Pull misc2-6-36 into release branch
[sfrench/cifs-2.6.git] / drivers / net / e1000 / e1000_main.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2006 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include "e1000.h"
30 #include <net/ip6_checksum.h>
31
32 char e1000_driver_name[] = "e1000";
33 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
34 #define DRV_VERSION "7.3.21-k6-NAPI"
35 const char e1000_driver_version[] = DRV_VERSION;
36 static const char e1000_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
37
38 /* e1000_pci_tbl - PCI Device ID Table
39  *
40  * Last entry must be all 0s
41  *
42  * Macro expands to...
43  *   {PCI_DEVICE(PCI_VENDOR_ID_INTEL, device_id)}
44  */
45 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
46         INTEL_E1000_ETHERNET_DEVICE(0x1000),
47         INTEL_E1000_ETHERNET_DEVICE(0x1001),
48         INTEL_E1000_ETHERNET_DEVICE(0x1004),
49         INTEL_E1000_ETHERNET_DEVICE(0x1008),
50         INTEL_E1000_ETHERNET_DEVICE(0x1009),
51         INTEL_E1000_ETHERNET_DEVICE(0x100C),
52         INTEL_E1000_ETHERNET_DEVICE(0x100D),
53         INTEL_E1000_ETHERNET_DEVICE(0x100E),
54         INTEL_E1000_ETHERNET_DEVICE(0x100F),
55         INTEL_E1000_ETHERNET_DEVICE(0x1010),
56         INTEL_E1000_ETHERNET_DEVICE(0x1011),
57         INTEL_E1000_ETHERNET_DEVICE(0x1012),
58         INTEL_E1000_ETHERNET_DEVICE(0x1013),
59         INTEL_E1000_ETHERNET_DEVICE(0x1014),
60         INTEL_E1000_ETHERNET_DEVICE(0x1015),
61         INTEL_E1000_ETHERNET_DEVICE(0x1016),
62         INTEL_E1000_ETHERNET_DEVICE(0x1017),
63         INTEL_E1000_ETHERNET_DEVICE(0x1018),
64         INTEL_E1000_ETHERNET_DEVICE(0x1019),
65         INTEL_E1000_ETHERNET_DEVICE(0x101A),
66         INTEL_E1000_ETHERNET_DEVICE(0x101D),
67         INTEL_E1000_ETHERNET_DEVICE(0x101E),
68         INTEL_E1000_ETHERNET_DEVICE(0x1026),
69         INTEL_E1000_ETHERNET_DEVICE(0x1027),
70         INTEL_E1000_ETHERNET_DEVICE(0x1028),
71         INTEL_E1000_ETHERNET_DEVICE(0x1075),
72         INTEL_E1000_ETHERNET_DEVICE(0x1076),
73         INTEL_E1000_ETHERNET_DEVICE(0x1077),
74         INTEL_E1000_ETHERNET_DEVICE(0x1078),
75         INTEL_E1000_ETHERNET_DEVICE(0x1079),
76         INTEL_E1000_ETHERNET_DEVICE(0x107A),
77         INTEL_E1000_ETHERNET_DEVICE(0x107B),
78         INTEL_E1000_ETHERNET_DEVICE(0x107C),
79         INTEL_E1000_ETHERNET_DEVICE(0x108A),
80         INTEL_E1000_ETHERNET_DEVICE(0x1099),
81         INTEL_E1000_ETHERNET_DEVICE(0x10B5),
82         /* required last entry */
83         {0,}
84 };
85
86 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
87
88 int e1000_up(struct e1000_adapter *adapter);
89 void e1000_down(struct e1000_adapter *adapter);
90 void e1000_reinit_locked(struct e1000_adapter *adapter);
91 void e1000_reset(struct e1000_adapter *adapter);
92 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx);
93 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
94 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
95 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
96 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
97 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
98                              struct e1000_tx_ring *txdr);
99 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
100                              struct e1000_rx_ring *rxdr);
101 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
102                              struct e1000_tx_ring *tx_ring);
103 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
104                              struct e1000_rx_ring *rx_ring);
105 void e1000_update_stats(struct e1000_adapter *adapter);
106
107 static int e1000_init_module(void);
108 static void e1000_exit_module(void);
109 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
110 static void __devexit e1000_remove(struct pci_dev *pdev);
111 static int e1000_alloc_queues(struct e1000_adapter *adapter);
112 static int e1000_sw_init(struct e1000_adapter *adapter);
113 static int e1000_open(struct net_device *netdev);
114 static int e1000_close(struct net_device *netdev);
115 static void e1000_configure_tx(struct e1000_adapter *adapter);
116 static void e1000_configure_rx(struct e1000_adapter *adapter);
117 static void e1000_setup_rctl(struct e1000_adapter *adapter);
118 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
119 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
120 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
121                                 struct e1000_tx_ring *tx_ring);
122 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
123                                 struct e1000_rx_ring *rx_ring);
124 static void e1000_set_rx_mode(struct net_device *netdev);
125 static void e1000_update_phy_info(unsigned long data);
126 static void e1000_watchdog(unsigned long data);
127 static void e1000_82547_tx_fifo_stall(unsigned long data);
128 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
129                                     struct net_device *netdev);
130 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
131 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
132 static int e1000_set_mac(struct net_device *netdev, void *p);
133 static irqreturn_t e1000_intr(int irq, void *data);
134 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
135                                struct e1000_tx_ring *tx_ring);
136 static int e1000_clean(struct napi_struct *napi, int budget);
137 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
138                                struct e1000_rx_ring *rx_ring,
139                                int *work_done, int work_to_do);
140 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
141                                      struct e1000_rx_ring *rx_ring,
142                                      int *work_done, int work_to_do);
143 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
144                                    struct e1000_rx_ring *rx_ring,
145                                    int cleaned_count);
146 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
147                                          struct e1000_rx_ring *rx_ring,
148                                          int cleaned_count);
149 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
150 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
151                            int cmd);
152 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
153 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
154 static void e1000_tx_timeout(struct net_device *dev);
155 static void e1000_reset_task(struct work_struct *work);
156 static void e1000_smartspeed(struct e1000_adapter *adapter);
157 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
158                                        struct sk_buff *skb);
159
160 static void e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp);
161 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid);
162 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid);
163 static void e1000_restore_vlan(struct e1000_adapter *adapter);
164
165 #ifdef CONFIG_PM
166 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
167 static int e1000_resume(struct pci_dev *pdev);
168 #endif
169 static void e1000_shutdown(struct pci_dev *pdev);
170
171 #ifdef CONFIG_NET_POLL_CONTROLLER
172 /* for netdump / net console */
173 static void e1000_netpoll (struct net_device *netdev);
174 #endif
175
176 #define COPYBREAK_DEFAULT 256
177 static unsigned int copybreak __read_mostly = COPYBREAK_DEFAULT;
178 module_param(copybreak, uint, 0644);
179 MODULE_PARM_DESC(copybreak,
180         "Maximum size of packet that is copied to a new buffer on receive");
181
182 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
183                      pci_channel_state_t state);
184 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
185 static void e1000_io_resume(struct pci_dev *pdev);
186
187 static struct pci_error_handlers e1000_err_handler = {
188         .error_detected = e1000_io_error_detected,
189         .slot_reset = e1000_io_slot_reset,
190         .resume = e1000_io_resume,
191 };
192
193 static struct pci_driver e1000_driver = {
194         .name     = e1000_driver_name,
195         .id_table = e1000_pci_tbl,
196         .probe    = e1000_probe,
197         .remove   = __devexit_p(e1000_remove),
198 #ifdef CONFIG_PM
199         /* Power Managment Hooks */
200         .suspend  = e1000_suspend,
201         .resume   = e1000_resume,
202 #endif
203         .shutdown = e1000_shutdown,
204         .err_handler = &e1000_err_handler
205 };
206
207 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
208 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
209 MODULE_LICENSE("GPL");
210 MODULE_VERSION(DRV_VERSION);
211
212 static int debug = NETIF_MSG_DRV | NETIF_MSG_PROBE;
213 module_param(debug, int, 0);
214 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
215
216 /**
217  * e1000_get_hw_dev - return device
218  * used by hardware layer to print debugging information
219  *
220  **/
221 struct net_device *e1000_get_hw_dev(struct e1000_hw *hw)
222 {
223         struct e1000_adapter *adapter = hw->back;
224         return adapter->netdev;
225 }
226
227 /**
228  * e1000_init_module - Driver Registration Routine
229  *
230  * e1000_init_module is the first routine called when the driver is
231  * loaded. All it does is register with the PCI subsystem.
232  **/
233
234 static int __init e1000_init_module(void)
235 {
236         int ret;
237         pr_info("%s - version %s\n", e1000_driver_string, e1000_driver_version);
238
239         pr_info("%s\n", e1000_copyright);
240
241         ret = pci_register_driver(&e1000_driver);
242         if (copybreak != COPYBREAK_DEFAULT) {
243                 if (copybreak == 0)
244                         pr_info("copybreak disabled\n");
245                 else
246                         pr_info("copybreak enabled for "
247                                    "packets <= %u bytes\n", copybreak);
248         }
249         return ret;
250 }
251
252 module_init(e1000_init_module);
253
254 /**
255  * e1000_exit_module - Driver Exit Cleanup Routine
256  *
257  * e1000_exit_module is called just before the driver is removed
258  * from memory.
259  **/
260
261 static void __exit e1000_exit_module(void)
262 {
263         pci_unregister_driver(&e1000_driver);
264 }
265
266 module_exit(e1000_exit_module);
267
268 static int e1000_request_irq(struct e1000_adapter *adapter)
269 {
270         struct net_device *netdev = adapter->netdev;
271         irq_handler_t handler = e1000_intr;
272         int irq_flags = IRQF_SHARED;
273         int err;
274
275         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
276                           netdev);
277         if (err) {
278                 e_err("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:%dMHz:%d-bit) %pM\n",
1051                ((hw->bus_type == e1000_bus_type_pcix) ? "-X" : ""),
1052                ((hw->bus_speed == e1000_bus_speed_133) ? 133 :
1053                 (hw->bus_speed == e1000_bus_speed_120) ? 120 :
1054                 (hw->bus_speed == e1000_bus_speed_100) ? 100 :
1055                 (hw->bus_speed == e1000_bus_speed_66) ? 66 : 33),
1056                ((hw->bus_width == e1000_bus_width_64) ? 64 : 32),
1057                netdev->dev_addr);
1058
1059         /* carrier off reporting is important to ethtool even BEFORE open */
1060         netif_carrier_off(netdev);
1061
1062         e_info("Intel(R) PRO/1000 Network Connection\n");
1063
1064         cards_found++;
1065         return 0;
1066
1067 err_register:
1068 err_eeprom:
1069         e1000_phy_hw_reset(hw);
1070
1071         if (hw->flash_address)
1072                 iounmap(hw->flash_address);
1073         kfree(adapter->tx_ring);
1074         kfree(adapter->rx_ring);
1075 err_sw_init:
1076         iounmap(hw->hw_addr);
1077 err_ioremap:
1078         free_netdev(netdev);
1079 err_alloc_etherdev:
1080         pci_release_selected_regions(pdev, bars);
1081 err_pci_reg:
1082 err_dma:
1083         pci_disable_device(pdev);
1084         return err;
1085 }
1086
1087 /**
1088  * e1000_remove - Device Removal Routine
1089  * @pdev: PCI device information struct
1090  *
1091  * e1000_remove is called by the PCI subsystem to alert the driver
1092  * that it should release a PCI device.  The could be caused by a
1093  * Hot-Plug event, or because the driver is going to be removed from
1094  * memory.
1095  **/
1096
1097 static void __devexit e1000_remove(struct pci_dev *pdev)
1098 {
1099         struct net_device *netdev = pci_get_drvdata(pdev);
1100         struct e1000_adapter *adapter = netdev_priv(netdev);
1101         struct e1000_hw *hw = &adapter->hw;
1102
1103         set_bit(__E1000_DOWN, &adapter->flags);
1104         del_timer_sync(&adapter->tx_fifo_stall_timer);
1105         del_timer_sync(&adapter->watchdog_timer);
1106         del_timer_sync(&adapter->phy_info_timer);
1107
1108         cancel_work_sync(&adapter->reset_task);
1109
1110         e1000_release_manageability(adapter);
1111
1112         unregister_netdev(netdev);
1113
1114         e1000_phy_hw_reset(hw);
1115
1116         kfree(adapter->tx_ring);
1117         kfree(adapter->rx_ring);
1118
1119         iounmap(hw->hw_addr);
1120         if (hw->flash_address)
1121                 iounmap(hw->flash_address);
1122         pci_release_selected_regions(pdev, adapter->bars);
1123
1124         free_netdev(netdev);
1125
1126         pci_disable_device(pdev);
1127 }
1128
1129 /**
1130  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1131  * @adapter: board private structure to initialize
1132  *
1133  * e1000_sw_init initializes the Adapter private data structure.
1134  * Fields are initialized based on PCI device information and
1135  * OS network device settings (MTU size).
1136  **/
1137
1138 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
1139 {
1140         struct e1000_hw *hw = &adapter->hw;
1141         struct net_device *netdev = adapter->netdev;
1142         struct pci_dev *pdev = adapter->pdev;
1143
1144         /* PCI config space info */
1145
1146         hw->vendor_id = pdev->vendor;
1147         hw->device_id = pdev->device;
1148         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1149         hw->subsystem_id = pdev->subsystem_device;
1150         hw->revision_id = pdev->revision;
1151
1152         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
1153
1154         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1155         hw->max_frame_size = netdev->mtu +
1156                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
1157         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
1158
1159         /* identify the MAC */
1160
1161         if (e1000_set_mac_type(hw)) {
1162                 e_err("Unknown MAC Type\n");
1163                 return -EIO;
1164         }
1165
1166         switch (hw->mac_type) {
1167         default:
1168                 break;
1169         case e1000_82541:
1170         case e1000_82547:
1171         case e1000_82541_rev_2:
1172         case e1000_82547_rev_2:
1173                 hw->phy_init_script = 1;
1174                 break;
1175         }
1176
1177         e1000_set_media_type(hw);
1178
1179         hw->wait_autoneg_complete = false;
1180         hw->tbi_compatibility_en = true;
1181         hw->adaptive_ifs = true;
1182
1183         /* Copper options */
1184
1185         if (hw->media_type == e1000_media_type_copper) {
1186                 hw->mdix = AUTO_ALL_MODES;
1187                 hw->disable_polarity_correction = false;
1188                 hw->master_slave = E1000_MASTER_SLAVE;
1189         }
1190
1191         adapter->num_tx_queues = 1;
1192         adapter->num_rx_queues = 1;
1193
1194         if (e1000_alloc_queues(adapter)) {
1195                 e_err("Unable to allocate memory for queues\n");
1196                 return -ENOMEM;
1197         }
1198
1199         /* Explicitly disable IRQ since the NIC can be in any state. */
1200         e1000_irq_disable(adapter);
1201
1202         spin_lock_init(&adapter->stats_lock);
1203
1204         set_bit(__E1000_DOWN, &adapter->flags);
1205
1206         return 0;
1207 }
1208
1209 /**
1210  * e1000_alloc_queues - Allocate memory for all rings
1211  * @adapter: board private structure to initialize
1212  *
1213  * We allocate one ring per queue at run-time since we don't know the
1214  * number of queues at compile-time.
1215  **/
1216
1217 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1218 {
1219         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1220                                    sizeof(struct e1000_tx_ring), GFP_KERNEL);
1221         if (!adapter->tx_ring)
1222                 return -ENOMEM;
1223
1224         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1225                                    sizeof(struct e1000_rx_ring), GFP_KERNEL);
1226         if (!adapter->rx_ring) {
1227                 kfree(adapter->tx_ring);
1228                 return -ENOMEM;
1229         }
1230
1231         return E1000_SUCCESS;
1232 }
1233
1234 /**
1235  * e1000_open - Called when a network interface is made active
1236  * @netdev: network interface device structure
1237  *
1238  * Returns 0 on success, negative value on failure
1239  *
1240  * The open entry point is called when a network interface is made
1241  * active by the system (IFF_UP).  At this point all resources needed
1242  * for transmit and receive operations are allocated, the interrupt
1243  * handler is registered with the OS, the watchdog timer is started,
1244  * and the stack is notified that the interface is ready.
1245  **/
1246
1247 static int e1000_open(struct net_device *netdev)
1248 {
1249         struct e1000_adapter *adapter = netdev_priv(netdev);
1250         struct e1000_hw *hw = &adapter->hw;
1251         int err;
1252
1253         /* disallow open during test */
1254         if (test_bit(__E1000_TESTING, &adapter->flags))
1255                 return -EBUSY;
1256
1257         netif_carrier_off(netdev);
1258
1259         /* allocate transmit descriptors */
1260         err = e1000_setup_all_tx_resources(adapter);
1261         if (err)
1262                 goto err_setup_tx;
1263
1264         /* allocate receive descriptors */
1265         err = e1000_setup_all_rx_resources(adapter);
1266         if (err)
1267                 goto err_setup_rx;
1268
1269         e1000_power_up_phy(adapter);
1270
1271         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1272         if ((hw->mng_cookie.status &
1273                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1274                 e1000_update_mng_vlan(adapter);
1275         }
1276
1277         /* before we allocate an interrupt, we must be ready to handle it.
1278          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1279          * as soon as we call pci_request_irq, so we have to setup our
1280          * clean_rx handler before we do so.  */
1281         e1000_configure(adapter);
1282
1283         err = e1000_request_irq(adapter);
1284         if (err)
1285                 goto err_req_irq;
1286
1287         /* From here on the code is the same as e1000_up() */
1288         clear_bit(__E1000_DOWN, &adapter->flags);
1289
1290         napi_enable(&adapter->napi);
1291
1292         e1000_irq_enable(adapter);
1293
1294         netif_start_queue(netdev);
1295
1296         /* fire a link status change interrupt to start the watchdog */
1297         ew32(ICS, E1000_ICS_LSC);
1298
1299         return E1000_SUCCESS;
1300
1301 err_req_irq:
1302         e1000_power_down_phy(adapter);
1303         e1000_free_all_rx_resources(adapter);
1304 err_setup_rx:
1305         e1000_free_all_tx_resources(adapter);
1306 err_setup_tx:
1307         e1000_reset(adapter);
1308
1309         return err;
1310 }
1311
1312 /**
1313  * e1000_close - Disables a network interface
1314  * @netdev: network interface device structure
1315  *
1316  * Returns 0, this is not allowed to fail
1317  *
1318  * The close entry point is called when an interface is de-activated
1319  * by the OS.  The hardware is still under the drivers control, but
1320  * needs to be disabled.  A global MAC reset is issued to stop the
1321  * hardware, and all transmit and receive resources are freed.
1322  **/
1323
1324 static int e1000_close(struct net_device *netdev)
1325 {
1326         struct e1000_adapter *adapter = netdev_priv(netdev);
1327         struct e1000_hw *hw = &adapter->hw;
1328
1329         WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1330         e1000_down(adapter);
1331         e1000_power_down_phy(adapter);
1332         e1000_free_irq(adapter);
1333
1334         e1000_free_all_tx_resources(adapter);
1335         e1000_free_all_rx_resources(adapter);
1336
1337         /* kill manageability vlan ID if supported, but not if a vlan with
1338          * the same ID is registered on the host OS (let 8021q kill it) */
1339         if ((hw->mng_cookie.status &
1340                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1341              !(adapter->vlgrp &&
1342                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id))) {
1343                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1344         }
1345
1346         return 0;
1347 }
1348
1349 /**
1350  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1351  * @adapter: address of board private structure
1352  * @start: address of beginning of memory
1353  * @len: length of memory
1354  **/
1355 static bool e1000_check_64k_bound(struct e1000_adapter *adapter, void *start,
1356                                   unsigned long len)
1357 {
1358         struct e1000_hw *hw = &adapter->hw;
1359         unsigned long begin = (unsigned long)start;
1360         unsigned long end = begin + len;
1361
1362         /* First rev 82545 and 82546 need to not allow any memory
1363          * write location to cross 64k boundary due to errata 23 */
1364         if (hw->mac_type == e1000_82545 ||
1365             hw->mac_type == e1000_82546) {
1366                 return ((begin ^ (end - 1)) >> 16) != 0 ? false : true;
1367         }
1368
1369         return true;
1370 }
1371
1372 /**
1373  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1374  * @adapter: board private structure
1375  * @txdr:    tx descriptor ring (for a specific queue) to setup
1376  *
1377  * Return 0 on success, negative on failure
1378  **/
1379
1380 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
1381                                     struct e1000_tx_ring *txdr)
1382 {
1383         struct pci_dev *pdev = adapter->pdev;
1384         int size;
1385
1386         size = sizeof(struct e1000_buffer) * txdr->count;
1387         txdr->buffer_info = vmalloc(size);
1388         if (!txdr->buffer_info) {
1389                 e_err("Unable to allocate memory for the Tx descriptor ring\n");
1390                 return -ENOMEM;
1391         }
1392         memset(txdr->buffer_info, 0, size);
1393
1394         /* round up to nearest 4K */
1395
1396         txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1397         txdr->size = ALIGN(txdr->size, 4096);
1398
1399         txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size, &txdr->dma,
1400                                         GFP_KERNEL);
1401         if (!txdr->desc) {
1402 setup_tx_desc_die:
1403                 vfree(txdr->buffer_info);
1404                 e_err("Unable to allocate memory for the Tx descriptor ring\n");
1405                 return -ENOMEM;
1406         }
1407
1408         /* Fix for errata 23, can't cross 64kB boundary */
1409         if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1410                 void *olddesc = txdr->desc;
1411                 dma_addr_t olddma = txdr->dma;
1412                 e_err("txdr align check failed: %u bytes at %p\n",
1413                       txdr->size, txdr->desc);
1414                 /* Try again, without freeing the previous */
1415                 txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size,
1416                                                 &txdr->dma, GFP_KERNEL);
1417                 /* Failed allocation, critical failure */
1418                 if (!txdr->desc) {
1419                         dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1420                                           olddma);
1421                         goto setup_tx_desc_die;
1422                 }
1423
1424                 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1425                         /* give up */
1426                         dma_free_coherent(&pdev->dev, txdr->size, txdr->desc,
1427                                           txdr->dma);
1428                         dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1429                                           olddma);
1430                         e_err("Unable to allocate aligned memory "
1431                               "for the transmit descriptor ring\n");
1432                         vfree(txdr->buffer_info);
1433                         return -ENOMEM;
1434                 } else {
1435                         /* Free old allocation, new allocation was successful */
1436                         dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1437                                           olddma);
1438                 }
1439         }
1440         memset(txdr->desc, 0, txdr->size);
1441
1442         txdr->next_to_use = 0;
1443         txdr->next_to_clean = 0;
1444
1445         return 0;
1446 }
1447
1448 /**
1449  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1450  *                                (Descriptors) for all queues
1451  * @adapter: board private structure
1452  *
1453  * Return 0 on success, negative on failure
1454  **/
1455
1456 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1457 {
1458         int i, err = 0;
1459
1460         for (i = 0; i < adapter->num_tx_queues; i++) {
1461                 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1462                 if (err) {
1463                         e_err("Allocation for Tx Queue %u failed\n", i);
1464                         for (i-- ; i >= 0; i--)
1465                                 e1000_free_tx_resources(adapter,
1466                                                         &adapter->tx_ring[i]);
1467                         break;
1468                 }
1469         }
1470
1471         return err;
1472 }
1473
1474 /**
1475  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1476  * @adapter: board private structure
1477  *
1478  * Configure the Tx unit of the MAC after a reset.
1479  **/
1480
1481 static void e1000_configure_tx(struct e1000_adapter *adapter)
1482 {
1483         u64 tdba;
1484         struct e1000_hw *hw = &adapter->hw;
1485         u32 tdlen, tctl, tipg;
1486         u32 ipgr1, ipgr2;
1487
1488         /* Setup the HW Tx Head and Tail descriptor pointers */
1489
1490         switch (adapter->num_tx_queues) {
1491         case 1:
1492         default:
1493                 tdba = adapter->tx_ring[0].dma;
1494                 tdlen = adapter->tx_ring[0].count *
1495                         sizeof(struct e1000_tx_desc);
1496                 ew32(TDLEN, tdlen);
1497                 ew32(TDBAH, (tdba >> 32));
1498                 ew32(TDBAL, (tdba & 0x00000000ffffffffULL));
1499                 ew32(TDT, 0);
1500                 ew32(TDH, 0);
1501                 adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ? E1000_TDH : E1000_82542_TDH);
1502                 adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ? E1000_TDT : E1000_82542_TDT);
1503                 break;
1504         }
1505
1506         /* Set the default values for the Tx Inter Packet Gap timer */
1507         if ((hw->media_type == e1000_media_type_fiber ||
1508              hw->media_type == e1000_media_type_internal_serdes))
1509                 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1510         else
1511                 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1512
1513         switch (hw->mac_type) {
1514         case e1000_82542_rev2_0:
1515         case e1000_82542_rev2_1:
1516                 tipg = DEFAULT_82542_TIPG_IPGT;
1517                 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1518                 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1519                 break;
1520         default:
1521                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1522                 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1523                 break;
1524         }
1525         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1526         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1527         ew32(TIPG, tipg);
1528
1529         /* Set the Tx Interrupt Delay register */
1530
1531         ew32(TIDV, adapter->tx_int_delay);
1532         if (hw->mac_type >= e1000_82540)
1533                 ew32(TADV, adapter->tx_abs_int_delay);
1534
1535         /* Program the Transmit Control Register */
1536
1537         tctl = er32(TCTL);
1538         tctl &= ~E1000_TCTL_CT;
1539         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1540                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1541
1542         e1000_config_collision_dist(hw);
1543
1544         /* Setup Transmit Descriptor Settings for eop descriptor */
1545         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1546
1547         /* only set IDE if we are delaying interrupts using the timers */
1548         if (adapter->tx_int_delay)
1549                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1550
1551         if (hw->mac_type < e1000_82543)
1552                 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1553         else
1554                 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1555
1556         /* Cache if we're 82544 running in PCI-X because we'll
1557          * need this to apply a workaround later in the send path. */
1558         if (hw->mac_type == e1000_82544 &&
1559             hw->bus_type == e1000_bus_type_pcix)
1560                 adapter->pcix_82544 = 1;
1561
1562         ew32(TCTL, tctl);
1563
1564 }
1565
1566 /**
1567  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1568  * @adapter: board private structure
1569  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1570  *
1571  * Returns 0 on success, negative on failure
1572  **/
1573
1574 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
1575                                     struct e1000_rx_ring *rxdr)
1576 {
1577         struct pci_dev *pdev = adapter->pdev;
1578         int size, desc_len;
1579
1580         size = sizeof(struct e1000_buffer) * rxdr->count;
1581         rxdr->buffer_info = vmalloc(size);
1582         if (!rxdr->buffer_info) {
1583                 e_err("Unable to allocate memory for the Rx descriptor ring\n");
1584                 return -ENOMEM;
1585         }
1586         memset(rxdr->buffer_info, 0, size);
1587
1588         desc_len = sizeof(struct e1000_rx_desc);
1589
1590         /* Round up to nearest 4K */
1591
1592         rxdr->size = rxdr->count * desc_len;
1593         rxdr->size = ALIGN(rxdr->size, 4096);
1594
1595         rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size, &rxdr->dma,
1596                                         GFP_KERNEL);
1597
1598         if (!rxdr->desc) {
1599                 e_err("Unable to allocate memory for the Rx descriptor ring\n");
1600 setup_rx_desc_die:
1601                 vfree(rxdr->buffer_info);
1602                 return -ENOMEM;
1603         }
1604
1605         /* Fix for errata 23, can't cross 64kB boundary */
1606         if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1607                 void *olddesc = rxdr->desc;
1608                 dma_addr_t olddma = rxdr->dma;
1609                 e_err("rxdr align check failed: %u bytes at %p\n",
1610                       rxdr->size, rxdr->desc);
1611                 /* Try again, without freeing the previous */
1612                 rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size,
1613                                                 &rxdr->dma, GFP_KERNEL);
1614                 /* Failed allocation, critical failure */
1615                 if (!rxdr->desc) {
1616                         dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1617                                           olddma);
1618                         e_err("Unable to allocate memory for the Rx descriptor "
1619                               "ring\n");
1620                         goto setup_rx_desc_die;
1621                 }
1622
1623                 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1624                         /* give up */
1625                         dma_free_coherent(&pdev->dev, rxdr->size, rxdr->desc,
1626                                           rxdr->dma);
1627                         dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1628                                           olddma);
1629                         e_err("Unable to allocate aligned memory for the Rx "
1630                               "descriptor ring\n");
1631                         goto setup_rx_desc_die;
1632                 } else {
1633                         /* Free old allocation, new allocation was successful */
1634                         dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1635                                           olddma);
1636                 }
1637         }
1638         memset(rxdr->desc, 0, rxdr->size);
1639
1640         rxdr->next_to_clean = 0;
1641         rxdr->next_to_use = 0;
1642         rxdr->rx_skb_top = NULL;
1643
1644         return 0;
1645 }
1646
1647 /**
1648  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1649  *                                (Descriptors) for all queues
1650  * @adapter: board private structure
1651  *
1652  * Return 0 on success, negative on failure
1653  **/
1654
1655 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1656 {
1657         int i, err = 0;
1658
1659         for (i = 0; i < adapter->num_rx_queues; i++) {
1660                 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1661                 if (err) {
1662                         e_err("Allocation for Rx Queue %u failed\n", i);
1663                         for (i-- ; i >= 0; i--)
1664                                 e1000_free_rx_resources(adapter,
1665                                                         &adapter->rx_ring[i]);
1666                         break;
1667                 }
1668         }
1669
1670         return err;
1671 }
1672
1673 /**
1674  * e1000_setup_rctl - configure the receive control registers
1675  * @adapter: Board private structure
1676  **/
1677 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1678 {
1679         struct e1000_hw *hw = &adapter->hw;
1680         u32 rctl;
1681
1682         rctl = er32(RCTL);
1683
1684         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1685
1686         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1687                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1688                 (hw->mc_filter_type << E1000_RCTL_MO_SHIFT);
1689
1690         if (hw->tbi_compatibility_on == 1)
1691                 rctl |= E1000_RCTL_SBP;
1692         else
1693                 rctl &= ~E1000_RCTL_SBP;
1694
1695         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1696                 rctl &= ~E1000_RCTL_LPE;
1697         else
1698                 rctl |= E1000_RCTL_LPE;
1699
1700         /* Setup buffer sizes */
1701         rctl &= ~E1000_RCTL_SZ_4096;
1702         rctl |= E1000_RCTL_BSEX;
1703         switch (adapter->rx_buffer_len) {
1704                 case E1000_RXBUFFER_2048:
1705                 default:
1706                         rctl |= E1000_RCTL_SZ_2048;
1707                         rctl &= ~E1000_RCTL_BSEX;
1708                         break;
1709                 case E1000_RXBUFFER_4096:
1710                         rctl |= E1000_RCTL_SZ_4096;
1711                         break;
1712                 case E1000_RXBUFFER_8192:
1713                         rctl |= E1000_RCTL_SZ_8192;
1714                         break;
1715                 case E1000_RXBUFFER_16384:
1716                         rctl |= E1000_RCTL_SZ_16384;
1717                         break;
1718         }
1719
1720         ew32(RCTL, rctl);
1721 }
1722
1723 /**
1724  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1725  * @adapter: board private structure
1726  *
1727  * Configure the Rx unit of the MAC after a reset.
1728  **/
1729
1730 static void e1000_configure_rx(struct e1000_adapter *adapter)
1731 {
1732         u64 rdba;
1733         struct e1000_hw *hw = &adapter->hw;
1734         u32 rdlen, rctl, rxcsum;
1735
1736         if (adapter->netdev->mtu > ETH_DATA_LEN) {
1737                 rdlen = adapter->rx_ring[0].count *
1738                         sizeof(struct e1000_rx_desc);
1739                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
1740                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
1741         } else {
1742                 rdlen = adapter->rx_ring[0].count *
1743                         sizeof(struct e1000_rx_desc);
1744                 adapter->clean_rx = e1000_clean_rx_irq;
1745                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1746         }
1747
1748         /* disable receives while setting up the descriptors */
1749         rctl = er32(RCTL);
1750         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1751
1752         /* set the Receive Delay Timer Register */
1753         ew32(RDTR, adapter->rx_int_delay);
1754
1755         if (hw->mac_type >= e1000_82540) {
1756                 ew32(RADV, adapter->rx_abs_int_delay);
1757                 if (adapter->itr_setting != 0)
1758                         ew32(ITR, 1000000000 / (adapter->itr * 256));
1759         }
1760
1761         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1762          * the Base and Length of the Rx Descriptor Ring */
1763         switch (adapter->num_rx_queues) {
1764         case 1:
1765         default:
1766                 rdba = adapter->rx_ring[0].dma;
1767                 ew32(RDLEN, rdlen);
1768                 ew32(RDBAH, (rdba >> 32));
1769                 ew32(RDBAL, (rdba & 0x00000000ffffffffULL));
1770                 ew32(RDT, 0);
1771                 ew32(RDH, 0);
1772                 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ? E1000_RDH : E1000_82542_RDH);
1773                 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ? E1000_RDT : E1000_82542_RDT);
1774                 break;
1775         }
1776
1777         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
1778         if (hw->mac_type >= e1000_82543) {
1779                 rxcsum = er32(RXCSUM);
1780                 if (adapter->rx_csum)
1781                         rxcsum |= E1000_RXCSUM_TUOFL;
1782                 else
1783                         /* don't need to clear IPPCSE as it defaults to 0 */
1784                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1785                 ew32(RXCSUM, rxcsum);
1786         }
1787
1788         /* Enable Receives */
1789         ew32(RCTL, rctl);
1790 }
1791
1792 /**
1793  * e1000_free_tx_resources - Free Tx Resources per Queue
1794  * @adapter: board private structure
1795  * @tx_ring: Tx descriptor ring for a specific queue
1796  *
1797  * Free all transmit software resources
1798  **/
1799
1800 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
1801                                     struct e1000_tx_ring *tx_ring)
1802 {
1803         struct pci_dev *pdev = adapter->pdev;
1804
1805         e1000_clean_tx_ring(adapter, tx_ring);
1806
1807         vfree(tx_ring->buffer_info);
1808         tx_ring->buffer_info = NULL;
1809
1810         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1811                           tx_ring->dma);
1812
1813         tx_ring->desc = NULL;
1814 }
1815
1816 /**
1817  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1818  * @adapter: board private structure
1819  *
1820  * Free all transmit software resources
1821  **/
1822
1823 void e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1824 {
1825         int i;
1826
1827         for (i = 0; i < adapter->num_tx_queues; i++)
1828                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1829 }
1830
1831 static void e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1832                                              struct e1000_buffer *buffer_info)
1833 {
1834         if (buffer_info->dma) {
1835                 if (buffer_info->mapped_as_page)
1836                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1837                                        buffer_info->length, DMA_TO_DEVICE);
1838                 else
1839                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1840                                          buffer_info->length,
1841                                          DMA_TO_DEVICE);
1842                 buffer_info->dma = 0;
1843         }
1844         if (buffer_info->skb) {
1845                 dev_kfree_skb_any(buffer_info->skb);
1846                 buffer_info->skb = NULL;
1847         }
1848         buffer_info->time_stamp = 0;
1849         /* buffer_info must be completely set up in the transmit path */
1850 }
1851
1852 /**
1853  * e1000_clean_tx_ring - Free Tx Buffers
1854  * @adapter: board private structure
1855  * @tx_ring: ring to be cleaned
1856  **/
1857
1858 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
1859                                 struct e1000_tx_ring *tx_ring)
1860 {
1861         struct e1000_hw *hw = &adapter->hw;
1862         struct e1000_buffer *buffer_info;
1863         unsigned long size;
1864         unsigned int i;
1865
1866         /* Free all the Tx ring sk_buffs */
1867
1868         for (i = 0; i < tx_ring->count; i++) {
1869                 buffer_info = &tx_ring->buffer_info[i];
1870                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
1871         }
1872
1873         size = sizeof(struct e1000_buffer) * tx_ring->count;
1874         memset(tx_ring->buffer_info, 0, size);
1875
1876         /* Zero out the descriptor ring */
1877
1878         memset(tx_ring->desc, 0, tx_ring->size);
1879
1880         tx_ring->next_to_use = 0;
1881         tx_ring->next_to_clean = 0;
1882         tx_ring->last_tx_tso = 0;
1883
1884         writel(0, hw->hw_addr + tx_ring->tdh);
1885         writel(0, hw->hw_addr + tx_ring->tdt);
1886 }
1887
1888 /**
1889  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
1890  * @adapter: board private structure
1891  **/
1892
1893 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
1894 {
1895         int i;
1896
1897         for (i = 0; i < adapter->num_tx_queues; i++)
1898                 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1899 }
1900
1901 /**
1902  * e1000_free_rx_resources - Free Rx Resources
1903  * @adapter: board private structure
1904  * @rx_ring: ring to clean the resources from
1905  *
1906  * Free all receive software resources
1907  **/
1908
1909 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
1910                                     struct e1000_rx_ring *rx_ring)
1911 {
1912         struct pci_dev *pdev = adapter->pdev;
1913
1914         e1000_clean_rx_ring(adapter, rx_ring);
1915
1916         vfree(rx_ring->buffer_info);
1917         rx_ring->buffer_info = NULL;
1918
1919         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1920                           rx_ring->dma);
1921
1922         rx_ring->desc = NULL;
1923 }
1924
1925 /**
1926  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
1927  * @adapter: board private structure
1928  *
1929  * Free all receive software resources
1930  **/
1931
1932 void e1000_free_all_rx_resources(struct e1000_adapter *adapter)
1933 {
1934         int i;
1935
1936         for (i = 0; i < adapter->num_rx_queues; i++)
1937                 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
1938 }
1939
1940 /**
1941  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1942  * @adapter: board private structure
1943  * @rx_ring: ring to free buffers from
1944  **/
1945
1946 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
1947                                 struct e1000_rx_ring *rx_ring)
1948 {
1949         struct e1000_hw *hw = &adapter->hw;
1950         struct e1000_buffer *buffer_info;
1951         struct pci_dev *pdev = adapter->pdev;
1952         unsigned long size;
1953         unsigned int i;
1954
1955         /* Free all the Rx ring sk_buffs */
1956         for (i = 0; i < rx_ring->count; i++) {
1957                 buffer_info = &rx_ring->buffer_info[i];
1958                 if (buffer_info->dma &&
1959                     adapter->clean_rx == e1000_clean_rx_irq) {
1960                         dma_unmap_single(&pdev->dev, buffer_info->dma,
1961                                          buffer_info->length,
1962                                          DMA_FROM_DEVICE);
1963                 } else if (buffer_info->dma &&
1964                            adapter->clean_rx == e1000_clean_jumbo_rx_irq) {
1965                         dma_unmap_page(&pdev->dev, buffer_info->dma,
1966                                        buffer_info->length,
1967                                        DMA_FROM_DEVICE);
1968                 }
1969
1970                 buffer_info->dma = 0;
1971                 if (buffer_info->page) {
1972                         put_page(buffer_info->page);
1973                         buffer_info->page = NULL;
1974                 }
1975                 if (buffer_info->skb) {
1976                         dev_kfree_skb(buffer_info->skb);
1977                         buffer_info->skb = NULL;
1978                 }
1979         }
1980
1981         /* there also may be some cached data from a chained receive */
1982         if (rx_ring->rx_skb_top) {
1983                 dev_kfree_skb(rx_ring->rx_skb_top);
1984                 rx_ring->rx_skb_top = NULL;
1985         }
1986
1987         size = sizeof(struct e1000_buffer) * rx_ring->count;
1988         memset(rx_ring->buffer_info, 0, size);
1989
1990         /* Zero out the descriptor ring */
1991         memset(rx_ring->desc, 0, rx_ring->size);
1992
1993         rx_ring->next_to_clean = 0;
1994         rx_ring->next_to_use = 0;
1995
1996         writel(0, hw->hw_addr + rx_ring->rdh);
1997         writel(0, hw->hw_addr + rx_ring->rdt);
1998 }
1999
2000 /**
2001  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2002  * @adapter: board private structure
2003  **/
2004
2005 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2006 {
2007         int i;
2008
2009         for (i = 0; i < adapter->num_rx_queues; i++)
2010                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2011 }
2012
2013 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2014  * and memory write and invalidate disabled for certain operations
2015  */
2016 static void e1000_enter_82542_rst(struct e1000_adapter *adapter)
2017 {
2018         struct e1000_hw *hw = &adapter->hw;
2019         struct net_device *netdev = adapter->netdev;
2020         u32 rctl;
2021
2022         e1000_pci_clear_mwi(hw);
2023
2024         rctl = er32(RCTL);
2025         rctl |= E1000_RCTL_RST;
2026         ew32(RCTL, rctl);
2027         E1000_WRITE_FLUSH();
2028         mdelay(5);
2029
2030         if (netif_running(netdev))
2031                 e1000_clean_all_rx_rings(adapter);
2032 }
2033
2034 static void e1000_leave_82542_rst(struct e1000_adapter *adapter)
2035 {
2036         struct e1000_hw *hw = &adapter->hw;
2037         struct net_device *netdev = adapter->netdev;
2038         u32 rctl;
2039
2040         rctl = er32(RCTL);
2041         rctl &= ~E1000_RCTL_RST;
2042         ew32(RCTL, rctl);
2043         E1000_WRITE_FLUSH();
2044         mdelay(5);
2045
2046         if (hw->pci_cmd_word & PCI_COMMAND_INVALIDATE)
2047                 e1000_pci_set_mwi(hw);
2048
2049         if (netif_running(netdev)) {
2050                 /* No need to loop, because 82542 supports only 1 queue */
2051                 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2052                 e1000_configure_rx(adapter);
2053                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2054         }
2055 }
2056
2057 /**
2058  * e1000_set_mac - Change the Ethernet Address of the NIC
2059  * @netdev: network interface device structure
2060  * @p: pointer to an address structure
2061  *
2062  * Returns 0 on success, negative on failure
2063  **/
2064
2065 static int e1000_set_mac(struct net_device *netdev, void *p)
2066 {
2067         struct e1000_adapter *adapter = netdev_priv(netdev);
2068         struct e1000_hw *hw = &adapter->hw;
2069         struct sockaddr *addr = p;
2070
2071         if (!is_valid_ether_addr(addr->sa_data))
2072                 return -EADDRNOTAVAIL;
2073
2074         /* 82542 2.0 needs to be in reset to write receive address registers */
2075
2076         if (hw->mac_type == e1000_82542_rev2_0)
2077                 e1000_enter_82542_rst(adapter);
2078
2079         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2080         memcpy(hw->mac_addr, addr->sa_data, netdev->addr_len);
2081
2082         e1000_rar_set(hw, hw->mac_addr, 0);
2083
2084         if (hw->mac_type == e1000_82542_rev2_0)
2085                 e1000_leave_82542_rst(adapter);
2086
2087         return 0;
2088 }
2089
2090 /**
2091  * e1000_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2092  * @netdev: network interface device structure
2093  *
2094  * The set_rx_mode entry point is called whenever the unicast or multicast
2095  * address lists or the network interface flags are updated. This routine is
2096  * responsible for configuring the hardware for proper unicast, multicast,
2097  * promiscuous mode, and all-multi behavior.
2098  **/
2099
2100 static void e1000_set_rx_mode(struct net_device *netdev)
2101 {
2102         struct e1000_adapter *adapter = netdev_priv(netdev);
2103         struct e1000_hw *hw = &adapter->hw;
2104         struct netdev_hw_addr *ha;
2105         bool use_uc = false;
2106         u32 rctl;
2107         u32 hash_value;
2108         int i, rar_entries = E1000_RAR_ENTRIES;
2109         int mta_reg_count = E1000_NUM_MTA_REGISTERS;
2110         u32 *mcarray = kcalloc(mta_reg_count, sizeof(u32), GFP_ATOMIC);
2111
2112         if (!mcarray) {
2113                 e_err("memory allocation failed\n");
2114                 return;
2115         }
2116
2117         /* Check for Promiscuous and All Multicast modes */
2118
2119         rctl = er32(RCTL);
2120
2121         if (netdev->flags & IFF_PROMISC) {
2122                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2123                 rctl &= ~E1000_RCTL_VFE;
2124         } else {
2125                 if (netdev->flags & IFF_ALLMULTI)
2126                         rctl |= E1000_RCTL_MPE;
2127                 else
2128                         rctl &= ~E1000_RCTL_MPE;
2129                 /* Enable VLAN filter if there is a VLAN */
2130                 if (adapter->vlgrp)
2131                         rctl |= E1000_RCTL_VFE;
2132         }
2133
2134         if (netdev_uc_count(netdev) > rar_entries - 1) {
2135                 rctl |= E1000_RCTL_UPE;
2136         } else if (!(netdev->flags & IFF_PROMISC)) {
2137                 rctl &= ~E1000_RCTL_UPE;
2138                 use_uc = true;
2139         }
2140
2141         ew32(RCTL, rctl);
2142
2143         /* 82542 2.0 needs to be in reset to write receive address registers */
2144
2145         if (hw->mac_type == e1000_82542_rev2_0)
2146                 e1000_enter_82542_rst(adapter);
2147
2148         /* load the first 14 addresses into the exact filters 1-14. Unicast
2149          * addresses take precedence to avoid disabling unicast filtering
2150          * when possible.
2151          *
2152          * RAR 0 is used for the station MAC adddress
2153          * if there are not 14 addresses, go ahead and clear the filters
2154          */
2155         i = 1;
2156         if (use_uc)
2157                 netdev_for_each_uc_addr(ha, netdev) {
2158                         if (i == rar_entries)
2159                                 break;
2160                         e1000_rar_set(hw, ha->addr, i++);
2161                 }
2162
2163         netdev_for_each_mc_addr(ha, netdev) {
2164                 if (i == rar_entries) {
2165                         /* load any remaining addresses into the hash table */
2166                         u32 hash_reg, hash_bit, mta;
2167                         hash_value = e1000_hash_mc_addr(hw, ha->addr);
2168                         hash_reg = (hash_value >> 5) & 0x7F;
2169                         hash_bit = hash_value & 0x1F;
2170                         mta = (1 << hash_bit);
2171                         mcarray[hash_reg] |= mta;
2172                 } else {
2173                         e1000_rar_set(hw, ha->addr, i++);
2174                 }
2175         }
2176
2177         for (; i < rar_entries; i++) {
2178                 E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2179                 E1000_WRITE_FLUSH();
2180                 E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2181                 E1000_WRITE_FLUSH();
2182         }
2183
2184         /* write the hash table completely, write from bottom to avoid
2185          * both stupid write combining chipsets, and flushing each write */
2186         for (i = mta_reg_count - 1; i >= 0 ; i--) {
2187                 /*
2188                  * If we are on an 82544 has an errata where writing odd
2189                  * offsets overwrites the previous even offset, but writing
2190                  * backwards over the range solves the issue by always
2191                  * writing the odd offset first
2192                  */
2193                 E1000_WRITE_REG_ARRAY(hw, MTA, i, mcarray[i]);
2194         }
2195         E1000_WRITE_FLUSH();
2196
2197         if (hw->mac_type == e1000_82542_rev2_0)
2198                 e1000_leave_82542_rst(adapter);
2199
2200         kfree(mcarray);
2201 }
2202
2203 /* Need to wait a few seconds after link up to get diagnostic information from
2204  * the phy */
2205
2206 static void e1000_update_phy_info(unsigned long data)
2207 {
2208         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2209         struct e1000_hw *hw = &adapter->hw;
2210         e1000_phy_get_info(hw, &adapter->phy_info);
2211 }
2212
2213 /**
2214  * e1000_82547_tx_fifo_stall - Timer Call-back
2215  * @data: pointer to adapter cast into an unsigned long
2216  **/
2217
2218 static void e1000_82547_tx_fifo_stall(unsigned long data)
2219 {
2220         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2221         struct e1000_hw *hw = &adapter->hw;
2222         struct net_device *netdev = adapter->netdev;
2223         u32 tctl;
2224
2225         if (atomic_read(&adapter->tx_fifo_stall)) {
2226                 if ((er32(TDT) == er32(TDH)) &&
2227                    (er32(TDFT) == er32(TDFH)) &&
2228                    (er32(TDFTS) == er32(TDFHS))) {
2229                         tctl = er32(TCTL);
2230                         ew32(TCTL, tctl & ~E1000_TCTL_EN);
2231                         ew32(TDFT, adapter->tx_head_addr);
2232                         ew32(TDFH, adapter->tx_head_addr);
2233                         ew32(TDFTS, adapter->tx_head_addr);
2234                         ew32(TDFHS, adapter->tx_head_addr);
2235                         ew32(TCTL, tctl);
2236                         E1000_WRITE_FLUSH();
2237
2238                         adapter->tx_fifo_head = 0;
2239                         atomic_set(&adapter->tx_fifo_stall, 0);
2240                         netif_wake_queue(netdev);
2241                 } else if (!test_bit(__E1000_DOWN, &adapter->flags)) {
2242                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2243                 }
2244         }
2245 }
2246
2247 bool e1000_has_link(struct e1000_adapter *adapter)
2248 {
2249         struct e1000_hw *hw = &adapter->hw;
2250         bool link_active = false;
2251
2252         /* get_link_status is set on LSC (link status) interrupt or
2253          * rx sequence error interrupt.  get_link_status will stay
2254          * false until the e1000_check_for_link establishes link
2255          * for copper adapters ONLY
2256          */
2257         switch (hw->media_type) {
2258         case e1000_media_type_copper:
2259                 if (hw->get_link_status) {
2260                         e1000_check_for_link(hw);
2261                         link_active = !hw->get_link_status;
2262                 } else {
2263                         link_active = true;
2264                 }
2265                 break;
2266         case e1000_media_type_fiber:
2267                 e1000_check_for_link(hw);
2268                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2269                 break;
2270         case e1000_media_type_internal_serdes:
2271                 e1000_check_for_link(hw);
2272                 link_active = hw->serdes_has_link;
2273                 break;
2274         default:
2275                 break;
2276         }
2277
2278         return link_active;
2279 }
2280
2281 /**
2282  * e1000_watchdog - Timer Call-back
2283  * @data: pointer to adapter cast into an unsigned long
2284  **/
2285 static void e1000_watchdog(unsigned long data)
2286 {
2287         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2288         struct e1000_hw *hw = &adapter->hw;
2289         struct net_device *netdev = adapter->netdev;
2290         struct e1000_tx_ring *txdr = adapter->tx_ring;
2291         u32 link, tctl;
2292
2293         link = e1000_has_link(adapter);
2294         if ((netif_carrier_ok(netdev)) && link)
2295                 goto link_up;
2296
2297         if (link) {
2298                 if (!netif_carrier_ok(netdev)) {
2299                         u32 ctrl;
2300                         bool txb2b = true;
2301                         /* update snapshot of PHY registers on LSC */
2302                         e1000_get_speed_and_duplex(hw,
2303                                                    &adapter->link_speed,
2304                                                    &adapter->link_duplex);
2305
2306                         ctrl = er32(CTRL);
2307                         pr_info("%s NIC Link is Up %d Mbps %s, "
2308                                 "Flow Control: %s\n",
2309                                 netdev->name,
2310                                 adapter->link_speed,
2311                                 adapter->link_duplex == FULL_DUPLEX ?
2312                                 "Full Duplex" : "Half Duplex",
2313                                 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2314                                 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2315                                 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2316                                 E1000_CTRL_TFCE) ? "TX" : "None")));
2317
2318                         /* adjust timeout factor according to speed/duplex */
2319                         adapter->tx_timeout_factor = 1;
2320                         switch (adapter->link_speed) {
2321                         case SPEED_10:
2322                                 txb2b = false;
2323                                 adapter->tx_timeout_factor = 16;
2324                                 break;
2325                         case SPEED_100:
2326                                 txb2b = false;
2327                                 /* maybe add some timeout factor ? */
2328                                 break;
2329                         }
2330
2331                         /* enable transmits in the hardware */
2332                         tctl = er32(TCTL);
2333                         tctl |= E1000_TCTL_EN;
2334                         ew32(TCTL, tctl);
2335
2336                         netif_carrier_on(netdev);
2337                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2338                                 mod_timer(&adapter->phy_info_timer,
2339                                           round_jiffies(jiffies + 2 * HZ));
2340                         adapter->smartspeed = 0;
2341                 }
2342         } else {
2343                 if (netif_carrier_ok(netdev)) {
2344                         adapter->link_speed = 0;
2345                         adapter->link_duplex = 0;
2346                         pr_info("%s NIC Link is Down\n",
2347                                 netdev->name);
2348                         netif_carrier_off(netdev);
2349
2350                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2351                                 mod_timer(&adapter->phy_info_timer,
2352                                           round_jiffies(jiffies + 2 * HZ));
2353                 }
2354
2355                 e1000_smartspeed(adapter);
2356         }
2357
2358 link_up:
2359         e1000_update_stats(adapter);
2360
2361         hw->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2362         adapter->tpt_old = adapter->stats.tpt;
2363         hw->collision_delta = adapter->stats.colc - adapter->colc_old;
2364         adapter->colc_old = adapter->stats.colc;
2365
2366         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2367         adapter->gorcl_old = adapter->stats.gorcl;
2368         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2369         adapter->gotcl_old = adapter->stats.gotcl;
2370
2371         e1000_update_adaptive(hw);
2372
2373         if (!netif_carrier_ok(netdev)) {
2374                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2375                         /* We've lost link, so the controller stops DMA,
2376                          * but we've got queued Tx work that's never going
2377                          * to get done, so reset controller to flush Tx.
2378                          * (Do the reset outside of interrupt context). */
2379                         adapter->tx_timeout_count++;
2380                         schedule_work(&adapter->reset_task);
2381                         /* return immediately since reset is imminent */
2382                         return;
2383                 }
2384         }
2385
2386         /* Simple mode for Interrupt Throttle Rate (ITR) */
2387         if (hw->mac_type >= e1000_82540 && adapter->itr_setting == 4) {
2388                 /*
2389                  * Symmetric Tx/Rx gets a reduced ITR=2000;
2390                  * Total asymmetrical Tx or Rx gets ITR=8000;
2391                  * everyone else is between 2000-8000.
2392                  */
2393                 u32 goc = (adapter->gotcl + adapter->gorcl) / 10000;
2394                 u32 dif = (adapter->gotcl > adapter->gorcl ?
2395                             adapter->gotcl - adapter->gorcl :
2396                             adapter->gorcl - adapter->gotcl) / 10000;
2397                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
2398
2399                 ew32(ITR, 1000000000 / (itr * 256));
2400         }
2401
2402         /* Cause software interrupt to ensure rx ring is cleaned */
2403         ew32(ICS, E1000_ICS_RXDMT0);
2404
2405         /* Force detection of hung controller every watchdog period */
2406         adapter->detect_tx_hung = true;
2407
2408         /* Reset the timer */
2409         if (!test_bit(__E1000_DOWN, &adapter->flags))
2410                 mod_timer(&adapter->watchdog_timer,
2411                           round_jiffies(jiffies + 2 * HZ));
2412 }
2413
2414 enum latency_range {
2415         lowest_latency = 0,
2416         low_latency = 1,
2417         bulk_latency = 2,
2418         latency_invalid = 255
2419 };
2420
2421 /**
2422  * e1000_update_itr - update the dynamic ITR value based on statistics
2423  * @adapter: pointer to adapter
2424  * @itr_setting: current adapter->itr
2425  * @packets: the number of packets during this measurement interval
2426  * @bytes: the number of bytes during this measurement interval
2427  *
2428  *      Stores a new ITR value based on packets and byte
2429  *      counts during the last interrupt.  The advantage of per interrupt
2430  *      computation is faster updates and more accurate ITR for the current
2431  *      traffic pattern.  Constants in this function were computed
2432  *      based on theoretical maximum wire speed and thresholds were set based
2433  *      on testing data as well as attempting to minimize response time
2434  *      while increasing bulk throughput.
2435  *      this functionality is controlled by the InterruptThrottleRate module
2436  *      parameter (see e1000_param.c)
2437  **/
2438 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2439                                      u16 itr_setting, int packets, int bytes)
2440 {
2441         unsigned int retval = itr_setting;
2442         struct e1000_hw *hw = &adapter->hw;
2443
2444         if (unlikely(hw->mac_type < e1000_82540))
2445                 goto update_itr_done;
2446
2447         if (packets == 0)
2448                 goto update_itr_done;
2449
2450         switch (itr_setting) {
2451         case lowest_latency:
2452                 /* jumbo frames get bulk treatment*/
2453                 if (bytes/packets > 8000)
2454                         retval = bulk_latency;
2455                 else if ((packets < 5) && (bytes > 512))
2456                         retval = low_latency;
2457                 break;
2458         case low_latency:  /* 50 usec aka 20000 ints/s */
2459                 if (bytes > 10000) {
2460                         /* jumbo frames need bulk latency setting */
2461                         if (bytes/packets > 8000)
2462                                 retval = bulk_latency;
2463                         else if ((packets < 10) || ((bytes/packets) > 1200))
2464                                 retval = bulk_latency;
2465                         else if ((packets > 35))
2466                                 retval = lowest_latency;
2467                 } else if (bytes/packets > 2000)
2468                         retval = bulk_latency;
2469                 else if (packets <= 2 && bytes < 512)
2470                         retval = lowest_latency;
2471                 break;
2472         case bulk_latency: /* 250 usec aka 4000 ints/s */
2473                 if (bytes > 25000) {
2474                         if (packets > 35)
2475                                 retval = low_latency;
2476                 } else if (bytes < 6000) {
2477                         retval = low_latency;
2478                 }
2479                 break;
2480         }
2481
2482 update_itr_done:
2483         return retval;
2484 }
2485
2486 static void e1000_set_itr(struct e1000_adapter *adapter)
2487 {
2488         struct e1000_hw *hw = &adapter->hw;
2489         u16 current_itr;
2490         u32 new_itr = adapter->itr;
2491
2492         if (unlikely(hw->mac_type < e1000_82540))
2493                 return;
2494
2495         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2496         if (unlikely(adapter->link_speed != SPEED_1000)) {
2497                 current_itr = 0;
2498                 new_itr = 4000;
2499                 goto set_itr_now;
2500         }
2501
2502         adapter->tx_itr = e1000_update_itr(adapter,
2503                                     adapter->tx_itr,
2504                                     adapter->total_tx_packets,
2505                                     adapter->total_tx_bytes);
2506         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2507         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2508                 adapter->tx_itr = low_latency;
2509
2510         adapter->rx_itr = e1000_update_itr(adapter,
2511                                     adapter->rx_itr,
2512                                     adapter->total_rx_packets,
2513                                     adapter->total_rx_bytes);
2514         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2515         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2516                 adapter->rx_itr = low_latency;
2517
2518         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2519
2520         switch (current_itr) {
2521         /* counts and packets in update_itr are dependent on these numbers */
2522         case lowest_latency:
2523                 new_itr = 70000;
2524                 break;
2525         case low_latency:
2526                 new_itr = 20000; /* aka hwitr = ~200 */
2527                 break;
2528         case bulk_latency:
2529                 new_itr = 4000;
2530                 break;
2531         default:
2532                 break;
2533         }
2534
2535 set_itr_now:
2536         if (new_itr != adapter->itr) {
2537                 /* this attempts to bias the interrupt rate towards Bulk
2538                  * by adding intermediate steps when interrupt rate is
2539                  * increasing */
2540                 new_itr = new_itr > adapter->itr ?
2541                              min(adapter->itr + (new_itr >> 2), new_itr) :
2542                              new_itr;
2543                 adapter->itr = new_itr;
2544                 ew32(ITR, 1000000000 / (new_itr * 256));
2545         }
2546 }
2547
2548 #define E1000_TX_FLAGS_CSUM             0x00000001
2549 #define E1000_TX_FLAGS_VLAN             0x00000002
2550 #define E1000_TX_FLAGS_TSO              0x00000004
2551 #define E1000_TX_FLAGS_IPV4             0x00000008
2552 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2553 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2554
2555 static int e1000_tso(struct e1000_adapter *adapter,
2556                      struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2557 {
2558         struct e1000_context_desc *context_desc;
2559         struct e1000_buffer *buffer_info;
2560         unsigned int i;
2561         u32 cmd_length = 0;
2562         u16 ipcse = 0, tucse, mss;
2563         u8 ipcss, ipcso, tucss, tucso, hdr_len;
2564         int err;
2565
2566         if (skb_is_gso(skb)) {
2567                 if (skb_header_cloned(skb)) {
2568                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2569                         if (err)
2570                                 return err;
2571                 }
2572
2573                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2574                 mss = skb_shinfo(skb)->gso_size;
2575                 if (skb->protocol == htons(ETH_P_IP)) {
2576                         struct iphdr *iph = ip_hdr(skb);
2577                         iph->tot_len = 0;
2578                         iph->check = 0;
2579                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2580                                                                  iph->daddr, 0,
2581                                                                  IPPROTO_TCP,
2582                                                                  0);
2583                         cmd_length = E1000_TXD_CMD_IP;
2584                         ipcse = skb_transport_offset(skb) - 1;
2585                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2586                         ipv6_hdr(skb)->payload_len = 0;
2587                         tcp_hdr(skb)->check =
2588                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2589                                                  &ipv6_hdr(skb)->daddr,
2590                                                  0, IPPROTO_TCP, 0);
2591                         ipcse = 0;
2592                 }
2593                 ipcss = skb_network_offset(skb);
2594                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2595                 tucss = skb_transport_offset(skb);
2596                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2597                 tucse = 0;
2598
2599                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2600                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2601
2602                 i = tx_ring->next_to_use;
2603                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2604                 buffer_info = &tx_ring->buffer_info[i];
2605
2606                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2607                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2608                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2609                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2610                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2611                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2612                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2613                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2614                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2615
2616                 buffer_info->time_stamp = jiffies;
2617                 buffer_info->next_to_watch = i;
2618
2619                 if (++i == tx_ring->count) i = 0;
2620                 tx_ring->next_to_use = i;
2621
2622                 return true;
2623         }
2624         return false;
2625 }
2626
2627 static bool e1000_tx_csum(struct e1000_adapter *adapter,
2628                           struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2629 {
2630         struct e1000_context_desc *context_desc;
2631         struct e1000_buffer *buffer_info;
2632         unsigned int i;
2633         u8 css;
2634         u32 cmd_len = E1000_TXD_CMD_DEXT;
2635
2636         if (skb->ip_summed != CHECKSUM_PARTIAL)
2637                 return false;
2638
2639         switch (skb->protocol) {
2640         case cpu_to_be16(ETH_P_IP):
2641                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2642                         cmd_len |= E1000_TXD_CMD_TCP;
2643                 break;
2644         case cpu_to_be16(ETH_P_IPV6):
2645                 /* XXX not handling all IPV6 headers */
2646                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2647                         cmd_len |= E1000_TXD_CMD_TCP;
2648                 break;
2649         default:
2650                 if (unlikely(net_ratelimit()))
2651                         e_warn("checksum_partial proto=%x!\n", skb->protocol);
2652                 break;
2653         }
2654
2655         css = skb_transport_offset(skb);
2656
2657         i = tx_ring->next_to_use;
2658         buffer_info = &tx_ring->buffer_info[i];
2659         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2660
2661         context_desc->lower_setup.ip_config = 0;
2662         context_desc->upper_setup.tcp_fields.tucss = css;
2663         context_desc->upper_setup.tcp_fields.tucso =
2664                 css + skb->csum_offset;
2665         context_desc->upper_setup.tcp_fields.tucse = 0;
2666         context_desc->tcp_seg_setup.data = 0;
2667         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
2668
2669         buffer_info->time_stamp = jiffies;
2670         buffer_info->next_to_watch = i;
2671
2672         if (unlikely(++i == tx_ring->count)) i = 0;
2673         tx_ring->next_to_use = i;
2674
2675         return true;
2676 }
2677
2678 #define E1000_MAX_TXD_PWR       12
2679 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2680
2681 static int e1000_tx_map(struct e1000_adapter *adapter,
2682                         struct e1000_tx_ring *tx_ring,
2683                         struct sk_buff *skb, unsigned int first,
2684                         unsigned int max_per_txd, unsigned int nr_frags,
2685                         unsigned int mss)
2686 {
2687         struct e1000_hw *hw = &adapter->hw;
2688         struct pci_dev *pdev = adapter->pdev;
2689         struct e1000_buffer *buffer_info;
2690         unsigned int len = skb_headlen(skb);
2691         unsigned int offset = 0, size, count = 0, i;
2692         unsigned int f;
2693
2694         i = tx_ring->next_to_use;
2695
2696         while (len) {
2697                 buffer_info = &tx_ring->buffer_info[i];
2698                 size = min(len, max_per_txd);
2699                 /* Workaround for Controller erratum --
2700                  * descriptor for non-tso packet in a linear SKB that follows a
2701                  * tso gets written back prematurely before the data is fully
2702                  * DMA'd to the controller */
2703                 if (!skb->data_len && tx_ring->last_tx_tso &&
2704                     !skb_is_gso(skb)) {
2705                         tx_ring->last_tx_tso = 0;
2706                         size -= 4;
2707                 }
2708
2709                 /* Workaround for premature desc write-backs
2710                  * in TSO mode.  Append 4-byte sentinel desc */
2711                 if (unlikely(mss && !nr_frags && size == len && size > 8))
2712                         size -= 4;
2713                 /* work-around for errata 10 and it applies
2714                  * to all controllers in PCI-X mode
2715                  * The fix is to make sure that the first descriptor of a
2716                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2717                  */
2718                 if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
2719                                 (size > 2015) && count == 0))
2720                         size = 2015;
2721
2722                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
2723                  * terminating buffers within evenly-aligned dwords. */
2724                 if (unlikely(adapter->pcix_82544 &&
2725                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2726                    size > 4))
2727                         size -= 4;
2728
2729                 buffer_info->length = size;
2730                 /* set time_stamp *before* dma to help avoid a possible race */
2731                 buffer_info->time_stamp = jiffies;
2732                 buffer_info->mapped_as_page = false;
2733                 buffer_info->dma = dma_map_single(&pdev->dev,
2734                                                   skb->data + offset,
2735                                                   size, DMA_TO_DEVICE);
2736                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2737                         goto dma_error;
2738                 buffer_info->next_to_watch = i;
2739
2740                 len -= size;
2741                 offset += size;
2742                 count++;
2743                 if (len) {
2744                         i++;
2745                         if (unlikely(i == tx_ring->count))
2746                                 i = 0;
2747                 }
2748         }
2749
2750         for (f = 0; f < nr_frags; f++) {
2751                 struct skb_frag_struct *frag;
2752
2753                 frag = &skb_shinfo(skb)->frags[f];
2754                 len = frag->size;
2755                 offset = frag->page_offset;
2756
2757                 while (len) {
2758                         i++;
2759                         if (unlikely(i == tx_ring->count))
2760                                 i = 0;
2761
2762                         buffer_info = &tx_ring->buffer_info[i];
2763                         size = min(len, max_per_txd);
2764                         /* Workaround for premature desc write-backs
2765                          * in TSO mode.  Append 4-byte sentinel desc */
2766                         if (unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
2767                                 size -= 4;
2768                         /* Workaround for potential 82544 hang in PCI-X.
2769                          * Avoid terminating buffers within evenly-aligned
2770                          * dwords. */
2771                         if (unlikely(adapter->pcix_82544 &&
2772                             !((unsigned long)(page_to_phys(frag->page) + offset
2773                                               + size - 1) & 4) &&
2774                             size > 4))
2775                                 size -= 4;
2776
2777                         buffer_info->length = size;
2778                         buffer_info->time_stamp = jiffies;
2779                         buffer_info->mapped_as_page = true;
2780                         buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
2781                                                         offset, size,
2782                                                         DMA_TO_DEVICE);
2783                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2784                                 goto dma_error;
2785                         buffer_info->next_to_watch = i;
2786
2787                         len -= size;
2788                         offset += size;
2789                         count++;
2790                 }
2791         }
2792
2793         tx_ring->buffer_info[i].skb = skb;
2794         tx_ring->buffer_info[first].next_to_watch = i;
2795
2796         return count;
2797
2798 dma_error:
2799         dev_err(&pdev->dev, "TX DMA map failed\n");
2800         buffer_info->dma = 0;
2801         if (count)
2802                 count--;
2803
2804         while (count--) {
2805                 if (i==0)
2806                         i += tx_ring->count;
2807                 i--;
2808                 buffer_info = &tx_ring->buffer_info[i];
2809                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2810         }
2811
2812         return 0;
2813 }
2814
2815 static void e1000_tx_queue(struct e1000_adapter *adapter,
2816                            struct e1000_tx_ring *tx_ring, int tx_flags,
2817                            int count)
2818 {
2819         struct e1000_hw *hw = &adapter->hw;
2820         struct e1000_tx_desc *tx_desc = NULL;
2821         struct e1000_buffer *buffer_info;
2822         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2823         unsigned int i;
2824
2825         if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
2826                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2827                              E1000_TXD_CMD_TSE;
2828                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2829
2830                 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
2831                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2832         }
2833
2834         if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
2835                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
2836                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2837         }
2838
2839         if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
2840                 txd_lower |= E1000_TXD_CMD_VLE;
2841                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
2842         }
2843
2844         i = tx_ring->next_to_use;
2845
2846         while (count--) {
2847                 buffer_info = &tx_ring->buffer_info[i];
2848                 tx_desc = E1000_TX_DESC(*tx_ring, i);
2849                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
2850                 tx_desc->lower.data =
2851                         cpu_to_le32(txd_lower | buffer_info->length);
2852                 tx_desc->upper.data = cpu_to_le32(txd_upper);
2853                 if (unlikely(++i == tx_ring->count)) i = 0;
2854         }
2855
2856         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
2857
2858         /* Force memory writes to complete before letting h/w
2859          * know there are new descriptors to fetch.  (Only
2860          * applicable for weak-ordered memory model archs,
2861          * such as IA-64). */
2862         wmb();
2863
2864         tx_ring->next_to_use = i;
2865         writel(i, hw->hw_addr + tx_ring->tdt);
2866         /* we need this if more than one processor can write to our tail
2867          * at a time, it syncronizes IO on IA64/Altix systems */
2868         mmiowb();
2869 }
2870
2871 /**
2872  * 82547 workaround to avoid controller hang in half-duplex environment.
2873  * The workaround is to avoid queuing a large packet that would span
2874  * the internal Tx FIFO ring boundary by notifying the stack to resend
2875  * the packet at a later time.  This gives the Tx FIFO an opportunity to
2876  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
2877  * to the beginning of the Tx FIFO.
2878  **/
2879
2880 #define E1000_FIFO_HDR                  0x10
2881 #define E1000_82547_PAD_LEN             0x3E0
2882
2883 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
2884                                        struct sk_buff *skb)
2885 {
2886         u32 fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
2887         u32 skb_fifo_len = skb->len + E1000_FIFO_HDR;
2888
2889         skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
2890
2891         if (adapter->link_duplex != HALF_DUPLEX)
2892                 goto no_fifo_stall_required;
2893
2894         if (atomic_read(&adapter->tx_fifo_stall))
2895                 return 1;
2896
2897         if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
2898                 atomic_set(&adapter->tx_fifo_stall, 1);
2899                 return 1;
2900         }
2901
2902 no_fifo_stall_required:
2903         adapter->tx_fifo_head += skb_fifo_len;
2904         if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
2905                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
2906         return 0;
2907 }
2908
2909 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
2910 {
2911         struct e1000_adapter *adapter = netdev_priv(netdev);
2912         struct e1000_tx_ring *tx_ring = adapter->tx_ring;
2913
2914         netif_stop_queue(netdev);
2915         /* Herbert's original patch had:
2916          *  smp_mb__after_netif_stop_queue();
2917          * but since that doesn't exist yet, just open code it. */
2918         smp_mb();
2919
2920         /* We need to check again in a case another CPU has just
2921          * made room available. */
2922         if (likely(E1000_DESC_UNUSED(tx_ring) < size))
2923                 return -EBUSY;
2924
2925         /* A reprieve! */
2926         netif_start_queue(netdev);
2927         ++adapter->restart_queue;
2928         return 0;
2929 }
2930
2931 static int e1000_maybe_stop_tx(struct net_device *netdev,
2932                                struct e1000_tx_ring *tx_ring, int size)
2933 {
2934         if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
2935                 return 0;
2936         return __e1000_maybe_stop_tx(netdev, size);
2937 }
2938
2939 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
2940 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
2941                                     struct net_device *netdev)
2942 {
2943         struct e1000_adapter *adapter = netdev_priv(netdev);
2944         struct e1000_hw *hw = &adapter->hw;
2945         struct e1000_tx_ring *tx_ring;
2946         unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
2947         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
2948         unsigned int tx_flags = 0;
2949         unsigned int len = skb_headlen(skb);
2950         unsigned int nr_frags;
2951         unsigned int mss;
2952         int count = 0;
2953         int tso;
2954         unsigned int f;
2955
2956         /* This goes back to the question of how to logically map a tx queue
2957          * to a flow.  Right now, performance is impacted slightly negatively
2958          * if using multiple tx queues.  If the stack breaks away from a
2959          * single qdisc implementation, we can look at this again. */
2960         tx_ring = adapter->tx_ring;
2961
2962         if (unlikely(skb->len <= 0)) {
2963                 dev_kfree_skb_any(skb);
2964                 return NETDEV_TX_OK;
2965         }
2966
2967         mss = skb_shinfo(skb)->gso_size;
2968         /* The controller does a simple calculation to
2969          * make sure there is enough room in the FIFO before
2970          * initiating the DMA for each buffer.  The calc is:
2971          * 4 = ceil(buffer len/mss).  To make sure we don't
2972          * overrun the FIFO, adjust the max buffer len if mss
2973          * drops. */
2974         if (mss) {
2975                 u8 hdr_len;
2976                 max_per_txd = min(mss << 2, max_per_txd);
2977                 max_txd_pwr = fls(max_per_txd) - 1;
2978
2979                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2980                 if (skb->data_len && hdr_len == len) {
2981                         switch (hw->mac_type) {
2982                                 unsigned int pull_size;
2983                         case e1000_82544:
2984                                 /* Make sure we have room to chop off 4 bytes,
2985                                  * and that the end alignment will work out to
2986                                  * this hardware's requirements
2987                                  * NOTE: this is a TSO only workaround
2988                                  * if end byte alignment not correct move us
2989                                  * into the next dword */
2990                                 if ((unsigned long)(skb_tail_pointer(skb) - 1) & 4)
2991                                         break;
2992                                 /* fall through */
2993                                 pull_size = min((unsigned int)4, skb->data_len);
2994                                 if (!__pskb_pull_tail(skb, pull_size)) {
2995                                         e_err("__pskb_pull_tail failed.\n");
2996                                         dev_kfree_skb_any(skb);
2997                                         return NETDEV_TX_OK;
2998                                 }
2999                                 len = skb_headlen(skb);
3000                                 break;
3001                         default:
3002                                 /* do nothing */
3003                                 break;
3004                         }
3005                 }
3006         }
3007
3008         /* reserve a descriptor for the offload context */
3009         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3010                 count++;
3011         count++;
3012
3013         /* Controller Erratum workaround */
3014         if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3015                 count++;
3016
3017         count += TXD_USE_COUNT(len, max_txd_pwr);
3018
3019         if (adapter->pcix_82544)
3020                 count++;
3021
3022         /* work-around for errata 10 and it applies to all controllers
3023          * in PCI-X mode, so add one more descriptor to the count
3024          */
3025         if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
3026                         (len > 2015)))
3027                 count++;
3028
3029         nr_frags = skb_shinfo(skb)->nr_frags;
3030         for (f = 0; f < nr_frags; f++)
3031                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3032                                        max_txd_pwr);
3033         if (adapter->pcix_82544)
3034                 count += nr_frags;
3035
3036         /* need: count + 2 desc gap to keep tail from touching
3037          * head, otherwise try next time */
3038         if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2)))
3039                 return NETDEV_TX_BUSY;
3040
3041         if (unlikely(hw->mac_type == e1000_82547)) {
3042                 if (unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
3043                         netif_stop_queue(netdev);
3044                         if (!test_bit(__E1000_DOWN, &adapter->flags))
3045                                 mod_timer(&adapter->tx_fifo_stall_timer,
3046                                           jiffies + 1);
3047                         return NETDEV_TX_BUSY;
3048                 }
3049         }
3050
3051         if (unlikely(adapter->vlgrp && vlan_tx_tag_present(skb))) {
3052                 tx_flags |= E1000_TX_FLAGS_VLAN;
3053                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3054         }
3055
3056         first = tx_ring->next_to_use;
3057
3058         tso = e1000_tso(adapter, tx_ring, skb);
3059         if (tso < 0) {
3060                 dev_kfree_skb_any(skb);
3061                 return NETDEV_TX_OK;
3062         }
3063
3064         if (likely(tso)) {
3065                 if (likely(hw->mac_type != e1000_82544))
3066                         tx_ring->last_tx_tso = 1;
3067                 tx_flags |= E1000_TX_FLAGS_TSO;
3068         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3069                 tx_flags |= E1000_TX_FLAGS_CSUM;
3070
3071         if (likely(skb->protocol == htons(ETH_P_IP)))
3072                 tx_flags |= E1000_TX_FLAGS_IPV4;
3073
3074         count = e1000_tx_map(adapter, tx_ring, skb, first, max_per_txd,
3075                              nr_frags, mss);
3076
3077         if (count) {
3078                 e1000_tx_queue(adapter, tx_ring, tx_flags, count);
3079                 /* Make sure there is space in the ring for the next send. */
3080                 e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3081
3082         } else {
3083                 dev_kfree_skb_any(skb);
3084                 tx_ring->buffer_info[first].time_stamp = 0;
3085                 tx_ring->next_to_use = first;
3086         }
3087
3088         return NETDEV_TX_OK;
3089 }
3090
3091 /**
3092  * e1000_tx_timeout - Respond to a Tx Hang
3093  * @netdev: network interface device structure
3094  **/
3095
3096 static void e1000_tx_timeout(struct net_device *netdev)
3097 {
3098         struct e1000_adapter *adapter = netdev_priv(netdev);
3099
3100         /* Do the reset outside of interrupt context */
3101         adapter->tx_timeout_count++;
3102         schedule_work(&adapter->reset_task);
3103 }
3104
3105 static void e1000_reset_task(struct work_struct *work)
3106 {
3107         struct e1000_adapter *adapter =
3108                 container_of(work, struct e1000_adapter, reset_task);
3109
3110         e1000_reinit_locked(adapter);
3111 }
3112
3113 /**
3114  * e1000_get_stats - Get System Network Statistics
3115  * @netdev: network interface device structure
3116  *
3117  * Returns the address of the device statistics structure.
3118  * The statistics are actually updated from the timer callback.
3119  **/
3120
3121 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3122 {
3123         /* only return the current stats */
3124         return &netdev->stats;
3125 }
3126
3127 /**
3128  * e1000_change_mtu - Change the Maximum Transfer Unit
3129  * @netdev: network interface device structure
3130  * @new_mtu: new value for maximum frame size
3131  *
3132  * Returns 0 on success, negative on failure
3133  **/
3134
3135 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3136 {
3137         struct e1000_adapter *adapter = netdev_priv(netdev);
3138         struct e1000_hw *hw = &adapter->hw;
3139         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3140
3141         if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3142             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3143                 e_err("Invalid MTU setting\n");
3144                 return -EINVAL;
3145         }
3146
3147         /* Adapter-specific max frame size limits. */
3148         switch (hw->mac_type) {
3149         case e1000_undefined ... e1000_82542_rev2_1:
3150                 if (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)) {
3151                         e_err("Jumbo Frames not supported.\n");
3152                         return -EINVAL;
3153                 }
3154                 break;
3155         default:
3156                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3157                 break;
3158         }
3159
3160         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
3161                 msleep(1);
3162         /* e1000_down has a dependency on max_frame_size */
3163         hw->max_frame_size = max_frame;
3164         if (netif_running(netdev))
3165                 e1000_down(adapter);
3166
3167         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3168          * means we reserve 2 more, this pushes us to allocate from the next
3169          * larger slab size.
3170          * i.e. RXBUFFER_2048 --> size-4096 slab
3171          *  however with the new *_jumbo_rx* routines, jumbo receives will use
3172          *  fragmented skbs */
3173
3174         if (max_frame <= E1000_RXBUFFER_2048)
3175                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3176         else
3177 #if (PAGE_SIZE >= E1000_RXBUFFER_16384)
3178                 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3179 #elif (PAGE_SIZE >= E1000_RXBUFFER_4096)
3180                 adapter->rx_buffer_len = PAGE_SIZE;
3181 #endif
3182
3183         /* adjust allocation if LPE protects us, and we aren't using SBP */
3184         if (!hw->tbi_compatibility_on &&
3185             ((max_frame == (ETH_FRAME_LEN + ETH_FCS_LEN)) ||
3186              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3187                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3188
3189         pr_info("%s changing MTU from %d to %d\n",
3190                 netdev->name, netdev->mtu, new_mtu);
3191         netdev->mtu = new_mtu;
3192
3193         if (netif_running(netdev))
3194                 e1000_up(adapter);
3195         else
3196                 e1000_reset(adapter);
3197
3198         clear_bit(__E1000_RESETTING, &adapter->flags);
3199
3200         return 0;
3201 }
3202
3203 /**
3204  * e1000_update_stats - Update the board statistics counters
3205  * @adapter: board private structure
3206  **/
3207
3208 void e1000_update_stats(struct e1000_adapter *adapter)
3209 {
3210         struct net_device *netdev = adapter->netdev;
3211         struct e1000_hw *hw = &adapter->hw;
3212         struct pci_dev *pdev = adapter->pdev;
3213         unsigned long flags;
3214         u16 phy_tmp;
3215
3216 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3217
3218         /*
3219          * Prevent stats update while adapter is being reset, or if the pci
3220          * connection is down.
3221          */
3222         if (adapter->link_speed == 0)
3223                 return;
3224         if (pci_channel_offline(pdev))
3225                 return;
3226
3227         spin_lock_irqsave(&adapter->stats_lock, flags);
3228
3229         /* these counters are modified from e1000_tbi_adjust_stats,
3230          * called from the interrupt context, so they must only
3231          * be written while holding adapter->stats_lock
3232          */
3233
3234         adapter->stats.crcerrs += er32(CRCERRS);
3235         adapter->stats.gprc += er32(GPRC);
3236         adapter->stats.gorcl += er32(GORCL);
3237         adapter->stats.gorch += er32(GORCH);
3238         adapter->stats.bprc += er32(BPRC);
3239         adapter->stats.mprc += er32(MPRC);
3240         adapter->stats.roc += er32(ROC);
3241
3242         adapter->stats.prc64 += er32(PRC64);
3243         adapter->stats.prc127 += er32(PRC127);
3244         adapter->stats.prc255 += er32(PRC255);
3245         adapter->stats.prc511 += er32(PRC511);
3246         adapter->stats.prc1023 += er32(PRC1023);
3247         adapter->stats.prc1522 += er32(PRC1522);
3248
3249         adapter->stats.symerrs += er32(SYMERRS);
3250         adapter->stats.mpc += er32(MPC);
3251         adapter->stats.scc += er32(SCC);
3252         adapter->stats.ecol += er32(ECOL);
3253         adapter->stats.mcc += er32(MCC);
3254         adapter->stats.latecol += er32(LATECOL);
3255         adapter->stats.dc += er32(DC);
3256         adapter->stats.sec += er32(SEC);
3257         adapter->stats.rlec += er32(RLEC);
3258         adapter->stats.xonrxc += er32(XONRXC);
3259         adapter->stats.xontxc += er32(XONTXC);
3260         adapter->stats.xoffrxc += er32(XOFFRXC);
3261         adapter->stats.xofftxc += er32(XOFFTXC);
3262         adapter->stats.fcruc += er32(FCRUC);
3263         adapter->stats.gptc += er32(GPTC);
3264         adapter->stats.gotcl += er32(GOTCL);
3265         adapter->stats.gotch += er32(GOTCH);
3266         adapter->stats.rnbc += er32(RNBC);
3267         adapter->stats.ruc += er32(RUC);
3268         adapter->stats.rfc += er32(RFC);
3269         adapter->stats.rjc += er32(RJC);
3270         adapter->stats.torl += er32(TORL);
3271         adapter->stats.torh += er32(TORH);
3272         adapter->stats.totl += er32(TOTL);
3273         adapter->stats.toth += er32(TOTH);
3274         adapter->stats.tpr += er32(TPR);
3275
3276         adapter->stats.ptc64 += er32(PTC64);
3277         adapter->stats.ptc127 += er32(PTC127);
3278         adapter->stats.ptc255 += er32(PTC255);
3279         adapter->stats.ptc511 += er32(PTC511);
3280         adapter->stats.ptc1023 += er32(PTC1023);
3281         adapter->stats.ptc1522 += er32(PTC1522);
3282
3283         adapter->stats.mptc += er32(MPTC);
3284         adapter->stats.bptc += er32(BPTC);
3285
3286         /* used for adaptive IFS */
3287
3288         hw->tx_packet_delta = er32(TPT);
3289         adapter->stats.tpt += hw->tx_packet_delta;
3290         hw->collision_delta = er32(COLC);
3291         adapter->stats.colc += hw->collision_delta;
3292
3293         if (hw->mac_type >= e1000_82543) {
3294                 adapter->stats.algnerrc += er32(ALGNERRC);
3295                 adapter->stats.rxerrc += er32(RXERRC);
3296                 adapter->stats.tncrs += er32(TNCRS);
3297                 adapter->stats.cexterr += er32(CEXTERR);
3298                 adapter->stats.tsctc += er32(TSCTC);
3299                 adapter->stats.tsctfc += er32(TSCTFC);
3300         }
3301
3302         /* Fill out the OS statistics structure */
3303         netdev->stats.multicast = adapter->stats.mprc;
3304         netdev->stats.collisions = adapter->stats.colc;
3305
3306         /* Rx Errors */
3307
3308         /* RLEC on some newer hardware can be incorrect so build
3309         * our own version based on RUC and ROC */
3310         netdev->stats.rx_errors = adapter->stats.rxerrc +
3311                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3312                 adapter->stats.ruc + adapter->stats.roc +
3313                 adapter->stats.cexterr;
3314         adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3315         netdev->stats.rx_length_errors = adapter->stats.rlerrc;
3316         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3317         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3318         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3319
3320         /* Tx Errors */
3321         adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3322         netdev->stats.tx_errors = adapter->stats.txerrc;
3323         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3324         netdev->stats.tx_window_errors = adapter->stats.latecol;
3325         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3326         if (hw->bad_tx_carr_stats_fd &&
3327             adapter->link_duplex == FULL_DUPLEX) {
3328                 netdev->stats.tx_carrier_errors = 0;
3329                 adapter->stats.tncrs = 0;
3330         }
3331
3332         /* Tx Dropped needs to be maintained elsewhere */
3333
3334         /* Phy Stats */
3335         if (hw->media_type == e1000_media_type_copper) {
3336                 if ((adapter->link_speed == SPEED_1000) &&
3337                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3338                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3339                         adapter->phy_stats.idle_errors += phy_tmp;
3340                 }
3341
3342                 if ((hw->mac_type <= e1000_82546) &&
3343                    (hw->phy_type == e1000_phy_m88) &&
3344                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3345                         adapter->phy_stats.receive_errors += phy_tmp;
3346         }
3347
3348         /* Management Stats */
3349         if (hw->has_smbus) {
3350                 adapter->stats.mgptc += er32(MGTPTC);
3351                 adapter->stats.mgprc += er32(MGTPRC);
3352                 adapter->stats.mgpdc += er32(MGTPDC);
3353         }
3354
3355         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3356 }
3357
3358 /**
3359  * e1000_intr - Interrupt Handler
3360  * @irq: interrupt number
3361  * @data: pointer to a network interface device structure
3362  **/
3363
3364 static irqreturn_t e1000_intr(int irq, void *data)
3365 {
3366         struct net_device *netdev = data;
3367         struct e1000_adapter *adapter = netdev_priv(netdev);
3368         struct e1000_hw *hw = &adapter->hw;
3369         u32 icr = er32(ICR);
3370
3371         if (unlikely((!icr) || test_bit(__E1000_DOWN, &adapter->flags)))
3372                 return IRQ_NONE;  /* Not our interrupt */
3373
3374         if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3375                 hw->get_link_status = 1;
3376                 /* guard against interrupt when we're going down */
3377                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3378                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3379         }
3380
3381         /* disable interrupts, without the synchronize_irq bit */
3382         ew32(IMC, ~0);
3383         E1000_WRITE_FLUSH();
3384
3385         if (likely(napi_schedule_prep(&adapter->napi))) {
3386                 adapter->total_tx_bytes = 0;
3387                 adapter->total_tx_packets = 0;
3388                 adapter->total_rx_bytes = 0;
3389                 adapter->total_rx_packets = 0;
3390                 __napi_schedule(&adapter->napi);
3391         } else {
3392                 /* this really should not happen! if it does it is basically a
3393                  * bug, but not a hard error, so enable ints and continue */
3394                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3395                         e1000_irq_enable(adapter);
3396         }
3397
3398         return IRQ_HANDLED;
3399 }
3400
3401 /**
3402  * e1000_clean - NAPI Rx polling callback
3403  * @adapter: board private structure
3404  **/
3405 static int e1000_clean(struct napi_struct *napi, int budget)
3406 {
3407         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
3408         int tx_clean_complete = 0, work_done = 0;
3409
3410         tx_clean_complete = e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);
3411
3412         adapter->clean_rx(adapter, &adapter->rx_ring[0], &work_done, budget);
3413
3414         if (!tx_clean_complete)
3415                 work_done = budget;
3416
3417         /* If budget not fully consumed, exit the polling mode */
3418         if (work_done < budget) {
3419                 if (likely(adapter->itr_setting & 3))
3420                         e1000_set_itr(adapter);
3421                 napi_complete(napi);
3422                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3423                         e1000_irq_enable(adapter);
3424         }
3425
3426         return work_done;
3427 }
3428
3429 /**
3430  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3431  * @adapter: board private structure
3432  **/
3433 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
3434                                struct e1000_tx_ring *tx_ring)
3435 {
3436         struct e1000_hw *hw = &adapter->hw;
3437         struct net_device *netdev = adapter->netdev;
3438         struct e1000_tx_desc *tx_desc, *eop_desc;
3439         struct e1000_buffer *buffer_info;
3440         unsigned int i, eop;
3441         unsigned int count = 0;
3442         unsigned int total_tx_bytes=0, total_tx_packets=0;
3443
3444         i = tx_ring->next_to_clean;
3445         eop = tx_ring->buffer_info[i].next_to_watch;
3446         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3447
3448         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
3449                (count < tx_ring->count)) {
3450                 bool cleaned = false;
3451                 for ( ; !cleaned; count++) {
3452                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3453                         buffer_info = &tx_ring->buffer_info[i];
3454                         cleaned = (i == eop);
3455
3456                         if (cleaned) {
3457                                 struct sk_buff *skb = buffer_info->skb;
3458                                 unsigned int segs, bytecount;
3459                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3460                                 /* multiply data chunks by size of headers */
3461                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3462                                             skb->len;
3463                                 total_tx_packets += segs;
3464                                 total_tx_bytes += bytecount;
3465                         }
3466                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3467                         tx_desc->upper.data = 0;
3468
3469                         if (unlikely(++i == tx_ring->count)) i = 0;
3470                 }
3471
3472                 eop = tx_ring->buffer_info[i].next_to_watch;
3473                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3474         }
3475
3476         tx_ring->next_to_clean = i;
3477
3478 #define TX_WAKE_THRESHOLD 32
3479         if (unlikely(count && netif_carrier_ok(netdev) &&
3480                      E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
3481                 /* Make sure that anybody stopping the queue after this
3482                  * sees the new next_to_clean.
3483                  */
3484                 smp_mb();
3485
3486                 if (netif_queue_stopped(netdev) &&
3487                     !(test_bit(__E1000_DOWN, &adapter->flags))) {
3488                         netif_wake_queue(netdev);
3489                         ++adapter->restart_queue;
3490                 }
3491         }
3492
3493         if (adapter->detect_tx_hung) {
3494                 /* Detect a transmit hang in hardware, this serializes the
3495                  * check with the clearing of time_stamp and movement of i */
3496                 adapter->detect_tx_hung = false;
3497                 if (tx_ring->buffer_info[eop].time_stamp &&
3498                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
3499                                (adapter->tx_timeout_factor * HZ)) &&
3500                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
3501
3502                         /* detected Tx unit hang */
3503                         e_err("Detected Tx Unit Hang\n"
3504                               "  Tx Queue             <%lu>\n"
3505                               "  TDH                  <%x>\n"
3506                               "  TDT                  <%x>\n"
3507                               "  next_to_use          <%x>\n"
3508                               "  next_to_clean        <%x>\n"
3509                               "buffer_info[next_to_clean]\n"
3510                               "  time_stamp           <%lx>\n"
3511                               "  next_to_watch        <%x>\n"
3512                               "  jiffies              <%lx>\n"
3513                               "  next_to_watch.status <%x>\n",
3514                                 (unsigned long)((tx_ring - adapter->tx_ring) /
3515                                         sizeof(struct e1000_tx_ring)),
3516                                 readl(hw->hw_addr + tx_ring->tdh),
3517                                 readl(hw->hw_addr + tx_ring->tdt),
3518                                 tx_ring->next_to_use,
3519                                 tx_ring->next_to_clean,
3520                                 tx_ring->buffer_info[eop].time_stamp,
3521                                 eop,
3522                                 jiffies,
3523                                 eop_desc->upper.fields.status);
3524                         netif_stop_queue(netdev);
3525                 }
3526         }
3527         adapter->total_tx_bytes += total_tx_bytes;
3528         adapter->total_tx_packets += total_tx_packets;
3529         netdev->stats.tx_bytes += total_tx_bytes;
3530         netdev->stats.tx_packets += total_tx_packets;
3531         return (count < tx_ring->count);
3532 }
3533
3534 /**
3535  * e1000_rx_checksum - Receive Checksum Offload for 82543
3536  * @adapter:     board private structure
3537  * @status_err:  receive descriptor status and error fields
3538  * @csum:        receive descriptor csum field
3539  * @sk_buff:     socket buffer with received data
3540  **/
3541
3542 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
3543                               u32 csum, struct sk_buff *skb)
3544 {
3545         struct e1000_hw *hw = &adapter->hw;
3546         u16 status = (u16)status_err;
3547         u8 errors = (u8)(status_err >> 24);
3548         skb->ip_summed = CHECKSUM_NONE;
3549
3550         /* 82543 or newer only */
3551         if (unlikely(hw->mac_type < e1000_82543)) return;
3552         /* Ignore Checksum bit is set */
3553         if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
3554         /* TCP/UDP checksum error bit is set */
3555         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
3556                 /* let the stack verify checksum errors */
3557                 adapter->hw_csum_err++;
3558                 return;
3559         }
3560         /* TCP/UDP Checksum has not been calculated */
3561         if (!(status & E1000_RXD_STAT_TCPCS))
3562                 return;
3563
3564         /* It must be a TCP or UDP packet with a valid checksum */
3565         if (likely(status & E1000_RXD_STAT_TCPCS)) {
3566                 /* TCP checksum is good */
3567                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3568         }
3569         adapter->hw_csum_good++;
3570 }
3571
3572 /**
3573  * e1000_consume_page - helper function
3574  **/
3575 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
3576                                u16 length)
3577 {
3578         bi->page = NULL;
3579         skb->len += length;
3580         skb->data_len += length;
3581         skb->truesize += length;
3582 }
3583
3584 /**
3585  * e1000_receive_skb - helper function to handle rx indications
3586  * @adapter: board private structure
3587  * @status: descriptor status field as written by hardware
3588  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3589  * @skb: pointer to sk_buff to be indicated to stack
3590  */
3591 static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
3592                               __le16 vlan, struct sk_buff *skb)
3593 {
3594         if (unlikely(adapter->vlgrp && (status & E1000_RXD_STAT_VP))) {
3595                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3596                                          le16_to_cpu(vlan) &
3597                                          E1000_RXD_SPC_VLAN_MASK);
3598         } else {
3599                 netif_receive_skb(skb);
3600         }
3601 }
3602
3603 /**
3604  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
3605  * @adapter: board private structure
3606  * @rx_ring: ring to clean
3607  * @work_done: amount of napi work completed this call
3608  * @work_to_do: max amount of work allowed for this call to do
3609  *
3610  * the return value indicates whether actual cleaning was done, there
3611  * is no guarantee that everything was cleaned
3612  */
3613 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
3614                                      struct e1000_rx_ring *rx_ring,
3615                                      int *work_done, int work_to_do)
3616 {
3617         struct e1000_hw *hw = &adapter->hw;
3618         struct net_device *netdev = adapter->netdev;
3619         struct pci_dev *pdev = adapter->pdev;
3620         struct e1000_rx_desc *rx_desc, *next_rxd;
3621         struct e1000_buffer *buffer_info, *next_buffer;
3622         unsigned long irq_flags;
3623         u32 length;
3624         unsigned int i;
3625         int cleaned_count = 0;
3626         bool cleaned = false;
3627         unsigned int total_rx_bytes=0, total_rx_packets=0;
3628
3629         i = rx_ring->next_to_clean;
3630         rx_desc = E1000_RX_DESC(*rx_ring, i);
3631         buffer_info = &rx_ring->buffer_info[i];
3632
3633         while (rx_desc->status & E1000_RXD_STAT_DD) {
3634                 struct sk_buff *skb;
3635                 u8 status;
3636
3637                 if (*work_done >= work_to_do)
3638                         break;
3639                 (*work_done)++;
3640
3641                 status = rx_desc->status;
3642                 skb = buffer_info->skb;
3643                 buffer_info->skb = NULL;
3644
3645                 if (++i == rx_ring->count) i = 0;
3646                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3647                 prefetch(next_rxd);
3648
3649                 next_buffer = &rx_ring->buffer_info[i];
3650
3651                 cleaned = true;
3652                 cleaned_count++;
3653                 dma_unmap_page(&pdev->dev, buffer_info->dma,
3654                                buffer_info->length, DMA_FROM_DEVICE);
3655                 buffer_info->dma = 0;
3656
3657                 length = le16_to_cpu(rx_desc->length);
3658
3659                 /* errors is only valid for DD + EOP descriptors */
3660                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
3661                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
3662                         u8 last_byte = *(skb->data + length - 1);
3663                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3664                                        last_byte)) {
3665                                 spin_lock_irqsave(&adapter->stats_lock,
3666                                                   irq_flags);
3667                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3668                                                        length, skb->data);
3669                                 spin_unlock_irqrestore(&adapter->stats_lock,
3670                                                        irq_flags);
3671                                 length--;
3672                         } else {
3673                                 /* recycle both page and skb */
3674                                 buffer_info->skb = skb;
3675                                 /* an error means any chain goes out the window
3676                                  * too */
3677                                 if (rx_ring->rx_skb_top)
3678                                         dev_kfree_skb(rx_ring->rx_skb_top);
3679                                 rx_ring->rx_skb_top = NULL;
3680                                 goto next_desc;
3681                         }
3682                 }
3683
3684 #define rxtop rx_ring->rx_skb_top
3685                 if (!(status & E1000_RXD_STAT_EOP)) {
3686                         /* this descriptor is only the beginning (or middle) */
3687                         if (!rxtop) {
3688                                 /* this is the beginning of a chain */
3689                                 rxtop = skb;
3690                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
3691                                                    0, length);
3692                         } else {
3693                                 /* this is the middle of a chain */
3694                                 skb_fill_page_desc(rxtop,
3695                                     skb_shinfo(rxtop)->nr_frags,
3696                                     buffer_info->page, 0, length);
3697                                 /* re-use the skb, only consumed the page */
3698                                 buffer_info->skb = skb;
3699                         }
3700                         e1000_consume_page(buffer_info, rxtop, length);
3701                         goto next_desc;
3702                 } else {
3703                         if (rxtop) {
3704                                 /* end of the chain */
3705                                 skb_fill_page_desc(rxtop,
3706                                     skb_shinfo(rxtop)->nr_frags,
3707                                     buffer_info->page, 0, length);
3708                                 /* re-use the current skb, we only consumed the
3709                                  * page */
3710                                 buffer_info->skb = skb;
3711                                 skb = rxtop;
3712                                 rxtop = NULL;
3713                                 e1000_consume_page(buffer_info, skb, length);
3714                         } else {
3715                                 /* no chain, got EOP, this buf is the packet
3716                                  * copybreak to save the put_page/alloc_page */
3717                                 if (length <= copybreak &&
3718                                     skb_tailroom(skb) >= length) {
3719                                         u8 *vaddr;
3720                                         vaddr = kmap_atomic(buffer_info->page,
3721                                                             KM_SKB_DATA_SOFTIRQ);
3722                                         memcpy(skb_tail_pointer(skb), vaddr, length);
3723                                         kunmap_atomic(vaddr,
3724                                                       KM_SKB_DATA_SOFTIRQ);
3725                                         /* re-use the page, so don't erase
3726                                          * buffer_info->page */
3727                                         skb_put(skb, length);
3728                                 } else {
3729                                         skb_fill_page_desc(skb, 0,
3730                                                            buffer_info->page, 0,
3731                                                            length);
3732                                         e1000_consume_page(buffer_info, skb,
3733                                                            length);
3734                                 }
3735                         }
3736                 }
3737
3738                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
3739                 e1000_rx_checksum(adapter,
3740                                   (u32)(status) |
3741                                   ((u32)(rx_desc->errors) << 24),
3742                                   le16_to_cpu(rx_desc->csum), skb);
3743
3744                 pskb_trim(skb, skb->len - 4);
3745
3746                 /* probably a little skewed due to removing CRC */
3747                 total_rx_bytes += skb->len;
3748                 total_rx_packets++;
3749
3750                 /* eth type trans needs skb->data to point to something */
3751                 if (!pskb_may_pull(skb, ETH_HLEN)) {
3752                         e_err("pskb_may_pull failed.\n");
3753                         dev_kfree_skb(skb);
3754                         goto next_desc;
3755                 }
3756
3757                 skb->protocol = eth_type_trans(skb, netdev);
3758
3759                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3760
3761 next_desc:
3762                 rx_desc->status = 0;
3763
3764                 /* return some buffers to hardware, one at a time is too slow */
3765                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3766                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3767                         cleaned_count = 0;
3768                 }
3769
3770                 /* use prefetched values */
3771                 rx_desc = next_rxd;
3772                 buffer_info = next_buffer;
3773         }
3774         rx_ring->next_to_clean = i;
3775
3776         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3777         if (cleaned_count)
3778                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3779
3780         adapter->total_rx_packets += total_rx_packets;
3781         adapter->total_rx_bytes += total_rx_bytes;
3782         netdev->stats.rx_bytes += total_rx_bytes;
3783         netdev->stats.rx_packets += total_rx_packets;
3784         return cleaned;
3785 }
3786
3787 /*
3788  * this should improve performance for small packets with large amounts
3789  * of reassembly being done in the stack
3790  */
3791 static void e1000_check_copybreak(struct net_device *netdev,
3792                                  struct e1000_buffer *buffer_info,
3793                                  u32 length, struct sk_buff **skb)
3794 {
3795         struct sk_buff *new_skb;
3796
3797         if (length > copybreak)
3798                 return;
3799
3800         new_skb = netdev_alloc_skb_ip_align(netdev, length);
3801         if (!new_skb)
3802                 return;
3803
3804         skb_copy_to_linear_data_offset(new_skb, -NET_IP_ALIGN,
3805                                        (*skb)->data - NET_IP_ALIGN,
3806                                        length + NET_IP_ALIGN);
3807         /* save the skb in buffer_info as good */
3808         buffer_info->skb = *skb;
3809         *skb = new_skb;
3810 }
3811
3812 /**
3813  * e1000_clean_rx_irq - Send received data up the network stack; legacy
3814  * @adapter: board private structure
3815  * @rx_ring: ring to clean
3816  * @work_done: amount of napi work completed this call
3817  * @work_to_do: max amount of work allowed for this call to do
3818  */
3819 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
3820                                struct e1000_rx_ring *rx_ring,
3821                                int *work_done, int work_to_do)
3822 {
3823         struct e1000_hw *hw = &adapter->hw;
3824         struct net_device *netdev = adapter->netdev;
3825         struct pci_dev *pdev = adapter->pdev;
3826         struct e1000_rx_desc *rx_desc, *next_rxd;
3827         struct e1000_buffer *buffer_info, *next_buffer;
3828         unsigned long flags;
3829         u32 length;
3830         unsigned int i;
3831         int cleaned_count = 0;
3832         bool cleaned = false;
3833         unsigned int total_rx_bytes=0, total_rx_packets=0;
3834
3835         i = rx_ring->next_to_clean;
3836         rx_desc = E1000_RX_DESC(*rx_ring, i);
3837         buffer_info = &rx_ring->buffer_info[i];
3838
3839         while (rx_desc->status & E1000_RXD_STAT_DD) {
3840                 struct sk_buff *skb;
3841                 u8 status;
3842
3843                 if (*work_done >= work_to_do)
3844                         break;
3845                 (*work_done)++;
3846
3847                 status = rx_desc->status;
3848                 skb = buffer_info->skb;
3849                 buffer_info->skb = NULL;
3850
3851                 prefetch(skb->data - NET_IP_ALIGN);
3852
3853                 if (++i == rx_ring->count) i = 0;
3854                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3855                 prefetch(next_rxd);
3856
3857                 next_buffer = &rx_ring->buffer_info[i];
3858
3859                 cleaned = true;
3860                 cleaned_count++;
3861                 dma_unmap_single(&pdev->dev, buffer_info->dma,
3862                                  buffer_info->length, DMA_FROM_DEVICE);
3863                 buffer_info->dma = 0;
3864
3865                 length = le16_to_cpu(rx_desc->length);
3866                 /* !EOP means multiple descriptors were used to store a single
3867                  * packet, if thats the case we need to toss it.  In fact, we
3868                  * to toss every packet with the EOP bit clear and the next
3869                  * frame that _does_ have the EOP bit set, as it is by
3870                  * definition only a frame fragment
3871                  */
3872                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
3873                         adapter->discarding = true;
3874
3875                 if (adapter->discarding) {
3876                         /* All receives must fit into a single buffer */
3877                         e_info("Receive packet consumed multiple buffers\n");
3878                         /* recycle */
3879                         buffer_info->skb = skb;
3880                         if (status & E1000_RXD_STAT_EOP)
3881                                 adapter->discarding = false;
3882                         goto next_desc;
3883                 }
3884
3885                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
3886                         u8 last_byte = *(skb->data + length - 1);
3887                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3888                                        last_byte)) {
3889                                 spin_lock_irqsave(&adapter->stats_lock, flags);
3890                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3891                                                        length, skb->data);
3892                                 spin_unlock_irqrestore(&adapter->stats_lock,
3893                                                        flags);
3894                                 length--;
3895                         } else {
3896                                 /* recycle */
3897                                 buffer_info->skb = skb;
3898                                 goto next_desc;
3899                         }
3900                 }
3901
3902                 /* adjust length to remove Ethernet CRC, this must be
3903                  * done after the TBI_ACCEPT workaround above */
3904                 length -= 4;
3905
3906                 /* probably a little skewed due to removing CRC */
3907                 total_rx_bytes += length;
3908                 total_rx_packets++;
3909
3910                 e1000_check_copybreak(netdev, buffer_info, length, &skb);
3911
3912                 skb_put(skb, length);
3913
3914                 /* Receive Checksum Offload */
3915                 e1000_rx_checksum(adapter,
3916                                   (u32)(status) |
3917                                   ((u32)(rx_desc->errors) << 24),
3918                                   le16_to_cpu(rx_desc->csum), skb);
3919
3920                 skb->protocol = eth_type_trans(skb, netdev);
3921
3922                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3923
3924 next_desc:
3925                 rx_desc->status = 0;
3926
3927                 /* return some buffers to hardware, one at a time is too slow */
3928                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3929                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3930                         cleaned_count = 0;
3931                 }
3932
3933                 /* use prefetched values */
3934                 rx_desc = next_rxd;
3935                 buffer_info = next_buffer;
3936         }
3937         rx_ring->next_to_clean = i;
3938
3939         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3940         if (cleaned_count)
3941                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3942
3943         adapter->total_rx_packets += total_rx_packets;
3944         adapter->total_rx_bytes += total_rx_bytes;
3945         netdev->stats.rx_bytes += total_rx_bytes;
3946         netdev->stats.rx_packets += total_rx_packets;
3947         return cleaned;
3948 }
3949
3950 /**
3951  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
3952  * @adapter: address of board private structure
3953  * @rx_ring: pointer to receive ring structure
3954  * @cleaned_count: number of buffers to allocate this pass
3955  **/
3956
3957 static void
3958 e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
3959                              struct e1000_rx_ring *rx_ring, int cleaned_count)
3960 {
3961         struct net_device *netdev = adapter->netdev;
3962         struct pci_dev *pdev = adapter->pdev;
3963         struct e1000_rx_desc *rx_desc;
3964         struct e1000_buffer *buffer_info;
3965         struct sk_buff *skb;
3966         unsigned int i;
3967         unsigned int bufsz = 256 - 16 /*for skb_reserve */ ;
3968
3969         i = rx_ring->next_to_use;
3970         buffer_info = &rx_ring->buffer_info[i];
3971
3972         while (cleaned_count--) {
3973                 skb = buffer_info->skb;
3974                 if (skb) {
3975                         skb_trim(skb, 0);
3976                         goto check_page;
3977                 }
3978
3979                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
3980                 if (unlikely(!skb)) {
3981                         /* Better luck next round */
3982                         adapter->alloc_rx_buff_failed++;
3983                         break;
3984                 }
3985
3986                 /* Fix for errata 23, can't cross 64kB boundary */
3987                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3988                         struct sk_buff *oldskb = skb;
3989                         e_err("skb align check failed: %u bytes at %p\n",
3990                               bufsz, skb->data);
3991                         /* Try again, without freeing the previous */
3992                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
3993                         /* Failed allocation, critical failure */
3994                         if (!skb) {
3995                                 dev_kfree_skb(oldskb);
3996                                 adapter->alloc_rx_buff_failed++;
3997                                 break;
3998                         }
3999
4000                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4001                                 /* give up */
4002                                 dev_kfree_skb(skb);
4003                                 dev_kfree_skb(oldskb);
4004                                 break; /* while (cleaned_count--) */
4005                         }
4006
4007                         /* Use new allocation */
4008                         dev_kfree_skb(oldskb);
4009                 }
4010                 buffer_info->skb = skb;
4011                 buffer_info->length = adapter->rx_buffer_len;
4012 check_page:
4013                 /* allocate a new page if necessary */
4014                 if (!buffer_info->page) {
4015                         buffer_info->page = alloc_page(GFP_ATOMIC);
4016                         if (unlikely(!buffer_info->page)) {
4017                                 adapter->alloc_rx_buff_failed++;
4018                                 break;
4019                         }
4020                 }
4021
4022                 if (!buffer_info->dma) {
4023                         buffer_info->dma = dma_map_page(&pdev->dev,
4024                                                         buffer_info->page, 0,
4025                                                         buffer_info->length,
4026                                                         DMA_FROM_DEVICE);
4027                         if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4028                                 put_page(buffer_info->page);
4029                                 dev_kfree_skb(skb);
4030                                 buffer_info->page = NULL;
4031                                 buffer_info->skb = NULL;
4032                                 buffer_info->dma = 0;
4033                                 adapter->alloc_rx_buff_failed++;
4034                                 break; /* while !buffer_info->skb */
4035                         }
4036                 }
4037
4038                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4039                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4040
4041                 if (unlikely(++i == rx_ring->count))
4042                         i = 0;
4043                 buffer_info = &rx_ring->buffer_info[i];
4044         }
4045
4046         if (likely(rx_ring->next_to_use != i)) {
4047                 rx_ring->next_to_use = i;
4048                 if (unlikely(i-- == 0))
4049                         i = (rx_ring->count - 1);
4050
4051                 /* Force memory writes to complete before letting h/w
4052                  * know there are new descriptors to fetch.  (Only
4053                  * applicable for weak-ordered memory model archs,
4054                  * such as IA-64). */
4055                 wmb();
4056                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4057         }
4058 }
4059
4060 /**
4061  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4062  * @adapter: address of board private structure
4063  **/
4064
4065 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4066                                    struct e1000_rx_ring *rx_ring,
4067                                    int cleaned_count)
4068 {
4069         struct e1000_hw *hw = &adapter->hw;
4070         struct net_device *netdev = adapter->netdev;
4071         struct pci_dev *pdev = adapter->pdev;
4072         struct e1000_rx_desc *rx_desc;
4073         struct e1000_buffer *buffer_info;
4074         struct sk_buff *skb;
4075         unsigned int i;
4076         unsigned int bufsz = adapter->rx_buffer_len;
4077
4078         i = rx_ring->next_to_use;
4079         buffer_info = &rx_ring->buffer_info[i];
4080
4081         while (cleaned_count--) {
4082                 skb = buffer_info->skb;
4083                 if (skb) {
4084                         skb_trim(skb, 0);
4085                         goto map_skb;
4086                 }
4087
4088                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4089                 if (unlikely(!skb)) {
4090                         /* Better luck next round */
4091                         adapter->alloc_rx_buff_failed++;
4092                         break;
4093                 }
4094
4095                 /* Fix for errata 23, can't cross 64kB boundary */
4096                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4097                         struct sk_buff *oldskb = skb;
4098                         e_err("skb align check failed: %u bytes at %p\n",
4099                               bufsz, skb->data);
4100                         /* Try again, without freeing the previous */
4101                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4102                         /* Failed allocation, critical failure */
4103                         if (!skb) {
4104                                 dev_kfree_skb(oldskb);
4105                                 adapter->alloc_rx_buff_failed++;
4106                                 break;
4107                         }
4108
4109                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4110                                 /* give up */
4111                                 dev_kfree_skb(skb);
4112                                 dev_kfree_skb(oldskb);
4113                                 adapter->alloc_rx_buff_failed++;
4114                                 break; /* while !buffer_info->skb */
4115                         }
4116
4117                         /* Use new allocation */
4118                         dev_kfree_skb(oldskb);
4119                 }
4120                 buffer_info->skb = skb;
4121                 buffer_info->length = adapter->rx_buffer_len;
4122 map_skb:
4123                 buffer_info->dma = dma_map_single(&pdev->dev,
4124                                                   skb->data,
4125                                                   buffer_info->length,
4126                                                   DMA_FROM_DEVICE);
4127                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4128                         dev_kfree_skb(skb);
4129                         buffer_info->skb = NULL;
4130                         buffer_info->dma = 0;
4131                         adapter->alloc_rx_buff_failed++;
4132                         break; /* while !buffer_info->skb */
4133                 }
4134
4135                 /*
4136                  * XXX if it was allocated cleanly it will never map to a
4137                  * boundary crossing
4138                  */
4139
4140                 /* Fix for errata 23, can't cross 64kB boundary */
4141                 if (!e1000_check_64k_bound(adapter,
4142                                         (void *)(unsigned long)buffer_info->dma,
4143                                         adapter->rx_buffer_len)) {
4144                         e_err("dma align check failed: %u bytes at %p\n",
4145                               adapter->rx_buffer_len,
4146                               (void *)(unsigned long)buffer_info->dma);
4147                         dev_kfree_skb(skb);
4148                         buffer_info->skb = NULL;
4149
4150                         dma_unmap_single(&pdev->dev, buffer_info->dma,
4151                                          adapter->rx_buffer_len,
4152                                          DMA_FROM_DEVICE);
4153                         buffer_info->dma = 0;
4154
4155                         adapter->alloc_rx_buff_failed++;
4156                         break; /* while !buffer_info->skb */
4157                 }
4158                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4159                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4160
4161                 if (unlikely(++i == rx_ring->count))
4162                         i = 0;
4163                 buffer_info = &rx_ring->buffer_info[i];
4164         }
4165
4166         if (likely(rx_ring->next_to_use != i)) {
4167                 rx_ring->next_to_use = i;
4168                 if (unlikely(i-- == 0))
4169                         i = (rx_ring->count - 1);
4170
4171                 /* Force memory writes to complete before letting h/w
4172                  * know there are new descriptors to fetch.  (Only
4173                  * applicable for weak-ordered memory model archs,
4174                  * such as IA-64). */
4175                 wmb();
4176                 writel(i, hw->hw_addr + rx_ring->rdt);
4177         }
4178 }
4179
4180 /**
4181  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4182  * @adapter:
4183  **/
4184
4185 static void e1000_smartspeed(struct e1000_adapter *adapter)
4186 {
4187         struct e1000_hw *hw = &adapter->hw;
4188         u16 phy_status;
4189         u16 phy_ctrl;
4190
4191         if ((hw->phy_type != e1000_phy_igp) || !hw->autoneg ||
4192            !(hw->autoneg_advertised & ADVERTISE_1000_FULL))
4193                 return;
4194
4195         if (adapter->smartspeed == 0) {
4196                 /* If Master/Slave config fault is asserted twice,
4197                  * we assume back-to-back */
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_STATUS, &phy_status);
4201                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4202                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4203                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4204                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4205                         e1000_write_phy_reg(hw, PHY_1000T_CTRL,
4206                                             phy_ctrl);
4207                         adapter->smartspeed++;
4208                         if (!e1000_phy_setup_autoneg(hw) &&
4209                            !e1000_read_phy_reg(hw, PHY_CTRL,
4210                                                &phy_ctrl)) {
4211                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4212                                              MII_CR_RESTART_AUTO_NEG);
4213                                 e1000_write_phy_reg(hw, PHY_CTRL,
4214                                                     phy_ctrl);
4215                         }
4216                 }
4217                 return;
4218         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4219                 /* If still no link, perhaps using 2/3 pair cable */
4220                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4221                 phy_ctrl |= CR_1000T_MS_ENABLE;
4222                 e1000_write_phy_reg(hw, PHY_1000T_CTRL, phy_ctrl);
4223                 if (!e1000_phy_setup_autoneg(hw) &&
4224                    !e1000_read_phy_reg(hw, PHY_CTRL, &phy_ctrl)) {
4225                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4226                                      MII_CR_RESTART_AUTO_NEG);
4227                         e1000_write_phy_reg(hw, PHY_CTRL, phy_ctrl);
4228                 }
4229         }
4230         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4231         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4232                 adapter->smartspeed = 0;
4233 }
4234
4235 /**
4236  * e1000_ioctl -
4237  * @netdev:
4238  * @ifreq:
4239  * @cmd:
4240  **/
4241
4242 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4243 {
4244         switch (cmd) {
4245         case SIOCGMIIPHY:
4246         case SIOCGMIIREG:
4247         case SIOCSMIIREG:
4248                 return e1000_mii_ioctl(netdev, ifr, cmd);
4249         default:
4250                 return -EOPNOTSUPP;
4251         }
4252 }
4253
4254 /**
4255  * e1000_mii_ioctl -
4256  * @netdev:
4257  * @ifreq:
4258  * @cmd:
4259  **/
4260
4261 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4262                            int cmd)
4263 {
4264         struct e1000_adapter *adapter = netdev_priv(netdev);
4265         struct e1000_hw *hw = &adapter->hw;
4266         struct mii_ioctl_data *data = if_mii(ifr);
4267         int retval;
4268         u16 mii_reg;
4269         u16 spddplx;
4270         unsigned long flags;
4271
4272         if (hw->media_type != e1000_media_type_copper)
4273                 return -EOPNOTSUPP;
4274
4275         switch (cmd) {
4276         case SIOCGMIIPHY:
4277                 data->phy_id = hw->phy_addr;
4278                 break;
4279         case SIOCGMIIREG:
4280                 spin_lock_irqsave(&adapter->stats_lock, flags);
4281                 if (e1000_read_phy_reg(hw, data->reg_num & 0x1F,
4282                                    &data->val_out)) {
4283                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4284                         return -EIO;
4285                 }
4286                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4287                 break;
4288         case SIOCSMIIREG:
4289                 if (data->reg_num & ~(0x1F))
4290                         return -EFAULT;
4291                 mii_reg = data->val_in;
4292                 spin_lock_irqsave(&adapter->stats_lock, flags);
4293                 if (e1000_write_phy_reg(hw, data->reg_num,
4294                                         mii_reg)) {
4295                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4296                         return -EIO;
4297                 }
4298                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4299                 if (hw->media_type == e1000_media_type_copper) {
4300                         switch (data->reg_num) {
4301                         case PHY_CTRL:
4302                                 if (mii_reg & MII_CR_POWER_DOWN)
4303                                         break;
4304                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4305                                         hw->autoneg = 1;
4306                                         hw->autoneg_advertised = 0x2F;
4307                                 } else {
4308                                         if (mii_reg & 0x40)
4309                                                 spddplx = SPEED_1000;
4310                                         else if (mii_reg & 0x2000)
4311                                                 spddplx = SPEED_100;
4312                                         else
4313                                                 spddplx = SPEED_10;
4314                                         spddplx += (mii_reg & 0x100)
4315                                                    ? DUPLEX_FULL :
4316                                                    DUPLEX_HALF;
4317                                         retval = e1000_set_spd_dplx(adapter,
4318                                                                     spddplx);
4319                                         if (retval)
4320                                                 return retval;
4321                                 }
4322                                 if (netif_running(adapter->netdev))
4323                                         e1000_reinit_locked(adapter);
4324                                 else
4325                                         e1000_reset(adapter);
4326                                 break;
4327                         case M88E1000_PHY_SPEC_CTRL:
4328                         case M88E1000_EXT_PHY_SPEC_CTRL:
4329                                 if (e1000_phy_reset(hw))
4330                                         return -EIO;
4331                                 break;
4332                         }
4333                 } else {
4334                         switch (data->reg_num) {
4335                         case PHY_CTRL:
4336                                 if (mii_reg & MII_CR_POWER_DOWN)
4337                                         break;
4338                                 if (netif_running(adapter->netdev))
4339                                         e1000_reinit_locked(adapter);
4340                                 else
4341                                         e1000_reset(adapter);
4342                                 break;
4343                         }
4344                 }
4345                 break;
4346         default:
4347                 return -EOPNOTSUPP;
4348         }
4349         return E1000_SUCCESS;
4350 }
4351
4352 void e1000_pci_set_mwi(struct e1000_hw *hw)
4353 {
4354         struct e1000_adapter *adapter = hw->back;
4355         int ret_val = pci_set_mwi(adapter->pdev);
4356
4357         if (ret_val)
4358                 e_err("Error in setting MWI\n");
4359 }
4360
4361 void e1000_pci_clear_mwi(struct e1000_hw *hw)
4362 {
4363         struct e1000_adapter *adapter = hw->back;
4364
4365         pci_clear_mwi(adapter->pdev);
4366 }
4367
4368 int e1000_pcix_get_mmrbc(struct e1000_hw *hw)
4369 {
4370         struct e1000_adapter *adapter = hw->back;
4371         return pcix_get_mmrbc(adapter->pdev);
4372 }
4373
4374 void e1000_pcix_set_mmrbc(struct e1000_hw *hw, int mmrbc)
4375 {
4376         struct e1000_adapter *adapter = hw->back;
4377         pcix_set_mmrbc(adapter->pdev, mmrbc);
4378 }
4379
4380 void e1000_io_write(struct e1000_hw *hw, unsigned long port, u32 value)
4381 {
4382         outl(value, port);
4383 }
4384
4385 static void e1000_vlan_rx_register(struct net_device *netdev,
4386                                    struct vlan_group *grp)
4387 {
4388         struct e1000_adapter *adapter = netdev_priv(netdev);
4389         struct e1000_hw *hw = &adapter->hw;
4390         u32 ctrl, rctl;
4391
4392         if (!test_bit(__E1000_DOWN, &adapter->flags))
4393                 e1000_irq_disable(adapter);
4394         adapter->vlgrp = grp;
4395
4396         if (grp) {
4397                 /* enable VLAN tag insert/strip */
4398                 ctrl = er32(CTRL);
4399                 ctrl |= E1000_CTRL_VME;
4400                 ew32(CTRL, ctrl);
4401
4402                 /* enable VLAN receive filtering */
4403                 rctl = er32(RCTL);
4404                 rctl &= ~E1000_RCTL_CFIEN;
4405                 if (!(netdev->flags & IFF_PROMISC))
4406                         rctl |= E1000_RCTL_VFE;
4407                 ew32(RCTL, rctl);
4408                 e1000_update_mng_vlan(adapter);
4409         } else {
4410                 /* disable VLAN tag insert/strip */
4411                 ctrl = er32(CTRL);
4412                 ctrl &= ~E1000_CTRL_VME;
4413                 ew32(CTRL, ctrl);
4414
4415                 /* disable VLAN receive filtering */
4416                 rctl = er32(RCTL);
4417                 rctl &= ~E1000_RCTL_VFE;
4418                 ew32(RCTL, rctl);
4419
4420                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
4421                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4422                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4423                 }
4424         }
4425
4426         if (!test_bit(__E1000_DOWN, &adapter->flags))
4427                 e1000_irq_enable(adapter);
4428 }
4429
4430 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4431 {
4432         struct e1000_adapter *adapter = netdev_priv(netdev);
4433         struct e1000_hw *hw = &adapter->hw;
4434         u32 vfta, index;
4435
4436         if ((hw->mng_cookie.status &
4437              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4438             (vid == adapter->mng_vlan_id))
4439                 return;
4440         /* add VID to filter table */
4441         index = (vid >> 5) & 0x7F;
4442         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4443         vfta |= (1 << (vid & 0x1F));
4444         e1000_write_vfta(hw, index, vfta);
4445 }
4446
4447 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
4448 {
4449         struct e1000_adapter *adapter = netdev_priv(netdev);
4450         struct e1000_hw *hw = &adapter->hw;
4451         u32 vfta, index;
4452
4453         if (!test_bit(__E1000_DOWN, &adapter->flags))
4454                 e1000_irq_disable(adapter);
4455         vlan_group_set_device(adapter->vlgrp, vid, NULL);
4456         if (!test_bit(__E1000_DOWN, &adapter->flags))
4457                 e1000_irq_enable(adapter);
4458
4459         /* remove VID from filter table */
4460         index = (vid >> 5) & 0x7F;
4461         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4462         vfta &= ~(1 << (vid & 0x1F));
4463         e1000_write_vfta(hw, index, vfta);
4464 }
4465
4466 static void e1000_restore_vlan(struct e1000_adapter *adapter)
4467 {
4468         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
4469
4470         if (adapter->vlgrp) {
4471                 u16 vid;
4472                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
4473                         if (!vlan_group_get_device(adapter->vlgrp, vid))
4474                                 continue;
4475                         e1000_vlan_rx_add_vid(adapter->netdev, vid);
4476                 }
4477         }
4478 }
4479
4480 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx)
4481 {
4482         struct e1000_hw *hw = &adapter->hw;
4483
4484         hw->autoneg = 0;
4485
4486         /* Fiber NICs only allow 1000 gbps Full duplex */
4487         if ((hw->media_type == e1000_media_type_fiber) &&
4488                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
4489                 e_err("Unsupported Speed/Duplex configuration\n");
4490                 return -EINVAL;
4491         }
4492
4493         switch (spddplx) {
4494         case SPEED_10 + DUPLEX_HALF:
4495                 hw->forced_speed_duplex = e1000_10_half;
4496                 break;
4497         case SPEED_10 + DUPLEX_FULL:
4498                 hw->forced_speed_duplex = e1000_10_full;
4499                 break;
4500         case SPEED_100 + DUPLEX_HALF:
4501                 hw->forced_speed_duplex = e1000_100_half;
4502                 break;
4503         case SPEED_100 + DUPLEX_FULL:
4504                 hw->forced_speed_duplex = e1000_100_full;
4505                 break;
4506         case SPEED_1000 + DUPLEX_FULL:
4507                 hw->autoneg = 1;
4508                 hw->autoneg_advertised = ADVERTISE_1000_FULL;
4509                 break;
4510         case SPEED_1000 + DUPLEX_HALF: /* not supported */
4511         default:
4512                 e_err("Unsupported Speed/Duplex configuration\n");
4513                 return -EINVAL;
4514         }
4515         return 0;
4516 }
4517
4518 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4519 {
4520         struct net_device *netdev = pci_get_drvdata(pdev);
4521         struct e1000_adapter *adapter = netdev_priv(netdev);
4522         struct e1000_hw *hw = &adapter->hw;
4523         u32 ctrl, ctrl_ext, rctl, status;
4524         u32 wufc = adapter->wol;
4525 #ifdef CONFIG_PM
4526         int retval = 0;
4527 #endif
4528
4529         netif_device_detach(netdev);
4530
4531         if (netif_running(netdev)) {
4532                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
4533                 e1000_down(adapter);
4534         }
4535
4536 #ifdef CONFIG_PM
4537         retval = pci_save_state(pdev);
4538         if (retval)
4539                 return retval;
4540 #endif
4541
4542         status = er32(STATUS);
4543         if (status & E1000_STATUS_LU)
4544                 wufc &= ~E1000_WUFC_LNKC;
4545
4546         if (wufc) {
4547                 e1000_setup_rctl(adapter);
4548                 e1000_set_rx_mode(netdev);
4549
4550                 /* turn on all-multi mode if wake on multicast is enabled */
4551                 if (wufc & E1000_WUFC_MC) {
4552                         rctl = er32(RCTL);
4553                         rctl |= E1000_RCTL_MPE;
4554                         ew32(RCTL, rctl);
4555                 }
4556
4557                 if (hw->mac_type >= e1000_82540) {
4558                         ctrl = er32(CTRL);
4559                         /* advertise wake from D3Cold */
4560                         #define E1000_CTRL_ADVD3WUC 0x00100000
4561                         /* phy power management enable */
4562                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4563                         ctrl |= E1000_CTRL_ADVD3WUC |
4564                                 E1000_CTRL_EN_PHY_PWR_MGMT;
4565                         ew32(CTRL, ctrl);
4566                 }
4567
4568                 if (hw->media_type == e1000_media_type_fiber ||
4569                     hw->media_type == e1000_media_type_internal_serdes) {
4570                         /* keep the laser running in D3 */
4571                         ctrl_ext = er32(CTRL_EXT);
4572                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4573                         ew32(CTRL_EXT, ctrl_ext);
4574                 }
4575
4576                 ew32(WUC, E1000_WUC_PME_EN);
4577                 ew32(WUFC, wufc);
4578         } else {
4579                 ew32(WUC, 0);
4580                 ew32(WUFC, 0);
4581         }
4582
4583         e1000_release_manageability(adapter);
4584
4585         *enable_wake = !!wufc;
4586
4587         /* make sure adapter isn't asleep if manageability is enabled */
4588         if (adapter->en_mng_pt)
4589                 *enable_wake = true;
4590
4591         if (netif_running(netdev))
4592                 e1000_free_irq(adapter);
4593
4594         pci_disable_device(pdev);
4595
4596         return 0;
4597 }
4598
4599 #ifdef CONFIG_PM
4600 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4601 {
4602         int retval;
4603         bool wake;
4604
4605         retval = __e1000_shutdown(pdev, &wake);
4606         if (retval)
4607                 return retval;
4608
4609         if (wake) {
4610                 pci_prepare_to_sleep(pdev);
4611         } else {
4612                 pci_wake_from_d3(pdev, false);
4613                 pci_set_power_state(pdev, PCI_D3hot);
4614         }
4615
4616         return 0;
4617 }
4618
4619 static int e1000_resume(struct pci_dev *pdev)
4620 {
4621         struct net_device *netdev = pci_get_drvdata(pdev);
4622         struct e1000_adapter *adapter = netdev_priv(netdev);
4623         struct e1000_hw *hw = &adapter->hw;
4624         u32 err;
4625
4626         pci_set_power_state(pdev, PCI_D0);
4627         pci_restore_state(pdev);
4628         pci_save_state(pdev);
4629
4630         if (adapter->need_ioport)
4631                 err = pci_enable_device(pdev);
4632         else
4633                 err = pci_enable_device_mem(pdev);
4634         if (err) {
4635                 pr_err("Cannot enable PCI device from suspend\n");
4636                 return err;
4637         }
4638         pci_set_master(pdev);
4639
4640         pci_enable_wake(pdev, PCI_D3hot, 0);
4641         pci_enable_wake(pdev, PCI_D3cold, 0);
4642
4643         if (netif_running(netdev)) {
4644                 err = e1000_request_irq(adapter);
4645                 if (err)
4646                         return err;
4647         }
4648
4649         e1000_power_up_phy(adapter);
4650         e1000_reset(adapter);
4651         ew32(WUS, ~0);
4652
4653         e1000_init_manageability(adapter);
4654
4655         if (netif_running(netdev))
4656                 e1000_up(adapter);
4657
4658         netif_device_attach(netdev);
4659
4660         return 0;
4661 }
4662 #endif
4663
4664 static void e1000_shutdown(struct pci_dev *pdev)
4665 {
4666         bool wake;
4667
4668         __e1000_shutdown(pdev, &wake);
4669
4670         if (system_state == SYSTEM_POWER_OFF) {
4671                 pci_wake_from_d3(pdev, wake);
4672                 pci_set_power_state(pdev, PCI_D3hot);
4673         }
4674 }
4675
4676 #ifdef CONFIG_NET_POLL_CONTROLLER
4677 /*
4678  * Polling 'interrupt' - used by things like netconsole to send skbs
4679  * without having to re-enable interrupts. It's not called while
4680  * the interrupt routine is executing.
4681  */
4682 static void e1000_netpoll(struct net_device *netdev)
4683 {
4684         struct e1000_adapter *adapter = netdev_priv(netdev);
4685
4686         disable_irq(adapter->pdev->irq);
4687         e1000_intr(adapter->pdev->irq, netdev);
4688         enable_irq(adapter->pdev->irq);
4689 }
4690 #endif
4691
4692 /**
4693  * e1000_io_error_detected - called when PCI error is detected
4694  * @pdev: Pointer to PCI device
4695  * @state: The current pci connection state
4696  *
4697  * This function is called after a PCI bus error affecting
4698  * this device has been detected.
4699  */
4700 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4701                                                 pci_channel_state_t state)
4702 {
4703         struct net_device *netdev = pci_get_drvdata(pdev);
4704         struct e1000_adapter *adapter = netdev_priv(netdev);
4705
4706         netif_device_detach(netdev);
4707
4708         if (state == pci_channel_io_perm_failure)
4709                 return PCI_ERS_RESULT_DISCONNECT;
4710
4711         if (netif_running(netdev))
4712                 e1000_down(adapter);
4713         pci_disable_device(pdev);
4714
4715         /* Request a slot slot reset. */
4716         return PCI_ERS_RESULT_NEED_RESET;
4717 }
4718
4719 /**
4720  * e1000_io_slot_reset - called after the pci bus has been reset.
4721  * @pdev: Pointer to PCI device
4722  *
4723  * Restart the card from scratch, as if from a cold-boot. Implementation
4724  * resembles the first-half of the e1000_resume routine.
4725  */
4726 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4727 {
4728         struct net_device *netdev = pci_get_drvdata(pdev);
4729         struct e1000_adapter *adapter = netdev_priv(netdev);
4730         struct e1000_hw *hw = &adapter->hw;
4731         int err;
4732
4733         if (adapter->need_ioport)
4734                 err = pci_enable_device(pdev);
4735         else
4736                 err = pci_enable_device_mem(pdev);
4737         if (err) {
4738                 pr_err("Cannot re-enable PCI device after reset.\n");
4739                 return PCI_ERS_RESULT_DISCONNECT;
4740         }
4741         pci_set_master(pdev);
4742
4743         pci_enable_wake(pdev, PCI_D3hot, 0);
4744         pci_enable_wake(pdev, PCI_D3cold, 0);
4745
4746         e1000_reset(adapter);
4747         ew32(WUS, ~0);
4748
4749         return PCI_ERS_RESULT_RECOVERED;
4750 }
4751
4752 /**
4753  * e1000_io_resume - called when traffic can start flowing again.
4754  * @pdev: Pointer to PCI device
4755  *
4756  * This callback is called when the error recovery driver tells us that
4757  * its OK to resume normal operation. Implementation resembles the
4758  * second-half of the e1000_resume routine.
4759  */
4760 static void e1000_io_resume(struct pci_dev *pdev)
4761 {
4762         struct net_device *netdev = pci_get_drvdata(pdev);
4763         struct e1000_adapter *adapter = netdev_priv(netdev);
4764
4765         e1000_init_manageability(adapter);
4766
4767         if (netif_running(netdev)) {
4768                 if (e1000_up(adapter)) {
4769                         pr_info("can't bring device back up after reset\n");
4770                         return;
4771                 }
4772         }
4773
4774         netif_device_attach(netdev);
4775 }
4776
4777 /* e1000_main.c */