e1000: update version to 7.3.20-k2
[linux-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 #ifndef CONFIG_E1000_NAPI
35 #define DRIVERNAPI
36 #else
37 #define DRIVERNAPI "-NAPI"
38 #endif
39 #define DRV_VERSION "7.3.20-k2"DRIVERNAPI
40 char e1000_driver_version[] = DRV_VERSION;
41 static char e1000_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
42
43 /* e1000_pci_tbl - PCI Device ID Table
44  *
45  * Last entry must be all 0s
46  *
47  * Macro expands to...
48  *   {PCI_DEVICE(PCI_VENDOR_ID_INTEL, device_id)}
49  */
50 static struct pci_device_id e1000_pci_tbl[] = {
51         INTEL_E1000_ETHERNET_DEVICE(0x1000),
52         INTEL_E1000_ETHERNET_DEVICE(0x1001),
53         INTEL_E1000_ETHERNET_DEVICE(0x1004),
54         INTEL_E1000_ETHERNET_DEVICE(0x1008),
55         INTEL_E1000_ETHERNET_DEVICE(0x1009),
56         INTEL_E1000_ETHERNET_DEVICE(0x100C),
57         INTEL_E1000_ETHERNET_DEVICE(0x100D),
58         INTEL_E1000_ETHERNET_DEVICE(0x100E),
59         INTEL_E1000_ETHERNET_DEVICE(0x100F),
60         INTEL_E1000_ETHERNET_DEVICE(0x1010),
61         INTEL_E1000_ETHERNET_DEVICE(0x1011),
62         INTEL_E1000_ETHERNET_DEVICE(0x1012),
63         INTEL_E1000_ETHERNET_DEVICE(0x1013),
64         INTEL_E1000_ETHERNET_DEVICE(0x1014),
65         INTEL_E1000_ETHERNET_DEVICE(0x1015),
66         INTEL_E1000_ETHERNET_DEVICE(0x1016),
67         INTEL_E1000_ETHERNET_DEVICE(0x1017),
68         INTEL_E1000_ETHERNET_DEVICE(0x1018),
69         INTEL_E1000_ETHERNET_DEVICE(0x1019),
70         INTEL_E1000_ETHERNET_DEVICE(0x101A),
71         INTEL_E1000_ETHERNET_DEVICE(0x101D),
72         INTEL_E1000_ETHERNET_DEVICE(0x101E),
73         INTEL_E1000_ETHERNET_DEVICE(0x1026),
74         INTEL_E1000_ETHERNET_DEVICE(0x1027),
75         INTEL_E1000_ETHERNET_DEVICE(0x1028),
76         INTEL_E1000_ETHERNET_DEVICE(0x1049),
77         INTEL_E1000_ETHERNET_DEVICE(0x104A),
78         INTEL_E1000_ETHERNET_DEVICE(0x104B),
79         INTEL_E1000_ETHERNET_DEVICE(0x104C),
80         INTEL_E1000_ETHERNET_DEVICE(0x104D),
81         INTEL_E1000_ETHERNET_DEVICE(0x105E),
82         INTEL_E1000_ETHERNET_DEVICE(0x105F),
83         INTEL_E1000_ETHERNET_DEVICE(0x1060),
84         INTEL_E1000_ETHERNET_DEVICE(0x1075),
85         INTEL_E1000_ETHERNET_DEVICE(0x1076),
86         INTEL_E1000_ETHERNET_DEVICE(0x1077),
87         INTEL_E1000_ETHERNET_DEVICE(0x1078),
88         INTEL_E1000_ETHERNET_DEVICE(0x1079),
89         INTEL_E1000_ETHERNET_DEVICE(0x107A),
90         INTEL_E1000_ETHERNET_DEVICE(0x107B),
91         INTEL_E1000_ETHERNET_DEVICE(0x107C),
92         INTEL_E1000_ETHERNET_DEVICE(0x107D),
93         INTEL_E1000_ETHERNET_DEVICE(0x107E),
94         INTEL_E1000_ETHERNET_DEVICE(0x107F),
95         INTEL_E1000_ETHERNET_DEVICE(0x108A),
96         INTEL_E1000_ETHERNET_DEVICE(0x108B),
97         INTEL_E1000_ETHERNET_DEVICE(0x108C),
98         INTEL_E1000_ETHERNET_DEVICE(0x1096),
99         INTEL_E1000_ETHERNET_DEVICE(0x1098),
100         INTEL_E1000_ETHERNET_DEVICE(0x1099),
101         INTEL_E1000_ETHERNET_DEVICE(0x109A),
102         INTEL_E1000_ETHERNET_DEVICE(0x10A4),
103         INTEL_E1000_ETHERNET_DEVICE(0x10B5),
104         INTEL_E1000_ETHERNET_DEVICE(0x10B9),
105         INTEL_E1000_ETHERNET_DEVICE(0x10BA),
106         INTEL_E1000_ETHERNET_DEVICE(0x10BB),
107         INTEL_E1000_ETHERNET_DEVICE(0x10BC),
108         INTEL_E1000_ETHERNET_DEVICE(0x10C4),
109         INTEL_E1000_ETHERNET_DEVICE(0x10C5),
110         /* required last entry */
111         {0,}
112 };
113
114 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
115
116 int e1000_up(struct e1000_adapter *adapter);
117 void e1000_down(struct e1000_adapter *adapter);
118 void e1000_reinit_locked(struct e1000_adapter *adapter);
119 void e1000_reset(struct e1000_adapter *adapter);
120 int e1000_set_spd_dplx(struct e1000_adapter *adapter, uint16_t spddplx);
121 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
122 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
123 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
124 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
125 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
126                              struct e1000_tx_ring *txdr);
127 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
128                              struct e1000_rx_ring *rxdr);
129 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
130                              struct e1000_tx_ring *tx_ring);
131 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
132                              struct e1000_rx_ring *rx_ring);
133 void e1000_update_stats(struct e1000_adapter *adapter);
134
135 static int e1000_init_module(void);
136 static void e1000_exit_module(void);
137 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
138 static void __devexit e1000_remove(struct pci_dev *pdev);
139 static int e1000_alloc_queues(struct e1000_adapter *adapter);
140 static int e1000_sw_init(struct e1000_adapter *adapter);
141 static int e1000_open(struct net_device *netdev);
142 static int e1000_close(struct net_device *netdev);
143 static void e1000_configure_tx(struct e1000_adapter *adapter);
144 static void e1000_configure_rx(struct e1000_adapter *adapter);
145 static void e1000_setup_rctl(struct e1000_adapter *adapter);
146 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
147 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
148 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
149                                 struct e1000_tx_ring *tx_ring);
150 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
151                                 struct e1000_rx_ring *rx_ring);
152 static void e1000_set_multi(struct net_device *netdev);
153 static void e1000_update_phy_info(unsigned long data);
154 static void e1000_watchdog(unsigned long data);
155 static void e1000_82547_tx_fifo_stall(unsigned long data);
156 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
157 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
158 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
159 static int e1000_set_mac(struct net_device *netdev, void *p);
160 static irqreturn_t e1000_intr(int irq, void *data);
161 #ifdef CONFIG_PCI_MSI
162 static irqreturn_t e1000_intr_msi(int irq, void *data);
163 #endif
164 static boolean_t e1000_clean_tx_irq(struct e1000_adapter *adapter,
165                                     struct e1000_tx_ring *tx_ring);
166 #ifdef CONFIG_E1000_NAPI
167 static int e1000_clean(struct net_device *poll_dev, int *budget);
168 static boolean_t e1000_clean_rx_irq(struct e1000_adapter *adapter,
169                                     struct e1000_rx_ring *rx_ring,
170                                     int *work_done, int work_to_do);
171 static boolean_t e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
172                                        struct e1000_rx_ring *rx_ring,
173                                        int *work_done, int work_to_do);
174 #else
175 static boolean_t e1000_clean_rx_irq(struct e1000_adapter *adapter,
176                                     struct e1000_rx_ring *rx_ring);
177 static boolean_t e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
178                                        struct e1000_rx_ring *rx_ring);
179 #endif
180 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
181                                    struct e1000_rx_ring *rx_ring,
182                                    int cleaned_count);
183 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
184                                       struct e1000_rx_ring *rx_ring,
185                                       int cleaned_count);
186 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
187 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
188                            int cmd);
189 void e1000_set_ethtool_ops(struct net_device *netdev);
190 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
191 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
192 static void e1000_tx_timeout(struct net_device *dev);
193 static void e1000_reset_task(struct work_struct *work);
194 static void e1000_smartspeed(struct e1000_adapter *adapter);
195 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
196                                        struct sk_buff *skb);
197
198 static void e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp);
199 static void e1000_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid);
200 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid);
201 static void e1000_restore_vlan(struct e1000_adapter *adapter);
202
203 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
204 #ifdef CONFIG_PM
205 static int e1000_resume(struct pci_dev *pdev);
206 #endif
207 static void e1000_shutdown(struct pci_dev *pdev);
208
209 #ifdef CONFIG_NET_POLL_CONTROLLER
210 /* for netdump / net console */
211 static void e1000_netpoll (struct net_device *netdev);
212 #endif
213
214 extern void e1000_check_options(struct e1000_adapter *adapter);
215
216 #define COPYBREAK_DEFAULT 256
217 static unsigned int copybreak __read_mostly = COPYBREAK_DEFAULT;
218 module_param(copybreak, uint, 0644);
219 MODULE_PARM_DESC(copybreak,
220         "Maximum size of packet that is copied to a new buffer on receive");
221
222 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
223                      pci_channel_state_t state);
224 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
225 static void e1000_io_resume(struct pci_dev *pdev);
226
227 static struct pci_error_handlers e1000_err_handler = {
228         .error_detected = e1000_io_error_detected,
229         .slot_reset = e1000_io_slot_reset,
230         .resume = e1000_io_resume,
231 };
232
233 static struct pci_driver e1000_driver = {
234         .name     = e1000_driver_name,
235         .id_table = e1000_pci_tbl,
236         .probe    = e1000_probe,
237         .remove   = __devexit_p(e1000_remove),
238 #ifdef CONFIG_PM
239         /* Power Managment Hooks */
240         .suspend  = e1000_suspend,
241         .resume   = e1000_resume,
242 #endif
243         .shutdown = e1000_shutdown,
244         .err_handler = &e1000_err_handler
245 };
246
247 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
248 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
249 MODULE_LICENSE("GPL");
250 MODULE_VERSION(DRV_VERSION);
251
252 static int debug = NETIF_MSG_DRV | NETIF_MSG_PROBE;
253 module_param(debug, int, 0);
254 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
255
256 /**
257  * e1000_init_module - Driver Registration Routine
258  *
259  * e1000_init_module is the first routine called when the driver is
260  * loaded. All it does is register with the PCI subsystem.
261  **/
262
263 static int __init
264 e1000_init_module(void)
265 {
266         int ret;
267         printk(KERN_INFO "%s - version %s\n",
268                e1000_driver_string, e1000_driver_version);
269
270         printk(KERN_INFO "%s\n", e1000_copyright);
271
272         ret = pci_register_driver(&e1000_driver);
273         if (copybreak != COPYBREAK_DEFAULT) {
274                 if (copybreak == 0)
275                         printk(KERN_INFO "e1000: copybreak disabled\n");
276                 else
277                         printk(KERN_INFO "e1000: copybreak enabled for "
278                                "packets <= %u bytes\n", copybreak);
279         }
280         return ret;
281 }
282
283 module_init(e1000_init_module);
284
285 /**
286  * e1000_exit_module - Driver Exit Cleanup Routine
287  *
288  * e1000_exit_module is called just before the driver is removed
289  * from memory.
290  **/
291
292 static void __exit
293 e1000_exit_module(void)
294 {
295         pci_unregister_driver(&e1000_driver);
296 }
297
298 module_exit(e1000_exit_module);
299
300 static int e1000_request_irq(struct e1000_adapter *adapter)
301 {
302         struct net_device *netdev = adapter->netdev;
303         int flags, err = 0;
304
305         flags = IRQF_SHARED;
306 #ifdef CONFIG_PCI_MSI
307         if (adapter->hw.mac_type >= e1000_82571) {
308                 adapter->have_msi = TRUE;
309                 if ((err = pci_enable_msi(adapter->pdev))) {
310                         DPRINTK(PROBE, ERR,
311                          "Unable to allocate MSI interrupt Error: %d\n", err);
312                         adapter->have_msi = FALSE;
313                 }
314         }
315         if (adapter->have_msi) {
316                 flags &= ~IRQF_SHARED;
317                 err = request_irq(adapter->pdev->irq, &e1000_intr_msi, flags,
318                                   netdev->name, netdev);
319                 if (err)
320                         DPRINTK(PROBE, ERR,
321                                "Unable to allocate interrupt Error: %d\n", err);
322         } else
323 #endif
324         if ((err = request_irq(adapter->pdev->irq, &e1000_intr, flags,
325                                netdev->name, netdev)))
326                 DPRINTK(PROBE, ERR,
327                         "Unable to allocate interrupt Error: %d\n", err);
328
329         return err;
330 }
331
332 static void e1000_free_irq(struct e1000_adapter *adapter)
333 {
334         struct net_device *netdev = adapter->netdev;
335
336         free_irq(adapter->pdev->irq, netdev);
337
338 #ifdef CONFIG_PCI_MSI
339         if (adapter->have_msi)
340                 pci_disable_msi(adapter->pdev);
341 #endif
342 }
343
344 /**
345  * e1000_irq_disable - Mask off interrupt generation on the NIC
346  * @adapter: board private structure
347  **/
348
349 static void
350 e1000_irq_disable(struct e1000_adapter *adapter)
351 {
352         atomic_inc(&adapter->irq_sem);
353         E1000_WRITE_REG(&adapter->hw, IMC, ~0);
354         E1000_WRITE_FLUSH(&adapter->hw);
355         synchronize_irq(adapter->pdev->irq);
356 }
357
358 /**
359  * e1000_irq_enable - Enable default interrupt generation settings
360  * @adapter: board private structure
361  **/
362
363 static void
364 e1000_irq_enable(struct e1000_adapter *adapter)
365 {
366         if (likely(atomic_dec_and_test(&adapter->irq_sem))) {
367                 E1000_WRITE_REG(&adapter->hw, IMS, IMS_ENABLE_MASK);
368                 E1000_WRITE_FLUSH(&adapter->hw);
369         }
370 }
371
372 static void
373 e1000_update_mng_vlan(struct e1000_adapter *adapter)
374 {
375         struct net_device *netdev = adapter->netdev;
376         uint16_t vid = adapter->hw.mng_cookie.vlan_id;
377         uint16_t old_vid = adapter->mng_vlan_id;
378         if (adapter->vlgrp) {
379                 if (!adapter->vlgrp->vlan_devices[vid]) {
380                         if (adapter->hw.mng_cookie.status &
381                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
382                                 e1000_vlan_rx_add_vid(netdev, vid);
383                                 adapter->mng_vlan_id = vid;
384                         } else
385                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
386
387                         if ((old_vid != (uint16_t)E1000_MNG_VLAN_NONE) &&
388                                         (vid != old_vid) &&
389                                         !adapter->vlgrp->vlan_devices[old_vid])
390                                 e1000_vlan_rx_kill_vid(netdev, old_vid);
391                 } else
392                         adapter->mng_vlan_id = vid;
393         }
394 }
395
396 /**
397  * e1000_release_hw_control - release control of the h/w to f/w
398  * @adapter: address of board private structure
399  *
400  * e1000_release_hw_control resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
401  * For ASF and Pass Through versions of f/w this means that the
402  * driver is no longer loaded. For AMT version (only with 82573) i
403  * of the f/w this means that the network i/f is closed.
404  *
405  **/
406
407 static void
408 e1000_release_hw_control(struct e1000_adapter *adapter)
409 {
410         uint32_t ctrl_ext;
411         uint32_t swsm;
412         uint32_t extcnf;
413
414         /* Let firmware taken over control of h/w */
415         switch (adapter->hw.mac_type) {
416         case e1000_82571:
417         case e1000_82572:
418         case e1000_80003es2lan:
419                 ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
420                 E1000_WRITE_REG(&adapter->hw, CTRL_EXT,
421                                 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
422                 break;
423         case e1000_82573:
424                 swsm = E1000_READ_REG(&adapter->hw, SWSM);
425                 E1000_WRITE_REG(&adapter->hw, SWSM,
426                                 swsm & ~E1000_SWSM_DRV_LOAD);
427         case e1000_ich8lan:
428                 extcnf = E1000_READ_REG(&adapter->hw, CTRL_EXT);
429                 E1000_WRITE_REG(&adapter->hw, CTRL_EXT,
430                                 extcnf & ~E1000_CTRL_EXT_DRV_LOAD);
431                 break;
432         default:
433                 break;
434         }
435 }
436
437 /**
438  * e1000_get_hw_control - get control of the h/w from f/w
439  * @adapter: address of board private structure
440  *
441  * e1000_get_hw_control sets {CTRL_EXT|FWSM}:DRV_LOAD bit.
442  * For ASF and Pass Through versions of f/w this means that
443  * the driver is loaded. For AMT version (only with 82573)
444  * of the f/w this means that the network i/f is open.
445  *
446  **/
447
448 static void
449 e1000_get_hw_control(struct e1000_adapter *adapter)
450 {
451         uint32_t ctrl_ext;
452         uint32_t swsm;
453         uint32_t extcnf;
454
455         /* Let firmware know the driver has taken over */
456         switch (adapter->hw.mac_type) {
457         case e1000_82571:
458         case e1000_82572:
459         case e1000_80003es2lan:
460                 ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
461                 E1000_WRITE_REG(&adapter->hw, CTRL_EXT,
462                                 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
463                 break;
464         case e1000_82573:
465                 swsm = E1000_READ_REG(&adapter->hw, SWSM);
466                 E1000_WRITE_REG(&adapter->hw, SWSM,
467                                 swsm | E1000_SWSM_DRV_LOAD);
468                 break;
469         case e1000_ich8lan:
470                 extcnf = E1000_READ_REG(&adapter->hw, EXTCNF_CTRL);
471                 E1000_WRITE_REG(&adapter->hw, EXTCNF_CTRL,
472                                 extcnf | E1000_EXTCNF_CTRL_SWFLAG);
473                 break;
474         default:
475                 break;
476         }
477 }
478
479 static void
480 e1000_init_manageability(struct e1000_adapter *adapter)
481 {
482         if (adapter->en_mng_pt) {
483                 uint32_t manc = E1000_READ_REG(&adapter->hw, MANC);
484
485                 /* disable hardware interception of ARP */
486                 manc &= ~(E1000_MANC_ARP_EN);
487
488                 /* enable receiving management packets to the host */
489                 /* this will probably generate destination unreachable messages
490                  * from the host OS, but the packets will be handled on SMBUS */
491                 if (adapter->hw.has_manc2h) {
492                         uint32_t manc2h = E1000_READ_REG(&adapter->hw, MANC2H);
493
494                         manc |= E1000_MANC_EN_MNG2HOST;
495 #define E1000_MNG2HOST_PORT_623 (1 << 5)
496 #define E1000_MNG2HOST_PORT_664 (1 << 6)
497                         manc2h |= E1000_MNG2HOST_PORT_623;
498                         manc2h |= E1000_MNG2HOST_PORT_664;
499                         E1000_WRITE_REG(&adapter->hw, MANC2H, manc2h);
500                 }
501
502                 E1000_WRITE_REG(&adapter->hw, MANC, manc);
503         }
504 }
505
506 static void
507 e1000_release_manageability(struct e1000_adapter *adapter)
508 {
509         if (adapter->en_mng_pt) {
510                 uint32_t manc = E1000_READ_REG(&adapter->hw, MANC);
511
512                 /* re-enable hardware interception of ARP */
513                 manc |= E1000_MANC_ARP_EN;
514
515                 if (adapter->hw.has_manc2h)
516                         manc &= ~E1000_MANC_EN_MNG2HOST;
517
518                 /* don't explicitly have to mess with MANC2H since
519                  * MANC has an enable disable that gates MANC2H */
520
521                 E1000_WRITE_REG(&adapter->hw, MANC, manc);
522         }
523 }
524
525 int
526 e1000_up(struct e1000_adapter *adapter)
527 {
528         struct net_device *netdev = adapter->netdev;
529         int i;
530
531         /* hardware has been reset, we need to reload some things */
532
533         e1000_set_multi(netdev);
534
535         e1000_restore_vlan(adapter);
536         e1000_init_manageability(adapter);
537
538         e1000_configure_tx(adapter);
539         e1000_setup_rctl(adapter);
540         e1000_configure_rx(adapter);
541         /* call E1000_DESC_UNUSED which always leaves
542          * at least 1 descriptor unused to make sure
543          * next_to_use != next_to_clean */
544         for (i = 0; i < adapter->num_rx_queues; i++) {
545                 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
546                 adapter->alloc_rx_buf(adapter, ring,
547                                       E1000_DESC_UNUSED(ring));
548         }
549
550         adapter->tx_queue_len = netdev->tx_queue_len;
551
552 #ifdef CONFIG_E1000_NAPI
553         netif_poll_enable(netdev);
554 #endif
555         e1000_irq_enable(adapter);
556
557         clear_bit(__E1000_DOWN, &adapter->flags);
558
559         /* fire a link change interrupt to start the watchdog */
560         E1000_WRITE_REG(&adapter->hw, ICS, E1000_ICS_LSC);
561         return 0;
562 }
563
564 /**
565  * e1000_power_up_phy - restore link in case the phy was powered down
566  * @adapter: address of board private structure
567  *
568  * The phy may be powered down to save power and turn off link when the
569  * driver is unloaded and wake on lan is not enabled (among others)
570  * *** this routine MUST be followed by a call to e1000_reset ***
571  *
572  **/
573
574 void e1000_power_up_phy(struct e1000_adapter *adapter)
575 {
576         uint16_t mii_reg = 0;
577
578         /* Just clear the power down bit to wake the phy back up */
579         if (adapter->hw.media_type == e1000_media_type_copper) {
580                 /* according to the manual, the phy will retain its
581                  * settings across a power-down/up cycle */
582                 e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &mii_reg);
583                 mii_reg &= ~MII_CR_POWER_DOWN;
584                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL, mii_reg);
585         }
586 }
587
588 static void e1000_power_down_phy(struct e1000_adapter *adapter)
589 {
590         /* Power down the PHY so no link is implied when interface is down *
591          * The PHY cannot be powered down if any of the following is TRUE *
592          * (a) WoL is enabled
593          * (b) AMT is active
594          * (c) SoL/IDER session is active */
595         if (!adapter->wol && adapter->hw.mac_type >= e1000_82540 &&
596            adapter->hw.media_type == e1000_media_type_copper) {
597                 uint16_t mii_reg = 0;
598
599                 switch (adapter->hw.mac_type) {
600                 case e1000_82540:
601                 case e1000_82545:
602                 case e1000_82545_rev_3:
603                 case e1000_82546:
604                 case e1000_82546_rev_3:
605                 case e1000_82541:
606                 case e1000_82541_rev_2:
607                 case e1000_82547:
608                 case e1000_82547_rev_2:
609                         if (E1000_READ_REG(&adapter->hw, MANC) &
610                             E1000_MANC_SMBUS_EN)
611                                 goto out;
612                         break;
613                 case e1000_82571:
614                 case e1000_82572:
615                 case e1000_82573:
616                 case e1000_80003es2lan:
617                 case e1000_ich8lan:
618                         if (e1000_check_mng_mode(&adapter->hw) ||
619                             e1000_check_phy_reset_block(&adapter->hw))
620                                 goto out;
621                         break;
622                 default:
623                         goto out;
624                 }
625                 e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &mii_reg);
626                 mii_reg |= MII_CR_POWER_DOWN;
627                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL, mii_reg);
628                 mdelay(1);
629         }
630 out:
631         return;
632 }
633
634 void
635 e1000_down(struct e1000_adapter *adapter)
636 {
637         struct net_device *netdev = adapter->netdev;
638
639         /* signal that we're down so the interrupt handler does not
640          * reschedule our watchdog timer */
641         set_bit(__E1000_DOWN, &adapter->flags);
642
643         e1000_irq_disable(adapter);
644
645         del_timer_sync(&adapter->tx_fifo_stall_timer);
646         del_timer_sync(&adapter->watchdog_timer);
647         del_timer_sync(&adapter->phy_info_timer);
648
649 #ifdef CONFIG_E1000_NAPI
650         netif_poll_disable(netdev);
651 #endif
652         netdev->tx_queue_len = adapter->tx_queue_len;
653         adapter->link_speed = 0;
654         adapter->link_duplex = 0;
655         netif_carrier_off(netdev);
656         netif_stop_queue(netdev);
657
658         e1000_reset(adapter);
659         e1000_clean_all_tx_rings(adapter);
660         e1000_clean_all_rx_rings(adapter);
661 }
662
663 void
664 e1000_reinit_locked(struct e1000_adapter *adapter)
665 {
666         WARN_ON(in_interrupt());
667         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
668                 msleep(1);
669         e1000_down(adapter);
670         e1000_up(adapter);
671         clear_bit(__E1000_RESETTING, &adapter->flags);
672 }
673
674 void
675 e1000_reset(struct e1000_adapter *adapter)
676 {
677         uint32_t pba = 0, tx_space, min_tx_space, min_rx_space;
678         uint16_t fc_high_water_mark = E1000_FC_HIGH_DIFF;
679         boolean_t legacy_pba_adjust = FALSE;
680
681         /* Repartition Pba for greater than 9k mtu
682          * To take effect CTRL.RST is required.
683          */
684
685         switch (adapter->hw.mac_type) {
686         case e1000_82542_rev2_0:
687         case e1000_82542_rev2_1:
688         case e1000_82543:
689         case e1000_82544:
690         case e1000_82540:
691         case e1000_82541:
692         case e1000_82541_rev_2:
693                 legacy_pba_adjust = TRUE;
694                 pba = E1000_PBA_48K;
695                 break;
696         case e1000_82545:
697         case e1000_82545_rev_3:
698         case e1000_82546:
699         case e1000_82546_rev_3:
700                 pba = E1000_PBA_48K;
701                 break;
702         case e1000_82547:
703         case e1000_82547_rev_2:
704                 legacy_pba_adjust = TRUE;
705                 pba = E1000_PBA_30K;
706                 break;
707         case e1000_82571:
708         case e1000_82572:
709         case e1000_80003es2lan:
710                 pba = E1000_PBA_38K;
711                 break;
712         case e1000_82573:
713                 pba = E1000_PBA_20K;
714                 break;
715         case e1000_ich8lan:
716                 pba = E1000_PBA_8K;
717         case e1000_undefined:
718         case e1000_num_macs:
719                 break;
720         }
721
722         if (legacy_pba_adjust == TRUE) {
723                 if (adapter->netdev->mtu > E1000_RXBUFFER_8192)
724                         pba -= 8; /* allocate more FIFO for Tx */
725
726                 if (adapter->hw.mac_type == e1000_82547) {
727                         adapter->tx_fifo_head = 0;
728                         adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
729                         adapter->tx_fifo_size =
730                                 (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
731                         atomic_set(&adapter->tx_fifo_stall, 0);
732                 }
733         } else if (adapter->hw.max_frame_size > MAXIMUM_ETHERNET_FRAME_SIZE) {
734                 /* adjust PBA for jumbo frames */
735                 E1000_WRITE_REG(&adapter->hw, PBA, pba);
736
737                 /* To maintain wire speed transmits, the Tx FIFO should be
738                  * large enough to accomodate two full transmit packets,
739                  * rounded up to the next 1KB and expressed in KB.  Likewise,
740                  * the Rx FIFO should be large enough to accomodate at least
741                  * one full receive packet and is similarly rounded up and
742                  * expressed in KB. */
743                 pba = E1000_READ_REG(&adapter->hw, PBA);
744                 /* upper 16 bits has Tx packet buffer allocation size in KB */
745                 tx_space = pba >> 16;
746                 /* lower 16 bits has Rx packet buffer allocation size in KB */
747                 pba &= 0xffff;
748                 /* don't include ethernet FCS because hardware appends/strips */
749                 min_rx_space = adapter->netdev->mtu + ENET_HEADER_SIZE +
750                                VLAN_TAG_SIZE;
751                 min_tx_space = min_rx_space;
752                 min_tx_space *= 2;
753                 E1000_ROUNDUP(min_tx_space, 1024);
754                 min_tx_space >>= 10;
755                 E1000_ROUNDUP(min_rx_space, 1024);
756                 min_rx_space >>= 10;
757
758                 /* If current Tx allocation is less than the min Tx FIFO size,
759                  * and the min Tx FIFO size is less than the current Rx FIFO
760                  * allocation, take space away from current Rx allocation */
761                 if (tx_space < min_tx_space &&
762                     ((min_tx_space - tx_space) < pba)) {
763                         pba = pba - (min_tx_space - tx_space);
764
765                         /* PCI/PCIx hardware has PBA alignment constraints */
766                         switch (adapter->hw.mac_type) {
767                         case e1000_82545 ... e1000_82546_rev_3:
768                                 pba &= ~(E1000_PBA_8K - 1);
769                                 break;
770                         default:
771                                 break;
772                         }
773
774                         /* if short on rx space, rx wins and must trump tx
775                          * adjustment or use Early Receive if available */
776                         if (pba < min_rx_space) {
777                                 switch (adapter->hw.mac_type) {
778                                 case e1000_82573:
779                                         /* ERT enabled in e1000_configure_rx */
780                                         break;
781                                 default:
782                                         pba = min_rx_space;
783                                         break;
784                                 }
785                         }
786                 }
787         }
788
789         E1000_WRITE_REG(&adapter->hw, PBA, pba);
790
791         /* flow control settings */
792         /* Set the FC high water mark to 90% of the FIFO size.
793          * Required to clear last 3 LSB */
794         fc_high_water_mark = ((pba * 9216)/10) & 0xFFF8;
795         /* We can't use 90% on small FIFOs because the remainder
796          * would be less than 1 full frame.  In this case, we size
797          * it to allow at least a full frame above the high water
798          *  mark. */
799         if (pba < E1000_PBA_16K)
800                 fc_high_water_mark = (pba * 1024) - 1600;
801
802         adapter->hw.fc_high_water = fc_high_water_mark;
803         adapter->hw.fc_low_water = fc_high_water_mark - 8;
804         if (adapter->hw.mac_type == e1000_80003es2lan)
805                 adapter->hw.fc_pause_time = 0xFFFF;
806         else
807                 adapter->hw.fc_pause_time = E1000_FC_PAUSE_TIME;
808         adapter->hw.fc_send_xon = 1;
809         adapter->hw.fc = adapter->hw.original_fc;
810
811         /* Allow time for pending master requests to run */
812         e1000_reset_hw(&adapter->hw);
813         if (adapter->hw.mac_type >= e1000_82544)
814                 E1000_WRITE_REG(&adapter->hw, WUC, 0);
815
816         if (e1000_init_hw(&adapter->hw))
817                 DPRINTK(PROBE, ERR, "Hardware Error\n");
818         e1000_update_mng_vlan(adapter);
819
820         /* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
821         if (adapter->hw.mac_type >= e1000_82544 &&
822             adapter->hw.mac_type <= e1000_82547_rev_2 &&
823             adapter->hw.autoneg == 1 &&
824             adapter->hw.autoneg_advertised == ADVERTISE_1000_FULL) {
825                 uint32_t ctrl = E1000_READ_REG(&adapter->hw, CTRL);
826                 /* clear phy power management bit if we are in gig only mode,
827                  * which if enabled will attempt negotiation to 100Mb, which
828                  * can cause a loss of link at power off or driver unload */
829                 ctrl &= ~E1000_CTRL_SWDPIN3;
830                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
831         }
832
833         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
834         E1000_WRITE_REG(&adapter->hw, VET, ETHERNET_IEEE_VLAN_TYPE);
835
836         e1000_reset_adaptive(&adapter->hw);
837         e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
838
839         if (!adapter->smart_power_down &&
840             (adapter->hw.mac_type == e1000_82571 ||
841              adapter->hw.mac_type == e1000_82572)) {
842                 uint16_t phy_data = 0;
843                 /* speed up time to link by disabling smart power down, ignore
844                  * the return value of this function because there is nothing
845                  * different we would do if it failed */
846                 e1000_read_phy_reg(&adapter->hw, IGP02E1000_PHY_POWER_MGMT,
847                                    &phy_data);
848                 phy_data &= ~IGP02E1000_PM_SPD;
849                 e1000_write_phy_reg(&adapter->hw, IGP02E1000_PHY_POWER_MGMT,
850                                     phy_data);
851         }
852
853         e1000_release_manageability(adapter);
854 }
855
856 /**
857  * e1000_probe - Device Initialization Routine
858  * @pdev: PCI device information struct
859  * @ent: entry in e1000_pci_tbl
860  *
861  * Returns 0 on success, negative on failure
862  *
863  * e1000_probe initializes an adapter identified by a pci_dev structure.
864  * The OS initialization, configuring of the adapter private structure,
865  * and a hardware reset occur.
866  **/
867
868 static int __devinit
869 e1000_probe(struct pci_dev *pdev,
870             const struct pci_device_id *ent)
871 {
872         struct net_device *netdev;
873         struct e1000_adapter *adapter;
874         unsigned long mmio_start, mmio_len;
875         unsigned long flash_start, flash_len;
876
877         static int cards_found = 0;
878         static int global_quad_port_a = 0; /* global ksp3 port a indication */
879         int i, err, pci_using_dac;
880         uint16_t eeprom_data = 0;
881         uint16_t eeprom_apme_mask = E1000_EEPROM_APME;
882         if ((err = pci_enable_device(pdev)))
883                 return err;
884
885         if (!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK)) &&
886             !(err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK))) {
887                 pci_using_dac = 1;
888         } else {
889                 if ((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK)) &&
890                     (err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK))) {
891                         E1000_ERR("No usable DMA configuration, aborting\n");
892                         goto err_dma;
893                 }
894                 pci_using_dac = 0;
895         }
896
897         if ((err = pci_request_regions(pdev, e1000_driver_name)))
898                 goto err_pci_reg;
899
900         pci_set_master(pdev);
901
902         err = -ENOMEM;
903         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
904         if (!netdev)
905                 goto err_alloc_etherdev;
906
907         SET_MODULE_OWNER(netdev);
908         SET_NETDEV_DEV(netdev, &pdev->dev);
909
910         pci_set_drvdata(pdev, netdev);
911         adapter = netdev_priv(netdev);
912         adapter->netdev = netdev;
913         adapter->pdev = pdev;
914         adapter->hw.back = adapter;
915         adapter->msg_enable = (1 << debug) - 1;
916
917         mmio_start = pci_resource_start(pdev, BAR_0);
918         mmio_len = pci_resource_len(pdev, BAR_0);
919
920         err = -EIO;
921         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
922         if (!adapter->hw.hw_addr)
923                 goto err_ioremap;
924
925         for (i = BAR_1; i <= BAR_5; i++) {
926                 if (pci_resource_len(pdev, i) == 0)
927                         continue;
928                 if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
929                         adapter->hw.io_base = pci_resource_start(pdev, i);
930                         break;
931                 }
932         }
933
934         netdev->open = &e1000_open;
935         netdev->stop = &e1000_close;
936         netdev->hard_start_xmit = &e1000_xmit_frame;
937         netdev->get_stats = &e1000_get_stats;
938         netdev->set_multicast_list = &e1000_set_multi;
939         netdev->set_mac_address = &e1000_set_mac;
940         netdev->change_mtu = &e1000_change_mtu;
941         netdev->do_ioctl = &e1000_ioctl;
942         e1000_set_ethtool_ops(netdev);
943         netdev->tx_timeout = &e1000_tx_timeout;
944         netdev->watchdog_timeo = 5 * HZ;
945 #ifdef CONFIG_E1000_NAPI
946         netdev->poll = &e1000_clean;
947         netdev->weight = 64;
948 #endif
949         netdev->vlan_rx_register = e1000_vlan_rx_register;
950         netdev->vlan_rx_add_vid = e1000_vlan_rx_add_vid;
951         netdev->vlan_rx_kill_vid = e1000_vlan_rx_kill_vid;
952 #ifdef CONFIG_NET_POLL_CONTROLLER
953         netdev->poll_controller = e1000_netpoll;
954 #endif
955         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
956
957         netdev->mem_start = mmio_start;
958         netdev->mem_end = mmio_start + mmio_len;
959         netdev->base_addr = adapter->hw.io_base;
960
961         adapter->bd_number = cards_found;
962
963         /* setup the private structure */
964
965         if ((err = e1000_sw_init(adapter)))
966                 goto err_sw_init;
967
968         err = -EIO;
969         /* Flash BAR mapping must happen after e1000_sw_init
970          * because it depends on mac_type */
971         if ((adapter->hw.mac_type == e1000_ich8lan) &&
972            (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
973                 flash_start = pci_resource_start(pdev, 1);
974                 flash_len = pci_resource_len(pdev, 1);
975                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
976                 if (!adapter->hw.flash_address)
977                         goto err_flashmap;
978         }
979
980         if (e1000_check_phy_reset_block(&adapter->hw))
981                 DPRINTK(PROBE, INFO, "PHY reset is blocked due to SOL/IDER session.\n");
982
983         if (adapter->hw.mac_type >= e1000_82543) {
984                 netdev->features = NETIF_F_SG |
985                                    NETIF_F_HW_CSUM |
986                                    NETIF_F_HW_VLAN_TX |
987                                    NETIF_F_HW_VLAN_RX |
988                                    NETIF_F_HW_VLAN_FILTER;
989                 if (adapter->hw.mac_type == e1000_ich8lan)
990                         netdev->features &= ~NETIF_F_HW_VLAN_FILTER;
991         }
992
993 #ifdef NETIF_F_TSO
994         if ((adapter->hw.mac_type >= e1000_82544) &&
995            (adapter->hw.mac_type != e1000_82547))
996                 netdev->features |= NETIF_F_TSO;
997
998 #ifdef NETIF_F_TSO6
999         if (adapter->hw.mac_type > e1000_82547_rev_2)
1000                 netdev->features |= NETIF_F_TSO6;
1001 #endif
1002 #endif
1003         if (pci_using_dac)
1004                 netdev->features |= NETIF_F_HIGHDMA;
1005
1006         netdev->features |= NETIF_F_LLTX;
1007
1008         adapter->en_mng_pt = e1000_enable_mng_pass_thru(&adapter->hw);
1009
1010         /* initialize eeprom parameters */
1011
1012         if (e1000_init_eeprom_params(&adapter->hw)) {
1013                 E1000_ERR("EEPROM initialization failed\n");
1014                 goto err_eeprom;
1015         }
1016
1017         /* before reading the EEPROM, reset the controller to
1018          * put the device in a known good starting state */
1019
1020         e1000_reset_hw(&adapter->hw);
1021
1022         /* make sure the EEPROM is good */
1023
1024         if (e1000_validate_eeprom_checksum(&adapter->hw) < 0) {
1025                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
1026                 goto err_eeprom;
1027         }
1028
1029         /* copy the MAC address out of the EEPROM */
1030
1031         if (e1000_read_mac_addr(&adapter->hw))
1032                 DPRINTK(PROBE, ERR, "EEPROM Read Error\n");
1033         memcpy(netdev->dev_addr, adapter->hw.mac_addr, netdev->addr_len);
1034         memcpy(netdev->perm_addr, adapter->hw.mac_addr, netdev->addr_len);
1035
1036         if (!is_valid_ether_addr(netdev->perm_addr)) {
1037                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
1038                 goto err_eeprom;
1039         }
1040
1041         e1000_get_bus_info(&adapter->hw);
1042
1043         init_timer(&adapter->tx_fifo_stall_timer);
1044         adapter->tx_fifo_stall_timer.function = &e1000_82547_tx_fifo_stall;
1045         adapter->tx_fifo_stall_timer.data = (unsigned long) adapter;
1046
1047         init_timer(&adapter->watchdog_timer);
1048         adapter->watchdog_timer.function = &e1000_watchdog;
1049         adapter->watchdog_timer.data = (unsigned long) adapter;
1050
1051         init_timer(&adapter->phy_info_timer);
1052         adapter->phy_info_timer.function = &e1000_update_phy_info;
1053         adapter->phy_info_timer.data = (unsigned long) adapter;
1054
1055         INIT_WORK(&adapter->reset_task, e1000_reset_task);
1056
1057         e1000_check_options(adapter);
1058
1059         /* Initial Wake on LAN setting
1060          * If APM wake is enabled in the EEPROM,
1061          * enable the ACPI Magic Packet filter
1062          */
1063
1064         switch (adapter->hw.mac_type) {
1065         case e1000_82542_rev2_0:
1066         case e1000_82542_rev2_1:
1067         case e1000_82543:
1068                 break;
1069         case e1000_82544:
1070                 e1000_read_eeprom(&adapter->hw,
1071                         EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
1072                 eeprom_apme_mask = E1000_EEPROM_82544_APM;
1073                 break;
1074         case e1000_ich8lan:
1075                 e1000_read_eeprom(&adapter->hw,
1076                         EEPROM_INIT_CONTROL1_REG, 1, &eeprom_data);
1077                 eeprom_apme_mask = E1000_EEPROM_ICH8_APME;
1078                 break;
1079         case e1000_82546:
1080         case e1000_82546_rev_3:
1081         case e1000_82571:
1082         case e1000_80003es2lan:
1083                 if (E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_FUNC_1){
1084                         e1000_read_eeprom(&adapter->hw,
1085                                 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1086                         break;
1087                 }
1088                 /* Fall Through */
1089         default:
1090                 e1000_read_eeprom(&adapter->hw,
1091                         EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1092                 break;
1093         }
1094         if (eeprom_data & eeprom_apme_mask)
1095                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1096
1097         /* now that we have the eeprom settings, apply the special cases
1098          * where the eeprom may be wrong or the board simply won't support
1099          * wake on lan on a particular port */
1100         switch (pdev->device) {
1101         case E1000_DEV_ID_82546GB_PCIE:
1102                 adapter->eeprom_wol = 0;
1103                 break;
1104         case E1000_DEV_ID_82546EB_FIBER:
1105         case E1000_DEV_ID_82546GB_FIBER:
1106         case E1000_DEV_ID_82571EB_FIBER:
1107                 /* Wake events only supported on port A for dual fiber
1108                  * regardless of eeprom setting */
1109                 if (E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_FUNC_1)
1110                         adapter->eeprom_wol = 0;
1111                 break;
1112         case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
1113         case E1000_DEV_ID_82571EB_QUAD_COPPER:
1114         case E1000_DEV_ID_82571EB_QUAD_COPPER_LOWPROFILE:
1115                 /* if quad port adapter, disable WoL on all but port A */
1116                 if (global_quad_port_a != 0)
1117                         adapter->eeprom_wol = 0;
1118                 else
1119                         adapter->quad_port_a = 1;
1120                 /* Reset for multiple quad port adapters */
1121                 if (++global_quad_port_a == 4)
1122                         global_quad_port_a = 0;
1123                 break;
1124         }
1125
1126         /* initialize the wol settings based on the eeprom settings */
1127         adapter->wol = adapter->eeprom_wol;
1128
1129         /* print bus type/speed/width info */
1130         {
1131         struct e1000_hw *hw = &adapter->hw;
1132         DPRINTK(PROBE, INFO, "(PCI%s:%s:%s) ",
1133                 ((hw->bus_type == e1000_bus_type_pcix) ? "-X" :
1134                  (hw->bus_type == e1000_bus_type_pci_express ? " Express":"")),
1135                 ((hw->bus_speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
1136                  (hw->bus_speed == e1000_bus_speed_133) ? "133MHz" :
1137                  (hw->bus_speed == e1000_bus_speed_120) ? "120MHz" :
1138                  (hw->bus_speed == e1000_bus_speed_100) ? "100MHz" :
1139                  (hw->bus_speed == e1000_bus_speed_66) ? "66MHz" : "33MHz"),
1140                 ((hw->bus_width == e1000_bus_width_64) ? "64-bit" :
1141                  (hw->bus_width == e1000_bus_width_pciex_4) ? "Width x4" :
1142                  (hw->bus_width == e1000_bus_width_pciex_1) ? "Width x1" :
1143                  "32-bit"));
1144         }
1145
1146         for (i = 0; i < 6; i++)
1147                 printk("%2.2x%c", netdev->dev_addr[i], i == 5 ? '\n' : ':');
1148
1149         /* reset the hardware with the new settings */
1150         e1000_reset(adapter);
1151
1152         /* If the controller is 82573 and f/w is AMT, do not set
1153          * DRV_LOAD until the interface is up.  For all other cases,
1154          * let the f/w know that the h/w is now under the control
1155          * of the driver. */
1156         if (adapter->hw.mac_type != e1000_82573 ||
1157             !e1000_check_mng_mode(&adapter->hw))
1158                 e1000_get_hw_control(adapter);
1159
1160         strcpy(netdev->name, "eth%d");
1161         if ((err = register_netdev(netdev)))
1162                 goto err_register;
1163
1164         /* tell the stack to leave us alone until e1000_open() is called */
1165         netif_carrier_off(netdev);
1166         netif_stop_queue(netdev);
1167
1168         DPRINTK(PROBE, INFO, "Intel(R) PRO/1000 Network Connection\n");
1169
1170         cards_found++;
1171         return 0;
1172
1173 err_register:
1174         e1000_release_hw_control(adapter);
1175 err_eeprom:
1176         if (!e1000_check_phy_reset_block(&adapter->hw))
1177                 e1000_phy_hw_reset(&adapter->hw);
1178
1179         if (adapter->hw.flash_address)
1180                 iounmap(adapter->hw.flash_address);
1181 err_flashmap:
1182 #ifdef CONFIG_E1000_NAPI
1183         for (i = 0; i < adapter->num_rx_queues; i++)
1184                 dev_put(&adapter->polling_netdev[i]);
1185 #endif
1186
1187         kfree(adapter->tx_ring);
1188         kfree(adapter->rx_ring);
1189 #ifdef CONFIG_E1000_NAPI
1190         kfree(adapter->polling_netdev);
1191 #endif
1192 err_sw_init:
1193         iounmap(adapter->hw.hw_addr);
1194 err_ioremap:
1195         free_netdev(netdev);
1196 err_alloc_etherdev:
1197         pci_release_regions(pdev);
1198 err_pci_reg:
1199 err_dma:
1200         pci_disable_device(pdev);
1201         return err;
1202 }
1203
1204 /**
1205  * e1000_remove - Device Removal Routine
1206  * @pdev: PCI device information struct
1207  *
1208  * e1000_remove is called by the PCI subsystem to alert the driver
1209  * that it should release a PCI device.  The could be caused by a
1210  * Hot-Plug event, or because the driver is going to be removed from
1211  * memory.
1212  **/
1213
1214 static void __devexit
1215 e1000_remove(struct pci_dev *pdev)
1216 {
1217         struct net_device *netdev = pci_get_drvdata(pdev);
1218         struct e1000_adapter *adapter = netdev_priv(netdev);
1219 #ifdef CONFIG_E1000_NAPI
1220         int i;
1221 #endif
1222
1223         flush_scheduled_work();
1224
1225         e1000_release_manageability(adapter);
1226
1227         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
1228          * would have already happened in close and is redundant. */
1229         e1000_release_hw_control(adapter);
1230
1231         unregister_netdev(netdev);
1232 #ifdef CONFIG_E1000_NAPI
1233         for (i = 0; i < adapter->num_rx_queues; i++)
1234                 dev_put(&adapter->polling_netdev[i]);
1235 #endif
1236
1237         if (!e1000_check_phy_reset_block(&adapter->hw))
1238                 e1000_phy_hw_reset(&adapter->hw);
1239
1240         kfree(adapter->tx_ring);
1241         kfree(adapter->rx_ring);
1242 #ifdef CONFIG_E1000_NAPI
1243         kfree(adapter->polling_netdev);
1244 #endif
1245
1246         iounmap(adapter->hw.hw_addr);
1247         if (adapter->hw.flash_address)
1248                 iounmap(adapter->hw.flash_address);
1249         pci_release_regions(pdev);
1250
1251         free_netdev(netdev);
1252
1253         pci_disable_device(pdev);
1254 }
1255
1256 /**
1257  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1258  * @adapter: board private structure to initialize
1259  *
1260  * e1000_sw_init initializes the Adapter private data structure.
1261  * Fields are initialized based on PCI device information and
1262  * OS network device settings (MTU size).
1263  **/
1264
1265 static int __devinit
1266 e1000_sw_init(struct e1000_adapter *adapter)
1267 {
1268         struct e1000_hw *hw = &adapter->hw;
1269         struct net_device *netdev = adapter->netdev;
1270         struct pci_dev *pdev = adapter->pdev;
1271 #ifdef CONFIG_E1000_NAPI
1272         int i;
1273 #endif
1274
1275         /* PCI config space info */
1276
1277         hw->vendor_id = pdev->vendor;
1278         hw->device_id = pdev->device;
1279         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1280         hw->subsystem_id = pdev->subsystem_device;
1281
1282         pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
1283
1284         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
1285
1286         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1287         adapter->rx_ps_bsize0 = E1000_RXBUFFER_128;
1288         hw->max_frame_size = netdev->mtu +
1289                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
1290         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
1291
1292         /* identify the MAC */
1293
1294         if (e1000_set_mac_type(hw)) {
1295                 DPRINTK(PROBE, ERR, "Unknown MAC Type\n");
1296                 return -EIO;
1297         }
1298
1299         switch (hw->mac_type) {
1300         default:
1301                 break;
1302         case e1000_82541:
1303         case e1000_82547:
1304         case e1000_82541_rev_2:
1305         case e1000_82547_rev_2:
1306                 hw->phy_init_script = 1;
1307                 break;
1308         }
1309
1310         e1000_set_media_type(hw);
1311
1312         hw->wait_autoneg_complete = FALSE;
1313         hw->tbi_compatibility_en = TRUE;
1314         hw->adaptive_ifs = TRUE;
1315
1316         /* Copper options */
1317
1318         if (hw->media_type == e1000_media_type_copper) {
1319                 hw->mdix = AUTO_ALL_MODES;
1320                 hw->disable_polarity_correction = FALSE;
1321                 hw->master_slave = E1000_MASTER_SLAVE;
1322         }
1323
1324         adapter->num_tx_queues = 1;
1325         adapter->num_rx_queues = 1;
1326
1327         if (e1000_alloc_queues(adapter)) {
1328                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
1329                 return -ENOMEM;
1330         }
1331
1332 #ifdef CONFIG_E1000_NAPI
1333         for (i = 0; i < adapter->num_rx_queues; i++) {
1334                 adapter->polling_netdev[i].priv = adapter;
1335                 adapter->polling_netdev[i].poll = &e1000_clean;
1336                 adapter->polling_netdev[i].weight = 64;
1337                 dev_hold(&adapter->polling_netdev[i]);
1338                 set_bit(__LINK_STATE_START, &adapter->polling_netdev[i].state);
1339         }
1340         spin_lock_init(&adapter->tx_queue_lock);
1341 #endif
1342
1343         atomic_set(&adapter->irq_sem, 1);
1344         spin_lock_init(&adapter->stats_lock);
1345
1346         set_bit(__E1000_DOWN, &adapter->flags);
1347
1348         return 0;
1349 }
1350
1351 /**
1352  * e1000_alloc_queues - Allocate memory for all rings
1353  * @adapter: board private structure to initialize
1354  *
1355  * We allocate one ring per queue at run-time since we don't know the
1356  * number of queues at compile-time.  The polling_netdev array is
1357  * intended for Multiqueue, but should work fine with a single queue.
1358  **/
1359
1360 static int __devinit
1361 e1000_alloc_queues(struct e1000_adapter *adapter)
1362 {
1363         int size;
1364
1365         size = sizeof(struct e1000_tx_ring) * adapter->num_tx_queues;
1366         adapter->tx_ring = kmalloc(size, GFP_KERNEL);
1367         if (!adapter->tx_ring)
1368                 return -ENOMEM;
1369         memset(adapter->tx_ring, 0, size);
1370
1371         size = sizeof(struct e1000_rx_ring) * adapter->num_rx_queues;
1372         adapter->rx_ring = kmalloc(size, GFP_KERNEL);
1373         if (!adapter->rx_ring) {
1374                 kfree(adapter->tx_ring);
1375                 return -ENOMEM;
1376         }
1377         memset(adapter->rx_ring, 0, size);
1378
1379 #ifdef CONFIG_E1000_NAPI
1380         size = sizeof(struct net_device) * adapter->num_rx_queues;
1381         adapter->polling_netdev = kmalloc(size, GFP_KERNEL);
1382         if (!adapter->polling_netdev) {
1383                 kfree(adapter->tx_ring);
1384                 kfree(adapter->rx_ring);
1385                 return -ENOMEM;
1386         }
1387         memset(adapter->polling_netdev, 0, size);
1388 #endif
1389
1390         return E1000_SUCCESS;
1391 }
1392
1393 /**
1394  * e1000_open - Called when a network interface is made active
1395  * @netdev: network interface device structure
1396  *
1397  * Returns 0 on success, negative value on failure
1398  *
1399  * The open entry point is called when a network interface is made
1400  * active by the system (IFF_UP).  At this point all resources needed
1401  * for transmit and receive operations are allocated, the interrupt
1402  * handler is registered with the OS, the watchdog timer is started,
1403  * and the stack is notified that the interface is ready.
1404  **/
1405
1406 static int
1407 e1000_open(struct net_device *netdev)
1408 {
1409         struct e1000_adapter *adapter = netdev_priv(netdev);
1410         int err;
1411
1412         /* disallow open during test */
1413         if (test_bit(__E1000_TESTING, &adapter->flags))
1414                 return -EBUSY;
1415
1416         /* allocate transmit descriptors */
1417         if ((err = e1000_setup_all_tx_resources(adapter)))
1418                 goto err_setup_tx;
1419
1420         /* allocate receive descriptors */
1421         if ((err = e1000_setup_all_rx_resources(adapter)))
1422                 goto err_setup_rx;
1423
1424         err = e1000_request_irq(adapter);
1425         if (err)
1426                 goto err_req_irq;
1427
1428         e1000_power_up_phy(adapter);
1429
1430         if ((err = e1000_up(adapter)))
1431                 goto err_up;
1432         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1433         if ((adapter->hw.mng_cookie.status &
1434                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1435                 e1000_update_mng_vlan(adapter);
1436         }
1437
1438         /* If AMT is enabled, let the firmware know that the network
1439          * interface is now open */
1440         if (adapter->hw.mac_type == e1000_82573 &&
1441             e1000_check_mng_mode(&adapter->hw))
1442                 e1000_get_hw_control(adapter);
1443
1444         return E1000_SUCCESS;
1445
1446 err_up:
1447         e1000_power_down_phy(adapter);
1448         e1000_free_irq(adapter);
1449 err_req_irq:
1450         e1000_free_all_rx_resources(adapter);
1451 err_setup_rx:
1452         e1000_free_all_tx_resources(adapter);
1453 err_setup_tx:
1454         e1000_reset(adapter);
1455
1456         return err;
1457 }
1458
1459 /**
1460  * e1000_close - Disables a network interface
1461  * @netdev: network interface device structure
1462  *
1463  * Returns 0, this is not allowed to fail
1464  *
1465  * The close entry point is called when an interface is de-activated
1466  * by the OS.  The hardware is still under the drivers control, but
1467  * needs to be disabled.  A global MAC reset is issued to stop the
1468  * hardware, and all transmit and receive resources are freed.
1469  **/
1470
1471 static int
1472 e1000_close(struct net_device *netdev)
1473 {
1474         struct e1000_adapter *adapter = netdev_priv(netdev);
1475
1476         WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1477         e1000_down(adapter);
1478         e1000_power_down_phy(adapter);
1479         e1000_free_irq(adapter);
1480
1481         e1000_free_all_tx_resources(adapter);
1482         e1000_free_all_rx_resources(adapter);
1483
1484         /* kill manageability vlan ID if supported, but not if a vlan with
1485          * the same ID is registered on the host OS (let 8021q kill it) */
1486         if ((adapter->hw.mng_cookie.status &
1487                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1488              !(adapter->vlgrp &&
1489                           adapter->vlgrp->vlan_devices[adapter->mng_vlan_id])) {
1490                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1491         }
1492
1493         /* If AMT is enabled, let the firmware know that the network
1494          * interface is now closed */
1495         if (adapter->hw.mac_type == e1000_82573 &&
1496             e1000_check_mng_mode(&adapter->hw))
1497                 e1000_release_hw_control(adapter);
1498
1499         return 0;
1500 }
1501
1502 /**
1503  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1504  * @adapter: address of board private structure
1505  * @start: address of beginning of memory
1506  * @len: length of memory
1507  **/
1508 static boolean_t
1509 e1000_check_64k_bound(struct e1000_adapter *adapter,
1510                       void *start, unsigned long len)
1511 {
1512         unsigned long begin = (unsigned long) start;
1513         unsigned long end = begin + len;
1514
1515         /* First rev 82545 and 82546 need to not allow any memory
1516          * write location to cross 64k boundary due to errata 23 */
1517         if (adapter->hw.mac_type == e1000_82545 ||
1518             adapter->hw.mac_type == e1000_82546) {
1519                 return ((begin ^ (end - 1)) >> 16) != 0 ? FALSE : TRUE;
1520         }
1521
1522         return TRUE;
1523 }
1524
1525 /**
1526  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1527  * @adapter: board private structure
1528  * @txdr:    tx descriptor ring (for a specific queue) to setup
1529  *
1530  * Return 0 on success, negative on failure
1531  **/
1532
1533 static int
1534 e1000_setup_tx_resources(struct e1000_adapter *adapter,
1535                          struct e1000_tx_ring *txdr)
1536 {
1537         struct pci_dev *pdev = adapter->pdev;
1538         int size;
1539
1540         size = sizeof(struct e1000_buffer) * txdr->count;
1541         txdr->buffer_info = vmalloc(size);
1542         if (!txdr->buffer_info) {
1543                 DPRINTK(PROBE, ERR,
1544                 "Unable to allocate memory for the transmit descriptor ring\n");
1545                 return -ENOMEM;
1546         }
1547         memset(txdr->buffer_info, 0, size);
1548
1549         /* round up to nearest 4K */
1550
1551         txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1552         E1000_ROUNDUP(txdr->size, 4096);
1553
1554         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1555         if (!txdr->desc) {
1556 setup_tx_desc_die:
1557                 vfree(txdr->buffer_info);
1558                 DPRINTK(PROBE, ERR,
1559                 "Unable to allocate memory for the transmit descriptor ring\n");
1560                 return -ENOMEM;
1561         }
1562
1563         /* Fix for errata 23, can't cross 64kB boundary */
1564         if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1565                 void *olddesc = txdr->desc;
1566                 dma_addr_t olddma = txdr->dma;
1567                 DPRINTK(TX_ERR, ERR, "txdr align check failed: %u bytes "
1568                                      "at %p\n", txdr->size, txdr->desc);
1569                 /* Try again, without freeing the previous */
1570                 txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1571                 /* Failed allocation, critical failure */
1572                 if (!txdr->desc) {
1573                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1574                         goto setup_tx_desc_die;
1575                 }
1576
1577                 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1578                         /* give up */
1579                         pci_free_consistent(pdev, txdr->size, txdr->desc,
1580                                             txdr->dma);
1581                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1582                         DPRINTK(PROBE, ERR,
1583                                 "Unable to allocate aligned memory "
1584                                 "for the transmit descriptor ring\n");
1585                         vfree(txdr->buffer_info);
1586                         return -ENOMEM;
1587                 } else {
1588                         /* Free old allocation, new allocation was successful */
1589                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1590                 }
1591         }
1592         memset(txdr->desc, 0, txdr->size);
1593
1594         txdr->next_to_use = 0;
1595         txdr->next_to_clean = 0;
1596         spin_lock_init(&txdr->tx_lock);
1597
1598         return 0;
1599 }
1600
1601 /**
1602  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1603  *                                (Descriptors) for all queues
1604  * @adapter: board private structure
1605  *
1606  * Return 0 on success, negative on failure
1607  **/
1608
1609 int
1610 e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1611 {
1612         int i, err = 0;
1613
1614         for (i = 0; i < adapter->num_tx_queues; i++) {
1615                 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1616                 if (err) {
1617                         DPRINTK(PROBE, ERR,
1618                                 "Allocation for Tx Queue %u failed\n", i);
1619                         for (i-- ; i >= 0; i--)
1620                                 e1000_free_tx_resources(adapter,
1621                                                         &adapter->tx_ring[i]);
1622                         break;
1623                 }
1624         }
1625
1626         return err;
1627 }
1628
1629 /**
1630  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1631  * @adapter: board private structure
1632  *
1633  * Configure the Tx unit of the MAC after a reset.
1634  **/
1635
1636 static void
1637 e1000_configure_tx(struct e1000_adapter *adapter)
1638 {
1639         uint64_t tdba;
1640         struct e1000_hw *hw = &adapter->hw;
1641         uint32_t tdlen, tctl, tipg, tarc;
1642         uint32_t ipgr1, ipgr2;
1643
1644         /* Setup the HW Tx Head and Tail descriptor pointers */
1645
1646         switch (adapter->num_tx_queues) {
1647         case 1:
1648         default:
1649                 tdba = adapter->tx_ring[0].dma;
1650                 tdlen = adapter->tx_ring[0].count *
1651                         sizeof(struct e1000_tx_desc);
1652                 E1000_WRITE_REG(hw, TDLEN, tdlen);
1653                 E1000_WRITE_REG(hw, TDBAH, (tdba >> 32));
1654                 E1000_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
1655                 E1000_WRITE_REG(hw, TDT, 0);
1656                 E1000_WRITE_REG(hw, TDH, 0);
1657                 adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ? E1000_TDH : E1000_82542_TDH);
1658                 adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ? E1000_TDT : E1000_82542_TDT);
1659                 break;
1660         }
1661
1662         /* Set the default values for the Tx Inter Packet Gap timer */
1663         if (adapter->hw.mac_type <= e1000_82547_rev_2 &&
1664             (hw->media_type == e1000_media_type_fiber ||
1665              hw->media_type == e1000_media_type_internal_serdes))
1666                 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1667         else
1668                 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1669
1670         switch (hw->mac_type) {
1671         case e1000_82542_rev2_0:
1672         case e1000_82542_rev2_1:
1673                 tipg = DEFAULT_82542_TIPG_IPGT;
1674                 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1675                 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1676                 break;
1677         case e1000_80003es2lan:
1678                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1679                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2;
1680                 break;
1681         default:
1682                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1683                 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1684                 break;
1685         }
1686         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1687         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1688         E1000_WRITE_REG(hw, TIPG, tipg);
1689
1690         /* Set the Tx Interrupt Delay register */
1691
1692         E1000_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
1693         if (hw->mac_type >= e1000_82540)
1694                 E1000_WRITE_REG(hw, TADV, adapter->tx_abs_int_delay);
1695
1696         /* Program the Transmit Control Register */
1697
1698         tctl = E1000_READ_REG(hw, TCTL);
1699         tctl &= ~E1000_TCTL_CT;
1700         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1701                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1702
1703         if (hw->mac_type == e1000_82571 || hw->mac_type == e1000_82572) {
1704                 tarc = E1000_READ_REG(hw, TARC0);
1705                 /* set the speed mode bit, we'll clear it if we're not at
1706                  * gigabit link later */
1707                 tarc |= (1 << 21);
1708                 E1000_WRITE_REG(hw, TARC0, tarc);
1709         } else if (hw->mac_type == e1000_80003es2lan) {
1710                 tarc = E1000_READ_REG(hw, TARC0);
1711                 tarc |= 1;
1712                 E1000_WRITE_REG(hw, TARC0, tarc);
1713                 tarc = E1000_READ_REG(hw, TARC1);
1714                 tarc |= 1;
1715                 E1000_WRITE_REG(hw, TARC1, tarc);
1716         }
1717
1718         e1000_config_collision_dist(hw);
1719
1720         /* Setup Transmit Descriptor Settings for eop descriptor */
1721         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1722
1723         /* only set IDE if we are delaying interrupts using the timers */
1724         if (adapter->tx_int_delay)
1725                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1726
1727         if (hw->mac_type < e1000_82543)
1728                 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1729         else
1730                 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1731
1732         /* Cache if we're 82544 running in PCI-X because we'll
1733          * need this to apply a workaround later in the send path. */
1734         if (hw->mac_type == e1000_82544 &&
1735             hw->bus_type == e1000_bus_type_pcix)
1736                 adapter->pcix_82544 = 1;
1737
1738         E1000_WRITE_REG(hw, TCTL, tctl);
1739
1740 }
1741
1742 /**
1743  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1744  * @adapter: board private structure
1745  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1746  *
1747  * Returns 0 on success, negative on failure
1748  **/
1749
1750 static int
1751 e1000_setup_rx_resources(struct e1000_adapter *adapter,
1752                          struct e1000_rx_ring *rxdr)
1753 {
1754         struct pci_dev *pdev = adapter->pdev;
1755         int size, desc_len;
1756
1757         size = sizeof(struct e1000_buffer) * rxdr->count;
1758         rxdr->buffer_info = vmalloc(size);
1759         if (!rxdr->buffer_info) {
1760                 DPRINTK(PROBE, ERR,
1761                 "Unable to allocate memory for the receive descriptor ring\n");
1762                 return -ENOMEM;
1763         }
1764         memset(rxdr->buffer_info, 0, size);
1765
1766         size = sizeof(struct e1000_ps_page) * rxdr->count;
1767         rxdr->ps_page = kmalloc(size, GFP_KERNEL);
1768         if (!rxdr->ps_page) {
1769                 vfree(rxdr->buffer_info);
1770                 DPRINTK(PROBE, ERR,
1771                 "Unable to allocate memory for the receive descriptor ring\n");
1772                 return -ENOMEM;
1773         }
1774         memset(rxdr->ps_page, 0, size);
1775
1776         size = sizeof(struct e1000_ps_page_dma) * rxdr->count;
1777         rxdr->ps_page_dma = kmalloc(size, GFP_KERNEL);
1778         if (!rxdr->ps_page_dma) {
1779                 vfree(rxdr->buffer_info);
1780                 kfree(rxdr->ps_page);
1781                 DPRINTK(PROBE, ERR,
1782                 "Unable to allocate memory for the receive descriptor ring\n");
1783                 return -ENOMEM;
1784         }
1785         memset(rxdr->ps_page_dma, 0, size);
1786
1787         if (adapter->hw.mac_type <= e1000_82547_rev_2)
1788                 desc_len = sizeof(struct e1000_rx_desc);
1789         else
1790                 desc_len = sizeof(union e1000_rx_desc_packet_split);
1791
1792         /* Round up to nearest 4K */
1793
1794         rxdr->size = rxdr->count * desc_len;
1795         E1000_ROUNDUP(rxdr->size, 4096);
1796
1797         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1798
1799         if (!rxdr->desc) {
1800                 DPRINTK(PROBE, ERR,
1801                 "Unable to allocate memory for the receive descriptor ring\n");
1802 setup_rx_desc_die:
1803                 vfree(rxdr->buffer_info);
1804                 kfree(rxdr->ps_page);
1805                 kfree(rxdr->ps_page_dma);
1806                 return -ENOMEM;
1807         }
1808
1809         /* Fix for errata 23, can't cross 64kB boundary */
1810         if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1811                 void *olddesc = rxdr->desc;
1812                 dma_addr_t olddma = rxdr->dma;
1813                 DPRINTK(RX_ERR, ERR, "rxdr align check failed: %u bytes "
1814                                      "at %p\n", rxdr->size, rxdr->desc);
1815                 /* Try again, without freeing the previous */
1816                 rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1817                 /* Failed allocation, critical failure */
1818                 if (!rxdr->desc) {
1819                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1820                         DPRINTK(PROBE, ERR,
1821                                 "Unable to allocate memory "
1822                                 "for the receive descriptor ring\n");
1823                         goto setup_rx_desc_die;
1824                 }
1825
1826                 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1827                         /* give up */
1828                         pci_free_consistent(pdev, rxdr->size, rxdr->desc,
1829                                             rxdr->dma);
1830                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1831                         DPRINTK(PROBE, ERR,
1832                                 "Unable to allocate aligned memory "
1833                                 "for the receive descriptor ring\n");
1834                         goto setup_rx_desc_die;
1835                 } else {
1836                         /* Free old allocation, new allocation was successful */
1837                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1838                 }
1839         }
1840         memset(rxdr->desc, 0, rxdr->size);
1841
1842         rxdr->next_to_clean = 0;
1843         rxdr->next_to_use = 0;
1844
1845         return 0;
1846 }
1847
1848 /**
1849  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1850  *                                (Descriptors) for all queues
1851  * @adapter: board private structure
1852  *
1853  * Return 0 on success, negative on failure
1854  **/
1855
1856 int
1857 e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1858 {
1859         int i, err = 0;
1860
1861         for (i = 0; i < adapter->num_rx_queues; i++) {
1862                 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1863                 if (err) {
1864                         DPRINTK(PROBE, ERR,
1865                                 "Allocation for Rx Queue %u failed\n", i);
1866                         for (i-- ; i >= 0; i--)
1867                                 e1000_free_rx_resources(adapter,
1868                                                         &adapter->rx_ring[i]);
1869                         break;
1870                 }
1871         }
1872
1873         return err;
1874 }
1875
1876 /**
1877  * e1000_setup_rctl - configure the receive control registers
1878  * @adapter: Board private structure
1879  **/
1880 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1881                         (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1882 static void
1883 e1000_setup_rctl(struct e1000_adapter *adapter)
1884 {
1885         uint32_t rctl, rfctl;
1886         uint32_t psrctl = 0;
1887 #ifndef CONFIG_E1000_DISABLE_PACKET_SPLIT
1888         uint32_t pages = 0;
1889 #endif
1890
1891         rctl = E1000_READ_REG(&adapter->hw, RCTL);
1892
1893         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1894
1895         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1896                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1897                 (adapter->hw.mc_filter_type << E1000_RCTL_MO_SHIFT);
1898
1899         if (adapter->hw.tbi_compatibility_on == 1)
1900                 rctl |= E1000_RCTL_SBP;
1901         else
1902                 rctl &= ~E1000_RCTL_SBP;
1903
1904         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1905                 rctl &= ~E1000_RCTL_LPE;
1906         else
1907                 rctl |= E1000_RCTL_LPE;
1908
1909         /* Setup buffer sizes */
1910         rctl &= ~E1000_RCTL_SZ_4096;
1911         rctl |= E1000_RCTL_BSEX;
1912         switch (adapter->rx_buffer_len) {
1913                 case E1000_RXBUFFER_256:
1914                         rctl |= E1000_RCTL_SZ_256;
1915                         rctl &= ~E1000_RCTL_BSEX;
1916                         break;
1917                 case E1000_RXBUFFER_512:
1918                         rctl |= E1000_RCTL_SZ_512;
1919                         rctl &= ~E1000_RCTL_BSEX;
1920                         break;
1921                 case E1000_RXBUFFER_1024:
1922                         rctl |= E1000_RCTL_SZ_1024;
1923                         rctl &= ~E1000_RCTL_BSEX;
1924                         break;
1925                 case E1000_RXBUFFER_2048:
1926                 default:
1927                         rctl |= E1000_RCTL_SZ_2048;
1928                         rctl &= ~E1000_RCTL_BSEX;
1929                         break;
1930                 case E1000_RXBUFFER_4096:
1931                         rctl |= E1000_RCTL_SZ_4096;
1932                         break;
1933                 case E1000_RXBUFFER_8192:
1934                         rctl |= E1000_RCTL_SZ_8192;
1935                         break;
1936                 case E1000_RXBUFFER_16384:
1937                         rctl |= E1000_RCTL_SZ_16384;
1938                         break;
1939         }
1940
1941 #ifndef CONFIG_E1000_DISABLE_PACKET_SPLIT
1942         /* 82571 and greater support packet-split where the protocol
1943          * header is placed in skb->data and the packet data is
1944          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
1945          * In the case of a non-split, skb->data is linearly filled,
1946          * followed by the page buffers.  Therefore, skb->data is
1947          * sized to hold the largest protocol header.
1948          */
1949         /* allocations using alloc_page take too long for regular MTU
1950          * so only enable packet split for jumbo frames */
1951         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
1952         if ((adapter->hw.mac_type >= e1000_82571) && (pages <= 3) &&
1953             PAGE_SIZE <= 16384 && (rctl & E1000_RCTL_LPE))
1954                 adapter->rx_ps_pages = pages;
1955         else
1956                 adapter->rx_ps_pages = 0;
1957 #endif
1958         if (adapter->rx_ps_pages) {
1959                 /* Configure extra packet-split registers */
1960                 rfctl = E1000_READ_REG(&adapter->hw, RFCTL);
1961                 rfctl |= E1000_RFCTL_EXTEN;
1962                 /* disable packet split support for IPv6 extension headers,
1963                  * because some malformed IPv6 headers can hang the RX */
1964                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
1965                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
1966
1967                 E1000_WRITE_REG(&adapter->hw, RFCTL, rfctl);
1968
1969                 rctl |= E1000_RCTL_DTYP_PS;
1970
1971                 psrctl |= adapter->rx_ps_bsize0 >>
1972                         E1000_PSRCTL_BSIZE0_SHIFT;
1973
1974                 switch (adapter->rx_ps_pages) {
1975                 case 3:
1976                         psrctl |= PAGE_SIZE <<
1977                                 E1000_PSRCTL_BSIZE3_SHIFT;
1978                 case 2:
1979                         psrctl |= PAGE_SIZE <<
1980                                 E1000_PSRCTL_BSIZE2_SHIFT;
1981                 case 1:
1982                         psrctl |= PAGE_SIZE >>
1983                                 E1000_PSRCTL_BSIZE1_SHIFT;
1984                         break;
1985                 }
1986
1987                 E1000_WRITE_REG(&adapter->hw, PSRCTL, psrctl);
1988         }
1989
1990         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
1991 }
1992
1993 /**
1994  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1995  * @adapter: board private structure
1996  *
1997  * Configure the Rx unit of the MAC after a reset.
1998  **/
1999
2000 static void
2001 e1000_configure_rx(struct e1000_adapter *adapter)
2002 {
2003         uint64_t rdba;
2004         struct e1000_hw *hw = &adapter->hw;
2005         uint32_t rdlen, rctl, rxcsum, ctrl_ext;
2006
2007         if (adapter->rx_ps_pages) {
2008                 /* this is a 32 byte descriptor */
2009                 rdlen = adapter->rx_ring[0].count *
2010                         sizeof(union e1000_rx_desc_packet_split);
2011                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2012                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2013         } else {
2014                 rdlen = adapter->rx_ring[0].count *
2015                         sizeof(struct e1000_rx_desc);
2016                 adapter->clean_rx = e1000_clean_rx_irq;
2017                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2018         }
2019
2020         /* disable receives while setting up the descriptors */
2021         rctl = E1000_READ_REG(hw, RCTL);
2022         E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
2023
2024         /* set the Receive Delay Timer Register */
2025         E1000_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
2026
2027         if (hw->mac_type >= e1000_82540) {
2028                 E1000_WRITE_REG(hw, RADV, adapter->rx_abs_int_delay);
2029                 if (adapter->itr_setting != 0)
2030                         E1000_WRITE_REG(hw, ITR,
2031                                 1000000000 / (adapter->itr * 256));
2032         }
2033
2034         if (hw->mac_type >= e1000_82571) {
2035                 ctrl_ext = E1000_READ_REG(hw, CTRL_EXT);
2036                 /* Reset delay timers after every interrupt */
2037                 ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2038 #ifdef CONFIG_E1000_NAPI
2039                 /* Auto-Mask interrupts upon ICR access */
2040                 ctrl_ext |= E1000_CTRL_EXT_IAME;
2041                 E1000_WRITE_REG(hw, IAM, 0xffffffff);
2042 #endif
2043                 E1000_WRITE_REG(hw, CTRL_EXT, ctrl_ext);
2044                 E1000_WRITE_FLUSH(hw);
2045         }
2046
2047         /* Setup the HW Rx Head and Tail Descriptor Pointers and
2048          * the Base and Length of the Rx Descriptor Ring */
2049         switch (adapter->num_rx_queues) {
2050         case 1:
2051         default:
2052                 rdba = adapter->rx_ring[0].dma;
2053                 E1000_WRITE_REG(hw, RDLEN, rdlen);
2054                 E1000_WRITE_REG(hw, RDBAH, (rdba >> 32));
2055                 E1000_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
2056                 E1000_WRITE_REG(hw, RDT, 0);
2057                 E1000_WRITE_REG(hw, RDH, 0);
2058                 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ? E1000_RDH : E1000_82542_RDH);
2059                 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ? E1000_RDT : E1000_82542_RDT);
2060                 break;
2061         }
2062
2063         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
2064         if (hw->mac_type >= e1000_82543) {
2065                 rxcsum = E1000_READ_REG(hw, RXCSUM);
2066                 if (adapter->rx_csum == TRUE) {
2067                         rxcsum |= E1000_RXCSUM_TUOFL;
2068
2069                         /* Enable 82571 IPv4 payload checksum for UDP fragments
2070                          * Must be used in conjunction with packet-split. */
2071                         if ((hw->mac_type >= e1000_82571) &&
2072                             (adapter->rx_ps_pages)) {
2073                                 rxcsum |= E1000_RXCSUM_IPPCSE;
2074                         }
2075                 } else {
2076                         rxcsum &= ~E1000_RXCSUM_TUOFL;
2077                         /* don't need to clear IPPCSE as it defaults to 0 */
2078                 }
2079                 E1000_WRITE_REG(hw, RXCSUM, rxcsum);
2080         }
2081
2082         /* enable early receives on 82573, only takes effect if using > 2048
2083          * byte total frame size.  for example only for jumbo frames */
2084 #define E1000_ERT_2048 0x100
2085         if (hw->mac_type == e1000_82573)
2086                 E1000_WRITE_REG(hw, ERT, E1000_ERT_2048);
2087
2088         /* Enable Receives */
2089         E1000_WRITE_REG(hw, RCTL, rctl);
2090 }
2091
2092 /**
2093  * e1000_free_tx_resources - Free Tx Resources per Queue
2094  * @adapter: board private structure
2095  * @tx_ring: Tx descriptor ring for a specific queue
2096  *
2097  * Free all transmit software resources
2098  **/
2099
2100 static void
2101 e1000_free_tx_resources(struct e1000_adapter *adapter,
2102                         struct e1000_tx_ring *tx_ring)
2103 {
2104         struct pci_dev *pdev = adapter->pdev;
2105
2106         e1000_clean_tx_ring(adapter, tx_ring);
2107
2108         vfree(tx_ring->buffer_info);
2109         tx_ring->buffer_info = NULL;
2110
2111         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
2112
2113         tx_ring->desc = NULL;
2114 }
2115
2116 /**
2117  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
2118  * @adapter: board private structure
2119  *
2120  * Free all transmit software resources
2121  **/
2122
2123 void
2124 e1000_free_all_tx_resources(struct e1000_adapter *adapter)
2125 {
2126         int i;
2127
2128         for (i = 0; i < adapter->num_tx_queues; i++)
2129                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
2130 }
2131
2132 static void
2133 e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
2134                         struct e1000_buffer *buffer_info)
2135 {
2136         if (buffer_info->dma) {
2137                 pci_unmap_page(adapter->pdev,
2138                                 buffer_info->dma,
2139                                 buffer_info->length,
2140                                 PCI_DMA_TODEVICE);
2141                 buffer_info->dma = 0;
2142         }
2143         if (buffer_info->skb) {
2144                 dev_kfree_skb_any(buffer_info->skb);
2145                 buffer_info->skb = NULL;
2146         }
2147         /* buffer_info must be completely set up in the transmit path */
2148 }
2149
2150 /**
2151  * e1000_clean_tx_ring - Free Tx Buffers
2152  * @adapter: board private structure
2153  * @tx_ring: ring to be cleaned
2154  **/
2155
2156 static void
2157 e1000_clean_tx_ring(struct e1000_adapter *adapter,
2158                     struct e1000_tx_ring *tx_ring)
2159 {
2160         struct e1000_buffer *buffer_info;
2161         unsigned long size;
2162         unsigned int i;
2163
2164         /* Free all the Tx ring sk_buffs */
2165
2166         for (i = 0; i < tx_ring->count; i++) {
2167                 buffer_info = &tx_ring->buffer_info[i];
2168                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2169         }
2170
2171         size = sizeof(struct e1000_buffer) * tx_ring->count;
2172         memset(tx_ring->buffer_info, 0, size);
2173
2174         /* Zero out the descriptor ring */
2175
2176         memset(tx_ring->desc, 0, tx_ring->size);
2177
2178         tx_ring->next_to_use = 0;
2179         tx_ring->next_to_clean = 0;
2180         tx_ring->last_tx_tso = 0;
2181
2182         writel(0, adapter->hw.hw_addr + tx_ring->tdh);
2183         writel(0, adapter->hw.hw_addr + tx_ring->tdt);
2184 }
2185
2186 /**
2187  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
2188  * @adapter: board private structure
2189  **/
2190
2191 static void
2192 e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
2193 {
2194         int i;
2195
2196         for (i = 0; i < adapter->num_tx_queues; i++)
2197                 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2198 }
2199
2200 /**
2201  * e1000_free_rx_resources - Free Rx Resources
2202  * @adapter: board private structure
2203  * @rx_ring: ring to clean the resources from
2204  *
2205  * Free all receive software resources
2206  **/
2207
2208 static void
2209 e1000_free_rx_resources(struct e1000_adapter *adapter,
2210                         struct e1000_rx_ring *rx_ring)
2211 {
2212         struct pci_dev *pdev = adapter->pdev;
2213
2214         e1000_clean_rx_ring(adapter, rx_ring);
2215
2216         vfree(rx_ring->buffer_info);
2217         rx_ring->buffer_info = NULL;
2218         kfree(rx_ring->ps_page);
2219         rx_ring->ps_page = NULL;
2220         kfree(rx_ring->ps_page_dma);
2221         rx_ring->ps_page_dma = NULL;
2222
2223         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2224
2225         rx_ring->desc = NULL;
2226 }
2227
2228 /**
2229  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
2230  * @adapter: board private structure
2231  *
2232  * Free all receive software resources
2233  **/
2234
2235 void
2236 e1000_free_all_rx_resources(struct e1000_adapter *adapter)
2237 {
2238         int i;
2239
2240         for (i = 0; i < adapter->num_rx_queues; i++)
2241                 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
2242 }
2243
2244 /**
2245  * e1000_clean_rx_ring - Free Rx Buffers per Queue
2246  * @adapter: board private structure
2247  * @rx_ring: ring to free buffers from
2248  **/
2249
2250 static void
2251 e1000_clean_rx_ring(struct e1000_adapter *adapter,
2252                     struct e1000_rx_ring *rx_ring)
2253 {
2254         struct e1000_buffer *buffer_info;
2255         struct e1000_ps_page *ps_page;
2256         struct e1000_ps_page_dma *ps_page_dma;
2257         struct pci_dev *pdev = adapter->pdev;
2258         unsigned long size;
2259         unsigned int i, j;
2260
2261         /* Free all the Rx ring sk_buffs */
2262         for (i = 0; i < rx_ring->count; i++) {
2263                 buffer_info = &rx_ring->buffer_info[i];
2264                 if (buffer_info->skb) {
2265                         pci_unmap_single(pdev,
2266                                          buffer_info->dma,
2267                                          buffer_info->length,
2268                                          PCI_DMA_FROMDEVICE);
2269
2270                         dev_kfree_skb(buffer_info->skb);
2271                         buffer_info->skb = NULL;
2272                 }
2273                 ps_page = &rx_ring->ps_page[i];
2274                 ps_page_dma = &rx_ring->ps_page_dma[i];
2275                 for (j = 0; j < adapter->rx_ps_pages; j++) {
2276                         if (!ps_page->ps_page[j]) break;
2277                         pci_unmap_page(pdev,
2278                                        ps_page_dma->ps_page_dma[j],
2279                                        PAGE_SIZE, PCI_DMA_FROMDEVICE);
2280                         ps_page_dma->ps_page_dma[j] = 0;
2281                         put_page(ps_page->ps_page[j]);
2282                         ps_page->ps_page[j] = NULL;
2283                 }
2284         }
2285
2286         size = sizeof(struct e1000_buffer) * rx_ring->count;
2287         memset(rx_ring->buffer_info, 0, size);
2288         size = sizeof(struct e1000_ps_page) * rx_ring->count;
2289         memset(rx_ring->ps_page, 0, size);
2290         size = sizeof(struct e1000_ps_page_dma) * rx_ring->count;
2291         memset(rx_ring->ps_page_dma, 0, size);
2292
2293         /* Zero out the descriptor ring */
2294
2295         memset(rx_ring->desc, 0, rx_ring->size);
2296
2297         rx_ring->next_to_clean = 0;
2298         rx_ring->next_to_use = 0;
2299
2300         writel(0, adapter->hw.hw_addr + rx_ring->rdh);
2301         writel(0, adapter->hw.hw_addr + rx_ring->rdt);
2302 }
2303
2304 /**
2305  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2306  * @adapter: board private structure
2307  **/
2308
2309 static void
2310 e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2311 {
2312         int i;
2313
2314         for (i = 0; i < adapter->num_rx_queues; i++)
2315                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2316 }
2317
2318 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2319  * and memory write and invalidate disabled for certain operations
2320  */
2321 static void
2322 e1000_enter_82542_rst(struct e1000_adapter *adapter)
2323 {
2324         struct net_device *netdev = adapter->netdev;
2325         uint32_t rctl;
2326
2327         e1000_pci_clear_mwi(&adapter->hw);
2328
2329         rctl = E1000_READ_REG(&adapter->hw, RCTL);
2330         rctl |= E1000_RCTL_RST;
2331         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
2332         E1000_WRITE_FLUSH(&adapter->hw);
2333         mdelay(5);
2334
2335         if (netif_running(netdev))
2336                 e1000_clean_all_rx_rings(adapter);
2337 }
2338
2339 static void
2340 e1000_leave_82542_rst(struct e1000_adapter *adapter)
2341 {
2342         struct net_device *netdev = adapter->netdev;
2343         uint32_t rctl;
2344
2345         rctl = E1000_READ_REG(&adapter->hw, RCTL);
2346         rctl &= ~E1000_RCTL_RST;
2347         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
2348         E1000_WRITE_FLUSH(&adapter->hw);
2349         mdelay(5);
2350
2351         if (adapter->hw.pci_cmd_word & PCI_COMMAND_INVALIDATE)
2352                 e1000_pci_set_mwi(&adapter->hw);
2353
2354         if (netif_running(netdev)) {
2355                 /* No need to loop, because 82542 supports only 1 queue */
2356                 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2357                 e1000_configure_rx(adapter);
2358                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2359         }
2360 }
2361
2362 /**
2363  * e1000_set_mac - Change the Ethernet Address of the NIC
2364  * @netdev: network interface device structure
2365  * @p: pointer to an address structure
2366  *
2367  * Returns 0 on success, negative on failure
2368  **/
2369
2370 static int
2371 e1000_set_mac(struct net_device *netdev, void *p)
2372 {
2373         struct e1000_adapter *adapter = netdev_priv(netdev);
2374         struct sockaddr *addr = p;
2375
2376         if (!is_valid_ether_addr(addr->sa_data))
2377                 return -EADDRNOTAVAIL;
2378
2379         /* 82542 2.0 needs to be in reset to write receive address registers */
2380
2381         if (adapter->hw.mac_type == e1000_82542_rev2_0)
2382                 e1000_enter_82542_rst(adapter);
2383
2384         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2385         memcpy(adapter->hw.mac_addr, addr->sa_data, netdev->addr_len);
2386
2387         e1000_rar_set(&adapter->hw, adapter->hw.mac_addr, 0);
2388
2389         /* With 82571 controllers, LAA may be overwritten (with the default)
2390          * due to controller reset from the other port. */
2391         if (adapter->hw.mac_type == e1000_82571) {
2392                 /* activate the work around */
2393                 adapter->hw.laa_is_present = 1;
2394
2395                 /* Hold a copy of the LAA in RAR[14] This is done so that
2396                  * between the time RAR[0] gets clobbered  and the time it
2397                  * gets fixed (in e1000_watchdog), the actual LAA is in one
2398                  * of the RARs and no incoming packets directed to this port
2399                  * are dropped. Eventaully the LAA will be in RAR[0] and
2400                  * RAR[14] */
2401                 e1000_rar_set(&adapter->hw, adapter->hw.mac_addr,
2402                                         E1000_RAR_ENTRIES - 1);
2403         }
2404
2405         if (adapter->hw.mac_type == e1000_82542_rev2_0)
2406                 e1000_leave_82542_rst(adapter);
2407
2408         return 0;
2409 }
2410
2411 /**
2412  * e1000_set_multi - Multicast and Promiscuous mode set
2413  * @netdev: network interface device structure
2414  *
2415  * The set_multi entry point is called whenever the multicast address
2416  * list or the network interface flags are updated.  This routine is
2417  * responsible for configuring the hardware for proper multicast,
2418  * promiscuous mode, and all-multi behavior.
2419  **/
2420
2421 static void
2422 e1000_set_multi(struct net_device *netdev)
2423 {
2424         struct e1000_adapter *adapter = netdev_priv(netdev);
2425         struct e1000_hw *hw = &adapter->hw;
2426         struct dev_mc_list *mc_ptr;
2427         uint32_t rctl;
2428         uint32_t hash_value;
2429         int i, rar_entries = E1000_RAR_ENTRIES;
2430         int mta_reg_count = (hw->mac_type == e1000_ich8lan) ?
2431                                 E1000_NUM_MTA_REGISTERS_ICH8LAN :
2432                                 E1000_NUM_MTA_REGISTERS;
2433
2434         if (adapter->hw.mac_type == e1000_ich8lan)
2435                 rar_entries = E1000_RAR_ENTRIES_ICH8LAN;
2436
2437         /* reserve RAR[14] for LAA over-write work-around */
2438         if (adapter->hw.mac_type == e1000_82571)
2439                 rar_entries--;
2440
2441         /* Check for Promiscuous and All Multicast modes */
2442
2443         rctl = E1000_READ_REG(hw, RCTL);
2444
2445         if (netdev->flags & IFF_PROMISC) {
2446                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2447         } else if (netdev->flags & IFF_ALLMULTI) {
2448                 rctl |= E1000_RCTL_MPE;
2449                 rctl &= ~E1000_RCTL_UPE;
2450         } else {
2451                 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2452         }
2453
2454         E1000_WRITE_REG(hw, RCTL, rctl);
2455
2456         /* 82542 2.0 needs to be in reset to write receive address registers */
2457
2458         if (hw->mac_type == e1000_82542_rev2_0)
2459                 e1000_enter_82542_rst(adapter);
2460
2461         /* load the first 14 multicast address into the exact filters 1-14
2462          * RAR 0 is used for the station MAC adddress
2463          * if there are not 14 addresses, go ahead and clear the filters
2464          * -- with 82571 controllers only 0-13 entries are filled here
2465          */
2466         mc_ptr = netdev->mc_list;
2467
2468         for (i = 1; i < rar_entries; i++) {
2469                 if (mc_ptr) {
2470                         e1000_rar_set(hw, mc_ptr->dmi_addr, i);
2471                         mc_ptr = mc_ptr->next;
2472                 } else {
2473                         E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2474                         E1000_WRITE_FLUSH(hw);
2475                         E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2476                         E1000_WRITE_FLUSH(hw);
2477                 }
2478         }
2479
2480         /* clear the old settings from the multicast hash table */
2481
2482         for (i = 0; i < mta_reg_count; i++) {
2483                 E1000_WRITE_REG_ARRAY(hw, MTA, i, 0);
2484                 E1000_WRITE_FLUSH(hw);
2485         }
2486
2487         /* load any remaining addresses into the hash table */
2488
2489         for (; mc_ptr; mc_ptr = mc_ptr->next) {
2490                 hash_value = e1000_hash_mc_addr(hw, mc_ptr->dmi_addr);
2491                 e1000_mta_set(hw, hash_value);
2492         }
2493
2494         if (hw->mac_type == e1000_82542_rev2_0)
2495                 e1000_leave_82542_rst(adapter);
2496 }
2497
2498 /* Need to wait a few seconds after link up to get diagnostic information from
2499  * the phy */
2500
2501 static void
2502 e1000_update_phy_info(unsigned long data)
2503 {
2504         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2505         e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
2506 }
2507
2508 /**
2509  * e1000_82547_tx_fifo_stall - Timer Call-back
2510  * @data: pointer to adapter cast into an unsigned long
2511  **/
2512
2513 static void
2514 e1000_82547_tx_fifo_stall(unsigned long data)
2515 {
2516         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2517         struct net_device *netdev = adapter->netdev;
2518         uint32_t tctl;
2519
2520         if (atomic_read(&adapter->tx_fifo_stall)) {
2521                 if ((E1000_READ_REG(&adapter->hw, TDT) ==
2522                     E1000_READ_REG(&adapter->hw, TDH)) &&
2523                    (E1000_READ_REG(&adapter->hw, TDFT) ==
2524                     E1000_READ_REG(&adapter->hw, TDFH)) &&
2525                    (E1000_READ_REG(&adapter->hw, TDFTS) ==
2526                     E1000_READ_REG(&adapter->hw, TDFHS))) {
2527                         tctl = E1000_READ_REG(&adapter->hw, TCTL);
2528                         E1000_WRITE_REG(&adapter->hw, TCTL,
2529                                         tctl & ~E1000_TCTL_EN);
2530                         E1000_WRITE_REG(&adapter->hw, TDFT,
2531                                         adapter->tx_head_addr);
2532                         E1000_WRITE_REG(&adapter->hw, TDFH,
2533                                         adapter->tx_head_addr);
2534                         E1000_WRITE_REG(&adapter->hw, TDFTS,
2535                                         adapter->tx_head_addr);
2536                         E1000_WRITE_REG(&adapter->hw, TDFHS,
2537                                         adapter->tx_head_addr);
2538                         E1000_WRITE_REG(&adapter->hw, TCTL, tctl);
2539                         E1000_WRITE_FLUSH(&adapter->hw);
2540
2541                         adapter->tx_fifo_head = 0;
2542                         atomic_set(&adapter->tx_fifo_stall, 0);
2543                         netif_wake_queue(netdev);
2544                 } else {
2545                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2546                 }
2547         }
2548 }
2549
2550 /**
2551  * e1000_watchdog - Timer Call-back
2552  * @data: pointer to adapter cast into an unsigned long
2553  **/
2554 static void
2555 e1000_watchdog(unsigned long data)
2556 {
2557         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2558         struct net_device *netdev = adapter->netdev;
2559         struct e1000_tx_ring *txdr = adapter->tx_ring;
2560         uint32_t link, tctl;
2561         int32_t ret_val;
2562
2563         ret_val = e1000_check_for_link(&adapter->hw);
2564         if ((ret_val == E1000_ERR_PHY) &&
2565             (adapter->hw.phy_type == e1000_phy_igp_3) &&
2566             (E1000_READ_REG(&adapter->hw, CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
2567                 /* See e1000_kumeran_lock_loss_workaround() */
2568                 DPRINTK(LINK, INFO,
2569                         "Gigabit has been disabled, downgrading speed\n");
2570         }
2571
2572         if (adapter->hw.mac_type == e1000_82573) {
2573                 e1000_enable_tx_pkt_filtering(&adapter->hw);
2574                 if (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id)
2575                         e1000_update_mng_vlan(adapter);
2576         }
2577
2578         if ((adapter->hw.media_type == e1000_media_type_internal_serdes) &&
2579            !(E1000_READ_REG(&adapter->hw, TXCW) & E1000_TXCW_ANE))
2580                 link = !adapter->hw.serdes_link_down;
2581         else
2582                 link = E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_LU;
2583
2584         if (link) {
2585                 if (!netif_carrier_ok(netdev)) {
2586                         uint32_t ctrl;
2587                         boolean_t txb2b = 1;
2588                         e1000_get_speed_and_duplex(&adapter->hw,
2589                                                    &adapter->link_speed,
2590                                                    &adapter->link_duplex);
2591
2592                         ctrl = E1000_READ_REG(&adapter->hw, CTRL);
2593                         DPRINTK(LINK, INFO, "NIC Link is Up %d Mbps %s, "
2594                                 "Flow Control: %s\n",
2595                                 adapter->link_speed,
2596                                 adapter->link_duplex == FULL_DUPLEX ?
2597                                 "Full Duplex" : "Half Duplex",
2598                                 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2599                                 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2600                                 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2601                                 E1000_CTRL_TFCE) ? "TX" : "None" )));
2602
2603                         /* tweak tx_queue_len according to speed/duplex
2604                          * and adjust the timeout factor */
2605                         netdev->tx_queue_len = adapter->tx_queue_len;
2606                         adapter->tx_timeout_factor = 1;
2607                         switch (adapter->link_speed) {
2608                         case SPEED_10:
2609                                 txb2b = 0;
2610                                 netdev->tx_queue_len = 10;
2611                                 adapter->tx_timeout_factor = 8;
2612                                 break;
2613                         case SPEED_100:
2614                                 txb2b = 0;
2615                                 netdev->tx_queue_len = 100;
2616                                 /* maybe add some timeout factor ? */
2617                                 break;
2618                         }
2619
2620                         if ((adapter->hw.mac_type == e1000_82571 ||
2621                              adapter->hw.mac_type == e1000_82572) &&
2622                             txb2b == 0) {
2623                                 uint32_t tarc0;
2624                                 tarc0 = E1000_READ_REG(&adapter->hw, TARC0);
2625                                 tarc0 &= ~(1 << 21);
2626                                 E1000_WRITE_REG(&adapter->hw, TARC0, tarc0);
2627                         }
2628
2629 #ifdef NETIF_F_TSO
2630                         /* disable TSO for pcie and 10/100 speeds, to avoid
2631                          * some hardware issues */
2632                         if (!adapter->tso_force &&
2633                             adapter->hw.bus_type == e1000_bus_type_pci_express){
2634                                 switch (adapter->link_speed) {
2635                                 case SPEED_10:
2636                                 case SPEED_100:
2637                                         DPRINTK(PROBE,INFO,
2638                                         "10/100 speed: disabling TSO\n");
2639                                         netdev->features &= ~NETIF_F_TSO;
2640 #ifdef NETIF_F_TSO6
2641                                         netdev->features &= ~NETIF_F_TSO6;
2642 #endif
2643                                         break;
2644                                 case SPEED_1000:
2645                                         netdev->features |= NETIF_F_TSO;
2646 #ifdef NETIF_F_TSO6
2647                                         netdev->features |= NETIF_F_TSO6;
2648 #endif
2649                                         break;
2650                                 default:
2651                                         /* oops */
2652                                         break;
2653                                 }
2654                         }
2655 #endif
2656
2657                         /* enable transmits in the hardware, need to do this
2658                          * after setting TARC0 */
2659                         tctl = E1000_READ_REG(&adapter->hw, TCTL);
2660                         tctl |= E1000_TCTL_EN;
2661                         E1000_WRITE_REG(&adapter->hw, TCTL, tctl);
2662
2663                         netif_carrier_on(netdev);
2664                         netif_wake_queue(netdev);
2665                         mod_timer(&adapter->phy_info_timer, jiffies + 2 * HZ);
2666                         adapter->smartspeed = 0;
2667                 } else {
2668                         /* make sure the receive unit is started */
2669                         if (adapter->hw.rx_needs_kicking) {
2670                                 struct e1000_hw *hw = &adapter->hw;
2671                                 uint32_t rctl = E1000_READ_REG(hw, RCTL);
2672                                 E1000_WRITE_REG(hw, RCTL, rctl | E1000_RCTL_EN);
2673                         }
2674                 }
2675         } else {
2676                 if (netif_carrier_ok(netdev)) {
2677                         adapter->link_speed = 0;
2678                         adapter->link_duplex = 0;
2679                         DPRINTK(LINK, INFO, "NIC Link is Down\n");
2680                         netif_carrier_off(netdev);
2681                         netif_stop_queue(netdev);
2682                         mod_timer(&adapter->phy_info_timer, jiffies + 2 * HZ);
2683
2684                         /* 80003ES2LAN workaround--
2685                          * For packet buffer work-around on link down event;
2686                          * disable receives in the ISR and
2687                          * reset device here in the watchdog
2688                          */
2689                         if (adapter->hw.mac_type == e1000_80003es2lan)
2690                                 /* reset device */
2691                                 schedule_work(&adapter->reset_task);
2692                 }
2693
2694                 e1000_smartspeed(adapter);
2695         }
2696
2697         e1000_update_stats(adapter);
2698
2699         adapter->hw.tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2700         adapter->tpt_old = adapter->stats.tpt;
2701         adapter->hw.collision_delta = adapter->stats.colc - adapter->colc_old;
2702         adapter->colc_old = adapter->stats.colc;
2703
2704         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2705         adapter->gorcl_old = adapter->stats.gorcl;
2706         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2707         adapter->gotcl_old = adapter->stats.gotcl;
2708
2709         e1000_update_adaptive(&adapter->hw);
2710
2711         if (!netif_carrier_ok(netdev)) {
2712                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2713                         /* We've lost link, so the controller stops DMA,
2714                          * but we've got queued Tx work that's never going
2715                          * to get done, so reset controller to flush Tx.
2716                          * (Do the reset outside of interrupt context). */
2717                         adapter->tx_timeout_count++;
2718                         schedule_work(&adapter->reset_task);
2719                 }
2720         }
2721
2722         /* Cause software interrupt to ensure rx ring is cleaned */
2723         E1000_WRITE_REG(&adapter->hw, ICS, E1000_ICS_RXDMT0);
2724
2725         /* Force detection of hung controller every watchdog period */
2726         adapter->detect_tx_hung = TRUE;
2727
2728         /* With 82571 controllers, LAA may be overwritten due to controller
2729          * reset from the other port. Set the appropriate LAA in RAR[0] */
2730         if (adapter->hw.mac_type == e1000_82571 && adapter->hw.laa_is_present)
2731                 e1000_rar_set(&adapter->hw, adapter->hw.mac_addr, 0);
2732
2733         /* Reset the timer */
2734         mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
2735 }
2736
2737 enum latency_range {
2738         lowest_latency = 0,
2739         low_latency = 1,
2740         bulk_latency = 2,
2741         latency_invalid = 255
2742 };
2743
2744 /**
2745  * e1000_update_itr - update the dynamic ITR value based on statistics
2746  *      Stores a new ITR value based on packets and byte
2747  *      counts during the last interrupt.  The advantage of per interrupt
2748  *      computation is faster updates and more accurate ITR for the current
2749  *      traffic pattern.  Constants in this function were computed
2750  *      based on theoretical maximum wire speed and thresholds were set based
2751  *      on testing data as well as attempting to minimize response time
2752  *      while increasing bulk throughput.
2753  *      this functionality is controlled by the InterruptThrottleRate module
2754  *      parameter (see e1000_param.c)
2755  * @adapter: pointer to adapter
2756  * @itr_setting: current adapter->itr
2757  * @packets: the number of packets during this measurement interval
2758  * @bytes: the number of bytes during this measurement interval
2759  **/
2760 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2761                                    uint16_t itr_setting,
2762                                    int packets,
2763                                    int bytes)
2764 {
2765         unsigned int retval = itr_setting;
2766         struct e1000_hw *hw = &adapter->hw;
2767
2768         if (unlikely(hw->mac_type < e1000_82540))
2769                 goto update_itr_done;
2770
2771         if (packets == 0)
2772                 goto update_itr_done;
2773
2774         switch (itr_setting) {
2775         case lowest_latency:
2776                 /* jumbo frames get bulk treatment*/
2777                 if (bytes/packets > 8000)
2778                         retval = bulk_latency;
2779                 else if ((packets < 5) && (bytes > 512))
2780                         retval = low_latency;
2781                 break;
2782         case low_latency:  /* 50 usec aka 20000 ints/s */
2783                 if (bytes > 10000) {
2784                         /* jumbo frames need bulk latency setting */
2785                         if (bytes/packets > 8000)
2786                                 retval = bulk_latency;
2787                         else if ((packets < 10) || ((bytes/packets) > 1200))
2788                                 retval = bulk_latency;
2789                         else if ((packets > 35))
2790                                 retval = lowest_latency;
2791                 } else if (bytes/packets > 2000)
2792                         retval = bulk_latency;
2793                 else if (packets <= 2 && bytes < 512)
2794                         retval = lowest_latency;
2795                 break;
2796         case bulk_latency: /* 250 usec aka 4000 ints/s */
2797                 if (bytes > 25000) {
2798                         if (packets > 35)
2799                                 retval = low_latency;
2800                 } else if (bytes < 6000) {
2801                         retval = low_latency;
2802                 }
2803                 break;
2804         }
2805
2806 update_itr_done:
2807         return retval;
2808 }
2809
2810 static void e1000_set_itr(struct e1000_adapter *adapter)
2811 {
2812         struct e1000_hw *hw = &adapter->hw;
2813         uint16_t current_itr;
2814         uint32_t new_itr = adapter->itr;
2815
2816         if (unlikely(hw->mac_type < e1000_82540))
2817                 return;
2818
2819         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2820         if (unlikely(adapter->link_speed != SPEED_1000)) {
2821                 current_itr = 0;
2822                 new_itr = 4000;
2823                 goto set_itr_now;
2824         }
2825
2826         adapter->tx_itr = e1000_update_itr(adapter,
2827                                     adapter->tx_itr,
2828                                     adapter->total_tx_packets,
2829                                     adapter->total_tx_bytes);
2830         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2831         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2832                 adapter->tx_itr = low_latency;
2833
2834         adapter->rx_itr = e1000_update_itr(adapter,
2835                                     adapter->rx_itr,
2836                                     adapter->total_rx_packets,
2837                                     adapter->total_rx_bytes);
2838         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2839         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2840                 adapter->rx_itr = low_latency;
2841
2842         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2843
2844         switch (current_itr) {
2845         /* counts and packets in update_itr are dependent on these numbers */
2846         case lowest_latency:
2847                 new_itr = 70000;
2848                 break;
2849         case low_latency:
2850                 new_itr = 20000; /* aka hwitr = ~200 */
2851                 break;
2852         case bulk_latency:
2853                 new_itr = 4000;
2854                 break;
2855         default:
2856                 break;
2857         }
2858
2859 set_itr_now:
2860         if (new_itr != adapter->itr) {
2861                 /* this attempts to bias the interrupt rate towards Bulk
2862                  * by adding intermediate steps when interrupt rate is
2863                  * increasing */
2864                 new_itr = new_itr > adapter->itr ?
2865                              min(adapter->itr + (new_itr >> 2), new_itr) :
2866                              new_itr;
2867                 adapter->itr = new_itr;
2868                 E1000_WRITE_REG(hw, ITR, 1000000000 / (new_itr * 256));
2869         }
2870
2871         return;
2872 }
2873
2874 #define E1000_TX_FLAGS_CSUM             0x00000001
2875 #define E1000_TX_FLAGS_VLAN             0x00000002
2876 #define E1000_TX_FLAGS_TSO              0x00000004
2877 #define E1000_TX_FLAGS_IPV4             0x00000008
2878 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2879 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2880
2881 static int
2882 e1000_tso(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2883           struct sk_buff *skb)
2884 {
2885 #ifdef NETIF_F_TSO
2886         struct e1000_context_desc *context_desc;
2887         struct e1000_buffer *buffer_info;
2888         unsigned int i;
2889         uint32_t cmd_length = 0;
2890         uint16_t ipcse = 0, tucse, mss;
2891         uint8_t ipcss, ipcso, tucss, tucso, hdr_len;
2892         int err;
2893
2894         if (skb_is_gso(skb)) {
2895                 if (skb_header_cloned(skb)) {
2896                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2897                         if (err)
2898                                 return err;
2899                 }
2900
2901                 hdr_len = ((skb->h.raw - skb->data) + (skb->h.th->doff << 2));
2902                 mss = skb_shinfo(skb)->gso_size;
2903                 if (skb->protocol == htons(ETH_P_IP)) {
2904                         skb->nh.iph->tot_len = 0;
2905                         skb->nh.iph->check = 0;
2906                         skb->h.th->check =
2907                                 ~csum_tcpudp_magic(skb->nh.iph->saddr,
2908                                                    skb->nh.iph->daddr,
2909                                                    0,
2910                                                    IPPROTO_TCP,
2911                                                    0);
2912                         cmd_length = E1000_TXD_CMD_IP;
2913                         ipcse = skb->h.raw - skb->data - 1;
2914 #ifdef NETIF_F_TSO6
2915                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2916                         skb->nh.ipv6h->payload_len = 0;
2917                         skb->h.th->check =
2918                                 ~csum_ipv6_magic(&skb->nh.ipv6h->saddr,
2919                                                  &skb->nh.ipv6h->daddr,
2920                                                  0,
2921                                                  IPPROTO_TCP,
2922                                                  0);
2923                         ipcse = 0;
2924 #endif
2925                 }
2926                 ipcss = skb->nh.raw - skb->data;
2927                 ipcso = (void *)&(skb->nh.iph->check) - (void *)skb->data;
2928                 tucss = skb->h.raw - skb->data;
2929                 tucso = (void *)&(skb->h.th->check) - (void *)skb->data;
2930                 tucse = 0;
2931
2932                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2933                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2934
2935                 i = tx_ring->next_to_use;
2936                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2937                 buffer_info = &tx_ring->buffer_info[i];
2938
2939                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2940                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2941                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2942                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2943                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2944                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2945                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2946                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2947                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2948
2949                 buffer_info->time_stamp = jiffies;
2950                 buffer_info->next_to_watch = i;
2951
2952                 if (++i == tx_ring->count) i = 0;
2953                 tx_ring->next_to_use = i;
2954
2955                 return TRUE;
2956         }
2957 #endif
2958
2959         return FALSE;
2960 }
2961
2962 static boolean_t
2963 e1000_tx_csum(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2964               struct sk_buff *skb)
2965 {
2966         struct e1000_context_desc *context_desc;
2967         struct e1000_buffer *buffer_info;
2968         unsigned int i;
2969         uint8_t css;
2970
2971         if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
2972                 css = skb->h.raw - skb->data;
2973
2974                 i = tx_ring->next_to_use;
2975                 buffer_info = &tx_ring->buffer_info[i];
2976                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2977
2978                 context_desc->lower_setup.ip_config = 0;
2979                 context_desc->upper_setup.tcp_fields.tucss = css;
2980                 context_desc->upper_setup.tcp_fields.tucso = css + skb->csum;
2981                 context_desc->upper_setup.tcp_fields.tucse = 0;
2982                 context_desc->tcp_seg_setup.data = 0;
2983                 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
2984
2985                 buffer_info->time_stamp = jiffies;
2986                 buffer_info->next_to_watch = i;
2987
2988                 if (unlikely(++i == tx_ring->count)) i = 0;
2989                 tx_ring->next_to_use = i;
2990
2991                 return TRUE;
2992         }
2993
2994         return FALSE;
2995 }
2996
2997 #define E1000_MAX_TXD_PWR       12
2998 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2999
3000 static int
3001 e1000_tx_map(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
3002              struct sk_buff *skb, unsigned int first, unsigned int max_per_txd,
3003              unsigned int nr_frags, unsigned int mss)
3004 {
3005         struct e1000_buffer *buffer_info;
3006         unsigned int len = skb->len;
3007         unsigned int offset = 0, size, count = 0, i;
3008         unsigned int f;
3009         len -= skb->data_len;
3010
3011         i = tx_ring->next_to_use;
3012
3013         while (len) {
3014                 buffer_info = &tx_ring->buffer_info[i];
3015                 size = min(len, max_per_txd);
3016 #ifdef NETIF_F_TSO
3017                 /* Workaround for Controller erratum --
3018                  * descriptor for non-tso packet in a linear SKB that follows a
3019                  * tso gets written back prematurely before the data is fully
3020                  * DMA'd to the controller */
3021                 if (!skb->data_len && tx_ring->last_tx_tso &&
3022                     !skb_is_gso(skb)) {
3023                         tx_ring->last_tx_tso = 0;
3024                         size -= 4;
3025                 }
3026
3027                 /* Workaround for premature desc write-backs
3028                  * in TSO mode.  Append 4-byte sentinel desc */
3029                 if (unlikely(mss && !nr_frags && size == len && size > 8))
3030                         size -= 4;
3031 #endif
3032                 /* work-around for errata 10 and it applies
3033                  * to all controllers in PCI-X mode
3034                  * The fix is to make sure that the first descriptor of a
3035                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
3036                  */
3037                 if (unlikely((adapter->hw.bus_type == e1000_bus_type_pcix) &&
3038                                 (size > 2015) && count == 0))
3039                         size = 2015;
3040
3041                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
3042                  * terminating buffers within evenly-aligned dwords. */
3043                 if (unlikely(adapter->pcix_82544 &&
3044                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
3045                    size > 4))
3046                         size -= 4;
3047
3048                 buffer_info->length = size;
3049                 buffer_info->dma =
3050                         pci_map_single(adapter->pdev,
3051                                 skb->data + offset,
3052                                 size,
3053                                 PCI_DMA_TODEVICE);
3054                 buffer_info->time_stamp = jiffies;
3055                 buffer_info->next_to_watch = i;
3056
3057                 len -= size;
3058                 offset += size;
3059                 count++;
3060                 if (unlikely(++i == tx_ring->count)) i = 0;
3061         }
3062
3063         for (f = 0; f < nr_frags; f++) {
3064                 struct skb_frag_struct *frag;
3065
3066                 frag = &skb_shinfo(skb)->frags[f];
3067                 len = frag->size;
3068                 offset = frag->page_offset;
3069
3070                 while (len) {
3071                         buffer_info = &tx_ring->buffer_info[i];
3072                         size = min(len, max_per_txd);
3073 #ifdef NETIF_F_TSO
3074                         /* Workaround for premature desc write-backs
3075                          * in TSO mode.  Append 4-byte sentinel desc */
3076                         if (unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
3077                                 size -= 4;
3078 #endif
3079                         /* Workaround for potential 82544 hang in PCI-X.
3080                          * Avoid terminating buffers within evenly-aligned
3081                          * dwords. */
3082                         if (unlikely(adapter->pcix_82544 &&
3083                            !((unsigned long)(frag->page+offset+size-1) & 4) &&
3084                            size > 4))
3085                                 size -= 4;
3086
3087                         buffer_info->length = size;
3088                         buffer_info->dma =
3089                                 pci_map_page(adapter->pdev,
3090                                         frag->page,
3091                                         offset,
3092                                         size,
3093                                         PCI_DMA_TODEVICE);
3094                         buffer_info->time_stamp = jiffies;
3095                         buffer_info->next_to_watch = i;
3096
3097                         len -= size;
3098                         offset += size;
3099                         count++;
3100                         if (unlikely(++i == tx_ring->count)) i = 0;
3101                 }
3102         }
3103
3104         i = (i == 0) ? tx_ring->count - 1 : i - 1;
3105         tx_ring->buffer_info[i].skb = skb;
3106         tx_ring->buffer_info[first].next_to_watch = i;
3107
3108         return count;
3109 }
3110
3111 static void
3112 e1000_tx_queue(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
3113                int tx_flags, int count)
3114 {
3115         struct e1000_tx_desc *tx_desc = NULL;
3116         struct e1000_buffer *buffer_info;
3117         uint32_t txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3118         unsigned int i;
3119
3120         if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
3121                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3122                              E1000_TXD_CMD_TSE;
3123                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3124
3125                 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
3126                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3127         }
3128
3129         if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
3130                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3131                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3132         }
3133
3134         if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
3135                 txd_lower |= E1000_TXD_CMD_VLE;
3136                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3137         }
3138
3139         i = tx_ring->next_to_use;
3140
3141         while (count--) {
3142                 buffer_info = &tx_ring->buffer_info[i];
3143                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3144                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3145                 tx_desc->lower.data =
3146                         cpu_to_le32(txd_lower | buffer_info->length);
3147                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3148                 if (unlikely(++i == tx_ring->count)) i = 0;
3149         }
3150
3151         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3152
3153         /* Force memory writes to complete before letting h/w
3154          * know there are new descriptors to fetch.  (Only
3155          * applicable for weak-ordered memory model archs,
3156          * such as IA-64). */
3157         wmb();
3158
3159         tx_ring->next_to_use = i;
3160         writel(i, adapter->hw.hw_addr + tx_ring->tdt);
3161         /* we need this if more than one processor can write to our tail
3162          * at a time, it syncronizes IO on IA64/Altix systems */
3163         mmiowb();
3164 }
3165
3166 /**
3167  * 82547 workaround to avoid controller hang in half-duplex environment.
3168  * The workaround is to avoid queuing a large packet that would span
3169  * the internal Tx FIFO ring boundary by notifying the stack to resend
3170  * the packet at a later time.  This gives the Tx FIFO an opportunity to
3171  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
3172  * to the beginning of the Tx FIFO.
3173  **/
3174
3175 #define E1000_FIFO_HDR                  0x10
3176 #define E1000_82547_PAD_LEN             0x3E0
3177
3178 static int
3179 e1000_82547_fifo_workaround(struct e1000_adapter *adapter, struct sk_buff *skb)
3180 {
3181         uint32_t fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
3182         uint32_t skb_fifo_len = skb->len + E1000_FIFO_HDR;
3183
3184         E1000_ROUNDUP(skb_fifo_len, E1000_FIFO_HDR);
3185
3186         if (adapter->link_duplex != HALF_DUPLEX)
3187                 goto no_fifo_stall_required;
3188
3189         if (atomic_read(&adapter->tx_fifo_stall))
3190                 return 1;
3191
3192         if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
3193                 atomic_set(&adapter->tx_fifo_stall, 1);
3194                 return 1;
3195         }
3196
3197 no_fifo_stall_required:
3198         adapter->tx_fifo_head += skb_fifo_len;
3199         if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
3200                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
3201         return 0;
3202 }
3203
3204 #define MINIMUM_DHCP_PACKET_SIZE 282
3205 static int
3206 e1000_transfer_dhcp_info(struct e1000_adapter *adapter, struct sk_buff *skb)
3207 {
3208         struct e1000_hw *hw =  &adapter->hw;
3209         uint16_t length, offset;
3210         if (vlan_tx_tag_present(skb)) {
3211                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
3212                         ( adapter->hw.mng_cookie.status &
3213                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) )
3214                         return 0;
3215         }
3216         if (skb->len > MINIMUM_DHCP_PACKET_SIZE) {
3217                 struct ethhdr *eth = (struct ethhdr *) skb->data;
3218                 if ((htons(ETH_P_IP) == eth->h_proto)) {
3219                         const struct iphdr *ip =
3220                                 (struct iphdr *)((uint8_t *)skb->data+14);
3221                         if (IPPROTO_UDP == ip->protocol) {
3222                                 struct udphdr *udp =
3223                                         (struct udphdr *)((uint8_t *)ip +
3224                                                 (ip->ihl << 2));
3225                                 if (ntohs(udp->dest) == 67) {
3226                                         offset = (uint8_t *)udp + 8 - skb->data;
3227                                         length = skb->len - offset;
3228
3229                                         return e1000_mng_write_dhcp_info(hw,
3230                                                         (uint8_t *)udp + 8,
3231                                                         length);
3232                                 }
3233                         }
3234                 }
3235         }
3236         return 0;
3237 }
3238
3239 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3240 {
3241         struct e1000_adapter *adapter = netdev_priv(netdev);
3242         struct e1000_tx_ring *tx_ring = adapter->tx_ring;
3243
3244         netif_stop_queue(netdev);
3245         /* Herbert's original patch had:
3246          *  smp_mb__after_netif_stop_queue();
3247          * but since that doesn't exist yet, just open code it. */
3248         smp_mb();
3249
3250         /* We need to check again in a case another CPU has just
3251          * made room available. */
3252         if (likely(E1000_DESC_UNUSED(tx_ring) < size))
3253                 return -EBUSY;
3254
3255         /* A reprieve! */
3256         netif_start_queue(netdev);
3257         ++adapter->restart_queue;
3258         return 0;
3259 }
3260
3261 static int e1000_maybe_stop_tx(struct net_device *netdev,
3262                                struct e1000_tx_ring *tx_ring, int size)
3263 {
3264         if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
3265                 return 0;
3266         return __e1000_maybe_stop_tx(netdev, size);
3267 }
3268
3269 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3270 static int
3271 e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3272 {
3273         struct e1000_adapter *adapter = netdev_priv(netdev);
3274         struct e1000_tx_ring *tx_ring;
3275         unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
3276         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3277         unsigned int tx_flags = 0;
3278         unsigned int len = skb->len;
3279         unsigned long flags;
3280         unsigned int nr_frags = 0;
3281         unsigned int mss = 0;
3282         int count = 0;
3283         int tso;
3284         unsigned int f;
3285         len -= skb->data_len;
3286
3287         /* This goes back to the question of how to logically map a tx queue
3288          * to a flow.  Right now, performance is impacted slightly negatively
3289          * if using multiple tx queues.  If the stack breaks away from a
3290          * single qdisc implementation, we can look at this again. */
3291         tx_ring = adapter->tx_ring;
3292
3293         if (unlikely(skb->len <= 0)) {
3294                 dev_kfree_skb_any(skb);
3295                 return NETDEV_TX_OK;
3296         }
3297
3298         /* 82571 and newer doesn't need the workaround that limited descriptor
3299          * length to 4kB */
3300         if (adapter->hw.mac_type >= e1000_82571)
3301                 max_per_txd = 8192;
3302
3303 #ifdef NETIF_F_TSO
3304         mss = skb_shinfo(skb)->gso_size;
3305         /* The controller does a simple calculation to
3306          * make sure there is enough room in the FIFO before
3307          * initiating the DMA for each buffer.  The calc is:
3308          * 4 = ceil(buffer len/mss).  To make sure we don't
3309          * overrun the FIFO, adjust the max buffer len if mss
3310          * drops. */
3311         if (mss) {
3312                 uint8_t hdr_len;
3313                 max_per_txd = min(mss << 2, max_per_txd);
3314                 max_txd_pwr = fls(max_per_txd) - 1;
3315
3316                 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
3317                 * points to just header, pull a few bytes of payload from
3318                 * frags into skb->data */
3319                 hdr_len = ((skb->h.raw - skb->data) + (skb->h.th->doff << 2));
3320                 if (skb->data_len && (hdr_len == (skb->len - skb->data_len))) {
3321                         switch (adapter->hw.mac_type) {
3322                                 unsigned int pull_size;
3323                         case e1000_82544:
3324                                 /* Make sure we have room to chop off 4 bytes,
3325                                  * and that the end alignment will work out to
3326                                  * this hardware's requirements
3327                                  * NOTE: this is a TSO only workaround
3328                                  * if end byte alignment not correct move us
3329                                  * into the next dword */
3330                                 if ((unsigned long)(skb->tail - 1) & 4)
3331                                         break;
3332                                 /* fall through */
3333                         case e1000_82571:
3334                         case e1000_82572:
3335                         case e1000_82573:
3336                         case e1000_ich8lan:
3337                                 pull_size = min((unsigned int)4, skb->data_len);
3338                                 if (!__pskb_pull_tail(skb, pull_size)) {
3339                                         DPRINTK(DRV, ERR,
3340                                                 "__pskb_pull_tail failed.\n");
3341                                         dev_kfree_skb_any(skb);
3342                                         return NETDEV_TX_OK;
3343                                 }
3344                                 len = skb->len - skb->data_len;
3345                                 break;
3346                         default:
3347                                 /* do nothing */
3348                                 break;
3349                         }
3350                 }
3351         }
3352
3353         /* reserve a descriptor for the offload context */
3354         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3355                 count++;
3356         count++;
3357 #else
3358         if (skb->ip_summed == CHECKSUM_PARTIAL)
3359                 count++;
3360 #endif
3361
3362 #ifdef NETIF_F_TSO
3363         /* Controller Erratum workaround */
3364         if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3365                 count++;
3366 #endif
3367
3368         count += TXD_USE_COUNT(len, max_txd_pwr);
3369
3370         if (adapter->pcix_82544)
3371                 count++;
3372
3373         /* work-around for errata 10 and it applies to all controllers
3374          * in PCI-X mode, so add one more descriptor to the count
3375          */
3376         if (unlikely((adapter->hw.bus_type == e1000_bus_type_pcix) &&
3377                         (len > 2015)))
3378                 count++;
3379
3380         nr_frags = skb_shinfo(skb)->nr_frags;
3381         for (f = 0; f < nr_frags; f++)
3382                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3383                                        max_txd_pwr);
3384         if (adapter->pcix_82544)
3385                 count += nr_frags;
3386
3387
3388         if (adapter->hw.tx_pkt_filtering &&
3389             (adapter->hw.mac_type == e1000_82573))
3390                 e1000_transfer_dhcp_info(adapter, skb);
3391
3392         local_irq_save(flags);
3393         if (!spin_trylock(&tx_ring->tx_lock)) {
3394                 /* Collision - tell upper layer to requeue */
3395                 local_irq_restore(flags);
3396                 return NETDEV_TX_LOCKED;
3397         }
3398
3399         /* need: count + 2 desc gap to keep tail from touching
3400          * head, otherwise try next time */
3401         if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2))) {
3402                 spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3403                 return NETDEV_TX_BUSY;
3404         }
3405
3406         if (unlikely(adapter->hw.mac_type == e1000_82547)) {
3407                 if (unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
3408                         netif_stop_queue(netdev);
3409                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
3410                         spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3411                         return NETDEV_TX_BUSY;
3412                 }
3413         }
3414
3415         if (unlikely(adapter->vlgrp && vlan_tx_tag_present(skb))) {
3416                 tx_flags |= E1000_TX_FLAGS_VLAN;
3417                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3418         }
3419
3420         first = tx_ring->next_to_use;
3421
3422         tso = e1000_tso(adapter, tx_ring, skb);
3423         if (tso < 0) {
3424                 dev_kfree_skb_any(skb);
3425                 spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3426                 return NETDEV_TX_OK;
3427         }
3428
3429         if (likely(tso)) {
3430                 tx_ring->last_tx_tso = 1;
3431                 tx_flags |= E1000_TX_FLAGS_TSO;
3432         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3433                 tx_flags |= E1000_TX_FLAGS_CSUM;
3434
3435         /* Old method was to assume IPv4 packet by default if TSO was enabled.
3436          * 82571 hardware supports TSO capabilities for IPv6 as well...
3437          * no longer assume, we must. */
3438         if (likely(skb->protocol == htons(ETH_P_IP)))
3439                 tx_flags |= E1000_TX_FLAGS_IPV4;
3440
3441         e1000_tx_queue(adapter, tx_ring, tx_flags,
3442                        e1000_tx_map(adapter, tx_ring, skb, first,
3443                                     max_per_txd, nr_frags, mss));
3444
3445         netdev->trans_start = jiffies;
3446
3447         /* Make sure there is space in the ring for the next send. */
3448         e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3449
3450         spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3451         return NETDEV_TX_OK;
3452 }
3453
3454 /**
3455  * e1000_tx_timeout - Respond to a Tx Hang
3456  * @netdev: network interface device structure
3457  **/
3458
3459 static void
3460 e1000_tx_timeout(struct net_device *netdev)
3461 {
3462         struct e1000_adapter *adapter = netdev_priv(netdev);
3463
3464         /* Do the reset outside of interrupt context */
3465         adapter->tx_timeout_count++;
3466         schedule_work(&adapter->reset_task);
3467 }
3468
3469 static void
3470 e1000_reset_task(struct work_struct *work)
3471 {
3472         struct e1000_adapter *adapter =
3473                 container_of(work, struct e1000_adapter, reset_task);
3474
3475         e1000_reinit_locked(adapter);
3476 }
3477
3478 /**
3479  * e1000_get_stats - Get System Network Statistics
3480  * @netdev: network interface device structure
3481  *
3482  * Returns the address of the device statistics structure.
3483  * The statistics are actually updated from the timer callback.
3484  **/
3485
3486 static struct net_device_stats *
3487 e1000_get_stats(struct net_device *netdev)
3488 {
3489         struct e1000_adapter *adapter = netdev_priv(netdev);
3490
3491         /* only return the current stats */
3492         return &adapter->net_stats;
3493 }
3494
3495 /**
3496  * e1000_change_mtu - Change the Maximum Transfer Unit
3497  * @netdev: network interface device structure
3498  * @new_mtu: new value for maximum frame size
3499  *
3500  * Returns 0 on success, negative on failure
3501  **/
3502
3503 static int
3504 e1000_change_mtu(struct net_device *netdev, int new_mtu)
3505 {
3506         struct e1000_adapter *adapter = netdev_priv(netdev);
3507         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3508         uint16_t eeprom_data = 0;
3509
3510         if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3511             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3512                 DPRINTK(PROBE, ERR, "Invalid MTU setting\n");
3513                 return -EINVAL;
3514         }
3515
3516         /* Adapter-specific max frame size limits. */
3517         switch (adapter->hw.mac_type) {
3518         case e1000_undefined ... e1000_82542_rev2_1:
3519         case e1000_ich8lan:
3520                 if (max_frame > MAXIMUM_ETHERNET_FRAME_SIZE) {
3521                         DPRINTK(PROBE, ERR, "Jumbo Frames not supported.\n");
3522                         return -EINVAL;
3523                 }
3524                 break;
3525         case e1000_82573:
3526                 /* Jumbo Frames not supported if:
3527                  * - this is not an 82573L device
3528                  * - ASPM is enabled in any way (0x1A bits 3:2) */
3529                 e1000_read_eeprom(&adapter->hw, EEPROM_INIT_3GIO_3, 1,
3530                                   &eeprom_data);
3531                 if ((adapter->hw.device_id != E1000_DEV_ID_82573L) ||
3532                     (eeprom_data & EEPROM_WORD1A_ASPM_MASK)) {
3533                         if (max_frame > MAXIMUM_ETHERNET_FRAME_SIZE) {
3534                                 DPRINTK(PROBE, ERR,
3535                                         "Jumbo Frames not supported.\n");
3536                                 return -EINVAL;
3537                         }
3538                         break;
3539                 }
3540                 /* ERT will be enabled later to enable wire speed receives */
3541
3542                 /* fall through to get support */
3543         case e1000_82571:
3544         case e1000_82572:
3545         case e1000_80003es2lan:
3546 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3547                 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3548                         DPRINTK(PROBE, ERR, "MTU > 9216 not supported.\n");
3549                         return -EINVAL;
3550                 }
3551                 break;
3552         default:
3553                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3554                 break;
3555         }
3556
3557         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3558          * means we reserve 2 more, this pushes us to allocate from the next
3559          * larger slab size
3560          * i.e. RXBUFFER_2048 --> size-4096 slab */
3561
3562         if (max_frame <= E1000_RXBUFFER_256)
3563                 adapter->rx_buffer_len = E1000_RXBUFFER_256;
3564         else if (max_frame <= E1000_RXBUFFER_512)
3565                 adapter->rx_buffer_len = E1000_RXBUFFER_512;
3566         else if (max_frame <= E1000_RXBUFFER_1024)
3567                 adapter->rx_buffer_len = E1000_RXBUFFER_1024;
3568         else if (max_frame <= E1000_RXBUFFER_2048)
3569                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3570         else if (max_frame <= E1000_RXBUFFER_4096)
3571                 adapter->rx_buffer_len = E1000_RXBUFFER_4096;
3572         else if (max_frame <= E1000_RXBUFFER_8192)
3573                 adapter->rx_buffer_len = E1000_RXBUFFER_8192;
3574         else if (max_frame <= E1000_RXBUFFER_16384)
3575                 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3576
3577         /* adjust allocation if LPE protects us, and we aren't using SBP */
3578         if (!adapter->hw.tbi_compatibility_on &&
3579             ((max_frame == MAXIMUM_ETHERNET_FRAME_SIZE) ||
3580              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3581                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3582
3583         netdev->mtu = new_mtu;
3584         adapter->hw.max_frame_size = max_frame;
3585
3586         if (netif_running(netdev))
3587                 e1000_reinit_locked(adapter);
3588
3589         return 0;
3590 }
3591
3592 /**
3593  * e1000_update_stats - Update the board statistics counters
3594  * @adapter: board private structure
3595  **/
3596
3597 void
3598 e1000_update_stats(struct e1000_adapter *adapter)
3599 {
3600         struct e1000_hw *hw = &adapter->hw;
3601         struct pci_dev *pdev = adapter->pdev;
3602         unsigned long flags;
3603         uint16_t phy_tmp;
3604
3605 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3606
3607         /*
3608          * Prevent stats update while adapter is being reset, or if the pci
3609          * connection is down.
3610          */
3611         if (adapter->link_speed == 0)
3612                 return;
3613         if (pdev->error_state && pdev->error_state != pci_channel_io_normal)
3614                 return;
3615
3616         spin_lock_irqsave(&adapter->stats_lock, flags);
3617
3618         /* these counters are modified from e1000_adjust_tbi_stats,
3619          * called from the interrupt context, so they must only
3620          * be written while holding adapter->stats_lock
3621          */
3622
3623         adapter->stats.crcerrs += E1000_READ_REG(hw, CRCERRS);
3624         adapter->stats.gprc += E1000_READ_REG(hw, GPRC);
3625         adapter->stats.gorcl += E1000_READ_REG(hw, GORCL);
3626         adapter->stats.gorch += E1000_READ_REG(hw, GORCH);
3627         adapter->stats.bprc += E1000_READ_REG(hw, BPRC);
3628         adapter->stats.mprc += E1000_READ_REG(hw, MPRC);
3629         adapter->stats.roc += E1000_READ_REG(hw, ROC);
3630
3631         if (adapter->hw.mac_type != e1000_ich8lan) {
3632                 adapter->stats.prc64 += E1000_READ_REG(hw, PRC64);
3633                 adapter->stats.prc127 += E1000_READ_REG(hw, PRC127);
3634                 adapter->stats.prc255 += E1000_READ_REG(hw, PRC255);
3635                 adapter->stats.prc511 += E1000_READ_REG(hw, PRC511);
3636                 adapter->stats.prc1023 += E1000_READ_REG(hw, PRC1023);
3637                 adapter->stats.prc1522 += E1000_READ_REG(hw, PRC1522);
3638         }
3639
3640         adapter->stats.symerrs += E1000_READ_REG(hw, SYMERRS);
3641         adapter->stats.mpc += E1000_READ_REG(hw, MPC);
3642         adapter->stats.scc += E1000_READ_REG(hw, SCC);
3643         adapter->stats.ecol += E1000_READ_REG(hw, ECOL);
3644         adapter->stats.mcc += E1000_READ_REG(hw, MCC);
3645         adapter->stats.latecol += E1000_READ_REG(hw, LATECOL);
3646         adapter->stats.dc += E1000_READ_REG(hw, DC);
3647         adapter->stats.sec += E1000_READ_REG(hw, SEC);
3648         adapter->stats.rlec += E1000_READ_REG(hw, RLEC);
3649         adapter->stats.xonrxc += E1000_READ_REG(hw, XONRXC);
3650         adapter->stats.xontxc += E1000_READ_REG(hw, XONTXC);
3651         adapter->stats.xoffrxc += E1000_READ_REG(hw, XOFFRXC);
3652         adapter->stats.xofftxc += E1000_READ_REG(hw, XOFFTXC);
3653         adapter->stats.fcruc += E1000_READ_REG(hw, FCRUC);
3654         adapter->stats.gptc += E1000_READ_REG(hw, GPTC);
3655         adapter->stats.gotcl += E1000_READ_REG(hw, GOTCL);
3656         adapter->stats.gotch += E1000_READ_REG(hw, GOTCH);
3657         adapter->stats.rnbc += E1000_READ_REG(hw, RNBC);
3658         adapter->stats.ruc += E1000_READ_REG(hw, RUC);
3659         adapter->stats.rfc += E1000_READ_REG(hw, RFC);
3660         adapter->stats.rjc += E1000_READ_REG(hw, RJC);
3661         adapter->stats.torl += E1000_READ_REG(hw, TORL);
3662         adapter->stats.torh += E1000_READ_REG(hw, TORH);
3663         adapter->stats.totl += E1000_READ_REG(hw, TOTL);
3664         adapter->stats.toth += E1000_READ_REG(hw, TOTH);
3665         adapter->stats.tpr += E1000_READ_REG(hw, TPR);
3666
3667         if (adapter->hw.mac_type != e1000_ich8lan) {
3668                 adapter->stats.ptc64 += E1000_READ_REG(hw, PTC64);
3669                 adapter->stats.ptc127 += E1000_READ_REG(hw, PTC127);
3670                 adapter->stats.ptc255 += E1000_READ_REG(hw, PTC255);
3671                 adapter->stats.ptc511 += E1000_READ_REG(hw, PTC511);
3672                 adapter->stats.ptc1023 += E1000_READ_REG(hw, PTC1023);
3673                 adapter->stats.ptc1522 += E1000_READ_REG(hw, PTC1522);
3674         }
3675
3676         adapter->stats.mptc += E1000_READ_REG(hw, MPTC);
3677         adapter->stats.bptc += E1000_READ_REG(hw, BPTC);
3678
3679         /* used for adaptive IFS */
3680
3681         hw->tx_packet_delta = E1000_READ_REG(hw, TPT);
3682         adapter->stats.tpt += hw->tx_packet_delta;
3683         hw->collision_delta = E1000_READ_REG(hw, COLC);
3684         adapter->stats.colc += hw->collision_delta;
3685
3686         if (hw->mac_type >= e1000_82543) {
3687                 adapter->stats.algnerrc += E1000_READ_REG(hw, ALGNERRC);
3688                 adapter->stats.rxerrc += E1000_READ_REG(hw, RXERRC);
3689                 adapter->stats.tncrs += E1000_READ_REG(hw, TNCRS);
3690                 adapter->stats.cexterr += E1000_READ_REG(hw, CEXTERR);
3691                 adapter->stats.tsctc += E1000_READ_REG(hw, TSCTC);
3692                 adapter->stats.tsctfc += E1000_READ_REG(hw, TSCTFC);
3693         }
3694         if (hw->mac_type > e1000_82547_rev_2) {
3695                 adapter->stats.iac += E1000_READ_REG(hw, IAC);
3696                 adapter->stats.icrxoc += E1000_READ_REG(hw, ICRXOC);
3697
3698                 if (adapter->hw.mac_type != e1000_ich8lan) {
3699                         adapter->stats.icrxptc += E1000_READ_REG(hw, ICRXPTC);
3700                         adapter->stats.icrxatc += E1000_READ_REG(hw, ICRXATC);
3701                         adapter->stats.ictxptc += E1000_READ_REG(hw, ICTXPTC);
3702                         adapter->stats.ictxatc += E1000_READ_REG(hw, ICTXATC);
3703                         adapter->stats.ictxqec += E1000_READ_REG(hw, ICTXQEC);
3704                         adapter->stats.ictxqmtc += E1000_READ_REG(hw, ICTXQMTC);
3705                         adapter->stats.icrxdmtc += E1000_READ_REG(hw, ICRXDMTC);
3706                 }
3707         }
3708
3709         /* Fill out the OS statistics structure */
3710         adapter->net_stats.rx_packets = adapter->stats.gprc;
3711         adapter->net_stats.tx_packets = adapter->stats.gptc;
3712         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
3713         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
3714         adapter->net_stats.multicast = adapter->stats.mprc;
3715         adapter->net_stats.collisions = adapter->stats.colc;
3716
3717         /* Rx Errors */
3718
3719         /* RLEC on some newer hardware can be incorrect so build
3720         * our own version based on RUC and ROC */
3721         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3722                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3723                 adapter->stats.ruc + adapter->stats.roc +
3724                 adapter->stats.cexterr;
3725         adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3726         adapter->net_stats.rx_length_errors = adapter->stats.rlerrc;
3727         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3728         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3729         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3730
3731         /* Tx Errors */
3732         adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3733         adapter->net_stats.tx_errors = adapter->stats.txerrc;
3734         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3735         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3736         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3737         if (adapter->hw.bad_tx_carr_stats_fd &&
3738             adapter->link_duplex == FULL_DUPLEX) {
3739                 adapter->net_stats.tx_carrier_errors = 0;
3740                 adapter->stats.tncrs = 0;
3741         }
3742
3743         /* Tx Dropped needs to be maintained elsewhere */
3744
3745         /* Phy Stats */
3746         if (hw->media_type == e1000_media_type_copper) {
3747                 if ((adapter->link_speed == SPEED_1000) &&
3748                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3749                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3750                         adapter->phy_stats.idle_errors += phy_tmp;
3751                 }
3752
3753                 if ((hw->mac_type <= e1000_82546) &&
3754                    (hw->phy_type == e1000_phy_m88) &&
3755                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3756                         adapter->phy_stats.receive_errors += phy_tmp;
3757         }
3758
3759         /* Management Stats */
3760         if (adapter->hw.has_smbus) {
3761                 adapter->stats.mgptc += E1000_READ_REG(hw, MGTPTC);
3762                 adapter->stats.mgprc += E1000_READ_REG(hw, MGTPRC);
3763                 adapter->stats.mgpdc += E1000_READ_REG(hw, MGTPDC);
3764         }
3765
3766         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3767 }
3768 #ifdef CONFIG_PCI_MSI
3769
3770 /**
3771  * e1000_intr_msi - Interrupt Handler
3772  * @irq: interrupt number
3773  * @data: pointer to a network interface device structure
3774  **/
3775
3776 static irqreturn_t
3777 e1000_intr_msi(int irq, void *data)
3778 {
3779         struct net_device *netdev = data;
3780         struct e1000_adapter *adapter = netdev_priv(netdev);
3781         struct e1000_hw *hw = &adapter->hw;
3782 #ifndef CONFIG_E1000_NAPI
3783         int i;
3784 #endif
3785         uint32_t icr = E1000_READ_REG(hw, ICR);
3786
3787 #ifdef CONFIG_E1000_NAPI
3788         /* read ICR disables interrupts using IAM, so keep up with our
3789          * enable/disable accounting */
3790         atomic_inc(&adapter->irq_sem);
3791 #endif
3792         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3793                 hw->get_link_status = 1;
3794                 /* 80003ES2LAN workaround-- For packet buffer work-around on
3795                  * link down event; disable receives here in the ISR and reset
3796                  * adapter in watchdog */
3797                 if (netif_carrier_ok(netdev) &&
3798                     (adapter->hw.mac_type == e1000_80003es2lan)) {
3799                         /* disable receives */
3800                         uint32_t rctl = E1000_READ_REG(hw, RCTL);
3801                         E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
3802                 }
3803                 /* guard against interrupt when we're going down */
3804                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3805                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3806         }
3807
3808 #ifdef CONFIG_E1000_NAPI
3809         if (likely(netif_rx_schedule_prep(netdev))) {
3810                 adapter->total_tx_bytes = 0;
3811                 adapter->total_tx_packets = 0;
3812                 adapter->total_rx_bytes = 0;
3813                 adapter->total_rx_packets = 0;
3814                 __netif_rx_schedule(netdev);
3815         } else
3816                 e1000_irq_enable(adapter);
3817 #else
3818         adapter->total_tx_bytes = 0;
3819         adapter->total_rx_bytes = 0;
3820         adapter->total_tx_packets = 0;
3821         adapter->total_rx_packets = 0;
3822
3823         for (i = 0; i < E1000_MAX_INTR; i++)
3824                 if (unlikely(!adapter->clean_rx(adapter, adapter->rx_ring) &
3825                    e1000_clean_tx_irq(adapter, adapter->tx_ring)))
3826                         break;
3827
3828         if (likely(adapter->itr_setting & 3))
3829                 e1000_set_itr(adapter);
3830 #endif
3831
3832         return IRQ_HANDLED;
3833 }
3834 #endif
3835
3836 /**
3837  * e1000_intr - Interrupt Handler
3838  * @irq: interrupt number
3839  * @data: pointer to a network interface device structure
3840  **/
3841
3842 static irqreturn_t
3843 e1000_intr(int irq, void *data)
3844 {
3845         struct net_device *netdev = data;
3846         struct e1000_adapter *adapter = netdev_priv(netdev);
3847         struct e1000_hw *hw = &adapter->hw;
3848         uint32_t rctl, icr = E1000_READ_REG(hw, ICR);
3849 #ifndef CONFIG_E1000_NAPI
3850         int i;
3851 #endif
3852         if (unlikely(!icr))
3853                 return IRQ_NONE;  /* Not our interrupt */
3854
3855 #ifdef CONFIG_E1000_NAPI
3856         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
3857          * not set, then the adapter didn't send an interrupt */
3858         if (unlikely(hw->mac_type >= e1000_82571 &&
3859                      !(icr & E1000_ICR_INT_ASSERTED)))
3860                 return IRQ_NONE;
3861
3862         /* Interrupt Auto-Mask...upon reading ICR,
3863          * interrupts are masked.  No need for the
3864          * IMC write, but it does mean we should
3865          * account for it ASAP. */
3866         if (likely(hw->mac_type >= e1000_82571))
3867                 atomic_inc(&adapter->irq_sem);
3868 #endif
3869
3870         if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3871                 hw->get_link_status = 1;
3872                 /* 80003ES2LAN workaround--
3873                  * For packet buffer work-around on link down event;
3874                  * disable receives here in the ISR and
3875                  * reset adapter in watchdog
3876                  */
3877                 if (netif_carrier_ok(netdev) &&
3878                     (adapter->hw.mac_type == e1000_80003es2lan)) {
3879                         /* disable receives */
3880                         rctl = E1000_READ_REG(hw, RCTL);
3881                         E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
3882                 }
3883                 /* guard against interrupt when we're going down */
3884                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3885                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3886         }
3887
3888 #ifdef CONFIG_E1000_NAPI
3889         if (unlikely(hw->mac_type < e1000_82571)) {
3890                 /* disable interrupts, without the synchronize_irq bit */
3891                 atomic_inc(&adapter->irq_sem);
3892                 E1000_WRITE_REG(hw, IMC, ~0);
3893                 E1000_WRITE_FLUSH(hw);
3894         }
3895         if (likely(netif_rx_schedule_prep(netdev))) {
3896                 adapter->total_tx_bytes = 0;
3897                 adapter->total_tx_packets = 0;
3898                 adapter->total_rx_bytes = 0;
3899                 adapter->total_rx_packets = 0;
3900                 __netif_rx_schedule(netdev);
3901         } else
3902                 /* this really should not happen! if it does it is basically a
3903                  * bug, but not a hard error, so enable ints and continue */
3904                 e1000_irq_enable(adapter);
3905 #else
3906         /* Writing IMC and IMS is needed for 82547.
3907          * Due to Hub Link bus being occupied, an interrupt
3908          * de-assertion message is not able to be sent.
3909          * When an interrupt assertion message is generated later,
3910          * two messages are re-ordered and sent out.
3911          * That causes APIC to think 82547 is in de-assertion
3912          * state, while 82547 is in assertion state, resulting
3913          * in dead lock. Writing IMC forces 82547 into
3914          * de-assertion state.
3915          */
3916         if (hw->mac_type == e1000_82547 || hw->mac_type == e1000_82547_rev_2) {
3917                 atomic_inc(&adapter->irq_sem);
3918                 E1000_WRITE_REG(hw, IMC, ~0);
3919         }
3920
3921         adapter->total_tx_bytes = 0;
3922         adapter->total_rx_bytes = 0;
3923         adapter->total_tx_packets = 0;
3924         adapter->total_rx_packets = 0;
3925
3926         for (i = 0; i < E1000_MAX_INTR; i++)
3927                 if (unlikely(!adapter->clean_rx(adapter, adapter->rx_ring) &
3928                    e1000_clean_tx_irq(adapter, adapter->tx_ring)))
3929                         break;
3930
3931         if (likely(adapter->itr_setting & 3))
3932                 e1000_set_itr(adapter);
3933
3934         if (hw->mac_type == e1000_82547 || hw->mac_type == e1000_82547_rev_2)
3935                 e1000_irq_enable(adapter);
3936
3937 #endif
3938         return IRQ_HANDLED;
3939 }
3940
3941 #ifdef CONFIG_E1000_NAPI
3942 /**
3943  * e1000_clean - NAPI Rx polling callback
3944  * @adapter: board private structure
3945  **/
3946
3947 static int
3948 e1000_clean(struct net_device *poll_dev, int *budget)
3949 {
3950         struct e1000_adapter *adapter;
3951         int work_to_do = min(*budget, poll_dev->quota);
3952         int tx_cleaned = 0, work_done = 0;
3953
3954         /* Must NOT use netdev_priv macro here. */
3955         adapter = poll_dev->priv;
3956
3957         /* Keep link state information with original netdev */
3958         if (!netif_carrier_ok(poll_dev))
3959                 goto quit_polling;
3960
3961         /* e1000_clean is called per-cpu.  This lock protects
3962          * tx_ring[0] from being cleaned by multiple cpus
3963          * simultaneously.  A failure obtaining the lock means
3964          * tx_ring[0] is currently being cleaned anyway. */
3965         if (spin_trylock(&adapter->tx_queue_lock)) {
3966                 tx_cleaned = e1000_clean_tx_irq(adapter,
3967                                                 &adapter->tx_ring[0]);
3968                 spin_unlock(&adapter->tx_queue_lock);
3969         }
3970
3971         adapter->clean_rx(adapter, &adapter->rx_ring[0],
3972                           &work_done, work_to_do);
3973
3974         *budget -= work_done;
3975         poll_dev->quota -= work_done;
3976
3977         /* If no Tx and not enough Rx work done, exit the polling mode */
3978         if ((tx_cleaned && (work_done < work_to_do)) ||
3979            !netif_running(poll_dev)) {
3980 quit_polling:
3981                 if (likely(adapter->itr_setting & 3))
3982                         e1000_set_itr(adapter);
3983                 netif_rx_complete(poll_dev);
3984                 e1000_irq_enable(adapter);
3985                 return 0;
3986         }
3987
3988         return 1;
3989 }
3990
3991 #endif
3992 /**
3993  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3994  * @adapter: board private structure
3995  **/
3996
3997 static boolean_t
3998 e1000_clean_tx_irq(struct e1000_adapter *adapter,
3999                    struct e1000_tx_ring *tx_ring)
4000 {
4001         struct net_device *netdev = adapter->netdev;
4002         struct e1000_tx_desc *tx_desc, *eop_desc;
4003         struct e1000_buffer *buffer_info;
4004         unsigned int i, eop;
4005 #ifdef CONFIG_E1000_NAPI
4006         unsigned int count = 0;
4007 #endif
4008         boolean_t cleaned = TRUE;
4009         unsigned int total_tx_bytes=0, total_tx_packets=0;
4010
4011         i = tx_ring->next_to_clean;
4012         eop = tx_ring->buffer_info[i].next_to_watch;
4013         eop_desc = E1000_TX_DESC(*tx_ring, eop);
4014
4015         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
4016                 for (cleaned = FALSE; !cleaned; ) {
4017                         tx_desc = E1000_TX_DESC(*tx_ring, i);
4018                         buffer_info = &tx_ring->buffer_info[i];
4019                         cleaned = (i == eop);
4020
4021                         if (cleaned) {
4022                                 struct sk_buff *skb = buffer_info->skb;
4023                                 unsigned int segs, bytecount;
4024                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
4025                                 /* multiply data chunks by size of headers */
4026                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
4027                                             skb->len;
4028                                 total_tx_packets += segs;
4029                                 total_tx_bytes += bytecount;
4030                         }
4031                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
4032                         tx_desc->upper.data = 0;
4033
4034                         if (unlikely(++i == tx_ring->count)) i = 0;
4035                 }
4036
4037                 eop = tx_ring->buffer_info[i].next_to_watch;
4038                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
4039 #ifdef CONFIG_E1000_NAPI
4040 #define E1000_TX_WEIGHT 64
4041                 /* weight of a sort for tx, to avoid endless transmit cleanup */
4042                 if (count++ == E1000_TX_WEIGHT) {
4043                         cleaned = FALSE;
4044                         break;
4045                 }
4046 #endif
4047         }
4048
4049         tx_ring->next_to_clean = i;
4050
4051 #define TX_WAKE_THRESHOLD 32
4052         if (unlikely(cleaned && netif_carrier_ok(netdev) &&
4053                      E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
4054                 /* Make sure that anybody stopping the queue after this
4055                  * sees the new next_to_clean.
4056                  */
4057                 smp_mb();
4058                 if (netif_queue_stopped(netdev)) {
4059                         netif_wake_queue(netdev);
4060                         ++adapter->restart_queue;
4061                 }
4062         }
4063
4064         if (adapter->detect_tx_hung) {
4065                 /* Detect a transmit hang in hardware, this serializes the
4066                  * check with the clearing of time_stamp and movement of i */
4067                 adapter->detect_tx_hung = FALSE;
4068                 if (tx_ring->buffer_info[eop].dma &&
4069                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
4070                                (adapter->tx_timeout_factor * HZ))
4071                     && !(E1000_READ_REG(&adapter->hw, STATUS) &
4072                          E1000_STATUS_TXOFF)) {
4073
4074                         /* detected Tx unit hang */
4075                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
4076                                         "  Tx Queue             <%lu>\n"
4077                                         "  TDH                  <%x>\n"
4078                                         "  TDT                  <%x>\n"
4079                                         "  next_to_use          <%x>\n"
4080                                         "  next_to_clean        <%x>\n"
4081                                         "buffer_info[next_to_clean]\n"
4082                                         "  time_stamp           <%lx>\n"
4083                                         "  next_to_watch        <%x>\n"
4084                                         "  jiffies              <%lx>\n"
4085                                         "  next_to_watch.status <%x>\n",
4086                                 (unsigned long)((tx_ring - adapter->tx_ring) /
4087                                         sizeof(struct e1000_tx_ring)),
4088                                 readl(adapter->hw.hw_addr + tx_ring->tdh),
4089                                 readl(adapter->hw.hw_addr + tx_ring->tdt),
4090                                 tx_ring->next_to_use,
4091                                 tx_ring->next_to_clean,
4092                                 tx_ring->buffer_info[eop].time_stamp,
4093                                 eop,
4094                                 jiffies,
4095                                 eop_desc->upper.fields.status);
4096                         netif_stop_queue(netdev);
4097                 }
4098         }
4099         adapter->total_tx_bytes += total_tx_bytes;
4100         adapter->total_tx_packets += total_tx_packets;
4101         return cleaned;
4102 }
4103
4104 /**
4105  * e1000_rx_checksum - Receive Checksum Offload for 82543
4106  * @adapter:     board private structure
4107  * @status_err:  receive descriptor status and error fields
4108  * @csum:        receive descriptor csum field
4109  * @sk_buff:     socket buffer with received data
4110  **/
4111
4112 static void
4113 e1000_rx_checksum(struct e1000_adapter *adapter,
4114                   uint32_t status_err, uint32_t csum,
4115                   struct sk_buff *skb)
4116 {
4117         uint16_t status = (uint16_t)status_err;
4118         uint8_t errors = (uint8_t)(status_err >> 24);
4119         skb->ip_summed = CHECKSUM_NONE;
4120
4121         /* 82543 or newer only */
4122         if (unlikely(adapter->hw.mac_type < e1000_82543)) return;
4123         /* Ignore Checksum bit is set */
4124         if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
4125         /* TCP/UDP checksum error bit is set */
4126         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
4127                 /* let the stack verify checksum errors */
4128                 adapter->hw_csum_err++;
4129                 return;
4130         }
4131         /* TCP/UDP Checksum has not been calculated */
4132         if (adapter->hw.mac_type <= e1000_82547_rev_2) {
4133                 if (!(status & E1000_RXD_STAT_TCPCS))
4134                         return;
4135         } else {
4136                 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
4137                         return;
4138         }
4139         /* It must be a TCP or UDP packet with a valid checksum */
4140         if (likely(status & E1000_RXD_STAT_TCPCS)) {
4141                 /* TCP checksum is good */
4142                 skb->ip_summed = CHECKSUM_UNNECESSARY;
4143         } else if (adapter->hw.mac_type > e1000_82547_rev_2) {
4144                 /* IP fragment with UDP payload */
4145                 /* Hardware complements the payload checksum, so we undo it
4146                  * and then put the value in host order for further stack use.
4147                  */
4148                 csum = ntohl(csum ^ 0xFFFF);
4149                 skb->csum = csum;
4150                 skb->ip_summed = CHECKSUM_COMPLETE;
4151         }
4152         adapter->hw_csum_good++;
4153 }
4154
4155 /**
4156  * e1000_clean_rx_irq - Send received data up the network stack; legacy
4157  * @adapter: board private structure
4158  **/
4159
4160 static boolean_t
4161 #ifdef CONFIG_E1000_NAPI
4162 e1000_clean_rx_irq(struct e1000_adapter *adapter,
4163                    struct e1000_rx_ring *rx_ring,
4164                    int *work_done, int work_to_do)
4165 #else
4166 e1000_clean_rx_irq(struct e1000_adapter *adapter,
4167                    struct e1000_rx_ring *rx_ring)
4168 #endif
4169 {
4170         struct net_device *netdev = adapter->netdev;
4171         struct pci_dev *pdev = adapter->pdev;
4172         struct e1000_rx_desc *rx_desc, *next_rxd;
4173         struct e1000_buffer *buffer_info, *next_buffer;
4174         unsigned long flags;
4175         uint32_t length;
4176         uint8_t last_byte;
4177         unsigned int i;
4178         int cleaned_count = 0;
4179         boolean_t cleaned = FALSE;
4180         unsigned int total_rx_bytes=0, total_rx_packets=0;
4181
4182         i = rx_ring->next_to_clean;
4183         rx_desc = E1000_RX_DESC(*rx_ring, i);
4184         buffer_info = &rx_ring->buffer_info[i];
4185
4186         while (rx_desc->status & E1000_RXD_STAT_DD) {
4187                 struct sk_buff *skb;
4188                 u8 status;
4189
4190 #ifdef CONFIG_E1000_NAPI
4191                 if (*work_done >= work_to_do)
4192                         break;
4193                 (*work_done)++;
4194 #endif
4195                 status = rx_desc->status;
4196                 skb = buffer_info->skb;
4197                 buffer_info->skb = NULL;
4198
4199                 prefetch(skb->data - NET_IP_ALIGN);
4200
4201                 if (++i == rx_ring->count) i = 0;
4202                 next_rxd = E1000_RX_DESC(*rx_ring, i);
4203                 prefetch(next_rxd);
4204
4205                 next_buffer = &rx_ring->buffer_info[i];
4206
4207                 cleaned = TRUE;
4208                 cleaned_count++;
4209                 pci_unmap_single(pdev,
4210                                  buffer_info->dma,
4211                                  buffer_info->length,
4212                                  PCI_DMA_FROMDEVICE);
4213
4214                 length = le16_to_cpu(rx_desc->length);
4215
4216                 if (unlikely(!(status & E1000_RXD_STAT_EOP))) {
4217                         /* All receives must fit into a single buffer */
4218                         E1000_DBG("%s: Receive packet consumed multiple"
4219                                   " buffers\n", netdev->name);
4220                         /* recycle */
4221                         buffer_info->skb = skb;
4222                         goto next_desc;
4223                 }
4224
4225                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
4226                         last_byte = *(skb->data + length - 1);
4227                         if (TBI_ACCEPT(&adapter->hw, status,
4228                                       rx_desc->errors, length, last_byte)) {
4229                                 spin_lock_irqsave(&adapter->stats_lock, flags);
4230                                 e1000_tbi_adjust_stats(&adapter->hw,
4231                                                        &adapter->stats,
4232                                                        length, skb->data);
4233                                 spin_unlock_irqrestore(&adapter->stats_lock,
4234                                                        flags);
4235                                 length--;
4236                         } else {
4237                                 /* recycle */
4238                                 buffer_info->skb = skb;
4239                                 goto next_desc;
4240                         }
4241                 }
4242
4243                 /* adjust length to remove Ethernet CRC, this must be
4244                  * done after the TBI_ACCEPT workaround above */
4245                 length -= 4;
4246
4247                 /* probably a little skewed due to removing CRC */
4248                 total_rx_bytes += length;
4249                 total_rx_packets++;
4250
4251                 /* code added for copybreak, this should improve
4252                  * performance for small packets with large amounts
4253                  * of reassembly being done in the stack */
4254                 if (length < copybreak) {
4255                         struct sk_buff *new_skb =
4256                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
4257                         if (new_skb) {
4258                                 skb_reserve(new_skb, NET_IP_ALIGN);
4259                                 memcpy(new_skb->data - NET_IP_ALIGN,
4260                                        skb->data - NET_IP_ALIGN,
4261                                        length + NET_IP_ALIGN);
4262                                 /* save the skb in buffer_info as good */
4263                                 buffer_info->skb = skb;
4264                                 skb = new_skb;
4265                         }
4266                         /* else just continue with the old one */
4267                 }
4268                 /* end copybreak code */
4269                 skb_put(skb, length);
4270
4271                 /* Receive Checksum Offload */
4272                 e1000_rx_checksum(adapter,
4273                                   (uint32_t)(status) |
4274                                   ((uint32_t)(rx_desc->errors) << 24),
4275                                   le16_to_cpu(rx_desc->csum), skb);
4276
4277                 skb->protocol = eth_type_trans(skb, netdev);
4278 #ifdef CONFIG_E1000_NAPI
4279                 if (unlikely(adapter->vlgrp &&
4280                             (status & E1000_RXD_STAT_VP))) {
4281                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
4282                                                  le16_to_cpu(rx_desc->special) &
4283                                                  E1000_RXD_SPC_VLAN_MASK);
4284                 } else {
4285                         netif_receive_skb(skb);
4286                 }
4287 #else /* CONFIG_E1000_NAPI */
4288                 if (unlikely(adapter->vlgrp &&
4289                             (status & E1000_RXD_STAT_VP))) {
4290                         vlan_hwaccel_rx(skb, adapter->vlgrp,
4291                                         le16_to_cpu(rx_desc->special) &
4292                                         E1000_RXD_SPC_VLAN_MASK);
4293                 } else {
4294                         netif_rx(skb);
4295                 }
4296 #endif /* CONFIG_E1000_NAPI */
4297                 netdev->last_rx = jiffies;
4298
4299 next_desc:
4300                 rx_desc->status = 0;
4301
4302                 /* return some buffers to hardware, one at a time is too slow */
4303                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4304                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4305                         cleaned_count = 0;
4306                 }
4307
4308                 /* use prefetched values */
4309                 rx_desc = next_rxd;
4310                 buffer_info = next_buffer;
4311         }
4312         rx_ring->next_to_clean = i;
4313
4314         cleaned_count = E1000_DESC_UNUSED(rx_ring);
4315         if (cleaned_count)
4316                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4317
4318         adapter->total_rx_packets += total_rx_packets;
4319         adapter->total_rx_bytes += total_rx_bytes;
4320         return cleaned;
4321 }
4322
4323 /**
4324  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
4325  * @adapter: board private structure
4326  **/
4327
4328 static boolean_t
4329 #ifdef CONFIG_E1000_NAPI
4330 e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
4331                       struct e1000_rx_ring *rx_ring,
4332                       int *work_done, int work_to_do)
4333 #else
4334 e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
4335                       struct e1000_rx_ring *rx_ring)
4336 #endif
4337 {
4338         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
4339         struct net_device *netdev = adapter->netdev;
4340         struct pci_dev *pdev = adapter->pdev;
4341         struct e1000_buffer *buffer_info, *next_buffer;
4342         struct e1000_ps_page *ps_page;
4343         struct e1000_ps_page_dma *ps_page_dma;
4344         struct sk_buff *skb;
4345         unsigned int i, j;
4346         uint32_t length, staterr;
4347         int cleaned_count = 0;
4348         boolean_t cleaned = FALSE;
4349         unsigned int total_rx_bytes=0, total_rx_packets=0;
4350
4351         i = rx_ring->next_to_clean;
4352         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
4353         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
4354         buffer_info = &rx_ring->buffer_info[i];
4355
4356         while (staterr & E1000_RXD_STAT_DD) {
4357                 ps_page = &rx_ring->ps_page[i];
4358                 ps_page_dma = &rx_ring->ps_page_dma[i];
4359 #ifdef CONFIG_E1000_NAPI
4360                 if (unlikely(*work_done >= work_to_do))
4361                         break;
4362                 (*work_done)++;
4363 #endif
4364                 skb = buffer_info->skb;
4365
4366                 /* in the packet split case this is header only */
4367                 prefetch(skb->data - NET_IP_ALIGN);
4368
4369                 if (++i == rx_ring->count) i = 0;
4370                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
4371                 prefetch(next_rxd);
4372
4373                 next_buffer = &rx_ring->buffer_info[i];
4374
4375                 cleaned = TRUE;
4376                 cleaned_count++;
4377                 pci_unmap_single(pdev, buffer_info->dma,
4378                                  buffer_info->length,
4379                                  PCI_DMA_FROMDEVICE);
4380
4381                 if (unlikely(!(staterr & E1000_RXD_STAT_EOP))) {
4382                         E1000_DBG("%s: Packet Split buffers didn't pick up"
4383                                   " the full packet\n", netdev->name);
4384                         dev_kfree_skb_irq(skb);
4385                         goto next_desc;
4386                 }
4387
4388                 if (unlikely(staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK)) {
4389                         dev_kfree_skb_irq(skb);
4390                         goto next_desc;
4391                 }
4392
4393                 length = le16_to_cpu(rx_desc->wb.middle.length0);
4394
4395                 if (unlikely(!length)) {
4396                         E1000_DBG("%s: Last part of the packet spanning"
4397                                   " multiple descriptors\n", netdev->name);
4398                         dev_kfree_skb_irq(skb);
4399                         goto next_desc;
4400                 }
4401
4402                 /* Good Receive */
4403                 skb_put(skb, length);
4404
4405                 {
4406                 /* this looks ugly, but it seems compiler issues make it
4407                    more efficient than reusing j */
4408                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
4409
4410                 /* page alloc/put takes too long and effects small packet
4411                  * throughput, so unsplit small packets and save the alloc/put*/
4412                 if (l1 && (l1 <= copybreak) && ((length + l1) <= adapter->rx_ps_bsize0)) {
4413                         u8 *vaddr;
4414                         /* there is no documentation about how to call
4415                          * kmap_atomic, so we can't hold the mapping
4416                          * very long */
4417                         pci_dma_sync_single_for_cpu(pdev,
4418                                 ps_page_dma->ps_page_dma[0],
4419                                 PAGE_SIZE,
4420                                 PCI_DMA_FROMDEVICE);
4421                         vaddr = kmap_atomic(ps_page->ps_page[0],
4422                                             KM_SKB_DATA_SOFTIRQ);
4423                         memcpy(skb->tail, vaddr, l1);
4424                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
4425                         pci_dma_sync_single_for_device(pdev,
4426                                 ps_page_dma->ps_page_dma[0],
4427                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
4428                         /* remove the CRC */
4429                         l1 -= 4;
4430                         skb_put(skb, l1);
4431                         goto copydone;
4432                 } /* if */
4433                 }
4434
4435                 for (j = 0; j < adapter->rx_ps_pages; j++) {
4436                         if (!(length= le16_to_cpu(rx_desc->wb.upper.length[j])))
4437                                 break;
4438                         pci_unmap_page(pdev, ps_page_dma->ps_page_dma[j],
4439                                         PAGE_SIZE, PCI_DMA_FROMDEVICE);
4440                         ps_page_dma->ps_page_dma[j] = 0;
4441                         skb_fill_page_desc(skb, j, ps_page->ps_page[j], 0,
4442                                            length);
4443                         ps_page->ps_page[j] = NULL;
4444                         skb->len += length;
4445                         skb->data_len += length;
4446                         skb->truesize += length;
4447                 }
4448
4449                 /* strip the ethernet crc, problem is we're using pages now so
4450                  * this whole operation can get a little cpu intensive */
4451                 pskb_trim(skb, skb->len - 4);
4452
4453 copydone:
4454                 total_rx_bytes += skb->len;
4455                 total_rx_packets++;
4456
4457                 e1000_rx_checksum(adapter, staterr,
4458                                   le16_to_cpu(rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
4459                 skb->protocol = eth_type_trans(skb, netdev);
4460
4461                 if (likely(rx_desc->wb.upper.header_status &
4462                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP)))
4463                         adapter->rx_hdr_split++;
4464 #ifdef CONFIG_E1000_NAPI
4465                 if (unlikely(adapter->vlgrp && (staterr & E1000_RXD_STAT_VP))) {
4466                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
4467                                 le16_to_cpu(rx_desc->wb.middle.vlan) &
4468                                 E1000_RXD_SPC_VLAN_MASK);
4469                 } else {
4470                         netif_receive_skb(skb);
4471                 }
4472 #else /* CONFIG_E1000_NAPI */
4473                 if (unlikely(adapter->vlgrp && (staterr & E1000_RXD_STAT_VP))) {
4474                         vlan_hwaccel_rx(skb, adapter->vlgrp,
4475                                 le16_to_cpu(rx_desc->wb.middle.vlan) &
4476                                 E1000_RXD_SPC_VLAN_MASK);
4477                 } else {
4478                         netif_rx(skb);
4479                 }
4480 #endif /* CONFIG_E1000_NAPI */
4481                 netdev->last_rx = jiffies;
4482
4483 next_desc:
4484                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
4485                 buffer_info->skb = NULL;
4486
4487                 /* return some buffers to hardware, one at a time is too slow */
4488                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4489                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4490                         cleaned_count = 0;
4491                 }
4492
4493                 /* use prefetched values */
4494                 rx_desc = next_rxd;
4495                 buffer_info = next_buffer;
4496
4497                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
4498         }
4499         rx_ring->next_to_clean = i;
4500
4501         cleaned_count = E1000_DESC_UNUSED(rx_ring);
4502         if (cleaned_count)
4503                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4504
4505         adapter->total_rx_packets += total_rx_packets;
4506         adapter->total_rx_bytes += total_rx_bytes;
4507         return cleaned;
4508 }
4509
4510 /**
4511  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4512  * @adapter: address of board private structure
4513  **/
4514
4515 static void
4516 e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4517                        struct e1000_rx_ring *rx_ring,
4518                        int cleaned_count)
4519 {
4520         struct net_device *netdev = adapter->netdev;
4521         struct pci_dev *pdev = adapter->pdev;
4522         struct e1000_rx_desc *rx_desc;
4523         struct e1000_buffer *buffer_info;
4524         struct sk_buff *skb;
4525         unsigned int i;
4526         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
4527
4528         i = rx_ring->next_to_use;
4529         buffer_info = &rx_ring->buffer_info[i];
4530
4531         while (cleaned_count--) {
4532                 skb = buffer_info->skb;
4533                 if (skb) {
4534                         skb_trim(skb, 0);
4535                         goto map_skb;
4536                 }
4537
4538                 skb = netdev_alloc_skb(netdev, bufsz);
4539                 if (unlikely(!skb)) {
4540                         /* Better luck next round */
4541                         adapter->alloc_rx_buff_failed++;
4542                         break;
4543                 }
4544
4545                 /* Fix for errata 23, can't cross 64kB boundary */
4546                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4547                         struct sk_buff *oldskb = skb;
4548                         DPRINTK(RX_ERR, ERR, "skb align check failed: %u bytes "
4549                                              "at %p\n", bufsz, skb->data);
4550                         /* Try again, without freeing the previous */
4551                         skb = netdev_alloc_skb(netdev, bufsz);
4552                         /* Failed allocation, critical failure */
4553                         if (!skb) {
4554                                 dev_kfree_skb(oldskb);
4555                                 break;
4556                         }
4557
4558                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4559                                 /* give up */
4560                                 dev_kfree_skb(skb);
4561                                 dev_kfree_skb(oldskb);
4562                                 break; /* while !buffer_info->skb */
4563                         }
4564
4565                         /* Use new allocation */
4566                         dev_kfree_skb(oldskb);
4567                 }
4568                 /* Make buffer alignment 2 beyond a 16 byte boundary
4569                  * this will result in a 16 byte aligned IP header after
4570                  * the 14 byte MAC header is removed
4571                  */
4572                 skb_reserve(skb, NET_IP_ALIGN);
4573
4574                 buffer_info->skb = skb;
4575                 buffer_info->length = adapter->rx_buffer_len;
4576 map_skb:
4577                 buffer_info->dma = pci_map_single(pdev,
4578                                                   skb->data,
4579                                                   adapter->rx_buffer_len,
4580                                                   PCI_DMA_FROMDEVICE);
4581
4582                 /* Fix for errata 23, can't cross 64kB boundary */
4583                 if (!e1000_check_64k_bound(adapter,
4584                                         (void *)(unsigned long)buffer_info->dma,
4585                                         adapter->rx_buffer_len)) {
4586                         DPRINTK(RX_ERR, ERR,
4587                                 "dma align check failed: %u bytes at %p\n",
4588                                 adapter->rx_buffer_len,
4589                                 (void *)(unsigned long)buffer_info->dma);
4590                         dev_kfree_skb(skb);
4591                         buffer_info->skb = NULL;
4592
4593                         pci_unmap_single(pdev, buffer_info->dma,
4594                                          adapter->rx_buffer_len,
4595                                          PCI_DMA_FROMDEVICE);
4596
4597                         break; /* while !buffer_info->skb */
4598                 }
4599                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4600                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4601
4602                 if (unlikely(++i == rx_ring->count))
4603                         i = 0;
4604                 buffer_info = &rx_ring->buffer_info[i];
4605         }
4606
4607         if (likely(rx_ring->next_to_use != i)) {
4608                 rx_ring->next_to_use = i;
4609                 if (unlikely(i-- == 0))
4610                         i = (rx_ring->count - 1);
4611
4612                 /* Force memory writes to complete before letting h/w
4613                  * know there are new descriptors to fetch.  (Only
4614                  * applicable for weak-ordered memory model archs,
4615                  * such as IA-64). */
4616                 wmb();
4617                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4618         }
4619 }
4620
4621 /**
4622  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
4623  * @adapter: address of board private structure
4624  **/
4625
4626 static void
4627 e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
4628                           struct e1000_rx_ring *rx_ring,
4629                           int cleaned_count)
4630 {
4631         struct net_device *netdev = adapter->netdev;
4632         struct pci_dev *pdev = adapter->pdev;
4633         union e1000_rx_desc_packet_split *rx_desc;
4634         struct e1000_buffer *buffer_info;
4635         struct e1000_ps_page *ps_page;
4636         struct e1000_ps_page_dma *ps_page_dma;
4637         struct sk_buff *skb;
4638         unsigned int i, j;
4639
4640         i = rx_ring->next_to_use;
4641         buffer_info = &rx_ring->buffer_info[i];
4642         ps_page = &rx_ring->ps_page[i];
4643         ps_page_dma = &rx_ring->ps_page_dma[i];
4644
4645         while (cleaned_count--) {
4646                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
4647
4648                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
4649                         if (j < adapter->rx_ps_pages) {
4650                                 if (likely(!ps_page->ps_page[j])) {
4651                                         ps_page->ps_page[j] =
4652                                                 alloc_page(GFP_ATOMIC);
4653                                         if (unlikely(!ps_page->ps_page[j])) {
4654                                                 adapter->alloc_rx_buff_failed++;
4655                                                 goto no_buffers;
4656                                         }
4657                                         ps_page_dma->ps_page_dma[j] =
4658                                                 pci_map_page(pdev,
4659                                                             ps_page->ps_page[j],
4660                                                             0, PAGE_SIZE,
4661                                                             PCI_DMA_FROMDEVICE);
4662                                 }
4663                                 /* Refresh the desc even if buffer_addrs didn't
4664                                  * change because each write-back erases
4665                                  * this info.
4666                                  */
4667                                 rx_desc->read.buffer_addr[j+1] =
4668                                      cpu_to_le64(ps_page_dma->ps_page_dma[j]);
4669                         } else
4670                                 rx_desc->read.buffer_addr[j+1] = ~0;
4671                 }
4672
4673                 skb = netdev_alloc_skb(netdev,
4674                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
4675
4676                 if (unlikely(!skb)) {
4677                         adapter->alloc_rx_buff_failed++;
4678                         break;
4679                 }
4680
4681                 /* Make buffer alignment 2 beyond a 16 byte boundary
4682                  * this will result in a 16 byte aligned IP header after
4683                  * the 14 byte MAC header is removed
4684                  */
4685                 skb_reserve(skb, NET_IP_ALIGN);
4686
4687                 buffer_info->skb = skb;
4688                 buffer_info->length = adapter->rx_ps_bsize0;
4689                 buffer_info->dma = pci_map_single(pdev, skb->data,
4690                                                   adapter->rx_ps_bsize0,
4691                                                   PCI_DMA_FROMDEVICE);
4692
4693                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
4694
4695                 if (unlikely(++i == rx_ring->count)) i = 0;
4696                 buffer_info = &rx_ring->buffer_info[i];
4697                 ps_page = &rx_ring->ps_page[i];
4698                 ps_page_dma = &rx_ring->ps_page_dma[i];
4699         }
4700
4701 no_buffers:
4702         if (likely(rx_ring->next_to_use != i)) {
4703                 rx_ring->next_to_use = i;
4704                 if (unlikely(i-- == 0)) i = (rx_ring->count - 1);
4705
4706                 /* Force memory writes to complete before letting h/w
4707                  * know there are new descriptors to fetch.  (Only
4708                  * applicable for weak-ordered memory model archs,
4709                  * such as IA-64). */
4710                 wmb();
4711                 /* Hardware increments by 16 bytes, but packet split
4712                  * descriptors are 32 bytes...so we increment tail
4713                  * twice as much.
4714                  */
4715                 writel(i<<1, adapter->hw.hw_addr + rx_ring->rdt);
4716         }
4717 }
4718
4719 /**
4720  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4721  * @adapter:
4722  **/
4723
4724 static void
4725 e1000_smartspeed(struct e1000_adapter *adapter)
4726 {
4727         uint16_t phy_status;
4728         uint16_t phy_ctrl;
4729
4730         if ((adapter->hw.phy_type != e1000_phy_igp) || !adapter->hw.autoneg ||
4731            !(adapter->hw.autoneg_advertised & ADVERTISE_1000_FULL))
4732                 return;
4733
4734         if (adapter->smartspeed == 0) {
4735                 /* If Master/Slave config fault is asserted twice,
4736                  * we assume back-to-back */
4737                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_status);
4738                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4739                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_status);
4740                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4741                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_ctrl);
4742                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4743                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4744                         e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL,
4745                                             phy_ctrl);
4746                         adapter->smartspeed++;
4747                         if (!e1000_phy_setup_autoneg(&adapter->hw) &&
4748                            !e1000_read_phy_reg(&adapter->hw, PHY_CTRL,
4749                                                &phy_ctrl)) {
4750                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4751                                              MII_CR_RESTART_AUTO_NEG);
4752                                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL,
4753                                                     phy_ctrl);
4754                         }
4755                 }
4756                 return;
4757         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4758                 /* If still no link, perhaps using 2/3 pair cable */
4759                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_ctrl);
4760                 phy_ctrl |= CR_1000T_MS_ENABLE;
4761                 e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL, phy_ctrl);
4762                 if (!e1000_phy_setup_autoneg(&adapter->hw) &&
4763                    !e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &phy_ctrl)) {
4764                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4765                                      MII_CR_RESTART_AUTO_NEG);
4766                         e1000_write_phy_reg(&adapter->hw, PHY_CTRL, phy_ctrl);
4767                 }
4768         }
4769         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4770         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4771                 adapter->smartspeed = 0;
4772 }
4773
4774 /**
4775  * e1000_ioctl -
4776  * @netdev:
4777  * @ifreq:
4778  * @cmd:
4779  **/
4780
4781 static int
4782 e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4783 {
4784         switch (cmd) {
4785         case SIOCGMIIPHY:
4786         case SIOCGMIIREG:
4787         case SIOCSMIIREG:
4788                 return e1000_mii_ioctl(netdev, ifr, cmd);
4789         default:
4790                 return -EOPNOTSUPP;
4791         }
4792 }
4793
4794 /**
4795  * e1000_mii_ioctl -
4796  * @netdev:
4797  * @ifreq:
4798  * @cmd:
4799  **/
4800
4801 static int
4802 e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4803 {
4804         struct e1000_adapter *adapter = netdev_priv(netdev);
4805         struct mii_ioctl_data *data = if_mii(ifr);
4806         int retval;
4807         uint16_t mii_reg;
4808         uint16_t spddplx;
4809         unsigned long flags;
4810
4811         if (adapter->hw.media_type != e1000_media_type_copper)
4812                 return -EOPNOTSUPP;
4813
4814         switch (cmd) {
4815         case SIOCGMIIPHY:
4816                 data->phy_id = adapter->hw.phy_addr;
4817                 break;
4818         case SIOCGMIIREG:
4819                 if (!capable(CAP_NET_ADMIN))
4820                         return -EPERM;
4821                 spin_lock_irqsave(&adapter->stats_lock, flags);
4822                 if (e1000_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
4823                                    &data->val_out)) {
4824                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4825                         return -EIO;
4826                 }
4827                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4828                 break;
4829         case SIOCSMIIREG:
4830                 if (!capable(CAP_NET_ADMIN))
4831                         return -EPERM;
4832                 if (data->reg_num & ~(0x1F))
4833                         return -EFAULT;
4834                 mii_reg = data->val_in;
4835                 spin_lock_irqsave(&adapter->stats_lock, flags);
4836                 if (e1000_write_phy_reg(&adapter->hw, data->reg_num,
4837                                         mii_reg)) {
4838                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4839                         return -EIO;
4840                 }
4841                 if (adapter->hw.media_type == e1000_media_type_copper) {
4842                         switch (data->reg_num) {
4843                         case PHY_CTRL:
4844                                 if (mii_reg & MII_CR_POWER_DOWN)
4845                                         break;
4846                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4847                                         adapter->hw.autoneg = 1;
4848                                         adapter->hw.autoneg_advertised = 0x2F;
4849                                 } else {
4850                                         if (mii_reg & 0x40)
4851                                                 spddplx = SPEED_1000;
4852                                         else if (mii_reg & 0x2000)
4853                                                 spddplx = SPEED_100;
4854                                         else
4855                                                 spddplx = SPEED_10;
4856                                         spddplx += (mii_reg & 0x100)
4857                                                    ? DUPLEX_FULL :
4858                                                    DUPLEX_HALF;
4859                                         retval = e1000_set_spd_dplx(adapter,
4860                                                                     spddplx);
4861                                         if (retval) {
4862                                                 spin_unlock_irqrestore(
4863                                                         &adapter->stats_lock,
4864                                                         flags);
4865                                                 return retval;
4866                                         }
4867                                 }
4868                                 if (netif_running(adapter->netdev))
4869                                         e1000_reinit_locked(adapter);
4870                                 else
4871                                         e1000_reset(adapter);
4872                                 break;
4873                         case M88E1000_PHY_SPEC_CTRL:
4874                         case M88E1000_EXT_PHY_SPEC_CTRL:
4875                                 if (e1000_phy_reset(&adapter->hw)) {
4876                                         spin_unlock_irqrestore(
4877                                                 &adapter->stats_lock, flags);
4878                                         return -EIO;
4879                                 }
4880                                 break;
4881                         }
4882                 } else {
4883                         switch (data->reg_num) {
4884                         case PHY_CTRL:
4885                                 if (mii_reg & MII_CR_POWER_DOWN)
4886                                         break;
4887                                 if (netif_running(adapter->netdev))
4888                                         e1000_reinit_locked(adapter);
4889                                 else
4890                                         e1000_reset(adapter);
4891                                 break;
4892                         }
4893                 }
4894                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4895                 break;
4896         default:
4897                 return -EOPNOTSUPP;
4898         }
4899         return E1000_SUCCESS;
4900 }
4901
4902 void
4903 e1000_pci_set_mwi(struct e1000_hw *hw)
4904 {
4905         struct e1000_adapter *adapter = hw->back;
4906         int ret_val = pci_set_mwi(adapter->pdev);
4907
4908         if (ret_val)
4909                 DPRINTK(PROBE, ERR, "Error in setting MWI\n");
4910 }
4911
4912 void
4913 e1000_pci_clear_mwi(struct e1000_hw *hw)
4914 {
4915         struct e1000_adapter *adapter = hw->back;
4916
4917         pci_clear_mwi(adapter->pdev);
4918 }
4919
4920 void
4921 e1000_read_pci_cfg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4922 {
4923         struct e1000_adapter *adapter = hw->back;
4924
4925         pci_read_config_word(adapter->pdev, reg, value);
4926 }
4927
4928 void
4929 e1000_write_pci_cfg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4930 {
4931         struct e1000_adapter *adapter = hw->back;
4932
4933         pci_write_config_word(adapter->pdev, reg, *value);
4934 }
4935
4936 int32_t
4937 e1000_read_pcie_cap_reg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4938 {
4939     struct e1000_adapter *adapter = hw->back;
4940     uint16_t cap_offset;
4941
4942     cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
4943     if (!cap_offset)
4944         return -E1000_ERR_CONFIG;
4945
4946     pci_read_config_word(adapter->pdev, cap_offset + reg, value);
4947
4948     return E1000_SUCCESS;
4949 }
4950
4951 void
4952 e1000_io_write(struct e1000_hw *hw, unsigned long port, uint32_t value)
4953 {
4954         outl(value, port);
4955 }
4956
4957 static void
4958 e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
4959 {
4960         struct e1000_adapter *adapter = netdev_priv(netdev);
4961         uint32_t ctrl, rctl;
4962
4963         e1000_irq_disable(adapter);
4964         adapter->vlgrp = grp;
4965
4966         if (grp) {
4967                 /* enable VLAN tag insert/strip */
4968                 ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4969                 ctrl |= E1000_CTRL_VME;
4970                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4971
4972                 if (adapter->hw.mac_type != e1000_ich8lan) {
4973                         /* enable VLAN receive filtering */
4974                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
4975                         rctl |= E1000_RCTL_VFE;
4976                         rctl &= ~E1000_RCTL_CFIEN;
4977                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4978                         e1000_update_mng_vlan(adapter);
4979                 }
4980         } else {
4981                 /* disable VLAN tag insert/strip */
4982                 ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4983                 ctrl &= ~E1000_CTRL_VME;
4984                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4985
4986                 if (adapter->hw.mac_type != e1000_ich8lan) {
4987                         /* disable VLAN filtering */
4988                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
4989                         rctl &= ~E1000_RCTL_VFE;
4990                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4991                         if (adapter->mng_vlan_id !=
4992                             (uint16_t)E1000_MNG_VLAN_NONE) {
4993                                 e1000_vlan_rx_kill_vid(netdev,
4994                                                        adapter->mng_vlan_id);
4995                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4996                         }
4997                 }
4998         }
4999
5000         e1000_irq_enable(adapter);
5001 }
5002
5003 static void
5004 e1000_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid)
5005 {
5006         struct e1000_adapter *adapter = netdev_priv(netdev);
5007         uint32_t vfta, index;
5008
5009         if ((adapter->hw.mng_cookie.status &
5010              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
5011             (vid == adapter->mng_vlan_id))
5012                 return;
5013         /* add VID to filter table */
5014         index = (vid >> 5) & 0x7F;
5015         vfta = E1000_READ_REG_ARRAY(&adapter->hw, VFTA, index);
5016         vfta |= (1 << (vid & 0x1F));
5017         e1000_write_vfta(&adapter->hw, index, vfta);
5018 }
5019
5020 static void
5021 e1000_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid)
5022 {
5023         struct e1000_adapter *adapter = netdev_priv(netdev);
5024         uint32_t vfta, index;
5025
5026         e1000_irq_disable(adapter);
5027
5028         if (adapter->vlgrp)
5029                 adapter->vlgrp->vlan_devices[vid] = NULL;
5030
5031         e1000_irq_enable(adapter);
5032
5033         if ((adapter->hw.mng_cookie.status &
5034              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
5035             (vid == adapter->mng_vlan_id)) {
5036                 /* release control to f/w */
5037                 e1000_release_hw_control(adapter);
5038                 return;
5039         }
5040
5041         /* remove VID from filter table */
5042         index = (vid >> 5) & 0x7F;
5043         vfta = E1000_READ_REG_ARRAY(&adapter->hw, VFTA, index);
5044         vfta &= ~(1 << (vid & 0x1F));
5045         e1000_write_vfta(&adapter->hw, index, vfta);
5046 }
5047
5048 static void
5049 e1000_restore_vlan(struct e1000_adapter *adapter)
5050 {
5051         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
5052
5053         if (adapter->vlgrp) {
5054                 uint16_t vid;
5055                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
5056                         if (!adapter->vlgrp->vlan_devices[vid])
5057                                 continue;
5058                         e1000_vlan_rx_add_vid(adapter->netdev, vid);
5059                 }
5060         }
5061 }
5062
5063 int
5064 e1000_set_spd_dplx(struct e1000_adapter *adapter, uint16_t spddplx)
5065 {
5066         adapter->hw.autoneg = 0;
5067
5068         /* Fiber NICs only allow 1000 gbps Full duplex */
5069         if ((adapter->hw.media_type == e1000_media_type_fiber) &&
5070                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
5071                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
5072                 return -EINVAL;
5073         }
5074
5075         switch (spddplx) {
5076         case SPEED_10 + DUPLEX_HALF:
5077                 adapter->hw.forced_speed_duplex = e1000_10_half;
5078                 break;
5079         case SPEED_10 + DUPLEX_FULL:
5080                 adapter->hw.forced_speed_duplex = e1000_10_full;
5081                 break;
5082         case SPEED_100 + DUPLEX_HALF:
5083                 adapter->hw.forced_speed_duplex = e1000_100_half;
5084                 break;
5085         case SPEED_100 + DUPLEX_FULL:
5086                 adapter->hw.forced_speed_duplex = e1000_100_full;
5087                 break;
5088         case SPEED_1000 + DUPLEX_FULL:
5089                 adapter->hw.autoneg = 1;
5090                 adapter->hw.autoneg_advertised = ADVERTISE_1000_FULL;
5091                 break;
5092         case SPEED_1000 + DUPLEX_HALF: /* not supported */
5093         default:
5094                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
5095                 return -EINVAL;
5096         }
5097         return 0;
5098 }
5099
5100 #ifdef CONFIG_PM
5101 /* Save/restore 16 or 64 dwords of PCI config space depending on which
5102  * bus we're on (PCI(X) vs. PCI-E)
5103  */
5104 #define PCIE_CONFIG_SPACE_LEN 256
5105 #define PCI_CONFIG_SPACE_LEN 64
5106 static int
5107 e1000_pci_save_state(struct e1000_adapter *adapter)
5108 {
5109         struct pci_dev *dev = adapter->pdev;
5110         int size;
5111         int i;
5112
5113         if (adapter->hw.mac_type >= e1000_82571)
5114                 size = PCIE_CONFIG_SPACE_LEN;
5115         else
5116                 size = PCI_CONFIG_SPACE_LEN;
5117
5118         WARN_ON(adapter->config_space != NULL);
5119
5120         adapter->config_space = kmalloc(size, GFP_KERNEL);
5121         if (!adapter->config_space) {
5122                 DPRINTK(PROBE, ERR, "unable to allocate %d bytes\n", size);
5123                 return -ENOMEM;
5124         }
5125         for (i = 0; i < (size / 4); i++)
5126                 pci_read_config_dword(dev, i * 4, &adapter->config_space[i]);
5127         return 0;
5128 }
5129
5130 static void
5131 e1000_pci_restore_state(struct e1000_adapter *adapter)
5132 {
5133         struct pci_dev *dev = adapter->pdev;
5134         int size;
5135         int i;
5136
5137         if (adapter->config_space == NULL)
5138                 return;
5139
5140         if (adapter->hw.mac_type >= e1000_82571)
5141                 size = PCIE_CONFIG_SPACE_LEN;
5142         else
5143                 size = PCI_CONFIG_SPACE_LEN;
5144         for (i = 0; i < (size / 4); i++)
5145                 pci_write_config_dword(dev, i * 4, adapter->config_space[i]);
5146         kfree(adapter->config_space);
5147         adapter->config_space = NULL;
5148         return;
5149 }
5150 #endif /* CONFIG_PM */
5151
5152 static int
5153 e1000_suspend(struct pci_dev *pdev, pm_message_t state)
5154 {
5155         struct net_device *netdev = pci_get_drvdata(pdev);
5156         struct e1000_adapter *adapter = netdev_priv(netdev);
5157         uint32_t ctrl, ctrl_ext, rctl, status;
5158         uint32_t wufc = adapter->wol;
5159 #ifdef CONFIG_PM
5160         int retval = 0;
5161 #endif
5162
5163         netif_device_detach(netdev);
5164
5165         if (netif_running(netdev)) {
5166                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
5167                 e1000_down(adapter);
5168         }
5169
5170 #ifdef CONFIG_PM
5171         /* Implement our own version of pci_save_state(pdev) because pci-
5172          * express adapters have 256-byte config spaces. */
5173         retval = e1000_pci_save_state(adapter);
5174         if (retval)
5175                 return retval;
5176 #endif
5177
5178         status = E1000_READ_REG(&adapter->hw, STATUS);
5179         if (status & E1000_STATUS_LU)
5180                 wufc &= ~E1000_WUFC_LNKC;
5181
5182         if (wufc) {
5183                 e1000_setup_rctl(adapter);
5184                 e1000_set_multi(netdev);
5185
5186                 /* turn on all-multi mode if wake on multicast is enabled */
5187                 if (wufc & E1000_WUFC_MC) {
5188                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
5189                         rctl |= E1000_RCTL_MPE;
5190                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
5191                 }
5192
5193                 if (adapter->hw.mac_type >= e1000_82540) {
5194                         ctrl = E1000_READ_REG(&adapter->hw, CTRL);
5195                         /* advertise wake from D3Cold */
5196                         #define E1000_CTRL_ADVD3WUC 0x00100000
5197                         /* phy power management enable */
5198                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5199                         ctrl |= E1000_CTRL_ADVD3WUC |
5200                                 E1000_CTRL_EN_PHY_PWR_MGMT;
5201                         E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
5202                 }
5203
5204                 if (adapter->hw.media_type == e1000_media_type_fiber ||
5205                    adapter->hw.media_type == e1000_media_type_internal_serdes) {
5206                         /* keep the laser running in D3 */
5207                         ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
5208                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
5209                         E1000_WRITE_REG(&adapter->hw, CTRL_EXT, ctrl_ext);
5210                 }
5211
5212                 /* Allow time for pending master requests to run */
5213                 e1000_disable_pciex_master(&adapter->hw);
5214
5215                 E1000_WRITE_REG(&adapter->hw, WUC, E1000_WUC_PME_EN);
5216                 E1000_WRITE_REG(&adapter->hw, WUFC, wufc);
5217                 pci_enable_wake(pdev, PCI_D3hot, 1);
5218                 pci_enable_wake(pdev, PCI_D3cold, 1);
5219         } else {
5220                 E1000_WRITE_REG(&adapter->hw, WUC, 0);
5221                 E1000_WRITE_REG(&adapter->hw, WUFC, 0);
5222                 pci_enable_wake(pdev, PCI_D3hot, 0);
5223                 pci_enable_wake(pdev, PCI_D3cold, 0);
5224         }
5225
5226         e1000_release_manageability(adapter);
5227
5228         /* make sure adapter isn't asleep if manageability is enabled */
5229         if (adapter->en_mng_pt) {
5230                 pci_enable_wake(pdev, PCI_D3hot, 1);
5231                 pci_enable_wake(pdev, PCI_D3cold, 1);
5232         }
5233
5234         if (adapter->hw.phy_type == e1000_phy_igp_3)
5235                 e1000_phy_powerdown_workaround(&adapter->hw);
5236
5237         if (netif_running(netdev))
5238                 e1000_free_irq(adapter);
5239
5240         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
5241          * would have already happened in close and is redundant. */
5242         e1000_release_hw_control(adapter);
5243
5244         pci_disable_device(pdev);
5245
5246         pci_set_power_state(pdev, pci_choose_state(pdev, state));
5247
5248         return 0;
5249 }
5250
5251 #ifdef CONFIG_PM
5252 static int
5253 e1000_resume(struct pci_dev *pdev)
5254 {
5255         struct net_device *netdev = pci_get_drvdata(pdev);
5256         struct e1000_adapter *adapter = netdev_priv(netdev);
5257         uint32_t err;
5258
5259         pci_set_power_state(pdev, PCI_D0);
5260         e1000_pci_restore_state(adapter);
5261         if ((err = pci_enable_device(pdev))) {
5262                 printk(KERN_ERR "e1000: Cannot enable PCI device from suspend\n");
5263                 return err;
5264         }
5265         pci_set_master(pdev);
5266
5267         pci_enable_wake(pdev, PCI_D3hot, 0);
5268         pci_enable_wake(pdev, PCI_D3cold, 0);
5269
5270         if (netif_running(netdev) && (err = e1000_request_irq(adapter)))
5271                 return err;
5272
5273         e1000_power_up_phy(adapter);
5274         e1000_reset(adapter);
5275         E1000_WRITE_REG(&adapter->hw, WUS, ~0);
5276
5277         e1000_init_manageability(adapter);
5278
5279         if (netif_running(netdev))
5280                 e1000_up(adapter);
5281
5282         netif_device_attach(netdev);
5283
5284         /* If the controller is 82573 and f/w is AMT, do not set
5285          * DRV_LOAD until the interface is up.  For all other cases,
5286          * let the f/w know that the h/w is now under the control
5287          * of the driver. */
5288         if (adapter->hw.mac_type != e1000_82573 ||
5289             !e1000_check_mng_mode(&adapter->hw))
5290                 e1000_get_hw_control(adapter);
5291
5292         return 0;
5293 }
5294 #endif
5295
5296 static void e1000_shutdown(struct pci_dev *pdev)
5297 {
5298         e1000_suspend(pdev, PMSG_SUSPEND);
5299 }
5300
5301 #ifdef CONFIG_NET_POLL_CONTROLLER
5302 /*
5303  * Polling 'interrupt' - used by things like netconsole to send skbs
5304  * without having to re-enable interrupts. It's not called while
5305  * the interrupt routine is executing.
5306  */
5307 static void
5308 e1000_netpoll(struct net_device *netdev)
5309 {
5310         struct e1000_adapter *adapter = netdev_priv(netdev);
5311
5312         disable_irq(adapter->pdev->irq);
5313         e1000_intr(adapter->pdev->irq, netdev);
5314         e1000_clean_tx_irq(adapter, adapter->tx_ring);
5315 #ifndef CONFIG_E1000_NAPI
5316         adapter->clean_rx(adapter, adapter->rx_ring);
5317 #endif
5318         enable_irq(adapter->pdev->irq);
5319 }
5320 #endif
5321
5322 /**
5323  * e1000_io_error_detected - called when PCI error is detected
5324  * @pdev: Pointer to PCI device
5325  * @state: The current pci conneection state
5326  *
5327  * This function is called after a PCI bus error affecting
5328  * this device has been detected.
5329  */
5330 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
5331 {
5332         struct net_device *netdev = pci_get_drvdata(pdev);
5333         struct e1000_adapter *adapter = netdev->priv;
5334
5335         netif_device_detach(netdev);
5336
5337         if (netif_running(netdev))
5338                 e1000_down(adapter);
5339         pci_disable_device(pdev);
5340
5341         /* Request a slot slot reset. */
5342         return PCI_ERS_RESULT_NEED_RESET;
5343 }
5344
5345 /**
5346  * e1000_io_slot_reset - called after the pci bus has been reset.
5347  * @pdev: Pointer to PCI device
5348  *
5349  * Restart the card from scratch, as if from a cold-boot. Implementation
5350  * resembles the first-half of the e1000_resume routine.
5351  */
5352 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5353 {
5354         struct net_device *netdev = pci_get_drvdata(pdev);
5355         struct e1000_adapter *adapter = netdev->priv;
5356
5357         if (pci_enable_device(pdev)) {
5358                 printk(KERN_ERR "e1000: Cannot re-enable PCI device after reset.\n");
5359                 return PCI_ERS_RESULT_DISCONNECT;
5360         }
5361         pci_set_master(pdev);
5362
5363         pci_enable_wake(pdev, PCI_D3hot, 0);
5364         pci_enable_wake(pdev, PCI_D3cold, 0);
5365
5366         e1000_reset(adapter);
5367         E1000_WRITE_REG(&adapter->hw, WUS, ~0);
5368
5369         return PCI_ERS_RESULT_RECOVERED;
5370 }
5371
5372 /**
5373  * e1000_io_resume - called when traffic can start flowing again.
5374  * @pdev: Pointer to PCI device
5375  *
5376  * This callback is called when the error recovery driver tells us that
5377  * its OK to resume normal operation. Implementation resembles the
5378  * second-half of the e1000_resume routine.
5379  */
5380 static void e1000_io_resume(struct pci_dev *pdev)
5381 {
5382         struct net_device *netdev = pci_get_drvdata(pdev);
5383         struct e1000_adapter *adapter = netdev->priv;
5384
5385         e1000_init_manageability(adapter);
5386
5387         if (netif_running(netdev)) {
5388                 if (e1000_up(adapter)) {
5389                         printk("e1000: can't bring device back up after reset\n");
5390                         return;
5391                 }
5392         }
5393
5394         netif_device_attach(netdev);
5395
5396         /* If the controller is 82573 and f/w is AMT, do not set
5397          * DRV_LOAD until the interface is up.  For all other cases,
5398          * let the f/w know that the h/w is now under the control
5399          * of the driver. */
5400         if (adapter->hw.mac_type != e1000_82573 ||
5401             !e1000_check_mng_mode(&adapter->hw))
5402                 e1000_get_hw_control(adapter);
5403
5404 }
5405
5406 /* e1000_main.c */