]> bbs.cooldavid.org Git - net-next-2.6.git/blob - drivers/net/e1000/e1000_main.c
5cc39ed289c62234d2f56cd9ad4c1eb841c26b8e
[net-next-2.6.git] / drivers / net / e1000 / e1000_main.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2006 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include "e1000.h"
30 #include <net/ip6_checksum.h>
31
32 char e1000_driver_name[] = "e1000";
33 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
34 #define DRV_VERSION "7.3.21-k6-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_get_hw_dev - return device
218  * used by hardware layer to print debugging information
219  *
220  **/
221 struct net_device *e1000_get_hw_dev(struct e1000_hw *hw)
222 {
223         struct e1000_adapter *adapter = hw->back;
224         return adapter->netdev;
225 }
226
227 /**
228  * e1000_init_module - Driver Registration Routine
229  *
230  * e1000_init_module is the first routine called when the driver is
231  * loaded. All it does is register with the PCI subsystem.
232  **/
233
234 static int __init e1000_init_module(void)
235 {
236         int ret;
237         pr_info("%s - version %s\n", e1000_driver_string, e1000_driver_version);
238
239         pr_info("%s\n", e1000_copyright);
240
241         ret = pci_register_driver(&e1000_driver);
242         if (copybreak != COPYBREAK_DEFAULT) {
243                 if (copybreak == 0)
244                         pr_info("copybreak disabled\n");
245                 else
246                         pr_info("copybreak enabled for "
247                                    "packets <= %u bytes\n", copybreak);
248         }
249         return ret;
250 }
251
252 module_init(e1000_init_module);
253
254 /**
255  * e1000_exit_module - Driver Exit Cleanup Routine
256  *
257  * e1000_exit_module is called just before the driver is removed
258  * from memory.
259  **/
260
261 static void __exit e1000_exit_module(void)
262 {
263         pci_unregister_driver(&e1000_driver);
264 }
265
266 module_exit(e1000_exit_module);
267
268 static int e1000_request_irq(struct e1000_adapter *adapter)
269 {
270         struct net_device *netdev = adapter->netdev;
271         irq_handler_t handler = e1000_intr;
272         int irq_flags = IRQF_SHARED;
273         int err;
274
275         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
276                           netdev);
277         if (err) {
278                 e_err(probe, "Unable to allocate interrupt Error: %d\n", err);
279         }
280
281         return err;
282 }
283
284 static void e1000_free_irq(struct e1000_adapter *adapter)
285 {
286         struct net_device *netdev = adapter->netdev;
287
288         free_irq(adapter->pdev->irq, netdev);
289 }
290
291 /**
292  * e1000_irq_disable - Mask off interrupt generation on the NIC
293  * @adapter: board private structure
294  **/
295
296 static void e1000_irq_disable(struct e1000_adapter *adapter)
297 {
298         struct e1000_hw *hw = &adapter->hw;
299
300         ew32(IMC, ~0);
301         E1000_WRITE_FLUSH();
302         synchronize_irq(adapter->pdev->irq);
303 }
304
305 /**
306  * e1000_irq_enable - Enable default interrupt generation settings
307  * @adapter: board private structure
308  **/
309
310 static void e1000_irq_enable(struct e1000_adapter *adapter)
311 {
312         struct e1000_hw *hw = &adapter->hw;
313
314         ew32(IMS, IMS_ENABLE_MASK);
315         E1000_WRITE_FLUSH();
316 }
317
318 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
319 {
320         struct e1000_hw *hw = &adapter->hw;
321         struct net_device *netdev = adapter->netdev;
322         u16 vid = hw->mng_cookie.vlan_id;
323         u16 old_vid = adapter->mng_vlan_id;
324         if (adapter->vlgrp) {
325                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
326                         if (hw->mng_cookie.status &
327                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
328                                 e1000_vlan_rx_add_vid(netdev, vid);
329                                 adapter->mng_vlan_id = vid;
330                         } else
331                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
332
333                         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
334                                         (vid != old_vid) &&
335                             !vlan_group_get_device(adapter->vlgrp, old_vid))
336                                 e1000_vlan_rx_kill_vid(netdev, old_vid);
337                 } else
338                         adapter->mng_vlan_id = vid;
339         }
340 }
341
342 static void e1000_init_manageability(struct e1000_adapter *adapter)
343 {
344         struct e1000_hw *hw = &adapter->hw;
345
346         if (adapter->en_mng_pt) {
347                 u32 manc = er32(MANC);
348
349                 /* disable hardware interception of ARP */
350                 manc &= ~(E1000_MANC_ARP_EN);
351
352                 ew32(MANC, manc);
353         }
354 }
355
356 static void e1000_release_manageability(struct e1000_adapter *adapter)
357 {
358         struct e1000_hw *hw = &adapter->hw;
359
360         if (adapter->en_mng_pt) {
361                 u32 manc = er32(MANC);
362
363                 /* re-enable hardware interception of ARP */
364                 manc |= E1000_MANC_ARP_EN;
365
366                 ew32(MANC, manc);
367         }
368 }
369
370 /**
371  * e1000_configure - configure the hardware for RX and TX
372  * @adapter = private board structure
373  **/
374 static void e1000_configure(struct e1000_adapter *adapter)
375 {
376         struct net_device *netdev = adapter->netdev;
377         int i;
378
379         e1000_set_rx_mode(netdev);
380
381         e1000_restore_vlan(adapter);
382         e1000_init_manageability(adapter);
383
384         e1000_configure_tx(adapter);
385         e1000_setup_rctl(adapter);
386         e1000_configure_rx(adapter);
387         /* call E1000_DESC_UNUSED which always leaves
388          * at least 1 descriptor unused to make sure
389          * next_to_use != next_to_clean */
390         for (i = 0; i < adapter->num_rx_queues; i++) {
391                 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
392                 adapter->alloc_rx_buf(adapter, ring,
393                                       E1000_DESC_UNUSED(ring));
394         }
395 }
396
397 int e1000_up(struct e1000_adapter *adapter)
398 {
399         struct e1000_hw *hw = &adapter->hw;
400
401         /* hardware has been reset, we need to reload some things */
402         e1000_configure(adapter);
403
404         clear_bit(__E1000_DOWN, &adapter->flags);
405
406         napi_enable(&adapter->napi);
407
408         e1000_irq_enable(adapter);
409
410         netif_wake_queue(adapter->netdev);
411
412         /* fire a link change interrupt to start the watchdog */
413         ew32(ICS, E1000_ICS_LSC);
414         return 0;
415 }
416
417 /**
418  * e1000_power_up_phy - restore link in case the phy was powered down
419  * @adapter: address of board private structure
420  *
421  * The phy may be powered down to save power and turn off link when the
422  * driver is unloaded and wake on lan is not enabled (among others)
423  * *** this routine MUST be followed by a call to e1000_reset ***
424  *
425  **/
426
427 void e1000_power_up_phy(struct e1000_adapter *adapter)
428 {
429         struct e1000_hw *hw = &adapter->hw;
430         u16 mii_reg = 0;
431
432         /* Just clear the power down bit to wake the phy back up */
433         if (hw->media_type == e1000_media_type_copper) {
434                 /* according to the manual, the phy will retain its
435                  * settings across a power-down/up cycle */
436                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
437                 mii_reg &= ~MII_CR_POWER_DOWN;
438                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
439         }
440 }
441
442 static void e1000_power_down_phy(struct e1000_adapter *adapter)
443 {
444         struct e1000_hw *hw = &adapter->hw;
445
446         /* Power down the PHY so no link is implied when interface is down *
447          * The PHY cannot be powered down if any of the following is true *
448          * (a) WoL is enabled
449          * (b) AMT is active
450          * (c) SoL/IDER session is active */
451         if (!adapter->wol && hw->mac_type >= e1000_82540 &&
452            hw->media_type == e1000_media_type_copper) {
453                 u16 mii_reg = 0;
454
455                 switch (hw->mac_type) {
456                 case e1000_82540:
457                 case e1000_82545:
458                 case e1000_82545_rev_3:
459                 case e1000_82546:
460                 case e1000_82546_rev_3:
461                 case e1000_82541:
462                 case e1000_82541_rev_2:
463                 case e1000_82547:
464                 case e1000_82547_rev_2:
465                         if (er32(MANC) & E1000_MANC_SMBUS_EN)
466                                 goto out;
467                         break;
468                 default:
469                         goto out;
470                 }
471                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
472                 mii_reg |= MII_CR_POWER_DOWN;
473                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
474                 mdelay(1);
475         }
476 out:
477         return;
478 }
479
480 void e1000_down(struct e1000_adapter *adapter)
481 {
482         struct e1000_hw *hw = &adapter->hw;
483         struct net_device *netdev = adapter->netdev;
484         u32 rctl, tctl;
485
486         /* signal that we're down so the interrupt handler does not
487          * reschedule our watchdog timer */
488         set_bit(__E1000_DOWN, &adapter->flags);
489
490         /* disable receives in the hardware */
491         rctl = er32(RCTL);
492         ew32(RCTL, rctl & ~E1000_RCTL_EN);
493         /* flush and sleep below */
494
495         netif_tx_disable(netdev);
496
497         /* disable transmits in the hardware */
498         tctl = er32(TCTL);
499         tctl &= ~E1000_TCTL_EN;
500         ew32(TCTL, tctl);
501         /* flush both disables and wait for them to finish */
502         E1000_WRITE_FLUSH();
503         msleep(10);
504
505         napi_disable(&adapter->napi);
506
507         e1000_irq_disable(adapter);
508
509         del_timer_sync(&adapter->tx_fifo_stall_timer);
510         del_timer_sync(&adapter->watchdog_timer);
511         del_timer_sync(&adapter->phy_info_timer);
512
513         adapter->link_speed = 0;
514         adapter->link_duplex = 0;
515         netif_carrier_off(netdev);
516
517         e1000_reset(adapter);
518         e1000_clean_all_tx_rings(adapter);
519         e1000_clean_all_rx_rings(adapter);
520 }
521
522 void e1000_reinit_locked(struct e1000_adapter *adapter)
523 {
524         WARN_ON(in_interrupt());
525         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
526                 msleep(1);
527         e1000_down(adapter);
528         e1000_up(adapter);
529         clear_bit(__E1000_RESETTING, &adapter->flags);
530 }
531
532 void e1000_reset(struct e1000_adapter *adapter)
533 {
534         struct e1000_hw *hw = &adapter->hw;
535         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
536         bool legacy_pba_adjust = false;
537         u16 hwm;
538
539         /* Repartition Pba for greater than 9k mtu
540          * To take effect CTRL.RST is required.
541          */
542
543         switch (hw->mac_type) {
544         case e1000_82542_rev2_0:
545         case e1000_82542_rev2_1:
546         case e1000_82543:
547         case e1000_82544:
548         case e1000_82540:
549         case e1000_82541:
550         case e1000_82541_rev_2:
551                 legacy_pba_adjust = true;
552                 pba = E1000_PBA_48K;
553                 break;
554         case e1000_82545:
555         case e1000_82545_rev_3:
556         case e1000_82546:
557         case e1000_82546_rev_3:
558                 pba = E1000_PBA_48K;
559                 break;
560         case e1000_82547:
561         case e1000_82547_rev_2:
562                 legacy_pba_adjust = true;
563                 pba = E1000_PBA_30K;
564                 break;
565         case e1000_undefined:
566         case e1000_num_macs:
567                 break;
568         }
569
570         if (legacy_pba_adjust) {
571                 if (hw->max_frame_size > E1000_RXBUFFER_8192)
572                         pba -= 8; /* allocate more FIFO for Tx */
573
574                 if (hw->mac_type == e1000_82547) {
575                         adapter->tx_fifo_head = 0;
576                         adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
577                         adapter->tx_fifo_size =
578                                 (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
579                         atomic_set(&adapter->tx_fifo_stall, 0);
580                 }
581         } else if (hw->max_frame_size >  ETH_FRAME_LEN + ETH_FCS_LEN) {
582                 /* adjust PBA for jumbo frames */
583                 ew32(PBA, pba);
584
585                 /* To maintain wire speed transmits, the Tx FIFO should be
586                  * large enough to accommodate two full transmit packets,
587                  * rounded up to the next 1KB and expressed in KB.  Likewise,
588                  * the Rx FIFO should be large enough to accommodate at least
589                  * one full receive packet and is similarly rounded up and
590                  * expressed in KB. */
591                 pba = er32(PBA);
592                 /* upper 16 bits has Tx packet buffer allocation size in KB */
593                 tx_space = pba >> 16;
594                 /* lower 16 bits has Rx packet buffer allocation size in KB */
595                 pba &= 0xffff;
596                 /*
597                  * the tx fifo also stores 16 bytes of information about the tx
598                  * but don't include ethernet FCS because hardware appends it
599                  */
600                 min_tx_space = (hw->max_frame_size +
601                                 sizeof(struct e1000_tx_desc) -
602                                 ETH_FCS_LEN) * 2;
603                 min_tx_space = ALIGN(min_tx_space, 1024);
604                 min_tx_space >>= 10;
605                 /* software strips receive CRC, so leave room for it */
606                 min_rx_space = hw->max_frame_size;
607                 min_rx_space = ALIGN(min_rx_space, 1024);
608                 min_rx_space >>= 10;
609
610                 /* If current Tx allocation is less than the min Tx FIFO size,
611                  * and the min Tx FIFO size is less than the current Rx FIFO
612                  * allocation, take space away from current Rx allocation */
613                 if (tx_space < min_tx_space &&
614                     ((min_tx_space - tx_space) < pba)) {
615                         pba = pba - (min_tx_space - tx_space);
616
617                         /* PCI/PCIx hardware has PBA alignment constraints */
618                         switch (hw->mac_type) {
619                         case e1000_82545 ... e1000_82546_rev_3:
620                                 pba &= ~(E1000_PBA_8K - 1);
621                                 break;
622                         default:
623                                 break;
624                         }
625
626                         /* if short on rx space, rx wins and must trump tx
627                          * adjustment or use Early Receive if available */
628                         if (pba < min_rx_space)
629                                 pba = min_rx_space;
630                 }
631         }
632
633         ew32(PBA, pba);
634
635         /*
636          * flow control settings:
637          * The high water mark must be low enough to fit one full frame
638          * (or the size used for early receive) above it in the Rx FIFO.
639          * Set it to the lower of:
640          * - 90% of the Rx FIFO size, and
641          * - the full Rx FIFO size minus the early receive size (for parts
642          *   with ERT support assuming ERT set to E1000_ERT_2048), or
643          * - the full Rx FIFO size minus one full frame
644          */
645         hwm = min(((pba << 10) * 9 / 10),
646                   ((pba << 10) - hw->max_frame_size));
647
648         hw->fc_high_water = hwm & 0xFFF8;       /* 8-byte granularity */
649         hw->fc_low_water = hw->fc_high_water - 8;
650         hw->fc_pause_time = E1000_FC_PAUSE_TIME;
651         hw->fc_send_xon = 1;
652         hw->fc = hw->original_fc;
653
654         /* Allow time for pending master requests to run */
655         e1000_reset_hw(hw);
656         if (hw->mac_type >= e1000_82544)
657                 ew32(WUC, 0);
658
659         if (e1000_init_hw(hw))
660                 e_dev_err("Hardware Error\n");
661         e1000_update_mng_vlan(adapter);
662
663         /* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
664         if (hw->mac_type >= e1000_82544 &&
665             hw->autoneg == 1 &&
666             hw->autoneg_advertised == ADVERTISE_1000_FULL) {
667                 u32 ctrl = er32(CTRL);
668                 /* clear phy power management bit if we are in gig only mode,
669                  * which if enabled will attempt negotiation to 100Mb, which
670                  * can cause a loss of link at power off or driver unload */
671                 ctrl &= ~E1000_CTRL_SWDPIN3;
672                 ew32(CTRL, ctrl);
673         }
674
675         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
676         ew32(VET, ETHERNET_IEEE_VLAN_TYPE);
677
678         e1000_reset_adaptive(hw);
679         e1000_phy_get_info(hw, &adapter->phy_info);
680
681         e1000_release_manageability(adapter);
682 }
683
684 /**
685  *  Dump the eeprom for users having checksum issues
686  **/
687 static void e1000_dump_eeprom(struct e1000_adapter *adapter)
688 {
689         struct net_device *netdev = adapter->netdev;
690         struct ethtool_eeprom eeprom;
691         const struct ethtool_ops *ops = netdev->ethtool_ops;
692         u8 *data;
693         int i;
694         u16 csum_old, csum_new = 0;
695
696         eeprom.len = ops->get_eeprom_len(netdev);
697         eeprom.offset = 0;
698
699         data = kmalloc(eeprom.len, GFP_KERNEL);
700         if (!data) {
701                 pr_err("Unable to allocate memory to dump EEPROM data\n");
702                 return;
703         }
704
705         ops->get_eeprom(netdev, &eeprom, data);
706
707         csum_old = (data[EEPROM_CHECKSUM_REG * 2]) +
708                    (data[EEPROM_CHECKSUM_REG * 2 + 1] << 8);
709         for (i = 0; i < EEPROM_CHECKSUM_REG * 2; i += 2)
710                 csum_new += data[i] + (data[i + 1] << 8);
711         csum_new = EEPROM_SUM - csum_new;
712
713         pr_err("/*********************/\n");
714         pr_err("Current EEPROM Checksum : 0x%04x\n", csum_old);
715         pr_err("Calculated              : 0x%04x\n", csum_new);
716
717         pr_err("Offset    Values\n");
718         pr_err("========  ======\n");
719         print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 16, 1, data, 128, 0);
720
721         pr_err("Include this output when contacting your support provider.\n");
722         pr_err("This is not a software error! Something bad happened to\n");
723         pr_err("your hardware or EEPROM image. Ignoring this problem could\n");
724         pr_err("result in further problems, possibly loss of data,\n");
725         pr_err("corruption or system hangs!\n");
726         pr_err("The MAC Address will be reset to 00:00:00:00:00:00,\n");
727         pr_err("which is invalid and requires you to set the proper MAC\n");
728         pr_err("address manually before continuing to enable this network\n");
729         pr_err("device. Please inspect the EEPROM dump and report the\n");
730         pr_err("issue to your hardware vendor or Intel Customer Support.\n");
731         pr_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 (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
830             !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
831                 pci_using_dac = 1;
832         } else {
833                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
834                 if (err) {
835                         err = dma_set_coherent_mask(&pdev->dev,
836                                                     DMA_BIT_MASK(32));
837                         if (err) {
838                                 pr_err("No usable DMA config, 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                 e_err(probe, "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                 e_err(probe, "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                         e_err(probe, "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                 e_err(probe, "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         /* reset the hardware with the new settings */
1042         e1000_reset(adapter);
1043
1044         strcpy(netdev->name, "eth%d");
1045         err = register_netdev(netdev);
1046         if (err)
1047                 goto err_register;
1048
1049         /* print bus type/speed/width info */
1050         e_info(probe, "(PCI%s:%dMHz:%d-bit) %pM\n",
1051                ((hw->bus_type == e1000_bus_type_pcix) ? "-X" : ""),
1052                ((hw->bus_speed == e1000_bus_speed_133) ? 133 :
1053                 (hw->bus_speed == e1000_bus_speed_120) ? 120 :
1054                 (hw->bus_speed == e1000_bus_speed_100) ? 100 :
1055                 (hw->bus_speed == e1000_bus_speed_66) ? 66 : 33),
1056                ((hw->bus_width == e1000_bus_width_64) ? 64 : 32),
1057                netdev->dev_addr);
1058
1059         /* carrier off reporting is important to ethtool even BEFORE open */
1060         netif_carrier_off(netdev);
1061
1062         e_info(probe, "Intel(R) PRO/1000 Network Connection\n");
1063
1064         cards_found++;
1065         return 0;
1066
1067 err_register:
1068 err_eeprom:
1069         e1000_phy_hw_reset(hw);
1070
1071         if (hw->flash_address)
1072                 iounmap(hw->flash_address);
1073         kfree(adapter->tx_ring);
1074         kfree(adapter->rx_ring);
1075 err_sw_init:
1076         iounmap(hw->hw_addr);
1077 err_ioremap:
1078         free_netdev(netdev);
1079 err_alloc_etherdev:
1080         pci_release_selected_regions(pdev, bars);
1081 err_pci_reg:
1082 err_dma:
1083         pci_disable_device(pdev);
1084         return err;
1085 }
1086
1087 /**
1088  * e1000_remove - Device Removal Routine
1089  * @pdev: PCI device information struct
1090  *
1091  * e1000_remove is called by the PCI subsystem to alert the driver
1092  * that it should release a PCI device.  The could be caused by a
1093  * Hot-Plug event, or because the driver is going to be removed from
1094  * memory.
1095  **/
1096
1097 static void __devexit e1000_remove(struct pci_dev *pdev)
1098 {
1099         struct net_device *netdev = pci_get_drvdata(pdev);
1100         struct e1000_adapter *adapter = netdev_priv(netdev);
1101         struct e1000_hw *hw = &adapter->hw;
1102
1103         set_bit(__E1000_DOWN, &adapter->flags);
1104         del_timer_sync(&adapter->tx_fifo_stall_timer);
1105         del_timer_sync(&adapter->watchdog_timer);
1106         del_timer_sync(&adapter->phy_info_timer);
1107
1108         cancel_work_sync(&adapter->reset_task);
1109
1110         e1000_release_manageability(adapter);
1111
1112         unregister_netdev(netdev);
1113
1114         e1000_phy_hw_reset(hw);
1115
1116         kfree(adapter->tx_ring);
1117         kfree(adapter->rx_ring);
1118
1119         iounmap(hw->hw_addr);
1120         if (hw->flash_address)
1121                 iounmap(hw->flash_address);
1122         pci_release_selected_regions(pdev, adapter->bars);
1123
1124         free_netdev(netdev);
1125
1126         pci_disable_device(pdev);
1127 }
1128
1129 /**
1130  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1131  * @adapter: board private structure to initialize
1132  *
1133  * e1000_sw_init initializes the Adapter private data structure.
1134  * Fields are initialized based on PCI device information and
1135  * OS network device settings (MTU size).
1136  **/
1137
1138 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
1139 {
1140         struct e1000_hw *hw = &adapter->hw;
1141         struct net_device *netdev = adapter->netdev;
1142         struct pci_dev *pdev = adapter->pdev;
1143
1144         /* PCI config space info */
1145
1146         hw->vendor_id = pdev->vendor;
1147         hw->device_id = pdev->device;
1148         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1149         hw->subsystem_id = pdev->subsystem_device;
1150         hw->revision_id = pdev->revision;
1151
1152         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
1153
1154         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1155         hw->max_frame_size = netdev->mtu +
1156                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
1157         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
1158
1159         /* identify the MAC */
1160
1161         if (e1000_set_mac_type(hw)) {
1162                 e_err(probe, "Unknown MAC Type\n");
1163                 return -EIO;
1164         }
1165
1166         switch (hw->mac_type) {
1167         default:
1168                 break;
1169         case e1000_82541:
1170         case e1000_82547:
1171         case e1000_82541_rev_2:
1172         case e1000_82547_rev_2:
1173                 hw->phy_init_script = 1;
1174                 break;
1175         }
1176
1177         e1000_set_media_type(hw);
1178
1179         hw->wait_autoneg_complete = false;
1180         hw->tbi_compatibility_en = true;
1181         hw->adaptive_ifs = true;
1182
1183         /* Copper options */
1184
1185         if (hw->media_type == e1000_media_type_copper) {
1186                 hw->mdix = AUTO_ALL_MODES;
1187                 hw->disable_polarity_correction = false;
1188                 hw->master_slave = E1000_MASTER_SLAVE;
1189         }
1190
1191         adapter->num_tx_queues = 1;
1192         adapter->num_rx_queues = 1;
1193
1194         if (e1000_alloc_queues(adapter)) {
1195                 e_err(probe, "Unable to allocate memory for queues\n");
1196                 return -ENOMEM;
1197         }
1198
1199         /* Explicitly disable IRQ since the NIC can be in any state. */
1200         e1000_irq_disable(adapter);
1201
1202         spin_lock_init(&adapter->stats_lock);
1203
1204         set_bit(__E1000_DOWN, &adapter->flags);
1205
1206         return 0;
1207 }
1208
1209 /**
1210  * e1000_alloc_queues - Allocate memory for all rings
1211  * @adapter: board private structure to initialize
1212  *
1213  * We allocate one ring per queue at run-time since we don't know the
1214  * number of queues at compile-time.
1215  **/
1216
1217 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1218 {
1219         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1220                                    sizeof(struct e1000_tx_ring), GFP_KERNEL);
1221         if (!adapter->tx_ring)
1222                 return -ENOMEM;
1223
1224         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1225                                    sizeof(struct e1000_rx_ring), GFP_KERNEL);
1226         if (!adapter->rx_ring) {
1227                 kfree(adapter->tx_ring);
1228                 return -ENOMEM;
1229         }
1230
1231         return E1000_SUCCESS;
1232 }
1233
1234 /**
1235  * e1000_open - Called when a network interface is made active
1236  * @netdev: network interface device structure
1237  *
1238  * Returns 0 on success, negative value on failure
1239  *
1240  * The open entry point is called when a network interface is made
1241  * active by the system (IFF_UP).  At this point all resources needed
1242  * for transmit and receive operations are allocated, the interrupt
1243  * handler is registered with the OS, the watchdog timer is started,
1244  * and the stack is notified that the interface is ready.
1245  **/
1246
1247 static int e1000_open(struct net_device *netdev)
1248 {
1249         struct e1000_adapter *adapter = netdev_priv(netdev);
1250         struct e1000_hw *hw = &adapter->hw;
1251         int err;
1252
1253         /* disallow open during test */
1254         if (test_bit(__E1000_TESTING, &adapter->flags))
1255                 return -EBUSY;
1256
1257         netif_carrier_off(netdev);
1258
1259         /* allocate transmit descriptors */
1260         err = e1000_setup_all_tx_resources(adapter);
1261         if (err)
1262                 goto err_setup_tx;
1263
1264         /* allocate receive descriptors */
1265         err = e1000_setup_all_rx_resources(adapter);
1266         if (err)
1267                 goto err_setup_rx;
1268
1269         e1000_power_up_phy(adapter);
1270
1271         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1272         if ((hw->mng_cookie.status &
1273                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1274                 e1000_update_mng_vlan(adapter);
1275         }
1276
1277         /* before we allocate an interrupt, we must be ready to handle it.
1278          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1279          * as soon as we call pci_request_irq, so we have to setup our
1280          * clean_rx handler before we do so.  */
1281         e1000_configure(adapter);
1282
1283         err = e1000_request_irq(adapter);
1284         if (err)
1285                 goto err_req_irq;
1286
1287         /* From here on the code is the same as e1000_up() */
1288         clear_bit(__E1000_DOWN, &adapter->flags);
1289
1290         napi_enable(&adapter->napi);
1291
1292         e1000_irq_enable(adapter);
1293
1294         netif_start_queue(netdev);
1295
1296         /* fire a link status change interrupt to start the watchdog */
1297         ew32(ICS, E1000_ICS_LSC);
1298
1299         return E1000_SUCCESS;
1300
1301 err_req_irq:
1302         e1000_power_down_phy(adapter);
1303         e1000_free_all_rx_resources(adapter);
1304 err_setup_rx:
1305         e1000_free_all_tx_resources(adapter);
1306 err_setup_tx:
1307         e1000_reset(adapter);
1308
1309         return err;
1310 }
1311
1312 /**
1313  * e1000_close - Disables a network interface
1314  * @netdev: network interface device structure
1315  *
1316  * Returns 0, this is not allowed to fail
1317  *
1318  * The close entry point is called when an interface is de-activated
1319  * by the OS.  The hardware is still under the drivers control, but
1320  * needs to be disabled.  A global MAC reset is issued to stop the
1321  * hardware, and all transmit and receive resources are freed.
1322  **/
1323
1324 static int e1000_close(struct net_device *netdev)
1325 {
1326         struct e1000_adapter *adapter = netdev_priv(netdev);
1327         struct e1000_hw *hw = &adapter->hw;
1328
1329         WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1330         e1000_down(adapter);
1331         e1000_power_down_phy(adapter);
1332         e1000_free_irq(adapter);
1333
1334         e1000_free_all_tx_resources(adapter);
1335         e1000_free_all_rx_resources(adapter);
1336
1337         /* kill manageability vlan ID if supported, but not if a vlan with
1338          * the same ID is registered on the host OS (let 8021q kill it) */
1339         if ((hw->mng_cookie.status &
1340                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1341              !(adapter->vlgrp &&
1342                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id))) {
1343                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1344         }
1345
1346         return 0;
1347 }
1348
1349 /**
1350  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1351  * @adapter: address of board private structure
1352  * @start: address of beginning of memory
1353  * @len: length of memory
1354  **/
1355 static bool e1000_check_64k_bound(struct e1000_adapter *adapter, void *start,
1356                                   unsigned long len)
1357 {
1358         struct e1000_hw *hw = &adapter->hw;
1359         unsigned long begin = (unsigned long)start;
1360         unsigned long end = begin + len;
1361
1362         /* First rev 82545 and 82546 need to not allow any memory
1363          * write location to cross 64k boundary due to errata 23 */
1364         if (hw->mac_type == e1000_82545 ||
1365             hw->mac_type == e1000_82546) {
1366                 return ((begin ^ (end - 1)) >> 16) != 0 ? false : true;
1367         }
1368
1369         return true;
1370 }
1371
1372 /**
1373  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1374  * @adapter: board private structure
1375  * @txdr:    tx descriptor ring (for a specific queue) to setup
1376  *
1377  * Return 0 on success, negative on failure
1378  **/
1379
1380 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
1381                                     struct e1000_tx_ring *txdr)
1382 {
1383         struct pci_dev *pdev = adapter->pdev;
1384         int size;
1385
1386         size = sizeof(struct e1000_buffer) * txdr->count;
1387         txdr->buffer_info = vmalloc(size);
1388         if (!txdr->buffer_info) {
1389                 e_err(probe, "Unable to allocate memory for the Tx descriptor "
1390                       "ring\n");
1391                 return -ENOMEM;
1392         }
1393         memset(txdr->buffer_info, 0, size);
1394
1395         /* round up to nearest 4K */
1396
1397         txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1398         txdr->size = ALIGN(txdr->size, 4096);
1399
1400         txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size, &txdr->dma,
1401                                         GFP_KERNEL);
1402         if (!txdr->desc) {
1403 setup_tx_desc_die:
1404                 vfree(txdr->buffer_info);
1405                 e_err(probe, "Unable to allocate memory for the Tx descriptor "
1406                       "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                 e_err(tx_err, "txdr align check failed: %u bytes at %p\n",
1415                       txdr->size, txdr->desc);
1416                 /* Try again, without freeing the previous */
1417                 txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size,
1418                                                 &txdr->dma, GFP_KERNEL);
1419                 /* Failed allocation, critical failure */
1420                 if (!txdr->desc) {
1421                         dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1422                                           olddma);
1423                         goto setup_tx_desc_die;
1424                 }
1425
1426                 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1427                         /* give up */
1428                         dma_free_coherent(&pdev->dev, txdr->size, txdr->desc,
1429                                           txdr->dma);
1430                         dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1431                                           olddma);
1432                         e_err(probe, "Unable to allocate aligned memory "
1433                               "for the transmit descriptor ring\n");
1434                         vfree(txdr->buffer_info);
1435                         return -ENOMEM;
1436                 } else {
1437                         /* Free old allocation, new allocation was successful */
1438                         dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1439                                           olddma);
1440                 }
1441         }
1442         memset(txdr->desc, 0, txdr->size);
1443
1444         txdr->next_to_use = 0;
1445         txdr->next_to_clean = 0;
1446
1447         return 0;
1448 }
1449
1450 /**
1451  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1452  *                                (Descriptors) for all queues
1453  * @adapter: board private structure
1454  *
1455  * Return 0 on success, negative on failure
1456  **/
1457
1458 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1459 {
1460         int i, err = 0;
1461
1462         for (i = 0; i < adapter->num_tx_queues; i++) {
1463                 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1464                 if (err) {
1465                         e_err(probe, "Allocation for Tx Queue %u failed\n", i);
1466                         for (i-- ; i >= 0; i--)
1467                                 e1000_free_tx_resources(adapter,
1468                                                         &adapter->tx_ring[i]);
1469                         break;
1470                 }
1471         }
1472
1473         return err;
1474 }
1475
1476 /**
1477  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1478  * @adapter: board private structure
1479  *
1480  * Configure the Tx unit of the MAC after a reset.
1481  **/
1482
1483 static void e1000_configure_tx(struct e1000_adapter *adapter)
1484 {
1485         u64 tdba;
1486         struct e1000_hw *hw = &adapter->hw;
1487         u32 tdlen, tctl, tipg;
1488         u32 ipgr1, ipgr2;
1489
1490         /* Setup the HW Tx Head and Tail descriptor pointers */
1491
1492         switch (adapter->num_tx_queues) {
1493         case 1:
1494         default:
1495                 tdba = adapter->tx_ring[0].dma;
1496                 tdlen = adapter->tx_ring[0].count *
1497                         sizeof(struct e1000_tx_desc);
1498                 ew32(TDLEN, tdlen);
1499                 ew32(TDBAH, (tdba >> 32));
1500                 ew32(TDBAL, (tdba & 0x00000000ffffffffULL));
1501                 ew32(TDT, 0);
1502                 ew32(TDH, 0);
1503                 adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ? E1000_TDH : E1000_82542_TDH);
1504                 adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ? E1000_TDT : E1000_82542_TDT);
1505                 break;
1506         }
1507
1508         /* Set the default values for the Tx Inter Packet Gap timer */
1509         if ((hw->media_type == e1000_media_type_fiber ||
1510              hw->media_type == e1000_media_type_internal_serdes))
1511                 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1512         else
1513                 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1514
1515         switch (hw->mac_type) {
1516         case e1000_82542_rev2_0:
1517         case e1000_82542_rev2_1:
1518                 tipg = DEFAULT_82542_TIPG_IPGT;
1519                 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1520                 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1521                 break;
1522         default:
1523                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1524                 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1525                 break;
1526         }
1527         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1528         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1529         ew32(TIPG, tipg);
1530
1531         /* Set the Tx Interrupt Delay register */
1532
1533         ew32(TIDV, adapter->tx_int_delay);
1534         if (hw->mac_type >= e1000_82540)
1535                 ew32(TADV, adapter->tx_abs_int_delay);
1536
1537         /* Program the Transmit Control Register */
1538
1539         tctl = er32(TCTL);
1540         tctl &= ~E1000_TCTL_CT;
1541         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1542                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1543
1544         e1000_config_collision_dist(hw);
1545
1546         /* Setup Transmit Descriptor Settings for eop descriptor */
1547         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1548
1549         /* only set IDE if we are delaying interrupts using the timers */
1550         if (adapter->tx_int_delay)
1551                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1552
1553         if (hw->mac_type < e1000_82543)
1554                 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1555         else
1556                 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1557
1558         /* Cache if we're 82544 running in PCI-X because we'll
1559          * need this to apply a workaround later in the send path. */
1560         if (hw->mac_type == e1000_82544 &&
1561             hw->bus_type == e1000_bus_type_pcix)
1562                 adapter->pcix_82544 = 1;
1563
1564         ew32(TCTL, tctl);
1565
1566 }
1567
1568 /**
1569  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1570  * @adapter: board private structure
1571  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1572  *
1573  * Returns 0 on success, negative on failure
1574  **/
1575
1576 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
1577                                     struct e1000_rx_ring *rxdr)
1578 {
1579         struct pci_dev *pdev = adapter->pdev;
1580         int size, desc_len;
1581
1582         size = sizeof(struct e1000_buffer) * rxdr->count;
1583         rxdr->buffer_info = vmalloc(size);
1584         if (!rxdr->buffer_info) {
1585                 e_err(probe, "Unable to allocate memory for the Rx descriptor "
1586                       "ring\n");
1587                 return -ENOMEM;
1588         }
1589         memset(rxdr->buffer_info, 0, size);
1590
1591         desc_len = sizeof(struct e1000_rx_desc);
1592
1593         /* Round up to nearest 4K */
1594
1595         rxdr->size = rxdr->count * desc_len;
1596         rxdr->size = ALIGN(rxdr->size, 4096);
1597
1598         rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size, &rxdr->dma,
1599                                         GFP_KERNEL);
1600
1601         if (!rxdr->desc) {
1602                 e_err(probe, "Unable to allocate memory for the Rx descriptor "
1603                       "ring\n");
1604 setup_rx_desc_die:
1605                 vfree(rxdr->buffer_info);
1606                 return -ENOMEM;
1607         }
1608
1609         /* Fix for errata 23, can't cross 64kB boundary */
1610         if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1611                 void *olddesc = rxdr->desc;
1612                 dma_addr_t olddma = rxdr->dma;
1613                 e_err(rx_err, "rxdr align check failed: %u bytes at %p\n",
1614                       rxdr->size, rxdr->desc);
1615                 /* Try again, without freeing the previous */
1616                 rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size,
1617                                                 &rxdr->dma, GFP_KERNEL);
1618                 /* Failed allocation, critical failure */
1619                 if (!rxdr->desc) {
1620                         dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1621                                           olddma);
1622                         e_err(probe, "Unable to allocate memory for the Rx "
1623                               "descriptor ring\n");
1624                         goto setup_rx_desc_die;
1625                 }
1626
1627                 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1628                         /* give up */
1629                         dma_free_coherent(&pdev->dev, rxdr->size, rxdr->desc,
1630                                           rxdr->dma);
1631                         dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1632                                           olddma);
1633                         e_err(probe, "Unable to allocate aligned memory for "
1634                               "the Rx descriptor ring\n");
1635                         goto setup_rx_desc_die;
1636                 } else {
1637                         /* Free old allocation, new allocation was successful */
1638                         dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1639                                           olddma);
1640                 }
1641         }
1642         memset(rxdr->desc, 0, rxdr->size);
1643
1644         rxdr->next_to_clean = 0;
1645         rxdr->next_to_use = 0;
1646         rxdr->rx_skb_top = NULL;
1647
1648         return 0;
1649 }
1650
1651 /**
1652  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1653  *                                (Descriptors) for all queues
1654  * @adapter: board private structure
1655  *
1656  * Return 0 on success, negative on failure
1657  **/
1658
1659 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1660 {
1661         int i, err = 0;
1662
1663         for (i = 0; i < adapter->num_rx_queues; i++) {
1664                 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1665                 if (err) {
1666                         e_err(probe, "Allocation for Rx Queue %u failed\n", i);
1667                         for (i-- ; i >= 0; i--)
1668                                 e1000_free_rx_resources(adapter,
1669                                                         &adapter->rx_ring[i]);
1670                         break;
1671                 }
1672         }
1673
1674         return err;
1675 }
1676
1677 /**
1678  * e1000_setup_rctl - configure the receive control registers
1679  * @adapter: Board private structure
1680  **/
1681 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1682 {
1683         struct e1000_hw *hw = &adapter->hw;
1684         u32 rctl;
1685
1686         rctl = er32(RCTL);
1687
1688         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1689
1690         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1691                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1692                 (hw->mc_filter_type << E1000_RCTL_MO_SHIFT);
1693
1694         if (hw->tbi_compatibility_on == 1)
1695                 rctl |= E1000_RCTL_SBP;
1696         else
1697                 rctl &= ~E1000_RCTL_SBP;
1698
1699         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1700                 rctl &= ~E1000_RCTL_LPE;
1701         else
1702                 rctl |= E1000_RCTL_LPE;
1703
1704         /* Setup buffer sizes */
1705         rctl &= ~E1000_RCTL_SZ_4096;
1706         rctl |= E1000_RCTL_BSEX;
1707         switch (adapter->rx_buffer_len) {
1708                 case E1000_RXBUFFER_2048:
1709                 default:
1710                         rctl |= E1000_RCTL_SZ_2048;
1711                         rctl &= ~E1000_RCTL_BSEX;
1712                         break;
1713                 case E1000_RXBUFFER_4096:
1714                         rctl |= E1000_RCTL_SZ_4096;
1715                         break;
1716                 case E1000_RXBUFFER_8192:
1717                         rctl |= E1000_RCTL_SZ_8192;
1718                         break;
1719                 case E1000_RXBUFFER_16384:
1720                         rctl |= E1000_RCTL_SZ_16384;
1721                         break;
1722         }
1723
1724         ew32(RCTL, rctl);
1725 }
1726
1727 /**
1728  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1729  * @adapter: board private structure
1730  *
1731  * Configure the Rx unit of the MAC after a reset.
1732  **/
1733
1734 static void e1000_configure_rx(struct e1000_adapter *adapter)
1735 {
1736         u64 rdba;
1737         struct e1000_hw *hw = &adapter->hw;
1738         u32 rdlen, rctl, rxcsum;
1739
1740         if (adapter->netdev->mtu > ETH_DATA_LEN) {
1741                 rdlen = adapter->rx_ring[0].count *
1742                         sizeof(struct e1000_rx_desc);
1743                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
1744                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
1745         } else {
1746                 rdlen = adapter->rx_ring[0].count *
1747                         sizeof(struct e1000_rx_desc);
1748                 adapter->clean_rx = e1000_clean_rx_irq;
1749                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1750         }
1751
1752         /* disable receives while setting up the descriptors */
1753         rctl = er32(RCTL);
1754         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1755
1756         /* set the Receive Delay Timer Register */
1757         ew32(RDTR, adapter->rx_int_delay);
1758
1759         if (hw->mac_type >= e1000_82540) {
1760                 ew32(RADV, adapter->rx_abs_int_delay);
1761                 if (adapter->itr_setting != 0)
1762                         ew32(ITR, 1000000000 / (adapter->itr * 256));
1763         }
1764
1765         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1766          * the Base and Length of the Rx Descriptor Ring */
1767         switch (adapter->num_rx_queues) {
1768         case 1:
1769         default:
1770                 rdba = adapter->rx_ring[0].dma;
1771                 ew32(RDLEN, rdlen);
1772                 ew32(RDBAH, (rdba >> 32));
1773                 ew32(RDBAL, (rdba & 0x00000000ffffffffULL));
1774                 ew32(RDT, 0);
1775                 ew32(RDH, 0);
1776                 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ? E1000_RDH : E1000_82542_RDH);
1777                 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ? E1000_RDT : E1000_82542_RDT);
1778                 break;
1779         }
1780
1781         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
1782         if (hw->mac_type >= e1000_82543) {
1783                 rxcsum = er32(RXCSUM);
1784                 if (adapter->rx_csum)
1785                         rxcsum |= E1000_RXCSUM_TUOFL;
1786                 else
1787                         /* don't need to clear IPPCSE as it defaults to 0 */
1788                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1789                 ew32(RXCSUM, rxcsum);
1790         }
1791
1792         /* Enable Receives */
1793         ew32(RCTL, rctl);
1794 }
1795
1796 /**
1797  * e1000_free_tx_resources - Free Tx Resources per Queue
1798  * @adapter: board private structure
1799  * @tx_ring: Tx descriptor ring for a specific queue
1800  *
1801  * Free all transmit software resources
1802  **/
1803
1804 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
1805                                     struct e1000_tx_ring *tx_ring)
1806 {
1807         struct pci_dev *pdev = adapter->pdev;
1808
1809         e1000_clean_tx_ring(adapter, tx_ring);
1810
1811         vfree(tx_ring->buffer_info);
1812         tx_ring->buffer_info = NULL;
1813
1814         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1815                           tx_ring->dma);
1816
1817         tx_ring->desc = NULL;
1818 }
1819
1820 /**
1821  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1822  * @adapter: board private structure
1823  *
1824  * Free all transmit software resources
1825  **/
1826
1827 void e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1828 {
1829         int i;
1830
1831         for (i = 0; i < adapter->num_tx_queues; i++)
1832                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1833 }
1834
1835 static void e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1836                                              struct e1000_buffer *buffer_info)
1837 {
1838         if (buffer_info->dma) {
1839                 if (buffer_info->mapped_as_page)
1840                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1841                                        buffer_info->length, DMA_TO_DEVICE);
1842                 else
1843                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1844                                          buffer_info->length,
1845                                          DMA_TO_DEVICE);
1846                 buffer_info->dma = 0;
1847         }
1848         if (buffer_info->skb) {
1849                 dev_kfree_skb_any(buffer_info->skb);
1850                 buffer_info->skb = NULL;
1851         }
1852         buffer_info->time_stamp = 0;
1853         /* buffer_info must be completely set up in the transmit path */
1854 }
1855
1856 /**
1857  * e1000_clean_tx_ring - Free Tx Buffers
1858  * @adapter: board private structure
1859  * @tx_ring: ring to be cleaned
1860  **/
1861
1862 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
1863                                 struct e1000_tx_ring *tx_ring)
1864 {
1865         struct e1000_hw *hw = &adapter->hw;
1866         struct e1000_buffer *buffer_info;
1867         unsigned long size;
1868         unsigned int i;
1869
1870         /* Free all the Tx ring sk_buffs */
1871
1872         for (i = 0; i < tx_ring->count; i++) {
1873                 buffer_info = &tx_ring->buffer_info[i];
1874                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
1875         }
1876
1877         size = sizeof(struct e1000_buffer) * tx_ring->count;
1878         memset(tx_ring->buffer_info, 0, size);
1879
1880         /* Zero out the descriptor ring */
1881
1882         memset(tx_ring->desc, 0, tx_ring->size);
1883
1884         tx_ring->next_to_use = 0;
1885         tx_ring->next_to_clean = 0;
1886         tx_ring->last_tx_tso = 0;
1887
1888         writel(0, hw->hw_addr + tx_ring->tdh);
1889         writel(0, hw->hw_addr + tx_ring->tdt);
1890 }
1891
1892 /**
1893  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
1894  * @adapter: board private structure
1895  **/
1896
1897 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
1898 {
1899         int i;
1900
1901         for (i = 0; i < adapter->num_tx_queues; i++)
1902                 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1903 }
1904
1905 /**
1906  * e1000_free_rx_resources - Free Rx Resources
1907  * @adapter: board private structure
1908  * @rx_ring: ring to clean the resources from
1909  *
1910  * Free all receive software resources
1911  **/
1912
1913 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
1914                                     struct e1000_rx_ring *rx_ring)
1915 {
1916         struct pci_dev *pdev = adapter->pdev;
1917
1918         e1000_clean_rx_ring(adapter, rx_ring);
1919
1920         vfree(rx_ring->buffer_info);
1921         rx_ring->buffer_info = NULL;
1922
1923         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1924                           rx_ring->dma);
1925
1926         rx_ring->desc = NULL;
1927 }
1928
1929 /**
1930  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
1931  * @adapter: board private structure
1932  *
1933  * Free all receive software resources
1934  **/
1935
1936 void e1000_free_all_rx_resources(struct e1000_adapter *adapter)
1937 {
1938         int i;
1939
1940         for (i = 0; i < adapter->num_rx_queues; i++)
1941                 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
1942 }
1943
1944 /**
1945  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1946  * @adapter: board private structure
1947  * @rx_ring: ring to free buffers from
1948  **/
1949
1950 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
1951                                 struct e1000_rx_ring *rx_ring)
1952 {
1953         struct e1000_hw *hw = &adapter->hw;
1954         struct e1000_buffer *buffer_info;
1955         struct pci_dev *pdev = adapter->pdev;
1956         unsigned long size;
1957         unsigned int i;
1958
1959         /* Free all the Rx ring sk_buffs */
1960         for (i = 0; i < rx_ring->count; i++) {
1961                 buffer_info = &rx_ring->buffer_info[i];
1962                 if (buffer_info->dma &&
1963                     adapter->clean_rx == e1000_clean_rx_irq) {
1964                         dma_unmap_single(&pdev->dev, buffer_info->dma,
1965                                          buffer_info->length,
1966                                          DMA_FROM_DEVICE);
1967                 } else if (buffer_info->dma &&
1968                            adapter->clean_rx == e1000_clean_jumbo_rx_irq) {
1969                         dma_unmap_page(&pdev->dev, buffer_info->dma,
1970                                        buffer_info->length,
1971                                        DMA_FROM_DEVICE);
1972                 }
1973
1974                 buffer_info->dma = 0;
1975                 if (buffer_info->page) {
1976                         put_page(buffer_info->page);
1977                         buffer_info->page = NULL;
1978                 }
1979                 if (buffer_info->skb) {
1980                         dev_kfree_skb(buffer_info->skb);
1981                         buffer_info->skb = NULL;
1982                 }
1983         }
1984
1985         /* there also may be some cached data from a chained receive */
1986         if (rx_ring->rx_skb_top) {
1987                 dev_kfree_skb(rx_ring->rx_skb_top);
1988                 rx_ring->rx_skb_top = NULL;
1989         }
1990
1991         size = sizeof(struct e1000_buffer) * rx_ring->count;
1992         memset(rx_ring->buffer_info, 0, size);
1993
1994         /* Zero out the descriptor ring */
1995         memset(rx_ring->desc, 0, rx_ring->size);
1996
1997         rx_ring->next_to_clean = 0;
1998         rx_ring->next_to_use = 0;
1999
2000         writel(0, hw->hw_addr + rx_ring->rdh);
2001         writel(0, hw->hw_addr + rx_ring->rdt);
2002 }
2003
2004 /**
2005  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2006  * @adapter: board private structure
2007  **/
2008
2009 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2010 {
2011         int i;
2012
2013         for (i = 0; i < adapter->num_rx_queues; i++)
2014                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2015 }
2016
2017 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2018  * and memory write and invalidate disabled for certain operations
2019  */
2020 static void e1000_enter_82542_rst(struct e1000_adapter *adapter)
2021 {
2022         struct e1000_hw *hw = &adapter->hw;
2023         struct net_device *netdev = adapter->netdev;
2024         u32 rctl;
2025
2026         e1000_pci_clear_mwi(hw);
2027
2028         rctl = er32(RCTL);
2029         rctl |= E1000_RCTL_RST;
2030         ew32(RCTL, rctl);
2031         E1000_WRITE_FLUSH();
2032         mdelay(5);
2033
2034         if (netif_running(netdev))
2035                 e1000_clean_all_rx_rings(adapter);
2036 }
2037
2038 static void e1000_leave_82542_rst(struct e1000_adapter *adapter)
2039 {
2040         struct e1000_hw *hw = &adapter->hw;
2041         struct net_device *netdev = adapter->netdev;
2042         u32 rctl;
2043
2044         rctl = er32(RCTL);
2045         rctl &= ~E1000_RCTL_RST;
2046         ew32(RCTL, rctl);
2047         E1000_WRITE_FLUSH();
2048         mdelay(5);
2049
2050         if (hw->pci_cmd_word & PCI_COMMAND_INVALIDATE)
2051                 e1000_pci_set_mwi(hw);
2052
2053         if (netif_running(netdev)) {
2054                 /* No need to loop, because 82542 supports only 1 queue */
2055                 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2056                 e1000_configure_rx(adapter);
2057                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2058         }
2059 }
2060
2061 /**
2062  * e1000_set_mac - Change the Ethernet Address of the NIC
2063  * @netdev: network interface device structure
2064  * @p: pointer to an address structure
2065  *
2066  * Returns 0 on success, negative on failure
2067  **/
2068
2069 static int e1000_set_mac(struct net_device *netdev, void *p)
2070 {
2071         struct e1000_adapter *adapter = netdev_priv(netdev);
2072         struct e1000_hw *hw = &adapter->hw;
2073         struct sockaddr *addr = p;
2074
2075         if (!is_valid_ether_addr(addr->sa_data))
2076                 return -EADDRNOTAVAIL;
2077
2078         /* 82542 2.0 needs to be in reset to write receive address registers */
2079
2080         if (hw->mac_type == e1000_82542_rev2_0)
2081                 e1000_enter_82542_rst(adapter);
2082
2083         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2084         memcpy(hw->mac_addr, addr->sa_data, netdev->addr_len);
2085
2086         e1000_rar_set(hw, hw->mac_addr, 0);
2087
2088         if (hw->mac_type == e1000_82542_rev2_0)
2089                 e1000_leave_82542_rst(adapter);
2090
2091         return 0;
2092 }
2093
2094 /**
2095  * e1000_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2096  * @netdev: network interface device structure
2097  *
2098  * The set_rx_mode entry point is called whenever the unicast or multicast
2099  * address lists or the network interface flags are updated. This routine is
2100  * responsible for configuring the hardware for proper unicast, multicast,
2101  * promiscuous mode, and all-multi behavior.
2102  **/
2103
2104 static void e1000_set_rx_mode(struct net_device *netdev)
2105 {
2106         struct e1000_adapter *adapter = netdev_priv(netdev);
2107         struct e1000_hw *hw = &adapter->hw;
2108         struct netdev_hw_addr *ha;
2109         bool use_uc = false;
2110         u32 rctl;
2111         u32 hash_value;
2112         int i, rar_entries = E1000_RAR_ENTRIES;
2113         int mta_reg_count = E1000_NUM_MTA_REGISTERS;
2114         u32 *mcarray = kcalloc(mta_reg_count, sizeof(u32), GFP_ATOMIC);
2115
2116         if (!mcarray) {
2117                 e_err(probe, "memory allocation failed\n");
2118                 return;
2119         }
2120
2121         /* Check for Promiscuous and All Multicast modes */
2122
2123         rctl = er32(RCTL);
2124
2125         if (netdev->flags & IFF_PROMISC) {
2126                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2127                 rctl &= ~E1000_RCTL_VFE;
2128         } else {
2129                 if (netdev->flags & IFF_ALLMULTI)
2130                         rctl |= E1000_RCTL_MPE;
2131                 else
2132                         rctl &= ~E1000_RCTL_MPE;
2133                 /* Enable VLAN filter if there is a VLAN */
2134                 if (adapter->vlgrp)
2135                         rctl |= E1000_RCTL_VFE;
2136         }
2137
2138         if (netdev_uc_count(netdev) > rar_entries - 1) {
2139                 rctl |= E1000_RCTL_UPE;
2140         } else if (!(netdev->flags & IFF_PROMISC)) {
2141                 rctl &= ~E1000_RCTL_UPE;
2142                 use_uc = true;
2143         }
2144
2145         ew32(RCTL, rctl);
2146
2147         /* 82542 2.0 needs to be in reset to write receive address registers */
2148
2149         if (hw->mac_type == e1000_82542_rev2_0)
2150                 e1000_enter_82542_rst(adapter);
2151
2152         /* load the first 14 addresses into the exact filters 1-14. Unicast
2153          * addresses take precedence to avoid disabling unicast filtering
2154          * when possible.
2155          *
2156          * RAR 0 is used for the station MAC adddress
2157          * if there are not 14 addresses, go ahead and clear the filters
2158          */
2159         i = 1;
2160         if (use_uc)
2161                 netdev_for_each_uc_addr(ha, netdev) {
2162                         if (i == rar_entries)
2163                                 break;
2164                         e1000_rar_set(hw, ha->addr, i++);
2165                 }
2166
2167         netdev_for_each_mc_addr(ha, netdev) {
2168                 if (i == rar_entries) {
2169                         /* load any remaining addresses into the hash table */
2170                         u32 hash_reg, hash_bit, mta;
2171                         hash_value = e1000_hash_mc_addr(hw, ha->addr);
2172                         hash_reg = (hash_value >> 5) & 0x7F;
2173                         hash_bit = hash_value & 0x1F;
2174                         mta = (1 << hash_bit);
2175                         mcarray[hash_reg] |= mta;
2176                 } else {
2177                         e1000_rar_set(hw, ha->addr, i++);
2178                 }
2179         }
2180
2181         for (; i < rar_entries; i++) {
2182                 E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2183                 E1000_WRITE_FLUSH();
2184                 E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2185                 E1000_WRITE_FLUSH();
2186         }
2187
2188         /* write the hash table completely, write from bottom to avoid
2189          * both stupid write combining chipsets, and flushing each write */
2190         for (i = mta_reg_count - 1; i >= 0 ; i--) {
2191                 /*
2192                  * If we are on an 82544 has an errata where writing odd
2193                  * offsets overwrites the previous even offset, but writing
2194                  * backwards over the range solves the issue by always
2195                  * writing the odd offset first
2196                  */
2197                 E1000_WRITE_REG_ARRAY(hw, MTA, i, mcarray[i]);
2198         }
2199         E1000_WRITE_FLUSH();
2200
2201         if (hw->mac_type == e1000_82542_rev2_0)
2202                 e1000_leave_82542_rst(adapter);
2203
2204         kfree(mcarray);
2205 }
2206
2207 /* Need to wait a few seconds after link up to get diagnostic information from
2208  * the phy */
2209
2210 static void e1000_update_phy_info(unsigned long data)
2211 {
2212         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2213         struct e1000_hw *hw = &adapter->hw;
2214         e1000_phy_get_info(hw, &adapter->phy_info);
2215 }
2216
2217 /**
2218  * e1000_82547_tx_fifo_stall - Timer Call-back
2219  * @data: pointer to adapter cast into an unsigned long
2220  **/
2221
2222 static void e1000_82547_tx_fifo_stall(unsigned long data)
2223 {
2224         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2225         struct e1000_hw *hw = &adapter->hw;
2226         struct net_device *netdev = adapter->netdev;
2227         u32 tctl;
2228
2229         if (atomic_read(&adapter->tx_fifo_stall)) {
2230                 if ((er32(TDT) == er32(TDH)) &&
2231                    (er32(TDFT) == er32(TDFH)) &&
2232                    (er32(TDFTS) == er32(TDFHS))) {
2233                         tctl = er32(TCTL);
2234                         ew32(TCTL, tctl & ~E1000_TCTL_EN);
2235                         ew32(TDFT, adapter->tx_head_addr);
2236                         ew32(TDFH, adapter->tx_head_addr);
2237                         ew32(TDFTS, adapter->tx_head_addr);
2238                         ew32(TDFHS, adapter->tx_head_addr);
2239                         ew32(TCTL, tctl);
2240                         E1000_WRITE_FLUSH();
2241
2242                         adapter->tx_fifo_head = 0;
2243                         atomic_set(&adapter->tx_fifo_stall, 0);
2244                         netif_wake_queue(netdev);
2245                 } else if (!test_bit(__E1000_DOWN, &adapter->flags)) {
2246                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2247                 }
2248         }
2249 }
2250
2251 bool e1000_has_link(struct e1000_adapter *adapter)
2252 {
2253         struct e1000_hw *hw = &adapter->hw;
2254         bool link_active = false;
2255
2256         /* get_link_status is set on LSC (link status) interrupt or
2257          * rx sequence error interrupt.  get_link_status will stay
2258          * false until the e1000_check_for_link establishes link
2259          * for copper adapters ONLY
2260          */
2261         switch (hw->media_type) {
2262         case e1000_media_type_copper:
2263                 if (hw->get_link_status) {
2264                         e1000_check_for_link(hw);
2265                         link_active = !hw->get_link_status;
2266                 } else {
2267                         link_active = true;
2268                 }
2269                 break;
2270         case e1000_media_type_fiber:
2271                 e1000_check_for_link(hw);
2272                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2273                 break;
2274         case e1000_media_type_internal_serdes:
2275                 e1000_check_for_link(hw);
2276                 link_active = hw->serdes_has_link;
2277                 break;
2278         default:
2279                 break;
2280         }
2281
2282         return link_active;
2283 }
2284
2285 /**
2286  * e1000_watchdog - Timer Call-back
2287  * @data: pointer to adapter cast into an unsigned long
2288  **/
2289 static void e1000_watchdog(unsigned long data)
2290 {
2291         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2292         struct e1000_hw *hw = &adapter->hw;
2293         struct net_device *netdev = adapter->netdev;
2294         struct e1000_tx_ring *txdr = adapter->tx_ring;
2295         u32 link, tctl;
2296
2297         link = e1000_has_link(adapter);
2298         if ((netif_carrier_ok(netdev)) && link)
2299                 goto link_up;
2300
2301         if (link) {
2302                 if (!netif_carrier_ok(netdev)) {
2303                         u32 ctrl;
2304                         bool txb2b = true;
2305                         /* update snapshot of PHY registers on LSC */
2306                         e1000_get_speed_and_duplex(hw,
2307                                                    &adapter->link_speed,
2308                                                    &adapter->link_duplex);
2309
2310                         ctrl = er32(CTRL);
2311                         pr_info("%s NIC Link is Up %d Mbps %s, "
2312                                 "Flow Control: %s\n",
2313                                 netdev->name,
2314                                 adapter->link_speed,
2315                                 adapter->link_duplex == FULL_DUPLEX ?
2316                                 "Full Duplex" : "Half Duplex",
2317                                 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2318                                 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2319                                 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2320                                 E1000_CTRL_TFCE) ? "TX" : "None")));
2321
2322                         /* adjust timeout factor according to speed/duplex */
2323                         adapter->tx_timeout_factor = 1;
2324                         switch (adapter->link_speed) {
2325                         case SPEED_10:
2326                                 txb2b = false;
2327                                 adapter->tx_timeout_factor = 16;
2328                                 break;
2329                         case SPEED_100:
2330                                 txb2b = false;
2331                                 /* maybe add some timeout factor ? */
2332                                 break;
2333                         }
2334
2335                         /* enable transmits in the hardware */
2336                         tctl = er32(TCTL);
2337                         tctl |= E1000_TCTL_EN;
2338                         ew32(TCTL, tctl);
2339
2340                         netif_carrier_on(netdev);
2341                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2342                                 mod_timer(&adapter->phy_info_timer,
2343                                           round_jiffies(jiffies + 2 * HZ));
2344                         adapter->smartspeed = 0;
2345                 }
2346         } else {
2347                 if (netif_carrier_ok(netdev)) {
2348                         adapter->link_speed = 0;
2349                         adapter->link_duplex = 0;
2350                         pr_info("%s NIC Link is Down\n",
2351                                 netdev->name);
2352                         netif_carrier_off(netdev);
2353
2354                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2355                                 mod_timer(&adapter->phy_info_timer,
2356                                           round_jiffies(jiffies + 2 * HZ));
2357                 }
2358
2359                 e1000_smartspeed(adapter);
2360         }
2361
2362 link_up:
2363         e1000_update_stats(adapter);
2364
2365         hw->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2366         adapter->tpt_old = adapter->stats.tpt;
2367         hw->collision_delta = adapter->stats.colc - adapter->colc_old;
2368         adapter->colc_old = adapter->stats.colc;
2369
2370         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2371         adapter->gorcl_old = adapter->stats.gorcl;
2372         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2373         adapter->gotcl_old = adapter->stats.gotcl;
2374
2375         e1000_update_adaptive(hw);
2376
2377         if (!netif_carrier_ok(netdev)) {
2378                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2379                         /* We've lost link, so the controller stops DMA,
2380                          * but we've got queued Tx work that's never going
2381                          * to get done, so reset controller to flush Tx.
2382                          * (Do the reset outside of interrupt context). */
2383                         adapter->tx_timeout_count++;
2384                         schedule_work(&adapter->reset_task);
2385                         /* return immediately since reset is imminent */
2386                         return;
2387                 }
2388         }
2389
2390         /* Simple mode for Interrupt Throttle Rate (ITR) */
2391         if (hw->mac_type >= e1000_82540 && adapter->itr_setting == 4) {
2392                 /*
2393                  * Symmetric Tx/Rx gets a reduced ITR=2000;
2394                  * Total asymmetrical Tx or Rx gets ITR=8000;
2395                  * everyone else is between 2000-8000.
2396                  */
2397                 u32 goc = (adapter->gotcl + adapter->gorcl) / 10000;
2398                 u32 dif = (adapter->gotcl > adapter->gorcl ?
2399                             adapter->gotcl - adapter->gorcl :
2400                             adapter->gorcl - adapter->gotcl) / 10000;
2401                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
2402
2403                 ew32(ITR, 1000000000 / (itr * 256));
2404         }
2405
2406         /* Cause software interrupt to ensure rx ring is cleaned */
2407         ew32(ICS, E1000_ICS_RXDMT0);
2408
2409         /* Force detection of hung controller every watchdog period */
2410         adapter->detect_tx_hung = true;
2411
2412         /* Reset the timer */
2413         if (!test_bit(__E1000_DOWN, &adapter->flags))
2414                 mod_timer(&adapter->watchdog_timer,
2415                           round_jiffies(jiffies + 2 * HZ));
2416 }
2417
2418 enum latency_range {
2419         lowest_latency = 0,
2420         low_latency = 1,
2421         bulk_latency = 2,
2422         latency_invalid = 255
2423 };
2424
2425 /**
2426  * e1000_update_itr - update the dynamic ITR value based on statistics
2427  * @adapter: pointer to adapter
2428  * @itr_setting: current adapter->itr
2429  * @packets: the number of packets during this measurement interval
2430  * @bytes: the number of bytes during this measurement interval
2431  *
2432  *      Stores a new ITR value based on packets and byte
2433  *      counts during the last interrupt.  The advantage of per interrupt
2434  *      computation is faster updates and more accurate ITR for the current
2435  *      traffic pattern.  Constants in this function were computed
2436  *      based on theoretical maximum wire speed and thresholds were set based
2437  *      on testing data as well as attempting to minimize response time
2438  *      while increasing bulk throughput.
2439  *      this functionality is controlled by the InterruptThrottleRate module
2440  *      parameter (see e1000_param.c)
2441  **/
2442 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2443                                      u16 itr_setting, int packets, int bytes)
2444 {
2445         unsigned int retval = itr_setting;
2446         struct e1000_hw *hw = &adapter->hw;
2447
2448         if (unlikely(hw->mac_type < e1000_82540))
2449                 goto update_itr_done;
2450
2451         if (packets == 0)
2452                 goto update_itr_done;
2453
2454         switch (itr_setting) {
2455         case lowest_latency:
2456                 /* jumbo frames get bulk treatment*/
2457                 if (bytes/packets > 8000)
2458                         retval = bulk_latency;
2459                 else if ((packets < 5) && (bytes > 512))
2460                         retval = low_latency;
2461                 break;
2462         case low_latency:  /* 50 usec aka 20000 ints/s */
2463                 if (bytes > 10000) {
2464                         /* jumbo frames need bulk latency setting */
2465                         if (bytes/packets > 8000)
2466                                 retval = bulk_latency;
2467                         else if ((packets < 10) || ((bytes/packets) > 1200))
2468                                 retval = bulk_latency;
2469                         else if ((packets > 35))
2470                                 retval = lowest_latency;
2471                 } else if (bytes/packets > 2000)
2472                         retval = bulk_latency;
2473                 else if (packets <= 2 && bytes < 512)
2474                         retval = lowest_latency;
2475                 break;
2476         case bulk_latency: /* 250 usec aka 4000 ints/s */
2477                 if (bytes > 25000) {
2478                         if (packets > 35)
2479                                 retval = low_latency;
2480                 } else if (bytes < 6000) {
2481                         retval = low_latency;
2482                 }
2483                 break;
2484         }
2485
2486 update_itr_done:
2487         return retval;
2488 }
2489
2490 static void e1000_set_itr(struct e1000_adapter *adapter)
2491 {
2492         struct e1000_hw *hw = &adapter->hw;
2493         u16 current_itr;
2494         u32 new_itr = adapter->itr;
2495
2496         if (unlikely(hw->mac_type < e1000_82540))
2497                 return;
2498
2499         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2500         if (unlikely(adapter->link_speed != SPEED_1000)) {
2501                 current_itr = 0;
2502                 new_itr = 4000;
2503                 goto set_itr_now;
2504         }
2505
2506         adapter->tx_itr = e1000_update_itr(adapter,
2507                                     adapter->tx_itr,
2508                                     adapter->total_tx_packets,
2509                                     adapter->total_tx_bytes);
2510         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2511         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2512                 adapter->tx_itr = low_latency;
2513
2514         adapter->rx_itr = e1000_update_itr(adapter,
2515                                     adapter->rx_itr,
2516                                     adapter->total_rx_packets,
2517                                     adapter->total_rx_bytes);
2518         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2519         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2520                 adapter->rx_itr = low_latency;
2521
2522         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2523
2524         switch (current_itr) {
2525         /* counts and packets in update_itr are dependent on these numbers */
2526         case lowest_latency:
2527                 new_itr = 70000;
2528                 break;
2529         case low_latency:
2530                 new_itr = 20000; /* aka hwitr = ~200 */
2531                 break;
2532         case bulk_latency:
2533                 new_itr = 4000;
2534                 break;
2535         default:
2536                 break;
2537         }
2538
2539 set_itr_now:
2540         if (new_itr != adapter->itr) {
2541                 /* this attempts to bias the interrupt rate towards Bulk
2542                  * by adding intermediate steps when interrupt rate is
2543                  * increasing */
2544                 new_itr = new_itr > adapter->itr ?
2545                              min(adapter->itr + (new_itr >> 2), new_itr) :
2546                              new_itr;
2547                 adapter->itr = new_itr;
2548                 ew32(ITR, 1000000000 / (new_itr * 256));
2549         }
2550 }
2551
2552 #define E1000_TX_FLAGS_CSUM             0x00000001
2553 #define E1000_TX_FLAGS_VLAN             0x00000002
2554 #define E1000_TX_FLAGS_TSO              0x00000004
2555 #define E1000_TX_FLAGS_IPV4             0x00000008
2556 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2557 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2558
2559 static int e1000_tso(struct e1000_adapter *adapter,
2560                      struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2561 {
2562         struct e1000_context_desc *context_desc;
2563         struct e1000_buffer *buffer_info;
2564         unsigned int i;
2565         u32 cmd_length = 0;
2566         u16 ipcse = 0, tucse, mss;
2567         u8 ipcss, ipcso, tucss, tucso, hdr_len;
2568         int err;
2569
2570         if (skb_is_gso(skb)) {
2571                 if (skb_header_cloned(skb)) {
2572                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2573                         if (err)
2574                                 return err;
2575                 }
2576
2577                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2578                 mss = skb_shinfo(skb)->gso_size;
2579                 if (skb->protocol == htons(ETH_P_IP)) {
2580                         struct iphdr *iph = ip_hdr(skb);
2581                         iph->tot_len = 0;
2582                         iph->check = 0;
2583                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2584                                                                  iph->daddr, 0,
2585                                                                  IPPROTO_TCP,
2586                                                                  0);
2587                         cmd_length = E1000_TXD_CMD_IP;
2588                         ipcse = skb_transport_offset(skb) - 1;
2589                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2590                         ipv6_hdr(skb)->payload_len = 0;
2591                         tcp_hdr(skb)->check =
2592                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2593                                                  &ipv6_hdr(skb)->daddr,
2594                                                  0, IPPROTO_TCP, 0);
2595                         ipcse = 0;
2596                 }
2597                 ipcss = skb_network_offset(skb);
2598                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2599                 tucss = skb_transport_offset(skb);
2600                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2601                 tucse = 0;
2602
2603                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2604                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2605
2606                 i = tx_ring->next_to_use;
2607                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2608                 buffer_info = &tx_ring->buffer_info[i];
2609
2610                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2611                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2612                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2613                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2614                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2615                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2616                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2617                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2618                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2619
2620                 buffer_info->time_stamp = jiffies;
2621                 buffer_info->next_to_watch = i;
2622
2623                 if (++i == tx_ring->count) i = 0;
2624                 tx_ring->next_to_use = i;
2625
2626                 return true;
2627         }
2628         return false;
2629 }
2630
2631 static bool e1000_tx_csum(struct e1000_adapter *adapter,
2632                           struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2633 {
2634         struct e1000_context_desc *context_desc;
2635         struct e1000_buffer *buffer_info;
2636         unsigned int i;
2637         u8 css;
2638         u32 cmd_len = E1000_TXD_CMD_DEXT;
2639
2640         if (skb->ip_summed != CHECKSUM_PARTIAL)
2641                 return false;
2642
2643         switch (skb->protocol) {
2644         case cpu_to_be16(ETH_P_IP):
2645                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2646                         cmd_len |= E1000_TXD_CMD_TCP;
2647                 break;
2648         case cpu_to_be16(ETH_P_IPV6):
2649                 /* XXX not handling all IPV6 headers */
2650                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2651                         cmd_len |= E1000_TXD_CMD_TCP;
2652                 break;
2653         default:
2654                 if (unlikely(net_ratelimit()))
2655                         e_warn(drv, "checksum_partial proto=%x!\n",
2656                                skb->protocol);
2657                 break;
2658         }
2659
2660         css = skb_transport_offset(skb);
2661
2662         i = tx_ring->next_to_use;
2663         buffer_info = &tx_ring->buffer_info[i];
2664         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2665
2666         context_desc->lower_setup.ip_config = 0;
2667         context_desc->upper_setup.tcp_fields.tucss = css;
2668         context_desc->upper_setup.tcp_fields.tucso =
2669                 css + skb->csum_offset;
2670         context_desc->upper_setup.tcp_fields.tucse = 0;
2671         context_desc->tcp_seg_setup.data = 0;
2672         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
2673
2674         buffer_info->time_stamp = jiffies;
2675         buffer_info->next_to_watch = i;
2676
2677         if (unlikely(++i == tx_ring->count)) i = 0;
2678         tx_ring->next_to_use = i;
2679
2680         return true;
2681 }
2682
2683 #define E1000_MAX_TXD_PWR       12
2684 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2685
2686 static int e1000_tx_map(struct e1000_adapter *adapter,
2687                         struct e1000_tx_ring *tx_ring,
2688                         struct sk_buff *skb, unsigned int first,
2689                         unsigned int max_per_txd, unsigned int nr_frags,
2690                         unsigned int mss)
2691 {
2692         struct e1000_hw *hw = &adapter->hw;
2693         struct pci_dev *pdev = adapter->pdev;
2694         struct e1000_buffer *buffer_info;
2695         unsigned int len = skb_headlen(skb);
2696         unsigned int offset = 0, size, count = 0, i;
2697         unsigned int f;
2698
2699         i = tx_ring->next_to_use;
2700
2701         while (len) {
2702                 buffer_info = &tx_ring->buffer_info[i];
2703                 size = min(len, max_per_txd);
2704                 /* Workaround for Controller erratum --
2705                  * descriptor for non-tso packet in a linear SKB that follows a
2706                  * tso gets written back prematurely before the data is fully
2707                  * DMA'd to the controller */
2708                 if (!skb->data_len && tx_ring->last_tx_tso &&
2709                     !skb_is_gso(skb)) {
2710                         tx_ring->last_tx_tso = 0;
2711                         size -= 4;
2712                 }
2713
2714                 /* Workaround for premature desc write-backs
2715                  * in TSO mode.  Append 4-byte sentinel desc */
2716                 if (unlikely(mss && !nr_frags && size == len && size > 8))
2717                         size -= 4;
2718                 /* work-around for errata 10 and it applies
2719                  * to all controllers in PCI-X mode
2720                  * The fix is to make sure that the first descriptor of a
2721                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2722                  */
2723                 if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
2724                                 (size > 2015) && count == 0))
2725                         size = 2015;
2726
2727                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
2728                  * terminating buffers within evenly-aligned dwords. */
2729                 if (unlikely(adapter->pcix_82544 &&
2730                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2731                    size > 4))
2732                         size -= 4;
2733
2734                 buffer_info->length = size;
2735                 /* set time_stamp *before* dma to help avoid a possible race */
2736                 buffer_info->time_stamp = jiffies;
2737                 buffer_info->mapped_as_page = false;
2738                 buffer_info->dma = dma_map_single(&pdev->dev,
2739                                                   skb->data + offset,
2740                                                   size, DMA_TO_DEVICE);
2741                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2742                         goto dma_error;
2743                 buffer_info->next_to_watch = i;
2744
2745                 len -= size;
2746                 offset += size;
2747                 count++;
2748                 if (len) {
2749                         i++;
2750                         if (unlikely(i == tx_ring->count))
2751                                 i = 0;
2752                 }
2753         }
2754
2755         for (f = 0; f < nr_frags; f++) {
2756                 struct skb_frag_struct *frag;
2757
2758                 frag = &skb_shinfo(skb)->frags[f];
2759                 len = frag->size;
2760                 offset = frag->page_offset;
2761
2762                 while (len) {
2763                         i++;
2764                         if (unlikely(i == tx_ring->count))
2765                                 i = 0;
2766
2767                         buffer_info = &tx_ring->buffer_info[i];
2768                         size = min(len, max_per_txd);
2769                         /* Workaround for premature desc write-backs
2770                          * in TSO mode.  Append 4-byte sentinel desc */
2771                         if (unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
2772                                 size -= 4;
2773                         /* Workaround for potential 82544 hang in PCI-X.
2774                          * Avoid terminating buffers within evenly-aligned
2775                          * dwords. */
2776                         if (unlikely(adapter->pcix_82544 &&
2777                             !((unsigned long)(page_to_phys(frag->page) + offset
2778                                               + size - 1) & 4) &&
2779                             size > 4))
2780                                 size -= 4;
2781
2782                         buffer_info->length = size;
2783                         buffer_info->time_stamp = jiffies;
2784                         buffer_info->mapped_as_page = true;
2785                         buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
2786                                                         offset, size,
2787                                                         DMA_TO_DEVICE);
2788                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2789                                 goto dma_error;
2790                         buffer_info->next_to_watch = i;
2791
2792                         len -= size;
2793                         offset += size;
2794                         count++;
2795                 }
2796         }
2797
2798         tx_ring->buffer_info[i].skb = skb;
2799         tx_ring->buffer_info[first].next_to_watch = i;
2800
2801         return count;
2802
2803 dma_error:
2804         dev_err(&pdev->dev, "TX DMA map failed\n");
2805         buffer_info->dma = 0;
2806         if (count)
2807                 count--;
2808
2809         while (count--) {
2810                 if (i==0)
2811                         i += tx_ring->count;
2812                 i--;
2813                 buffer_info = &tx_ring->buffer_info[i];
2814                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2815         }
2816
2817         return 0;
2818 }
2819
2820 static void e1000_tx_queue(struct e1000_adapter *adapter,
2821                            struct e1000_tx_ring *tx_ring, int tx_flags,
2822                            int count)
2823 {
2824         struct e1000_hw *hw = &adapter->hw;
2825         struct e1000_tx_desc *tx_desc = NULL;
2826         struct e1000_buffer *buffer_info;
2827         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2828         unsigned int i;
2829
2830         if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
2831                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2832                              E1000_TXD_CMD_TSE;
2833                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2834
2835                 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
2836                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2837         }
2838
2839         if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
2840                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
2841                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2842         }
2843
2844         if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
2845                 txd_lower |= E1000_TXD_CMD_VLE;
2846                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
2847         }
2848
2849         i = tx_ring->next_to_use;
2850
2851         while (count--) {
2852                 buffer_info = &tx_ring->buffer_info[i];
2853                 tx_desc = E1000_TX_DESC(*tx_ring, i);
2854                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
2855                 tx_desc->lower.data =
2856                         cpu_to_le32(txd_lower | buffer_info->length);
2857                 tx_desc->upper.data = cpu_to_le32(txd_upper);
2858                 if (unlikely(++i == tx_ring->count)) i = 0;
2859         }
2860
2861         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
2862
2863         /* Force memory writes to complete before letting h/w
2864          * know there are new descriptors to fetch.  (Only
2865          * applicable for weak-ordered memory model archs,
2866          * such as IA-64). */
2867         wmb();
2868
2869         tx_ring->next_to_use = i;
2870         writel(i, hw->hw_addr + tx_ring->tdt);
2871         /* we need this if more than one processor can write to our tail
2872          * at a time, it syncronizes IO on IA64/Altix systems */
2873         mmiowb();
2874 }
2875
2876 /**
2877  * 82547 workaround to avoid controller hang in half-duplex environment.
2878  * The workaround is to avoid queuing a large packet that would span
2879  * the internal Tx FIFO ring boundary by notifying the stack to resend
2880  * the packet at a later time.  This gives the Tx FIFO an opportunity to
2881  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
2882  * to the beginning of the Tx FIFO.
2883  **/
2884
2885 #define E1000_FIFO_HDR                  0x10
2886 #define E1000_82547_PAD_LEN             0x3E0
2887
2888 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
2889                                        struct sk_buff *skb)
2890 {
2891         u32 fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
2892         u32 skb_fifo_len = skb->len + E1000_FIFO_HDR;
2893
2894         skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
2895
2896         if (adapter->link_duplex != HALF_DUPLEX)
2897                 goto no_fifo_stall_required;
2898
2899         if (atomic_read(&adapter->tx_fifo_stall))
2900                 return 1;
2901
2902         if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
2903                 atomic_set(&adapter->tx_fifo_stall, 1);
2904                 return 1;
2905         }
2906
2907 no_fifo_stall_required:
2908         adapter->tx_fifo_head += skb_fifo_len;
2909         if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
2910                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
2911         return 0;
2912 }
2913
2914 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
2915 {
2916         struct e1000_adapter *adapter = netdev_priv(netdev);
2917         struct e1000_tx_ring *tx_ring = adapter->tx_ring;
2918
2919         netif_stop_queue(netdev);
2920         /* Herbert's original patch had:
2921          *  smp_mb__after_netif_stop_queue();
2922          * but since that doesn't exist yet, just open code it. */
2923         smp_mb();
2924
2925         /* We need to check again in a case another CPU has just
2926          * made room available. */
2927         if (likely(E1000_DESC_UNUSED(tx_ring) < size))
2928                 return -EBUSY;
2929
2930         /* A reprieve! */
2931         netif_start_queue(netdev);
2932         ++adapter->restart_queue;
2933         return 0;
2934 }
2935
2936 static int e1000_maybe_stop_tx(struct net_device *netdev,
2937                                struct e1000_tx_ring *tx_ring, int size)
2938 {
2939         if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
2940                 return 0;
2941         return __e1000_maybe_stop_tx(netdev, size);
2942 }
2943
2944 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
2945 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
2946                                     struct net_device *netdev)
2947 {
2948         struct e1000_adapter *adapter = netdev_priv(netdev);
2949         struct e1000_hw *hw = &adapter->hw;
2950         struct e1000_tx_ring *tx_ring;
2951         unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
2952         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
2953         unsigned int tx_flags = 0;
2954         unsigned int len = skb_headlen(skb);
2955         unsigned int nr_frags;
2956         unsigned int mss;
2957         int count = 0;
2958         int tso;
2959         unsigned int f;
2960
2961         /* This goes back to the question of how to logically map a tx queue
2962          * to a flow.  Right now, performance is impacted slightly negatively
2963          * if using multiple tx queues.  If the stack breaks away from a
2964          * single qdisc implementation, we can look at this again. */
2965         tx_ring = adapter->tx_ring;
2966
2967         if (unlikely(skb->len <= 0)) {
2968                 dev_kfree_skb_any(skb);
2969                 return NETDEV_TX_OK;
2970         }
2971
2972         mss = skb_shinfo(skb)->gso_size;
2973         /* The controller does a simple calculation to
2974          * make sure there is enough room in the FIFO before
2975          * initiating the DMA for each buffer.  The calc is:
2976          * 4 = ceil(buffer len/mss).  To make sure we don't
2977          * overrun the FIFO, adjust the max buffer len if mss
2978          * drops. */
2979         if (mss) {
2980                 u8 hdr_len;
2981                 max_per_txd = min(mss << 2, max_per_txd);
2982                 max_txd_pwr = fls(max_per_txd) - 1;
2983
2984                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2985                 if (skb->data_len && hdr_len == len) {
2986                         switch (hw->mac_type) {
2987                                 unsigned int pull_size;
2988                         case e1000_82544:
2989                                 /* Make sure we have room to chop off 4 bytes,
2990                                  * and that the end alignment will work out to
2991                                  * this hardware's requirements
2992                                  * NOTE: this is a TSO only workaround
2993                                  * if end byte alignment not correct move us
2994                                  * into the next dword */
2995                                 if ((unsigned long)(skb_tail_pointer(skb) - 1) & 4)
2996                                         break;
2997                                 /* fall through */
2998                                 pull_size = min((unsigned int)4, skb->data_len);
2999                                 if (!__pskb_pull_tail(skb, pull_size)) {
3000                                         e_err(drv, "__pskb_pull_tail "
3001                                               "failed.\n");
3002                                         dev_kfree_skb_any(skb);
3003                                         return NETDEV_TX_OK;
3004                                 }
3005                                 len = skb_headlen(skb);
3006                                 break;
3007                         default:
3008                                 /* do nothing */
3009                                 break;
3010                         }
3011                 }
3012         }
3013
3014         /* reserve a descriptor for the offload context */
3015         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3016                 count++;
3017         count++;
3018
3019         /* Controller Erratum workaround */
3020         if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3021                 count++;
3022
3023         count += TXD_USE_COUNT(len, max_txd_pwr);
3024
3025         if (adapter->pcix_82544)
3026                 count++;
3027
3028         /* work-around for errata 10 and it applies to all controllers
3029          * in PCI-X mode, so add one more descriptor to the count
3030          */
3031         if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
3032                         (len > 2015)))
3033                 count++;
3034
3035         nr_frags = skb_shinfo(skb)->nr_frags;
3036         for (f = 0; f < nr_frags; f++)
3037                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3038                                        max_txd_pwr);
3039         if (adapter->pcix_82544)
3040                 count += nr_frags;
3041
3042         /* need: count + 2 desc gap to keep tail from touching
3043          * head, otherwise try next time */
3044         if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2)))
3045                 return NETDEV_TX_BUSY;
3046
3047         if (unlikely(hw->mac_type == e1000_82547)) {
3048                 if (unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
3049                         netif_stop_queue(netdev);
3050                         if (!test_bit(__E1000_DOWN, &adapter->flags))
3051                                 mod_timer(&adapter->tx_fifo_stall_timer,
3052                                           jiffies + 1);
3053                         return NETDEV_TX_BUSY;
3054                 }
3055         }
3056
3057         if (unlikely(adapter->vlgrp && vlan_tx_tag_present(skb))) {
3058                 tx_flags |= E1000_TX_FLAGS_VLAN;
3059                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3060         }
3061
3062         first = tx_ring->next_to_use;
3063
3064         tso = e1000_tso(adapter, tx_ring, skb);
3065         if (tso < 0) {
3066                 dev_kfree_skb_any(skb);
3067                 return NETDEV_TX_OK;
3068         }
3069
3070         if (likely(tso)) {
3071                 if (likely(hw->mac_type != e1000_82544))
3072                         tx_ring->last_tx_tso = 1;
3073                 tx_flags |= E1000_TX_FLAGS_TSO;
3074         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3075                 tx_flags |= E1000_TX_FLAGS_CSUM;
3076
3077         if (likely(skb->protocol == htons(ETH_P_IP)))
3078                 tx_flags |= E1000_TX_FLAGS_IPV4;
3079
3080         count = e1000_tx_map(adapter, tx_ring, skb, first, max_per_txd,
3081                              nr_frags, mss);
3082
3083         if (count) {
3084                 e1000_tx_queue(adapter, tx_ring, tx_flags, count);
3085                 /* Make sure there is space in the ring for the next send. */
3086                 e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3087
3088         } else {
3089                 dev_kfree_skb_any(skb);
3090                 tx_ring->buffer_info[first].time_stamp = 0;
3091                 tx_ring->next_to_use = first;
3092         }
3093
3094         return NETDEV_TX_OK;
3095 }
3096
3097 /**
3098  * e1000_tx_timeout - Respond to a Tx Hang
3099  * @netdev: network interface device structure
3100  **/
3101
3102 static void e1000_tx_timeout(struct net_device *netdev)
3103 {
3104         struct e1000_adapter *adapter = netdev_priv(netdev);
3105
3106         /* Do the reset outside of interrupt context */
3107         adapter->tx_timeout_count++;
3108         schedule_work(&adapter->reset_task);
3109 }
3110
3111 static void e1000_reset_task(struct work_struct *work)
3112 {
3113         struct e1000_adapter *adapter =
3114                 container_of(work, struct e1000_adapter, reset_task);
3115
3116         e1000_reinit_locked(adapter);
3117 }
3118
3119 /**
3120  * e1000_get_stats - Get System Network Statistics
3121  * @netdev: network interface device structure
3122  *
3123  * Returns the address of the device statistics structure.
3124  * The statistics are actually updated from the timer callback.
3125  **/
3126
3127 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3128 {
3129         /* only return the current stats */
3130         return &netdev->stats;
3131 }
3132
3133 /**
3134  * e1000_change_mtu - Change the Maximum Transfer Unit
3135  * @netdev: network interface device structure
3136  * @new_mtu: new value for maximum frame size
3137  *
3138  * Returns 0 on success, negative on failure
3139  **/
3140
3141 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3142 {
3143         struct e1000_adapter *adapter = netdev_priv(netdev);
3144         struct e1000_hw *hw = &adapter->hw;
3145         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3146
3147         if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3148             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3149                 e_err(probe, "Invalid MTU setting\n");
3150                 return -EINVAL;
3151         }
3152
3153         /* Adapter-specific max frame size limits. */
3154         switch (hw->mac_type) {
3155         case e1000_undefined ... e1000_82542_rev2_1:
3156                 if (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)) {
3157                         e_err(probe, "Jumbo Frames not supported.\n");
3158                         return -EINVAL;
3159                 }
3160                 break;
3161         default:
3162                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3163                 break;
3164         }
3165
3166         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
3167                 msleep(1);
3168         /* e1000_down has a dependency on max_frame_size */
3169         hw->max_frame_size = max_frame;
3170         if (netif_running(netdev))
3171                 e1000_down(adapter);
3172
3173         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3174          * means we reserve 2 more, this pushes us to allocate from the next
3175          * larger slab size.
3176          * i.e. RXBUFFER_2048 --> size-4096 slab
3177          *  however with the new *_jumbo_rx* routines, jumbo receives will use
3178          *  fragmented skbs */
3179
3180         if (max_frame <= E1000_RXBUFFER_2048)
3181                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3182         else
3183 #if (PAGE_SIZE >= E1000_RXBUFFER_16384)
3184                 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3185 #elif (PAGE_SIZE >= E1000_RXBUFFER_4096)
3186                 adapter->rx_buffer_len = PAGE_SIZE;
3187 #endif
3188
3189         /* adjust allocation if LPE protects us, and we aren't using SBP */
3190         if (!hw->tbi_compatibility_on &&
3191             ((max_frame == (ETH_FRAME_LEN + ETH_FCS_LEN)) ||
3192              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3193                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3194
3195         pr_info("%s changing MTU from %d to %d\n",
3196                 netdev->name, netdev->mtu, new_mtu);
3197         netdev->mtu = new_mtu;
3198
3199         if (netif_running(netdev))
3200                 e1000_up(adapter);
3201         else
3202                 e1000_reset(adapter);
3203
3204         clear_bit(__E1000_RESETTING, &adapter->flags);
3205
3206         return 0;
3207 }
3208
3209 /**
3210  * e1000_update_stats - Update the board statistics counters
3211  * @adapter: board private structure
3212  **/
3213
3214 void e1000_update_stats(struct e1000_adapter *adapter)
3215 {
3216         struct net_device *netdev = adapter->netdev;
3217         struct e1000_hw *hw = &adapter->hw;
3218         struct pci_dev *pdev = adapter->pdev;
3219         unsigned long flags;
3220         u16 phy_tmp;
3221
3222 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3223
3224         /*
3225          * Prevent stats update while adapter is being reset, or if the pci
3226          * connection is down.
3227          */
3228         if (adapter->link_speed == 0)
3229                 return;
3230         if (pci_channel_offline(pdev))
3231                 return;
3232
3233         spin_lock_irqsave(&adapter->stats_lock, flags);
3234
3235         /* these counters are modified from e1000_tbi_adjust_stats,
3236          * called from the interrupt context, so they must only
3237          * be written while holding adapter->stats_lock
3238          */
3239
3240         adapter->stats.crcerrs += er32(CRCERRS);
3241         adapter->stats.gprc += er32(GPRC);
3242         adapter->stats.gorcl += er32(GORCL);
3243         adapter->stats.gorch += er32(GORCH);
3244         adapter->stats.bprc += er32(BPRC);
3245         adapter->stats.mprc += er32(MPRC);
3246         adapter->stats.roc += er32(ROC);
3247
3248         adapter->stats.prc64 += er32(PRC64);
3249         adapter->stats.prc127 += er32(PRC127);
3250         adapter->stats.prc255 += er32(PRC255);
3251         adapter->stats.prc511 += er32(PRC511);
3252         adapter->stats.prc1023 += er32(PRC1023);
3253         adapter->stats.prc1522 += er32(PRC1522);
3254
3255         adapter->stats.symerrs += er32(SYMERRS);
3256         adapter->stats.mpc += er32(MPC);
3257         adapter->stats.scc += er32(SCC);
3258         adapter->stats.ecol += er32(ECOL);
3259         adapter->stats.mcc += er32(MCC);
3260         adapter->stats.latecol += er32(LATECOL);
3261         adapter->stats.dc += er32(DC);
3262         adapter->stats.sec += er32(SEC);
3263         adapter->stats.rlec += er32(RLEC);
3264         adapter->stats.xonrxc += er32(XONRXC);
3265         adapter->stats.xontxc += er32(XONTXC);
3266         adapter->stats.xoffrxc += er32(XOFFRXC);
3267         adapter->stats.xofftxc += er32(XOFFTXC);
3268         adapter->stats.fcruc += er32(FCRUC);
3269         adapter->stats.gptc += er32(GPTC);
3270         adapter->stats.gotcl += er32(GOTCL);
3271         adapter->stats.gotch += er32(GOTCH);
3272         adapter->stats.rnbc += er32(RNBC);
3273         adapter->stats.ruc += er32(RUC);
3274         adapter->stats.rfc += er32(RFC);
3275         adapter->stats.rjc += er32(RJC);
3276         adapter->stats.torl += er32(TORL);
3277         adapter->stats.torh += er32(TORH);
3278         adapter->stats.totl += er32(TOTL);
3279         adapter->stats.toth += er32(TOTH);
3280         adapter->stats.tpr += er32(TPR);
3281
3282         adapter->stats.ptc64 += er32(PTC64);
3283         adapter->stats.ptc127 += er32(PTC127);
3284         adapter->stats.ptc255 += er32(PTC255);
3285         adapter->stats.ptc511 += er32(PTC511);
3286         adapter->stats.ptc1023 += er32(PTC1023);
3287         adapter->stats.ptc1522 += er32(PTC1522);
3288
3289         adapter->stats.mptc += er32(MPTC);
3290         adapter->stats.bptc += er32(BPTC);
3291
3292         /* used for adaptive IFS */
3293
3294         hw->tx_packet_delta = er32(TPT);
3295         adapter->stats.tpt += hw->tx_packet_delta;
3296         hw->collision_delta = er32(COLC);
3297         adapter->stats.colc += hw->collision_delta;
3298
3299         if (hw->mac_type >= e1000_82543) {
3300                 adapter->stats.algnerrc += er32(ALGNERRC);
3301                 adapter->stats.rxerrc += er32(RXERRC);
3302                 adapter->stats.tncrs += er32(TNCRS);
3303                 adapter->stats.cexterr += er32(CEXTERR);
3304                 adapter->stats.tsctc += er32(TSCTC);
3305                 adapter->stats.tsctfc += er32(TSCTFC);
3306         }
3307
3308         /* Fill out the OS statistics structure */
3309         netdev->stats.multicast = adapter->stats.mprc;
3310         netdev->stats.collisions = adapter->stats.colc;
3311
3312         /* Rx Errors */
3313
3314         /* RLEC on some newer hardware can be incorrect so build
3315         * our own version based on RUC and ROC */
3316         netdev->stats.rx_errors = adapter->stats.rxerrc +
3317                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3318                 adapter->stats.ruc + adapter->stats.roc +
3319                 adapter->stats.cexterr;
3320         adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3321         netdev->stats.rx_length_errors = adapter->stats.rlerrc;
3322         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3323         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3324         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3325
3326         /* Tx Errors */
3327         adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3328         netdev->stats.tx_errors = adapter->stats.txerrc;
3329         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3330         netdev->stats.tx_window_errors = adapter->stats.latecol;
3331         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3332         if (hw->bad_tx_carr_stats_fd &&
3333             adapter->link_duplex == FULL_DUPLEX) {
3334                 netdev->stats.tx_carrier_errors = 0;
3335                 adapter->stats.tncrs = 0;
3336         }
3337
3338         /* Tx Dropped needs to be maintained elsewhere */
3339
3340         /* Phy Stats */
3341         if (hw->media_type == e1000_media_type_copper) {
3342                 if ((adapter->link_speed == SPEED_1000) &&
3343                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3344                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3345                         adapter->phy_stats.idle_errors += phy_tmp;
3346                 }
3347
3348                 if ((hw->mac_type <= e1000_82546) &&
3349                    (hw->phy_type == e1000_phy_m88) &&
3350                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3351                         adapter->phy_stats.receive_errors += phy_tmp;
3352         }
3353
3354         /* Management Stats */
3355         if (hw->has_smbus) {
3356                 adapter->stats.mgptc += er32(MGTPTC);
3357                 adapter->stats.mgprc += er32(MGTPRC);
3358                 adapter->stats.mgpdc += er32(MGTPDC);
3359         }
3360
3361         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3362 }
3363
3364 /**
3365  * e1000_intr - Interrupt Handler
3366  * @irq: interrupt number
3367  * @data: pointer to a network interface device structure
3368  **/
3369
3370 static irqreturn_t e1000_intr(int irq, void *data)
3371 {
3372         struct net_device *netdev = data;
3373         struct e1000_adapter *adapter = netdev_priv(netdev);
3374         struct e1000_hw *hw = &adapter->hw;
3375         u32 icr = er32(ICR);
3376
3377         if (unlikely((!icr) || test_bit(__E1000_DOWN, &adapter->flags)))
3378                 return IRQ_NONE;  /* Not our interrupt */
3379
3380         if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3381                 hw->get_link_status = 1;
3382                 /* guard against interrupt when we're going down */
3383                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3384                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3385         }
3386
3387         /* disable interrupts, without the synchronize_irq bit */
3388         ew32(IMC, ~0);
3389         E1000_WRITE_FLUSH();
3390
3391         if (likely(napi_schedule_prep(&adapter->napi))) {
3392                 adapter->total_tx_bytes = 0;
3393                 adapter->total_tx_packets = 0;
3394                 adapter->total_rx_bytes = 0;
3395                 adapter->total_rx_packets = 0;
3396                 __napi_schedule(&adapter->napi);
3397         } else {
3398                 /* this really should not happen! if it does it is basically a
3399                  * bug, but not a hard error, so enable ints and continue */
3400                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3401                         e1000_irq_enable(adapter);
3402         }
3403
3404         return IRQ_HANDLED;
3405 }
3406
3407 /**
3408  * e1000_clean - NAPI Rx polling callback
3409  * @adapter: board private structure
3410  **/
3411 static int e1000_clean(struct napi_struct *napi, int budget)
3412 {
3413         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
3414         int tx_clean_complete = 0, work_done = 0;
3415
3416         tx_clean_complete = e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);
3417
3418         adapter->clean_rx(adapter, &adapter->rx_ring[0], &work_done, budget);
3419
3420         if (!tx_clean_complete)
3421                 work_done = budget;
3422
3423         /* If budget not fully consumed, exit the polling mode */
3424         if (work_done < budget) {
3425                 if (likely(adapter->itr_setting & 3))
3426                         e1000_set_itr(adapter);
3427                 napi_complete(napi);
3428                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3429                         e1000_irq_enable(adapter);
3430         }
3431
3432         return work_done;
3433 }
3434
3435 /**
3436  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3437  * @adapter: board private structure
3438  **/
3439 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
3440                                struct e1000_tx_ring *tx_ring)
3441 {
3442         struct e1000_hw *hw = &adapter->hw;
3443         struct net_device *netdev = adapter->netdev;
3444         struct e1000_tx_desc *tx_desc, *eop_desc;
3445         struct e1000_buffer *buffer_info;
3446         unsigned int i, eop;
3447         unsigned int count = 0;
3448         unsigned int total_tx_bytes=0, total_tx_packets=0;
3449
3450         i = tx_ring->next_to_clean;
3451         eop = tx_ring->buffer_info[i].next_to_watch;
3452         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3453
3454         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
3455                (count < tx_ring->count)) {
3456                 bool cleaned = false;
3457                 rmb();  /* read buffer_info after eop_desc */
3458                 for ( ; !cleaned; count++) {
3459                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3460                         buffer_info = &tx_ring->buffer_info[i];
3461                         cleaned = (i == eop);
3462
3463                         if (cleaned) {
3464                                 struct sk_buff *skb = buffer_info->skb;
3465                                 unsigned int segs, bytecount;
3466                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3467                                 /* multiply data chunks by size of headers */
3468                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3469                                             skb->len;
3470                                 total_tx_packets += segs;
3471                                 total_tx_bytes += bytecount;
3472                         }
3473                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3474                         tx_desc->upper.data = 0;
3475
3476                         if (unlikely(++i == tx_ring->count)) i = 0;
3477                 }
3478
3479                 eop = tx_ring->buffer_info[i].next_to_watch;
3480                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3481         }
3482
3483         tx_ring->next_to_clean = i;
3484
3485 #define TX_WAKE_THRESHOLD 32
3486         if (unlikely(count && netif_carrier_ok(netdev) &&
3487                      E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
3488                 /* Make sure that anybody stopping the queue after this
3489                  * sees the new next_to_clean.
3490                  */
3491                 smp_mb();
3492
3493                 if (netif_queue_stopped(netdev) &&
3494                     !(test_bit(__E1000_DOWN, &adapter->flags))) {
3495                         netif_wake_queue(netdev);
3496                         ++adapter->restart_queue;
3497                 }
3498         }
3499
3500         if (adapter->detect_tx_hung) {
3501                 /* Detect a transmit hang in hardware, this serializes the
3502                  * check with the clearing of time_stamp and movement of i */
3503                 adapter->detect_tx_hung = false;
3504                 if (tx_ring->buffer_info[eop].time_stamp &&
3505                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
3506                                (adapter->tx_timeout_factor * HZ)) &&
3507                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
3508
3509                         /* detected Tx unit hang */
3510                         e_err(drv, "Detected Tx Unit Hang\n"
3511                               "  Tx Queue             <%lu>\n"
3512                               "  TDH                  <%x>\n"
3513                               "  TDT                  <%x>\n"
3514                               "  next_to_use          <%x>\n"
3515                               "  next_to_clean        <%x>\n"
3516                               "buffer_info[next_to_clean]\n"
3517                               "  time_stamp           <%lx>\n"
3518                               "  next_to_watch        <%x>\n"
3519                               "  jiffies              <%lx>\n"
3520                               "  next_to_watch.status <%x>\n",
3521                                 (unsigned long)((tx_ring - adapter->tx_ring) /
3522                                         sizeof(struct e1000_tx_ring)),
3523                                 readl(hw->hw_addr + tx_ring->tdh),
3524                                 readl(hw->hw_addr + tx_ring->tdt),
3525                                 tx_ring->next_to_use,
3526                                 tx_ring->next_to_clean,
3527                                 tx_ring->buffer_info[eop].time_stamp,
3528                                 eop,
3529                                 jiffies,
3530                                 eop_desc->upper.fields.status);
3531                         netif_stop_queue(netdev);
3532                 }
3533         }
3534         adapter->total_tx_bytes += total_tx_bytes;
3535         adapter->total_tx_packets += total_tx_packets;
3536         netdev->stats.tx_bytes += total_tx_bytes;
3537         netdev->stats.tx_packets += total_tx_packets;
3538         return (count < tx_ring->count);
3539 }
3540
3541 /**
3542  * e1000_rx_checksum - Receive Checksum Offload for 82543
3543  * @adapter:     board private structure
3544  * @status_err:  receive descriptor status and error fields
3545  * @csum:        receive descriptor csum field
3546  * @sk_buff:     socket buffer with received data
3547  **/
3548
3549 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
3550                               u32 csum, struct sk_buff *skb)
3551 {
3552         struct e1000_hw *hw = &adapter->hw;
3553         u16 status = (u16)status_err;
3554         u8 errors = (u8)(status_err >> 24);
3555         skb->ip_summed = CHECKSUM_NONE;
3556
3557         /* 82543 or newer only */
3558         if (unlikely(hw->mac_type < e1000_82543)) return;
3559         /* Ignore Checksum bit is set */
3560         if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
3561         /* TCP/UDP checksum error bit is set */
3562         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
3563                 /* let the stack verify checksum errors */
3564                 adapter->hw_csum_err++;
3565                 return;
3566         }
3567         /* TCP/UDP Checksum has not been calculated */
3568         if (!(status & E1000_RXD_STAT_TCPCS))
3569                 return;
3570
3571         /* It must be a TCP or UDP packet with a valid checksum */
3572         if (likely(status & E1000_RXD_STAT_TCPCS)) {
3573                 /* TCP checksum is good */
3574                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3575         }
3576         adapter->hw_csum_good++;
3577 }
3578
3579 /**
3580  * e1000_consume_page - helper function
3581  **/
3582 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
3583                                u16 length)
3584 {
3585         bi->page = NULL;
3586         skb->len += length;
3587         skb->data_len += length;
3588         skb->truesize += length;
3589 }
3590
3591 /**
3592  * e1000_receive_skb - helper function to handle rx indications
3593  * @adapter: board private structure
3594  * @status: descriptor status field as written by hardware
3595  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3596  * @skb: pointer to sk_buff to be indicated to stack
3597  */
3598 static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
3599                               __le16 vlan, struct sk_buff *skb)
3600 {
3601         if (unlikely(adapter->vlgrp && (status & E1000_RXD_STAT_VP))) {
3602                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3603                                          le16_to_cpu(vlan) &
3604                                          E1000_RXD_SPC_VLAN_MASK);
3605         } else {
3606                 netif_receive_skb(skb);
3607         }
3608 }
3609
3610 /**
3611  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
3612  * @adapter: board private structure
3613  * @rx_ring: ring to clean
3614  * @work_done: amount of napi work completed this call
3615  * @work_to_do: max amount of work allowed for this call to do
3616  *
3617  * the return value indicates whether actual cleaning was done, there
3618  * is no guarantee that everything was cleaned
3619  */
3620 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
3621                                      struct e1000_rx_ring *rx_ring,
3622                                      int *work_done, int work_to_do)
3623 {
3624         struct e1000_hw *hw = &adapter->hw;
3625         struct net_device *netdev = adapter->netdev;
3626         struct pci_dev *pdev = adapter->pdev;
3627         struct e1000_rx_desc *rx_desc, *next_rxd;
3628         struct e1000_buffer *buffer_info, *next_buffer;
3629         unsigned long irq_flags;
3630         u32 length;
3631         unsigned int i;
3632         int cleaned_count = 0;
3633         bool cleaned = false;
3634         unsigned int total_rx_bytes=0, total_rx_packets=0;
3635
3636         i = rx_ring->next_to_clean;
3637         rx_desc = E1000_RX_DESC(*rx_ring, i);
3638         buffer_info = &rx_ring->buffer_info[i];
3639
3640         while (rx_desc->status & E1000_RXD_STAT_DD) {
3641                 struct sk_buff *skb;
3642                 u8 status;
3643
3644                 if (*work_done >= work_to_do)
3645                         break;
3646                 (*work_done)++;
3647                 rmb(); /* read descriptor and rx_buffer_info after status DD */
3648
3649                 status = rx_desc->status;
3650                 skb = buffer_info->skb;
3651                 buffer_info->skb = NULL;
3652
3653                 if (++i == rx_ring->count) i = 0;
3654                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3655                 prefetch(next_rxd);
3656
3657                 next_buffer = &rx_ring->buffer_info[i];
3658
3659                 cleaned = true;
3660                 cleaned_count++;
3661                 dma_unmap_page(&pdev->dev, buffer_info->dma,
3662                                buffer_info->length, DMA_FROM_DEVICE);
3663                 buffer_info->dma = 0;
3664
3665                 length = le16_to_cpu(rx_desc->length);
3666
3667                 /* errors is only valid for DD + EOP descriptors */
3668                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
3669                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
3670                         u8 last_byte = *(skb->data + length - 1);
3671                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3672                                        last_byte)) {
3673                                 spin_lock_irqsave(&adapter->stats_lock,
3674                                                   irq_flags);
3675                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3676                                                        length, skb->data);
3677                                 spin_unlock_irqrestore(&adapter->stats_lock,
3678                                                        irq_flags);
3679                                 length--;
3680                         } else {
3681                                 /* recycle both page and skb */
3682                                 buffer_info->skb = skb;
3683                                 /* an error means any chain goes out the window
3684                                  * too */
3685                                 if (rx_ring->rx_skb_top)
3686                                         dev_kfree_skb(rx_ring->rx_skb_top);
3687                                 rx_ring->rx_skb_top = NULL;
3688                                 goto next_desc;
3689                         }
3690                 }
3691
3692 #define rxtop rx_ring->rx_skb_top
3693                 if (!(status & E1000_RXD_STAT_EOP)) {
3694                         /* this descriptor is only the beginning (or middle) */
3695                         if (!rxtop) {
3696                                 /* this is the beginning of a chain */
3697                                 rxtop = skb;
3698                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
3699                                                    0, length);
3700                         } else {
3701                                 /* this is the middle of a chain */
3702                                 skb_fill_page_desc(rxtop,
3703                                     skb_shinfo(rxtop)->nr_frags,
3704                                     buffer_info->page, 0, length);
3705                                 /* re-use the skb, only consumed the page */
3706                                 buffer_info->skb = skb;
3707                         }
3708                         e1000_consume_page(buffer_info, rxtop, length);
3709                         goto next_desc;
3710                 } else {
3711                         if (rxtop) {
3712                                 /* end of the chain */
3713                                 skb_fill_page_desc(rxtop,
3714                                     skb_shinfo(rxtop)->nr_frags,
3715                                     buffer_info->page, 0, length);
3716                                 /* re-use the current skb, we only consumed the
3717                                  * page */
3718                                 buffer_info->skb = skb;
3719                                 skb = rxtop;
3720                                 rxtop = NULL;
3721                                 e1000_consume_page(buffer_info, skb, length);
3722                         } else {
3723                                 /* no chain, got EOP, this buf is the packet
3724                                  * copybreak to save the put_page/alloc_page */
3725                                 if (length <= copybreak &&
3726                                     skb_tailroom(skb) >= length) {
3727                                         u8 *vaddr;
3728                                         vaddr = kmap_atomic(buffer_info->page,
3729                                                             KM_SKB_DATA_SOFTIRQ);
3730                                         memcpy(skb_tail_pointer(skb), vaddr, length);
3731                                         kunmap_atomic(vaddr,
3732                                                       KM_SKB_DATA_SOFTIRQ);
3733                                         /* re-use the page, so don't erase
3734                                          * buffer_info->page */
3735                                         skb_put(skb, length);
3736                                 } else {
3737                                         skb_fill_page_desc(skb, 0,
3738                                                            buffer_info->page, 0,
3739                                                            length);
3740                                         e1000_consume_page(buffer_info, skb,
3741                                                            length);
3742                                 }
3743                         }
3744                 }
3745
3746                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
3747                 e1000_rx_checksum(adapter,
3748                                   (u32)(status) |
3749                                   ((u32)(rx_desc->errors) << 24),
3750                                   le16_to_cpu(rx_desc->csum), skb);
3751
3752                 pskb_trim(skb, skb->len - 4);
3753
3754                 /* probably a little skewed due to removing CRC */
3755                 total_rx_bytes += skb->len;
3756                 total_rx_packets++;
3757
3758                 /* eth type trans needs skb->data to point to something */
3759                 if (!pskb_may_pull(skb, ETH_HLEN)) {
3760                         e_err(drv, "pskb_may_pull failed.\n");
3761                         dev_kfree_skb(skb);
3762                         goto next_desc;
3763                 }
3764
3765                 skb->protocol = eth_type_trans(skb, netdev);
3766
3767                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3768
3769 next_desc:
3770                 rx_desc->status = 0;
3771
3772                 /* return some buffers to hardware, one at a time is too slow */
3773                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3774                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3775                         cleaned_count = 0;
3776                 }
3777
3778                 /* use prefetched values */
3779                 rx_desc = next_rxd;
3780                 buffer_info = next_buffer;
3781         }
3782         rx_ring->next_to_clean = i;
3783
3784         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3785         if (cleaned_count)
3786                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3787
3788         adapter->total_rx_packets += total_rx_packets;
3789         adapter->total_rx_bytes += total_rx_bytes;
3790         netdev->stats.rx_bytes += total_rx_bytes;
3791         netdev->stats.rx_packets += total_rx_packets;
3792         return cleaned;
3793 }
3794
3795 /*
3796  * this should improve performance for small packets with large amounts
3797  * of reassembly being done in the stack
3798  */
3799 static void e1000_check_copybreak(struct net_device *netdev,
3800                                  struct e1000_buffer *buffer_info,
3801                                  u32 length, struct sk_buff **skb)
3802 {
3803         struct sk_buff *new_skb;
3804
3805         if (length > copybreak)
3806                 return;
3807
3808         new_skb = netdev_alloc_skb_ip_align(netdev, length);
3809         if (!new_skb)
3810                 return;
3811
3812         skb_copy_to_linear_data_offset(new_skb, -NET_IP_ALIGN,
3813                                        (*skb)->data - NET_IP_ALIGN,
3814                                        length + NET_IP_ALIGN);
3815         /* save the skb in buffer_info as good */
3816         buffer_info->skb = *skb;
3817         *skb = new_skb;
3818 }
3819
3820 /**
3821  * e1000_clean_rx_irq - Send received data up the network stack; legacy
3822  * @adapter: board private structure
3823  * @rx_ring: ring to clean
3824  * @work_done: amount of napi work completed this call
3825  * @work_to_do: max amount of work allowed for this call to do
3826  */
3827 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
3828                                struct e1000_rx_ring *rx_ring,
3829                                int *work_done, int work_to_do)
3830 {
3831         struct e1000_hw *hw = &adapter->hw;
3832         struct net_device *netdev = adapter->netdev;
3833         struct pci_dev *pdev = adapter->pdev;
3834         struct e1000_rx_desc *rx_desc, *next_rxd;
3835         struct e1000_buffer *buffer_info, *next_buffer;
3836         unsigned long flags;
3837         u32 length;
3838         unsigned int i;
3839         int cleaned_count = 0;
3840         bool cleaned = false;
3841         unsigned int total_rx_bytes=0, total_rx_packets=0;
3842
3843         i = rx_ring->next_to_clean;
3844         rx_desc = E1000_RX_DESC(*rx_ring, i);
3845         buffer_info = &rx_ring->buffer_info[i];
3846
3847         while (rx_desc->status & E1000_RXD_STAT_DD) {
3848                 struct sk_buff *skb;
3849                 u8 status;
3850
3851                 if (*work_done >= work_to_do)
3852                         break;
3853                 (*work_done)++;
3854                 rmb(); /* read descriptor and rx_buffer_info after status DD */
3855
3856                 status = rx_desc->status;
3857                 skb = buffer_info->skb;
3858                 buffer_info->skb = NULL;
3859
3860                 prefetch(skb->data - NET_IP_ALIGN);
3861
3862                 if (++i == rx_ring->count) i = 0;
3863                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3864                 prefetch(next_rxd);
3865
3866                 next_buffer = &rx_ring->buffer_info[i];
3867
3868                 cleaned = true;
3869                 cleaned_count++;
3870                 dma_unmap_single(&pdev->dev, buffer_info->dma,
3871                                  buffer_info->length, DMA_FROM_DEVICE);
3872                 buffer_info->dma = 0;
3873
3874                 length = le16_to_cpu(rx_desc->length);
3875                 /* !EOP means multiple descriptors were used to store a single
3876                  * packet, if thats the case we need to toss it.  In fact, we
3877                  * to toss every packet with the EOP bit clear and the next
3878                  * frame that _does_ have the EOP bit set, as it is by
3879                  * definition only a frame fragment
3880                  */
3881                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
3882                         adapter->discarding = true;
3883
3884                 if (adapter->discarding) {
3885                         /* All receives must fit into a single buffer */
3886                         e_dbg("Receive packet consumed multiple buffers\n");
3887                         /* recycle */
3888                         buffer_info->skb = skb;
3889                         if (status & E1000_RXD_STAT_EOP)
3890                                 adapter->discarding = false;
3891                         goto next_desc;
3892                 }
3893
3894                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
3895                         u8 last_byte = *(skb->data + length - 1);
3896                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3897                                        last_byte)) {
3898                                 spin_lock_irqsave(&adapter->stats_lock, flags);
3899                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3900                                                        length, skb->data);
3901                                 spin_unlock_irqrestore(&adapter->stats_lock,
3902                                                        flags);
3903                                 length--;
3904                         } else {
3905                                 /* recycle */
3906                                 buffer_info->skb = skb;
3907                                 goto next_desc;
3908                         }
3909                 }
3910
3911                 /* adjust length to remove Ethernet CRC, this must be
3912                  * done after the TBI_ACCEPT workaround above */
3913                 length -= 4;
3914
3915                 /* probably a little skewed due to removing CRC */
3916                 total_rx_bytes += length;
3917                 total_rx_packets++;
3918
3919                 e1000_check_copybreak(netdev, buffer_info, length, &skb);
3920
3921                 skb_put(skb, length);
3922
3923                 /* Receive Checksum Offload */
3924                 e1000_rx_checksum(adapter,
3925                                   (u32)(status) |
3926                                   ((u32)(rx_desc->errors) << 24),
3927                                   le16_to_cpu(rx_desc->csum), skb);
3928
3929                 skb->protocol = eth_type_trans(skb, netdev);
3930
3931                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3932
3933 next_desc:
3934                 rx_desc->status = 0;
3935
3936                 /* return some buffers to hardware, one at a time is too slow */
3937                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3938                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3939                         cleaned_count = 0;
3940                 }
3941
3942                 /* use prefetched values */
3943                 rx_desc = next_rxd;
3944                 buffer_info = next_buffer;
3945         }
3946         rx_ring->next_to_clean = i;
3947
3948         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3949         if (cleaned_count)
3950                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3951
3952         adapter->total_rx_packets += total_rx_packets;
3953         adapter->total_rx_bytes += total_rx_bytes;
3954         netdev->stats.rx_bytes += total_rx_bytes;
3955         netdev->stats.rx_packets += total_rx_packets;
3956         return cleaned;
3957 }
3958
3959 /**
3960  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
3961  * @adapter: address of board private structure
3962  * @rx_ring: pointer to receive ring structure
3963  * @cleaned_count: number of buffers to allocate this pass
3964  **/
3965
3966 static void
3967 e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
3968                              struct e1000_rx_ring *rx_ring, int cleaned_count)
3969 {
3970         struct net_device *netdev = adapter->netdev;
3971         struct pci_dev *pdev = adapter->pdev;
3972         struct e1000_rx_desc *rx_desc;
3973         struct e1000_buffer *buffer_info;
3974         struct sk_buff *skb;
3975         unsigned int i;
3976         unsigned int bufsz = 256 - 16 /*for skb_reserve */ ;
3977
3978         i = rx_ring->next_to_use;
3979         buffer_info = &rx_ring->buffer_info[i];
3980
3981         while (cleaned_count--) {
3982                 skb = buffer_info->skb;
3983                 if (skb) {
3984                         skb_trim(skb, 0);
3985                         goto check_page;
3986                 }
3987
3988                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
3989                 if (unlikely(!skb)) {
3990                         /* Better luck next round */
3991                         adapter->alloc_rx_buff_failed++;
3992                         break;
3993                 }
3994
3995                 /* Fix for errata 23, can't cross 64kB boundary */
3996                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3997                         struct sk_buff *oldskb = skb;
3998                         e_err(rx_err, "skb align check failed: %u bytes at "
3999                               "%p\n", bufsz, skb->data);
4000                         /* Try again, without freeing the previous */
4001                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4002                         /* Failed allocation, critical failure */
4003                         if (!skb) {
4004                                 dev_kfree_skb(oldskb);
4005                                 adapter->alloc_rx_buff_failed++;
4006                                 break;
4007                         }
4008
4009                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4010                                 /* give up */
4011                                 dev_kfree_skb(skb);
4012                                 dev_kfree_skb(oldskb);
4013                                 break; /* while (cleaned_count--) */
4014                         }
4015
4016                         /* Use new allocation */
4017                         dev_kfree_skb(oldskb);
4018                 }
4019                 buffer_info->skb = skb;
4020                 buffer_info->length = adapter->rx_buffer_len;
4021 check_page:
4022                 /* allocate a new page if necessary */
4023                 if (!buffer_info->page) {
4024                         buffer_info->page = alloc_page(GFP_ATOMIC);
4025                         if (unlikely(!buffer_info->page)) {
4026                                 adapter->alloc_rx_buff_failed++;
4027                                 break;
4028                         }
4029                 }
4030
4031                 if (!buffer_info->dma) {
4032                         buffer_info->dma = dma_map_page(&pdev->dev,
4033                                                         buffer_info->page, 0,
4034                                                         buffer_info->length,
4035                                                         DMA_FROM_DEVICE);
4036                         if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4037                                 put_page(buffer_info->page);
4038                                 dev_kfree_skb(skb);
4039                                 buffer_info->page = NULL;
4040                                 buffer_info->skb = NULL;
4041                                 buffer_info->dma = 0;
4042                                 adapter->alloc_rx_buff_failed++;
4043                                 break; /* while !buffer_info->skb */
4044                         }
4045                 }
4046
4047                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4048                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4049
4050                 if (unlikely(++i == rx_ring->count))
4051                         i = 0;
4052                 buffer_info = &rx_ring->buffer_info[i];
4053         }
4054
4055         if (likely(rx_ring->next_to_use != i)) {
4056                 rx_ring->next_to_use = i;
4057                 if (unlikely(i-- == 0))
4058                         i = (rx_ring->count - 1);
4059
4060                 /* Force memory writes to complete before letting h/w
4061                  * know there are new descriptors to fetch.  (Only
4062                  * applicable for weak-ordered memory model archs,
4063                  * such as IA-64). */
4064                 wmb();
4065                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4066         }
4067 }
4068
4069 /**
4070  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4071  * @adapter: address of board private structure
4072  **/
4073
4074 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4075                                    struct e1000_rx_ring *rx_ring,
4076                                    int cleaned_count)
4077 {
4078         struct e1000_hw *hw = &adapter->hw;
4079         struct net_device *netdev = adapter->netdev;
4080         struct pci_dev *pdev = adapter->pdev;
4081         struct e1000_rx_desc *rx_desc;
4082         struct e1000_buffer *buffer_info;
4083         struct sk_buff *skb;
4084         unsigned int i;
4085         unsigned int bufsz = adapter->rx_buffer_len;
4086
4087         i = rx_ring->next_to_use;
4088         buffer_info = &rx_ring->buffer_info[i];
4089
4090         while (cleaned_count--) {
4091                 skb = buffer_info->skb;
4092                 if (skb) {
4093                         skb_trim(skb, 0);
4094                         goto map_skb;
4095                 }
4096
4097                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4098                 if (unlikely(!skb)) {
4099                         /* Better luck next round */
4100                         adapter->alloc_rx_buff_failed++;
4101                         break;
4102                 }
4103
4104                 /* Fix for errata 23, can't cross 64kB boundary */
4105                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4106                         struct sk_buff *oldskb = skb;
4107                         e_err(rx_err, "skb align check failed: %u bytes at "
4108                               "%p\n", bufsz, skb->data);
4109                         /* Try again, without freeing the previous */
4110                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4111                         /* Failed allocation, critical failure */
4112                         if (!skb) {
4113                                 dev_kfree_skb(oldskb);
4114                                 adapter->alloc_rx_buff_failed++;
4115                                 break;
4116                         }
4117
4118                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4119                                 /* give up */
4120                                 dev_kfree_skb(skb);
4121                                 dev_kfree_skb(oldskb);
4122                                 adapter->alloc_rx_buff_failed++;
4123                                 break; /* while !buffer_info->skb */
4124                         }
4125
4126                         /* Use new allocation */
4127                         dev_kfree_skb(oldskb);
4128                 }
4129                 buffer_info->skb = skb;
4130                 buffer_info->length = adapter->rx_buffer_len;
4131 map_skb:
4132                 buffer_info->dma = dma_map_single(&pdev->dev,
4133                                                   skb->data,
4134                                                   buffer_info->length,
4135                                                   DMA_FROM_DEVICE);
4136                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4137                         dev_kfree_skb(skb);
4138                         buffer_info->skb = NULL;
4139                         buffer_info->dma = 0;
4140                         adapter->alloc_rx_buff_failed++;
4141                         break; /* while !buffer_info->skb */
4142                 }
4143
4144                 /*
4145                  * XXX if it was allocated cleanly it will never map to a
4146                  * boundary crossing
4147                  */
4148
4149                 /* Fix for errata 23, can't cross 64kB boundary */
4150                 if (!e1000_check_64k_bound(adapter,
4151                                         (void *)(unsigned long)buffer_info->dma,
4152                                         adapter->rx_buffer_len)) {
4153                         e_err(rx_err, "dma align check failed: %u bytes at "
4154                               "%p\n", adapter->rx_buffer_len,
4155                               (void *)(unsigned long)buffer_info->dma);
4156                         dev_kfree_skb(skb);
4157                         buffer_info->skb = NULL;
4158
4159                         dma_unmap_single(&pdev->dev, buffer_info->dma,
4160                                          adapter->rx_buffer_len,
4161                                          DMA_FROM_DEVICE);
4162                         buffer_info->dma = 0;
4163
4164                         adapter->alloc_rx_buff_failed++;
4165                         break; /* while !buffer_info->skb */
4166                 }
4167                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4168                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4169
4170                 if (unlikely(++i == rx_ring->count))
4171                         i = 0;
4172                 buffer_info = &rx_ring->buffer_info[i];
4173         }
4174
4175         if (likely(rx_ring->next_to_use != i)) {
4176                 rx_ring->next_to_use = i;
4177                 if (unlikely(i-- == 0))
4178                         i = (rx_ring->count - 1);
4179
4180                 /* Force memory writes to complete before letting h/w
4181                  * know there are new descriptors to fetch.  (Only
4182                  * applicable for weak-ordered memory model archs,
4183                  * such as IA-64). */
4184                 wmb();
4185                 writel(i, hw->hw_addr + rx_ring->rdt);
4186         }
4187 }
4188
4189 /**
4190  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4191  * @adapter:
4192  **/
4193
4194 static void e1000_smartspeed(struct e1000_adapter *adapter)
4195 {
4196         struct e1000_hw *hw = &adapter->hw;
4197         u16 phy_status;
4198         u16 phy_ctrl;
4199
4200         if ((hw->phy_type != e1000_phy_igp) || !hw->autoneg ||
4201            !(hw->autoneg_advertised & ADVERTISE_1000_FULL))
4202                 return;
4203
4204         if (adapter->smartspeed == 0) {
4205                 /* If Master/Slave config fault is asserted twice,
4206                  * we assume back-to-back */
4207                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4208                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4209                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4210                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4211                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4212                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4213                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4214                         e1000_write_phy_reg(hw, PHY_1000T_CTRL,
4215                                             phy_ctrl);
4216                         adapter->smartspeed++;
4217                         if (!e1000_phy_setup_autoneg(hw) &&
4218                            !e1000_read_phy_reg(hw, PHY_CTRL,
4219                                                &phy_ctrl)) {
4220                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4221                                              MII_CR_RESTART_AUTO_NEG);
4222                                 e1000_write_phy_reg(hw, PHY_CTRL,
4223                                                     phy_ctrl);
4224                         }
4225                 }
4226                 return;
4227         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4228                 /* If still no link, perhaps using 2/3 pair cable */
4229                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4230                 phy_ctrl |= CR_1000T_MS_ENABLE;
4231                 e1000_write_phy_reg(hw, PHY_1000T_CTRL, phy_ctrl);
4232                 if (!e1000_phy_setup_autoneg(hw) &&
4233                    !e1000_read_phy_reg(hw, PHY_CTRL, &phy_ctrl)) {
4234                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4235                                      MII_CR_RESTART_AUTO_NEG);
4236                         e1000_write_phy_reg(hw, PHY_CTRL, phy_ctrl);
4237                 }
4238         }
4239         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4240         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4241                 adapter->smartspeed = 0;
4242 }
4243
4244 /**
4245  * e1000_ioctl -
4246  * @netdev:
4247  * @ifreq:
4248  * @cmd:
4249  **/
4250
4251 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4252 {
4253         switch (cmd) {
4254         case SIOCGMIIPHY:
4255         case SIOCGMIIREG:
4256         case SIOCSMIIREG:
4257                 return e1000_mii_ioctl(netdev, ifr, cmd);
4258         default:
4259                 return -EOPNOTSUPP;
4260         }
4261 }
4262
4263 /**
4264  * e1000_mii_ioctl -
4265  * @netdev:
4266  * @ifreq:
4267  * @cmd:
4268  **/
4269
4270 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4271                            int cmd)
4272 {
4273         struct e1000_adapter *adapter = netdev_priv(netdev);
4274         struct e1000_hw *hw = &adapter->hw;
4275         struct mii_ioctl_data *data = if_mii(ifr);
4276         int retval;
4277         u16 mii_reg;
4278         u16 spddplx;
4279         unsigned long flags;
4280
4281         if (hw->media_type != e1000_media_type_copper)
4282                 return -EOPNOTSUPP;
4283
4284         switch (cmd) {
4285         case SIOCGMIIPHY:
4286                 data->phy_id = hw->phy_addr;
4287                 break;
4288         case SIOCGMIIREG:
4289                 spin_lock_irqsave(&adapter->stats_lock, flags);
4290                 if (e1000_read_phy_reg(hw, data->reg_num & 0x1F,
4291                                    &data->val_out)) {
4292                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4293                         return -EIO;
4294                 }
4295                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4296                 break;
4297         case SIOCSMIIREG:
4298                 if (data->reg_num & ~(0x1F))
4299                         return -EFAULT;
4300                 mii_reg = data->val_in;
4301                 spin_lock_irqsave(&adapter->stats_lock, flags);
4302                 if (e1000_write_phy_reg(hw, data->reg_num,
4303                                         mii_reg)) {
4304                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4305                         return -EIO;
4306                 }
4307                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4308                 if (hw->media_type == e1000_media_type_copper) {
4309                         switch (data->reg_num) {
4310                         case PHY_CTRL:
4311                                 if (mii_reg & MII_CR_POWER_DOWN)
4312                                         break;
4313                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4314                                         hw->autoneg = 1;
4315                                         hw->autoneg_advertised = 0x2F;
4316                                 } else {
4317                                         if (mii_reg & 0x40)
4318                                                 spddplx = SPEED_1000;
4319                                         else if (mii_reg & 0x2000)
4320                                                 spddplx = SPEED_100;
4321                                         else
4322                                                 spddplx = SPEED_10;
4323                                         spddplx += (mii_reg & 0x100)
4324                                                    ? DUPLEX_FULL :
4325                                                    DUPLEX_HALF;
4326                                         retval = e1000_set_spd_dplx(adapter,
4327                                                                     spddplx);
4328                                         if (retval)
4329                                                 return retval;
4330                                 }
4331                                 if (netif_running(adapter->netdev))
4332                                         e1000_reinit_locked(adapter);
4333                                 else
4334                                         e1000_reset(adapter);
4335                                 break;
4336                         case M88E1000_PHY_SPEC_CTRL:
4337                         case M88E1000_EXT_PHY_SPEC_CTRL:
4338                                 if (e1000_phy_reset(hw))
4339                                         return -EIO;
4340                                 break;
4341                         }
4342                 } else {
4343                         switch (data->reg_num) {
4344                         case PHY_CTRL:
4345                                 if (mii_reg & MII_CR_POWER_DOWN)
4346                                         break;
4347                                 if (netif_running(adapter->netdev))
4348                                         e1000_reinit_locked(adapter);
4349                                 else
4350                                         e1000_reset(adapter);
4351                                 break;
4352                         }
4353                 }
4354                 break;
4355         default:
4356                 return -EOPNOTSUPP;
4357         }
4358         return E1000_SUCCESS;
4359 }
4360
4361 void e1000_pci_set_mwi(struct e1000_hw *hw)
4362 {
4363         struct e1000_adapter *adapter = hw->back;
4364         int ret_val = pci_set_mwi(adapter->pdev);
4365
4366         if (ret_val)
4367                 e_err(probe, "Error in setting MWI\n");
4368 }
4369
4370 void e1000_pci_clear_mwi(struct e1000_hw *hw)
4371 {
4372         struct e1000_adapter *adapter = hw->back;
4373
4374         pci_clear_mwi(adapter->pdev);
4375 }
4376
4377 int e1000_pcix_get_mmrbc(struct e1000_hw *hw)
4378 {
4379         struct e1000_adapter *adapter = hw->back;
4380         return pcix_get_mmrbc(adapter->pdev);
4381 }
4382
4383 void e1000_pcix_set_mmrbc(struct e1000_hw *hw, int mmrbc)
4384 {
4385         struct e1000_adapter *adapter = hw->back;
4386         pcix_set_mmrbc(adapter->pdev, mmrbc);
4387 }
4388
4389 void e1000_io_write(struct e1000_hw *hw, unsigned long port, u32 value)
4390 {
4391         outl(value, port);
4392 }
4393
4394 static void e1000_vlan_rx_register(struct net_device *netdev,
4395                                    struct vlan_group *grp)
4396 {
4397         struct e1000_adapter *adapter = netdev_priv(netdev);
4398         struct e1000_hw *hw = &adapter->hw;
4399         u32 ctrl, rctl;
4400
4401         if (!test_bit(__E1000_DOWN, &adapter->flags))
4402                 e1000_irq_disable(adapter);
4403         adapter->vlgrp = grp;
4404
4405         if (grp) {
4406                 /* enable VLAN tag insert/strip */
4407                 ctrl = er32(CTRL);
4408                 ctrl |= E1000_CTRL_VME;
4409                 ew32(CTRL, ctrl);
4410
4411                 /* enable VLAN receive filtering */
4412                 rctl = er32(RCTL);
4413                 rctl &= ~E1000_RCTL_CFIEN;
4414                 if (!(netdev->flags & IFF_PROMISC))
4415                         rctl |= E1000_RCTL_VFE;
4416                 ew32(RCTL, rctl);
4417                 e1000_update_mng_vlan(adapter);
4418         } else {
4419                 /* disable VLAN tag insert/strip */
4420                 ctrl = er32(CTRL);
4421                 ctrl &= ~E1000_CTRL_VME;
4422                 ew32(CTRL, ctrl);
4423
4424                 /* disable VLAN receive filtering */
4425                 rctl = er32(RCTL);
4426                 rctl &= ~E1000_RCTL_VFE;
4427                 ew32(RCTL, rctl);
4428
4429                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
4430                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4431                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4432                 }
4433         }
4434
4435         if (!test_bit(__E1000_DOWN, &adapter->flags))
4436                 e1000_irq_enable(adapter);
4437 }
4438
4439 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4440 {
4441         struct e1000_adapter *adapter = netdev_priv(netdev);
4442         struct e1000_hw *hw = &adapter->hw;
4443         u32 vfta, index;
4444
4445         if ((hw->mng_cookie.status &
4446              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4447             (vid == adapter->mng_vlan_id))
4448                 return;
4449         /* add VID to filter table */
4450         index = (vid >> 5) & 0x7F;
4451         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4452         vfta |= (1 << (vid & 0x1F));
4453         e1000_write_vfta(hw, index, vfta);
4454 }
4455
4456 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
4457 {
4458         struct e1000_adapter *adapter = netdev_priv(netdev);
4459         struct e1000_hw *hw = &adapter->hw;
4460         u32 vfta, index;
4461
4462         if (!test_bit(__E1000_DOWN, &adapter->flags))
4463                 e1000_irq_disable(adapter);
4464         vlan_group_set_device(adapter->vlgrp, vid, NULL);
4465         if (!test_bit(__E1000_DOWN, &adapter->flags))
4466                 e1000_irq_enable(adapter);
4467
4468         /* remove VID from filter table */
4469         index = (vid >> 5) & 0x7F;
4470         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4471         vfta &= ~(1 << (vid & 0x1F));
4472         e1000_write_vfta(hw, index, vfta);
4473 }
4474
4475 static void e1000_restore_vlan(struct e1000_adapter *adapter)
4476 {
4477         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
4478
4479         if (adapter->vlgrp) {
4480                 u16 vid;
4481                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
4482                         if (!vlan_group_get_device(adapter->vlgrp, vid))
4483                                 continue;
4484                         e1000_vlan_rx_add_vid(adapter->netdev, vid);
4485                 }
4486         }
4487 }
4488
4489 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx)
4490 {
4491         struct e1000_hw *hw = &adapter->hw;
4492
4493         hw->autoneg = 0;
4494
4495         /* Fiber NICs only allow 1000 gbps Full duplex */
4496         if ((hw->media_type == e1000_media_type_fiber) &&
4497                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
4498                 e_err(probe, "Unsupported Speed/Duplex configuration\n");
4499                 return -EINVAL;
4500         }
4501
4502         switch (spddplx) {
4503         case SPEED_10 + DUPLEX_HALF:
4504                 hw->forced_speed_duplex = e1000_10_half;
4505                 break;
4506         case SPEED_10 + DUPLEX_FULL:
4507                 hw->forced_speed_duplex = e1000_10_full;
4508                 break;
4509         case SPEED_100 + DUPLEX_HALF:
4510                 hw->forced_speed_duplex = e1000_100_half;
4511                 break;
4512         case SPEED_100 + DUPLEX_FULL:
4513                 hw->forced_speed_duplex = e1000_100_full;
4514                 break;
4515         case SPEED_1000 + DUPLEX_FULL:
4516                 hw->autoneg = 1;
4517                 hw->autoneg_advertised = ADVERTISE_1000_FULL;
4518                 break;
4519         case SPEED_1000 + DUPLEX_HALF: /* not supported */
4520         default:
4521                 e_err(probe, "Unsupported Speed/Duplex configuration\n");
4522                 return -EINVAL;
4523         }
4524         return 0;
4525 }
4526
4527 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4528 {
4529         struct net_device *netdev = pci_get_drvdata(pdev);
4530         struct e1000_adapter *adapter = netdev_priv(netdev);
4531         struct e1000_hw *hw = &adapter->hw;
4532         u32 ctrl, ctrl_ext, rctl, status;
4533         u32 wufc = adapter->wol;
4534 #ifdef CONFIG_PM
4535         int retval = 0;
4536 #endif
4537
4538         netif_device_detach(netdev);
4539
4540         if (netif_running(netdev)) {
4541                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
4542                 e1000_down(adapter);
4543         }
4544
4545 #ifdef CONFIG_PM
4546         retval = pci_save_state(pdev);
4547         if (retval)
4548                 return retval;
4549 #endif
4550
4551         status = er32(STATUS);
4552         if (status & E1000_STATUS_LU)
4553                 wufc &= ~E1000_WUFC_LNKC;
4554
4555         if (wufc) {
4556                 e1000_setup_rctl(adapter);
4557                 e1000_set_rx_mode(netdev);
4558
4559                 /* turn on all-multi mode if wake on multicast is enabled */
4560                 if (wufc & E1000_WUFC_MC) {
4561                         rctl = er32(RCTL);
4562                         rctl |= E1000_RCTL_MPE;
4563                         ew32(RCTL, rctl);
4564                 }
4565
4566                 if (hw->mac_type >= e1000_82540) {
4567                         ctrl = er32(CTRL);
4568                         /* advertise wake from D3Cold */
4569                         #define E1000_CTRL_ADVD3WUC 0x00100000
4570                         /* phy power management enable */
4571                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4572                         ctrl |= E1000_CTRL_ADVD3WUC |
4573                                 E1000_CTRL_EN_PHY_PWR_MGMT;
4574                         ew32(CTRL, ctrl);
4575                 }
4576
4577                 if (hw->media_type == e1000_media_type_fiber ||
4578                     hw->media_type == e1000_media_type_internal_serdes) {
4579                         /* keep the laser running in D3 */
4580                         ctrl_ext = er32(CTRL_EXT);
4581                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4582                         ew32(CTRL_EXT, ctrl_ext);
4583                 }
4584
4585                 ew32(WUC, E1000_WUC_PME_EN);
4586                 ew32(WUFC, wufc);
4587         } else {
4588                 ew32(WUC, 0);
4589                 ew32(WUFC, 0);
4590         }
4591
4592         e1000_release_manageability(adapter);
4593
4594         *enable_wake = !!wufc;
4595
4596         /* make sure adapter isn't asleep if manageability is enabled */
4597         if (adapter->en_mng_pt)
4598                 *enable_wake = true;
4599
4600         if (netif_running(netdev))
4601                 e1000_free_irq(adapter);
4602
4603         pci_disable_device(pdev);
4604
4605         return 0;
4606 }
4607
4608 #ifdef CONFIG_PM
4609 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4610 {
4611         int retval;
4612         bool wake;
4613
4614         retval = __e1000_shutdown(pdev, &wake);
4615         if (retval)
4616                 return retval;
4617
4618         if (wake) {
4619                 pci_prepare_to_sleep(pdev);
4620         } else {
4621                 pci_wake_from_d3(pdev, false);
4622                 pci_set_power_state(pdev, PCI_D3hot);
4623         }
4624
4625         return 0;
4626 }
4627
4628 static int e1000_resume(struct pci_dev *pdev)
4629 {
4630         struct net_device *netdev = pci_get_drvdata(pdev);
4631         struct e1000_adapter *adapter = netdev_priv(netdev);
4632         struct e1000_hw *hw = &adapter->hw;
4633         u32 err;
4634
4635         pci_set_power_state(pdev, PCI_D0);
4636         pci_restore_state(pdev);
4637         pci_save_state(pdev);
4638
4639         if (adapter->need_ioport)
4640                 err = pci_enable_device(pdev);
4641         else
4642                 err = pci_enable_device_mem(pdev);
4643         if (err) {
4644                 pr_err("Cannot enable PCI device from suspend\n");
4645                 return err;
4646         }
4647         pci_set_master(pdev);
4648
4649         pci_enable_wake(pdev, PCI_D3hot, 0);
4650         pci_enable_wake(pdev, PCI_D3cold, 0);
4651
4652         if (netif_running(netdev)) {
4653                 err = e1000_request_irq(adapter);
4654                 if (err)
4655                         return err;
4656         }
4657
4658         e1000_power_up_phy(adapter);
4659         e1000_reset(adapter);
4660         ew32(WUS, ~0);
4661
4662         e1000_init_manageability(adapter);
4663
4664         if (netif_running(netdev))
4665                 e1000_up(adapter);
4666
4667         netif_device_attach(netdev);
4668
4669         return 0;
4670 }
4671 #endif
4672
4673 static void e1000_shutdown(struct pci_dev *pdev)
4674 {
4675         bool wake;
4676
4677         __e1000_shutdown(pdev, &wake);
4678
4679         if (system_state == SYSTEM_POWER_OFF) {
4680                 pci_wake_from_d3(pdev, wake);
4681                 pci_set_power_state(pdev, PCI_D3hot);
4682         }
4683 }
4684
4685 #ifdef CONFIG_NET_POLL_CONTROLLER
4686 /*
4687  * Polling 'interrupt' - used by things like netconsole to send skbs
4688  * without having to re-enable interrupts. It's not called while
4689  * the interrupt routine is executing.
4690  */
4691 static void e1000_netpoll(struct net_device *netdev)
4692 {
4693         struct e1000_adapter *adapter = netdev_priv(netdev);
4694
4695         disable_irq(adapter->pdev->irq);
4696         e1000_intr(adapter->pdev->irq, netdev);
4697         enable_irq(adapter->pdev->irq);
4698 }
4699 #endif
4700
4701 /**
4702  * e1000_io_error_detected - called when PCI error is detected
4703  * @pdev: Pointer to PCI device
4704  * @state: The current pci connection state
4705  *
4706  * This function is called after a PCI bus error affecting
4707  * this device has been detected.
4708  */
4709 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4710                                                 pci_channel_state_t state)
4711 {
4712         struct net_device *netdev = pci_get_drvdata(pdev);
4713         struct e1000_adapter *adapter = netdev_priv(netdev);
4714
4715         netif_device_detach(netdev);
4716
4717         if (state == pci_channel_io_perm_failure)
4718                 return PCI_ERS_RESULT_DISCONNECT;
4719
4720         if (netif_running(netdev))
4721                 e1000_down(adapter);
4722         pci_disable_device(pdev);
4723
4724         /* Request a slot slot reset. */
4725         return PCI_ERS_RESULT_NEED_RESET;
4726 }
4727
4728 /**
4729  * e1000_io_slot_reset - called after the pci bus has been reset.
4730  * @pdev: Pointer to PCI device
4731  *
4732  * Restart the card from scratch, as if from a cold-boot. Implementation
4733  * resembles the first-half of the e1000_resume routine.
4734  */
4735 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4736 {
4737         struct net_device *netdev = pci_get_drvdata(pdev);
4738         struct e1000_adapter *adapter = netdev_priv(netdev);
4739         struct e1000_hw *hw = &adapter->hw;
4740         int err;
4741
4742         if (adapter->need_ioport)
4743                 err = pci_enable_device(pdev);
4744         else
4745                 err = pci_enable_device_mem(pdev);
4746         if (err) {
4747                 pr_err("Cannot re-enable PCI device after reset.\n");
4748                 return PCI_ERS_RESULT_DISCONNECT;
4749         }
4750         pci_set_master(pdev);
4751
4752         pci_enable_wake(pdev, PCI_D3hot, 0);
4753         pci_enable_wake(pdev, PCI_D3cold, 0);
4754
4755         e1000_reset(adapter);
4756         ew32(WUS, ~0);
4757
4758         return PCI_ERS_RESULT_RECOVERED;
4759 }
4760
4761 /**
4762  * e1000_io_resume - called when traffic can start flowing again.
4763  * @pdev: Pointer to PCI device
4764  *
4765  * This callback is called when the error recovery driver tells us that
4766  * its OK to resume normal operation. Implementation resembles the
4767  * second-half of the e1000_resume routine.
4768  */
4769 static void e1000_io_resume(struct pci_dev *pdev)
4770 {
4771         struct net_device *netdev = pci_get_drvdata(pdev);
4772         struct e1000_adapter *adapter = netdev_priv(netdev);
4773
4774         e1000_init_manageability(adapter);
4775
4776         if (netif_running(netdev)) {
4777                 if (e1000_up(adapter)) {
4778                         pr_info("can't bring device back up after reset\n");
4779                         return;
4780                 }
4781         }
4782
4783         netif_device_attach(netdev);
4784 }
4785
4786 /* e1000_main.c */