8be6faee43e6e519519d015dd6c1d63a864d7b05
[safe/jmp/linux-2.6] / drivers / net / e1000 / e1000_main.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2006 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include "e1000.h"
30 #include <net/ip6_checksum.h>
31
32 char e1000_driver_name[] = "e1000";
33 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
34 #define DRV_VERSION "7.3.21-k5-NAPI"
35 const char e1000_driver_version[] = DRV_VERSION;
36 static const char e1000_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
37
38 /* e1000_pci_tbl - PCI Device ID Table
39  *
40  * Last entry must be all 0s
41  *
42  * Macro expands to...
43  *   {PCI_DEVICE(PCI_VENDOR_ID_INTEL, device_id)}
44  */
45 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
46         INTEL_E1000_ETHERNET_DEVICE(0x1000),
47         INTEL_E1000_ETHERNET_DEVICE(0x1001),
48         INTEL_E1000_ETHERNET_DEVICE(0x1004),
49         INTEL_E1000_ETHERNET_DEVICE(0x1008),
50         INTEL_E1000_ETHERNET_DEVICE(0x1009),
51         INTEL_E1000_ETHERNET_DEVICE(0x100C),
52         INTEL_E1000_ETHERNET_DEVICE(0x100D),
53         INTEL_E1000_ETHERNET_DEVICE(0x100E),
54         INTEL_E1000_ETHERNET_DEVICE(0x100F),
55         INTEL_E1000_ETHERNET_DEVICE(0x1010),
56         INTEL_E1000_ETHERNET_DEVICE(0x1011),
57         INTEL_E1000_ETHERNET_DEVICE(0x1012),
58         INTEL_E1000_ETHERNET_DEVICE(0x1013),
59         INTEL_E1000_ETHERNET_DEVICE(0x1014),
60         INTEL_E1000_ETHERNET_DEVICE(0x1015),
61         INTEL_E1000_ETHERNET_DEVICE(0x1016),
62         INTEL_E1000_ETHERNET_DEVICE(0x1017),
63         INTEL_E1000_ETHERNET_DEVICE(0x1018),
64         INTEL_E1000_ETHERNET_DEVICE(0x1019),
65         INTEL_E1000_ETHERNET_DEVICE(0x101A),
66         INTEL_E1000_ETHERNET_DEVICE(0x101D),
67         INTEL_E1000_ETHERNET_DEVICE(0x101E),
68         INTEL_E1000_ETHERNET_DEVICE(0x1026),
69         INTEL_E1000_ETHERNET_DEVICE(0x1027),
70         INTEL_E1000_ETHERNET_DEVICE(0x1028),
71         INTEL_E1000_ETHERNET_DEVICE(0x1075),
72         INTEL_E1000_ETHERNET_DEVICE(0x1076),
73         INTEL_E1000_ETHERNET_DEVICE(0x1077),
74         INTEL_E1000_ETHERNET_DEVICE(0x1078),
75         INTEL_E1000_ETHERNET_DEVICE(0x1079),
76         INTEL_E1000_ETHERNET_DEVICE(0x107A),
77         INTEL_E1000_ETHERNET_DEVICE(0x107B),
78         INTEL_E1000_ETHERNET_DEVICE(0x107C),
79         INTEL_E1000_ETHERNET_DEVICE(0x108A),
80         INTEL_E1000_ETHERNET_DEVICE(0x1099),
81         INTEL_E1000_ETHERNET_DEVICE(0x10B5),
82         /* required last entry */
83         {0,}
84 };
85
86 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
87
88 int e1000_up(struct e1000_adapter *adapter);
89 void e1000_down(struct e1000_adapter *adapter);
90 void e1000_reinit_locked(struct e1000_adapter *adapter);
91 void e1000_reset(struct e1000_adapter *adapter);
92 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx);
93 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
94 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
95 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
96 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
97 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
98                              struct e1000_tx_ring *txdr);
99 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
100                              struct e1000_rx_ring *rxdr);
101 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
102                              struct e1000_tx_ring *tx_ring);
103 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
104                              struct e1000_rx_ring *rx_ring);
105 void e1000_update_stats(struct e1000_adapter *adapter);
106
107 static int e1000_init_module(void);
108 static void e1000_exit_module(void);
109 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
110 static void __devexit e1000_remove(struct pci_dev *pdev);
111 static int e1000_alloc_queues(struct e1000_adapter *adapter);
112 static int e1000_sw_init(struct e1000_adapter *adapter);
113 static int e1000_open(struct net_device *netdev);
114 static int e1000_close(struct net_device *netdev);
115 static void e1000_configure_tx(struct e1000_adapter *adapter);
116 static void e1000_configure_rx(struct e1000_adapter *adapter);
117 static void e1000_setup_rctl(struct e1000_adapter *adapter);
118 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
119 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
120 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
121                                 struct e1000_tx_ring *tx_ring);
122 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
123                                 struct e1000_rx_ring *rx_ring);
124 static void e1000_set_rx_mode(struct net_device *netdev);
125 static void e1000_update_phy_info(unsigned long data);
126 static void e1000_watchdog(unsigned long data);
127 static void e1000_82547_tx_fifo_stall(unsigned long data);
128 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
129                                     struct net_device *netdev);
130 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
131 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
132 static int e1000_set_mac(struct net_device *netdev, void *p);
133 static irqreturn_t e1000_intr(int irq, void *data);
134 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
135                                struct e1000_tx_ring *tx_ring);
136 static int e1000_clean(struct napi_struct *napi, int budget);
137 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
138                                struct e1000_rx_ring *rx_ring,
139                                int *work_done, int work_to_do);
140 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
141                                      struct e1000_rx_ring *rx_ring,
142                                      int *work_done, int work_to_do);
143 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
144                                    struct e1000_rx_ring *rx_ring,
145                                    int cleaned_count);
146 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
147                                          struct e1000_rx_ring *rx_ring,
148                                          int cleaned_count);
149 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
150 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
151                            int cmd);
152 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
153 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
154 static void e1000_tx_timeout(struct net_device *dev);
155 static void e1000_reset_task(struct work_struct *work);
156 static void e1000_smartspeed(struct e1000_adapter *adapter);
157 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
158                                        struct sk_buff *skb);
159
160 static void e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp);
161 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid);
162 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid);
163 static void e1000_restore_vlan(struct e1000_adapter *adapter);
164
165 #ifdef CONFIG_PM
166 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
167 static int e1000_resume(struct pci_dev *pdev);
168 #endif
169 static void e1000_shutdown(struct pci_dev *pdev);
170
171 #ifdef CONFIG_NET_POLL_CONTROLLER
172 /* for netdump / net console */
173 static void e1000_netpoll (struct net_device *netdev);
174 #endif
175
176 #define COPYBREAK_DEFAULT 256
177 static unsigned int copybreak __read_mostly = COPYBREAK_DEFAULT;
178 module_param(copybreak, uint, 0644);
179 MODULE_PARM_DESC(copybreak,
180         "Maximum size of packet that is copied to a new buffer on receive");
181
182 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
183                      pci_channel_state_t state);
184 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
185 static void e1000_io_resume(struct pci_dev *pdev);
186
187 static struct pci_error_handlers e1000_err_handler = {
188         .error_detected = e1000_io_error_detected,
189         .slot_reset = e1000_io_slot_reset,
190         .resume = e1000_io_resume,
191 };
192
193 static struct pci_driver e1000_driver = {
194         .name     = e1000_driver_name,
195         .id_table = e1000_pci_tbl,
196         .probe    = e1000_probe,
197         .remove   = __devexit_p(e1000_remove),
198 #ifdef CONFIG_PM
199         /* Power Managment Hooks */
200         .suspend  = e1000_suspend,
201         .resume   = e1000_resume,
202 #endif
203         .shutdown = e1000_shutdown,
204         .err_handler = &e1000_err_handler
205 };
206
207 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
208 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
209 MODULE_LICENSE("GPL");
210 MODULE_VERSION(DRV_VERSION);
211
212 static int debug = NETIF_MSG_DRV | NETIF_MSG_PROBE;
213 module_param(debug, int, 0);
214 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
215
216 /**
217  * e1000_init_module - Driver Registration Routine
218  *
219  * e1000_init_module is the first routine called when the driver is
220  * loaded. All it does is register with the PCI subsystem.
221  **/
222
223 static int __init e1000_init_module(void)
224 {
225         int ret;
226         printk(KERN_INFO "%s - version %s\n",
227                e1000_driver_string, e1000_driver_version);
228
229         printk(KERN_INFO "%s\n", e1000_copyright);
230
231         ret = pci_register_driver(&e1000_driver);
232         if (copybreak != COPYBREAK_DEFAULT) {
233                 if (copybreak == 0)
234                         printk(KERN_INFO "e1000: copybreak disabled\n");
235                 else
236                         printk(KERN_INFO "e1000: copybreak enabled for "
237                                "packets <= %u bytes\n", copybreak);
238         }
239         return ret;
240 }
241
242 module_init(e1000_init_module);
243
244 /**
245  * e1000_exit_module - Driver Exit Cleanup Routine
246  *
247  * e1000_exit_module is called just before the driver is removed
248  * from memory.
249  **/
250
251 static void __exit e1000_exit_module(void)
252 {
253         pci_unregister_driver(&e1000_driver);
254 }
255
256 module_exit(e1000_exit_module);
257
258 static int e1000_request_irq(struct e1000_adapter *adapter)
259 {
260         struct net_device *netdev = adapter->netdev;
261         irq_handler_t handler = e1000_intr;
262         int irq_flags = IRQF_SHARED;
263         int err;
264
265         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
266                           netdev);
267         if (err) {
268                 DPRINTK(PROBE, ERR,
269                         "Unable to allocate interrupt Error: %d\n", err);
270         }
271
272         return err;
273 }
274
275 static void e1000_free_irq(struct e1000_adapter *adapter)
276 {
277         struct net_device *netdev = adapter->netdev;
278
279         free_irq(adapter->pdev->irq, netdev);
280 }
281
282 /**
283  * e1000_irq_disable - Mask off interrupt generation on the NIC
284  * @adapter: board private structure
285  **/
286
287 static void e1000_irq_disable(struct e1000_adapter *adapter)
288 {
289         struct e1000_hw *hw = &adapter->hw;
290
291         ew32(IMC, ~0);
292         E1000_WRITE_FLUSH();
293         synchronize_irq(adapter->pdev->irq);
294 }
295
296 /**
297  * e1000_irq_enable - Enable default interrupt generation settings
298  * @adapter: board private structure
299  **/
300
301 static void e1000_irq_enable(struct e1000_adapter *adapter)
302 {
303         struct e1000_hw *hw = &adapter->hw;
304
305         ew32(IMS, IMS_ENABLE_MASK);
306         E1000_WRITE_FLUSH();
307 }
308
309 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
310 {
311         struct e1000_hw *hw = &adapter->hw;
312         struct net_device *netdev = adapter->netdev;
313         u16 vid = hw->mng_cookie.vlan_id;
314         u16 old_vid = adapter->mng_vlan_id;
315         if (adapter->vlgrp) {
316                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
317                         if (hw->mng_cookie.status &
318                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
319                                 e1000_vlan_rx_add_vid(netdev, vid);
320                                 adapter->mng_vlan_id = vid;
321                         } else
322                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
323
324                         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
325                                         (vid != old_vid) &&
326                             !vlan_group_get_device(adapter->vlgrp, old_vid))
327                                 e1000_vlan_rx_kill_vid(netdev, old_vid);
328                 } else
329                         adapter->mng_vlan_id = vid;
330         }
331 }
332
333 static void e1000_init_manageability(struct e1000_adapter *adapter)
334 {
335         struct e1000_hw *hw = &adapter->hw;
336
337         if (adapter->en_mng_pt) {
338                 u32 manc = er32(MANC);
339
340                 /* disable hardware interception of ARP */
341                 manc &= ~(E1000_MANC_ARP_EN);
342
343                 ew32(MANC, manc);
344         }
345 }
346
347 static void e1000_release_manageability(struct e1000_adapter *adapter)
348 {
349         struct e1000_hw *hw = &adapter->hw;
350
351         if (adapter->en_mng_pt) {
352                 u32 manc = er32(MANC);
353
354                 /* re-enable hardware interception of ARP */
355                 manc |= E1000_MANC_ARP_EN;
356
357                 ew32(MANC, manc);
358         }
359 }
360
361 /**
362  * e1000_configure - configure the hardware for RX and TX
363  * @adapter = private board structure
364  **/
365 static void e1000_configure(struct e1000_adapter *adapter)
366 {
367         struct net_device *netdev = adapter->netdev;
368         int i;
369
370         e1000_set_rx_mode(netdev);
371
372         e1000_restore_vlan(adapter);
373         e1000_init_manageability(adapter);
374
375         e1000_configure_tx(adapter);
376         e1000_setup_rctl(adapter);
377         e1000_configure_rx(adapter);
378         /* call E1000_DESC_UNUSED which always leaves
379          * at least 1 descriptor unused to make sure
380          * next_to_use != next_to_clean */
381         for (i = 0; i < adapter->num_rx_queues; i++) {
382                 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
383                 adapter->alloc_rx_buf(adapter, ring,
384                                       E1000_DESC_UNUSED(ring));
385         }
386
387         adapter->tx_queue_len = netdev->tx_queue_len;
388 }
389
390 int e1000_up(struct e1000_adapter *adapter)
391 {
392         struct e1000_hw *hw = &adapter->hw;
393
394         /* hardware has been reset, we need to reload some things */
395         e1000_configure(adapter);
396
397         clear_bit(__E1000_DOWN, &adapter->flags);
398
399         napi_enable(&adapter->napi);
400
401         e1000_irq_enable(adapter);
402
403         netif_wake_queue(adapter->netdev);
404
405         /* fire a link change interrupt to start the watchdog */
406         ew32(ICS, E1000_ICS_LSC);
407         return 0;
408 }
409
410 /**
411  * e1000_power_up_phy - restore link in case the phy was powered down
412  * @adapter: address of board private structure
413  *
414  * The phy may be powered down to save power and turn off link when the
415  * driver is unloaded and wake on lan is not enabled (among others)
416  * *** this routine MUST be followed by a call to e1000_reset ***
417  *
418  **/
419
420 void e1000_power_up_phy(struct e1000_adapter *adapter)
421 {
422         struct e1000_hw *hw = &adapter->hw;
423         u16 mii_reg = 0;
424
425         /* Just clear the power down bit to wake the phy back up */
426         if (hw->media_type == e1000_media_type_copper) {
427                 /* according to the manual, the phy will retain its
428                  * settings across a power-down/up cycle */
429                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
430                 mii_reg &= ~MII_CR_POWER_DOWN;
431                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
432         }
433 }
434
435 static void e1000_power_down_phy(struct e1000_adapter *adapter)
436 {
437         struct e1000_hw *hw = &adapter->hw;
438
439         /* Power down the PHY so no link is implied when interface is down *
440          * The PHY cannot be powered down if any of the following is true *
441          * (a) WoL is enabled
442          * (b) AMT is active
443          * (c) SoL/IDER session is active */
444         if (!adapter->wol && hw->mac_type >= e1000_82540 &&
445            hw->media_type == e1000_media_type_copper) {
446                 u16 mii_reg = 0;
447
448                 switch (hw->mac_type) {
449                 case e1000_82540:
450                 case e1000_82545:
451                 case e1000_82545_rev_3:
452                 case e1000_82546:
453                 case e1000_82546_rev_3:
454                 case e1000_82541:
455                 case e1000_82541_rev_2:
456                 case e1000_82547:
457                 case e1000_82547_rev_2:
458                         if (er32(MANC) & E1000_MANC_SMBUS_EN)
459                                 goto out;
460                         break;
461                 default:
462                         goto out;
463                 }
464                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
465                 mii_reg |= MII_CR_POWER_DOWN;
466                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
467                 mdelay(1);
468         }
469 out:
470         return;
471 }
472
473 void e1000_down(struct e1000_adapter *adapter)
474 {
475         struct e1000_hw *hw = &adapter->hw;
476         struct net_device *netdev = adapter->netdev;
477         u32 rctl, tctl;
478
479         /* signal that we're down so the interrupt handler does not
480          * reschedule our watchdog timer */
481         set_bit(__E1000_DOWN, &adapter->flags);
482
483         /* disable receives in the hardware */
484         rctl = er32(RCTL);
485         ew32(RCTL, rctl & ~E1000_RCTL_EN);
486         /* flush and sleep below */
487
488         netif_tx_disable(netdev);
489
490         /* disable transmits in the hardware */
491         tctl = er32(TCTL);
492         tctl &= ~E1000_TCTL_EN;
493         ew32(TCTL, tctl);
494         /* flush both disables and wait for them to finish */
495         E1000_WRITE_FLUSH();
496         msleep(10);
497
498         napi_disable(&adapter->napi);
499
500         e1000_irq_disable(adapter);
501
502         del_timer_sync(&adapter->tx_fifo_stall_timer);
503         del_timer_sync(&adapter->watchdog_timer);
504         del_timer_sync(&adapter->phy_info_timer);
505
506         netdev->tx_queue_len = adapter->tx_queue_len;
507         adapter->link_speed = 0;
508         adapter->link_duplex = 0;
509         netif_carrier_off(netdev);
510
511         e1000_reset(adapter);
512         e1000_clean_all_tx_rings(adapter);
513         e1000_clean_all_rx_rings(adapter);
514 }
515
516 void e1000_reinit_locked(struct e1000_adapter *adapter)
517 {
518         WARN_ON(in_interrupt());
519         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
520                 msleep(1);
521         e1000_down(adapter);
522         e1000_up(adapter);
523         clear_bit(__E1000_RESETTING, &adapter->flags);
524 }
525
526 void e1000_reset(struct e1000_adapter *adapter)
527 {
528         struct e1000_hw *hw = &adapter->hw;
529         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
530         bool legacy_pba_adjust = false;
531         u16 hwm;
532
533         /* Repartition Pba for greater than 9k mtu
534          * To take effect CTRL.RST is required.
535          */
536
537         switch (hw->mac_type) {
538         case e1000_82542_rev2_0:
539         case e1000_82542_rev2_1:
540         case e1000_82543:
541         case e1000_82544:
542         case e1000_82540:
543         case e1000_82541:
544         case e1000_82541_rev_2:
545                 legacy_pba_adjust = true;
546                 pba = E1000_PBA_48K;
547                 break;
548         case e1000_82545:
549         case e1000_82545_rev_3:
550         case e1000_82546:
551         case e1000_82546_rev_3:
552                 pba = E1000_PBA_48K;
553                 break;
554         case e1000_82547:
555         case e1000_82547_rev_2:
556                 legacy_pba_adjust = true;
557                 pba = E1000_PBA_30K;
558                 break;
559         case e1000_undefined:
560         case e1000_num_macs:
561                 break;
562         }
563
564         if (legacy_pba_adjust) {
565                 if (hw->max_frame_size > E1000_RXBUFFER_8192)
566                         pba -= 8; /* allocate more FIFO for Tx */
567
568                 if (hw->mac_type == e1000_82547) {
569                         adapter->tx_fifo_head = 0;
570                         adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
571                         adapter->tx_fifo_size =
572                                 (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
573                         atomic_set(&adapter->tx_fifo_stall, 0);
574                 }
575         } else if (hw->max_frame_size >  ETH_FRAME_LEN + ETH_FCS_LEN) {
576                 /* adjust PBA for jumbo frames */
577                 ew32(PBA, pba);
578
579                 /* To maintain wire speed transmits, the Tx FIFO should be
580                  * large enough to accommodate two full transmit packets,
581                  * rounded up to the next 1KB and expressed in KB.  Likewise,
582                  * the Rx FIFO should be large enough to accommodate at least
583                  * one full receive packet and is similarly rounded up and
584                  * expressed in KB. */
585                 pba = er32(PBA);
586                 /* upper 16 bits has Tx packet buffer allocation size in KB */
587                 tx_space = pba >> 16;
588                 /* lower 16 bits has Rx packet buffer allocation size in KB */
589                 pba &= 0xffff;
590                 /*
591                  * the tx fifo also stores 16 bytes of information about the tx
592                  * but don't include ethernet FCS because hardware appends it
593                  */
594                 min_tx_space = (hw->max_frame_size +
595                                 sizeof(struct e1000_tx_desc) -
596                                 ETH_FCS_LEN) * 2;
597                 min_tx_space = ALIGN(min_tx_space, 1024);
598                 min_tx_space >>= 10;
599                 /* software strips receive CRC, so leave room for it */
600                 min_rx_space = hw->max_frame_size;
601                 min_rx_space = ALIGN(min_rx_space, 1024);
602                 min_rx_space >>= 10;
603
604                 /* If current Tx allocation is less than the min Tx FIFO size,
605                  * and the min Tx FIFO size is less than the current Rx FIFO
606                  * allocation, take space away from current Rx allocation */
607                 if (tx_space < min_tx_space &&
608                     ((min_tx_space - tx_space) < pba)) {
609                         pba = pba - (min_tx_space - tx_space);
610
611                         /* PCI/PCIx hardware has PBA alignment constraints */
612                         switch (hw->mac_type) {
613                         case e1000_82545 ... e1000_82546_rev_3:
614                                 pba &= ~(E1000_PBA_8K - 1);
615                                 break;
616                         default:
617                                 break;
618                         }
619
620                         /* if short on rx space, rx wins and must trump tx
621                          * adjustment or use Early Receive if available */
622                         if (pba < min_rx_space)
623                                 pba = min_rx_space;
624                 }
625         }
626
627         ew32(PBA, pba);
628
629         /*
630          * flow control settings:
631          * The high water mark must be low enough to fit one full frame
632          * (or the size used for early receive) above it in the Rx FIFO.
633          * Set it to the lower of:
634          * - 90% of the Rx FIFO size, and
635          * - the full Rx FIFO size minus the early receive size (for parts
636          *   with ERT support assuming ERT set to E1000_ERT_2048), or
637          * - the full Rx FIFO size minus one full frame
638          */
639         hwm = min(((pba << 10) * 9 / 10),
640                   ((pba << 10) - hw->max_frame_size));
641
642         hw->fc_high_water = hwm & 0xFFF8;       /* 8-byte granularity */
643         hw->fc_low_water = hw->fc_high_water - 8;
644         hw->fc_pause_time = E1000_FC_PAUSE_TIME;
645         hw->fc_send_xon = 1;
646         hw->fc = hw->original_fc;
647
648         /* Allow time for pending master requests to run */
649         e1000_reset_hw(hw);
650         if (hw->mac_type >= e1000_82544)
651                 ew32(WUC, 0);
652
653         if (e1000_init_hw(hw))
654                 DPRINTK(PROBE, ERR, "Hardware Error\n");
655         e1000_update_mng_vlan(adapter);
656
657         /* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
658         if (hw->mac_type >= e1000_82544 &&
659             hw->autoneg == 1 &&
660             hw->autoneg_advertised == ADVERTISE_1000_FULL) {
661                 u32 ctrl = er32(CTRL);
662                 /* clear phy power management bit if we are in gig only mode,
663                  * which if enabled will attempt negotiation to 100Mb, which
664                  * can cause a loss of link at power off or driver unload */
665                 ctrl &= ~E1000_CTRL_SWDPIN3;
666                 ew32(CTRL, ctrl);
667         }
668
669         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
670         ew32(VET, ETHERNET_IEEE_VLAN_TYPE);
671
672         e1000_reset_adaptive(hw);
673         e1000_phy_get_info(hw, &adapter->phy_info);
674
675         e1000_release_manageability(adapter);
676 }
677
678 /**
679  *  Dump the eeprom for users having checksum issues
680  **/
681 static void e1000_dump_eeprom(struct e1000_adapter *adapter)
682 {
683         struct net_device *netdev = adapter->netdev;
684         struct ethtool_eeprom eeprom;
685         const struct ethtool_ops *ops = netdev->ethtool_ops;
686         u8 *data;
687         int i;
688         u16 csum_old, csum_new = 0;
689
690         eeprom.len = ops->get_eeprom_len(netdev);
691         eeprom.offset = 0;
692
693         data = kmalloc(eeprom.len, GFP_KERNEL);
694         if (!data) {
695                 printk(KERN_ERR "Unable to allocate memory to dump EEPROM"
696                        " data\n");
697                 return;
698         }
699
700         ops->get_eeprom(netdev, &eeprom, data);
701
702         csum_old = (data[EEPROM_CHECKSUM_REG * 2]) +
703                    (data[EEPROM_CHECKSUM_REG * 2 + 1] << 8);
704         for (i = 0; i < EEPROM_CHECKSUM_REG * 2; i += 2)
705                 csum_new += data[i] + (data[i + 1] << 8);
706         csum_new = EEPROM_SUM - csum_new;
707
708         printk(KERN_ERR "/*********************/\n");
709         printk(KERN_ERR "Current EEPROM Checksum : 0x%04x\n", csum_old);
710         printk(KERN_ERR "Calculated              : 0x%04x\n", csum_new);
711
712         printk(KERN_ERR "Offset    Values\n");
713         printk(KERN_ERR "========  ======\n");
714         print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 16, 1, data, 128, 0);
715
716         printk(KERN_ERR "Include this output when contacting your support "
717                "provider.\n");
718         printk(KERN_ERR "This is not a software error! Something bad "
719                "happened to your hardware or\n");
720         printk(KERN_ERR "EEPROM image. Ignoring this "
721                "problem could result in further problems,\n");
722         printk(KERN_ERR "possibly loss of data, corruption or system hangs!\n");
723         printk(KERN_ERR "The MAC Address will be reset to 00:00:00:00:00:00, "
724                "which is invalid\n");
725         printk(KERN_ERR "and requires you to set the proper MAC "
726                "address manually before continuing\n");
727         printk(KERN_ERR "to enable this network device.\n");
728         printk(KERN_ERR "Please inspect the EEPROM dump and report the issue "
729                "to your hardware vendor\n");
730         printk(KERN_ERR "or Intel Customer Support.\n");
731         printk(KERN_ERR "/*********************/\n");
732
733         kfree(data);
734 }
735
736 /**
737  * e1000_is_need_ioport - determine if an adapter needs ioport resources or not
738  * @pdev: PCI device information struct
739  *
740  * Return true if an adapter needs ioport resources
741  **/
742 static int e1000_is_need_ioport(struct pci_dev *pdev)
743 {
744         switch (pdev->device) {
745         case E1000_DEV_ID_82540EM:
746         case E1000_DEV_ID_82540EM_LOM:
747         case E1000_DEV_ID_82540EP:
748         case E1000_DEV_ID_82540EP_LOM:
749         case E1000_DEV_ID_82540EP_LP:
750         case E1000_DEV_ID_82541EI:
751         case E1000_DEV_ID_82541EI_MOBILE:
752         case E1000_DEV_ID_82541ER:
753         case E1000_DEV_ID_82541ER_LOM:
754         case E1000_DEV_ID_82541GI:
755         case E1000_DEV_ID_82541GI_LF:
756         case E1000_DEV_ID_82541GI_MOBILE:
757         case E1000_DEV_ID_82544EI_COPPER:
758         case E1000_DEV_ID_82544EI_FIBER:
759         case E1000_DEV_ID_82544GC_COPPER:
760         case E1000_DEV_ID_82544GC_LOM:
761         case E1000_DEV_ID_82545EM_COPPER:
762         case E1000_DEV_ID_82545EM_FIBER:
763         case E1000_DEV_ID_82546EB_COPPER:
764         case E1000_DEV_ID_82546EB_FIBER:
765         case E1000_DEV_ID_82546EB_QUAD_COPPER:
766                 return true;
767         default:
768                 return false;
769         }
770 }
771
772 static const struct net_device_ops e1000_netdev_ops = {
773         .ndo_open               = e1000_open,
774         .ndo_stop               = e1000_close,
775         .ndo_start_xmit         = e1000_xmit_frame,
776         .ndo_get_stats          = e1000_get_stats,
777         .ndo_set_rx_mode        = e1000_set_rx_mode,
778         .ndo_set_mac_address    = e1000_set_mac,
779         .ndo_tx_timeout         = e1000_tx_timeout,
780         .ndo_change_mtu         = e1000_change_mtu,
781         .ndo_do_ioctl           = e1000_ioctl,
782         .ndo_validate_addr      = eth_validate_addr,
783
784         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
785         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
786         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
787 #ifdef CONFIG_NET_POLL_CONTROLLER
788         .ndo_poll_controller    = e1000_netpoll,
789 #endif
790 };
791
792 /**
793  * e1000_probe - Device Initialization Routine
794  * @pdev: PCI device information struct
795  * @ent: entry in e1000_pci_tbl
796  *
797  * Returns 0 on success, negative on failure
798  *
799  * e1000_probe initializes an adapter identified by a pci_dev structure.
800  * The OS initialization, configuring of the adapter private structure,
801  * and a hardware reset occur.
802  **/
803 static int __devinit e1000_probe(struct pci_dev *pdev,
804                                  const struct pci_device_id *ent)
805 {
806         struct net_device *netdev;
807         struct e1000_adapter *adapter;
808         struct e1000_hw *hw;
809
810         static int cards_found = 0;
811         static int global_quad_port_a = 0; /* global ksp3 port a indication */
812         int i, err, pci_using_dac;
813         u16 eeprom_data = 0;
814         u16 eeprom_apme_mask = E1000_EEPROM_APME;
815         int bars, need_ioport;
816
817         /* do not allocate ioport bars when not needed */
818         need_ioport = e1000_is_need_ioport(pdev);
819         if (need_ioport) {
820                 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
821                 err = pci_enable_device(pdev);
822         } else {
823                 bars = pci_select_bars(pdev, IORESOURCE_MEM);
824                 err = pci_enable_device_mem(pdev);
825         }
826         if (err)
827                 return err;
828
829         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) &&
830             !pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64))) {
831                 pci_using_dac = 1;
832         } else {
833                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
834                 if (err) {
835                         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
836                         if (err) {
837                                 E1000_ERR("No usable DMA configuration, "
838                                           "aborting\n");
839                                 goto err_dma;
840                         }
841                 }
842                 pci_using_dac = 0;
843         }
844
845         err = pci_request_selected_regions(pdev, bars, e1000_driver_name);
846         if (err)
847                 goto err_pci_reg;
848
849         pci_set_master(pdev);
850         err = pci_save_state(pdev);
851         if (err)
852                 goto err_alloc_etherdev;
853
854         err = -ENOMEM;
855         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
856         if (!netdev)
857                 goto err_alloc_etherdev;
858
859         SET_NETDEV_DEV(netdev, &pdev->dev);
860
861         pci_set_drvdata(pdev, netdev);
862         adapter = netdev_priv(netdev);
863         adapter->netdev = netdev;
864         adapter->pdev = pdev;
865         adapter->msg_enable = (1 << debug) - 1;
866         adapter->bars = bars;
867         adapter->need_ioport = need_ioport;
868
869         hw = &adapter->hw;
870         hw->back = adapter;
871
872         err = -EIO;
873         hw->hw_addr = pci_ioremap_bar(pdev, BAR_0);
874         if (!hw->hw_addr)
875                 goto err_ioremap;
876
877         if (adapter->need_ioport) {
878                 for (i = BAR_1; i <= BAR_5; i++) {
879                         if (pci_resource_len(pdev, i) == 0)
880                                 continue;
881                         if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
882                                 hw->io_base = pci_resource_start(pdev, i);
883                                 break;
884                         }
885                 }
886         }
887
888         netdev->netdev_ops = &e1000_netdev_ops;
889         e1000_set_ethtool_ops(netdev);
890         netdev->watchdog_timeo = 5 * HZ;
891         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
892
893         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
894
895         adapter->bd_number = cards_found;
896
897         /* setup the private structure */
898
899         err = e1000_sw_init(adapter);
900         if (err)
901                 goto err_sw_init;
902
903         err = -EIO;
904
905         if (hw->mac_type >= e1000_82543) {
906                 netdev->features = NETIF_F_SG |
907                                    NETIF_F_HW_CSUM |
908                                    NETIF_F_HW_VLAN_TX |
909                                    NETIF_F_HW_VLAN_RX |
910                                    NETIF_F_HW_VLAN_FILTER;
911         }
912
913         if ((hw->mac_type >= e1000_82544) &&
914            (hw->mac_type != e1000_82547))
915                 netdev->features |= NETIF_F_TSO;
916
917         if (pci_using_dac)
918                 netdev->features |= NETIF_F_HIGHDMA;
919
920         netdev->vlan_features |= NETIF_F_TSO;
921         netdev->vlan_features |= NETIF_F_HW_CSUM;
922         netdev->vlan_features |= NETIF_F_SG;
923
924         adapter->en_mng_pt = e1000_enable_mng_pass_thru(hw);
925
926         /* initialize eeprom parameters */
927         if (e1000_init_eeprom_params(hw)) {
928                 E1000_ERR("EEPROM initialization failed\n");
929                 goto err_eeprom;
930         }
931
932         /* before reading the EEPROM, reset the controller to
933          * put the device in a known good starting state */
934
935         e1000_reset_hw(hw);
936
937         /* make sure the EEPROM is good */
938         if (e1000_validate_eeprom_checksum(hw) < 0) {
939                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
940                 e1000_dump_eeprom(adapter);
941                 /*
942                  * set MAC address to all zeroes to invalidate and temporary
943                  * disable this device for the user. This blocks regular
944                  * traffic while still permitting ethtool ioctls from reaching
945                  * the hardware as well as allowing the user to run the
946                  * interface after manually setting a hw addr using
947                  * `ip set address`
948                  */
949                 memset(hw->mac_addr, 0, netdev->addr_len);
950         } else {
951                 /* copy the MAC address out of the EEPROM */
952                 if (e1000_read_mac_addr(hw))
953                         DPRINTK(PROBE, ERR, "EEPROM Read Error\n");
954         }
955         /* don't block initalization here due to bad MAC address */
956         memcpy(netdev->dev_addr, hw->mac_addr, netdev->addr_len);
957         memcpy(netdev->perm_addr, hw->mac_addr, netdev->addr_len);
958
959         if (!is_valid_ether_addr(netdev->perm_addr))
960                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
961
962         e1000_get_bus_info(hw);
963
964         init_timer(&adapter->tx_fifo_stall_timer);
965         adapter->tx_fifo_stall_timer.function = &e1000_82547_tx_fifo_stall;
966         adapter->tx_fifo_stall_timer.data = (unsigned long)adapter;
967
968         init_timer(&adapter->watchdog_timer);
969         adapter->watchdog_timer.function = &e1000_watchdog;
970         adapter->watchdog_timer.data = (unsigned long) adapter;
971
972         init_timer(&adapter->phy_info_timer);
973         adapter->phy_info_timer.function = &e1000_update_phy_info;
974         adapter->phy_info_timer.data = (unsigned long)adapter;
975
976         INIT_WORK(&adapter->reset_task, e1000_reset_task);
977
978         e1000_check_options(adapter);
979
980         /* Initial Wake on LAN setting
981          * If APM wake is enabled in the EEPROM,
982          * enable the ACPI Magic Packet filter
983          */
984
985         switch (hw->mac_type) {
986         case e1000_82542_rev2_0:
987         case e1000_82542_rev2_1:
988         case e1000_82543:
989                 break;
990         case e1000_82544:
991                 e1000_read_eeprom(hw,
992                         EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
993                 eeprom_apme_mask = E1000_EEPROM_82544_APM;
994                 break;
995         case e1000_82546:
996         case e1000_82546_rev_3:
997                 if (er32(STATUS) & E1000_STATUS_FUNC_1){
998                         e1000_read_eeprom(hw,
999                                 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1000                         break;
1001                 }
1002                 /* Fall Through */
1003         default:
1004                 e1000_read_eeprom(hw,
1005                         EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1006                 break;
1007         }
1008         if (eeprom_data & eeprom_apme_mask)
1009                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1010
1011         /* now that we have the eeprom settings, apply the special cases
1012          * where the eeprom may be wrong or the board simply won't support
1013          * wake on lan on a particular port */
1014         switch (pdev->device) {
1015         case E1000_DEV_ID_82546GB_PCIE:
1016                 adapter->eeprom_wol = 0;
1017                 break;
1018         case E1000_DEV_ID_82546EB_FIBER:
1019         case E1000_DEV_ID_82546GB_FIBER:
1020                 /* Wake events only supported on port A for dual fiber
1021                  * regardless of eeprom setting */
1022                 if (er32(STATUS) & E1000_STATUS_FUNC_1)
1023                         adapter->eeprom_wol = 0;
1024                 break;
1025         case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
1026                 /* if quad port adapter, disable WoL on all but port A */
1027                 if (global_quad_port_a != 0)
1028                         adapter->eeprom_wol = 0;
1029                 else
1030                         adapter->quad_port_a = 1;
1031                 /* Reset for multiple quad port adapters */
1032                 if (++global_quad_port_a == 4)
1033                         global_quad_port_a = 0;
1034                 break;
1035         }
1036
1037         /* initialize the wol settings based on the eeprom settings */
1038         adapter->wol = adapter->eeprom_wol;
1039         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1040
1041         /* print bus type/speed/width info */
1042         DPRINTK(PROBE, INFO, "(PCI%s:%s:%s) ",
1043                 ((hw->bus_type == e1000_bus_type_pcix) ? "-X" : ""),
1044                 ((hw->bus_speed == e1000_bus_speed_133) ? "133MHz" :
1045                  (hw->bus_speed == e1000_bus_speed_120) ? "120MHz" :
1046                  (hw->bus_speed == e1000_bus_speed_100) ? "100MHz" :
1047                  (hw->bus_speed == e1000_bus_speed_66) ? "66MHz" : "33MHz"),
1048                 ((hw->bus_width == e1000_bus_width_64) ? "64-bit" : "32-bit"));
1049
1050         printk("%pM\n", netdev->dev_addr);
1051
1052         /* reset the hardware with the new settings */
1053         e1000_reset(adapter);
1054
1055         strcpy(netdev->name, "eth%d");
1056         err = register_netdev(netdev);
1057         if (err)
1058                 goto err_register;
1059
1060         /* carrier off reporting is important to ethtool even BEFORE open */
1061         netif_carrier_off(netdev);
1062
1063         DPRINTK(PROBE, INFO, "Intel(R) PRO/1000 Network Connection\n");
1064
1065         cards_found++;
1066         return 0;
1067
1068 err_register:
1069 err_eeprom:
1070         e1000_phy_hw_reset(hw);
1071
1072         if (hw->flash_address)
1073                 iounmap(hw->flash_address);
1074         kfree(adapter->tx_ring);
1075         kfree(adapter->rx_ring);
1076 err_sw_init:
1077         iounmap(hw->hw_addr);
1078 err_ioremap:
1079         free_netdev(netdev);
1080 err_alloc_etherdev:
1081         pci_release_selected_regions(pdev, bars);
1082 err_pci_reg:
1083 err_dma:
1084         pci_disable_device(pdev);
1085         return err;
1086 }
1087
1088 /**
1089  * e1000_remove - Device Removal Routine
1090  * @pdev: PCI device information struct
1091  *
1092  * e1000_remove is called by the PCI subsystem to alert the driver
1093  * that it should release a PCI device.  The could be caused by a
1094  * Hot-Plug event, or because the driver is going to be removed from
1095  * memory.
1096  **/
1097
1098 static void __devexit e1000_remove(struct pci_dev *pdev)
1099 {
1100         struct net_device *netdev = pci_get_drvdata(pdev);
1101         struct e1000_adapter *adapter = netdev_priv(netdev);
1102         struct e1000_hw *hw = &adapter->hw;
1103
1104         set_bit(__E1000_DOWN, &adapter->flags);
1105         del_timer_sync(&adapter->tx_fifo_stall_timer);
1106         del_timer_sync(&adapter->watchdog_timer);
1107         del_timer_sync(&adapter->phy_info_timer);
1108
1109         cancel_work_sync(&adapter->reset_task);
1110
1111         e1000_release_manageability(adapter);
1112
1113         unregister_netdev(netdev);
1114
1115         e1000_phy_hw_reset(hw);
1116
1117         kfree(adapter->tx_ring);
1118         kfree(adapter->rx_ring);
1119
1120         iounmap(hw->hw_addr);
1121         if (hw->flash_address)
1122                 iounmap(hw->flash_address);
1123         pci_release_selected_regions(pdev, adapter->bars);
1124
1125         free_netdev(netdev);
1126
1127         pci_disable_device(pdev);
1128 }
1129
1130 /**
1131  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1132  * @adapter: board private structure to initialize
1133  *
1134  * e1000_sw_init initializes the Adapter private data structure.
1135  * Fields are initialized based on PCI device information and
1136  * OS network device settings (MTU size).
1137  **/
1138
1139 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
1140 {
1141         struct e1000_hw *hw = &adapter->hw;
1142         struct net_device *netdev = adapter->netdev;
1143         struct pci_dev *pdev = adapter->pdev;
1144
1145         /* PCI config space info */
1146
1147         hw->vendor_id = pdev->vendor;
1148         hw->device_id = pdev->device;
1149         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1150         hw->subsystem_id = pdev->subsystem_device;
1151         hw->revision_id = pdev->revision;
1152
1153         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
1154
1155         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1156         hw->max_frame_size = netdev->mtu +
1157                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
1158         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
1159
1160         /* identify the MAC */
1161
1162         if (e1000_set_mac_type(hw)) {
1163                 DPRINTK(PROBE, ERR, "Unknown MAC Type\n");
1164                 return -EIO;
1165         }
1166
1167         switch (hw->mac_type) {
1168         default:
1169                 break;
1170         case e1000_82541:
1171         case e1000_82547:
1172         case e1000_82541_rev_2:
1173         case e1000_82547_rev_2:
1174                 hw->phy_init_script = 1;
1175                 break;
1176         }
1177
1178         e1000_set_media_type(hw);
1179
1180         hw->wait_autoneg_complete = false;
1181         hw->tbi_compatibility_en = true;
1182         hw->adaptive_ifs = true;
1183
1184         /* Copper options */
1185
1186         if (hw->media_type == e1000_media_type_copper) {
1187                 hw->mdix = AUTO_ALL_MODES;
1188                 hw->disable_polarity_correction = false;
1189                 hw->master_slave = E1000_MASTER_SLAVE;
1190         }
1191
1192         adapter->num_tx_queues = 1;
1193         adapter->num_rx_queues = 1;
1194
1195         if (e1000_alloc_queues(adapter)) {
1196                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
1197                 return -ENOMEM;
1198         }
1199
1200         /* Explicitly disable IRQ since the NIC can be in any state. */
1201         e1000_irq_disable(adapter);
1202
1203         spin_lock_init(&adapter->stats_lock);
1204
1205         set_bit(__E1000_DOWN, &adapter->flags);
1206
1207         return 0;
1208 }
1209
1210 /**
1211  * e1000_alloc_queues - Allocate memory for all rings
1212  * @adapter: board private structure to initialize
1213  *
1214  * We allocate one ring per queue at run-time since we don't know the
1215  * number of queues at compile-time.
1216  **/
1217
1218 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1219 {
1220         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1221                                    sizeof(struct e1000_tx_ring), GFP_KERNEL);
1222         if (!adapter->tx_ring)
1223                 return -ENOMEM;
1224
1225         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1226                                    sizeof(struct e1000_rx_ring), GFP_KERNEL);
1227         if (!adapter->rx_ring) {
1228                 kfree(adapter->tx_ring);
1229                 return -ENOMEM;
1230         }
1231
1232         return E1000_SUCCESS;
1233 }
1234
1235 /**
1236  * e1000_open - Called when a network interface is made active
1237  * @netdev: network interface device structure
1238  *
1239  * Returns 0 on success, negative value on failure
1240  *
1241  * The open entry point is called when a network interface is made
1242  * active by the system (IFF_UP).  At this point all resources needed
1243  * for transmit and receive operations are allocated, the interrupt
1244  * handler is registered with the OS, the watchdog timer is started,
1245  * and the stack is notified that the interface is ready.
1246  **/
1247
1248 static int e1000_open(struct net_device *netdev)
1249 {
1250         struct e1000_adapter *adapter = netdev_priv(netdev);
1251         struct e1000_hw *hw = &adapter->hw;
1252         int err;
1253
1254         /* disallow open during test */
1255         if (test_bit(__E1000_TESTING, &adapter->flags))
1256                 return -EBUSY;
1257
1258         netif_carrier_off(netdev);
1259
1260         /* allocate transmit descriptors */
1261         err = e1000_setup_all_tx_resources(adapter);
1262         if (err)
1263                 goto err_setup_tx;
1264
1265         /* allocate receive descriptors */
1266         err = e1000_setup_all_rx_resources(adapter);
1267         if (err)
1268                 goto err_setup_rx;
1269
1270         e1000_power_up_phy(adapter);
1271
1272         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1273         if ((hw->mng_cookie.status &
1274                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1275                 e1000_update_mng_vlan(adapter);
1276         }
1277
1278         /* before we allocate an interrupt, we must be ready to handle it.
1279          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1280          * as soon as we call pci_request_irq, so we have to setup our
1281          * clean_rx handler before we do so.  */
1282         e1000_configure(adapter);
1283
1284         err = e1000_request_irq(adapter);
1285         if (err)
1286                 goto err_req_irq;
1287
1288         /* From here on the code is the same as e1000_up() */
1289         clear_bit(__E1000_DOWN, &adapter->flags);
1290
1291         napi_enable(&adapter->napi);
1292
1293         e1000_irq_enable(adapter);
1294
1295         netif_start_queue(netdev);
1296
1297         /* fire a link status change interrupt to start the watchdog */
1298         ew32(ICS, E1000_ICS_LSC);
1299
1300         return E1000_SUCCESS;
1301
1302 err_req_irq:
1303         e1000_power_down_phy(adapter);
1304         e1000_free_all_rx_resources(adapter);
1305 err_setup_rx:
1306         e1000_free_all_tx_resources(adapter);
1307 err_setup_tx:
1308         e1000_reset(adapter);
1309
1310         return err;
1311 }
1312
1313 /**
1314  * e1000_close - Disables a network interface
1315  * @netdev: network interface device structure
1316  *
1317  * Returns 0, this is not allowed to fail
1318  *
1319  * The close entry point is called when an interface is de-activated
1320  * by the OS.  The hardware is still under the drivers control, but
1321  * needs to be disabled.  A global MAC reset is issued to stop the
1322  * hardware, and all transmit and receive resources are freed.
1323  **/
1324
1325 static int e1000_close(struct net_device *netdev)
1326 {
1327         struct e1000_adapter *adapter = netdev_priv(netdev);
1328         struct e1000_hw *hw = &adapter->hw;
1329
1330         WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1331         e1000_down(adapter);
1332         e1000_power_down_phy(adapter);
1333         e1000_free_irq(adapter);
1334
1335         e1000_free_all_tx_resources(adapter);
1336         e1000_free_all_rx_resources(adapter);
1337
1338         /* kill manageability vlan ID if supported, but not if a vlan with
1339          * the same ID is registered on the host OS (let 8021q kill it) */
1340         if ((hw->mng_cookie.status &
1341                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1342              !(adapter->vlgrp &&
1343                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id))) {
1344                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1345         }
1346
1347         return 0;
1348 }
1349
1350 /**
1351  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1352  * @adapter: address of board private structure
1353  * @start: address of beginning of memory
1354  * @len: length of memory
1355  **/
1356 static bool e1000_check_64k_bound(struct e1000_adapter *adapter, void *start,
1357                                   unsigned long len)
1358 {
1359         struct e1000_hw *hw = &adapter->hw;
1360         unsigned long begin = (unsigned long)start;
1361         unsigned long end = begin + len;
1362
1363         /* First rev 82545 and 82546 need to not allow any memory
1364          * write location to cross 64k boundary due to errata 23 */
1365         if (hw->mac_type == e1000_82545 ||
1366             hw->mac_type == e1000_82546) {
1367                 return ((begin ^ (end - 1)) >> 16) != 0 ? false : true;
1368         }
1369
1370         return true;
1371 }
1372
1373 /**
1374  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1375  * @adapter: board private structure
1376  * @txdr:    tx descriptor ring (for a specific queue) to setup
1377  *
1378  * Return 0 on success, negative on failure
1379  **/
1380
1381 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
1382                                     struct e1000_tx_ring *txdr)
1383 {
1384         struct pci_dev *pdev = adapter->pdev;
1385         int size;
1386
1387         size = sizeof(struct e1000_buffer) * txdr->count;
1388         txdr->buffer_info = vmalloc(size);
1389         if (!txdr->buffer_info) {
1390                 DPRINTK(PROBE, ERR,
1391                 "Unable to allocate memory for the transmit descriptor ring\n");
1392                 return -ENOMEM;
1393         }
1394         memset(txdr->buffer_info, 0, size);
1395
1396         /* round up to nearest 4K */
1397
1398         txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1399         txdr->size = ALIGN(txdr->size, 4096);
1400
1401         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1402         if (!txdr->desc) {
1403 setup_tx_desc_die:
1404                 vfree(txdr->buffer_info);
1405                 DPRINTK(PROBE, ERR,
1406                 "Unable to allocate memory for the transmit descriptor ring\n");
1407                 return -ENOMEM;
1408         }
1409
1410         /* Fix for errata 23, can't cross 64kB boundary */
1411         if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1412                 void *olddesc = txdr->desc;
1413                 dma_addr_t olddma = txdr->dma;
1414                 DPRINTK(TX_ERR, ERR, "txdr align check failed: %u bytes "
1415                                      "at %p\n", txdr->size, txdr->desc);
1416                 /* Try again, without freeing the previous */
1417                 txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1418                 /* Failed allocation, critical failure */
1419                 if (!txdr->desc) {
1420                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1421                         goto setup_tx_desc_die;
1422                 }
1423
1424                 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1425                         /* give up */
1426                         pci_free_consistent(pdev, txdr->size, txdr->desc,
1427                                             txdr->dma);
1428                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1429                         DPRINTK(PROBE, ERR,
1430                                 "Unable to allocate aligned memory "
1431                                 "for the transmit descriptor ring\n");
1432                         vfree(txdr->buffer_info);
1433                         return -ENOMEM;
1434                 } else {
1435                         /* Free old allocation, new allocation was successful */
1436                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1437                 }
1438         }
1439         memset(txdr->desc, 0, txdr->size);
1440
1441         txdr->next_to_use = 0;
1442         txdr->next_to_clean = 0;
1443
1444         return 0;
1445 }
1446
1447 /**
1448  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1449  *                                (Descriptors) for all queues
1450  * @adapter: board private structure
1451  *
1452  * Return 0 on success, negative on failure
1453  **/
1454
1455 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1456 {
1457         int i, err = 0;
1458
1459         for (i = 0; i < adapter->num_tx_queues; i++) {
1460                 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1461                 if (err) {
1462                         DPRINTK(PROBE, ERR,
1463                                 "Allocation for Tx Queue %u failed\n", i);
1464                         for (i-- ; i >= 0; i--)
1465                                 e1000_free_tx_resources(adapter,
1466                                                         &adapter->tx_ring[i]);
1467                         break;
1468                 }
1469         }
1470
1471         return err;
1472 }
1473
1474 /**
1475  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1476  * @adapter: board private structure
1477  *
1478  * Configure the Tx unit of the MAC after a reset.
1479  **/
1480
1481 static void e1000_configure_tx(struct e1000_adapter *adapter)
1482 {
1483         u64 tdba;
1484         struct e1000_hw *hw = &adapter->hw;
1485         u32 tdlen, tctl, tipg;
1486         u32 ipgr1, ipgr2;
1487
1488         /* Setup the HW Tx Head and Tail descriptor pointers */
1489
1490         switch (adapter->num_tx_queues) {
1491         case 1:
1492         default:
1493                 tdba = adapter->tx_ring[0].dma;
1494                 tdlen = adapter->tx_ring[0].count *
1495                         sizeof(struct e1000_tx_desc);
1496                 ew32(TDLEN, tdlen);
1497                 ew32(TDBAH, (tdba >> 32));
1498                 ew32(TDBAL, (tdba & 0x00000000ffffffffULL));
1499                 ew32(TDT, 0);
1500                 ew32(TDH, 0);
1501                 adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ? E1000_TDH : E1000_82542_TDH);
1502                 adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ? E1000_TDT : E1000_82542_TDT);
1503                 break;
1504         }
1505
1506         /* Set the default values for the Tx Inter Packet Gap timer */
1507         if ((hw->media_type == e1000_media_type_fiber ||
1508              hw->media_type == e1000_media_type_internal_serdes))
1509                 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1510         else
1511                 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1512
1513         switch (hw->mac_type) {
1514         case e1000_82542_rev2_0:
1515         case e1000_82542_rev2_1:
1516                 tipg = DEFAULT_82542_TIPG_IPGT;
1517                 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1518                 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1519                 break;
1520         default:
1521                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1522                 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1523                 break;
1524         }
1525         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1526         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1527         ew32(TIPG, tipg);
1528
1529         /* Set the Tx Interrupt Delay register */
1530
1531         ew32(TIDV, adapter->tx_int_delay);
1532         if (hw->mac_type >= e1000_82540)
1533                 ew32(TADV, adapter->tx_abs_int_delay);
1534
1535         /* Program the Transmit Control Register */
1536
1537         tctl = er32(TCTL);
1538         tctl &= ~E1000_TCTL_CT;
1539         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1540                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1541
1542         e1000_config_collision_dist(hw);
1543
1544         /* Setup Transmit Descriptor Settings for eop descriptor */
1545         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1546
1547         /* only set IDE if we are delaying interrupts using the timers */
1548         if (adapter->tx_int_delay)
1549                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1550
1551         if (hw->mac_type < e1000_82543)
1552                 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1553         else
1554                 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1555
1556         /* Cache if we're 82544 running in PCI-X because we'll
1557          * need this to apply a workaround later in the send path. */
1558         if (hw->mac_type == e1000_82544 &&
1559             hw->bus_type == e1000_bus_type_pcix)
1560                 adapter->pcix_82544 = 1;
1561
1562         ew32(TCTL, tctl);
1563
1564 }
1565
1566 /**
1567  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1568  * @adapter: board private structure
1569  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1570  *
1571  * Returns 0 on success, negative on failure
1572  **/
1573
1574 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
1575                                     struct e1000_rx_ring *rxdr)
1576 {
1577         struct pci_dev *pdev = adapter->pdev;
1578         int size, desc_len;
1579
1580         size = sizeof(struct e1000_buffer) * rxdr->count;
1581         rxdr->buffer_info = vmalloc(size);
1582         if (!rxdr->buffer_info) {
1583                 DPRINTK(PROBE, ERR,
1584                 "Unable to allocate memory for the receive descriptor ring\n");
1585                 return -ENOMEM;
1586         }
1587         memset(rxdr->buffer_info, 0, size);
1588
1589         desc_len = sizeof(struct e1000_rx_desc);
1590
1591         /* Round up to nearest 4K */
1592
1593         rxdr->size = rxdr->count * desc_len;
1594         rxdr->size = ALIGN(rxdr->size, 4096);
1595
1596         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1597
1598         if (!rxdr->desc) {
1599                 DPRINTK(PROBE, ERR,
1600                 "Unable to allocate memory for the receive descriptor ring\n");
1601 setup_rx_desc_die:
1602                 vfree(rxdr->buffer_info);
1603                 return -ENOMEM;
1604         }
1605
1606         /* Fix for errata 23, can't cross 64kB boundary */
1607         if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1608                 void *olddesc = rxdr->desc;
1609                 dma_addr_t olddma = rxdr->dma;
1610                 DPRINTK(RX_ERR, ERR, "rxdr align check failed: %u bytes "
1611                                      "at %p\n", rxdr->size, rxdr->desc);
1612                 /* Try again, without freeing the previous */
1613                 rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1614                 /* Failed allocation, critical failure */
1615                 if (!rxdr->desc) {
1616                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1617                         DPRINTK(PROBE, ERR,
1618                                 "Unable to allocate memory "
1619                                 "for the receive descriptor ring\n");
1620                         goto setup_rx_desc_die;
1621                 }
1622
1623                 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1624                         /* give up */
1625                         pci_free_consistent(pdev, rxdr->size, rxdr->desc,
1626                                             rxdr->dma);
1627                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1628                         DPRINTK(PROBE, ERR,
1629                                 "Unable to allocate aligned memory "
1630                                 "for the receive descriptor ring\n");
1631                         goto setup_rx_desc_die;
1632                 } else {
1633                         /* Free old allocation, new allocation was successful */
1634                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1635                 }
1636         }
1637         memset(rxdr->desc, 0, rxdr->size);
1638
1639         rxdr->next_to_clean = 0;
1640         rxdr->next_to_use = 0;
1641         rxdr->rx_skb_top = NULL;
1642
1643         return 0;
1644 }
1645
1646 /**
1647  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1648  *                                (Descriptors) for all queues
1649  * @adapter: board private structure
1650  *
1651  * Return 0 on success, negative on failure
1652  **/
1653
1654 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1655 {
1656         int i, err = 0;
1657
1658         for (i = 0; i < adapter->num_rx_queues; i++) {
1659                 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1660                 if (err) {
1661                         DPRINTK(PROBE, ERR,
1662                                 "Allocation for Rx Queue %u failed\n", i);
1663                         for (i-- ; i >= 0; i--)
1664                                 e1000_free_rx_resources(adapter,
1665                                                         &adapter->rx_ring[i]);
1666                         break;
1667                 }
1668         }
1669
1670         return err;
1671 }
1672
1673 /**
1674  * e1000_setup_rctl - configure the receive control registers
1675  * @adapter: Board private structure
1676  **/
1677 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1678 {
1679         struct e1000_hw *hw = &adapter->hw;
1680         u32 rctl;
1681
1682         rctl = er32(RCTL);
1683
1684         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1685
1686         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1687                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1688                 (hw->mc_filter_type << E1000_RCTL_MO_SHIFT);
1689
1690         if (hw->tbi_compatibility_on == 1)
1691                 rctl |= E1000_RCTL_SBP;
1692         else
1693                 rctl &= ~E1000_RCTL_SBP;
1694
1695         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1696                 rctl &= ~E1000_RCTL_LPE;
1697         else
1698                 rctl |= E1000_RCTL_LPE;
1699
1700         /* Setup buffer sizes */
1701         rctl &= ~E1000_RCTL_SZ_4096;
1702         rctl |= E1000_RCTL_BSEX;
1703         switch (adapter->rx_buffer_len) {
1704                 case E1000_RXBUFFER_2048:
1705                 default:
1706                         rctl |= E1000_RCTL_SZ_2048;
1707                         rctl &= ~E1000_RCTL_BSEX;
1708                         break;
1709                 case E1000_RXBUFFER_4096:
1710                         rctl |= E1000_RCTL_SZ_4096;
1711                         break;
1712                 case E1000_RXBUFFER_8192:
1713                         rctl |= E1000_RCTL_SZ_8192;
1714                         break;
1715                 case E1000_RXBUFFER_16384:
1716                         rctl |= E1000_RCTL_SZ_16384;
1717                         break;
1718         }
1719
1720         ew32(RCTL, rctl);
1721 }
1722
1723 /**
1724  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1725  * @adapter: board private structure
1726  *
1727  * Configure the Rx unit of the MAC after a reset.
1728  **/
1729
1730 static void e1000_configure_rx(struct e1000_adapter *adapter)
1731 {
1732         u64 rdba;
1733         struct e1000_hw *hw = &adapter->hw;
1734         u32 rdlen, rctl, rxcsum;
1735
1736         if (adapter->netdev->mtu > ETH_DATA_LEN) {
1737                 rdlen = adapter->rx_ring[0].count *
1738                         sizeof(struct e1000_rx_desc);
1739                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
1740                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
1741         } else {
1742                 rdlen = adapter->rx_ring[0].count *
1743                         sizeof(struct e1000_rx_desc);
1744                 adapter->clean_rx = e1000_clean_rx_irq;
1745                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1746         }
1747
1748         /* disable receives while setting up the descriptors */
1749         rctl = er32(RCTL);
1750         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1751
1752         /* set the Receive Delay Timer Register */
1753         ew32(RDTR, adapter->rx_int_delay);
1754
1755         if (hw->mac_type >= e1000_82540) {
1756                 ew32(RADV, adapter->rx_abs_int_delay);
1757                 if (adapter->itr_setting != 0)
1758                         ew32(ITR, 1000000000 / (adapter->itr * 256));
1759         }
1760
1761         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1762          * the Base and Length of the Rx Descriptor Ring */
1763         switch (adapter->num_rx_queues) {
1764         case 1:
1765         default:
1766                 rdba = adapter->rx_ring[0].dma;
1767                 ew32(RDLEN, rdlen);
1768                 ew32(RDBAH, (rdba >> 32));
1769                 ew32(RDBAL, (rdba & 0x00000000ffffffffULL));
1770                 ew32(RDT, 0);
1771                 ew32(RDH, 0);
1772                 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ? E1000_RDH : E1000_82542_RDH);
1773                 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ? E1000_RDT : E1000_82542_RDT);
1774                 break;
1775         }
1776
1777         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
1778         if (hw->mac_type >= e1000_82543) {
1779                 rxcsum = er32(RXCSUM);
1780                 if (adapter->rx_csum)
1781                         rxcsum |= E1000_RXCSUM_TUOFL;
1782                 else
1783                         /* don't need to clear IPPCSE as it defaults to 0 */
1784                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1785                 ew32(RXCSUM, rxcsum);
1786         }
1787
1788         /* Enable Receives */
1789         ew32(RCTL, rctl);
1790 }
1791
1792 /**
1793  * e1000_free_tx_resources - Free Tx Resources per Queue
1794  * @adapter: board private structure
1795  * @tx_ring: Tx descriptor ring for a specific queue
1796  *
1797  * Free all transmit software resources
1798  **/
1799
1800 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
1801                                     struct e1000_tx_ring *tx_ring)
1802 {
1803         struct pci_dev *pdev = adapter->pdev;
1804
1805         e1000_clean_tx_ring(adapter, tx_ring);
1806
1807         vfree(tx_ring->buffer_info);
1808         tx_ring->buffer_info = NULL;
1809
1810         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
1811
1812         tx_ring->desc = NULL;
1813 }
1814
1815 /**
1816  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1817  * @adapter: board private structure
1818  *
1819  * Free all transmit software resources
1820  **/
1821
1822 void e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1823 {
1824         int i;
1825
1826         for (i = 0; i < adapter->num_tx_queues; i++)
1827                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1828 }
1829
1830 static void e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1831                                              struct e1000_buffer *buffer_info)
1832 {
1833         if (buffer_info->dma) {
1834                 if (buffer_info->mapped_as_page)
1835                         pci_unmap_page(adapter->pdev, buffer_info->dma,
1836                                        buffer_info->length, PCI_DMA_TODEVICE);
1837                 else
1838                         pci_unmap_single(adapter->pdev, buffer_info->dma,
1839                                          buffer_info->length,
1840                                          PCI_DMA_TODEVICE);
1841                 buffer_info->dma = 0;
1842         }
1843         if (buffer_info->skb) {
1844                 dev_kfree_skb_any(buffer_info->skb);
1845                 buffer_info->skb = NULL;
1846         }
1847         buffer_info->time_stamp = 0;
1848         /* buffer_info must be completely set up in the transmit path */
1849 }
1850
1851 /**
1852  * e1000_clean_tx_ring - Free Tx Buffers
1853  * @adapter: board private structure
1854  * @tx_ring: ring to be cleaned
1855  **/
1856
1857 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
1858                                 struct e1000_tx_ring *tx_ring)
1859 {
1860         struct e1000_hw *hw = &adapter->hw;
1861         struct e1000_buffer *buffer_info;
1862         unsigned long size;
1863         unsigned int i;
1864
1865         /* Free all the Tx ring sk_buffs */
1866
1867         for (i = 0; i < tx_ring->count; i++) {
1868                 buffer_info = &tx_ring->buffer_info[i];
1869                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
1870         }
1871
1872         size = sizeof(struct e1000_buffer) * tx_ring->count;
1873         memset(tx_ring->buffer_info, 0, size);
1874
1875         /* Zero out the descriptor ring */
1876
1877         memset(tx_ring->desc, 0, tx_ring->size);
1878
1879         tx_ring->next_to_use = 0;
1880         tx_ring->next_to_clean = 0;
1881         tx_ring->last_tx_tso = 0;
1882
1883         writel(0, hw->hw_addr + tx_ring->tdh);
1884         writel(0, hw->hw_addr + tx_ring->tdt);
1885 }
1886
1887 /**
1888  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
1889  * @adapter: board private structure
1890  **/
1891
1892 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
1893 {
1894         int i;
1895
1896         for (i = 0; i < adapter->num_tx_queues; i++)
1897                 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1898 }
1899
1900 /**
1901  * e1000_free_rx_resources - Free Rx Resources
1902  * @adapter: board private structure
1903  * @rx_ring: ring to clean the resources from
1904  *
1905  * Free all receive software resources
1906  **/
1907
1908 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
1909                                     struct e1000_rx_ring *rx_ring)
1910 {
1911         struct pci_dev *pdev = adapter->pdev;
1912
1913         e1000_clean_rx_ring(adapter, rx_ring);
1914
1915         vfree(rx_ring->buffer_info);
1916         rx_ring->buffer_info = NULL;
1917
1918         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
1919
1920         rx_ring->desc = NULL;
1921 }
1922
1923 /**
1924  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
1925  * @adapter: board private structure
1926  *
1927  * Free all receive software resources
1928  **/
1929
1930 void e1000_free_all_rx_resources(struct e1000_adapter *adapter)
1931 {
1932         int i;
1933
1934         for (i = 0; i < adapter->num_rx_queues; i++)
1935                 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
1936 }
1937
1938 /**
1939  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1940  * @adapter: board private structure
1941  * @rx_ring: ring to free buffers from
1942  **/
1943
1944 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
1945                                 struct e1000_rx_ring *rx_ring)
1946 {
1947         struct e1000_hw *hw = &adapter->hw;
1948         struct e1000_buffer *buffer_info;
1949         struct pci_dev *pdev = adapter->pdev;
1950         unsigned long size;
1951         unsigned int i;
1952
1953         /* Free all the Rx ring sk_buffs */
1954         for (i = 0; i < rx_ring->count; i++) {
1955                 buffer_info = &rx_ring->buffer_info[i];
1956                 if (buffer_info->dma &&
1957                     adapter->clean_rx == e1000_clean_rx_irq) {
1958                         pci_unmap_single(pdev, buffer_info->dma,
1959                                          buffer_info->length,
1960                                          PCI_DMA_FROMDEVICE);
1961                 } else if (buffer_info->dma &&
1962                            adapter->clean_rx == e1000_clean_jumbo_rx_irq) {
1963                         pci_unmap_page(pdev, buffer_info->dma,
1964                                        buffer_info->length,
1965                                        PCI_DMA_FROMDEVICE);
1966                 }
1967
1968                 buffer_info->dma = 0;
1969                 if (buffer_info->page) {
1970                         put_page(buffer_info->page);
1971                         buffer_info->page = NULL;
1972                 }
1973                 if (buffer_info->skb) {
1974                         dev_kfree_skb(buffer_info->skb);
1975                         buffer_info->skb = NULL;
1976                 }
1977         }
1978
1979         /* there also may be some cached data from a chained receive */
1980         if (rx_ring->rx_skb_top) {
1981                 dev_kfree_skb(rx_ring->rx_skb_top);
1982                 rx_ring->rx_skb_top = NULL;
1983         }
1984
1985         size = sizeof(struct e1000_buffer) * rx_ring->count;
1986         memset(rx_ring->buffer_info, 0, size);
1987
1988         /* Zero out the descriptor ring */
1989         memset(rx_ring->desc, 0, rx_ring->size);
1990
1991         rx_ring->next_to_clean = 0;
1992         rx_ring->next_to_use = 0;
1993
1994         writel(0, hw->hw_addr + rx_ring->rdh);
1995         writel(0, hw->hw_addr + rx_ring->rdt);
1996 }
1997
1998 /**
1999  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2000  * @adapter: board private structure
2001  **/
2002
2003 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2004 {
2005         int i;
2006
2007         for (i = 0; i < adapter->num_rx_queues; i++)
2008                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2009 }
2010
2011 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2012  * and memory write and invalidate disabled for certain operations
2013  */
2014 static void e1000_enter_82542_rst(struct e1000_adapter *adapter)
2015 {
2016         struct e1000_hw *hw = &adapter->hw;
2017         struct net_device *netdev = adapter->netdev;
2018         u32 rctl;
2019
2020         e1000_pci_clear_mwi(hw);
2021
2022         rctl = er32(RCTL);
2023         rctl |= E1000_RCTL_RST;
2024         ew32(RCTL, rctl);
2025         E1000_WRITE_FLUSH();
2026         mdelay(5);
2027
2028         if (netif_running(netdev))
2029                 e1000_clean_all_rx_rings(adapter);
2030 }
2031
2032 static void e1000_leave_82542_rst(struct e1000_adapter *adapter)
2033 {
2034         struct e1000_hw *hw = &adapter->hw;
2035         struct net_device *netdev = adapter->netdev;
2036         u32 rctl;
2037
2038         rctl = er32(RCTL);
2039         rctl &= ~E1000_RCTL_RST;
2040         ew32(RCTL, rctl);
2041         E1000_WRITE_FLUSH();
2042         mdelay(5);
2043
2044         if (hw->pci_cmd_word & PCI_COMMAND_INVALIDATE)
2045                 e1000_pci_set_mwi(hw);
2046
2047         if (netif_running(netdev)) {
2048                 /* No need to loop, because 82542 supports only 1 queue */
2049                 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2050                 e1000_configure_rx(adapter);
2051                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2052         }
2053 }
2054
2055 /**
2056  * e1000_set_mac - Change the Ethernet Address of the NIC
2057  * @netdev: network interface device structure
2058  * @p: pointer to an address structure
2059  *
2060  * Returns 0 on success, negative on failure
2061  **/
2062
2063 static int e1000_set_mac(struct net_device *netdev, void *p)
2064 {
2065         struct e1000_adapter *adapter = netdev_priv(netdev);
2066         struct e1000_hw *hw = &adapter->hw;
2067         struct sockaddr *addr = p;
2068
2069         if (!is_valid_ether_addr(addr->sa_data))
2070                 return -EADDRNOTAVAIL;
2071
2072         /* 82542 2.0 needs to be in reset to write receive address registers */
2073
2074         if (hw->mac_type == e1000_82542_rev2_0)
2075                 e1000_enter_82542_rst(adapter);
2076
2077         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2078         memcpy(hw->mac_addr, addr->sa_data, netdev->addr_len);
2079
2080         e1000_rar_set(hw, hw->mac_addr, 0);
2081
2082         if (hw->mac_type == e1000_82542_rev2_0)
2083                 e1000_leave_82542_rst(adapter);
2084
2085         return 0;
2086 }
2087
2088 /**
2089  * e1000_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2090  * @netdev: network interface device structure
2091  *
2092  * The set_rx_mode entry point is called whenever the unicast or multicast
2093  * address lists or the network interface flags are updated. This routine is
2094  * responsible for configuring the hardware for proper unicast, multicast,
2095  * promiscuous mode, and all-multi behavior.
2096  **/
2097
2098 static void e1000_set_rx_mode(struct net_device *netdev)
2099 {
2100         struct e1000_adapter *adapter = netdev_priv(netdev);
2101         struct e1000_hw *hw = &adapter->hw;
2102         struct netdev_hw_addr *ha;
2103         bool use_uc = false;
2104         struct dev_addr_list *mc_ptr;
2105         u32 rctl;
2106         u32 hash_value;
2107         int i, rar_entries = E1000_RAR_ENTRIES;
2108         int mta_reg_count = E1000_NUM_MTA_REGISTERS;
2109         u32 *mcarray = kcalloc(mta_reg_count, sizeof(u32), GFP_ATOMIC);
2110
2111         if (!mcarray) {
2112                 DPRINTK(PROBE, ERR, "memory allocation failed\n");
2113                 return;
2114         }
2115
2116         /* Check for Promiscuous and All Multicast modes */
2117
2118         rctl = er32(RCTL);
2119
2120         if (netdev->flags & IFF_PROMISC) {
2121                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2122                 rctl &= ~E1000_RCTL_VFE;
2123         } else {
2124                 if (netdev->flags & IFF_ALLMULTI)
2125                         rctl |= E1000_RCTL_MPE;
2126                 else
2127                         rctl &= ~E1000_RCTL_MPE;
2128                 /* Enable VLAN filter if there is a VLAN */
2129                 if (adapter->vlgrp)
2130                         rctl |= E1000_RCTL_VFE;
2131         }
2132
2133         if (netdev_uc_count(netdev) > rar_entries - 1) {
2134                 rctl |= E1000_RCTL_UPE;
2135         } else if (!(netdev->flags & IFF_PROMISC)) {
2136                 rctl &= ~E1000_RCTL_UPE;
2137                 use_uc = true;
2138         }
2139
2140         ew32(RCTL, rctl);
2141
2142         /* 82542 2.0 needs to be in reset to write receive address registers */
2143
2144         if (hw->mac_type == e1000_82542_rev2_0)
2145                 e1000_enter_82542_rst(adapter);
2146
2147         /* load the first 14 addresses into the exact filters 1-14. Unicast
2148          * addresses take precedence to avoid disabling unicast filtering
2149          * when possible.
2150          *
2151          * RAR 0 is used for the station MAC adddress
2152          * if there are not 14 addresses, go ahead and clear the filters
2153          */
2154         i = 1;
2155         if (use_uc)
2156                 netdev_for_each_uc_addr(ha, netdev) {
2157                         if (i == rar_entries)
2158                                 break;
2159                         e1000_rar_set(hw, ha->addr, i++);
2160                 }
2161
2162         WARN_ON(i == rar_entries);
2163
2164         netdev_for_each_mc_addr(mc_ptr, netdev) {
2165                 if (i == rar_entries) {
2166                         /* load any remaining addresses into the hash table */
2167                         u32 hash_reg, hash_bit, mta;
2168                         hash_value = e1000_hash_mc_addr(hw, mc_ptr->da_addr);
2169                         hash_reg = (hash_value >> 5) & 0x7F;
2170                         hash_bit = hash_value & 0x1F;
2171                         mta = (1 << hash_bit);
2172                         mcarray[hash_reg] |= mta;
2173                 } else {
2174                         e1000_rar_set(hw, mc_ptr->da_addr, i++);
2175                 }
2176         }
2177
2178         for (; i < rar_entries; i++) {
2179                 E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2180                 E1000_WRITE_FLUSH();
2181                 E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2182                 E1000_WRITE_FLUSH();
2183         }
2184
2185         /* write the hash table completely, write from bottom to avoid
2186          * both stupid write combining chipsets, and flushing each write */
2187         for (i = mta_reg_count - 1; i >= 0 ; i--) {
2188                 /*
2189                  * If we are on an 82544 has an errata where writing odd
2190                  * offsets overwrites the previous even offset, but writing
2191                  * backwards over the range solves the issue by always
2192                  * writing the odd offset first
2193                  */
2194                 E1000_WRITE_REG_ARRAY(hw, MTA, i, mcarray[i]);
2195         }
2196         E1000_WRITE_FLUSH();
2197
2198         if (hw->mac_type == e1000_82542_rev2_0)
2199                 e1000_leave_82542_rst(adapter);
2200
2201         kfree(mcarray);
2202 }
2203
2204 /* Need to wait a few seconds after link up to get diagnostic information from
2205  * the phy */
2206
2207 static void e1000_update_phy_info(unsigned long data)
2208 {
2209         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2210         struct e1000_hw *hw = &adapter->hw;
2211         e1000_phy_get_info(hw, &adapter->phy_info);
2212 }
2213
2214 /**
2215  * e1000_82547_tx_fifo_stall - Timer Call-back
2216  * @data: pointer to adapter cast into an unsigned long
2217  **/
2218
2219 static void e1000_82547_tx_fifo_stall(unsigned long data)
2220 {
2221         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2222         struct e1000_hw *hw = &adapter->hw;
2223         struct net_device *netdev = adapter->netdev;
2224         u32 tctl;
2225
2226         if (atomic_read(&adapter->tx_fifo_stall)) {
2227                 if ((er32(TDT) == er32(TDH)) &&
2228                    (er32(TDFT) == er32(TDFH)) &&
2229                    (er32(TDFTS) == er32(TDFHS))) {
2230                         tctl = er32(TCTL);
2231                         ew32(TCTL, tctl & ~E1000_TCTL_EN);
2232                         ew32(TDFT, adapter->tx_head_addr);
2233                         ew32(TDFH, adapter->tx_head_addr);
2234                         ew32(TDFTS, adapter->tx_head_addr);
2235                         ew32(TDFHS, adapter->tx_head_addr);
2236                         ew32(TCTL, tctl);
2237                         E1000_WRITE_FLUSH();
2238
2239                         adapter->tx_fifo_head = 0;
2240                         atomic_set(&adapter->tx_fifo_stall, 0);
2241                         netif_wake_queue(netdev);
2242                 } else if (!test_bit(__E1000_DOWN, &adapter->flags)) {
2243                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2244                 }
2245         }
2246 }
2247
2248 bool e1000_has_link(struct e1000_adapter *adapter)
2249 {
2250         struct e1000_hw *hw = &adapter->hw;
2251         bool link_active = false;
2252
2253         /* get_link_status is set on LSC (link status) interrupt or
2254          * rx sequence error interrupt.  get_link_status will stay
2255          * false until the e1000_check_for_link establishes link
2256          * for copper adapters ONLY
2257          */
2258         switch (hw->media_type) {
2259         case e1000_media_type_copper:
2260                 if (hw->get_link_status) {
2261                         e1000_check_for_link(hw);
2262                         link_active = !hw->get_link_status;
2263                 } else {
2264                         link_active = true;
2265                 }
2266                 break;
2267         case e1000_media_type_fiber:
2268                 e1000_check_for_link(hw);
2269                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2270                 break;
2271         case e1000_media_type_internal_serdes:
2272                 e1000_check_for_link(hw);
2273                 link_active = hw->serdes_has_link;
2274                 break;
2275         default:
2276                 break;
2277         }
2278
2279         return link_active;
2280 }
2281
2282 /**
2283  * e1000_watchdog - Timer Call-back
2284  * @data: pointer to adapter cast into an unsigned long
2285  **/
2286 static void e1000_watchdog(unsigned long data)
2287 {
2288         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2289         struct e1000_hw *hw = &adapter->hw;
2290         struct net_device *netdev = adapter->netdev;
2291         struct e1000_tx_ring *txdr = adapter->tx_ring;
2292         u32 link, tctl;
2293
2294         link = e1000_has_link(adapter);
2295         if ((netif_carrier_ok(netdev)) && link)
2296                 goto link_up;
2297
2298         if (link) {
2299                 if (!netif_carrier_ok(netdev)) {
2300                         u32 ctrl;
2301                         bool txb2b = true;
2302                         /* update snapshot of PHY registers on LSC */
2303                         e1000_get_speed_and_duplex(hw,
2304                                                    &adapter->link_speed,
2305                                                    &adapter->link_duplex);
2306
2307                         ctrl = er32(CTRL);
2308                         printk(KERN_INFO "e1000: %s NIC Link is Up %d Mbps %s, "
2309                                "Flow Control: %s\n",
2310                                netdev->name,
2311                                adapter->link_speed,
2312                                adapter->link_duplex == FULL_DUPLEX ?
2313                                 "Full Duplex" : "Half Duplex",
2314                                 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2315                                 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2316                                 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2317                                 E1000_CTRL_TFCE) ? "TX" : "None" )));
2318
2319                         /* tweak tx_queue_len according to speed/duplex
2320                          * and adjust the timeout factor */
2321                         netdev->tx_queue_len = adapter->tx_queue_len;
2322                         adapter->tx_timeout_factor = 1;
2323                         switch (adapter->link_speed) {
2324                         case SPEED_10:
2325                                 txb2b = false;
2326                                 netdev->tx_queue_len = 10;
2327                                 adapter->tx_timeout_factor = 16;
2328                                 break;
2329                         case SPEED_100:
2330                                 txb2b = false;
2331                                 netdev->tx_queue_len = 100;
2332                                 /* maybe add some timeout factor ? */
2333                                 break;
2334                         }
2335
2336                         /* enable transmits in the hardware */
2337                         tctl = er32(TCTL);
2338                         tctl |= E1000_TCTL_EN;
2339                         ew32(TCTL, tctl);
2340
2341                         netif_carrier_on(netdev);
2342                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2343                                 mod_timer(&adapter->phy_info_timer,
2344                                           round_jiffies(jiffies + 2 * HZ));
2345                         adapter->smartspeed = 0;
2346                 }
2347         } else {
2348                 if (netif_carrier_ok(netdev)) {
2349                         adapter->link_speed = 0;
2350                         adapter->link_duplex = 0;
2351                         printk(KERN_INFO "e1000: %s NIC Link is Down\n",
2352                                netdev->name);
2353                         netif_carrier_off(netdev);
2354
2355                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2356                                 mod_timer(&adapter->phy_info_timer,
2357                                           round_jiffies(jiffies + 2 * HZ));
2358                 }
2359
2360                 e1000_smartspeed(adapter);
2361         }
2362
2363 link_up:
2364         e1000_update_stats(adapter);
2365
2366         hw->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2367         adapter->tpt_old = adapter->stats.tpt;
2368         hw->collision_delta = adapter->stats.colc - adapter->colc_old;
2369         adapter->colc_old = adapter->stats.colc;
2370
2371         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2372         adapter->gorcl_old = adapter->stats.gorcl;
2373         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2374         adapter->gotcl_old = adapter->stats.gotcl;
2375
2376         e1000_update_adaptive(hw);
2377
2378         if (!netif_carrier_ok(netdev)) {
2379                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2380                         /* We've lost link, so the controller stops DMA,
2381                          * but we've got queued Tx work that's never going
2382                          * to get done, so reset controller to flush Tx.
2383                          * (Do the reset outside of interrupt context). */
2384                         adapter->tx_timeout_count++;
2385                         schedule_work(&adapter->reset_task);
2386                         /* return immediately since reset is imminent */
2387                         return;
2388                 }
2389         }
2390
2391         /* Cause software interrupt to ensure rx ring is cleaned */
2392         ew32(ICS, E1000_ICS_RXDMT0);
2393
2394         /* Force detection of hung controller every watchdog period */
2395         adapter->detect_tx_hung = true;
2396
2397         /* Reset the timer */
2398         if (!test_bit(__E1000_DOWN, &adapter->flags))
2399                 mod_timer(&adapter->watchdog_timer,
2400                           round_jiffies(jiffies + 2 * HZ));
2401 }
2402
2403 enum latency_range {
2404         lowest_latency = 0,
2405         low_latency = 1,
2406         bulk_latency = 2,
2407         latency_invalid = 255
2408 };
2409
2410 /**
2411  * e1000_update_itr - update the dynamic ITR value based on statistics
2412  * @adapter: pointer to adapter
2413  * @itr_setting: current adapter->itr
2414  * @packets: the number of packets during this measurement interval
2415  * @bytes: the number of bytes during this measurement interval
2416  *
2417  *      Stores a new ITR value based on packets and byte
2418  *      counts during the last interrupt.  The advantage of per interrupt
2419  *      computation is faster updates and more accurate ITR for the current
2420  *      traffic pattern.  Constants in this function were computed
2421  *      based on theoretical maximum wire speed and thresholds were set based
2422  *      on testing data as well as attempting to minimize response time
2423  *      while increasing bulk throughput.
2424  *      this functionality is controlled by the InterruptThrottleRate module
2425  *      parameter (see e1000_param.c)
2426  **/
2427 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2428                                      u16 itr_setting, int packets, int bytes)
2429 {
2430         unsigned int retval = itr_setting;
2431         struct e1000_hw *hw = &adapter->hw;
2432
2433         if (unlikely(hw->mac_type < e1000_82540))
2434                 goto update_itr_done;
2435
2436         if (packets == 0)
2437                 goto update_itr_done;
2438
2439         switch (itr_setting) {
2440         case lowest_latency:
2441                 /* jumbo frames get bulk treatment*/
2442                 if (bytes/packets > 8000)
2443                         retval = bulk_latency;
2444                 else if ((packets < 5) && (bytes > 512))
2445                         retval = low_latency;
2446                 break;
2447         case low_latency:  /* 50 usec aka 20000 ints/s */
2448                 if (bytes > 10000) {
2449                         /* jumbo frames need bulk latency setting */
2450                         if (bytes/packets > 8000)
2451                                 retval = bulk_latency;
2452                         else if ((packets < 10) || ((bytes/packets) > 1200))
2453                                 retval = bulk_latency;
2454                         else if ((packets > 35))
2455                                 retval = lowest_latency;
2456                 } else if (bytes/packets > 2000)
2457                         retval = bulk_latency;
2458                 else if (packets <= 2 && bytes < 512)
2459                         retval = lowest_latency;
2460                 break;
2461         case bulk_latency: /* 250 usec aka 4000 ints/s */
2462                 if (bytes > 25000) {
2463                         if (packets > 35)
2464                                 retval = low_latency;
2465                 } else if (bytes < 6000) {
2466                         retval = low_latency;
2467                 }
2468                 break;
2469         }
2470
2471 update_itr_done:
2472         return retval;
2473 }
2474
2475 static void e1000_set_itr(struct e1000_adapter *adapter)
2476 {
2477         struct e1000_hw *hw = &adapter->hw;
2478         u16 current_itr;
2479         u32 new_itr = adapter->itr;
2480
2481         if (unlikely(hw->mac_type < e1000_82540))
2482                 return;
2483
2484         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2485         if (unlikely(adapter->link_speed != SPEED_1000)) {
2486                 current_itr = 0;
2487                 new_itr = 4000;
2488                 goto set_itr_now;
2489         }
2490
2491         adapter->tx_itr = e1000_update_itr(adapter,
2492                                     adapter->tx_itr,
2493                                     adapter->total_tx_packets,
2494                                     adapter->total_tx_bytes);
2495         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2496         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2497                 adapter->tx_itr = low_latency;
2498
2499         adapter->rx_itr = e1000_update_itr(adapter,
2500                                     adapter->rx_itr,
2501                                     adapter->total_rx_packets,
2502                                     adapter->total_rx_bytes);
2503         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2504         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2505                 adapter->rx_itr = low_latency;
2506
2507         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2508
2509         switch (current_itr) {
2510         /* counts and packets in update_itr are dependent on these numbers */
2511         case lowest_latency:
2512                 new_itr = 70000;
2513                 break;
2514         case low_latency:
2515                 new_itr = 20000; /* aka hwitr = ~200 */
2516                 break;
2517         case bulk_latency:
2518                 new_itr = 4000;
2519                 break;
2520         default:
2521                 break;
2522         }
2523
2524 set_itr_now:
2525         if (new_itr != adapter->itr) {
2526                 /* this attempts to bias the interrupt rate towards Bulk
2527                  * by adding intermediate steps when interrupt rate is
2528                  * increasing */
2529                 new_itr = new_itr > adapter->itr ?
2530                              min(adapter->itr + (new_itr >> 2), new_itr) :
2531                              new_itr;
2532                 adapter->itr = new_itr;
2533                 ew32(ITR, 1000000000 / (new_itr * 256));
2534         }
2535
2536         return;
2537 }
2538
2539 #define E1000_TX_FLAGS_CSUM             0x00000001
2540 #define E1000_TX_FLAGS_VLAN             0x00000002
2541 #define E1000_TX_FLAGS_TSO              0x00000004
2542 #define E1000_TX_FLAGS_IPV4             0x00000008
2543 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2544 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2545
2546 static int e1000_tso(struct e1000_adapter *adapter,
2547                      struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2548 {
2549         struct e1000_context_desc *context_desc;
2550         struct e1000_buffer *buffer_info;
2551         unsigned int i;
2552         u32 cmd_length = 0;
2553         u16 ipcse = 0, tucse, mss;
2554         u8 ipcss, ipcso, tucss, tucso, hdr_len;
2555         int err;
2556
2557         if (skb_is_gso(skb)) {
2558                 if (skb_header_cloned(skb)) {
2559                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2560                         if (err)
2561                                 return err;
2562                 }
2563
2564                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2565                 mss = skb_shinfo(skb)->gso_size;
2566                 if (skb->protocol == htons(ETH_P_IP)) {
2567                         struct iphdr *iph = ip_hdr(skb);
2568                         iph->tot_len = 0;
2569                         iph->check = 0;
2570                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2571                                                                  iph->daddr, 0,
2572                                                                  IPPROTO_TCP,
2573                                                                  0);
2574                         cmd_length = E1000_TXD_CMD_IP;
2575                         ipcse = skb_transport_offset(skb) - 1;
2576                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2577                         ipv6_hdr(skb)->payload_len = 0;
2578                         tcp_hdr(skb)->check =
2579                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2580                                                  &ipv6_hdr(skb)->daddr,
2581                                                  0, IPPROTO_TCP, 0);
2582                         ipcse = 0;
2583                 }
2584                 ipcss = skb_network_offset(skb);
2585                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2586                 tucss = skb_transport_offset(skb);
2587                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2588                 tucse = 0;
2589
2590                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2591                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2592
2593                 i = tx_ring->next_to_use;
2594                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2595                 buffer_info = &tx_ring->buffer_info[i];
2596
2597                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2598                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2599                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2600                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2601                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2602                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2603                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2604                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2605                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2606
2607                 buffer_info->time_stamp = jiffies;
2608                 buffer_info->next_to_watch = i;
2609
2610                 if (++i == tx_ring->count) i = 0;
2611                 tx_ring->next_to_use = i;
2612
2613                 return true;
2614         }
2615         return false;
2616 }
2617
2618 static bool e1000_tx_csum(struct e1000_adapter *adapter,
2619                           struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2620 {
2621         struct e1000_context_desc *context_desc;
2622         struct e1000_buffer *buffer_info;
2623         unsigned int i;
2624         u8 css;
2625         u32 cmd_len = E1000_TXD_CMD_DEXT;
2626
2627         if (skb->ip_summed != CHECKSUM_PARTIAL)
2628                 return false;
2629
2630         switch (skb->protocol) {
2631         case cpu_to_be16(ETH_P_IP):
2632                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2633                         cmd_len |= E1000_TXD_CMD_TCP;
2634                 break;
2635         case cpu_to_be16(ETH_P_IPV6):
2636                 /* XXX not handling all IPV6 headers */
2637                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2638                         cmd_len |= E1000_TXD_CMD_TCP;
2639                 break;
2640         default:
2641                 if (unlikely(net_ratelimit()))
2642                         DPRINTK(DRV, WARNING,
2643                                 "checksum_partial proto=%x!\n", skb->protocol);
2644                 break;
2645         }
2646
2647         css = skb_transport_offset(skb);
2648
2649         i = tx_ring->next_to_use;
2650         buffer_info = &tx_ring->buffer_info[i];
2651         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2652
2653         context_desc->lower_setup.ip_config = 0;
2654         context_desc->upper_setup.tcp_fields.tucss = css;
2655         context_desc->upper_setup.tcp_fields.tucso =
2656                 css + skb->csum_offset;
2657         context_desc->upper_setup.tcp_fields.tucse = 0;
2658         context_desc->tcp_seg_setup.data = 0;
2659         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
2660
2661         buffer_info->time_stamp = jiffies;
2662         buffer_info->next_to_watch = i;
2663
2664         if (unlikely(++i == tx_ring->count)) i = 0;
2665         tx_ring->next_to_use = i;
2666
2667         return true;
2668 }
2669
2670 #define E1000_MAX_TXD_PWR       12
2671 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2672
2673 static int e1000_tx_map(struct e1000_adapter *adapter,
2674                         struct e1000_tx_ring *tx_ring,
2675                         struct sk_buff *skb, unsigned int first,
2676                         unsigned int max_per_txd, unsigned int nr_frags,
2677                         unsigned int mss)
2678 {
2679         struct e1000_hw *hw = &adapter->hw;
2680         struct pci_dev *pdev = adapter->pdev;
2681         struct e1000_buffer *buffer_info;
2682         unsigned int len = skb_headlen(skb);
2683         unsigned int offset = 0, size, count = 0, i;
2684         unsigned int f;
2685
2686         i = tx_ring->next_to_use;
2687
2688         while (len) {
2689                 buffer_info = &tx_ring->buffer_info[i];
2690                 size = min(len, max_per_txd);
2691                 /* Workaround for Controller erratum --
2692                  * descriptor for non-tso packet in a linear SKB that follows a
2693                  * tso gets written back prematurely before the data is fully
2694                  * DMA'd to the controller */
2695                 if (!skb->data_len && tx_ring->last_tx_tso &&
2696                     !skb_is_gso(skb)) {
2697                         tx_ring->last_tx_tso = 0;
2698                         size -= 4;
2699                 }
2700
2701                 /* Workaround for premature desc write-backs
2702                  * in TSO mode.  Append 4-byte sentinel desc */
2703                 if (unlikely(mss && !nr_frags && size == len && size > 8))
2704                         size -= 4;
2705                 /* work-around for errata 10 and it applies
2706                  * to all controllers in PCI-X mode
2707                  * The fix is to make sure that the first descriptor of a
2708                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2709                  */
2710                 if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
2711                                 (size > 2015) && count == 0))
2712                         size = 2015;
2713
2714                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
2715                  * terminating buffers within evenly-aligned dwords. */
2716                 if (unlikely(adapter->pcix_82544 &&
2717                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2718                    size > 4))
2719                         size -= 4;
2720
2721                 buffer_info->length = size;
2722                 /* set time_stamp *before* dma to help avoid a possible race */
2723                 buffer_info->time_stamp = jiffies;
2724                 buffer_info->mapped_as_page = false;
2725                 buffer_info->dma = pci_map_single(pdev, skb->data + offset,
2726                                                   size, PCI_DMA_TODEVICE);
2727                 if (pci_dma_mapping_error(pdev, buffer_info->dma))
2728                         goto dma_error;
2729                 buffer_info->next_to_watch = i;
2730
2731                 len -= size;
2732                 offset += size;
2733                 count++;
2734                 if (len) {
2735                         i++;
2736                         if (unlikely(i == tx_ring->count))
2737                                 i = 0;
2738                 }
2739         }
2740
2741         for (f = 0; f < nr_frags; f++) {
2742                 struct skb_frag_struct *frag;
2743
2744                 frag = &skb_shinfo(skb)->frags[f];
2745                 len = frag->size;
2746                 offset = frag->page_offset;
2747
2748                 while (len) {
2749                         i++;
2750                         if (unlikely(i == tx_ring->count))
2751                                 i = 0;
2752
2753                         buffer_info = &tx_ring->buffer_info[i];
2754                         size = min(len, max_per_txd);
2755                         /* Workaround for premature desc write-backs
2756                          * in TSO mode.  Append 4-byte sentinel desc */
2757                         if (unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
2758                                 size -= 4;
2759                         /* Workaround for potential 82544 hang in PCI-X.
2760                          * Avoid terminating buffers within evenly-aligned
2761                          * dwords. */
2762                         if (unlikely(adapter->pcix_82544 &&
2763                             !((unsigned long)(page_to_phys(frag->page) + offset
2764                                               + size - 1) & 4) &&
2765                             size > 4))
2766                                 size -= 4;
2767
2768                         buffer_info->length = size;
2769                         buffer_info->time_stamp = jiffies;
2770                         buffer_info->mapped_as_page = true;
2771                         buffer_info->dma = pci_map_page(pdev, frag->page,
2772                                                         offset, size,
2773                                                         PCI_DMA_TODEVICE);
2774                         if (pci_dma_mapping_error(pdev, buffer_info->dma))
2775                                 goto dma_error;
2776                         buffer_info->next_to_watch = i;
2777
2778                         len -= size;
2779                         offset += size;
2780                         count++;
2781                 }
2782         }
2783
2784         tx_ring->buffer_info[i].skb = skb;
2785         tx_ring->buffer_info[first].next_to_watch = i;
2786
2787         return count;
2788
2789 dma_error:
2790         dev_err(&pdev->dev, "TX DMA map failed\n");
2791         buffer_info->dma = 0;
2792         if (count)
2793                 count--;
2794
2795         while (count--) {
2796                 if (i==0)
2797                         i += tx_ring->count;
2798                 i--;
2799                 buffer_info = &tx_ring->buffer_info[i];
2800                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2801         }
2802
2803         return 0;
2804 }
2805
2806 static void e1000_tx_queue(struct e1000_adapter *adapter,
2807                            struct e1000_tx_ring *tx_ring, int tx_flags,
2808                            int count)
2809 {
2810         struct e1000_hw *hw = &adapter->hw;
2811         struct e1000_tx_desc *tx_desc = NULL;
2812         struct e1000_buffer *buffer_info;
2813         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2814         unsigned int i;
2815
2816         if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
2817                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2818                              E1000_TXD_CMD_TSE;
2819                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2820
2821                 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
2822                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2823         }
2824
2825         if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
2826                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
2827                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2828         }
2829
2830         if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
2831                 txd_lower |= E1000_TXD_CMD_VLE;
2832                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
2833         }
2834
2835         i = tx_ring->next_to_use;
2836
2837         while (count--) {
2838                 buffer_info = &tx_ring->buffer_info[i];
2839                 tx_desc = E1000_TX_DESC(*tx_ring, i);
2840                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
2841                 tx_desc->lower.data =
2842                         cpu_to_le32(txd_lower | buffer_info->length);
2843                 tx_desc->upper.data = cpu_to_le32(txd_upper);
2844                 if (unlikely(++i == tx_ring->count)) i = 0;
2845         }
2846
2847         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
2848
2849         /* Force memory writes to complete before letting h/w
2850          * know there are new descriptors to fetch.  (Only
2851          * applicable for weak-ordered memory model archs,
2852          * such as IA-64). */
2853         wmb();
2854
2855         tx_ring->next_to_use = i;
2856         writel(i, hw->hw_addr + tx_ring->tdt);
2857         /* we need this if more than one processor can write to our tail
2858          * at a time, it syncronizes IO on IA64/Altix systems */
2859         mmiowb();
2860 }
2861
2862 /**
2863  * 82547 workaround to avoid controller hang in half-duplex environment.
2864  * The workaround is to avoid queuing a large packet that would span
2865  * the internal Tx FIFO ring boundary by notifying the stack to resend
2866  * the packet at a later time.  This gives the Tx FIFO an opportunity to
2867  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
2868  * to the beginning of the Tx FIFO.
2869  **/
2870
2871 #define E1000_FIFO_HDR                  0x10
2872 #define E1000_82547_PAD_LEN             0x3E0
2873
2874 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
2875                                        struct sk_buff *skb)
2876 {
2877         u32 fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
2878         u32 skb_fifo_len = skb->len + E1000_FIFO_HDR;
2879
2880         skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
2881
2882         if (adapter->link_duplex != HALF_DUPLEX)
2883                 goto no_fifo_stall_required;
2884
2885         if (atomic_read(&adapter->tx_fifo_stall))
2886                 return 1;
2887
2888         if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
2889                 atomic_set(&adapter->tx_fifo_stall, 1);
2890                 return 1;
2891         }
2892
2893 no_fifo_stall_required:
2894         adapter->tx_fifo_head += skb_fifo_len;
2895         if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
2896                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
2897         return 0;
2898 }
2899
2900 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
2901 {
2902         struct e1000_adapter *adapter = netdev_priv(netdev);
2903         struct e1000_tx_ring *tx_ring = adapter->tx_ring;
2904
2905         netif_stop_queue(netdev);
2906         /* Herbert's original patch had:
2907          *  smp_mb__after_netif_stop_queue();
2908          * but since that doesn't exist yet, just open code it. */
2909         smp_mb();
2910
2911         /* We need to check again in a case another CPU has just
2912          * made room available. */
2913         if (likely(E1000_DESC_UNUSED(tx_ring) < size))
2914                 return -EBUSY;
2915
2916         /* A reprieve! */
2917         netif_start_queue(netdev);
2918         ++adapter->restart_queue;
2919         return 0;
2920 }
2921
2922 static int e1000_maybe_stop_tx(struct net_device *netdev,
2923                                struct e1000_tx_ring *tx_ring, int size)
2924 {
2925         if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
2926                 return 0;
2927         return __e1000_maybe_stop_tx(netdev, size);
2928 }
2929
2930 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
2931 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
2932                                     struct net_device *netdev)
2933 {
2934         struct e1000_adapter *adapter = netdev_priv(netdev);
2935         struct e1000_hw *hw = &adapter->hw;
2936         struct e1000_tx_ring *tx_ring;
2937         unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
2938         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
2939         unsigned int tx_flags = 0;
2940         unsigned int len = skb->len - skb->data_len;
2941         unsigned int nr_frags;
2942         unsigned int mss;
2943         int count = 0;
2944         int tso;
2945         unsigned int f;
2946
2947         /* This goes back to the question of how to logically map a tx queue
2948          * to a flow.  Right now, performance is impacted slightly negatively
2949          * if using multiple tx queues.  If the stack breaks away from a
2950          * single qdisc implementation, we can look at this again. */
2951         tx_ring = adapter->tx_ring;
2952
2953         if (unlikely(skb->len <= 0)) {
2954                 dev_kfree_skb_any(skb);
2955                 return NETDEV_TX_OK;
2956         }
2957
2958         mss = skb_shinfo(skb)->gso_size;
2959         /* The controller does a simple calculation to
2960          * make sure there is enough room in the FIFO before
2961          * initiating the DMA for each buffer.  The calc is:
2962          * 4 = ceil(buffer len/mss).  To make sure we don't
2963          * overrun the FIFO, adjust the max buffer len if mss
2964          * drops. */
2965         if (mss) {
2966                 u8 hdr_len;
2967                 max_per_txd = min(mss << 2, max_per_txd);
2968                 max_txd_pwr = fls(max_per_txd) - 1;
2969
2970                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2971                 if (skb->data_len && hdr_len == len) {
2972                         switch (hw->mac_type) {
2973                                 unsigned int pull_size;
2974                         case e1000_82544:
2975                                 /* Make sure we have room to chop off 4 bytes,
2976                                  * and that the end alignment will work out to
2977                                  * this hardware's requirements
2978                                  * NOTE: this is a TSO only workaround
2979                                  * if end byte alignment not correct move us
2980                                  * into the next dword */
2981                                 if ((unsigned long)(skb_tail_pointer(skb) - 1) & 4)
2982                                         break;
2983                                 /* fall through */
2984                                 pull_size = min((unsigned int)4, skb->data_len);
2985                                 if (!__pskb_pull_tail(skb, pull_size)) {
2986                                         DPRINTK(DRV, ERR,
2987                                                 "__pskb_pull_tail failed.\n");
2988                                         dev_kfree_skb_any(skb);
2989                                         return NETDEV_TX_OK;
2990                                 }
2991                                 len = skb->len - skb->data_len;
2992                                 break;
2993                         default:
2994                                 /* do nothing */
2995                                 break;
2996                         }
2997                 }
2998         }
2999
3000         /* reserve a descriptor for the offload context */
3001         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3002                 count++;
3003         count++;
3004
3005         /* Controller Erratum workaround */
3006         if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3007                 count++;
3008
3009         count += TXD_USE_COUNT(len, max_txd_pwr);
3010
3011         if (adapter->pcix_82544)
3012                 count++;
3013
3014         /* work-around for errata 10 and it applies to all controllers
3015          * in PCI-X mode, so add one more descriptor to the count
3016          */
3017         if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
3018                         (len > 2015)))
3019                 count++;
3020
3021         nr_frags = skb_shinfo(skb)->nr_frags;
3022         for (f = 0; f < nr_frags; f++)
3023                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3024                                        max_txd_pwr);
3025         if (adapter->pcix_82544)
3026                 count += nr_frags;
3027
3028         /* need: count + 2 desc gap to keep tail from touching
3029          * head, otherwise try next time */
3030         if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2)))
3031                 return NETDEV_TX_BUSY;
3032
3033         if (unlikely(hw->mac_type == e1000_82547)) {
3034                 if (unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
3035                         netif_stop_queue(netdev);
3036                         if (!test_bit(__E1000_DOWN, &adapter->flags))
3037                                 mod_timer(&adapter->tx_fifo_stall_timer,
3038                                           jiffies + 1);
3039                         return NETDEV_TX_BUSY;
3040                 }
3041         }
3042
3043         if (unlikely(adapter->vlgrp && vlan_tx_tag_present(skb))) {
3044                 tx_flags |= E1000_TX_FLAGS_VLAN;
3045                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3046         }
3047
3048         first = tx_ring->next_to_use;
3049
3050         tso = e1000_tso(adapter, tx_ring, skb);
3051         if (tso < 0) {
3052                 dev_kfree_skb_any(skb);
3053                 return NETDEV_TX_OK;
3054         }
3055
3056         if (likely(tso)) {
3057                 if (likely(hw->mac_type != e1000_82544))
3058                         tx_ring->last_tx_tso = 1;
3059                 tx_flags |= E1000_TX_FLAGS_TSO;
3060         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3061                 tx_flags |= E1000_TX_FLAGS_CSUM;
3062
3063         if (likely(skb->protocol == htons(ETH_P_IP)))
3064                 tx_flags |= E1000_TX_FLAGS_IPV4;
3065
3066         count = e1000_tx_map(adapter, tx_ring, skb, first, max_per_txd,
3067                              nr_frags, mss);
3068
3069         if (count) {
3070                 e1000_tx_queue(adapter, tx_ring, tx_flags, count);
3071                 /* Make sure there is space in the ring for the next send. */
3072                 e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3073
3074         } else {
3075                 dev_kfree_skb_any(skb);
3076                 tx_ring->buffer_info[first].time_stamp = 0;
3077                 tx_ring->next_to_use = first;
3078         }
3079
3080         return NETDEV_TX_OK;
3081 }
3082
3083 /**
3084  * e1000_tx_timeout - Respond to a Tx Hang
3085  * @netdev: network interface device structure
3086  **/
3087
3088 static void e1000_tx_timeout(struct net_device *netdev)
3089 {
3090         struct e1000_adapter *adapter = netdev_priv(netdev);
3091
3092         /* Do the reset outside of interrupt context */
3093         adapter->tx_timeout_count++;
3094         schedule_work(&adapter->reset_task);
3095 }
3096
3097 static void e1000_reset_task(struct work_struct *work)
3098 {
3099         struct e1000_adapter *adapter =
3100                 container_of(work, struct e1000_adapter, reset_task);
3101
3102         e1000_reinit_locked(adapter);
3103 }
3104
3105 /**
3106  * e1000_get_stats - Get System Network Statistics
3107  * @netdev: network interface device structure
3108  *
3109  * Returns the address of the device statistics structure.
3110  * The statistics are actually updated from the timer callback.
3111  **/
3112
3113 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3114 {
3115         /* only return the current stats */
3116         return &netdev->stats;
3117 }
3118
3119 /**
3120  * e1000_change_mtu - Change the Maximum Transfer Unit
3121  * @netdev: network interface device structure
3122  * @new_mtu: new value for maximum frame size
3123  *
3124  * Returns 0 on success, negative on failure
3125  **/
3126
3127 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3128 {
3129         struct e1000_adapter *adapter = netdev_priv(netdev);
3130         struct e1000_hw *hw = &adapter->hw;
3131         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3132
3133         if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3134             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3135                 DPRINTK(PROBE, ERR, "Invalid MTU setting\n");
3136                 return -EINVAL;
3137         }
3138
3139         /* Adapter-specific max frame size limits. */
3140         switch (hw->mac_type) {
3141         case e1000_undefined ... e1000_82542_rev2_1:
3142                 if (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)) {
3143                         DPRINTK(PROBE, ERR, "Jumbo Frames not supported.\n");
3144                         return -EINVAL;
3145                 }
3146                 break;
3147         default:
3148                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3149                 break;
3150         }
3151
3152         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
3153                 msleep(1);
3154         /* e1000_down has a dependency on max_frame_size */
3155         hw->max_frame_size = max_frame;
3156         if (netif_running(netdev))
3157                 e1000_down(adapter);
3158
3159         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3160          * means we reserve 2 more, this pushes us to allocate from the next
3161          * larger slab size.
3162          * i.e. RXBUFFER_2048 --> size-4096 slab
3163          *  however with the new *_jumbo_rx* routines, jumbo receives will use
3164          *  fragmented skbs */
3165
3166         if (max_frame <= E1000_RXBUFFER_2048)
3167                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3168         else
3169 #if (PAGE_SIZE >= E1000_RXBUFFER_16384)
3170                 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3171 #elif (PAGE_SIZE >= E1000_RXBUFFER_4096)
3172                 adapter->rx_buffer_len = PAGE_SIZE;
3173 #endif
3174
3175         /* adjust allocation if LPE protects us, and we aren't using SBP */
3176         if (!hw->tbi_compatibility_on &&
3177             ((max_frame == (ETH_FRAME_LEN + ETH_FCS_LEN)) ||
3178              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3179                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3180
3181         printk(KERN_INFO "e1000: %s changing MTU from %d to %d\n",
3182                netdev->name, netdev->mtu, new_mtu);
3183         netdev->mtu = new_mtu;
3184
3185         if (netif_running(netdev))
3186                 e1000_up(adapter);
3187         else
3188                 e1000_reset(adapter);
3189
3190         clear_bit(__E1000_RESETTING, &adapter->flags);
3191
3192         return 0;
3193 }
3194
3195 /**
3196  * e1000_update_stats - Update the board statistics counters
3197  * @adapter: board private structure
3198  **/
3199
3200 void e1000_update_stats(struct e1000_adapter *adapter)
3201 {
3202         struct net_device *netdev = adapter->netdev;
3203         struct e1000_hw *hw = &adapter->hw;
3204         struct pci_dev *pdev = adapter->pdev;
3205         unsigned long flags;
3206         u16 phy_tmp;
3207
3208 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3209
3210         /*
3211          * Prevent stats update while adapter is being reset, or if the pci
3212          * connection is down.
3213          */
3214         if (adapter->link_speed == 0)
3215                 return;
3216         if (pci_channel_offline(pdev))
3217                 return;
3218
3219         spin_lock_irqsave(&adapter->stats_lock, flags);
3220
3221         /* these counters are modified from e1000_tbi_adjust_stats,
3222          * called from the interrupt context, so they must only
3223          * be written while holding adapter->stats_lock
3224          */
3225
3226         adapter->stats.crcerrs += er32(CRCERRS);
3227         adapter->stats.gprc += er32(GPRC);
3228         adapter->stats.gorcl += er32(GORCL);
3229         adapter->stats.gorch += er32(GORCH);
3230         adapter->stats.bprc += er32(BPRC);
3231         adapter->stats.mprc += er32(MPRC);
3232         adapter->stats.roc += er32(ROC);
3233
3234         adapter->stats.prc64 += er32(PRC64);
3235         adapter->stats.prc127 += er32(PRC127);
3236         adapter->stats.prc255 += er32(PRC255);
3237         adapter->stats.prc511 += er32(PRC511);
3238         adapter->stats.prc1023 += er32(PRC1023);
3239         adapter->stats.prc1522 += er32(PRC1522);
3240
3241         adapter->stats.symerrs += er32(SYMERRS);
3242         adapter->stats.mpc += er32(MPC);
3243         adapter->stats.scc += er32(SCC);
3244         adapter->stats.ecol += er32(ECOL);
3245         adapter->stats.mcc += er32(MCC);
3246         adapter->stats.latecol += er32(LATECOL);
3247         adapter->stats.dc += er32(DC);
3248         adapter->stats.sec += er32(SEC);
3249         adapter->stats.rlec += er32(RLEC);
3250         adapter->stats.xonrxc += er32(XONRXC);
3251         adapter->stats.xontxc += er32(XONTXC);
3252         adapter->stats.xoffrxc += er32(XOFFRXC);
3253         adapter->stats.xofftxc += er32(XOFFTXC);
3254         adapter->stats.fcruc += er32(FCRUC);
3255         adapter->stats.gptc += er32(GPTC);
3256         adapter->stats.gotcl += er32(GOTCL);
3257         adapter->stats.gotch += er32(GOTCH);
3258         adapter->stats.rnbc += er32(RNBC);
3259         adapter->stats.ruc += er32(RUC);
3260         adapter->stats.rfc += er32(RFC);
3261         adapter->stats.rjc += er32(RJC);
3262         adapter->stats.torl += er32(TORL);
3263         adapter->stats.torh += er32(TORH);
3264         adapter->stats.totl += er32(TOTL);
3265         adapter->stats.toth += er32(TOTH);
3266         adapter->stats.tpr += er32(TPR);
3267
3268         adapter->stats.ptc64 += er32(PTC64);
3269         adapter->stats.ptc127 += er32(PTC127);
3270         adapter->stats.ptc255 += er32(PTC255);
3271         adapter->stats.ptc511 += er32(PTC511);
3272         adapter->stats.ptc1023 += er32(PTC1023);
3273         adapter->stats.ptc1522 += er32(PTC1522);
3274
3275         adapter->stats.mptc += er32(MPTC);
3276         adapter->stats.bptc += er32(BPTC);
3277
3278         /* used for adaptive IFS */
3279
3280         hw->tx_packet_delta = er32(TPT);
3281         adapter->stats.tpt += hw->tx_packet_delta;
3282         hw->collision_delta = er32(COLC);
3283         adapter->stats.colc += hw->collision_delta;
3284
3285         if (hw->mac_type >= e1000_82543) {
3286                 adapter->stats.algnerrc += er32(ALGNERRC);
3287                 adapter->stats.rxerrc += er32(RXERRC);
3288                 adapter->stats.tncrs += er32(TNCRS);
3289                 adapter->stats.cexterr += er32(CEXTERR);
3290                 adapter->stats.tsctc += er32(TSCTC);
3291                 adapter->stats.tsctfc += er32(TSCTFC);
3292         }
3293
3294         /* Fill out the OS statistics structure */
3295         netdev->stats.multicast = adapter->stats.mprc;
3296         netdev->stats.collisions = adapter->stats.colc;
3297
3298         /* Rx Errors */
3299
3300         /* RLEC on some newer hardware can be incorrect so build
3301         * our own version based on RUC and ROC */
3302         netdev->stats.rx_errors = adapter->stats.rxerrc +
3303                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3304                 adapter->stats.ruc + adapter->stats.roc +
3305                 adapter->stats.cexterr;
3306         adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3307         netdev->stats.rx_length_errors = adapter->stats.rlerrc;
3308         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3309         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3310         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3311
3312         /* Tx Errors */
3313         adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3314         netdev->stats.tx_errors = adapter->stats.txerrc;
3315         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3316         netdev->stats.tx_window_errors = adapter->stats.latecol;
3317         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3318         if (hw->bad_tx_carr_stats_fd &&
3319             adapter->link_duplex == FULL_DUPLEX) {
3320                 netdev->stats.tx_carrier_errors = 0;
3321                 adapter->stats.tncrs = 0;
3322         }
3323
3324         /* Tx Dropped needs to be maintained elsewhere */
3325
3326         /* Phy Stats */
3327         if (hw->media_type == e1000_media_type_copper) {
3328                 if ((adapter->link_speed == SPEED_1000) &&
3329                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3330                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3331                         adapter->phy_stats.idle_errors += phy_tmp;
3332                 }
3333
3334                 if ((hw->mac_type <= e1000_82546) &&
3335                    (hw->phy_type == e1000_phy_m88) &&
3336                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3337                         adapter->phy_stats.receive_errors += phy_tmp;
3338         }
3339
3340         /* Management Stats */
3341         if (hw->has_smbus) {
3342                 adapter->stats.mgptc += er32(MGTPTC);
3343                 adapter->stats.mgprc += er32(MGTPRC);
3344                 adapter->stats.mgpdc += er32(MGTPDC);
3345         }
3346
3347         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3348 }
3349
3350 /**
3351  * e1000_intr - Interrupt Handler
3352  * @irq: interrupt number
3353  * @data: pointer to a network interface device structure
3354  **/
3355
3356 static irqreturn_t e1000_intr(int irq, void *data)
3357 {
3358         struct net_device *netdev = data;
3359         struct e1000_adapter *adapter = netdev_priv(netdev);
3360         struct e1000_hw *hw = &adapter->hw;
3361         u32 icr = er32(ICR);
3362
3363         if (unlikely((!icr) || test_bit(__E1000_DOWN, &adapter->flags)))
3364                 return IRQ_NONE;  /* Not our interrupt */
3365
3366         if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3367                 hw->get_link_status = 1;
3368                 /* guard against interrupt when we're going down */
3369                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3370                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3371         }
3372
3373         /* disable interrupts, without the synchronize_irq bit */
3374         ew32(IMC, ~0);
3375         E1000_WRITE_FLUSH();
3376
3377         if (likely(napi_schedule_prep(&adapter->napi))) {
3378                 adapter->total_tx_bytes = 0;
3379                 adapter->total_tx_packets = 0;
3380                 adapter->total_rx_bytes = 0;
3381                 adapter->total_rx_packets = 0;
3382                 __napi_schedule(&adapter->napi);
3383         } else {
3384                 /* this really should not happen! if it does it is basically a
3385                  * bug, but not a hard error, so enable ints and continue */
3386                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3387                         e1000_irq_enable(adapter);
3388         }
3389
3390         return IRQ_HANDLED;
3391 }
3392
3393 /**
3394  * e1000_clean - NAPI Rx polling callback
3395  * @adapter: board private structure
3396  **/
3397 static int e1000_clean(struct napi_struct *napi, int budget)
3398 {
3399         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
3400         int tx_clean_complete = 0, work_done = 0;
3401
3402         tx_clean_complete = e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);
3403
3404         adapter->clean_rx(adapter, &adapter->rx_ring[0], &work_done, budget);
3405
3406         if (!tx_clean_complete)
3407                 work_done = budget;
3408
3409         /* If budget not fully consumed, exit the polling mode */
3410         if (work_done < budget) {
3411                 if (likely(adapter->itr_setting & 3))
3412                         e1000_set_itr(adapter);
3413                 napi_complete(napi);
3414                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3415                         e1000_irq_enable(adapter);
3416         }
3417
3418         return work_done;
3419 }
3420
3421 /**
3422  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3423  * @adapter: board private structure
3424  **/
3425 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
3426                                struct e1000_tx_ring *tx_ring)
3427 {
3428         struct e1000_hw *hw = &adapter->hw;
3429         struct net_device *netdev = adapter->netdev;
3430         struct e1000_tx_desc *tx_desc, *eop_desc;
3431         struct e1000_buffer *buffer_info;
3432         unsigned int i, eop;
3433         unsigned int count = 0;
3434         unsigned int total_tx_bytes=0, total_tx_packets=0;
3435
3436         i = tx_ring->next_to_clean;
3437         eop = tx_ring->buffer_info[i].next_to_watch;
3438         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3439
3440         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
3441                (count < tx_ring->count)) {
3442                 bool cleaned = false;
3443                 for ( ; !cleaned; count++) {
3444                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3445                         buffer_info = &tx_ring->buffer_info[i];
3446                         cleaned = (i == eop);
3447
3448                         if (cleaned) {
3449                                 struct sk_buff *skb = buffer_info->skb;
3450                                 unsigned int segs, bytecount;
3451                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3452                                 /* multiply data chunks by size of headers */
3453                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3454                                             skb->len;
3455                                 total_tx_packets += segs;
3456                                 total_tx_bytes += bytecount;
3457                         }
3458                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3459                         tx_desc->upper.data = 0;
3460
3461                         if (unlikely(++i == tx_ring->count)) i = 0;
3462                 }
3463
3464                 eop = tx_ring->buffer_info[i].next_to_watch;
3465                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3466         }
3467
3468         tx_ring->next_to_clean = i;
3469
3470 #define TX_WAKE_THRESHOLD 32
3471         if (unlikely(count && netif_carrier_ok(netdev) &&
3472                      E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
3473                 /* Make sure that anybody stopping the queue after this
3474                  * sees the new next_to_clean.
3475                  */
3476                 smp_mb();
3477
3478                 if (netif_queue_stopped(netdev) &&
3479                     !(test_bit(__E1000_DOWN, &adapter->flags))) {
3480                         netif_wake_queue(netdev);
3481                         ++adapter->restart_queue;
3482                 }
3483         }
3484
3485         if (adapter->detect_tx_hung) {
3486                 /* Detect a transmit hang in hardware, this serializes the
3487                  * check with the clearing of time_stamp and movement of i */
3488                 adapter->detect_tx_hung = false;
3489                 if (tx_ring->buffer_info[eop].time_stamp &&
3490                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
3491                                (adapter->tx_timeout_factor * HZ)) &&
3492                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
3493
3494                         /* detected Tx unit hang */
3495                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
3496                                         "  Tx Queue             <%lu>\n"
3497                                         "  TDH                  <%x>\n"
3498                                         "  TDT                  <%x>\n"
3499                                         "  next_to_use          <%x>\n"
3500                                         "  next_to_clean        <%x>\n"
3501                                         "buffer_info[next_to_clean]\n"
3502                                         "  time_stamp           <%lx>\n"
3503                                         "  next_to_watch        <%x>\n"
3504                                         "  jiffies              <%lx>\n"
3505                                         "  next_to_watch.status <%x>\n",
3506                                 (unsigned long)((tx_ring - adapter->tx_ring) /
3507                                         sizeof(struct e1000_tx_ring)),
3508                                 readl(hw->hw_addr + tx_ring->tdh),
3509                                 readl(hw->hw_addr + tx_ring->tdt),
3510                                 tx_ring->next_to_use,
3511                                 tx_ring->next_to_clean,
3512                                 tx_ring->buffer_info[eop].time_stamp,
3513                                 eop,
3514                                 jiffies,
3515                                 eop_desc->upper.fields.status);
3516                         netif_stop_queue(netdev);
3517                 }
3518         }
3519         adapter->total_tx_bytes += total_tx_bytes;
3520         adapter->total_tx_packets += total_tx_packets;
3521         netdev->stats.tx_bytes += total_tx_bytes;
3522         netdev->stats.tx_packets += total_tx_packets;
3523         return (count < tx_ring->count);
3524 }
3525
3526 /**
3527  * e1000_rx_checksum - Receive Checksum Offload for 82543
3528  * @adapter:     board private structure
3529  * @status_err:  receive descriptor status and error fields
3530  * @csum:        receive descriptor csum field
3531  * @sk_buff:     socket buffer with received data
3532  **/
3533
3534 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
3535                               u32 csum, struct sk_buff *skb)
3536 {
3537         struct e1000_hw *hw = &adapter->hw;
3538         u16 status = (u16)status_err;
3539         u8 errors = (u8)(status_err >> 24);
3540         skb->ip_summed = CHECKSUM_NONE;
3541
3542         /* 82543 or newer only */
3543         if (unlikely(hw->mac_type < e1000_82543)) return;
3544         /* Ignore Checksum bit is set */
3545         if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
3546         /* TCP/UDP checksum error bit is set */
3547         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
3548                 /* let the stack verify checksum errors */
3549                 adapter->hw_csum_err++;
3550                 return;
3551         }
3552         /* TCP/UDP Checksum has not been calculated */
3553         if (!(status & E1000_RXD_STAT_TCPCS))
3554                 return;
3555
3556         /* It must be a TCP or UDP packet with a valid checksum */
3557         if (likely(status & E1000_RXD_STAT_TCPCS)) {
3558                 /* TCP checksum is good */
3559                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3560         }
3561         adapter->hw_csum_good++;
3562 }
3563
3564 /**
3565  * e1000_consume_page - helper function
3566  **/
3567 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
3568                                u16 length)
3569 {
3570         bi->page = NULL;
3571         skb->len += length;
3572         skb->data_len += length;
3573         skb->truesize += length;
3574 }
3575
3576 /**
3577  * e1000_receive_skb - helper function to handle rx indications
3578  * @adapter: board private structure
3579  * @status: descriptor status field as written by hardware
3580  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3581  * @skb: pointer to sk_buff to be indicated to stack
3582  */
3583 static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
3584                               __le16 vlan, struct sk_buff *skb)
3585 {
3586         if (unlikely(adapter->vlgrp && (status & E1000_RXD_STAT_VP))) {
3587                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3588                                          le16_to_cpu(vlan) &
3589                                          E1000_RXD_SPC_VLAN_MASK);
3590         } else {
3591                 netif_receive_skb(skb);
3592         }
3593 }
3594
3595 /**
3596  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
3597  * @adapter: board private structure
3598  * @rx_ring: ring to clean
3599  * @work_done: amount of napi work completed this call
3600  * @work_to_do: max amount of work allowed for this call to do
3601  *
3602  * the return value indicates whether actual cleaning was done, there
3603  * is no guarantee that everything was cleaned
3604  */
3605 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
3606                                      struct e1000_rx_ring *rx_ring,
3607                                      int *work_done, int work_to_do)
3608 {
3609         struct e1000_hw *hw = &adapter->hw;
3610         struct net_device *netdev = adapter->netdev;
3611         struct pci_dev *pdev = adapter->pdev;
3612         struct e1000_rx_desc *rx_desc, *next_rxd;
3613         struct e1000_buffer *buffer_info, *next_buffer;
3614         unsigned long irq_flags;
3615         u32 length;
3616         unsigned int i;
3617         int cleaned_count = 0;
3618         bool cleaned = false;
3619         unsigned int total_rx_bytes=0, total_rx_packets=0;
3620
3621         i = rx_ring->next_to_clean;
3622         rx_desc = E1000_RX_DESC(*rx_ring, i);
3623         buffer_info = &rx_ring->buffer_info[i];
3624
3625         while (rx_desc->status & E1000_RXD_STAT_DD) {
3626                 struct sk_buff *skb;
3627                 u8 status;
3628
3629                 if (*work_done >= work_to_do)
3630                         break;
3631                 (*work_done)++;
3632
3633                 status = rx_desc->status;
3634                 skb = buffer_info->skb;
3635                 buffer_info->skb = NULL;
3636
3637                 if (++i == rx_ring->count) i = 0;
3638                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3639                 prefetch(next_rxd);
3640
3641                 next_buffer = &rx_ring->buffer_info[i];
3642
3643                 cleaned = true;
3644                 cleaned_count++;
3645                 pci_unmap_page(pdev, buffer_info->dma, buffer_info->length,
3646                                PCI_DMA_FROMDEVICE);
3647                 buffer_info->dma = 0;
3648
3649                 length = le16_to_cpu(rx_desc->length);
3650
3651                 /* errors is only valid for DD + EOP descriptors */
3652                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
3653                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
3654                         u8 last_byte = *(skb->data + length - 1);
3655                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3656                                        last_byte)) {
3657                                 spin_lock_irqsave(&adapter->stats_lock,
3658                                                   irq_flags);
3659                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3660                                                        length, skb->data);
3661                                 spin_unlock_irqrestore(&adapter->stats_lock,
3662                                                        irq_flags);
3663                                 length--;
3664                         } else {
3665                                 /* recycle both page and skb */
3666                                 buffer_info->skb = skb;
3667                                 /* an error means any chain goes out the window
3668                                  * too */
3669                                 if (rx_ring->rx_skb_top)
3670                                         dev_kfree_skb(rx_ring->rx_skb_top);
3671                                 rx_ring->rx_skb_top = NULL;
3672                                 goto next_desc;
3673                         }
3674                 }
3675
3676 #define rxtop rx_ring->rx_skb_top
3677                 if (!(status & E1000_RXD_STAT_EOP)) {
3678                         /* this descriptor is only the beginning (or middle) */
3679                         if (!rxtop) {
3680                                 /* this is the beginning of a chain */
3681                                 rxtop = skb;
3682                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
3683                                                    0, length);
3684                         } else {
3685                                 /* this is the middle of a chain */
3686                                 skb_fill_page_desc(rxtop,
3687                                     skb_shinfo(rxtop)->nr_frags,
3688                                     buffer_info->page, 0, length);
3689                                 /* re-use the skb, only consumed the page */
3690                                 buffer_info->skb = skb;
3691                         }
3692                         e1000_consume_page(buffer_info, rxtop, length);
3693                         goto next_desc;
3694                 } else {
3695                         if (rxtop) {
3696                                 /* end of the chain */
3697                                 skb_fill_page_desc(rxtop,
3698                                     skb_shinfo(rxtop)->nr_frags,
3699                                     buffer_info->page, 0, length);
3700                                 /* re-use the current skb, we only consumed the
3701                                  * page */
3702                                 buffer_info->skb = skb;
3703                                 skb = rxtop;
3704                                 rxtop = NULL;
3705                                 e1000_consume_page(buffer_info, skb, length);
3706                         } else {
3707                                 /* no chain, got EOP, this buf is the packet
3708                                  * copybreak to save the put_page/alloc_page */
3709                                 if (length <= copybreak &&
3710                                     skb_tailroom(skb) >= length) {
3711                                         u8 *vaddr;
3712                                         vaddr = kmap_atomic(buffer_info->page,
3713                                                             KM_SKB_DATA_SOFTIRQ);
3714                                         memcpy(skb_tail_pointer(skb), vaddr, length);
3715                                         kunmap_atomic(vaddr,
3716                                                       KM_SKB_DATA_SOFTIRQ);
3717                                         /* re-use the page, so don't erase
3718                                          * buffer_info->page */
3719                                         skb_put(skb, length);
3720                                 } else {
3721                                         skb_fill_page_desc(skb, 0,
3722                                                            buffer_info->page, 0,
3723                                                            length);
3724                                         e1000_consume_page(buffer_info, skb,
3725                                                            length);
3726                                 }
3727                         }
3728                 }
3729
3730                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
3731                 e1000_rx_checksum(adapter,
3732                                   (u32)(status) |
3733                                   ((u32)(rx_desc->errors) << 24),
3734                                   le16_to_cpu(rx_desc->csum), skb);
3735
3736                 pskb_trim(skb, skb->len - 4);
3737
3738                 /* probably a little skewed due to removing CRC */
3739                 total_rx_bytes += skb->len;
3740                 total_rx_packets++;
3741
3742                 /* eth type trans needs skb->data to point to something */
3743                 if (!pskb_may_pull(skb, ETH_HLEN)) {
3744                         DPRINTK(DRV, ERR, "pskb_may_pull failed.\n");
3745                         dev_kfree_skb(skb);
3746                         goto next_desc;
3747                 }
3748
3749                 skb->protocol = eth_type_trans(skb, netdev);
3750
3751                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3752
3753 next_desc:
3754                 rx_desc->status = 0;
3755
3756                 /* return some buffers to hardware, one at a time is too slow */
3757                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3758                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3759                         cleaned_count = 0;
3760                 }
3761
3762                 /* use prefetched values */
3763                 rx_desc = next_rxd;
3764                 buffer_info = next_buffer;
3765         }
3766         rx_ring->next_to_clean = i;
3767
3768         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3769         if (cleaned_count)
3770                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3771
3772         adapter->total_rx_packets += total_rx_packets;
3773         adapter->total_rx_bytes += total_rx_bytes;
3774         netdev->stats.rx_bytes += total_rx_bytes;
3775         netdev->stats.rx_packets += total_rx_packets;
3776         return cleaned;
3777 }
3778
3779 /**
3780  * e1000_clean_rx_irq - Send received data up the network stack; legacy
3781  * @adapter: board private structure
3782  * @rx_ring: ring to clean
3783  * @work_done: amount of napi work completed this call
3784  * @work_to_do: max amount of work allowed for this call to do
3785  */
3786 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
3787                                struct e1000_rx_ring *rx_ring,
3788                                int *work_done, int work_to_do)
3789 {
3790         struct e1000_hw *hw = &adapter->hw;
3791         struct net_device *netdev = adapter->netdev;
3792         struct pci_dev *pdev = adapter->pdev;
3793         struct e1000_rx_desc *rx_desc, *next_rxd;
3794         struct e1000_buffer *buffer_info, *next_buffer;
3795         unsigned long flags;
3796         u32 length;
3797         unsigned int i;
3798         int cleaned_count = 0;
3799         bool cleaned = false;
3800         unsigned int total_rx_bytes=0, total_rx_packets=0;
3801
3802         i = rx_ring->next_to_clean;
3803         rx_desc = E1000_RX_DESC(*rx_ring, i);
3804         buffer_info = &rx_ring->buffer_info[i];
3805
3806         while (rx_desc->status & E1000_RXD_STAT_DD) {
3807                 struct sk_buff *skb;
3808                 u8 status;
3809
3810                 if (*work_done >= work_to_do)
3811                         break;
3812                 (*work_done)++;
3813
3814                 status = rx_desc->status;
3815                 skb = buffer_info->skb;
3816                 buffer_info->skb = NULL;
3817
3818                 prefetch(skb->data - NET_IP_ALIGN);
3819
3820                 if (++i == rx_ring->count) i = 0;
3821                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3822                 prefetch(next_rxd);
3823
3824                 next_buffer = &rx_ring->buffer_info[i];
3825
3826                 cleaned = true;
3827                 cleaned_count++;
3828                 pci_unmap_single(pdev, buffer_info->dma, buffer_info->length,
3829                                  PCI_DMA_FROMDEVICE);
3830                 buffer_info->dma = 0;
3831
3832                 length = le16_to_cpu(rx_desc->length);
3833                 /* !EOP means multiple descriptors were used to store a single
3834                  * packet, if thats the case we need to toss it.  In fact, we
3835                  * to toss every packet with the EOP bit clear and the next
3836                  * frame that _does_ have the EOP bit set, as it is by
3837                  * definition only a frame fragment
3838                  */
3839                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
3840                         adapter->discarding = true;
3841
3842                 if (adapter->discarding) {
3843                         /* All receives must fit into a single buffer */
3844                         E1000_DBG("%s: Receive packet consumed multiple"
3845                                   " buffers\n", netdev->name);
3846                         /* recycle */
3847                         buffer_info->skb = skb;
3848                         if (status & E1000_RXD_STAT_EOP)
3849                                 adapter->discarding = false;
3850                         goto next_desc;
3851                 }
3852
3853                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
3854                         u8 last_byte = *(skb->data + length - 1);
3855                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3856                                        last_byte)) {
3857                                 spin_lock_irqsave(&adapter->stats_lock, flags);
3858                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3859                                                        length, skb->data);
3860                                 spin_unlock_irqrestore(&adapter->stats_lock,
3861                                                        flags);
3862                                 length--;
3863                         } else {
3864                                 /* recycle */
3865                                 buffer_info->skb = skb;
3866                                 goto next_desc;
3867                         }
3868                 }
3869
3870                 /* adjust length to remove Ethernet CRC, this must be
3871                  * done after the TBI_ACCEPT workaround above */
3872                 length -= 4;
3873
3874                 /* probably a little skewed due to removing CRC */
3875                 total_rx_bytes += length;
3876                 total_rx_packets++;
3877
3878                 /* code added for copybreak, this should improve
3879                  * performance for small packets with large amounts
3880                  * of reassembly being done in the stack */
3881                 if (length < copybreak) {
3882                         struct sk_buff *new_skb =
3883                             netdev_alloc_skb_ip_align(netdev, length);
3884                         if (new_skb) {
3885                                 skb_copy_to_linear_data_offset(new_skb,
3886                                                                -NET_IP_ALIGN,
3887                                                                (skb->data -
3888                                                                 NET_IP_ALIGN),
3889                                                                (length +
3890                                                                 NET_IP_ALIGN));
3891                                 /* save the skb in buffer_info as good */
3892                                 buffer_info->skb = skb;
3893                                 skb = new_skb;
3894                         }
3895                         /* else just continue with the old one */
3896                 }
3897                 /* end copybreak code */
3898                 skb_put(skb, length);
3899
3900                 /* Receive Checksum Offload */
3901                 e1000_rx_checksum(adapter,
3902                                   (u32)(status) |
3903                                   ((u32)(rx_desc->errors) << 24),
3904                                   le16_to_cpu(rx_desc->csum), skb);
3905
3906                 skb->protocol = eth_type_trans(skb, netdev);
3907
3908                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3909
3910 next_desc:
3911                 rx_desc->status = 0;
3912
3913                 /* return some buffers to hardware, one at a time is too slow */
3914                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3915                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3916                         cleaned_count = 0;
3917                 }
3918
3919                 /* use prefetched values */
3920                 rx_desc = next_rxd;
3921                 buffer_info = next_buffer;
3922         }
3923         rx_ring->next_to_clean = i;
3924
3925         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3926         if (cleaned_count)
3927                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3928
3929         adapter->total_rx_packets += total_rx_packets;
3930         adapter->total_rx_bytes += total_rx_bytes;
3931         netdev->stats.rx_bytes += total_rx_bytes;
3932         netdev->stats.rx_packets += total_rx_packets;
3933         return cleaned;
3934 }
3935
3936 /**
3937  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
3938  * @adapter: address of board private structure
3939  * @rx_ring: pointer to receive ring structure
3940  * @cleaned_count: number of buffers to allocate this pass
3941  **/
3942
3943 static void
3944 e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
3945                              struct e1000_rx_ring *rx_ring, int cleaned_count)
3946 {
3947         struct net_device *netdev = adapter->netdev;
3948         struct pci_dev *pdev = adapter->pdev;
3949         struct e1000_rx_desc *rx_desc;
3950         struct e1000_buffer *buffer_info;
3951         struct sk_buff *skb;
3952         unsigned int i;
3953         unsigned int bufsz = 256 - 16 /*for skb_reserve */ ;
3954
3955         i = rx_ring->next_to_use;
3956         buffer_info = &rx_ring->buffer_info[i];
3957
3958         while (cleaned_count--) {
3959                 skb = buffer_info->skb;
3960                 if (skb) {
3961                         skb_trim(skb, 0);
3962                         goto check_page;
3963                 }
3964
3965                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
3966                 if (unlikely(!skb)) {
3967                         /* Better luck next round */
3968                         adapter->alloc_rx_buff_failed++;
3969                         break;
3970                 }
3971
3972                 /* Fix for errata 23, can't cross 64kB boundary */
3973                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3974                         struct sk_buff *oldskb = skb;
3975                         DPRINTK(PROBE, ERR, "skb align check failed: %u bytes "
3976                                              "at %p\n", bufsz, skb->data);
3977                         /* Try again, without freeing the previous */
3978                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
3979                         /* Failed allocation, critical failure */
3980                         if (!skb) {
3981                                 dev_kfree_skb(oldskb);
3982                                 adapter->alloc_rx_buff_failed++;
3983                                 break;
3984                         }
3985
3986                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3987                                 /* give up */
3988                                 dev_kfree_skb(skb);
3989                                 dev_kfree_skb(oldskb);
3990                                 break; /* while (cleaned_count--) */
3991                         }
3992
3993                         /* Use new allocation */
3994                         dev_kfree_skb(oldskb);
3995                 }
3996                 buffer_info->skb = skb;
3997                 buffer_info->length = adapter->rx_buffer_len;
3998 check_page:
3999                 /* allocate a new page if necessary */
4000                 if (!buffer_info->page) {
4001                         buffer_info->page = alloc_page(GFP_ATOMIC);
4002                         if (unlikely(!buffer_info->page)) {
4003                                 adapter->alloc_rx_buff_failed++;
4004                                 break;
4005                         }
4006                 }
4007
4008                 if (!buffer_info->dma) {
4009                         buffer_info->dma = pci_map_page(pdev,
4010                                                         buffer_info->page, 0,
4011                                                         buffer_info->length,
4012                                                         PCI_DMA_FROMDEVICE);
4013                         if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
4014                                 put_page(buffer_info->page);
4015                                 dev_kfree_skb(skb);
4016                                 buffer_info->page = NULL;
4017                                 buffer_info->skb = NULL;
4018                                 buffer_info->dma = 0;
4019                                 adapter->alloc_rx_buff_failed++;
4020                                 break; /* while !buffer_info->skb */
4021                         }
4022                 }
4023
4024                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4025                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4026
4027                 if (unlikely(++i == rx_ring->count))
4028                         i = 0;
4029                 buffer_info = &rx_ring->buffer_info[i];
4030         }
4031
4032         if (likely(rx_ring->next_to_use != i)) {
4033                 rx_ring->next_to_use = i;
4034                 if (unlikely(i-- == 0))
4035                         i = (rx_ring->count - 1);
4036
4037                 /* Force memory writes to complete before letting h/w
4038                  * know there are new descriptors to fetch.  (Only
4039                  * applicable for weak-ordered memory model archs,
4040                  * such as IA-64). */
4041                 wmb();
4042                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4043         }
4044 }
4045
4046 /**
4047  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4048  * @adapter: address of board private structure
4049  **/
4050
4051 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4052                                    struct e1000_rx_ring *rx_ring,
4053                                    int cleaned_count)
4054 {
4055         struct e1000_hw *hw = &adapter->hw;
4056         struct net_device *netdev = adapter->netdev;
4057         struct pci_dev *pdev = adapter->pdev;
4058         struct e1000_rx_desc *rx_desc;
4059         struct e1000_buffer *buffer_info;
4060         struct sk_buff *skb;
4061         unsigned int i;
4062         unsigned int bufsz = adapter->rx_buffer_len;
4063
4064         i = rx_ring->next_to_use;
4065         buffer_info = &rx_ring->buffer_info[i];
4066
4067         while (cleaned_count--) {
4068                 skb = buffer_info->skb;
4069                 if (skb) {
4070                         skb_trim(skb, 0);
4071                         goto map_skb;
4072                 }
4073
4074                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4075                 if (unlikely(!skb)) {
4076                         /* Better luck next round */
4077                         adapter->alloc_rx_buff_failed++;
4078                         break;
4079                 }
4080
4081                 /* Fix for errata 23, can't cross 64kB boundary */
4082                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4083                         struct sk_buff *oldskb = skb;
4084                         DPRINTK(RX_ERR, ERR, "skb align check failed: %u bytes "
4085                                              "at %p\n", bufsz, skb->data);
4086                         /* Try again, without freeing the previous */
4087                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4088                         /* Failed allocation, critical failure */
4089                         if (!skb) {
4090                                 dev_kfree_skb(oldskb);
4091                                 adapter->alloc_rx_buff_failed++;
4092                                 break;
4093                         }
4094
4095                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4096                                 /* give up */
4097                                 dev_kfree_skb(skb);
4098                                 dev_kfree_skb(oldskb);
4099                                 adapter->alloc_rx_buff_failed++;
4100                                 break; /* while !buffer_info->skb */
4101                         }
4102
4103                         /* Use new allocation */
4104                         dev_kfree_skb(oldskb);
4105                 }
4106                 buffer_info->skb = skb;
4107                 buffer_info->length = adapter->rx_buffer_len;
4108 map_skb:
4109                 buffer_info->dma = pci_map_single(pdev,
4110                                                   skb->data,
4111                                                   buffer_info->length,
4112                                                   PCI_DMA_FROMDEVICE);
4113                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
4114                         dev_kfree_skb(skb);
4115                         buffer_info->skb = NULL;
4116                         buffer_info->dma = 0;
4117                         adapter->alloc_rx_buff_failed++;
4118                         break; /* while !buffer_info->skb */
4119                 }
4120
4121                 /*
4122                  * XXX if it was allocated cleanly it will never map to a
4123                  * boundary crossing
4124                  */
4125
4126                 /* Fix for errata 23, can't cross 64kB boundary */
4127                 if (!e1000_check_64k_bound(adapter,
4128                                         (void *)(unsigned long)buffer_info->dma,
4129                                         adapter->rx_buffer_len)) {
4130                         DPRINTK(RX_ERR, ERR,
4131                                 "dma align check failed: %u bytes at %p\n",
4132                                 adapter->rx_buffer_len,
4133                                 (void *)(unsigned long)buffer_info->dma);
4134                         dev_kfree_skb(skb);
4135                         buffer_info->skb = NULL;
4136
4137                         pci_unmap_single(pdev, buffer_info->dma,
4138                                          adapter->rx_buffer_len,
4139                                          PCI_DMA_FROMDEVICE);
4140                         buffer_info->dma = 0;
4141
4142                         adapter->alloc_rx_buff_failed++;
4143                         break; /* while !buffer_info->skb */
4144                 }
4145                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4146                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4147
4148                 if (unlikely(++i == rx_ring->count))
4149                         i = 0;
4150                 buffer_info = &rx_ring->buffer_info[i];
4151         }
4152
4153         if (likely(rx_ring->next_to_use != i)) {
4154                 rx_ring->next_to_use = i;
4155                 if (unlikely(i-- == 0))
4156                         i = (rx_ring->count - 1);
4157
4158                 /* Force memory writes to complete before letting h/w
4159                  * know there are new descriptors to fetch.  (Only
4160                  * applicable for weak-ordered memory model archs,
4161                  * such as IA-64). */
4162                 wmb();
4163                 writel(i, hw->hw_addr + rx_ring->rdt);
4164         }
4165 }
4166
4167 /**
4168  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4169  * @adapter:
4170  **/
4171
4172 static void e1000_smartspeed(struct e1000_adapter *adapter)
4173 {
4174         struct e1000_hw *hw = &adapter->hw;
4175         u16 phy_status;
4176         u16 phy_ctrl;
4177
4178         if ((hw->phy_type != e1000_phy_igp) || !hw->autoneg ||
4179            !(hw->autoneg_advertised & ADVERTISE_1000_FULL))
4180                 return;
4181
4182         if (adapter->smartspeed == 0) {
4183                 /* If Master/Slave config fault is asserted twice,
4184                  * we assume back-to-back */
4185                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4186                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4187                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4188                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4189                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4190                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4191                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4192                         e1000_write_phy_reg(hw, PHY_1000T_CTRL,
4193                                             phy_ctrl);
4194                         adapter->smartspeed++;
4195                         if (!e1000_phy_setup_autoneg(hw) &&
4196                            !e1000_read_phy_reg(hw, PHY_CTRL,
4197                                                &phy_ctrl)) {
4198                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4199                                              MII_CR_RESTART_AUTO_NEG);
4200                                 e1000_write_phy_reg(hw, PHY_CTRL,
4201                                                     phy_ctrl);
4202                         }
4203                 }
4204                 return;
4205         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4206                 /* If still no link, perhaps using 2/3 pair cable */
4207                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4208                 phy_ctrl |= CR_1000T_MS_ENABLE;
4209                 e1000_write_phy_reg(hw, PHY_1000T_CTRL, phy_ctrl);
4210                 if (!e1000_phy_setup_autoneg(hw) &&
4211                    !e1000_read_phy_reg(hw, PHY_CTRL, &phy_ctrl)) {
4212                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4213                                      MII_CR_RESTART_AUTO_NEG);
4214                         e1000_write_phy_reg(hw, PHY_CTRL, phy_ctrl);
4215                 }
4216         }
4217         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4218         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4219                 adapter->smartspeed = 0;
4220 }
4221
4222 /**
4223  * e1000_ioctl -
4224  * @netdev:
4225  * @ifreq:
4226  * @cmd:
4227  **/
4228
4229 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4230 {
4231         switch (cmd) {
4232         case SIOCGMIIPHY:
4233         case SIOCGMIIREG:
4234         case SIOCSMIIREG:
4235                 return e1000_mii_ioctl(netdev, ifr, cmd);
4236         default:
4237                 return -EOPNOTSUPP;
4238         }
4239 }
4240
4241 /**
4242  * e1000_mii_ioctl -
4243  * @netdev:
4244  * @ifreq:
4245  * @cmd:
4246  **/
4247
4248 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4249                            int cmd)
4250 {
4251         struct e1000_adapter *adapter = netdev_priv(netdev);
4252         struct e1000_hw *hw = &adapter->hw;
4253         struct mii_ioctl_data *data = if_mii(ifr);
4254         int retval;
4255         u16 mii_reg;
4256         u16 spddplx;
4257         unsigned long flags;
4258
4259         if (hw->media_type != e1000_media_type_copper)
4260                 return -EOPNOTSUPP;
4261
4262         switch (cmd) {
4263         case SIOCGMIIPHY:
4264                 data->phy_id = hw->phy_addr;
4265                 break;
4266         case SIOCGMIIREG:
4267                 spin_lock_irqsave(&adapter->stats_lock, flags);
4268                 if (e1000_read_phy_reg(hw, data->reg_num & 0x1F,
4269                                    &data->val_out)) {
4270                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4271                         return -EIO;
4272                 }
4273                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4274                 break;
4275         case SIOCSMIIREG:
4276                 if (data->reg_num & ~(0x1F))
4277                         return -EFAULT;
4278                 mii_reg = data->val_in;
4279                 spin_lock_irqsave(&adapter->stats_lock, flags);
4280                 if (e1000_write_phy_reg(hw, data->reg_num,
4281                                         mii_reg)) {
4282                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4283                         return -EIO;
4284                 }
4285                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4286                 if (hw->media_type == e1000_media_type_copper) {
4287                         switch (data->reg_num) {
4288                         case PHY_CTRL:
4289                                 if (mii_reg & MII_CR_POWER_DOWN)
4290                                         break;
4291                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4292                                         hw->autoneg = 1;
4293                                         hw->autoneg_advertised = 0x2F;
4294                                 } else {
4295                                         if (mii_reg & 0x40)
4296                                                 spddplx = SPEED_1000;
4297                                         else if (mii_reg & 0x2000)
4298                                                 spddplx = SPEED_100;
4299                                         else
4300                                                 spddplx = SPEED_10;
4301                                         spddplx += (mii_reg & 0x100)
4302                                                    ? DUPLEX_FULL :
4303                                                    DUPLEX_HALF;
4304                                         retval = e1000_set_spd_dplx(adapter,
4305                                                                     spddplx);
4306                                         if (retval)
4307                                                 return retval;
4308                                 }
4309                                 if (netif_running(adapter->netdev))
4310                                         e1000_reinit_locked(adapter);
4311                                 else
4312                                         e1000_reset(adapter);
4313                                 break;
4314                         case M88E1000_PHY_SPEC_CTRL:
4315                         case M88E1000_EXT_PHY_SPEC_CTRL:
4316                                 if (e1000_phy_reset(hw))
4317                                         return -EIO;
4318                                 break;
4319                         }
4320                 } else {
4321                         switch (data->reg_num) {
4322                         case PHY_CTRL:
4323                                 if (mii_reg & MII_CR_POWER_DOWN)
4324                                         break;
4325                                 if (netif_running(adapter->netdev))
4326                                         e1000_reinit_locked(adapter);
4327                                 else
4328                                         e1000_reset(adapter);
4329                                 break;
4330                         }
4331                 }
4332                 break;
4333         default:
4334                 return -EOPNOTSUPP;
4335         }
4336         return E1000_SUCCESS;
4337 }
4338
4339 void e1000_pci_set_mwi(struct e1000_hw *hw)
4340 {
4341         struct e1000_adapter *adapter = hw->back;
4342         int ret_val = pci_set_mwi(adapter->pdev);
4343
4344         if (ret_val)
4345                 DPRINTK(PROBE, ERR, "Error in setting MWI\n");
4346 }
4347
4348 void e1000_pci_clear_mwi(struct e1000_hw *hw)
4349 {
4350         struct e1000_adapter *adapter = hw->back;
4351
4352         pci_clear_mwi(adapter->pdev);
4353 }
4354
4355 int e1000_pcix_get_mmrbc(struct e1000_hw *hw)
4356 {
4357         struct e1000_adapter *adapter = hw->back;
4358         return pcix_get_mmrbc(adapter->pdev);
4359 }
4360
4361 void e1000_pcix_set_mmrbc(struct e1000_hw *hw, int mmrbc)
4362 {
4363         struct e1000_adapter *adapter = hw->back;
4364         pcix_set_mmrbc(adapter->pdev, mmrbc);
4365 }
4366
4367 void e1000_io_write(struct e1000_hw *hw, unsigned long port, u32 value)
4368 {
4369         outl(value, port);
4370 }
4371
4372 static void e1000_vlan_rx_register(struct net_device *netdev,
4373                                    struct vlan_group *grp)
4374 {
4375         struct e1000_adapter *adapter = netdev_priv(netdev);
4376         struct e1000_hw *hw = &adapter->hw;
4377         u32 ctrl, rctl;
4378
4379         if (!test_bit(__E1000_DOWN, &adapter->flags))
4380                 e1000_irq_disable(adapter);
4381         adapter->vlgrp = grp;
4382
4383         if (grp) {
4384                 /* enable VLAN tag insert/strip */
4385                 ctrl = er32(CTRL);
4386                 ctrl |= E1000_CTRL_VME;
4387                 ew32(CTRL, ctrl);
4388
4389                 /* enable VLAN receive filtering */
4390                 rctl = er32(RCTL);
4391                 rctl &= ~E1000_RCTL_CFIEN;
4392                 if (!(netdev->flags & IFF_PROMISC))
4393                         rctl |= E1000_RCTL_VFE;
4394                 ew32(RCTL, rctl);
4395                 e1000_update_mng_vlan(adapter);
4396         } else {
4397                 /* disable VLAN tag insert/strip */
4398                 ctrl = er32(CTRL);
4399                 ctrl &= ~E1000_CTRL_VME;
4400                 ew32(CTRL, ctrl);
4401
4402                 /* disable VLAN receive filtering */
4403                 rctl = er32(RCTL);
4404                 rctl &= ~E1000_RCTL_VFE;
4405                 ew32(RCTL, rctl);
4406
4407                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
4408                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4409                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4410                 }
4411         }
4412
4413         if (!test_bit(__E1000_DOWN, &adapter->flags))
4414                 e1000_irq_enable(adapter);
4415 }
4416
4417 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4418 {
4419         struct e1000_adapter *adapter = netdev_priv(netdev);
4420         struct e1000_hw *hw = &adapter->hw;
4421         u32 vfta, index;
4422
4423         if ((hw->mng_cookie.status &
4424              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4425             (vid == adapter->mng_vlan_id))
4426                 return;
4427         /* add VID to filter table */
4428         index = (vid >> 5) & 0x7F;
4429         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4430         vfta |= (1 << (vid & 0x1F));
4431         e1000_write_vfta(hw, index, vfta);
4432 }
4433
4434 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
4435 {
4436         struct e1000_adapter *adapter = netdev_priv(netdev);
4437         struct e1000_hw *hw = &adapter->hw;
4438         u32 vfta, index;
4439
4440         if (!test_bit(__E1000_DOWN, &adapter->flags))
4441                 e1000_irq_disable(adapter);
4442         vlan_group_set_device(adapter->vlgrp, vid, NULL);
4443         if (!test_bit(__E1000_DOWN, &adapter->flags))
4444                 e1000_irq_enable(adapter);
4445
4446         /* remove VID from filter table */
4447         index = (vid >> 5) & 0x7F;
4448         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4449         vfta &= ~(1 << (vid & 0x1F));
4450         e1000_write_vfta(hw, index, vfta);
4451 }
4452
4453 static void e1000_restore_vlan(struct e1000_adapter *adapter)
4454 {
4455         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
4456
4457         if (adapter->vlgrp) {
4458                 u16 vid;
4459                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
4460                         if (!vlan_group_get_device(adapter->vlgrp, vid))
4461                                 continue;
4462                         e1000_vlan_rx_add_vid(adapter->netdev, vid);
4463                 }
4464         }
4465 }
4466
4467 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx)
4468 {
4469         struct e1000_hw *hw = &adapter->hw;
4470
4471         hw->autoneg = 0;
4472
4473         /* Fiber NICs only allow 1000 gbps Full duplex */
4474         if ((hw->media_type == e1000_media_type_fiber) &&
4475                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
4476                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
4477                 return -EINVAL;
4478         }
4479
4480         switch (spddplx) {
4481         case SPEED_10 + DUPLEX_HALF:
4482                 hw->forced_speed_duplex = e1000_10_half;
4483                 break;
4484         case SPEED_10 + DUPLEX_FULL:
4485                 hw->forced_speed_duplex = e1000_10_full;
4486                 break;
4487         case SPEED_100 + DUPLEX_HALF:
4488                 hw->forced_speed_duplex = e1000_100_half;
4489                 break;
4490         case SPEED_100 + DUPLEX_FULL:
4491                 hw->forced_speed_duplex = e1000_100_full;
4492                 break;
4493         case SPEED_1000 + DUPLEX_FULL:
4494                 hw->autoneg = 1;
4495                 hw->autoneg_advertised = ADVERTISE_1000_FULL;
4496                 break;
4497         case SPEED_1000 + DUPLEX_HALF: /* not supported */
4498         default:
4499                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
4500                 return -EINVAL;
4501         }
4502         return 0;
4503 }
4504
4505 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4506 {
4507         struct net_device *netdev = pci_get_drvdata(pdev);
4508         struct e1000_adapter *adapter = netdev_priv(netdev);
4509         struct e1000_hw *hw = &adapter->hw;
4510         u32 ctrl, ctrl_ext, rctl, status;
4511         u32 wufc = adapter->wol;
4512 #ifdef CONFIG_PM
4513         int retval = 0;
4514 #endif
4515
4516         netif_device_detach(netdev);
4517
4518         if (netif_running(netdev)) {
4519                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
4520                 e1000_down(adapter);
4521         }
4522
4523 #ifdef CONFIG_PM
4524         retval = pci_save_state(pdev);
4525         if (retval)
4526                 return retval;
4527 #endif
4528
4529         status = er32(STATUS);
4530         if (status & E1000_STATUS_LU)
4531                 wufc &= ~E1000_WUFC_LNKC;
4532
4533         if (wufc) {
4534                 e1000_setup_rctl(adapter);
4535                 e1000_set_rx_mode(netdev);
4536
4537                 /* turn on all-multi mode if wake on multicast is enabled */
4538                 if (wufc & E1000_WUFC_MC) {
4539                         rctl = er32(RCTL);
4540                         rctl |= E1000_RCTL_MPE;
4541                         ew32(RCTL, rctl);
4542                 }
4543
4544                 if (hw->mac_type >= e1000_82540) {
4545                         ctrl = er32(CTRL);
4546                         /* advertise wake from D3Cold */
4547                         #define E1000_CTRL_ADVD3WUC 0x00100000
4548                         /* phy power management enable */
4549                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4550                         ctrl |= E1000_CTRL_ADVD3WUC |
4551                                 E1000_CTRL_EN_PHY_PWR_MGMT;
4552                         ew32(CTRL, ctrl);
4553                 }
4554
4555                 if (hw->media_type == e1000_media_type_fiber ||
4556                     hw->media_type == e1000_media_type_internal_serdes) {
4557                         /* keep the laser running in D3 */
4558                         ctrl_ext = er32(CTRL_EXT);
4559                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4560                         ew32(CTRL_EXT, ctrl_ext);
4561                 }
4562
4563                 ew32(WUC, E1000_WUC_PME_EN);
4564                 ew32(WUFC, wufc);
4565         } else {
4566                 ew32(WUC, 0);
4567                 ew32(WUFC, 0);
4568         }
4569
4570         e1000_release_manageability(adapter);
4571
4572         *enable_wake = !!wufc;
4573
4574         /* make sure adapter isn't asleep if manageability is enabled */
4575         if (adapter->en_mng_pt)
4576                 *enable_wake = true;
4577
4578         if (netif_running(netdev))
4579                 e1000_free_irq(adapter);
4580
4581         pci_disable_device(pdev);
4582
4583         return 0;
4584 }
4585
4586 #ifdef CONFIG_PM
4587 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4588 {
4589         int retval;
4590         bool wake;
4591
4592         retval = __e1000_shutdown(pdev, &wake);
4593         if (retval)
4594                 return retval;
4595
4596         if (wake) {
4597                 pci_prepare_to_sleep(pdev);
4598         } else {
4599                 pci_wake_from_d3(pdev, false);
4600                 pci_set_power_state(pdev, PCI_D3hot);
4601         }
4602
4603         return 0;
4604 }
4605
4606 static int e1000_resume(struct pci_dev *pdev)
4607 {
4608         struct net_device *netdev = pci_get_drvdata(pdev);
4609         struct e1000_adapter *adapter = netdev_priv(netdev);
4610         struct e1000_hw *hw = &adapter->hw;
4611         u32 err;
4612
4613         pci_set_power_state(pdev, PCI_D0);
4614         pci_restore_state(pdev);
4615         pci_save_state(pdev);
4616
4617         if (adapter->need_ioport)
4618                 err = pci_enable_device(pdev);
4619         else
4620                 err = pci_enable_device_mem(pdev);
4621         if (err) {
4622                 printk(KERN_ERR "e1000: Cannot enable PCI device from suspend\n");
4623                 return err;
4624         }
4625         pci_set_master(pdev);
4626
4627         pci_enable_wake(pdev, PCI_D3hot, 0);
4628         pci_enable_wake(pdev, PCI_D3cold, 0);
4629
4630         if (netif_running(netdev)) {
4631                 err = e1000_request_irq(adapter);
4632                 if (err)
4633                         return err;
4634         }
4635
4636         e1000_power_up_phy(adapter);
4637         e1000_reset(adapter);
4638         ew32(WUS, ~0);
4639
4640         e1000_init_manageability(adapter);
4641
4642         if (netif_running(netdev))
4643                 e1000_up(adapter);
4644
4645         netif_device_attach(netdev);
4646
4647         return 0;
4648 }
4649 #endif
4650
4651 static void e1000_shutdown(struct pci_dev *pdev)
4652 {
4653         bool wake;
4654
4655         __e1000_shutdown(pdev, &wake);
4656
4657         if (system_state == SYSTEM_POWER_OFF) {
4658                 pci_wake_from_d3(pdev, wake);
4659                 pci_set_power_state(pdev, PCI_D3hot);
4660         }
4661 }
4662
4663 #ifdef CONFIG_NET_POLL_CONTROLLER
4664 /*
4665  * Polling 'interrupt' - used by things like netconsole to send skbs
4666  * without having to re-enable interrupts. It's not called while
4667  * the interrupt routine is executing.
4668  */
4669 static void e1000_netpoll(struct net_device *netdev)
4670 {
4671         struct e1000_adapter *adapter = netdev_priv(netdev);
4672
4673         disable_irq(adapter->pdev->irq);
4674         e1000_intr(adapter->pdev->irq, netdev);
4675         enable_irq(adapter->pdev->irq);
4676 }
4677 #endif
4678
4679 /**
4680  * e1000_io_error_detected - called when PCI error is detected
4681  * @pdev: Pointer to PCI device
4682  * @state: The current pci connection state
4683  *
4684  * This function is called after a PCI bus error affecting
4685  * this device has been detected.
4686  */
4687 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4688                                                 pci_channel_state_t state)
4689 {
4690         struct net_device *netdev = pci_get_drvdata(pdev);
4691         struct e1000_adapter *adapter = netdev_priv(netdev);
4692
4693         netif_device_detach(netdev);
4694
4695         if (state == pci_channel_io_perm_failure)
4696                 return PCI_ERS_RESULT_DISCONNECT;
4697
4698         if (netif_running(netdev))
4699                 e1000_down(adapter);
4700         pci_disable_device(pdev);
4701
4702         /* Request a slot slot reset. */
4703         return PCI_ERS_RESULT_NEED_RESET;
4704 }
4705
4706 /**
4707  * e1000_io_slot_reset - called after the pci bus has been reset.
4708  * @pdev: Pointer to PCI device
4709  *
4710  * Restart the card from scratch, as if from a cold-boot. Implementation
4711  * resembles the first-half of the e1000_resume routine.
4712  */
4713 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4714 {
4715         struct net_device *netdev = pci_get_drvdata(pdev);
4716         struct e1000_adapter *adapter = netdev_priv(netdev);
4717         struct e1000_hw *hw = &adapter->hw;
4718         int err;
4719
4720         if (adapter->need_ioport)
4721                 err = pci_enable_device(pdev);
4722         else
4723                 err = pci_enable_device_mem(pdev);
4724         if (err) {
4725                 printk(KERN_ERR "e1000: Cannot re-enable PCI device after reset.\n");
4726                 return PCI_ERS_RESULT_DISCONNECT;
4727         }
4728         pci_set_master(pdev);
4729
4730         pci_enable_wake(pdev, PCI_D3hot, 0);
4731         pci_enable_wake(pdev, PCI_D3cold, 0);
4732
4733         e1000_reset(adapter);
4734         ew32(WUS, ~0);
4735
4736         return PCI_ERS_RESULT_RECOVERED;
4737 }
4738
4739 /**
4740  * e1000_io_resume - called when traffic can start flowing again.
4741  * @pdev: Pointer to PCI device
4742  *
4743  * This callback is called when the error recovery driver tells us that
4744  * its OK to resume normal operation. Implementation resembles the
4745  * second-half of the e1000_resume routine.
4746  */
4747 static void e1000_io_resume(struct pci_dev *pdev)
4748 {
4749         struct net_device *netdev = pci_get_drvdata(pdev);
4750         struct e1000_adapter *adapter = netdev_priv(netdev);
4751
4752         e1000_init_manageability(adapter);
4753
4754         if (netif_running(netdev)) {
4755                 if (e1000_up(adapter)) {
4756                         printk("e1000: can't bring device back up after reset\n");
4757                         return;
4758                 }
4759         }
4760
4761         netif_device_attach(netdev);
4762 }
4763
4764 /* e1000_main.c */