e1000/e1000e: implement a simple interrupt moderation
[linux-2.6-microblaze.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("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_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("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("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("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("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("(PCI%s:%s:%s) ",
1051                 ((hw->bus_type == e1000_bus_type_pcix) ? "-X" : ""),
1052                 ((hw->bus_speed == e1000_bus_speed_133) ? "133MHz" :
1053                  (hw->bus_speed == e1000_bus_speed_120) ? "120MHz" :
1054                  (hw->bus_speed == e1000_bus_speed_100) ? "100MHz" :
1055                  (hw->bus_speed == e1000_bus_speed_66) ? "66MHz" : "33MHz"),
1056                 ((hw->bus_width == e1000_bus_width_64) ? "64-bit" : "32-bit"));
1057
1058         e_info("%pM\n", netdev->dev_addr);
1059
1060         /* carrier off reporting is important to ethtool even BEFORE open */
1061         netif_carrier_off(netdev);
1062
1063         e_info("Intel(R) PRO/1000 Network Connection\n");
1064
1065         cards_found++;
1066         return 0;
1067
1068 err_register:
1069 err_eeprom:
1070         e1000_phy_hw_reset(hw);
1071
1072         if (hw->flash_address)
1073                 iounmap(hw->flash_address);
1074         kfree(adapter->tx_ring);
1075         kfree(adapter->rx_ring);
1076 err_sw_init:
1077         iounmap(hw->hw_addr);
1078 err_ioremap:
1079         free_netdev(netdev);
1080 err_alloc_etherdev:
1081         pci_release_selected_regions(pdev, bars);
1082 err_pci_reg:
1083 err_dma:
1084         pci_disable_device(pdev);
1085         return err;
1086 }
1087
1088 /**
1089  * e1000_remove - Device Removal Routine
1090  * @pdev: PCI device information struct
1091  *
1092  * e1000_remove is called by the PCI subsystem to alert the driver
1093  * that it should release a PCI device.  The could be caused by a
1094  * Hot-Plug event, or because the driver is going to be removed from
1095  * memory.
1096  **/
1097
1098 static void __devexit e1000_remove(struct pci_dev *pdev)
1099 {
1100         struct net_device *netdev = pci_get_drvdata(pdev);
1101         struct e1000_adapter *adapter = netdev_priv(netdev);
1102         struct e1000_hw *hw = &adapter->hw;
1103
1104         set_bit(__E1000_DOWN, &adapter->flags);
1105         del_timer_sync(&adapter->tx_fifo_stall_timer);
1106         del_timer_sync(&adapter->watchdog_timer);
1107         del_timer_sync(&adapter->phy_info_timer);
1108
1109         cancel_work_sync(&adapter->reset_task);
1110
1111         e1000_release_manageability(adapter);
1112
1113         unregister_netdev(netdev);
1114
1115         e1000_phy_hw_reset(hw);
1116
1117         kfree(adapter->tx_ring);
1118         kfree(adapter->rx_ring);
1119
1120         iounmap(hw->hw_addr);
1121         if (hw->flash_address)
1122                 iounmap(hw->flash_address);
1123         pci_release_selected_regions(pdev, adapter->bars);
1124
1125         free_netdev(netdev);
1126
1127         pci_disable_device(pdev);
1128 }
1129
1130 /**
1131  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1132  * @adapter: board private structure to initialize
1133  *
1134  * e1000_sw_init initializes the Adapter private data structure.
1135  * Fields are initialized based on PCI device information and
1136  * OS network device settings (MTU size).
1137  **/
1138
1139 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
1140 {
1141         struct e1000_hw *hw = &adapter->hw;
1142         struct net_device *netdev = adapter->netdev;
1143         struct pci_dev *pdev = adapter->pdev;
1144
1145         /* PCI config space info */
1146
1147         hw->vendor_id = pdev->vendor;
1148         hw->device_id = pdev->device;
1149         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1150         hw->subsystem_id = pdev->subsystem_device;
1151         hw->revision_id = pdev->revision;
1152
1153         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
1154
1155         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1156         hw->max_frame_size = netdev->mtu +
1157                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
1158         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
1159
1160         /* identify the MAC */
1161
1162         if (e1000_set_mac_type(hw)) {
1163                 e_err("Unknown MAC Type\n");
1164                 return -EIO;
1165         }
1166
1167         switch (hw->mac_type) {
1168         default:
1169                 break;
1170         case e1000_82541:
1171         case e1000_82547:
1172         case e1000_82541_rev_2:
1173         case e1000_82547_rev_2:
1174                 hw->phy_init_script = 1;
1175                 break;
1176         }
1177
1178         e1000_set_media_type(hw);
1179
1180         hw->wait_autoneg_complete = false;
1181         hw->tbi_compatibility_en = true;
1182         hw->adaptive_ifs = true;
1183
1184         /* Copper options */
1185
1186         if (hw->media_type == e1000_media_type_copper) {
1187                 hw->mdix = AUTO_ALL_MODES;
1188                 hw->disable_polarity_correction = false;
1189                 hw->master_slave = E1000_MASTER_SLAVE;
1190         }
1191
1192         adapter->num_tx_queues = 1;
1193         adapter->num_rx_queues = 1;
1194
1195         if (e1000_alloc_queues(adapter)) {
1196                 e_err("Unable to allocate memory for queues\n");
1197                 return -ENOMEM;
1198         }
1199
1200         /* Explicitly disable IRQ since the NIC can be in any state. */
1201         e1000_irq_disable(adapter);
1202
1203         spin_lock_init(&adapter->stats_lock);
1204
1205         set_bit(__E1000_DOWN, &adapter->flags);
1206
1207         return 0;
1208 }
1209
1210 /**
1211  * e1000_alloc_queues - Allocate memory for all rings
1212  * @adapter: board private structure to initialize
1213  *
1214  * We allocate one ring per queue at run-time since we don't know the
1215  * number of queues at compile-time.
1216  **/
1217
1218 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1219 {
1220         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1221                                    sizeof(struct e1000_tx_ring), GFP_KERNEL);
1222         if (!adapter->tx_ring)
1223                 return -ENOMEM;
1224
1225         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1226                                    sizeof(struct e1000_rx_ring), GFP_KERNEL);
1227         if (!adapter->rx_ring) {
1228                 kfree(adapter->tx_ring);
1229                 return -ENOMEM;
1230         }
1231
1232         return E1000_SUCCESS;
1233 }
1234
1235 /**
1236  * e1000_open - Called when a network interface is made active
1237  * @netdev: network interface device structure
1238  *
1239  * Returns 0 on success, negative value on failure
1240  *
1241  * The open entry point is called when a network interface is made
1242  * active by the system (IFF_UP).  At this point all resources needed
1243  * for transmit and receive operations are allocated, the interrupt
1244  * handler is registered with the OS, the watchdog timer is started,
1245  * and the stack is notified that the interface is ready.
1246  **/
1247
1248 static int e1000_open(struct net_device *netdev)
1249 {
1250         struct e1000_adapter *adapter = netdev_priv(netdev);
1251         struct e1000_hw *hw = &adapter->hw;
1252         int err;
1253
1254         /* disallow open during test */
1255         if (test_bit(__E1000_TESTING, &adapter->flags))
1256                 return -EBUSY;
1257
1258         netif_carrier_off(netdev);
1259
1260         /* allocate transmit descriptors */
1261         err = e1000_setup_all_tx_resources(adapter);
1262         if (err)
1263                 goto err_setup_tx;
1264
1265         /* allocate receive descriptors */
1266         err = e1000_setup_all_rx_resources(adapter);
1267         if (err)
1268                 goto err_setup_rx;
1269
1270         e1000_power_up_phy(adapter);
1271
1272         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1273         if ((hw->mng_cookie.status &
1274                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1275                 e1000_update_mng_vlan(adapter);
1276         }
1277
1278         /* before we allocate an interrupt, we must be ready to handle it.
1279          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1280          * as soon as we call pci_request_irq, so we have to setup our
1281          * clean_rx handler before we do so.  */
1282         e1000_configure(adapter);
1283
1284         err = e1000_request_irq(adapter);
1285         if (err)
1286                 goto err_req_irq;
1287
1288         /* From here on the code is the same as e1000_up() */
1289         clear_bit(__E1000_DOWN, &adapter->flags);
1290
1291         napi_enable(&adapter->napi);
1292
1293         e1000_irq_enable(adapter);
1294
1295         netif_start_queue(netdev);
1296
1297         /* fire a link status change interrupt to start the watchdog */
1298         ew32(ICS, E1000_ICS_LSC);
1299
1300         return E1000_SUCCESS;
1301
1302 err_req_irq:
1303         e1000_power_down_phy(adapter);
1304         e1000_free_all_rx_resources(adapter);
1305 err_setup_rx:
1306         e1000_free_all_tx_resources(adapter);
1307 err_setup_tx:
1308         e1000_reset(adapter);
1309
1310         return err;
1311 }
1312
1313 /**
1314  * e1000_close - Disables a network interface
1315  * @netdev: network interface device structure
1316  *
1317  * Returns 0, this is not allowed to fail
1318  *
1319  * The close entry point is called when an interface is de-activated
1320  * by the OS.  The hardware is still under the drivers control, but
1321  * needs to be disabled.  A global MAC reset is issued to stop the
1322  * hardware, and all transmit and receive resources are freed.
1323  **/
1324
1325 static int e1000_close(struct net_device *netdev)
1326 {
1327         struct e1000_adapter *adapter = netdev_priv(netdev);
1328         struct e1000_hw *hw = &adapter->hw;
1329
1330         WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1331         e1000_down(adapter);
1332         e1000_power_down_phy(adapter);
1333         e1000_free_irq(adapter);
1334
1335         e1000_free_all_tx_resources(adapter);
1336         e1000_free_all_rx_resources(adapter);
1337
1338         /* kill manageability vlan ID if supported, but not if a vlan with
1339          * the same ID is registered on the host OS (let 8021q kill it) */
1340         if ((hw->mng_cookie.status &
1341                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1342              !(adapter->vlgrp &&
1343                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id))) {
1344                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1345         }
1346
1347         return 0;
1348 }
1349
1350 /**
1351  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1352  * @adapter: address of board private structure
1353  * @start: address of beginning of memory
1354  * @len: length of memory
1355  **/
1356 static bool e1000_check_64k_bound(struct e1000_adapter *adapter, void *start,
1357                                   unsigned long len)
1358 {
1359         struct e1000_hw *hw = &adapter->hw;
1360         unsigned long begin = (unsigned long)start;
1361         unsigned long end = begin + len;
1362
1363         /* First rev 82545 and 82546 need to not allow any memory
1364          * write location to cross 64k boundary due to errata 23 */
1365         if (hw->mac_type == e1000_82545 ||
1366             hw->mac_type == e1000_82546) {
1367                 return ((begin ^ (end - 1)) >> 16) != 0 ? false : true;
1368         }
1369
1370         return true;
1371 }
1372
1373 /**
1374  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1375  * @adapter: board private structure
1376  * @txdr:    tx descriptor ring (for a specific queue) to setup
1377  *
1378  * Return 0 on success, negative on failure
1379  **/
1380
1381 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
1382                                     struct e1000_tx_ring *txdr)
1383 {
1384         struct pci_dev *pdev = adapter->pdev;
1385         int size;
1386
1387         size = sizeof(struct e1000_buffer) * txdr->count;
1388         txdr->buffer_info = vmalloc(size);
1389         if (!txdr->buffer_info) {
1390                 e_err("Unable to allocate memory for the Tx descriptor 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("Unable to allocate memory for the Tx descriptor ring\n");
1406                 return -ENOMEM;
1407         }
1408
1409         /* Fix for errata 23, can't cross 64kB boundary */
1410         if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1411                 void *olddesc = txdr->desc;
1412                 dma_addr_t olddma = txdr->dma;
1413                 e_err("txdr align check failed: %u bytes at %p\n",
1414                       txdr->size, txdr->desc);
1415                 /* Try again, without freeing the previous */
1416                 txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size,
1417                                                 &txdr->dma, GFP_KERNEL);
1418                 /* Failed allocation, critical failure */
1419                 if (!txdr->desc) {
1420                         dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1421                                           olddma);
1422                         goto setup_tx_desc_die;
1423                 }
1424
1425                 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1426                         /* give up */
1427                         dma_free_coherent(&pdev->dev, txdr->size, txdr->desc,
1428                                           txdr->dma);
1429                         dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1430                                           olddma);
1431                         e_err("Unable to allocate aligned memory "
1432                               "for the transmit descriptor ring\n");
1433                         vfree(txdr->buffer_info);
1434                         return -ENOMEM;
1435                 } else {
1436                         /* Free old allocation, new allocation was successful */
1437                         dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1438                                           olddma);
1439                 }
1440         }
1441         memset(txdr->desc, 0, txdr->size);
1442
1443         txdr->next_to_use = 0;
1444         txdr->next_to_clean = 0;
1445
1446         return 0;
1447 }
1448
1449 /**
1450  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1451  *                                (Descriptors) for all queues
1452  * @adapter: board private structure
1453  *
1454  * Return 0 on success, negative on failure
1455  **/
1456
1457 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1458 {
1459         int i, err = 0;
1460
1461         for (i = 0; i < adapter->num_tx_queues; i++) {
1462                 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1463                 if (err) {
1464                         e_err("Allocation for Tx Queue %u failed\n", i);
1465                         for (i-- ; i >= 0; i--)
1466                                 e1000_free_tx_resources(adapter,
1467                                                         &adapter->tx_ring[i]);
1468                         break;
1469                 }
1470         }
1471
1472         return err;
1473 }
1474
1475 /**
1476  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1477  * @adapter: board private structure
1478  *
1479  * Configure the Tx unit of the MAC after a reset.
1480  **/
1481
1482 static void e1000_configure_tx(struct e1000_adapter *adapter)
1483 {
1484         u64 tdba;
1485         struct e1000_hw *hw = &adapter->hw;
1486         u32 tdlen, tctl, tipg;
1487         u32 ipgr1, ipgr2;
1488
1489         /* Setup the HW Tx Head and Tail descriptor pointers */
1490
1491         switch (adapter->num_tx_queues) {
1492         case 1:
1493         default:
1494                 tdba = adapter->tx_ring[0].dma;
1495                 tdlen = adapter->tx_ring[0].count *
1496                         sizeof(struct e1000_tx_desc);
1497                 ew32(TDLEN, tdlen);
1498                 ew32(TDBAH, (tdba >> 32));
1499                 ew32(TDBAL, (tdba & 0x00000000ffffffffULL));
1500                 ew32(TDT, 0);
1501                 ew32(TDH, 0);
1502                 adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ? E1000_TDH : E1000_82542_TDH);
1503                 adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ? E1000_TDT : E1000_82542_TDT);
1504                 break;
1505         }
1506
1507         /* Set the default values for the Tx Inter Packet Gap timer */
1508         if ((hw->media_type == e1000_media_type_fiber ||
1509              hw->media_type == e1000_media_type_internal_serdes))
1510                 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1511         else
1512                 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1513
1514         switch (hw->mac_type) {
1515         case e1000_82542_rev2_0:
1516         case e1000_82542_rev2_1:
1517                 tipg = DEFAULT_82542_TIPG_IPGT;
1518                 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1519                 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1520                 break;
1521         default:
1522                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1523                 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1524                 break;
1525         }
1526         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1527         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1528         ew32(TIPG, tipg);
1529
1530         /* Set the Tx Interrupt Delay register */
1531
1532         ew32(TIDV, adapter->tx_int_delay);
1533         if (hw->mac_type >= e1000_82540)
1534                 ew32(TADV, adapter->tx_abs_int_delay);
1535
1536         /* Program the Transmit Control Register */
1537
1538         tctl = er32(TCTL);
1539         tctl &= ~E1000_TCTL_CT;
1540         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1541                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1542
1543         e1000_config_collision_dist(hw);
1544
1545         /* Setup Transmit Descriptor Settings for eop descriptor */
1546         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1547
1548         /* only set IDE if we are delaying interrupts using the timers */
1549         if (adapter->tx_int_delay)
1550                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1551
1552         if (hw->mac_type < e1000_82543)
1553                 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1554         else
1555                 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1556
1557         /* Cache if we're 82544 running in PCI-X because we'll
1558          * need this to apply a workaround later in the send path. */
1559         if (hw->mac_type == e1000_82544 &&
1560             hw->bus_type == e1000_bus_type_pcix)
1561                 adapter->pcix_82544 = 1;
1562
1563         ew32(TCTL, tctl);
1564
1565 }
1566
1567 /**
1568  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1569  * @adapter: board private structure
1570  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1571  *
1572  * Returns 0 on success, negative on failure
1573  **/
1574
1575 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
1576                                     struct e1000_rx_ring *rxdr)
1577 {
1578         struct pci_dev *pdev = adapter->pdev;
1579         int size, desc_len;
1580
1581         size = sizeof(struct e1000_buffer) * rxdr->count;
1582         rxdr->buffer_info = vmalloc(size);
1583         if (!rxdr->buffer_info) {
1584                 e_err("Unable to allocate memory for the Rx descriptor ring\n");
1585                 return -ENOMEM;
1586         }
1587         memset(rxdr->buffer_info, 0, size);
1588
1589         desc_len = sizeof(struct e1000_rx_desc);
1590
1591         /* Round up to nearest 4K */
1592
1593         rxdr->size = rxdr->count * desc_len;
1594         rxdr->size = ALIGN(rxdr->size, 4096);
1595
1596         rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size, &rxdr->dma,
1597                                         GFP_KERNEL);
1598
1599         if (!rxdr->desc) {
1600                 e_err("Unable to allocate memory for the Rx descriptor ring\n");
1601 setup_rx_desc_die:
1602                 vfree(rxdr->buffer_info);
1603                 return -ENOMEM;
1604         }
1605
1606         /* Fix for errata 23, can't cross 64kB boundary */
1607         if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1608                 void *olddesc = rxdr->desc;
1609                 dma_addr_t olddma = rxdr->dma;
1610                 e_err("rxdr align check failed: %u bytes at %p\n",
1611                       rxdr->size, rxdr->desc);
1612                 /* Try again, without freeing the previous */
1613                 rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size,
1614                                                 &rxdr->dma, GFP_KERNEL);
1615                 /* Failed allocation, critical failure */
1616                 if (!rxdr->desc) {
1617                         dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1618                                           olddma);
1619                         e_err("Unable to allocate memory for the Rx descriptor "
1620                               "ring\n");
1621                         goto setup_rx_desc_die;
1622                 }
1623
1624                 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1625                         /* give up */
1626                         dma_free_coherent(&pdev->dev, rxdr->size, rxdr->desc,
1627                                           rxdr->dma);
1628                         dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1629                                           olddma);
1630                         e_err("Unable to allocate aligned memory for the Rx "
1631                               "descriptor ring\n");
1632                         goto setup_rx_desc_die;
1633                 } else {
1634                         /* Free old allocation, new allocation was successful */
1635                         dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1636                                           olddma);
1637                 }
1638         }
1639         memset(rxdr->desc, 0, rxdr->size);
1640
1641         rxdr->next_to_clean = 0;
1642         rxdr->next_to_use = 0;
1643         rxdr->rx_skb_top = NULL;
1644
1645         return 0;
1646 }
1647
1648 /**
1649  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1650  *                                (Descriptors) for all queues
1651  * @adapter: board private structure
1652  *
1653  * Return 0 on success, negative on failure
1654  **/
1655
1656 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1657 {
1658         int i, err = 0;
1659
1660         for (i = 0; i < adapter->num_rx_queues; i++) {
1661                 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1662                 if (err) {
1663                         e_err("Allocation for Rx Queue %u failed\n", i);
1664                         for (i-- ; i >= 0; i--)
1665                                 e1000_free_rx_resources(adapter,
1666                                                         &adapter->rx_ring[i]);
1667                         break;
1668                 }
1669         }
1670
1671         return err;
1672 }
1673
1674 /**
1675  * e1000_setup_rctl - configure the receive control registers
1676  * @adapter: Board private structure
1677  **/
1678 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1679 {
1680         struct e1000_hw *hw = &adapter->hw;
1681         u32 rctl;
1682
1683         rctl = er32(RCTL);
1684
1685         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1686
1687         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1688                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1689                 (hw->mc_filter_type << E1000_RCTL_MO_SHIFT);
1690
1691         if (hw->tbi_compatibility_on == 1)
1692                 rctl |= E1000_RCTL_SBP;
1693         else
1694                 rctl &= ~E1000_RCTL_SBP;
1695
1696         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1697                 rctl &= ~E1000_RCTL_LPE;
1698         else
1699                 rctl |= E1000_RCTL_LPE;
1700
1701         /* Setup buffer sizes */
1702         rctl &= ~E1000_RCTL_SZ_4096;
1703         rctl |= E1000_RCTL_BSEX;
1704         switch (adapter->rx_buffer_len) {
1705                 case E1000_RXBUFFER_2048:
1706                 default:
1707                         rctl |= E1000_RCTL_SZ_2048;
1708                         rctl &= ~E1000_RCTL_BSEX;
1709                         break;
1710                 case E1000_RXBUFFER_4096:
1711                         rctl |= E1000_RCTL_SZ_4096;
1712                         break;
1713                 case E1000_RXBUFFER_8192:
1714                         rctl |= E1000_RCTL_SZ_8192;
1715                         break;
1716                 case E1000_RXBUFFER_16384:
1717                         rctl |= E1000_RCTL_SZ_16384;
1718                         break;
1719         }
1720
1721         ew32(RCTL, rctl);
1722 }
1723
1724 /**
1725  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1726  * @adapter: board private structure
1727  *
1728  * Configure the Rx unit of the MAC after a reset.
1729  **/
1730
1731 static void e1000_configure_rx(struct e1000_adapter *adapter)
1732 {
1733         u64 rdba;
1734         struct e1000_hw *hw = &adapter->hw;
1735         u32 rdlen, rctl, rxcsum;
1736
1737         if (adapter->netdev->mtu > ETH_DATA_LEN) {
1738                 rdlen = adapter->rx_ring[0].count *
1739                         sizeof(struct e1000_rx_desc);
1740                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
1741                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
1742         } else {
1743                 rdlen = adapter->rx_ring[0].count *
1744                         sizeof(struct e1000_rx_desc);
1745                 adapter->clean_rx = e1000_clean_rx_irq;
1746                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1747         }
1748
1749         /* disable receives while setting up the descriptors */
1750         rctl = er32(RCTL);
1751         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1752
1753         /* set the Receive Delay Timer Register */
1754         ew32(RDTR, adapter->rx_int_delay);
1755
1756         if (hw->mac_type >= e1000_82540) {
1757                 ew32(RADV, adapter->rx_abs_int_delay);
1758                 if (adapter->itr_setting != 0)
1759                         ew32(ITR, 1000000000 / (adapter->itr * 256));
1760         }
1761
1762         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1763          * the Base and Length of the Rx Descriptor Ring */
1764         switch (adapter->num_rx_queues) {
1765         case 1:
1766         default:
1767                 rdba = adapter->rx_ring[0].dma;
1768                 ew32(RDLEN, rdlen);
1769                 ew32(RDBAH, (rdba >> 32));
1770                 ew32(RDBAL, (rdba & 0x00000000ffffffffULL));
1771                 ew32(RDT, 0);
1772                 ew32(RDH, 0);
1773                 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ? E1000_RDH : E1000_82542_RDH);
1774                 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ? E1000_RDT : E1000_82542_RDT);
1775                 break;
1776         }
1777
1778         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
1779         if (hw->mac_type >= e1000_82543) {
1780                 rxcsum = er32(RXCSUM);
1781                 if (adapter->rx_csum)
1782                         rxcsum |= E1000_RXCSUM_TUOFL;
1783                 else
1784                         /* don't need to clear IPPCSE as it defaults to 0 */
1785                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1786                 ew32(RXCSUM, rxcsum);
1787         }
1788
1789         /* Enable Receives */
1790         ew32(RCTL, rctl);
1791 }
1792
1793 /**
1794  * e1000_free_tx_resources - Free Tx Resources per Queue
1795  * @adapter: board private structure
1796  * @tx_ring: Tx descriptor ring for a specific queue
1797  *
1798  * Free all transmit software resources
1799  **/
1800
1801 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
1802                                     struct e1000_tx_ring *tx_ring)
1803 {
1804         struct pci_dev *pdev = adapter->pdev;
1805
1806         e1000_clean_tx_ring(adapter, tx_ring);
1807
1808         vfree(tx_ring->buffer_info);
1809         tx_ring->buffer_info = NULL;
1810
1811         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1812                           tx_ring->dma);
1813
1814         tx_ring->desc = NULL;
1815 }
1816
1817 /**
1818  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1819  * @adapter: board private structure
1820  *
1821  * Free all transmit software resources
1822  **/
1823
1824 void e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1825 {
1826         int i;
1827
1828         for (i = 0; i < adapter->num_tx_queues; i++)
1829                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1830 }
1831
1832 static void e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1833                                              struct e1000_buffer *buffer_info)
1834 {
1835         if (buffer_info->dma) {
1836                 if (buffer_info->mapped_as_page)
1837                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1838                                        buffer_info->length, DMA_TO_DEVICE);
1839                 else
1840                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1841                                          buffer_info->length,
1842                                          DMA_TO_DEVICE);
1843                 buffer_info->dma = 0;
1844         }
1845         if (buffer_info->skb) {
1846                 dev_kfree_skb_any(buffer_info->skb);
1847                 buffer_info->skb = NULL;
1848         }
1849         buffer_info->time_stamp = 0;
1850         /* buffer_info must be completely set up in the transmit path */
1851 }
1852
1853 /**
1854  * e1000_clean_tx_ring - Free Tx Buffers
1855  * @adapter: board private structure
1856  * @tx_ring: ring to be cleaned
1857  **/
1858
1859 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
1860                                 struct e1000_tx_ring *tx_ring)
1861 {
1862         struct e1000_hw *hw = &adapter->hw;
1863         struct e1000_buffer *buffer_info;
1864         unsigned long size;
1865         unsigned int i;
1866
1867         /* Free all the Tx ring sk_buffs */
1868
1869         for (i = 0; i < tx_ring->count; i++) {
1870                 buffer_info = &tx_ring->buffer_info[i];
1871                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
1872         }
1873
1874         size = sizeof(struct e1000_buffer) * tx_ring->count;
1875         memset(tx_ring->buffer_info, 0, size);
1876
1877         /* Zero out the descriptor ring */
1878
1879         memset(tx_ring->desc, 0, tx_ring->size);
1880
1881         tx_ring->next_to_use = 0;
1882         tx_ring->next_to_clean = 0;
1883         tx_ring->last_tx_tso = 0;
1884
1885         writel(0, hw->hw_addr + tx_ring->tdh);
1886         writel(0, hw->hw_addr + tx_ring->tdt);
1887 }
1888
1889 /**
1890  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
1891  * @adapter: board private structure
1892  **/
1893
1894 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
1895 {
1896         int i;
1897
1898         for (i = 0; i < adapter->num_tx_queues; i++)
1899                 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1900 }
1901
1902 /**
1903  * e1000_free_rx_resources - Free Rx Resources
1904  * @adapter: board private structure
1905  * @rx_ring: ring to clean the resources from
1906  *
1907  * Free all receive software resources
1908  **/
1909
1910 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
1911                                     struct e1000_rx_ring *rx_ring)
1912 {
1913         struct pci_dev *pdev = adapter->pdev;
1914
1915         e1000_clean_rx_ring(adapter, rx_ring);
1916
1917         vfree(rx_ring->buffer_info);
1918         rx_ring->buffer_info = NULL;
1919
1920         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1921                           rx_ring->dma);
1922
1923         rx_ring->desc = NULL;
1924 }
1925
1926 /**
1927  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
1928  * @adapter: board private structure
1929  *
1930  * Free all receive software resources
1931  **/
1932
1933 void e1000_free_all_rx_resources(struct e1000_adapter *adapter)
1934 {
1935         int i;
1936
1937         for (i = 0; i < adapter->num_rx_queues; i++)
1938                 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
1939 }
1940
1941 /**
1942  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1943  * @adapter: board private structure
1944  * @rx_ring: ring to free buffers from
1945  **/
1946
1947 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
1948                                 struct e1000_rx_ring *rx_ring)
1949 {
1950         struct e1000_hw *hw = &adapter->hw;
1951         struct e1000_buffer *buffer_info;
1952         struct pci_dev *pdev = adapter->pdev;
1953         unsigned long size;
1954         unsigned int i;
1955
1956         /* Free all the Rx ring sk_buffs */
1957         for (i = 0; i < rx_ring->count; i++) {
1958                 buffer_info = &rx_ring->buffer_info[i];
1959                 if (buffer_info->dma &&
1960                     adapter->clean_rx == e1000_clean_rx_irq) {
1961                         dma_unmap_single(&pdev->dev, buffer_info->dma,
1962                                          buffer_info->length,
1963                                          DMA_FROM_DEVICE);
1964                 } else if (buffer_info->dma &&
1965                            adapter->clean_rx == e1000_clean_jumbo_rx_irq) {
1966                         dma_unmap_page(&pdev->dev, buffer_info->dma,
1967                                        buffer_info->length,
1968                                        DMA_FROM_DEVICE);
1969                 }
1970
1971                 buffer_info->dma = 0;
1972                 if (buffer_info->page) {
1973                         put_page(buffer_info->page);
1974                         buffer_info->page = NULL;
1975                 }
1976                 if (buffer_info->skb) {
1977                         dev_kfree_skb(buffer_info->skb);
1978                         buffer_info->skb = NULL;
1979                 }
1980         }
1981
1982         /* there also may be some cached data from a chained receive */
1983         if (rx_ring->rx_skb_top) {
1984                 dev_kfree_skb(rx_ring->rx_skb_top);
1985                 rx_ring->rx_skb_top = NULL;
1986         }
1987
1988         size = sizeof(struct e1000_buffer) * rx_ring->count;
1989         memset(rx_ring->buffer_info, 0, size);
1990
1991         /* Zero out the descriptor ring */
1992         memset(rx_ring->desc, 0, rx_ring->size);
1993
1994         rx_ring->next_to_clean = 0;
1995         rx_ring->next_to_use = 0;
1996
1997         writel(0, hw->hw_addr + rx_ring->rdh);
1998         writel(0, hw->hw_addr + rx_ring->rdt);
1999 }
2000
2001 /**
2002  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2003  * @adapter: board private structure
2004  **/
2005
2006 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2007 {
2008         int i;
2009
2010         for (i = 0; i < adapter->num_rx_queues; i++)
2011                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2012 }
2013
2014 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2015  * and memory write and invalidate disabled for certain operations
2016  */
2017 static void e1000_enter_82542_rst(struct e1000_adapter *adapter)
2018 {
2019         struct e1000_hw *hw = &adapter->hw;
2020         struct net_device *netdev = adapter->netdev;
2021         u32 rctl;
2022
2023         e1000_pci_clear_mwi(hw);
2024
2025         rctl = er32(RCTL);
2026         rctl |= E1000_RCTL_RST;
2027         ew32(RCTL, rctl);
2028         E1000_WRITE_FLUSH();
2029         mdelay(5);
2030
2031         if (netif_running(netdev))
2032                 e1000_clean_all_rx_rings(adapter);
2033 }
2034
2035 static void e1000_leave_82542_rst(struct e1000_adapter *adapter)
2036 {
2037         struct e1000_hw *hw = &adapter->hw;
2038         struct net_device *netdev = adapter->netdev;
2039         u32 rctl;
2040
2041         rctl = er32(RCTL);
2042         rctl &= ~E1000_RCTL_RST;
2043         ew32(RCTL, rctl);
2044         E1000_WRITE_FLUSH();
2045         mdelay(5);
2046
2047         if (hw->pci_cmd_word & PCI_COMMAND_INVALIDATE)
2048                 e1000_pci_set_mwi(hw);
2049
2050         if (netif_running(netdev)) {
2051                 /* No need to loop, because 82542 supports only 1 queue */
2052                 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2053                 e1000_configure_rx(adapter);
2054                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2055         }
2056 }
2057
2058 /**
2059  * e1000_set_mac - Change the Ethernet Address of the NIC
2060  * @netdev: network interface device structure
2061  * @p: pointer to an address structure
2062  *
2063  * Returns 0 on success, negative on failure
2064  **/
2065
2066 static int e1000_set_mac(struct net_device *netdev, void *p)
2067 {
2068         struct e1000_adapter *adapter = netdev_priv(netdev);
2069         struct e1000_hw *hw = &adapter->hw;
2070         struct sockaddr *addr = p;
2071
2072         if (!is_valid_ether_addr(addr->sa_data))
2073                 return -EADDRNOTAVAIL;
2074
2075         /* 82542 2.0 needs to be in reset to write receive address registers */
2076
2077         if (hw->mac_type == e1000_82542_rev2_0)
2078                 e1000_enter_82542_rst(adapter);
2079
2080         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2081         memcpy(hw->mac_addr, addr->sa_data, netdev->addr_len);
2082
2083         e1000_rar_set(hw, hw->mac_addr, 0);
2084
2085         if (hw->mac_type == e1000_82542_rev2_0)
2086                 e1000_leave_82542_rst(adapter);
2087
2088         return 0;
2089 }
2090
2091 /**
2092  * e1000_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2093  * @netdev: network interface device structure
2094  *
2095  * The set_rx_mode entry point is called whenever the unicast or multicast
2096  * address lists or the network interface flags are updated. This routine is
2097  * responsible for configuring the hardware for proper unicast, multicast,
2098  * promiscuous mode, and all-multi behavior.
2099  **/
2100
2101 static void e1000_set_rx_mode(struct net_device *netdev)
2102 {
2103         struct e1000_adapter *adapter = netdev_priv(netdev);
2104         struct e1000_hw *hw = &adapter->hw;
2105         struct netdev_hw_addr *ha;
2106         bool use_uc = false;
2107         u32 rctl;
2108         u32 hash_value;
2109         int i, rar_entries = E1000_RAR_ENTRIES;
2110         int mta_reg_count = E1000_NUM_MTA_REGISTERS;
2111         u32 *mcarray = kcalloc(mta_reg_count, sizeof(u32), GFP_ATOMIC);
2112
2113         if (!mcarray) {
2114                 e_err("memory allocation failed\n");
2115                 return;
2116         }
2117
2118         /* Check for Promiscuous and All Multicast modes */
2119
2120         rctl = er32(RCTL);
2121
2122         if (netdev->flags & IFF_PROMISC) {
2123                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2124                 rctl &= ~E1000_RCTL_VFE;
2125         } else {
2126                 if (netdev->flags & IFF_ALLMULTI)
2127                         rctl |= E1000_RCTL_MPE;
2128                 else
2129                         rctl &= ~E1000_RCTL_MPE;
2130                 /* Enable VLAN filter if there is a VLAN */
2131                 if (adapter->vlgrp)
2132                         rctl |= E1000_RCTL_VFE;
2133         }
2134
2135         if (netdev_uc_count(netdev) > rar_entries - 1) {
2136                 rctl |= E1000_RCTL_UPE;
2137         } else if (!(netdev->flags & IFF_PROMISC)) {
2138                 rctl &= ~E1000_RCTL_UPE;
2139                 use_uc = true;
2140         }
2141
2142         ew32(RCTL, rctl);
2143
2144         /* 82542 2.0 needs to be in reset to write receive address registers */
2145
2146         if (hw->mac_type == e1000_82542_rev2_0)
2147                 e1000_enter_82542_rst(adapter);
2148
2149         /* load the first 14 addresses into the exact filters 1-14. Unicast
2150          * addresses take precedence to avoid disabling unicast filtering
2151          * when possible.
2152          *
2153          * RAR 0 is used for the station MAC adddress
2154          * if there are not 14 addresses, go ahead and clear the filters
2155          */
2156         i = 1;
2157         if (use_uc)
2158                 netdev_for_each_uc_addr(ha, netdev) {
2159                         if (i == rar_entries)
2160                                 break;
2161                         e1000_rar_set(hw, ha->addr, i++);
2162                 }
2163
2164         WARN_ON(i == rar_entries);
2165
2166         netdev_for_each_mc_addr(ha, netdev) {
2167                 if (i == rar_entries) {
2168                         /* load any remaining addresses into the hash table */
2169                         u32 hash_reg, hash_bit, mta;
2170                         hash_value = e1000_hash_mc_addr(hw, ha->addr);
2171                         hash_reg = (hash_value >> 5) & 0x7F;
2172                         hash_bit = hash_value & 0x1F;
2173                         mta = (1 << hash_bit);
2174                         mcarray[hash_reg] |= mta;
2175                 } else {
2176                         e1000_rar_set(hw, ha->addr, i++);
2177                 }
2178         }
2179
2180         for (; i < rar_entries; i++) {
2181                 E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2182                 E1000_WRITE_FLUSH();
2183                 E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2184                 E1000_WRITE_FLUSH();
2185         }
2186
2187         /* write the hash table completely, write from bottom to avoid
2188          * both stupid write combining chipsets, and flushing each write */
2189         for (i = mta_reg_count - 1; i >= 0 ; i--) {
2190                 /*
2191                  * If we are on an 82544 has an errata where writing odd
2192                  * offsets overwrites the previous even offset, but writing
2193                  * backwards over the range solves the issue by always
2194                  * writing the odd offset first
2195                  */
2196                 E1000_WRITE_REG_ARRAY(hw, MTA, i, mcarray[i]);
2197         }
2198         E1000_WRITE_FLUSH();
2199
2200         if (hw->mac_type == e1000_82542_rev2_0)
2201                 e1000_leave_82542_rst(adapter);
2202
2203         kfree(mcarray);
2204 }
2205
2206 /* Need to wait a few seconds after link up to get diagnostic information from
2207  * the phy */
2208
2209 static void e1000_update_phy_info(unsigned long data)
2210 {
2211         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2212         struct e1000_hw *hw = &adapter->hw;
2213         e1000_phy_get_info(hw, &adapter->phy_info);
2214 }
2215
2216 /**
2217  * e1000_82547_tx_fifo_stall - Timer Call-back
2218  * @data: pointer to adapter cast into an unsigned long
2219  **/
2220
2221 static void e1000_82547_tx_fifo_stall(unsigned long data)
2222 {
2223         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2224         struct e1000_hw *hw = &adapter->hw;
2225         struct net_device *netdev = adapter->netdev;
2226         u32 tctl;
2227
2228         if (atomic_read(&adapter->tx_fifo_stall)) {
2229                 if ((er32(TDT) == er32(TDH)) &&
2230                    (er32(TDFT) == er32(TDFH)) &&
2231                    (er32(TDFTS) == er32(TDFHS))) {
2232                         tctl = er32(TCTL);
2233                         ew32(TCTL, tctl & ~E1000_TCTL_EN);
2234                         ew32(TDFT, adapter->tx_head_addr);
2235                         ew32(TDFH, adapter->tx_head_addr);
2236                         ew32(TDFTS, adapter->tx_head_addr);
2237                         ew32(TDFHS, adapter->tx_head_addr);
2238                         ew32(TCTL, tctl);
2239                         E1000_WRITE_FLUSH();
2240
2241                         adapter->tx_fifo_head = 0;
2242                         atomic_set(&adapter->tx_fifo_stall, 0);
2243                         netif_wake_queue(netdev);
2244                 } else if (!test_bit(__E1000_DOWN, &adapter->flags)) {
2245                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2246                 }
2247         }
2248 }
2249
2250 bool e1000_has_link(struct e1000_adapter *adapter)
2251 {
2252         struct e1000_hw *hw = &adapter->hw;
2253         bool link_active = false;
2254
2255         /* get_link_status is set on LSC (link status) interrupt or
2256          * rx sequence error interrupt.  get_link_status will stay
2257          * false until the e1000_check_for_link establishes link
2258          * for copper adapters ONLY
2259          */
2260         switch (hw->media_type) {
2261         case e1000_media_type_copper:
2262                 if (hw->get_link_status) {
2263                         e1000_check_for_link(hw);
2264                         link_active = !hw->get_link_status;
2265                 } else {
2266                         link_active = true;
2267                 }
2268                 break;
2269         case e1000_media_type_fiber:
2270                 e1000_check_for_link(hw);
2271                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2272                 break;
2273         case e1000_media_type_internal_serdes:
2274                 e1000_check_for_link(hw);
2275                 link_active = hw->serdes_has_link;
2276                 break;
2277         default:
2278                 break;
2279         }
2280
2281         return link_active;
2282 }
2283
2284 /**
2285  * e1000_watchdog - Timer Call-back
2286  * @data: pointer to adapter cast into an unsigned long
2287  **/
2288 static void e1000_watchdog(unsigned long data)
2289 {
2290         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2291         struct e1000_hw *hw = &adapter->hw;
2292         struct net_device *netdev = adapter->netdev;
2293         struct e1000_tx_ring *txdr = adapter->tx_ring;
2294         u32 link, tctl;
2295
2296         link = e1000_has_link(adapter);
2297         if ((netif_carrier_ok(netdev)) && link)
2298                 goto link_up;
2299
2300         if (link) {
2301                 if (!netif_carrier_ok(netdev)) {
2302                         u32 ctrl;
2303                         bool txb2b = true;
2304                         /* update snapshot of PHY registers on LSC */
2305                         e1000_get_speed_and_duplex(hw,
2306                                                    &adapter->link_speed,
2307                                                    &adapter->link_duplex);
2308
2309                         ctrl = er32(CTRL);
2310                         pr_info("%s NIC Link is Up %d Mbps %s, "
2311                                 "Flow Control: %s\n",
2312                                 netdev->name,
2313                                 adapter->link_speed,
2314                                 adapter->link_duplex == FULL_DUPLEX ?
2315                                 "Full Duplex" : "Half Duplex",
2316                                 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2317                                 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2318                                 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2319                                 E1000_CTRL_TFCE) ? "TX" : "None")));
2320
2321                         /* adjust timeout factor according to speed/duplex */
2322                         adapter->tx_timeout_factor = 1;
2323                         switch (adapter->link_speed) {
2324                         case SPEED_10:
2325                                 txb2b = false;
2326                                 adapter->tx_timeout_factor = 16;
2327                                 break;
2328                         case SPEED_100:
2329                                 txb2b = false;
2330                                 /* maybe add some timeout factor ? */
2331                                 break;
2332                         }
2333
2334                         /* enable transmits in the hardware */
2335                         tctl = er32(TCTL);
2336                         tctl |= E1000_TCTL_EN;
2337                         ew32(TCTL, tctl);
2338
2339                         netif_carrier_on(netdev);
2340                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2341                                 mod_timer(&adapter->phy_info_timer,
2342                                           round_jiffies(jiffies + 2 * HZ));
2343                         adapter->smartspeed = 0;
2344                 }
2345         } else {
2346                 if (netif_carrier_ok(netdev)) {
2347                         adapter->link_speed = 0;
2348                         adapter->link_duplex = 0;
2349                         pr_info("%s NIC Link is Down\n",
2350                                 netdev->name);
2351                         netif_carrier_off(netdev);
2352
2353                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2354                                 mod_timer(&adapter->phy_info_timer,
2355                                           round_jiffies(jiffies + 2 * HZ));
2356                 }
2357
2358                 e1000_smartspeed(adapter);
2359         }
2360
2361 link_up:
2362         e1000_update_stats(adapter);
2363
2364         hw->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2365         adapter->tpt_old = adapter->stats.tpt;
2366         hw->collision_delta = adapter->stats.colc - adapter->colc_old;
2367         adapter->colc_old = adapter->stats.colc;
2368
2369         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2370         adapter->gorcl_old = adapter->stats.gorcl;
2371         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2372         adapter->gotcl_old = adapter->stats.gotcl;
2373
2374         e1000_update_adaptive(hw);
2375
2376         if (!netif_carrier_ok(netdev)) {
2377                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2378                         /* We've lost link, so the controller stops DMA,
2379                          * but we've got queued Tx work that's never going
2380                          * to get done, so reset controller to flush Tx.
2381                          * (Do the reset outside of interrupt context). */
2382                         adapter->tx_timeout_count++;
2383                         schedule_work(&adapter->reset_task);
2384                         /* return immediately since reset is imminent */
2385                         return;
2386                 }
2387         }
2388
2389         /* Simple mode for Interrupt Throttle Rate (ITR) */
2390         if (hw->mac_type >= e1000_82540 && adapter->itr_setting == 4) {
2391                 /*
2392                  * Symmetric Tx/Rx gets a reduced ITR=2000;
2393                  * Total asymmetrical Tx or Rx gets ITR=8000;
2394                  * everyone else is between 2000-8000.
2395                  */
2396                 u32 goc = (adapter->gotcl + adapter->gorcl) / 10000;
2397                 u32 dif = (adapter->gotcl > adapter->gorcl ?
2398                             adapter->gotcl - adapter->gorcl :
2399                             adapter->gorcl - adapter->gotcl) / 10000;
2400                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
2401
2402                 ew32(ITR, 1000000000 / (itr * 256));
2403         }
2404
2405         /* Cause software interrupt to ensure rx ring is cleaned */
2406         ew32(ICS, E1000_ICS_RXDMT0);
2407
2408         /* Force detection of hung controller every watchdog period */
2409         adapter->detect_tx_hung = true;
2410
2411         /* Reset the timer */
2412         if (!test_bit(__E1000_DOWN, &adapter->flags))
2413                 mod_timer(&adapter->watchdog_timer,
2414                           round_jiffies(jiffies + 2 * HZ));
2415 }
2416
2417 enum latency_range {
2418         lowest_latency = 0,
2419         low_latency = 1,
2420         bulk_latency = 2,
2421         latency_invalid = 255
2422 };
2423
2424 /**
2425  * e1000_update_itr - update the dynamic ITR value based on statistics
2426  * @adapter: pointer to adapter
2427  * @itr_setting: current adapter->itr
2428  * @packets: the number of packets during this measurement interval
2429  * @bytes: the number of bytes during this measurement interval
2430  *
2431  *      Stores a new ITR value based on packets and byte
2432  *      counts during the last interrupt.  The advantage of per interrupt
2433  *      computation is faster updates and more accurate ITR for the current
2434  *      traffic pattern.  Constants in this function were computed
2435  *      based on theoretical maximum wire speed and thresholds were set based
2436  *      on testing data as well as attempting to minimize response time
2437  *      while increasing bulk throughput.
2438  *      this functionality is controlled by the InterruptThrottleRate module
2439  *      parameter (see e1000_param.c)
2440  **/
2441 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2442                                      u16 itr_setting, int packets, int bytes)
2443 {
2444         unsigned int retval = itr_setting;
2445         struct e1000_hw *hw = &adapter->hw;
2446
2447         if (unlikely(hw->mac_type < e1000_82540))
2448                 goto update_itr_done;
2449
2450         if (packets == 0)
2451                 goto update_itr_done;
2452
2453         switch (itr_setting) {
2454         case lowest_latency:
2455                 /* jumbo frames get bulk treatment*/
2456                 if (bytes/packets > 8000)
2457                         retval = bulk_latency;
2458                 else if ((packets < 5) && (bytes > 512))
2459                         retval = low_latency;
2460                 break;
2461         case low_latency:  /* 50 usec aka 20000 ints/s */
2462                 if (bytes > 10000) {
2463                         /* jumbo frames need bulk latency setting */
2464                         if (bytes/packets > 8000)
2465                                 retval = bulk_latency;
2466                         else if ((packets < 10) || ((bytes/packets) > 1200))
2467                                 retval = bulk_latency;
2468                         else if ((packets > 35))
2469                                 retval = lowest_latency;
2470                 } else if (bytes/packets > 2000)
2471                         retval = bulk_latency;
2472                 else if (packets <= 2 && bytes < 512)
2473                         retval = lowest_latency;
2474                 break;
2475         case bulk_latency: /* 250 usec aka 4000 ints/s */
2476                 if (bytes > 25000) {
2477                         if (packets > 35)
2478                                 retval = low_latency;
2479                 } else if (bytes < 6000) {
2480                         retval = low_latency;
2481                 }
2482                 break;
2483         }
2484
2485 update_itr_done:
2486         return retval;
2487 }
2488
2489 static void e1000_set_itr(struct e1000_adapter *adapter)
2490 {
2491         struct e1000_hw *hw = &adapter->hw;
2492         u16 current_itr;
2493         u32 new_itr = adapter->itr;
2494
2495         if (unlikely(hw->mac_type < e1000_82540))
2496                 return;
2497
2498         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2499         if (unlikely(adapter->link_speed != SPEED_1000)) {
2500                 current_itr = 0;
2501                 new_itr = 4000;
2502                 goto set_itr_now;
2503         }
2504
2505         adapter->tx_itr = e1000_update_itr(adapter,
2506                                     adapter->tx_itr,
2507                                     adapter->total_tx_packets,
2508                                     adapter->total_tx_bytes);
2509         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2510         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2511                 adapter->tx_itr = low_latency;
2512
2513         adapter->rx_itr = e1000_update_itr(adapter,
2514                                     adapter->rx_itr,
2515                                     adapter->total_rx_packets,
2516                                     adapter->total_rx_bytes);
2517         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2518         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2519                 adapter->rx_itr = low_latency;
2520
2521         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2522
2523         switch (current_itr) {
2524         /* counts and packets in update_itr are dependent on these numbers */
2525         case lowest_latency:
2526                 new_itr = 70000;
2527                 break;
2528         case low_latency:
2529                 new_itr = 20000; /* aka hwitr = ~200 */
2530                 break;
2531         case bulk_latency:
2532                 new_itr = 4000;
2533                 break;
2534         default:
2535                 break;
2536         }
2537
2538 set_itr_now:
2539         if (new_itr != adapter->itr) {
2540                 /* this attempts to bias the interrupt rate towards Bulk
2541                  * by adding intermediate steps when interrupt rate is
2542                  * increasing */
2543                 new_itr = new_itr > adapter->itr ?
2544                              min(adapter->itr + (new_itr >> 2), new_itr) :
2545                              new_itr;
2546                 adapter->itr = new_itr;
2547                 ew32(ITR, 1000000000 / (new_itr * 256));
2548         }
2549
2550         return;
2551 }
2552
2553 #define E1000_TX_FLAGS_CSUM             0x00000001
2554 #define E1000_TX_FLAGS_VLAN             0x00000002
2555 #define E1000_TX_FLAGS_TSO              0x00000004
2556 #define E1000_TX_FLAGS_IPV4             0x00000008
2557 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2558 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2559
2560 static int e1000_tso(struct e1000_adapter *adapter,
2561                      struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2562 {
2563         struct e1000_context_desc *context_desc;
2564         struct e1000_buffer *buffer_info;
2565         unsigned int i;
2566         u32 cmd_length = 0;
2567         u16 ipcse = 0, tucse, mss;
2568         u8 ipcss, ipcso, tucss, tucso, hdr_len;
2569         int err;
2570
2571         if (skb_is_gso(skb)) {
2572                 if (skb_header_cloned(skb)) {
2573                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2574                         if (err)
2575                                 return err;
2576                 }
2577
2578                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2579                 mss = skb_shinfo(skb)->gso_size;
2580                 if (skb->protocol == htons(ETH_P_IP)) {
2581                         struct iphdr *iph = ip_hdr(skb);
2582                         iph->tot_len = 0;
2583                         iph->check = 0;
2584                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2585                                                                  iph->daddr, 0,
2586                                                                  IPPROTO_TCP,
2587                                                                  0);
2588                         cmd_length = E1000_TXD_CMD_IP;
2589                         ipcse = skb_transport_offset(skb) - 1;
2590                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2591                         ipv6_hdr(skb)->payload_len = 0;
2592                         tcp_hdr(skb)->check =
2593                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2594                                                  &ipv6_hdr(skb)->daddr,
2595                                                  0, IPPROTO_TCP, 0);
2596                         ipcse = 0;
2597                 }
2598                 ipcss = skb_network_offset(skb);
2599                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2600                 tucss = skb_transport_offset(skb);
2601                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2602                 tucse = 0;
2603
2604                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2605                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2606
2607                 i = tx_ring->next_to_use;
2608                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2609                 buffer_info = &tx_ring->buffer_info[i];
2610
2611                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2612                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2613                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2614                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2615                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2616                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2617                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2618                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2619                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2620
2621                 buffer_info->time_stamp = jiffies;
2622                 buffer_info->next_to_watch = i;
2623
2624                 if (++i == tx_ring->count) i = 0;
2625                 tx_ring->next_to_use = i;
2626
2627                 return true;
2628         }
2629         return false;
2630 }
2631
2632 static bool e1000_tx_csum(struct e1000_adapter *adapter,
2633                           struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2634 {
2635         struct e1000_context_desc *context_desc;
2636         struct e1000_buffer *buffer_info;
2637         unsigned int i;
2638         u8 css;
2639         u32 cmd_len = E1000_TXD_CMD_DEXT;
2640
2641         if (skb->ip_summed != CHECKSUM_PARTIAL)
2642                 return false;
2643
2644         switch (skb->protocol) {
2645         case cpu_to_be16(ETH_P_IP):
2646                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2647                         cmd_len |= E1000_TXD_CMD_TCP;
2648                 break;
2649         case cpu_to_be16(ETH_P_IPV6):
2650                 /* XXX not handling all IPV6 headers */
2651                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2652                         cmd_len |= E1000_TXD_CMD_TCP;
2653                 break;
2654         default:
2655                 if (unlikely(net_ratelimit()))
2656                         e_warn("checksum_partial proto=%x!\n", 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("__pskb_pull_tail failed.\n");
3001                                         dev_kfree_skb_any(skb);
3002                                         return NETDEV_TX_OK;
3003                                 }
3004                                 len = skb_headlen(skb);
3005                                 break;
3006                         default:
3007                                 /* do nothing */
3008                                 break;
3009                         }
3010                 }
3011         }
3012
3013         /* reserve a descriptor for the offload context */
3014         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3015                 count++;
3016         count++;
3017
3018         /* Controller Erratum workaround */
3019         if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3020                 count++;
3021
3022         count += TXD_USE_COUNT(len, max_txd_pwr);
3023
3024         if (adapter->pcix_82544)
3025                 count++;
3026
3027         /* work-around for errata 10 and it applies to all controllers
3028          * in PCI-X mode, so add one more descriptor to the count
3029          */
3030         if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
3031                         (len > 2015)))
3032                 count++;
3033
3034         nr_frags = skb_shinfo(skb)->nr_frags;
3035         for (f = 0; f < nr_frags; f++)
3036                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3037                                        max_txd_pwr);
3038         if (adapter->pcix_82544)
3039                 count += nr_frags;
3040
3041         /* need: count + 2 desc gap to keep tail from touching
3042          * head, otherwise try next time */
3043         if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2)))
3044                 return NETDEV_TX_BUSY;
3045
3046         if (unlikely(hw->mac_type == e1000_82547)) {
3047                 if (unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
3048                         netif_stop_queue(netdev);
3049                         if (!test_bit(__E1000_DOWN, &adapter->flags))
3050                                 mod_timer(&adapter->tx_fifo_stall_timer,
3051                                           jiffies + 1);
3052                         return NETDEV_TX_BUSY;
3053                 }
3054         }
3055
3056         if (unlikely(adapter->vlgrp && vlan_tx_tag_present(skb))) {
3057                 tx_flags |= E1000_TX_FLAGS_VLAN;
3058                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3059         }
3060
3061         first = tx_ring->next_to_use;
3062
3063         tso = e1000_tso(adapter, tx_ring, skb);
3064         if (tso < 0) {
3065                 dev_kfree_skb_any(skb);
3066                 return NETDEV_TX_OK;
3067         }
3068
3069         if (likely(tso)) {
3070                 if (likely(hw->mac_type != e1000_82544))
3071                         tx_ring->last_tx_tso = 1;
3072                 tx_flags |= E1000_TX_FLAGS_TSO;
3073         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3074                 tx_flags |= E1000_TX_FLAGS_CSUM;
3075
3076         if (likely(skb->protocol == htons(ETH_P_IP)))
3077                 tx_flags |= E1000_TX_FLAGS_IPV4;
3078
3079         count = e1000_tx_map(adapter, tx_ring, skb, first, max_per_txd,
3080                              nr_frags, mss);
3081
3082         if (count) {
3083                 e1000_tx_queue(adapter, tx_ring, tx_flags, count);
3084                 /* Make sure there is space in the ring for the next send. */
3085                 e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3086
3087         } else {
3088                 dev_kfree_skb_any(skb);
3089                 tx_ring->buffer_info[first].time_stamp = 0;
3090                 tx_ring->next_to_use = first;
3091         }
3092
3093         return NETDEV_TX_OK;
3094 }
3095
3096 /**
3097  * e1000_tx_timeout - Respond to a Tx Hang
3098  * @netdev: network interface device structure
3099  **/
3100
3101 static void e1000_tx_timeout(struct net_device *netdev)
3102 {
3103         struct e1000_adapter *adapter = netdev_priv(netdev);
3104
3105         /* Do the reset outside of interrupt context */
3106         adapter->tx_timeout_count++;
3107         schedule_work(&adapter->reset_task);
3108 }
3109
3110 static void e1000_reset_task(struct work_struct *work)
3111 {
3112         struct e1000_adapter *adapter =
3113                 container_of(work, struct e1000_adapter, reset_task);
3114
3115         e1000_reinit_locked(adapter);
3116 }
3117
3118 /**
3119  * e1000_get_stats - Get System Network Statistics
3120  * @netdev: network interface device structure
3121  *
3122  * Returns the address of the device statistics structure.
3123  * The statistics are actually updated from the timer callback.
3124  **/
3125
3126 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3127 {
3128         /* only return the current stats */
3129         return &netdev->stats;
3130 }
3131
3132 /**
3133  * e1000_change_mtu - Change the Maximum Transfer Unit
3134  * @netdev: network interface device structure
3135  * @new_mtu: new value for maximum frame size
3136  *
3137  * Returns 0 on success, negative on failure
3138  **/
3139
3140 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3141 {
3142         struct e1000_adapter *adapter = netdev_priv(netdev);
3143         struct e1000_hw *hw = &adapter->hw;
3144         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3145
3146         if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3147             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3148                 e_err("Invalid MTU setting\n");
3149                 return -EINVAL;
3150         }
3151
3152         /* Adapter-specific max frame size limits. */
3153         switch (hw->mac_type) {
3154         case e1000_undefined ... e1000_82542_rev2_1:
3155                 if (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)) {
3156                         e_err("Jumbo Frames not supported.\n");
3157                         return -EINVAL;
3158                 }
3159                 break;
3160         default:
3161                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3162                 break;
3163         }
3164
3165         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
3166                 msleep(1);
3167         /* e1000_down has a dependency on max_frame_size */
3168         hw->max_frame_size = max_frame;
3169         if (netif_running(netdev))
3170                 e1000_down(adapter);
3171
3172         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3173          * means we reserve 2 more, this pushes us to allocate from the next
3174          * larger slab size.
3175          * i.e. RXBUFFER_2048 --> size-4096 slab
3176          *  however with the new *_jumbo_rx* routines, jumbo receives will use
3177          *  fragmented skbs */
3178
3179         if (max_frame <= E1000_RXBUFFER_2048)
3180                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3181         else
3182 #if (PAGE_SIZE >= E1000_RXBUFFER_16384)
3183                 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3184 #elif (PAGE_SIZE >= E1000_RXBUFFER_4096)
3185                 adapter->rx_buffer_len = PAGE_SIZE;
3186 #endif
3187
3188         /* adjust allocation if LPE protects us, and we aren't using SBP */
3189         if (!hw->tbi_compatibility_on &&
3190             ((max_frame == (ETH_FRAME_LEN + ETH_FCS_LEN)) ||
3191              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3192                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3193
3194         pr_info("%s changing MTU from %d to %d\n",
3195                 netdev->name, netdev->mtu, new_mtu);
3196         netdev->mtu = new_mtu;
3197
3198         if (netif_running(netdev))
3199                 e1000_up(adapter);
3200         else
3201                 e1000_reset(adapter);
3202
3203         clear_bit(__E1000_RESETTING, &adapter->flags);
3204
3205         return 0;
3206 }
3207
3208 /**
3209  * e1000_update_stats - Update the board statistics counters
3210  * @adapter: board private structure
3211  **/
3212
3213 void e1000_update_stats(struct e1000_adapter *adapter)
3214 {
3215         struct net_device *netdev = adapter->netdev;
3216         struct e1000_hw *hw = &adapter->hw;
3217         struct pci_dev *pdev = adapter->pdev;
3218         unsigned long flags;
3219         u16 phy_tmp;
3220
3221 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3222
3223         /*
3224          * Prevent stats update while adapter is being reset, or if the pci
3225          * connection is down.
3226          */
3227         if (adapter->link_speed == 0)
3228                 return;
3229         if (pci_channel_offline(pdev))
3230                 return;
3231
3232         spin_lock_irqsave(&adapter->stats_lock, flags);
3233
3234         /* these counters are modified from e1000_tbi_adjust_stats,
3235          * called from the interrupt context, so they must only
3236          * be written while holding adapter->stats_lock
3237          */
3238
3239         adapter->stats.crcerrs += er32(CRCERRS);
3240         adapter->stats.gprc += er32(GPRC);
3241         adapter->stats.gorcl += er32(GORCL);
3242         adapter->stats.gorch += er32(GORCH);
3243         adapter->stats.bprc += er32(BPRC);
3244         adapter->stats.mprc += er32(MPRC);
3245         adapter->stats.roc += er32(ROC);
3246
3247         adapter->stats.prc64 += er32(PRC64);
3248         adapter->stats.prc127 += er32(PRC127);
3249         adapter->stats.prc255 += er32(PRC255);
3250         adapter->stats.prc511 += er32(PRC511);
3251         adapter->stats.prc1023 += er32(PRC1023);
3252         adapter->stats.prc1522 += er32(PRC1522);
3253
3254         adapter->stats.symerrs += er32(SYMERRS);
3255         adapter->stats.mpc += er32(MPC);
3256         adapter->stats.scc += er32(SCC);
3257         adapter->stats.ecol += er32(ECOL);
3258         adapter->stats.mcc += er32(MCC);
3259         adapter->stats.latecol += er32(LATECOL);
3260         adapter->stats.dc += er32(DC);
3261         adapter->stats.sec += er32(SEC);
3262         adapter->stats.rlec += er32(RLEC);
3263         adapter->stats.xonrxc += er32(XONRXC);
3264         adapter->stats.xontxc += er32(XONTXC);
3265         adapter->stats.xoffrxc += er32(XOFFRXC);
3266         adapter->stats.xofftxc += er32(XOFFTXC);
3267         adapter->stats.fcruc += er32(FCRUC);
3268         adapter->stats.gptc += er32(GPTC);
3269         adapter->stats.gotcl += er32(GOTCL);
3270         adapter->stats.gotch += er32(GOTCH);
3271         adapter->stats.rnbc += er32(RNBC);
3272         adapter->stats.ruc += er32(RUC);
3273         adapter->stats.rfc += er32(RFC);
3274         adapter->stats.rjc += er32(RJC);
3275         adapter->stats.torl += er32(TORL);
3276         adapter->stats.torh += er32(TORH);
3277         adapter->stats.totl += er32(TOTL);
3278         adapter->stats.toth += er32(TOTH);
3279         adapter->stats.tpr += er32(TPR);
3280
3281         adapter->stats.ptc64 += er32(PTC64);
3282         adapter->stats.ptc127 += er32(PTC127);
3283         adapter->stats.ptc255 += er32(PTC255);
3284         adapter->stats.ptc511 += er32(PTC511);
3285         adapter->stats.ptc1023 += er32(PTC1023);
3286         adapter->stats.ptc1522 += er32(PTC1522);
3287
3288         adapter->stats.mptc += er32(MPTC);
3289         adapter->stats.bptc += er32(BPTC);
3290
3291         /* used for adaptive IFS */
3292
3293         hw->tx_packet_delta = er32(TPT);
3294         adapter->stats.tpt += hw->tx_packet_delta;
3295         hw->collision_delta = er32(COLC);
3296         adapter->stats.colc += hw->collision_delta;
3297
3298         if (hw->mac_type >= e1000_82543) {
3299                 adapter->stats.algnerrc += er32(ALGNERRC);
3300                 adapter->stats.rxerrc += er32(RXERRC);
3301                 adapter->stats.tncrs += er32(TNCRS);
3302                 adapter->stats.cexterr += er32(CEXTERR);
3303                 adapter->stats.tsctc += er32(TSCTC);
3304                 adapter->stats.tsctfc += er32(TSCTFC);
3305         }
3306
3307         /* Fill out the OS statistics structure */
3308         netdev->stats.multicast = adapter->stats.mprc;
3309         netdev->stats.collisions = adapter->stats.colc;
3310
3311         /* Rx Errors */
3312
3313         /* RLEC on some newer hardware can be incorrect so build
3314         * our own version based on RUC and ROC */
3315         netdev->stats.rx_errors = adapter->stats.rxerrc +
3316                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3317                 adapter->stats.ruc + adapter->stats.roc +
3318                 adapter->stats.cexterr;
3319         adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3320         netdev->stats.rx_length_errors = adapter->stats.rlerrc;
3321         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3322         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3323         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3324
3325         /* Tx Errors */
3326         adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3327         netdev->stats.tx_errors = adapter->stats.txerrc;
3328         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3329         netdev->stats.tx_window_errors = adapter->stats.latecol;
3330         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3331         if (hw->bad_tx_carr_stats_fd &&
3332             adapter->link_duplex == FULL_DUPLEX) {
3333                 netdev->stats.tx_carrier_errors = 0;
3334                 adapter->stats.tncrs = 0;
3335         }
3336
3337         /* Tx Dropped needs to be maintained elsewhere */
3338
3339         /* Phy Stats */
3340         if (hw->media_type == e1000_media_type_copper) {
3341                 if ((adapter->link_speed == SPEED_1000) &&
3342                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3343                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3344                         adapter->phy_stats.idle_errors += phy_tmp;
3345                 }
3346
3347                 if ((hw->mac_type <= e1000_82546) &&
3348                    (hw->phy_type == e1000_phy_m88) &&
3349                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3350                         adapter->phy_stats.receive_errors += phy_tmp;
3351         }
3352
3353         /* Management Stats */
3354         if (hw->has_smbus) {
3355                 adapter->stats.mgptc += er32(MGTPTC);
3356                 adapter->stats.mgprc += er32(MGTPRC);
3357                 adapter->stats.mgpdc += er32(MGTPDC);
3358         }
3359
3360         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3361 }
3362
3363 /**
3364  * e1000_intr - Interrupt Handler
3365  * @irq: interrupt number
3366  * @data: pointer to a network interface device structure
3367  **/
3368
3369 static irqreturn_t e1000_intr(int irq, void *data)
3370 {
3371         struct net_device *netdev = data;
3372         struct e1000_adapter *adapter = netdev_priv(netdev);
3373         struct e1000_hw *hw = &adapter->hw;
3374         u32 icr = er32(ICR);
3375
3376         if (unlikely((!icr) || test_bit(__E1000_DOWN, &adapter->flags)))
3377                 return IRQ_NONE;  /* Not our interrupt */
3378
3379         if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3380                 hw->get_link_status = 1;
3381                 /* guard against interrupt when we're going down */
3382                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3383                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3384         }
3385
3386         /* disable interrupts, without the synchronize_irq bit */
3387         ew32(IMC, ~0);
3388         E1000_WRITE_FLUSH();
3389
3390         if (likely(napi_schedule_prep(&adapter->napi))) {
3391                 adapter->total_tx_bytes = 0;
3392                 adapter->total_tx_packets = 0;
3393                 adapter->total_rx_bytes = 0;
3394                 adapter->total_rx_packets = 0;
3395                 __napi_schedule(&adapter->napi);
3396         } else {
3397                 /* this really should not happen! if it does it is basically a
3398                  * bug, but not a hard error, so enable ints and continue */
3399                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3400                         e1000_irq_enable(adapter);
3401         }
3402
3403         return IRQ_HANDLED;
3404 }
3405
3406 /**
3407  * e1000_clean - NAPI Rx polling callback
3408  * @adapter: board private structure
3409  **/
3410 static int e1000_clean(struct napi_struct *napi, int budget)
3411 {
3412         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
3413         int tx_clean_complete = 0, work_done = 0;
3414
3415         tx_clean_complete = e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);
3416
3417         adapter->clean_rx(adapter, &adapter->rx_ring[0], &work_done, budget);
3418
3419         if (!tx_clean_complete)
3420                 work_done = budget;
3421
3422         /* If budget not fully consumed, exit the polling mode */
3423         if (work_done < budget) {
3424                 if (likely(adapter->itr_setting & 3))
3425                         e1000_set_itr(adapter);
3426                 napi_complete(napi);
3427                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3428                         e1000_irq_enable(adapter);
3429         }
3430
3431         return work_done;
3432 }
3433
3434 /**
3435  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3436  * @adapter: board private structure
3437  **/
3438 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
3439                                struct e1000_tx_ring *tx_ring)
3440 {
3441         struct e1000_hw *hw = &adapter->hw;
3442         struct net_device *netdev = adapter->netdev;
3443         struct e1000_tx_desc *tx_desc, *eop_desc;
3444         struct e1000_buffer *buffer_info;
3445         unsigned int i, eop;
3446         unsigned int count = 0;
3447         unsigned int total_tx_bytes=0, total_tx_packets=0;
3448
3449         i = tx_ring->next_to_clean;
3450         eop = tx_ring->buffer_info[i].next_to_watch;
3451         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3452
3453         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
3454                (count < tx_ring->count)) {
3455                 bool cleaned = false;
3456                 for ( ; !cleaned; count++) {
3457                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3458                         buffer_info = &tx_ring->buffer_info[i];
3459                         cleaned = (i == eop);
3460
3461                         if (cleaned) {
3462                                 struct sk_buff *skb = buffer_info->skb;
3463                                 unsigned int segs, bytecount;
3464                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3465                                 /* multiply data chunks by size of headers */
3466                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3467                                             skb->len;
3468                                 total_tx_packets += segs;
3469                                 total_tx_bytes += bytecount;
3470                         }
3471                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3472                         tx_desc->upper.data = 0;
3473
3474                         if (unlikely(++i == tx_ring->count)) i = 0;
3475                 }
3476
3477                 eop = tx_ring->buffer_info[i].next_to_watch;
3478                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3479         }
3480
3481         tx_ring->next_to_clean = i;
3482
3483 #define TX_WAKE_THRESHOLD 32
3484         if (unlikely(count && netif_carrier_ok(netdev) &&
3485                      E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
3486                 /* Make sure that anybody stopping the queue after this
3487                  * sees the new next_to_clean.
3488                  */
3489                 smp_mb();
3490
3491                 if (netif_queue_stopped(netdev) &&
3492                     !(test_bit(__E1000_DOWN, &adapter->flags))) {
3493                         netif_wake_queue(netdev);
3494                         ++adapter->restart_queue;
3495                 }
3496         }
3497
3498         if (adapter->detect_tx_hung) {
3499                 /* Detect a transmit hang in hardware, this serializes the
3500                  * check with the clearing of time_stamp and movement of i */
3501                 adapter->detect_tx_hung = false;
3502                 if (tx_ring->buffer_info[eop].time_stamp &&
3503                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
3504                                (adapter->tx_timeout_factor * HZ)) &&
3505                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
3506
3507                         /* detected Tx unit hang */
3508                         e_err("Detected Tx Unit Hang\n"
3509                               "  Tx Queue             <%lu>\n"
3510                               "  TDH                  <%x>\n"
3511                               "  TDT                  <%x>\n"
3512                               "  next_to_use          <%x>\n"
3513                               "  next_to_clean        <%x>\n"
3514                               "buffer_info[next_to_clean]\n"
3515                               "  time_stamp           <%lx>\n"
3516                               "  next_to_watch        <%x>\n"
3517                               "  jiffies              <%lx>\n"
3518                               "  next_to_watch.status <%x>\n",
3519                                 (unsigned long)((tx_ring - adapter->tx_ring) /
3520                                         sizeof(struct e1000_tx_ring)),
3521                                 readl(hw->hw_addr + tx_ring->tdh),
3522                                 readl(hw->hw_addr + tx_ring->tdt),
3523                                 tx_ring->next_to_use,
3524                                 tx_ring->next_to_clean,
3525                                 tx_ring->buffer_info[eop].time_stamp,
3526                                 eop,
3527                                 jiffies,
3528                                 eop_desc->upper.fields.status);
3529                         netif_stop_queue(netdev);
3530                 }
3531         }
3532         adapter->total_tx_bytes += total_tx_bytes;
3533         adapter->total_tx_packets += total_tx_packets;
3534         netdev->stats.tx_bytes += total_tx_bytes;
3535         netdev->stats.tx_packets += total_tx_packets;
3536         return (count < tx_ring->count);
3537 }
3538
3539 /**
3540  * e1000_rx_checksum - Receive Checksum Offload for 82543
3541  * @adapter:     board private structure
3542  * @status_err:  receive descriptor status and error fields
3543  * @csum:        receive descriptor csum field
3544  * @sk_buff:     socket buffer with received data
3545  **/
3546
3547 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
3548                               u32 csum, struct sk_buff *skb)
3549 {
3550         struct e1000_hw *hw = &adapter->hw;
3551         u16 status = (u16)status_err;
3552         u8 errors = (u8)(status_err >> 24);
3553         skb->ip_summed = CHECKSUM_NONE;
3554
3555         /* 82543 or newer only */
3556         if (unlikely(hw->mac_type < e1000_82543)) return;
3557         /* Ignore Checksum bit is set */
3558         if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
3559         /* TCP/UDP checksum error bit is set */
3560         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
3561                 /* let the stack verify checksum errors */
3562                 adapter->hw_csum_err++;
3563                 return;
3564         }
3565         /* TCP/UDP Checksum has not been calculated */
3566         if (!(status & E1000_RXD_STAT_TCPCS))
3567                 return;
3568
3569         /* It must be a TCP or UDP packet with a valid checksum */
3570         if (likely(status & E1000_RXD_STAT_TCPCS)) {
3571                 /* TCP checksum is good */
3572                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3573         }
3574         adapter->hw_csum_good++;
3575 }
3576
3577 /**
3578  * e1000_consume_page - helper function
3579  **/
3580 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
3581                                u16 length)
3582 {
3583         bi->page = NULL;
3584         skb->len += length;
3585         skb->data_len += length;
3586         skb->truesize += length;
3587 }
3588
3589 /**
3590  * e1000_receive_skb - helper function to handle rx indications
3591  * @adapter: board private structure
3592  * @status: descriptor status field as written by hardware
3593  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3594  * @skb: pointer to sk_buff to be indicated to stack
3595  */
3596 static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
3597                               __le16 vlan, struct sk_buff *skb)
3598 {
3599         if (unlikely(adapter->vlgrp && (status & E1000_RXD_STAT_VP))) {
3600                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3601                                          le16_to_cpu(vlan) &
3602                                          E1000_RXD_SPC_VLAN_MASK);
3603         } else {
3604                 netif_receive_skb(skb);
3605         }
3606 }
3607
3608 /**
3609  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
3610  * @adapter: board private structure
3611  * @rx_ring: ring to clean
3612  * @work_done: amount of napi work completed this call
3613  * @work_to_do: max amount of work allowed for this call to do
3614  *
3615  * the return value indicates whether actual cleaning was done, there
3616  * is no guarantee that everything was cleaned
3617  */
3618 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
3619                                      struct e1000_rx_ring *rx_ring,
3620                                      int *work_done, int work_to_do)
3621 {
3622         struct e1000_hw *hw = &adapter->hw;
3623         struct net_device *netdev = adapter->netdev;
3624         struct pci_dev *pdev = adapter->pdev;
3625         struct e1000_rx_desc *rx_desc, *next_rxd;
3626         struct e1000_buffer *buffer_info, *next_buffer;
3627         unsigned long irq_flags;
3628         u32 length;
3629         unsigned int i;
3630         int cleaned_count = 0;
3631         bool cleaned = false;
3632         unsigned int total_rx_bytes=0, total_rx_packets=0;
3633
3634         i = rx_ring->next_to_clean;
3635         rx_desc = E1000_RX_DESC(*rx_ring, i);
3636         buffer_info = &rx_ring->buffer_info[i];
3637
3638         while (rx_desc->status & E1000_RXD_STAT_DD) {
3639                 struct sk_buff *skb;
3640                 u8 status;
3641
3642                 if (*work_done >= work_to_do)
3643                         break;
3644                 (*work_done)++;
3645
3646                 status = rx_desc->status;
3647                 skb = buffer_info->skb;
3648                 buffer_info->skb = NULL;
3649
3650                 if (++i == rx_ring->count) i = 0;
3651                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3652                 prefetch(next_rxd);
3653
3654                 next_buffer = &rx_ring->buffer_info[i];
3655
3656                 cleaned = true;
3657                 cleaned_count++;
3658                 dma_unmap_page(&pdev->dev, buffer_info->dma,
3659                                buffer_info->length, DMA_FROM_DEVICE);
3660                 buffer_info->dma = 0;
3661
3662                 length = le16_to_cpu(rx_desc->length);
3663
3664                 /* errors is only valid for DD + EOP descriptors */
3665                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
3666                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
3667                         u8 last_byte = *(skb->data + length - 1);
3668                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3669                                        last_byte)) {
3670                                 spin_lock_irqsave(&adapter->stats_lock,
3671                                                   irq_flags);
3672                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3673                                                        length, skb->data);
3674                                 spin_unlock_irqrestore(&adapter->stats_lock,
3675                                                        irq_flags);
3676                                 length--;
3677                         } else {
3678                                 /* recycle both page and skb */
3679                                 buffer_info->skb = skb;
3680                                 /* an error means any chain goes out the window
3681                                  * too */
3682                                 if (rx_ring->rx_skb_top)
3683                                         dev_kfree_skb(rx_ring->rx_skb_top);
3684                                 rx_ring->rx_skb_top = NULL;
3685                                 goto next_desc;
3686                         }
3687                 }
3688
3689 #define rxtop rx_ring->rx_skb_top
3690                 if (!(status & E1000_RXD_STAT_EOP)) {
3691                         /* this descriptor is only the beginning (or middle) */
3692                         if (!rxtop) {
3693                                 /* this is the beginning of a chain */
3694                                 rxtop = skb;
3695                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
3696                                                    0, length);
3697                         } else {
3698                                 /* this is the middle of a chain */
3699                                 skb_fill_page_desc(rxtop,
3700                                     skb_shinfo(rxtop)->nr_frags,
3701                                     buffer_info->page, 0, length);
3702                                 /* re-use the skb, only consumed the page */
3703                                 buffer_info->skb = skb;
3704                         }
3705                         e1000_consume_page(buffer_info, rxtop, length);
3706                         goto next_desc;
3707                 } else {
3708                         if (rxtop) {
3709                                 /* end of the chain */
3710                                 skb_fill_page_desc(rxtop,
3711                                     skb_shinfo(rxtop)->nr_frags,
3712                                     buffer_info->page, 0, length);
3713                                 /* re-use the current skb, we only consumed the
3714                                  * page */
3715                                 buffer_info->skb = skb;
3716                                 skb = rxtop;
3717                                 rxtop = NULL;
3718                                 e1000_consume_page(buffer_info, skb, length);
3719                         } else {
3720                                 /* no chain, got EOP, this buf is the packet
3721                                  * copybreak to save the put_page/alloc_page */
3722                                 if (length <= copybreak &&
3723                                     skb_tailroom(skb) >= length) {
3724                                         u8 *vaddr;
3725                                         vaddr = kmap_atomic(buffer_info->page,
3726                                                             KM_SKB_DATA_SOFTIRQ);
3727                                         memcpy(skb_tail_pointer(skb), vaddr, length);
3728                                         kunmap_atomic(vaddr,
3729                                                       KM_SKB_DATA_SOFTIRQ);
3730                                         /* re-use the page, so don't erase
3731                                          * buffer_info->page */
3732                                         skb_put(skb, length);
3733                                 } else {
3734                                         skb_fill_page_desc(skb, 0,
3735                                                            buffer_info->page, 0,
3736                                                            length);
3737                                         e1000_consume_page(buffer_info, skb,
3738                                                            length);
3739                                 }
3740                         }
3741                 }
3742
3743                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
3744                 e1000_rx_checksum(adapter,
3745                                   (u32)(status) |
3746                                   ((u32)(rx_desc->errors) << 24),
3747                                   le16_to_cpu(rx_desc->csum), skb);
3748
3749                 pskb_trim(skb, skb->len - 4);
3750
3751                 /* probably a little skewed due to removing CRC */
3752                 total_rx_bytes += skb->len;
3753                 total_rx_packets++;
3754
3755                 /* eth type trans needs skb->data to point to something */
3756                 if (!pskb_may_pull(skb, ETH_HLEN)) {
3757                         e_err("pskb_may_pull failed.\n");
3758                         dev_kfree_skb(skb);
3759                         goto next_desc;
3760                 }
3761
3762                 skb->protocol = eth_type_trans(skb, netdev);
3763
3764                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3765
3766 next_desc:
3767                 rx_desc->status = 0;
3768
3769                 /* return some buffers to hardware, one at a time is too slow */
3770                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3771                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3772                         cleaned_count = 0;
3773                 }
3774
3775                 /* use prefetched values */
3776                 rx_desc = next_rxd;
3777                 buffer_info = next_buffer;
3778         }
3779         rx_ring->next_to_clean = i;
3780
3781         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3782         if (cleaned_count)
3783                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3784
3785         adapter->total_rx_packets += total_rx_packets;
3786         adapter->total_rx_bytes += total_rx_bytes;
3787         netdev->stats.rx_bytes += total_rx_bytes;
3788         netdev->stats.rx_packets += total_rx_packets;
3789         return cleaned;
3790 }
3791
3792 /**
3793  * e1000_clean_rx_irq - Send received data up the network stack; legacy
3794  * @adapter: board private structure
3795  * @rx_ring: ring to clean
3796  * @work_done: amount of napi work completed this call
3797  * @work_to_do: max amount of work allowed for this call to do
3798  */
3799 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
3800                                struct e1000_rx_ring *rx_ring,
3801                                int *work_done, int work_to_do)
3802 {
3803         struct e1000_hw *hw = &adapter->hw;
3804         struct net_device *netdev = adapter->netdev;
3805         struct pci_dev *pdev = adapter->pdev;
3806         struct e1000_rx_desc *rx_desc, *next_rxd;
3807         struct e1000_buffer *buffer_info, *next_buffer;
3808         unsigned long flags;
3809         u32 length;
3810         unsigned int i;
3811         int cleaned_count = 0;
3812         bool cleaned = false;
3813         unsigned int total_rx_bytes=0, total_rx_packets=0;
3814
3815         i = rx_ring->next_to_clean;
3816         rx_desc = E1000_RX_DESC(*rx_ring, i);
3817         buffer_info = &rx_ring->buffer_info[i];
3818
3819         while (rx_desc->status & E1000_RXD_STAT_DD) {
3820                 struct sk_buff *skb;
3821                 u8 status;
3822
3823                 if (*work_done >= work_to_do)
3824                         break;
3825                 (*work_done)++;
3826
3827                 status = rx_desc->status;
3828                 skb = buffer_info->skb;
3829                 buffer_info->skb = NULL;
3830
3831                 prefetch(skb->data - NET_IP_ALIGN);
3832
3833                 if (++i == rx_ring->count) i = 0;
3834                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3835                 prefetch(next_rxd);
3836
3837                 next_buffer = &rx_ring->buffer_info[i];
3838
3839                 cleaned = true;
3840                 cleaned_count++;
3841                 dma_unmap_single(&pdev->dev, buffer_info->dma,
3842                                  buffer_info->length, DMA_FROM_DEVICE);
3843                 buffer_info->dma = 0;
3844
3845                 length = le16_to_cpu(rx_desc->length);
3846                 /* !EOP means multiple descriptors were used to store a single
3847                  * packet, if thats the case we need to toss it.  In fact, we
3848                  * to toss every packet with the EOP bit clear and the next
3849                  * frame that _does_ have the EOP bit set, as it is by
3850                  * definition only a frame fragment
3851                  */
3852                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
3853                         adapter->discarding = true;
3854
3855                 if (adapter->discarding) {
3856                         /* All receives must fit into a single buffer */
3857                         e_info("Receive packet consumed multiple buffers\n");
3858                         /* recycle */
3859                         buffer_info->skb = skb;
3860                         if (status & E1000_RXD_STAT_EOP)
3861                                 adapter->discarding = false;
3862                         goto next_desc;
3863                 }
3864
3865                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
3866                         u8 last_byte = *(skb->data + length - 1);
3867                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3868                                        last_byte)) {
3869                                 spin_lock_irqsave(&adapter->stats_lock, flags);
3870                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3871                                                        length, skb->data);
3872                                 spin_unlock_irqrestore(&adapter->stats_lock,
3873                                                        flags);
3874                                 length--;
3875                         } else {
3876                                 /* recycle */
3877                                 buffer_info->skb = skb;
3878                                 goto next_desc;
3879                         }
3880                 }
3881
3882                 /* adjust length to remove Ethernet CRC, this must be
3883                  * done after the TBI_ACCEPT workaround above */
3884                 length -= 4;
3885
3886                 /* probably a little skewed due to removing CRC */
3887                 total_rx_bytes += length;
3888                 total_rx_packets++;
3889
3890                 /* code added for copybreak, this should improve
3891                  * performance for small packets with large amounts
3892                  * of reassembly being done in the stack */
3893                 if (length < copybreak) {
3894                         struct sk_buff *new_skb =
3895                             netdev_alloc_skb_ip_align(netdev, length);
3896                         if (new_skb) {
3897                                 skb_copy_to_linear_data_offset(new_skb,
3898                                                                -NET_IP_ALIGN,
3899                                                                (skb->data -
3900                                                                 NET_IP_ALIGN),
3901                                                                (length +
3902                                                                 NET_IP_ALIGN));
3903                                 /* save the skb in buffer_info as good */
3904                                 buffer_info->skb = skb;
3905                                 skb = new_skb;
3906                         }
3907                         /* else just continue with the old one */
3908                 }
3909                 /* end copybreak code */
3910                 skb_put(skb, length);
3911
3912                 /* Receive Checksum Offload */
3913                 e1000_rx_checksum(adapter,
3914                                   (u32)(status) |
3915                                   ((u32)(rx_desc->errors) << 24),
3916                                   le16_to_cpu(rx_desc->csum), skb);
3917
3918                 skb->protocol = eth_type_trans(skb, netdev);
3919
3920                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3921
3922 next_desc:
3923                 rx_desc->status = 0;
3924
3925                 /* return some buffers to hardware, one at a time is too slow */
3926                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3927                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3928                         cleaned_count = 0;
3929                 }
3930
3931                 /* use prefetched values */
3932                 rx_desc = next_rxd;
3933                 buffer_info = next_buffer;
3934         }
3935         rx_ring->next_to_clean = i;
3936
3937         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3938         if (cleaned_count)
3939                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3940
3941         adapter->total_rx_packets += total_rx_packets;
3942         adapter->total_rx_bytes += total_rx_bytes;
3943         netdev->stats.rx_bytes += total_rx_bytes;
3944         netdev->stats.rx_packets += total_rx_packets;
3945         return cleaned;
3946 }
3947
3948 /**
3949  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
3950  * @adapter: address of board private structure
3951  * @rx_ring: pointer to receive ring structure
3952  * @cleaned_count: number of buffers to allocate this pass
3953  **/
3954
3955 static void
3956 e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
3957                              struct e1000_rx_ring *rx_ring, int cleaned_count)
3958 {
3959         struct net_device *netdev = adapter->netdev;
3960         struct pci_dev *pdev = adapter->pdev;
3961         struct e1000_rx_desc *rx_desc;
3962         struct e1000_buffer *buffer_info;
3963         struct sk_buff *skb;
3964         unsigned int i;
3965         unsigned int bufsz = 256 - 16 /*for skb_reserve */ ;
3966
3967         i = rx_ring->next_to_use;
3968         buffer_info = &rx_ring->buffer_info[i];
3969
3970         while (cleaned_count--) {
3971                 skb = buffer_info->skb;
3972                 if (skb) {
3973                         skb_trim(skb, 0);
3974                         goto check_page;
3975                 }
3976
3977                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
3978                 if (unlikely(!skb)) {
3979                         /* Better luck next round */
3980                         adapter->alloc_rx_buff_failed++;
3981                         break;
3982                 }
3983
3984                 /* Fix for errata 23, can't cross 64kB boundary */
3985                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3986                         struct sk_buff *oldskb = skb;
3987                         e_err("skb align check failed: %u bytes at %p\n",
3988                               bufsz, skb->data);
3989                         /* Try again, without freeing the previous */
3990                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
3991                         /* Failed allocation, critical failure */
3992                         if (!skb) {
3993                                 dev_kfree_skb(oldskb);
3994                                 adapter->alloc_rx_buff_failed++;
3995                                 break;
3996                         }
3997
3998                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3999                                 /* give up */
4000                                 dev_kfree_skb(skb);
4001                                 dev_kfree_skb(oldskb);
4002                                 break; /* while (cleaned_count--) */
4003                         }
4004
4005                         /* Use new allocation */
4006                         dev_kfree_skb(oldskb);
4007                 }
4008                 buffer_info->skb = skb;
4009                 buffer_info->length = adapter->rx_buffer_len;
4010 check_page:
4011                 /* allocate a new page if necessary */
4012                 if (!buffer_info->page) {
4013                         buffer_info->page = alloc_page(GFP_ATOMIC);
4014                         if (unlikely(!buffer_info->page)) {
4015                                 adapter->alloc_rx_buff_failed++;
4016                                 break;
4017                         }
4018                 }
4019
4020                 if (!buffer_info->dma) {
4021                         buffer_info->dma = dma_map_page(&pdev->dev,
4022                                                         buffer_info->page, 0,
4023                                                         buffer_info->length,
4024                                                         DMA_FROM_DEVICE);
4025                         if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4026                                 put_page(buffer_info->page);
4027                                 dev_kfree_skb(skb);
4028                                 buffer_info->page = NULL;
4029                                 buffer_info->skb = NULL;
4030                                 buffer_info->dma = 0;
4031                                 adapter->alloc_rx_buff_failed++;
4032                                 break; /* while !buffer_info->skb */
4033                         }
4034                 }
4035
4036                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4037                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4038
4039                 if (unlikely(++i == rx_ring->count))
4040                         i = 0;
4041                 buffer_info = &rx_ring->buffer_info[i];
4042         }
4043
4044         if (likely(rx_ring->next_to_use != i)) {
4045                 rx_ring->next_to_use = i;
4046                 if (unlikely(i-- == 0))
4047                         i = (rx_ring->count - 1);
4048
4049                 /* Force memory writes to complete before letting h/w
4050                  * know there are new descriptors to fetch.  (Only
4051                  * applicable for weak-ordered memory model archs,
4052                  * such as IA-64). */
4053                 wmb();
4054                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4055         }
4056 }
4057
4058 /**
4059  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4060  * @adapter: address of board private structure
4061  **/
4062
4063 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4064                                    struct e1000_rx_ring *rx_ring,
4065                                    int cleaned_count)
4066 {
4067         struct e1000_hw *hw = &adapter->hw;
4068         struct net_device *netdev = adapter->netdev;
4069         struct pci_dev *pdev = adapter->pdev;
4070         struct e1000_rx_desc *rx_desc;
4071         struct e1000_buffer *buffer_info;
4072         struct sk_buff *skb;
4073         unsigned int i;
4074         unsigned int bufsz = adapter->rx_buffer_len;
4075
4076         i = rx_ring->next_to_use;
4077         buffer_info = &rx_ring->buffer_info[i];
4078
4079         while (cleaned_count--) {
4080                 skb = buffer_info->skb;
4081                 if (skb) {
4082                         skb_trim(skb, 0);
4083                         goto map_skb;
4084                 }
4085
4086                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4087                 if (unlikely(!skb)) {
4088                         /* Better luck next round */
4089                         adapter->alloc_rx_buff_failed++;
4090                         break;
4091                 }
4092
4093                 /* Fix for errata 23, can't cross 64kB boundary */
4094                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4095                         struct sk_buff *oldskb = skb;
4096                         e_err("skb align check failed: %u bytes at %p\n",
4097                               bufsz, skb->data);
4098                         /* Try again, without freeing the previous */
4099                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4100                         /* Failed allocation, critical failure */
4101                         if (!skb) {
4102                                 dev_kfree_skb(oldskb);
4103                                 adapter->alloc_rx_buff_failed++;
4104                                 break;
4105                         }
4106
4107                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4108                                 /* give up */
4109                                 dev_kfree_skb(skb);
4110                                 dev_kfree_skb(oldskb);
4111                                 adapter->alloc_rx_buff_failed++;
4112                                 break; /* while !buffer_info->skb */
4113                         }
4114
4115                         /* Use new allocation */
4116                         dev_kfree_skb(oldskb);
4117                 }
4118                 buffer_info->skb = skb;
4119                 buffer_info->length = adapter->rx_buffer_len;
4120 map_skb:
4121                 buffer_info->dma = dma_map_single(&pdev->dev,
4122                                                   skb->data,
4123                                                   buffer_info->length,
4124                                                   DMA_FROM_DEVICE);
4125                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4126                         dev_kfree_skb(skb);
4127                         buffer_info->skb = NULL;
4128                         buffer_info->dma = 0;
4129                         adapter->alloc_rx_buff_failed++;
4130                         break; /* while !buffer_info->skb */
4131                 }
4132
4133                 /*
4134                  * XXX if it was allocated cleanly it will never map to a
4135                  * boundary crossing
4136                  */
4137
4138                 /* Fix for errata 23, can't cross 64kB boundary */
4139                 if (!e1000_check_64k_bound(adapter,
4140                                         (void *)(unsigned long)buffer_info->dma,
4141                                         adapter->rx_buffer_len)) {
4142                         e_err("dma align check failed: %u bytes at %p\n",
4143                               adapter->rx_buffer_len,
4144                               (void *)(unsigned long)buffer_info->dma);
4145                         dev_kfree_skb(skb);
4146                         buffer_info->skb = NULL;
4147
4148                         dma_unmap_single(&pdev->dev, buffer_info->dma,
4149                                          adapter->rx_buffer_len,
4150                                          DMA_FROM_DEVICE);
4151                         buffer_info->dma = 0;
4152
4153                         adapter->alloc_rx_buff_failed++;
4154                         break; /* while !buffer_info->skb */
4155                 }
4156                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4157                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4158
4159                 if (unlikely(++i == rx_ring->count))
4160                         i = 0;
4161                 buffer_info = &rx_ring->buffer_info[i];
4162         }
4163
4164         if (likely(rx_ring->next_to_use != i)) {
4165                 rx_ring->next_to_use = i;
4166                 if (unlikely(i-- == 0))
4167                         i = (rx_ring->count - 1);
4168
4169                 /* Force memory writes to complete before letting h/w
4170                  * know there are new descriptors to fetch.  (Only
4171                  * applicable for weak-ordered memory model archs,
4172                  * such as IA-64). */
4173                 wmb();
4174                 writel(i, hw->hw_addr + rx_ring->rdt);
4175         }
4176 }
4177
4178 /**
4179  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4180  * @adapter:
4181  **/
4182
4183 static void e1000_smartspeed(struct e1000_adapter *adapter)
4184 {
4185         struct e1000_hw *hw = &adapter->hw;
4186         u16 phy_status;
4187         u16 phy_ctrl;
4188
4189         if ((hw->phy_type != e1000_phy_igp) || !hw->autoneg ||
4190            !(hw->autoneg_advertised & ADVERTISE_1000_FULL))
4191                 return;
4192
4193         if (adapter->smartspeed == 0) {
4194                 /* If Master/Slave config fault is asserted twice,
4195                  * we assume back-to-back */
4196                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4197                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4198                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4199                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4200                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4201                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4202                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4203                         e1000_write_phy_reg(hw, PHY_1000T_CTRL,
4204                                             phy_ctrl);
4205                         adapter->smartspeed++;
4206                         if (!e1000_phy_setup_autoneg(hw) &&
4207                            !e1000_read_phy_reg(hw, PHY_CTRL,
4208                                                &phy_ctrl)) {
4209                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4210                                              MII_CR_RESTART_AUTO_NEG);
4211                                 e1000_write_phy_reg(hw, PHY_CTRL,
4212                                                     phy_ctrl);
4213                         }
4214                 }
4215                 return;
4216         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4217                 /* If still no link, perhaps using 2/3 pair cable */
4218                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4219                 phy_ctrl |= CR_1000T_MS_ENABLE;
4220                 e1000_write_phy_reg(hw, PHY_1000T_CTRL, phy_ctrl);
4221                 if (!e1000_phy_setup_autoneg(hw) &&
4222                    !e1000_read_phy_reg(hw, PHY_CTRL, &phy_ctrl)) {
4223                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4224                                      MII_CR_RESTART_AUTO_NEG);
4225                         e1000_write_phy_reg(hw, PHY_CTRL, phy_ctrl);
4226                 }
4227         }
4228         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4229         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4230                 adapter->smartspeed = 0;
4231 }
4232
4233 /**
4234  * e1000_ioctl -
4235  * @netdev:
4236  * @ifreq:
4237  * @cmd:
4238  **/
4239
4240 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4241 {
4242         switch (cmd) {
4243         case SIOCGMIIPHY:
4244         case SIOCGMIIREG:
4245         case SIOCSMIIREG:
4246                 return e1000_mii_ioctl(netdev, ifr, cmd);
4247         default:
4248                 return -EOPNOTSUPP;
4249         }
4250 }
4251
4252 /**
4253  * e1000_mii_ioctl -
4254  * @netdev:
4255  * @ifreq:
4256  * @cmd:
4257  **/
4258
4259 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4260                            int cmd)
4261 {
4262         struct e1000_adapter *adapter = netdev_priv(netdev);
4263         struct e1000_hw *hw = &adapter->hw;
4264         struct mii_ioctl_data *data = if_mii(ifr);
4265         int retval;
4266         u16 mii_reg;
4267         u16 spddplx;
4268         unsigned long flags;
4269
4270         if (hw->media_type != e1000_media_type_copper)
4271                 return -EOPNOTSUPP;
4272
4273         switch (cmd) {
4274         case SIOCGMIIPHY:
4275                 data->phy_id = hw->phy_addr;
4276                 break;
4277         case SIOCGMIIREG:
4278                 spin_lock_irqsave(&adapter->stats_lock, flags);
4279                 if (e1000_read_phy_reg(hw, data->reg_num & 0x1F,
4280                                    &data->val_out)) {
4281                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4282                         return -EIO;
4283                 }
4284                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4285                 break;
4286         case SIOCSMIIREG:
4287                 if (data->reg_num & ~(0x1F))
4288                         return -EFAULT;
4289                 mii_reg = data->val_in;
4290                 spin_lock_irqsave(&adapter->stats_lock, flags);
4291                 if (e1000_write_phy_reg(hw, data->reg_num,
4292                                         mii_reg)) {
4293                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4294                         return -EIO;
4295                 }
4296                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4297                 if (hw->media_type == e1000_media_type_copper) {
4298                         switch (data->reg_num) {
4299                         case PHY_CTRL:
4300                                 if (mii_reg & MII_CR_POWER_DOWN)
4301                                         break;
4302                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4303                                         hw->autoneg = 1;
4304                                         hw->autoneg_advertised = 0x2F;
4305                                 } else {
4306                                         if (mii_reg & 0x40)
4307                                                 spddplx = SPEED_1000;
4308                                         else if (mii_reg & 0x2000)
4309                                                 spddplx = SPEED_100;
4310                                         else
4311                                                 spddplx = SPEED_10;
4312                                         spddplx += (mii_reg & 0x100)
4313                                                    ? DUPLEX_FULL :
4314                                                    DUPLEX_HALF;
4315                                         retval = e1000_set_spd_dplx(adapter,
4316                                                                     spddplx);
4317                                         if (retval)
4318                                                 return retval;
4319                                 }
4320                                 if (netif_running(adapter->netdev))
4321                                         e1000_reinit_locked(adapter);
4322                                 else
4323                                         e1000_reset(adapter);
4324                                 break;
4325                         case M88E1000_PHY_SPEC_CTRL:
4326                         case M88E1000_EXT_PHY_SPEC_CTRL:
4327                                 if (e1000_phy_reset(hw))
4328                                         return -EIO;
4329                                 break;
4330                         }
4331                 } else {
4332                         switch (data->reg_num) {
4333                         case PHY_CTRL:
4334                                 if (mii_reg & MII_CR_POWER_DOWN)
4335                                         break;
4336                                 if (netif_running(adapter->netdev))
4337                                         e1000_reinit_locked(adapter);
4338                                 else
4339                                         e1000_reset(adapter);
4340                                 break;
4341                         }
4342                 }
4343                 break;
4344         default:
4345                 return -EOPNOTSUPP;
4346         }
4347         return E1000_SUCCESS;
4348 }
4349
4350 void e1000_pci_set_mwi(struct e1000_hw *hw)
4351 {
4352         struct e1000_adapter *adapter = hw->back;
4353         int ret_val = pci_set_mwi(adapter->pdev);
4354
4355         if (ret_val)
4356                 e_err("Error in setting MWI\n");
4357 }
4358
4359 void e1000_pci_clear_mwi(struct e1000_hw *hw)
4360 {
4361         struct e1000_adapter *adapter = hw->back;
4362
4363         pci_clear_mwi(adapter->pdev);
4364 }
4365
4366 int e1000_pcix_get_mmrbc(struct e1000_hw *hw)
4367 {
4368         struct e1000_adapter *adapter = hw->back;
4369         return pcix_get_mmrbc(adapter->pdev);
4370 }
4371
4372 void e1000_pcix_set_mmrbc(struct e1000_hw *hw, int mmrbc)
4373 {
4374         struct e1000_adapter *adapter = hw->back;
4375         pcix_set_mmrbc(adapter->pdev, mmrbc);
4376 }
4377
4378 void e1000_io_write(struct e1000_hw *hw, unsigned long port, u32 value)
4379 {
4380         outl(value, port);
4381 }
4382
4383 static void e1000_vlan_rx_register(struct net_device *netdev,
4384                                    struct vlan_group *grp)
4385 {
4386         struct e1000_adapter *adapter = netdev_priv(netdev);
4387         struct e1000_hw *hw = &adapter->hw;
4388         u32 ctrl, rctl;
4389
4390         if (!test_bit(__E1000_DOWN, &adapter->flags))
4391                 e1000_irq_disable(adapter);
4392         adapter->vlgrp = grp;
4393
4394         if (grp) {
4395                 /* enable VLAN tag insert/strip */
4396                 ctrl = er32(CTRL);
4397                 ctrl |= E1000_CTRL_VME;
4398                 ew32(CTRL, ctrl);
4399
4400                 /* enable VLAN receive filtering */
4401                 rctl = er32(RCTL);
4402                 rctl &= ~E1000_RCTL_CFIEN;
4403                 if (!(netdev->flags & IFF_PROMISC))
4404                         rctl |= E1000_RCTL_VFE;
4405                 ew32(RCTL, rctl);
4406                 e1000_update_mng_vlan(adapter);
4407         } else {
4408                 /* disable VLAN tag insert/strip */
4409                 ctrl = er32(CTRL);
4410                 ctrl &= ~E1000_CTRL_VME;
4411                 ew32(CTRL, ctrl);
4412
4413                 /* disable VLAN receive filtering */
4414                 rctl = er32(RCTL);
4415                 rctl &= ~E1000_RCTL_VFE;
4416                 ew32(RCTL, rctl);
4417
4418                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
4419                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4420                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4421                 }
4422         }
4423
4424         if (!test_bit(__E1000_DOWN, &adapter->flags))
4425                 e1000_irq_enable(adapter);
4426 }
4427
4428 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4429 {
4430         struct e1000_adapter *adapter = netdev_priv(netdev);
4431         struct e1000_hw *hw = &adapter->hw;
4432         u32 vfta, index;
4433
4434         if ((hw->mng_cookie.status &
4435              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4436             (vid == adapter->mng_vlan_id))
4437                 return;
4438         /* add VID to filter table */
4439         index = (vid >> 5) & 0x7F;
4440         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4441         vfta |= (1 << (vid & 0x1F));
4442         e1000_write_vfta(hw, index, vfta);
4443 }
4444
4445 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
4446 {
4447         struct e1000_adapter *adapter = netdev_priv(netdev);
4448         struct e1000_hw *hw = &adapter->hw;
4449         u32 vfta, index;
4450
4451         if (!test_bit(__E1000_DOWN, &adapter->flags))
4452                 e1000_irq_disable(adapter);
4453         vlan_group_set_device(adapter->vlgrp, vid, NULL);
4454         if (!test_bit(__E1000_DOWN, &adapter->flags))
4455                 e1000_irq_enable(adapter);
4456
4457         /* remove VID from filter table */
4458         index = (vid >> 5) & 0x7F;
4459         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4460         vfta &= ~(1 << (vid & 0x1F));
4461         e1000_write_vfta(hw, index, vfta);
4462 }
4463
4464 static void e1000_restore_vlan(struct e1000_adapter *adapter)
4465 {
4466         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
4467
4468         if (adapter->vlgrp) {
4469                 u16 vid;
4470                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
4471                         if (!vlan_group_get_device(adapter->vlgrp, vid))
4472                                 continue;
4473                         e1000_vlan_rx_add_vid(adapter->netdev, vid);
4474                 }
4475         }
4476 }
4477
4478 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx)
4479 {
4480         struct e1000_hw *hw = &adapter->hw;
4481
4482         hw->autoneg = 0;
4483
4484         /* Fiber NICs only allow 1000 gbps Full duplex */
4485         if ((hw->media_type == e1000_media_type_fiber) &&
4486                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
4487                 e_err("Unsupported Speed/Duplex configuration\n");
4488                 return -EINVAL;
4489         }
4490
4491         switch (spddplx) {
4492         case SPEED_10 + DUPLEX_HALF:
4493                 hw->forced_speed_duplex = e1000_10_half;
4494                 break;
4495         case SPEED_10 + DUPLEX_FULL:
4496                 hw->forced_speed_duplex = e1000_10_full;
4497                 break;
4498         case SPEED_100 + DUPLEX_HALF:
4499                 hw->forced_speed_duplex = e1000_100_half;
4500                 break;
4501         case SPEED_100 + DUPLEX_FULL:
4502                 hw->forced_speed_duplex = e1000_100_full;
4503                 break;
4504         case SPEED_1000 + DUPLEX_FULL:
4505                 hw->autoneg = 1;
4506                 hw->autoneg_advertised = ADVERTISE_1000_FULL;
4507                 break;
4508         case SPEED_1000 + DUPLEX_HALF: /* not supported */
4509         default:
4510                 e_err("Unsupported Speed/Duplex configuration\n");
4511                 return -EINVAL;
4512         }
4513         return 0;
4514 }
4515
4516 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4517 {
4518         struct net_device *netdev = pci_get_drvdata(pdev);
4519         struct e1000_adapter *adapter = netdev_priv(netdev);
4520         struct e1000_hw *hw = &adapter->hw;
4521         u32 ctrl, ctrl_ext, rctl, status;
4522         u32 wufc = adapter->wol;
4523 #ifdef CONFIG_PM
4524         int retval = 0;
4525 #endif
4526
4527         netif_device_detach(netdev);
4528
4529         if (netif_running(netdev)) {
4530                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
4531                 e1000_down(adapter);
4532         }
4533
4534 #ifdef CONFIG_PM
4535         retval = pci_save_state(pdev);
4536         if (retval)
4537                 return retval;
4538 #endif
4539
4540         status = er32(STATUS);
4541         if (status & E1000_STATUS_LU)
4542                 wufc &= ~E1000_WUFC_LNKC;
4543
4544         if (wufc) {
4545                 e1000_setup_rctl(adapter);
4546                 e1000_set_rx_mode(netdev);
4547
4548                 /* turn on all-multi mode if wake on multicast is enabled */
4549                 if (wufc & E1000_WUFC_MC) {
4550                         rctl = er32(RCTL);
4551                         rctl |= E1000_RCTL_MPE;
4552                         ew32(RCTL, rctl);
4553                 }
4554
4555                 if (hw->mac_type >= e1000_82540) {
4556                         ctrl = er32(CTRL);
4557                         /* advertise wake from D3Cold */
4558                         #define E1000_CTRL_ADVD3WUC 0x00100000
4559                         /* phy power management enable */
4560                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4561                         ctrl |= E1000_CTRL_ADVD3WUC |
4562                                 E1000_CTRL_EN_PHY_PWR_MGMT;
4563                         ew32(CTRL, ctrl);
4564                 }
4565
4566                 if (hw->media_type == e1000_media_type_fiber ||
4567                     hw->media_type == e1000_media_type_internal_serdes) {
4568                         /* keep the laser running in D3 */
4569                         ctrl_ext = er32(CTRL_EXT);
4570                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4571                         ew32(CTRL_EXT, ctrl_ext);
4572                 }
4573
4574                 ew32(WUC, E1000_WUC_PME_EN);
4575                 ew32(WUFC, wufc);
4576         } else {
4577                 ew32(WUC, 0);
4578                 ew32(WUFC, 0);
4579         }
4580
4581         e1000_release_manageability(adapter);
4582
4583         *enable_wake = !!wufc;
4584
4585         /* make sure adapter isn't asleep if manageability is enabled */
4586         if (adapter->en_mng_pt)
4587                 *enable_wake = true;
4588
4589         if (netif_running(netdev))
4590                 e1000_free_irq(adapter);
4591
4592         pci_disable_device(pdev);
4593
4594         return 0;
4595 }
4596
4597 #ifdef CONFIG_PM
4598 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4599 {
4600         int retval;
4601         bool wake;
4602
4603         retval = __e1000_shutdown(pdev, &wake);
4604         if (retval)
4605                 return retval;
4606
4607         if (wake) {
4608                 pci_prepare_to_sleep(pdev);
4609         } else {
4610                 pci_wake_from_d3(pdev, false);
4611                 pci_set_power_state(pdev, PCI_D3hot);
4612         }
4613
4614         return 0;
4615 }
4616
4617 static int e1000_resume(struct pci_dev *pdev)
4618 {
4619         struct net_device *netdev = pci_get_drvdata(pdev);
4620         struct e1000_adapter *adapter = netdev_priv(netdev);
4621         struct e1000_hw *hw = &adapter->hw;
4622         u32 err;
4623
4624         pci_set_power_state(pdev, PCI_D0);
4625         pci_restore_state(pdev);
4626         pci_save_state(pdev);
4627
4628         if (adapter->need_ioport)
4629                 err = pci_enable_device(pdev);
4630         else
4631                 err = pci_enable_device_mem(pdev);
4632         if (err) {
4633                 pr_err("Cannot enable PCI device from suspend\n");
4634                 return err;
4635         }
4636         pci_set_master(pdev);
4637
4638         pci_enable_wake(pdev, PCI_D3hot, 0);
4639         pci_enable_wake(pdev, PCI_D3cold, 0);
4640
4641         if (netif_running(netdev)) {
4642                 err = e1000_request_irq(adapter);
4643                 if (err)
4644                         return err;
4645         }
4646
4647         e1000_power_up_phy(adapter);
4648         e1000_reset(adapter);
4649         ew32(WUS, ~0);
4650
4651         e1000_init_manageability(adapter);
4652
4653         if (netif_running(netdev))
4654                 e1000_up(adapter);
4655
4656         netif_device_attach(netdev);
4657
4658         return 0;
4659 }
4660 #endif
4661
4662 static void e1000_shutdown(struct pci_dev *pdev)
4663 {
4664         bool wake;
4665
4666         __e1000_shutdown(pdev, &wake);
4667
4668         if (system_state == SYSTEM_POWER_OFF) {
4669                 pci_wake_from_d3(pdev, wake);
4670                 pci_set_power_state(pdev, PCI_D3hot);
4671         }
4672 }
4673
4674 #ifdef CONFIG_NET_POLL_CONTROLLER
4675 /*
4676  * Polling 'interrupt' - used by things like netconsole to send skbs
4677  * without having to re-enable interrupts. It's not called while
4678  * the interrupt routine is executing.
4679  */
4680 static void e1000_netpoll(struct net_device *netdev)
4681 {
4682         struct e1000_adapter *adapter = netdev_priv(netdev);
4683
4684         disable_irq(adapter->pdev->irq);
4685         e1000_intr(adapter->pdev->irq, netdev);
4686         enable_irq(adapter->pdev->irq);
4687 }
4688 #endif
4689
4690 /**
4691  * e1000_io_error_detected - called when PCI error is detected
4692  * @pdev: Pointer to PCI device
4693  * @state: The current pci connection state
4694  *
4695  * This function is called after a PCI bus error affecting
4696  * this device has been detected.
4697  */
4698 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4699                                                 pci_channel_state_t state)
4700 {
4701         struct net_device *netdev = pci_get_drvdata(pdev);
4702         struct e1000_adapter *adapter = netdev_priv(netdev);
4703
4704         netif_device_detach(netdev);
4705
4706         if (state == pci_channel_io_perm_failure)
4707                 return PCI_ERS_RESULT_DISCONNECT;
4708
4709         if (netif_running(netdev))
4710                 e1000_down(adapter);
4711         pci_disable_device(pdev);
4712
4713         /* Request a slot slot reset. */
4714         return PCI_ERS_RESULT_NEED_RESET;
4715 }
4716
4717 /**
4718  * e1000_io_slot_reset - called after the pci bus has been reset.
4719  * @pdev: Pointer to PCI device
4720  *
4721  * Restart the card from scratch, as if from a cold-boot. Implementation
4722  * resembles the first-half of the e1000_resume routine.
4723  */
4724 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4725 {
4726         struct net_device *netdev = pci_get_drvdata(pdev);
4727         struct e1000_adapter *adapter = netdev_priv(netdev);
4728         struct e1000_hw *hw = &adapter->hw;
4729         int err;
4730
4731         if (adapter->need_ioport)
4732                 err = pci_enable_device(pdev);
4733         else
4734                 err = pci_enable_device_mem(pdev);
4735         if (err) {
4736                 pr_err("Cannot re-enable PCI device after reset.\n");
4737                 return PCI_ERS_RESULT_DISCONNECT;
4738         }
4739         pci_set_master(pdev);
4740
4741         pci_enable_wake(pdev, PCI_D3hot, 0);
4742         pci_enable_wake(pdev, PCI_D3cold, 0);
4743
4744         e1000_reset(adapter);
4745         ew32(WUS, ~0);
4746
4747         return PCI_ERS_RESULT_RECOVERED;
4748 }
4749
4750 /**
4751  * e1000_io_resume - called when traffic can start flowing again.
4752  * @pdev: Pointer to PCI device
4753  *
4754  * This callback is called when the error recovery driver tells us that
4755  * its OK to resume normal operation. Implementation resembles the
4756  * second-half of the e1000_resume routine.
4757  */
4758 static void e1000_io_resume(struct pci_dev *pdev)
4759 {
4760         struct net_device *netdev = pci_get_drvdata(pdev);
4761         struct e1000_adapter *adapter = netdev_priv(netdev);
4762
4763         e1000_init_manageability(adapter);
4764
4765         if (netif_running(netdev)) {
4766                 if (e1000_up(adapter)) {
4767                         pr_info("can't bring device back up after reset\n");
4768                         return;
4769                 }
4770         }
4771
4772         netif_device_attach(netdev);
4773 }
4774
4775 /* e1000_main.c */