_detected - called when PCI error is detected
4622  * @pdev: Pointer to PCI device
4623  * @state: The current pci conneection state
4624  *
4625  * This function is called after a PCI bus error affecting
4626  * this device has been detected.
4627  */
4628 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4629                                                 pci_channel_state_t state)
4630 {
4631         struct net_device *netdev = pci_get_drvdata(pdev);
4632         struct e1000_adapter *adapter = netdev_priv(netdev);
4633
4634         netif_device_detach(netdev);
4635
4636         if (state == pci_channel_io_perm_failure)
4637                 return PCI_ERS_RESULT_DISCONNECT;
4638
4639         if (netif_running(netdev))
4640                 e1000_down(adapter);
4641         pci_disable_device(pdev);
4642
4643         /* Request a slot slot reset. */
4644         return PCI_ERS_RESULT_NEED_RESET;
4645 }
4646
4647 /**
4648  * e1000_io_slot_reset - called after the pci bus has been reset.
4649  * @pdev: Pointer to PCI device
4650  *
4651  * Restart the card from scratch, as if from a cold-boot. Implementation
4652  * resembles the first-half of the e1000_resume routine.
4653  */
4654 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4655 {
4656         struct net_device *netdev = pci_get_drvdata(pdev);
4657         struct e1000_adapter *adapter = netdev_priv(netdev);
4658         struct e1000_hw *hw = &adapter->hw;
4659         int err;
4660
4661         if (adapter->need_ioport)
4662                 err = pci_enable_device(pdev);
4663         else
4664                 err = pci_enable_device_mem(pdev);
4665         if (err) {
4666                 printk(KERN_ERR "e1000: Cannot re-enable PCI device after reset.\n");
4667                 return PCI_ERS_RESULT_DISCONNECT;
4668         }
4669         pci_set_master(pdev);
4670
4671         pci_enable_wake(pdev, PCI_D3hot, 0);
4672         pci_enable_wake(pdev, PCI_D3cold, 0);
4673
4674         e1000_reset(adapter);
4675         ew32(WUS, ~0);
4676
4677         return PCI_ERS_RESULT_RECOVERED;
4678 }
4679
4680 /**
4681  * e1000_io_resume - called when traffic can start flowing again.
4682  * @pdev: Pointer to PCI device
4683  *
4684  * This callback is called when the error recovery driver tells us that
4685  * its OK to resume normal operation. Implementation resembles the
4686  * second-half of the e1000_resume routine.
4687  */
4688 static void e1000_io_resume(struct pci_dev *pdev)
4689 {
4690         struct net_device *netdev = pci_get_drvdata(pdev);
4691         struct e1000_adapter *adapter = netdev_priv(netdev);
4692
4693         e1000_init_manageability(adapter);
4694
4695         if (netif_running(netdev)) {
4696                 if (e1000_up(adapter)) {
4697                         printk("e1000: can't bring device back up after reset\n");
4698                         return;
4699                 }
4700         }
4701
4702         netif_device_attach(netdev);
4703 }
4704
4705 /* e1000_main.c */