Merge master.kernel.org:/pub/scm/linux/kernel/git/lethal/sh-2.6
[linux-2.6] / drivers / net / e1000 / e1000_main.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2006 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include "e1000.h"
30 #include <net/ip6_checksum.h>
31
32 char e1000_driver_name[] = "e1000";
33 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
34 #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 static irqreturn_t e1000_intr_msi(int irq, void *data);
162 static boolean_t e1000_clean_tx_irq(struct e1000_adapter *adapter,
163                                     struct e1000_tx_ring *tx_ring);
164 #ifdef CONFIG_E1000_NAPI
165 static int e1000_clean(struct net_device *poll_dev, int *budget);
166 static boolean_t e1000_clean_rx_irq(struct e1000_adapter *adapter,
167                                     struct e1000_rx_ring *rx_ring,
168                                     int *work_done, int work_to_do);
169 static boolean_t e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
170                                        struct e1000_rx_ring *rx_ring,
171                                        int *work_done, int work_to_do);
172 #else
173 static boolean_t e1000_clean_rx_irq(struct e1000_adapter *adapter,
174                                     struct e1000_rx_ring *rx_ring);
175 static boolean_t e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
176                                        struct e1000_rx_ring *rx_ring);
177 #endif
178 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
179                                    struct e1000_rx_ring *rx_ring,
180                                    int cleaned_count);
181 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
182                                       struct e1000_rx_ring *rx_ring,
183                                       int cleaned_count);
184 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
185 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
186                            int cmd);
187 void e1000_set_ethtool_ops(struct net_device *netdev);
188 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
189 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
190 static void e1000_tx_timeout(struct net_device *dev);
191 static void e1000_reset_task(struct work_struct *work);
192 static void e1000_smartspeed(struct e1000_adapter *adapter);
193 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
194                                        struct sk_buff *skb);
195
196 static void e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp);
197 static void e1000_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid);
198 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid);
199 static void e1000_restore_vlan(struct e1000_adapter *adapter);
200
201 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
202 #ifdef CONFIG_PM
203 static int e1000_resume(struct pci_dev *pdev);
204 #endif
205 static void e1000_shutdown(struct pci_dev *pdev);
206
207 #ifdef CONFIG_NET_POLL_CONTROLLER
208 /* for netdump / net console */
209 static void e1000_netpoll (struct net_device *netdev);
210 #endif
211
212 extern void e1000_check_options(struct e1000_adapter *adapter);
213
214 #define COPYBREAK_DEFAULT 256
215 static unsigned int copybreak __read_mostly = COPYBREAK_DEFAULT;
216 module_param(copybreak, uint, 0644);
217 MODULE_PARM_DESC(copybreak,
218         "Maximum size of packet that is copied to a new buffer on receive");
219
220 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
221                      pci_channel_state_t state);
222 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
223 static void e1000_io_resume(struct pci_dev *pdev);
224
225 static struct pci_error_handlers e1000_err_handler = {
226         .error_detected = e1000_io_error_detected,
227         .slot_reset = e1000_io_slot_reset,
228         .resume = e1000_io_resume,
229 };
230
231 static struct pci_driver e1000_driver = {
232         .name     = e1000_driver_name,
233         .id_table = e1000_pci_tbl,
234         .probe    = e1000_probe,
235         .remove   = __devexit_p(e1000_remove),
236 #ifdef CONFIG_PM
237         /* Power Managment Hooks */
238         .suspend  = e1000_suspend,
239         .resume   = e1000_resume,
240 #endif
241         .shutdown = e1000_shutdown,
242         .err_handler = &e1000_err_handler
243 };
244
245 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
246 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
247 MODULE_LICENSE("GPL");
248 MODULE_VERSION(DRV_VERSION);
249
250 static int debug = NETIF_MSG_DRV | NETIF_MSG_PROBE;
251 module_param(debug, int, 0);
252 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
253
254 /**
255  * e1000_init_module - Driver Registration Routine
256  *
257  * e1000_init_module is the first routine called when the driver is
258  * loaded. All it does is register with the PCI subsystem.
259  **/
260
261 static int __init
262 e1000_init_module(void)
263 {
264         int ret;
265         printk(KERN_INFO "%s - version %s\n",
266                e1000_driver_string, e1000_driver_version);
267
268         printk(KERN_INFO "%s\n", e1000_copyright);
269
270         ret = pci_register_driver(&e1000_driver);
271         if (copybreak != COPYBREAK_DEFAULT) {
272                 if (copybreak == 0)
273                         printk(KERN_INFO "e1000: copybreak disabled\n");
274                 else
275                         printk(KERN_INFO "e1000: copybreak enabled for "
276                                "packets <= %u bytes\n", copybreak);
277         }
278         return ret;
279 }
280
281 module_init(e1000_init_module);
282
283 /**
284  * e1000_exit_module - Driver Exit Cleanup Routine
285  *
286  * e1000_exit_module is called just before the driver is removed
287  * from memory.
288  **/
289
290 static void __exit
291 e1000_exit_module(void)
292 {
293         pci_unregister_driver(&e1000_driver);
294 }
295
296 module_exit(e1000_exit_module);
297
298 static int e1000_request_irq(struct e1000_adapter *adapter)
299 {
300         struct net_device *netdev = adapter->netdev;
301         void (*handler) = &e1000_intr;
302         int irq_flags = IRQF_SHARED;
303         int err;
304
305         if (adapter->hw.mac_type >= e1000_82571) {
306                 adapter->have_msi = !pci_enable_msi(adapter->pdev);
307                 if (adapter->have_msi) {
308                         handler = &e1000_intr_msi;
309                         irq_flags = 0;
310                 }
311         }
312
313         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
314                           netdev);
315         if (err) {
316                 if (adapter->have_msi)
317                         pci_disable_msi(adapter->pdev);
318                 DPRINTK(PROBE, ERR,
319                         "Unable to allocate interrupt Error: %d\n", err);
320         }
321
322         return err;
323 }
324
325 static void e1000_free_irq(struct e1000_adapter *adapter)
326 {
327         struct net_device *netdev = adapter->netdev;
328
329         free_irq(adapter->pdev->irq, netdev);
330
331         if (adapter->have_msi)
332                 pci_disable_msi(adapter->pdev);
333 }
334
335 /**
336  * e1000_irq_disable - Mask off interrupt generation on the NIC
337  * @adapter: board private structure
338  **/
339
340 static void
341 e1000_irq_disable(struct e1000_adapter *adapter)
342 {
343         atomic_inc(&adapter->irq_sem);
344         E1000_WRITE_REG(&adapter->hw, IMC, ~0);
345         E1000_WRITE_FLUSH(&adapter->hw);
346         synchronize_irq(adapter->pdev->irq);
347 }
348
349 /**
350  * e1000_irq_enable - Enable default interrupt generation settings
351  * @adapter: board private structure
352  **/
353
354 static void
355 e1000_irq_enable(struct e1000_adapter *adapter)
356 {
357         if (likely(atomic_dec_and_test(&adapter->irq_sem))) {
358                 E1000_WRITE_REG(&adapter->hw, IMS, IMS_ENABLE_MASK);
359                 E1000_WRITE_FLUSH(&adapter->hw);
360         }
361 }
362
363 static void
364 e1000_update_mng_vlan(struct e1000_adapter *adapter)
365 {
366         struct net_device *netdev = adapter->netdev;
367         uint16_t vid = adapter->hw.mng_cookie.vlan_id;
368         uint16_t old_vid = adapter->mng_vlan_id;
369         if (adapter->vlgrp) {
370                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
371                         if (adapter->hw.mng_cookie.status &
372                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
373                                 e1000_vlan_rx_add_vid(netdev, vid);
374                                 adapter->mng_vlan_id = vid;
375                         } else
376                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
377
378                         if ((old_vid != (uint16_t)E1000_MNG_VLAN_NONE) &&
379                                         (vid != old_vid) &&
380                             !vlan_group_get_device(adapter->vlgrp, old_vid))
381                                 e1000_vlan_rx_kill_vid(netdev, old_vid);
382                 } else
383                         adapter->mng_vlan_id = vid;
384         }
385 }
386
387 /**
388  * e1000_release_hw_control - release control of the h/w to f/w
389  * @adapter: address of board private structure
390  *
391  * e1000_release_hw_control resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
392  * For ASF and Pass Through versions of f/w this means that the
393  * driver is no longer loaded. For AMT version (only with 82573) i
394  * of the f/w this means that the network i/f is closed.
395  *
396  **/
397
398 static void
399 e1000_release_hw_control(struct e1000_adapter *adapter)
400 {
401         uint32_t ctrl_ext;
402         uint32_t swsm;
403
404         /* Let firmware taken over control of h/w */
405         switch (adapter->hw.mac_type) {
406         case e1000_82573:
407                 swsm = E1000_READ_REG(&adapter->hw, SWSM);
408                 E1000_WRITE_REG(&adapter->hw, SWSM,
409                                 swsm & ~E1000_SWSM_DRV_LOAD);
410                 break;
411         case e1000_82571:
412         case e1000_82572:
413         case e1000_80003es2lan:
414         case e1000_ich8lan:
415                 ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
416                 E1000_WRITE_REG(&adapter->hw, CTRL_EXT,
417                                 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
418                 break;
419         default:
420                 break;
421         }
422 }
423
424 /**
425  * e1000_get_hw_control - get control of the h/w from f/w
426  * @adapter: address of board private structure
427  *
428  * e1000_get_hw_control sets {CTRL_EXT|FWSM}:DRV_LOAD bit.
429  * For ASF and Pass Through versions of f/w this means that
430  * the driver is loaded. For AMT version (only with 82573)
431  * of the f/w this means that the network i/f is open.
432  *
433  **/
434
435 static void
436 e1000_get_hw_control(struct e1000_adapter *adapter)
437 {
438         uint32_t ctrl_ext;
439         uint32_t swsm;
440
441         /* Let firmware know the driver has taken over */
442         switch (adapter->hw.mac_type) {
443         case e1000_82573:
444                 swsm = E1000_READ_REG(&adapter->hw, SWSM);
445                 E1000_WRITE_REG(&adapter->hw, SWSM,
446                                 swsm | E1000_SWSM_DRV_LOAD);
447                 break;
448         case e1000_82571:
449         case e1000_82572:
450         case e1000_80003es2lan:
451         case e1000_ich8lan:
452                 ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
453                 E1000_WRITE_REG(&adapter->hw, CTRL_EXT,
454                                 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
455                 break;
456         default:
457                 break;
458         }
459 }
460
461 static void
462 e1000_init_manageability(struct e1000_adapter *adapter)
463 {
464         if (adapter->en_mng_pt) {
465                 uint32_t manc = E1000_READ_REG(&adapter->hw, MANC);
466
467                 /* disable hardware interception of ARP */
468                 manc &= ~(E1000_MANC_ARP_EN);
469
470                 /* enable receiving management packets to the host */
471                 /* this will probably generate destination unreachable messages
472                  * from the host OS, but the packets will be handled on SMBUS */
473                 if (adapter->hw.has_manc2h) {
474                         uint32_t manc2h = E1000_READ_REG(&adapter->hw, MANC2H);
475
476                         manc |= E1000_MANC_EN_MNG2HOST;
477 #define E1000_MNG2HOST_PORT_623 (1 << 5)
478 #define E1000_MNG2HOST_PORT_664 (1 << 6)
479                         manc2h |= E1000_MNG2HOST_PORT_623;
480                         manc2h |= E1000_MNG2HOST_PORT_664;
481                         E1000_WRITE_REG(&adapter->hw, MANC2H, manc2h);
482                 }
483
484                 E1000_WRITE_REG(&adapter->hw, MANC, manc);
485         }
486 }
487
488 static void
489 e1000_release_manageability(struct e1000_adapter *adapter)
490 {
491         if (adapter->en_mng_pt) {
492                 uint32_t manc = E1000_READ_REG(&adapter->hw, MANC);
493
494                 /* re-enable hardware interception of ARP */
495                 manc |= E1000_MANC_ARP_EN;
496
497                 if (adapter->hw.has_manc2h)
498                         manc &= ~E1000_MANC_EN_MNG2HOST;
499
500                 /* don't explicitly have to mess with MANC2H since
501                  * MANC has an enable disable that gates MANC2H */
502
503                 E1000_WRITE_REG(&adapter->hw, MANC, manc);
504         }
505 }
506
507 /**
508  * e1000_configure - configure the hardware for RX and TX
509  * @adapter = private board structure
510  **/
511 static void e1000_configure(struct e1000_adapter *adapter)
512 {
513         struct net_device *netdev = adapter->netdev;
514         int i;
515
516         e1000_set_multi(netdev);
517
518         e1000_restore_vlan(adapter);
519         e1000_init_manageability(adapter);
520
521         e1000_configure_tx(adapter);
522         e1000_setup_rctl(adapter);
523         e1000_configure_rx(adapter);
524         /* call E1000_DESC_UNUSED which always leaves
525          * at least 1 descriptor unused to make sure
526          * next_to_use != next_to_clean */
527         for (i = 0; i < adapter->num_rx_queues; i++) {
528                 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
529                 adapter->alloc_rx_buf(adapter, ring,
530                                       E1000_DESC_UNUSED(ring));
531         }
532
533         adapter->tx_queue_len = netdev->tx_queue_len;
534 }
535
536 int e1000_up(struct e1000_adapter *adapter)
537 {
538         /* hardware has been reset, we need to reload some things */
539         e1000_configure(adapter);
540
541         clear_bit(__E1000_DOWN, &adapter->flags);
542
543 #ifdef CONFIG_E1000_NAPI
544         netif_poll_enable(adapter->netdev);
545 #endif
546         e1000_irq_enable(adapter);
547
548         /* fire a link change interrupt to start the watchdog */
549         E1000_WRITE_REG(&adapter->hw, ICS, E1000_ICS_LSC);
550         return 0;
551 }
552
553 /**
554  * e1000_power_up_phy - restore link in case the phy was powered down
555  * @adapter: address of board private structure
556  *
557  * The phy may be powered down to save power and turn off link when the
558  * driver is unloaded and wake on lan is not enabled (among others)
559  * *** this routine MUST be followed by a call to e1000_reset ***
560  *
561  **/
562
563 void e1000_power_up_phy(struct e1000_adapter *adapter)
564 {
565         uint16_t mii_reg = 0;
566
567         /* Just clear the power down bit to wake the phy back up */
568         if (adapter->hw.media_type == e1000_media_type_copper) {
569                 /* according to the manual, the phy will retain its
570                  * settings across a power-down/up cycle */
571                 e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &mii_reg);
572                 mii_reg &= ~MII_CR_POWER_DOWN;
573                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL, mii_reg);
574         }
575 }
576
577 static void e1000_power_down_phy(struct e1000_adapter *adapter)
578 {
579         /* Power down the PHY so no link is implied when interface is down *
580          * The PHY cannot be powered down if any of the following is TRUE *
581          * (a) WoL is enabled
582          * (b) AMT is active
583          * (c) SoL/IDER session is active */
584         if (!adapter->wol && adapter->hw.mac_type >= e1000_82540 &&
585            adapter->hw.media_type == e1000_media_type_copper) {
586                 uint16_t mii_reg = 0;
587
588                 switch (adapter->hw.mac_type) {
589                 case e1000_82540:
590                 case e1000_82545:
591                 case e1000_82545_rev_3:
592                 case e1000_82546:
593                 case e1000_82546_rev_3:
594                 case e1000_82541:
595                 case e1000_82541_rev_2:
596                 case e1000_82547:
597                 case e1000_82547_rev_2:
598                         if (E1000_READ_REG(&adapter->hw, MANC) &
599                             E1000_MANC_SMBUS_EN)
600                                 goto out;
601                         break;
602                 case e1000_82571:
603                 case e1000_82572:
604                 case e1000_82573:
605                 case e1000_80003es2lan:
606                 case e1000_ich8lan:
607                         if (e1000_check_mng_mode(&adapter->hw) ||
608                             e1000_check_phy_reset_block(&adapter->hw))
609                                 goto out;
610                         break;
611                 default:
612                         goto out;
613                 }
614                 e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &mii_reg);
615                 mii_reg |= MII_CR_POWER_DOWN;
616                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL, mii_reg);
617                 mdelay(1);
618         }
619 out:
620         return;
621 }
622
623 void
624 e1000_down(struct e1000_adapter *adapter)
625 {
626         struct net_device *netdev = adapter->netdev;
627
628         /* signal that we're down so the interrupt handler does not
629          * reschedule our watchdog timer */
630         set_bit(__E1000_DOWN, &adapter->flags);
631
632 #ifdef CONFIG_E1000_NAPI
633         netif_poll_disable(netdev);
634 #endif
635         e1000_irq_disable(adapter);
636
637         del_timer_sync(&adapter->tx_fifo_stall_timer);
638         del_timer_sync(&adapter->watchdog_timer);
639         del_timer_sync(&adapter->phy_info_timer);
640
641         netdev->tx_queue_len = adapter->tx_queue_len;
642         adapter->link_speed = 0;
643         adapter->link_duplex = 0;
644         netif_carrier_off(netdev);
645         netif_stop_queue(netdev);
646
647         e1000_reset(adapter);
648         e1000_clean_all_tx_rings(adapter);
649         e1000_clean_all_rx_rings(adapter);
650 }
651
652 void
653 e1000_reinit_locked(struct e1000_adapter *adapter)
654 {
655         WARN_ON(in_interrupt());
656         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
657                 msleep(1);
658         e1000_down(adapter);
659         e1000_up(adapter);
660         clear_bit(__E1000_RESETTING, &adapter->flags);
661 }
662
663 void
664 e1000_reset(struct e1000_adapter *adapter)
665 {
666         uint32_t pba = 0, tx_space, min_tx_space, min_rx_space;
667         uint16_t fc_high_water_mark = E1000_FC_HIGH_DIFF;
668         boolean_t legacy_pba_adjust = FALSE;
669
670         /* Repartition Pba for greater than 9k mtu
671          * To take effect CTRL.RST is required.
672          */
673
674         switch (adapter->hw.mac_type) {
675         case e1000_82542_rev2_0:
676         case e1000_82542_rev2_1:
677         case e1000_82543:
678         case e1000_82544:
679         case e1000_82540:
680         case e1000_82541:
681         case e1000_82541_rev_2:
682                 legacy_pba_adjust = TRUE;
683                 pba = E1000_PBA_48K;
684                 break;
685         case e1000_82545:
686         case e1000_82545_rev_3:
687         case e1000_82546:
688         case e1000_82546_rev_3:
689                 pba = E1000_PBA_48K;
690                 break;
691         case e1000_82547:
692         case e1000_82547_rev_2:
693                 legacy_pba_adjust = TRUE;
694                 pba = E1000_PBA_30K;
695                 break;
696         case e1000_82571:
697         case e1000_82572:
698         case e1000_80003es2lan:
699                 pba = E1000_PBA_38K;
700                 break;
701         case e1000_82573:
702                 pba = E1000_PBA_20K;
703                 break;
704         case e1000_ich8lan:
705                 pba = E1000_PBA_8K;
706         case e1000_undefined:
707         case e1000_num_macs:
708                 break;
709         }
710
711         if (legacy_pba_adjust == TRUE) {
712                 if (adapter->netdev->mtu > E1000_RXBUFFER_8192)
713                         pba -= 8; /* allocate more FIFO for Tx */
714
715                 if (adapter->hw.mac_type == e1000_82547) {
716                         adapter->tx_fifo_head = 0;
717                         adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
718                         adapter->tx_fifo_size =
719                                 (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
720                         atomic_set(&adapter->tx_fifo_stall, 0);
721                 }
722         } else if (adapter->hw.max_frame_size > MAXIMUM_ETHERNET_FRAME_SIZE) {
723                 /* adjust PBA for jumbo frames */
724                 E1000_WRITE_REG(&adapter->hw, PBA, pba);
725
726                 /* To maintain wire speed transmits, the Tx FIFO should be
727                  * large enough to accomodate two full transmit packets,
728                  * rounded up to the next 1KB and expressed in KB.  Likewise,
729                  * the Rx FIFO should be large enough to accomodate at least
730                  * one full receive packet and is similarly rounded up and
731                  * expressed in KB. */
732                 pba = E1000_READ_REG(&adapter->hw, PBA);
733                 /* upper 16 bits has Tx packet buffer allocation size in KB */
734                 tx_space = pba >> 16;
735                 /* lower 16 bits has Rx packet buffer allocation size in KB */
736                 pba &= 0xffff;
737                 /* don't include ethernet FCS because hardware appends/strips */
738                 min_rx_space = adapter->netdev->mtu + ENET_HEADER_SIZE +
739                                VLAN_TAG_SIZE;
740                 min_tx_space = min_rx_space;
741                 min_tx_space *= 2;
742                 min_tx_space = ALIGN(min_tx_space, 1024);
743                 min_tx_space >>= 10;
744                 min_rx_space = ALIGN(min_rx_space, 1024);
745                 min_rx_space >>= 10;
746
747                 /* If current Tx allocation is less than the min Tx FIFO size,
748                  * and the min Tx FIFO size is less than the current Rx FIFO
749                  * allocation, take space away from current Rx allocation */
750                 if (tx_space < min_tx_space &&
751                     ((min_tx_space - tx_space) < pba)) {
752                         pba = pba - (min_tx_space - tx_space);
753
754                         /* PCI/PCIx hardware has PBA alignment constraints */
755                         switch (adapter->hw.mac_type) {
756                         case e1000_82545 ... e1000_82546_rev_3:
757                                 pba &= ~(E1000_PBA_8K - 1);
758                                 break;
759                         default:
760                                 break;
761                         }
762
763                         /* if short on rx space, rx wins and must trump tx
764                          * adjustment or use Early Receive if available */
765                         if (pba < min_rx_space) {
766                                 switch (adapter->hw.mac_type) {
767                                 case e1000_82573:
768                                         /* ERT enabled in e1000_configure_rx */
769                                         break;
770                                 default:
771                                         pba = min_rx_space;
772                                         break;
773                                 }
774                         }
775                 }
776         }
777
778         E1000_WRITE_REG(&adapter->hw, PBA, pba);
779
780         /* flow control settings */
781         /* Set the FC high water mark to 90% of the FIFO size.
782          * Required to clear last 3 LSB */
783         fc_high_water_mark = ((pba * 9216)/10) & 0xFFF8;
784         /* We can't use 90% on small FIFOs because the remainder
785          * would be less than 1 full frame.  In this case, we size
786          * it to allow at least a full frame above the high water
787          *  mark. */
788         if (pba < E1000_PBA_16K)
789                 fc_high_water_mark = (pba * 1024) - 1600;
790
791         adapter->hw.fc_high_water = fc_high_water_mark;
792         adapter->hw.fc_low_water = fc_high_water_mark - 8;
793         if (adapter->hw.mac_type == e1000_80003es2lan)
794                 adapter->hw.fc_pause_time = 0xFFFF;
795         else
796                 adapter->hw.fc_pause_time = E1000_FC_PAUSE_TIME;
797         adapter->hw.fc_send_xon = 1;
798         adapter->hw.fc = adapter->hw.original_fc;
799
800         /* Allow time for pending master requests to run */
801         e1000_reset_hw(&adapter->hw);
802         if (adapter->hw.mac_type >= e1000_82544)
803                 E1000_WRITE_REG(&adapter->hw, WUC, 0);
804
805         if (e1000_init_hw(&adapter->hw))
806                 DPRINTK(PROBE, ERR, "Hardware Error\n");
807         e1000_update_mng_vlan(adapter);
808
809         /* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
810         if (adapter->hw.mac_type >= e1000_82544 &&
811             adapter->hw.mac_type <= e1000_82547_rev_2 &&
812             adapter->hw.autoneg == 1 &&
813             adapter->hw.autoneg_advertised == ADVERTISE_1000_FULL) {
814                 uint32_t ctrl = E1000_READ_REG(&adapter->hw, CTRL);
815                 /* clear phy power management bit if we are in gig only mode,
816                  * which if enabled will attempt negotiation to 100Mb, which
817                  * can cause a loss of link at power off or driver unload */
818                 ctrl &= ~E1000_CTRL_SWDPIN3;
819                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
820         }
821
822         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
823         E1000_WRITE_REG(&adapter->hw, VET, ETHERNET_IEEE_VLAN_TYPE);
824
825         e1000_reset_adaptive(&adapter->hw);
826         e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
827
828         if (!adapter->smart_power_down &&
829             (adapter->hw.mac_type == e1000_82571 ||
830              adapter->hw.mac_type == e1000_82572)) {
831                 uint16_t phy_data = 0;
832                 /* speed up time to link by disabling smart power down, ignore
833                  * the return value of this function because there is nothing
834                  * different we would do if it failed */
835                 e1000_read_phy_reg(&adapter->hw, IGP02E1000_PHY_POWER_MGMT,
836                                    &phy_data);
837                 phy_data &= ~IGP02E1000_PM_SPD;
838                 e1000_write_phy_reg(&adapter->hw, IGP02E1000_PHY_POWER_MGMT,
839                                     phy_data);
840         }
841
842         e1000_release_manageability(adapter);
843 }
844
845 /**
846  * e1000_probe - Device Initialization Routine
847  * @pdev: PCI device information struct
848  * @ent: entry in e1000_pci_tbl
849  *
850  * Returns 0 on success, negative on failure
851  *
852  * e1000_probe initializes an adapter identified by a pci_dev structure.
853  * The OS initialization, configuring of the adapter private structure,
854  * and a hardware reset occur.
855  **/
856
857 static int __devinit
858 e1000_probe(struct pci_dev *pdev,
859             const struct pci_device_id *ent)
860 {
861         struct net_device *netdev;
862         struct e1000_adapter *adapter;
863         unsigned long mmio_start, mmio_len;
864         unsigned long flash_start, flash_len;
865
866         static int cards_found = 0;
867         static int global_quad_port_a = 0; /* global ksp3 port a indication */
868         int i, err, pci_using_dac;
869         uint16_t eeprom_data = 0;
870         uint16_t eeprom_apme_mask = E1000_EEPROM_APME;
871         if ((err = pci_enable_device(pdev)))
872                 return err;
873
874         if (!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK)) &&
875             !(err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK))) {
876                 pci_using_dac = 1;
877         } else {
878                 if ((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK)) &&
879                     (err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK))) {
880                         E1000_ERR("No usable DMA configuration, aborting\n");
881                         goto err_dma;
882                 }
883                 pci_using_dac = 0;
884         }
885
886         if ((err = pci_request_regions(pdev, e1000_driver_name)))
887                 goto err_pci_reg;
888
889         pci_set_master(pdev);
890
891         err = -ENOMEM;
892         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
893         if (!netdev)
894                 goto err_alloc_etherdev;
895
896         SET_MODULE_OWNER(netdev);
897         SET_NETDEV_DEV(netdev, &pdev->dev);
898
899         pci_set_drvdata(pdev, netdev);
900         adapter = netdev_priv(netdev);
901         adapter->netdev = netdev;
902         adapter->pdev = pdev;
903         adapter->hw.back = adapter;
904         adapter->msg_enable = (1 << debug) - 1;
905
906         mmio_start = pci_resource_start(pdev, BAR_0);
907         mmio_len = pci_resource_len(pdev, BAR_0);
908
909         err = -EIO;
910         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
911         if (!adapter->hw.hw_addr)
912                 goto err_ioremap;
913
914         for (i = BAR_1; i <= BAR_5; i++) {
915                 if (pci_resource_len(pdev, i) == 0)
916                         continue;
917                 if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
918                         adapter->hw.io_base = pci_resource_start(pdev, i);
919                         break;
920                 }
921         }
922
923         netdev->open = &e1000_open;
924         netdev->stop = &e1000_close;
925         netdev->hard_start_xmit = &e1000_xmit_frame;
926         netdev->get_stats = &e1000_get_stats;
927         netdev->set_multicast_list = &e1000_set_multi;
928         netdev->set_mac_address = &e1000_set_mac;
929         netdev->change_mtu = &e1000_change_mtu;
930         netdev->do_ioctl = &e1000_ioctl;
931         e1000_set_ethtool_ops(netdev);
932         netdev->tx_timeout = &e1000_tx_timeout;
933         netdev->watchdog_timeo = 5 * HZ;
934 #ifdef CONFIG_E1000_NAPI
935         netdev->poll = &e1000_clean;
936         netdev->weight = 64;
937 #endif
938         netdev->vlan_rx_register = e1000_vlan_rx_register;
939         netdev->vlan_rx_add_vid = e1000_vlan_rx_add_vid;
940         netdev->vlan_rx_kill_vid = e1000_vlan_rx_kill_vid;
941 #ifdef CONFIG_NET_POLL_CONTROLLER
942         netdev->poll_controller = e1000_netpoll;
943 #endif
944         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
945
946         netdev->mem_start = mmio_start;
947         netdev->mem_end = mmio_start + mmio_len;
948         netdev->base_addr = adapter->hw.io_base;
949
950         adapter->bd_number = cards_found;
951
952         /* setup the private structure */
953
954         if ((err = e1000_sw_init(adapter)))
955                 goto err_sw_init;
956
957         err = -EIO;
958         /* Flash BAR mapping must happen after e1000_sw_init
959          * because it depends on mac_type */
960         if ((adapter->hw.mac_type == e1000_ich8lan) &&
961            (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
962                 flash_start = pci_resource_start(pdev, 1);
963                 flash_len = pci_resource_len(pdev, 1);
964                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
965                 if (!adapter->hw.flash_address)
966                         goto err_flashmap;
967         }
968
969         if (e1000_check_phy_reset_block(&adapter->hw))
970                 DPRINTK(PROBE, INFO, "PHY reset is blocked due to SOL/IDER session.\n");
971
972         if (adapter->hw.mac_type >= e1000_82543) {
973                 netdev->features = NETIF_F_SG |
974                                    NETIF_F_HW_CSUM |
975                                    NETIF_F_HW_VLAN_TX |
976                                    NETIF_F_HW_VLAN_RX |
977                                    NETIF_F_HW_VLAN_FILTER;
978                 if (adapter->hw.mac_type == e1000_ich8lan)
979                         netdev->features &= ~NETIF_F_HW_VLAN_FILTER;
980         }
981
982         if ((adapter->hw.mac_type >= e1000_82544) &&
983            (adapter->hw.mac_type != e1000_82547))
984                 netdev->features |= NETIF_F_TSO;
985
986         if (adapter->hw.mac_type > e1000_82547_rev_2)
987                 netdev->features |= NETIF_F_TSO6;
988         if (pci_using_dac)
989                 netdev->features |= NETIF_F_HIGHDMA;
990
991         netdev->features |= NETIF_F_LLTX;
992
993         adapter->en_mng_pt = e1000_enable_mng_pass_thru(&adapter->hw);
994
995         /* initialize eeprom parameters */
996
997         if (e1000_init_eeprom_params(&adapter->hw)) {
998                 E1000_ERR("EEPROM initialization failed\n");
999                 goto err_eeprom;
1000         }
1001
1002         /* before reading the EEPROM, reset the controller to
1003          * put the device in a known good starting state */
1004
1005         e1000_reset_hw(&adapter->hw);
1006
1007         /* make sure the EEPROM is good */
1008
1009         if (e1000_validate_eeprom_checksum(&adapter->hw) < 0) {
1010                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
1011                 goto err_eeprom;
1012         }
1013
1014         /* copy the MAC address out of the EEPROM */
1015
1016         if (e1000_read_mac_addr(&adapter->hw))
1017                 DPRINTK(PROBE, ERR, "EEPROM Read Error\n");
1018         memcpy(netdev->dev_addr, adapter->hw.mac_addr, netdev->addr_len);
1019         memcpy(netdev->perm_addr, adapter->hw.mac_addr, netdev->addr_len);
1020
1021         if (!is_valid_ether_addr(netdev->perm_addr)) {
1022                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
1023                 goto err_eeprom;
1024         }
1025
1026         e1000_get_bus_info(&adapter->hw);
1027
1028         init_timer(&adapter->tx_fifo_stall_timer);
1029         adapter->tx_fifo_stall_timer.function = &e1000_82547_tx_fifo_stall;
1030         adapter->tx_fifo_stall_timer.data = (unsigned long) adapter;
1031
1032         init_timer(&adapter->watchdog_timer);
1033         adapter->watchdog_timer.function = &e1000_watchdog;
1034         adapter->watchdog_timer.data = (unsigned long) adapter;
1035
1036         init_timer(&adapter->phy_info_timer);
1037         adapter->phy_info_timer.function = &e1000_update_phy_info;
1038         adapter->phy_info_timer.data = (unsigned long) adapter;
1039
1040         INIT_WORK(&adapter->reset_task, e1000_reset_task);
1041
1042         e1000_check_options(adapter);
1043
1044         /* Initial Wake on LAN setting
1045          * If APM wake is enabled in the EEPROM,
1046          * enable the ACPI Magic Packet filter
1047          */
1048
1049         switch (adapter->hw.mac_type) {
1050         case e1000_82542_rev2_0:
1051         case e1000_82542_rev2_1:
1052         case e1000_82543:
1053                 break;
1054         case e1000_82544:
1055                 e1000_read_eeprom(&adapter->hw,
1056                         EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
1057                 eeprom_apme_mask = E1000_EEPROM_82544_APM;
1058                 break;
1059         case e1000_ich8lan:
1060                 e1000_read_eeprom(&adapter->hw,
1061                         EEPROM_INIT_CONTROL1_REG, 1, &eeprom_data);
1062                 eeprom_apme_mask = E1000_EEPROM_ICH8_APME;
1063                 break;
1064         case e1000_82546:
1065         case e1000_82546_rev_3:
1066         case e1000_82571:
1067         case e1000_80003es2lan:
1068                 if (E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_FUNC_1){
1069                         e1000_read_eeprom(&adapter->hw,
1070                                 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1071                         break;
1072                 }
1073                 /* Fall Through */
1074         default:
1075                 e1000_read_eeprom(&adapter->hw,
1076                         EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1077                 break;
1078         }
1079         if (eeprom_data & eeprom_apme_mask)
1080                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1081
1082         /* now that we have the eeprom settings, apply the special cases
1083          * where the eeprom may be wrong or the board simply won't support
1084          * wake on lan on a particular port */
1085         switch (pdev->device) {
1086         case E1000_DEV_ID_82546GB_PCIE:
1087                 adapter->eeprom_wol = 0;
1088                 break;
1089         case E1000_DEV_ID_82546EB_FIBER:
1090         case E1000_DEV_ID_82546GB_FIBER:
1091         case E1000_DEV_ID_82571EB_FIBER:
1092                 /* Wake events only supported on port A for dual fiber
1093                  * regardless of eeprom setting */
1094                 if (E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_FUNC_1)
1095                         adapter->eeprom_wol = 0;
1096                 break;
1097         case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
1098         case E1000_DEV_ID_82571EB_QUAD_COPPER:
1099         case E1000_DEV_ID_82571EB_QUAD_COPPER_LOWPROFILE:
1100                 /* if quad port adapter, disable WoL on all but port A */
1101                 if (global_quad_port_a != 0)
1102                         adapter->eeprom_wol = 0;
1103                 else
1104                         adapter->quad_port_a = 1;
1105                 /* Reset for multiple quad port adapters */
1106                 if (++global_quad_port_a == 4)
1107                         global_quad_port_a = 0;
1108                 break;
1109         }
1110
1111         /* initialize the wol settings based on the eeprom settings */
1112         adapter->wol = adapter->eeprom_wol;
1113
1114         /* print bus type/speed/width info */
1115         {
1116         struct e1000_hw *hw = &adapter->hw;
1117         DPRINTK(PROBE, INFO, "(PCI%s:%s:%s) ",
1118                 ((hw->bus_type == e1000_bus_type_pcix) ? "-X" :
1119                  (hw->bus_type == e1000_bus_type_pci_express ? " Express":"")),
1120                 ((hw->bus_speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
1121                  (hw->bus_speed == e1000_bus_speed_133) ? "133MHz" :
1122                  (hw->bus_speed == e1000_bus_speed_120) ? "120MHz" :
1123                  (hw->bus_speed == e1000_bus_speed_100) ? "100MHz" :
1124                  (hw->bus_speed == e1000_bus_speed_66) ? "66MHz" : "33MHz"),
1125                 ((hw->bus_width == e1000_bus_width_64) ? "64-bit" :
1126                  (hw->bus_width == e1000_bus_width_pciex_4) ? "Width x4" :
1127                  (hw->bus_width == e1000_bus_width_pciex_1) ? "Width x1" :
1128                  "32-bit"));
1129         }
1130
1131         for (i = 0; i < 6; i++)
1132                 printk("%2.2x%c", netdev->dev_addr[i], i == 5 ? '\n' : ':');
1133
1134         /* reset the hardware with the new settings */
1135         e1000_reset(adapter);
1136
1137         /* If the controller is 82573 and f/w is AMT, do not set
1138          * DRV_LOAD until the interface is up.  For all other cases,
1139          * let the f/w know that the h/w is now under the control
1140          * of the driver. */
1141         if (adapter->hw.mac_type != e1000_82573 ||
1142             !e1000_check_mng_mode(&adapter->hw))
1143                 e1000_get_hw_control(adapter);
1144
1145         /* tell the stack to leave us alone until e1000_open() is called */
1146         netif_carrier_off(netdev);
1147         netif_stop_queue(netdev);
1148 #ifdef CONFIG_E1000_NAPI
1149         netif_poll_disable(netdev);
1150 #endif
1151
1152         strcpy(netdev->name, "eth%d");
1153         if ((err = register_netdev(netdev)))
1154                 goto err_register;
1155
1156         DPRINTK(PROBE, INFO, "Intel(R) PRO/1000 Network Connection\n");
1157
1158         cards_found++;
1159         return 0;
1160
1161 err_register:
1162         e1000_release_hw_control(adapter);
1163 err_eeprom:
1164         if (!e1000_check_phy_reset_block(&adapter->hw))
1165                 e1000_phy_hw_reset(&adapter->hw);
1166
1167         if (adapter->hw.flash_address)
1168                 iounmap(adapter->hw.flash_address);
1169 err_flashmap:
1170 #ifdef CONFIG_E1000_NAPI
1171         for (i = 0; i < adapter->num_rx_queues; i++)
1172                 dev_put(&adapter->polling_netdev[i]);
1173 #endif
1174
1175         kfree(adapter->tx_ring);
1176         kfree(adapter->rx_ring);
1177 #ifdef CONFIG_E1000_NAPI
1178         kfree(adapter->polling_netdev);
1179 #endif
1180 err_sw_init:
1181         iounmap(adapter->hw.hw_addr);
1182 err_ioremap:
1183         free_netdev(netdev);
1184 err_alloc_etherdev:
1185         pci_release_regions(pdev);
1186 err_pci_reg:
1187 err_dma:
1188         pci_disable_device(pdev);
1189         return err;
1190 }
1191
1192 /**
1193  * e1000_remove - Device Removal Routine
1194  * @pdev: PCI device information struct
1195  *
1196  * e1000_remove is called by the PCI subsystem to alert the driver
1197  * that it should release a PCI device.  The could be caused by a
1198  * Hot-Plug event, or because the driver is going to be removed from
1199  * memory.
1200  **/
1201
1202 static void __devexit
1203 e1000_remove(struct pci_dev *pdev)
1204 {
1205         struct net_device *netdev = pci_get_drvdata(pdev);
1206         struct e1000_adapter *adapter = netdev_priv(netdev);
1207 #ifdef CONFIG_E1000_NAPI
1208         int i;
1209 #endif
1210
1211         cancel_work_sync(&adapter->reset_task);
1212
1213         e1000_release_manageability(adapter);
1214
1215         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
1216          * would have already happened in close and is redundant. */
1217         e1000_release_hw_control(adapter);
1218
1219         unregister_netdev(netdev);
1220 #ifdef CONFIG_E1000_NAPI
1221         for (i = 0; i < adapter->num_rx_queues; i++)
1222                 dev_put(&adapter->polling_netdev[i]);
1223 #endif
1224
1225         if (!e1000_check_phy_reset_block(&adapter->hw))
1226                 e1000_phy_hw_reset(&adapter->hw);
1227
1228         kfree(adapter->tx_ring);
1229         kfree(adapter->rx_ring);
1230 #ifdef CONFIG_E1000_NAPI
1231         kfree(adapter->polling_netdev);
1232 #endif
1233
1234         iounmap(adapter->hw.hw_addr);
1235         if (adapter->hw.flash_address)
1236                 iounmap(adapter->hw.flash_address);
1237         pci_release_regions(pdev);
1238
1239         free_netdev(netdev);
1240
1241         pci_disable_device(pdev);
1242 }
1243
1244 /**
1245  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1246  * @adapter: board private structure to initialize
1247  *
1248  * e1000_sw_init initializes the Adapter private data structure.
1249  * Fields are initialized based on PCI device information and
1250  * OS network device settings (MTU size).
1251  **/
1252
1253 static int __devinit
1254 e1000_sw_init(struct e1000_adapter *adapter)
1255 {
1256         struct e1000_hw *hw = &adapter->hw;
1257         struct net_device *netdev = adapter->netdev;
1258         struct pci_dev *pdev = adapter->pdev;
1259 #ifdef CONFIG_E1000_NAPI
1260         int i;
1261 #endif
1262
1263         /* PCI config space info */
1264
1265         hw->vendor_id = pdev->vendor;
1266         hw->device_id = pdev->device;
1267         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1268         hw->subsystem_id = pdev->subsystem_device;
1269         hw->revision_id = pdev->revision;
1270
1271         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
1272
1273         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1274         adapter->rx_ps_bsize0 = E1000_RXBUFFER_128;
1275         hw->max_frame_size = netdev->mtu +
1276                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
1277         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
1278
1279         /* identify the MAC */
1280
1281         if (e1000_set_mac_type(hw)) {
1282                 DPRINTK(PROBE, ERR, "Unknown MAC Type\n");
1283                 return -EIO;
1284         }
1285
1286         switch (hw->mac_type) {
1287         default:
1288                 break;
1289         case e1000_82541:
1290         case e1000_82547:
1291         case e1000_82541_rev_2:
1292         case e1000_82547_rev_2:
1293                 hw->phy_init_script = 1;
1294                 break;
1295         }
1296
1297         e1000_set_media_type(hw);
1298
1299         hw->wait_autoneg_complete = FALSE;
1300         hw->tbi_compatibility_en = TRUE;
1301         hw->adaptive_ifs = TRUE;
1302
1303         /* Copper options */
1304
1305         if (hw->media_type == e1000_media_type_copper) {
1306                 hw->mdix = AUTO_ALL_MODES;
1307                 hw->disable_polarity_correction = FALSE;
1308                 hw->master_slave = E1000_MASTER_SLAVE;
1309         }
1310
1311         adapter->num_tx_queues = 1;
1312         adapter->num_rx_queues = 1;
1313
1314         if (e1000_alloc_queues(adapter)) {
1315                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
1316                 return -ENOMEM;
1317         }
1318
1319 #ifdef CONFIG_E1000_NAPI
1320         for (i = 0; i < adapter->num_rx_queues; i++) {
1321                 adapter->polling_netdev[i].priv = adapter;
1322                 adapter->polling_netdev[i].poll = &e1000_clean;
1323                 adapter->polling_netdev[i].weight = 64;
1324                 dev_hold(&adapter->polling_netdev[i]);
1325                 set_bit(__LINK_STATE_START, &adapter->polling_netdev[i].state);
1326         }
1327         spin_lock_init(&adapter->tx_queue_lock);
1328 #endif
1329
1330         /* Explicitly disable IRQ since the NIC can be in any state. */
1331         atomic_set(&adapter->irq_sem, 0);
1332         e1000_irq_disable(adapter);
1333
1334         spin_lock_init(&adapter->stats_lock);
1335
1336         set_bit(__E1000_DOWN, &adapter->flags);
1337
1338         return 0;
1339 }
1340
1341 /**
1342  * e1000_alloc_queues - Allocate memory for all rings
1343  * @adapter: board private structure to initialize
1344  *
1345  * We allocate one ring per queue at run-time since we don't know the
1346  * number of queues at compile-time.  The polling_netdev array is
1347  * intended for Multiqueue, but should work fine with a single queue.
1348  **/
1349
1350 static int __devinit
1351 e1000_alloc_queues(struct e1000_adapter *adapter)
1352 {
1353         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1354                                    sizeof(struct e1000_tx_ring), GFP_KERNEL);
1355         if (!adapter->tx_ring)
1356                 return -ENOMEM;
1357
1358         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1359                                    sizeof(struct e1000_rx_ring), GFP_KERNEL);
1360         if (!adapter->rx_ring) {
1361                 kfree(adapter->tx_ring);
1362                 return -ENOMEM;
1363         }
1364
1365 #ifdef CONFIG_E1000_NAPI
1366         adapter->polling_netdev = kcalloc(adapter->num_rx_queues,
1367                                           sizeof(struct net_device),
1368                                           GFP_KERNEL);
1369         if (!adapter->polling_netdev) {
1370                 kfree(adapter->tx_ring);
1371                 kfree(adapter->rx_ring);
1372                 return -ENOMEM;
1373         }
1374 #endif
1375
1376         return E1000_SUCCESS;
1377 }
1378
1379 /**
1380  * e1000_open - Called when a network interface is made active
1381  * @netdev: network interface device structure
1382  *
1383  * Returns 0 on success, negative value on failure
1384  *
1385  * The open entry point is called when a network interface is made
1386  * active by the system (IFF_UP).  At this point all resources needed
1387  * for transmit and receive operations are allocated, the interrupt
1388  * handler is registered with the OS, the watchdog timer is started,
1389  * and the stack is notified that the interface is ready.
1390  **/
1391
1392 static int
1393 e1000_open(struct net_device *netdev)
1394 {
1395         struct e1000_adapter *adapter = netdev_priv(netdev);
1396         int err;
1397
1398         /* disallow open during test */
1399         if (test_bit(__E1000_TESTING, &adapter->flags))
1400                 return -EBUSY;
1401
1402         /* allocate transmit descriptors */
1403         err = e1000_setup_all_tx_resources(adapter);
1404         if (err)
1405                 goto err_setup_tx;
1406
1407         /* allocate receive descriptors */
1408         err = e1000_setup_all_rx_resources(adapter);
1409         if (err)
1410                 goto err_setup_rx;
1411
1412         e1000_power_up_phy(adapter);
1413
1414         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1415         if ((adapter->hw.mng_cookie.status &
1416                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1417                 e1000_update_mng_vlan(adapter);
1418         }
1419
1420         /* If AMT is enabled, let the firmware know that the network
1421          * interface is now open */
1422         if (adapter->hw.mac_type == e1000_82573 &&
1423             e1000_check_mng_mode(&adapter->hw))
1424                 e1000_get_hw_control(adapter);
1425
1426         /* before we allocate an interrupt, we must be ready to handle it.
1427          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1428          * as soon as we call pci_request_irq, so we have to setup our
1429          * clean_rx handler before we do so.  */
1430         e1000_configure(adapter);
1431
1432         err = e1000_request_irq(adapter);
1433         if (err)
1434                 goto err_req_irq;
1435
1436         /* From here on the code is the same as e1000_up() */
1437         clear_bit(__E1000_DOWN, &adapter->flags);
1438
1439 #ifdef CONFIG_E1000_NAPI
1440         netif_poll_enable(netdev);
1441 #endif
1442
1443         e1000_irq_enable(adapter);
1444
1445         /* fire a link status change interrupt to start the watchdog */
1446         E1000_WRITE_REG(&adapter->hw, ICS, E1000_ICS_LSC);
1447
1448         return E1000_SUCCESS;
1449
1450 err_req_irq:
1451         e1000_release_hw_control(adapter);
1452         e1000_power_down_phy(adapter);
1453         e1000_free_all_rx_resources(adapter);
1454 err_setup_rx:
1455         e1000_free_all_tx_resources(adapter);
1456 err_setup_tx:
1457         e1000_reset(adapter);
1458
1459         return err;
1460 }
1461
1462 /**
1463  * e1000_close - Disables a network interface
1464  * @netdev: network interface device structure
1465  *
1466  * Returns 0, this is not allowed to fail
1467  *
1468  * The close entry point is called when an interface is de-activated
1469  * by the OS.  The hardware is still under the drivers control, but
1470  * needs to be disabled.  A global MAC reset is issued to stop the
1471  * hardware, and all transmit and receive resources are freed.
1472  **/
1473
1474 static int
1475 e1000_close(struct net_device *netdev)
1476 {
1477         struct e1000_adapter *adapter = netdev_priv(netdev);
1478
1479         WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1480         e1000_down(adapter);
1481         e1000_power_down_phy(adapter);
1482         e1000_free_irq(adapter);
1483
1484         e1000_free_all_tx_resources(adapter);
1485         e1000_free_all_rx_resources(adapter);
1486
1487         /* kill manageability vlan ID if supported, but not if a vlan with
1488          * the same ID is registered on the host OS (let 8021q kill it) */
1489         if ((adapter->hw.mng_cookie.status &
1490                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1491              !(adapter->vlgrp &&
1492                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id))) {
1493                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1494         }
1495
1496         /* If AMT is enabled, let the firmware know that the network
1497          * interface is now closed */
1498         if (adapter->hw.mac_type == e1000_82573 &&
1499             e1000_check_mng_mode(&adapter->hw))
1500                 e1000_release_hw_control(adapter);
1501
1502         return 0;
1503 }
1504
1505 /**
1506  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1507  * @adapter: address of board private structure
1508  * @start: address of beginning of memory
1509  * @len: length of memory
1510  **/
1511 static boolean_t
1512 e1000_check_64k_bound(struct e1000_adapter *adapter,
1513                       void *start, unsigned long len)
1514 {
1515         unsigned long begin = (unsigned long) start;
1516         unsigned long end = begin + len;
1517
1518         /* First rev 82545 and 82546 need to not allow any memory
1519          * write location to cross 64k boundary due to errata 23 */
1520         if (adapter->hw.mac_type == e1000_82545 ||
1521             adapter->hw.mac_type == e1000_82546) {
1522                 return ((begin ^ (end - 1)) >> 16) != 0 ? FALSE : TRUE;
1523         }
1524
1525         return TRUE;
1526 }
1527
1528 /**
1529  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1530  * @adapter: board private structure
1531  * @txdr:    tx descriptor ring (for a specific queue) to setup
1532  *
1533  * Return 0 on success, negative on failure
1534  **/
1535
1536 static int
1537 e1000_setup_tx_resources(struct e1000_adapter *adapter,
1538                          struct e1000_tx_ring *txdr)
1539 {
1540         struct pci_dev *pdev = adapter->pdev;
1541         int size;
1542
1543         size = sizeof(struct e1000_buffer) * txdr->count;
1544         txdr->buffer_info = vmalloc(size);
1545         if (!txdr->buffer_info) {
1546                 DPRINTK(PROBE, ERR,
1547                 "Unable to allocate memory for the transmit descriptor ring\n");
1548                 return -ENOMEM;
1549         }
1550         memset(txdr->buffer_info, 0, size);
1551
1552         /* round up to nearest 4K */
1553
1554         txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1555         txdr->size = ALIGN(txdr->size, 4096);
1556
1557         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1558         if (!txdr->desc) {
1559 setup_tx_desc_die:
1560                 vfree(txdr->buffer_info);
1561                 DPRINTK(PROBE, ERR,
1562                 "Unable to allocate memory for the transmit descriptor ring\n");
1563                 return -ENOMEM;
1564         }
1565
1566         /* Fix for errata 23, can't cross 64kB boundary */
1567         if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1568                 void *olddesc = txdr->desc;
1569                 dma_addr_t olddma = txdr->dma;
1570                 DPRINTK(TX_ERR, ERR, "txdr align check failed: %u bytes "
1571                                      "at %p\n", txdr->size, txdr->desc);
1572                 /* Try again, without freeing the previous */
1573                 txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1574                 /* Failed allocation, critical failure */
1575                 if (!txdr->desc) {
1576                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1577                         goto setup_tx_desc_die;
1578                 }
1579
1580                 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1581                         /* give up */
1582                         pci_free_consistent(pdev, txdr->size, txdr->desc,
1583                                             txdr->dma);
1584                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1585                         DPRINTK(PROBE, ERR,
1586                                 "Unable to allocate aligned memory "
1587                                 "for the transmit descriptor ring\n");
1588                         vfree(txdr->buffer_info);
1589                         return -ENOMEM;
1590                 } else {
1591                         /* Free old allocation, new allocation was successful */
1592                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1593                 }
1594         }
1595         memset(txdr->desc, 0, txdr->size);
1596
1597         txdr->next_to_use = 0;
1598         txdr->next_to_clean = 0;
1599         spin_lock_init(&txdr->tx_lock);
1600
1601         return 0;
1602 }
1603
1604 /**
1605  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1606  *                                (Descriptors) for all queues
1607  * @adapter: board private structure
1608  *
1609  * Return 0 on success, negative on failure
1610  **/
1611
1612 int
1613 e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1614 {
1615         int i, err = 0;
1616
1617         for (i = 0; i < adapter->num_tx_queues; i++) {
1618                 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1619                 if (err) {
1620                         DPRINTK(PROBE, ERR,
1621                                 "Allocation for Tx Queue %u failed\n", i);
1622                         for (i-- ; i >= 0; i--)
1623                                 e1000_free_tx_resources(adapter,
1624                                                         &adapter->tx_ring[i]);
1625                         break;
1626                 }
1627         }
1628
1629         return err;
1630 }
1631
1632 /**
1633  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1634  * @adapter: board private structure
1635  *
1636  * Configure the Tx unit of the MAC after a reset.
1637  **/
1638
1639 static void
1640 e1000_configure_tx(struct e1000_adapter *adapter)
1641 {
1642         uint64_t tdba;
1643         struct e1000_hw *hw = &adapter->hw;
1644         uint32_t tdlen, tctl, tipg, tarc;
1645         uint32_t ipgr1, ipgr2;
1646
1647         /* Setup the HW Tx Head and Tail descriptor pointers */
1648
1649         switch (adapter->num_tx_queues) {
1650         case 1:
1651         default:
1652                 tdba = adapter->tx_ring[0].dma;
1653                 tdlen = adapter->tx_ring[0].count *
1654                         sizeof(struct e1000_tx_desc);
1655                 E1000_WRITE_REG(hw, TDLEN, tdlen);
1656                 E1000_WRITE_REG(hw, TDBAH, (tdba >> 32));
1657                 E1000_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
1658                 E1000_WRITE_REG(hw, TDT, 0);
1659                 E1000_WRITE_REG(hw, TDH, 0);
1660                 adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ? E1000_TDH : E1000_82542_TDH);
1661                 adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ? E1000_TDT : E1000_82542_TDT);
1662                 break;
1663         }
1664
1665         /* Set the default values for the Tx Inter Packet Gap timer */
1666         if (adapter->hw.mac_type <= e1000_82547_rev_2 &&
1667             (hw->media_type == e1000_media_type_fiber ||
1668              hw->media_type == e1000_media_type_internal_serdes))
1669                 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1670         else
1671                 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1672
1673         switch (hw->mac_type) {
1674         case e1000_82542_rev2_0:
1675         case e1000_82542_rev2_1:
1676                 tipg = DEFAULT_82542_TIPG_IPGT;
1677                 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1678                 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1679                 break;
1680         case e1000_80003es2lan:
1681                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1682                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2;
1683                 break;
1684         default:
1685                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1686                 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1687                 break;
1688         }
1689         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1690         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1691         E1000_WRITE_REG(hw, TIPG, tipg);
1692
1693         /* Set the Tx Interrupt Delay register */
1694
1695         E1000_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
1696         if (hw->mac_type >= e1000_82540)
1697                 E1000_WRITE_REG(hw, TADV, adapter->tx_abs_int_delay);
1698
1699         /* Program the Transmit Control Register */
1700
1701         tctl = E1000_READ_REG(hw, TCTL);
1702         tctl &= ~E1000_TCTL_CT;
1703         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1704                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1705
1706         if (hw->mac_type == e1000_82571 || hw->mac_type == e1000_82572) {
1707                 tarc = E1000_READ_REG(hw, TARC0);
1708                 /* set the speed mode bit, we'll clear it if we're not at
1709                  * gigabit link later */
1710                 tarc |= (1 << 21);
1711                 E1000_WRITE_REG(hw, TARC0, tarc);
1712         } else if (hw->mac_type == e1000_80003es2lan) {
1713                 tarc = E1000_READ_REG(hw, TARC0);
1714                 tarc |= 1;
1715                 E1000_WRITE_REG(hw, TARC0, tarc);
1716                 tarc = E1000_READ_REG(hw, TARC1);
1717                 tarc |= 1;
1718                 E1000_WRITE_REG(hw, TARC1, tarc);
1719         }
1720
1721         e1000_config_collision_dist(hw);
1722
1723         /* Setup Transmit Descriptor Settings for eop descriptor */
1724         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1725
1726         /* only set IDE if we are delaying interrupts using the timers */
1727         if (adapter->tx_int_delay)
1728                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1729
1730         if (hw->mac_type < e1000_82543)
1731                 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1732         else
1733                 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1734
1735         /* Cache if we're 82544 running in PCI-X because we'll
1736          * need this to apply a workaround later in the send path. */
1737         if (hw->mac_type == e1000_82544 &&
1738             hw->bus_type == e1000_bus_type_pcix)
1739                 adapter->pcix_82544 = 1;
1740
1741         E1000_WRITE_REG(hw, TCTL, tctl);
1742
1743 }
1744
1745 /**
1746  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1747  * @adapter: board private structure
1748  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1749  *
1750  * Returns 0 on success, negative on failure
1751  **/
1752
1753 static int
1754 e1000_setup_rx_resources(struct e1000_adapter *adapter,
1755                          struct e1000_rx_ring *rxdr)
1756 {
1757         struct pci_dev *pdev = adapter->pdev;
1758         int size, desc_len;
1759
1760         size = sizeof(struct e1000_buffer) * rxdr->count;
1761         rxdr->buffer_info = vmalloc(size);
1762         if (!rxdr->buffer_info) {
1763                 DPRINTK(PROBE, ERR,
1764                 "Unable to allocate memory for the receive descriptor ring\n");
1765                 return -ENOMEM;
1766         }
1767         memset(rxdr->buffer_info, 0, size);
1768
1769         rxdr->ps_page = kcalloc(rxdr->count, sizeof(struct e1000_ps_page),
1770                                 GFP_KERNEL);
1771         if (!rxdr->ps_page) {
1772                 vfree(rxdr->buffer_info);
1773                 DPRINTK(PROBE, ERR,
1774                 "Unable to allocate memory for the receive descriptor ring\n");
1775                 return -ENOMEM;
1776         }
1777
1778         rxdr->ps_page_dma = kcalloc(rxdr->count,
1779                                     sizeof(struct e1000_ps_page_dma),
1780                                     GFP_KERNEL);
1781         if (!rxdr->ps_page_dma) {
1782                 vfree(rxdr->buffer_info);
1783                 kfree(rxdr->ps_page);
1784                 DPRINTK(PROBE, ERR,
1785                 "Unable to allocate memory for the receive descriptor ring\n");
1786                 return -ENOMEM;
1787         }
1788
1789         if (adapter->hw.mac_type <= e1000_82547_rev_2)
1790                 desc_len = sizeof(struct e1000_rx_desc);
1791         else
1792                 desc_len = sizeof(union e1000_rx_desc_packet_split);
1793
1794         /* Round up to nearest 4K */
1795
1796         rxdr->size = rxdr->count * desc_len;
1797         rxdr->size = ALIGN(rxdr->size, 4096);
1798
1799         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1800
1801         if (!rxdr->desc) {
1802                 DPRINTK(PROBE, ERR,
1803                 "Unable to allocate memory for the receive descriptor ring\n");
1804 setup_rx_desc_die:
1805                 vfree(rxdr->buffer_info);
1806                 kfree(rxdr->ps_page);
1807                 kfree(rxdr->ps_page_dma);
1808                 return -ENOMEM;
1809         }
1810
1811         /* Fix for errata 23, can't cross 64kB boundary */
1812         if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1813                 void *olddesc = rxdr->desc;
1814                 dma_addr_t olddma = rxdr->dma;
1815                 DPRINTK(RX_ERR, ERR, "rxdr align check failed: %u bytes "
1816                                      "at %p\n", rxdr->size, rxdr->desc);
1817                 /* Try again, without freeing the previous */
1818                 rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1819                 /* Failed allocation, critical failure */
1820                 if (!rxdr->desc) {
1821                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1822                         DPRINTK(PROBE, ERR,
1823                                 "Unable to allocate memory "
1824                                 "for the receive descriptor ring\n");
1825                         goto setup_rx_desc_die;
1826                 }
1827
1828                 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1829                         /* give up */
1830                         pci_free_consistent(pdev, rxdr->size, rxdr->desc,
1831                                             rxdr->dma);
1832                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1833                         DPRINTK(PROBE, ERR,
1834                                 "Unable to allocate aligned memory "
1835                                 "for the receive descriptor ring\n");
1836                         goto setup_rx_desc_die;
1837                 } else {
1838                         /* Free old allocation, new allocation was successful */
1839                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1840                 }
1841         }
1842         memset(rxdr->desc, 0, rxdr->size);
1843
1844         rxdr->next_to_clean = 0;
1845         rxdr->next_to_use = 0;
1846
1847         return 0;
1848 }
1849
1850 /**
1851  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1852  *                                (Descriptors) for all queues
1853  * @adapter: board private structure
1854  *
1855  * Return 0 on success, negative on failure
1856  **/
1857
1858 int
1859 e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1860 {
1861         int i, err = 0;
1862
1863         for (i = 0; i < adapter->num_rx_queues; i++) {
1864                 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1865                 if (err) {
1866                         DPRINTK(PROBE, ERR,
1867                                 "Allocation for Rx Queue %u failed\n", i);
1868                         for (i-- ; i >= 0; i--)
1869                                 e1000_free_rx_resources(adapter,
1870                                                         &adapter->rx_ring[i]);
1871                         break;
1872                 }
1873         }
1874
1875         return err;
1876 }
1877
1878 /**
1879  * e1000_setup_rctl - configure the receive control registers
1880  * @adapter: Board private structure
1881  **/
1882 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1883                         (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1884 static void
1885 e1000_setup_rctl(struct e1000_adapter *adapter)
1886 {
1887         uint32_t rctl, rfctl;
1888         uint32_t psrctl = 0;
1889 #ifndef CONFIG_E1000_DISABLE_PACKET_SPLIT
1890         uint32_t pages = 0;
1891 #endif
1892
1893         rctl = E1000_READ_REG(&adapter->hw, RCTL);
1894
1895         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1896
1897         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1898                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1899                 (adapter->hw.mc_filter_type << E1000_RCTL_MO_SHIFT);
1900
1901         if (adapter->hw.tbi_compatibility_on == 1)
1902                 rctl |= E1000_RCTL_SBP;
1903         else
1904                 rctl &= ~E1000_RCTL_SBP;
1905
1906         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1907                 rctl &= ~E1000_RCTL_LPE;
1908         else
1909                 rctl |= E1000_RCTL_LPE;
1910
1911         /* Setup buffer sizes */
1912         rctl &= ~E1000_RCTL_SZ_4096;
1913         rctl |= E1000_RCTL_BSEX;
1914         switch (adapter->rx_buffer_len) {
1915                 case E1000_RXBUFFER_256:
1916                         rctl |= E1000_RCTL_SZ_256;
1917                         rctl &= ~E1000_RCTL_BSEX;
1918                         break;
1919                 case E1000_RXBUFFER_512:
1920                         rctl |= E1000_RCTL_SZ_512;
1921                         rctl &= ~E1000_RCTL_BSEX;
1922                         break;
1923                 case E1000_RXBUFFER_1024:
1924                         rctl |= E1000_RCTL_SZ_1024;
1925                         rctl &= ~E1000_RCTL_BSEX;
1926                         break;
1927                 case E1000_RXBUFFER_2048:
1928                 default:
1929                         rctl |= E1000_RCTL_SZ_2048;
1930                         rctl &= ~E1000_RCTL_BSEX;
1931                         break;
1932                 case E1000_RXBUFFER_4096:
1933                         rctl |= E1000_RCTL_SZ_4096;
1934                         break;
1935                 case E1000_RXBUFFER_8192:
1936                         rctl |= E1000_RCTL_SZ_8192;
1937                         break;
1938                 case E1000_RXBUFFER_16384:
1939                         rctl |= E1000_RCTL_SZ_16384;
1940                         break;
1941         }
1942
1943 #ifndef CONFIG_E1000_DISABLE_PACKET_SPLIT
1944         /* 82571 and greater support packet-split where the protocol
1945          * header is placed in skb->data and the packet data is
1946          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
1947          * In the case of a non-split, skb->data is linearly filled,
1948          * followed by the page buffers.  Therefore, skb->data is
1949          * sized to hold the largest protocol header.
1950          */
1951         /* allocations using alloc_page take too long for regular MTU
1952          * so only enable packet split for jumbo frames */
1953         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
1954         if ((adapter->hw.mac_type >= e1000_82571) && (pages <= 3) &&
1955             PAGE_SIZE <= 16384 && (rctl & E1000_RCTL_LPE))
1956                 adapter->rx_ps_pages = pages;
1957         else
1958                 adapter->rx_ps_pages = 0;
1959 #endif
1960         if (adapter->rx_ps_pages) {
1961                 /* Configure extra packet-split registers */
1962                 rfctl = E1000_READ_REG(&adapter->hw, RFCTL);
1963                 rfctl |= E1000_RFCTL_EXTEN;
1964                 /* disable packet split support for IPv6 extension headers,
1965                  * because some malformed IPv6 headers can hang the RX */
1966                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
1967                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
1968
1969                 E1000_WRITE_REG(&adapter->hw, RFCTL, rfctl);
1970
1971                 rctl |= E1000_RCTL_DTYP_PS;
1972
1973                 psrctl |= adapter->rx_ps_bsize0 >>
1974                         E1000_PSRCTL_BSIZE0_SHIFT;
1975
1976                 switch (adapter->rx_ps_pages) {
1977                 case 3:
1978                         psrctl |= PAGE_SIZE <<
1979                                 E1000_PSRCTL_BSIZE3_SHIFT;
1980                 case 2:
1981                         psrctl |= PAGE_SIZE <<
1982                                 E1000_PSRCTL_BSIZE2_SHIFT;
1983                 case 1:
1984                         psrctl |= PAGE_SIZE >>
1985                                 E1000_PSRCTL_BSIZE1_SHIFT;
1986                         break;
1987                 }
1988
1989                 E1000_WRITE_REG(&adapter->hw, PSRCTL, psrctl);
1990         }
1991
1992         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
1993 }
1994
1995 /**
1996  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1997  * @adapter: board private structure
1998  *
1999  * Configure the Rx unit of the MAC after a reset.
2000  **/
2001
2002 static void
2003 e1000_configure_rx(struct e1000_adapter *adapter)
2004 {
2005         uint64_t rdba;
2006         struct e1000_hw *hw = &adapter->hw;
2007         uint32_t rdlen, rctl, rxcsum, ctrl_ext;
2008
2009         if (adapter->rx_ps_pages) {
2010                 /* this is a 32 byte descriptor */
2011                 rdlen = adapter->rx_ring[0].count *
2012                         sizeof(union e1000_rx_desc_packet_split);
2013                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2014                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2015         } else {
2016                 rdlen = adapter->rx_ring[0].count *
2017                         sizeof(struct e1000_rx_desc);
2018                 adapter->clean_rx = e1000_clean_rx_irq;
2019                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2020         }
2021
2022         /* disable receives while setting up the descriptors */
2023         rctl = E1000_READ_REG(hw, RCTL);
2024         E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
2025
2026         /* set the Receive Delay Timer Register */
2027         E1000_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
2028
2029         if (hw->mac_type >= e1000_82540) {
2030                 E1000_WRITE_REG(hw, RADV, adapter->rx_abs_int_delay);
2031                 if (adapter->itr_setting != 0)
2032                         E1000_WRITE_REG(hw, ITR,
2033                                 1000000000 / (adapter->itr * 256));
2034         }
2035
2036         if (hw->mac_type >= e1000_82571) {
2037                 ctrl_ext = E1000_READ_REG(hw, CTRL_EXT);
2038                 /* Reset delay timers after every interrupt */
2039                 ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2040 #ifdef CONFIG_E1000_NAPI
2041                 /* Auto-Mask interrupts upon ICR access */
2042                 ctrl_ext |= E1000_CTRL_EXT_IAME;
2043                 E1000_WRITE_REG(hw, IAM, 0xffffffff);
2044 #endif
2045                 E1000_WRITE_REG(hw, CTRL_EXT, ctrl_ext);
2046                 E1000_WRITE_FLUSH(hw);
2047         }
2048
2049         /* Setup the HW Rx Head and Tail Descriptor Pointers and
2050          * the Base and Length of the Rx Descriptor Ring */
2051         switch (adapter->num_rx_queues) {
2052         case 1:
2053         default:
2054                 rdba = adapter->rx_ring[0].dma;
2055                 E1000_WRITE_REG(hw, RDLEN, rdlen);
2056                 E1000_WRITE_REG(hw, RDBAH, (rdba >> 32));
2057                 E1000_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
2058                 E1000_WRITE_REG(hw, RDT, 0);
2059                 E1000_WRITE_REG(hw, RDH, 0);
2060                 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ? E1000_RDH : E1000_82542_RDH);
2061                 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ? E1000_RDT : E1000_82542_RDT);
2062                 break;
2063         }
2064
2065         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
2066         if (hw->mac_type >= e1000_82543) {
2067                 rxcsum = E1000_READ_REG(hw, RXCSUM);
2068                 if (adapter->rx_csum == TRUE) {
2069                         rxcsum |= E1000_RXCSUM_TUOFL;
2070
2071                         /* Enable 82571 IPv4 payload checksum for UDP fragments
2072                          * Must be used in conjunction with packet-split. */
2073                         if ((hw->mac_type >= e1000_82571) &&
2074                             (adapter->rx_ps_pages)) {
2075                                 rxcsum |= E1000_RXCSUM_IPPCSE;
2076                         }
2077                 } else {
2078                         rxcsum &= ~E1000_RXCSUM_TUOFL;
2079                         /* don't need to clear IPPCSE as it defaults to 0 */
2080                 }
2081                 E1000_WRITE_REG(hw, RXCSUM, rxcsum);
2082         }
2083
2084         /* enable early receives on 82573, only takes effect if using > 2048
2085          * byte total frame size.  for example only for jumbo frames */
2086 #define E1000_ERT_2048 0x100
2087         if (hw->mac_type == e1000_82573)
2088                 E1000_WRITE_REG(hw, ERT, E1000_ERT_2048);
2089
2090         /* Enable Receives */
2091         E1000_WRITE_REG(hw, RCTL, rctl);
2092 }
2093
2094 /**
2095  * e1000_free_tx_resources - Free Tx Resources per Queue
2096  * @adapter: board private structure
2097  * @tx_ring: Tx descriptor ring for a specific queue
2098  *
2099  * Free all transmit software resources
2100  **/
2101
2102 static void
2103 e1000_free_tx_resources(struct e1000_adapter *adapter,
2104                         struct e1000_tx_ring *tx_ring)
2105 {
2106         struct pci_dev *pdev = adapter->pdev;
2107
2108         e1000_clean_tx_ring(adapter, tx_ring);
2109
2110         vfree(tx_ring->buffer_info);
2111         tx_ring->buffer_info = NULL;
2112
2113         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
2114
2115         tx_ring->desc = NULL;
2116 }
2117
2118 /**
2119  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
2120  * @adapter: board private structure
2121  *
2122  * Free all transmit software resources
2123  **/
2124
2125 void
2126 e1000_free_all_tx_resources(struct e1000_adapter *adapter)
2127 {
2128         int i;
2129
2130         for (i = 0; i < adapter->num_tx_queues; i++)
2131                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
2132 }
2133
2134 static void
2135 e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
2136                         struct e1000_buffer *buffer_info)
2137 {
2138         if (buffer_info->dma) {
2139                 pci_unmap_page(adapter->pdev,
2140                                 buffer_info->dma,
2141                                 buffer_info->length,
2142                                 PCI_DMA_TODEVICE);
2143                 buffer_info->dma = 0;
2144         }
2145         if (buffer_info->skb) {
2146                 dev_kfree_skb_any(buffer_info->skb);
2147                 buffer_info->skb = NULL;
2148         }
2149         /* buffer_info must be completely set up in the transmit path */
2150 }
2151
2152 /**
2153  * e1000_clean_tx_ring - Free Tx Buffers
2154  * @adapter: board private structure
2155  * @tx_ring: ring to be cleaned
2156  **/
2157
2158 static void
2159 e1000_clean_tx_ring(struct e1000_adapter *adapter,
2160                     struct e1000_tx_ring *tx_ring)
2161 {
2162         struct e1000_buffer *buffer_info;
2163         unsigned long size;
2164         unsigned int i;
2165
2166         /* Free all the Tx ring sk_buffs */
2167
2168         for (i = 0; i < tx_ring->count; i++) {
2169                 buffer_info = &tx_ring->buffer_info[i];
2170                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2171         }
2172
2173         size = sizeof(struct e1000_buffer) * tx_ring->count;
2174         memset(tx_ring->buffer_info, 0, size);
2175
2176         /* Zero out the descriptor ring */
2177
2178         memset(tx_ring->desc, 0, tx_ring->size);
2179
2180         tx_ring->next_to_use = 0;
2181         tx_ring->next_to_clean = 0;
2182         tx_ring->last_tx_tso = 0;
2183
2184         writel(0, adapter->hw.hw_addr + tx_ring->tdh);
2185         writel(0, adapter->hw.hw_addr + tx_ring->tdt);
2186 }
2187
2188 /**
2189  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
2190  * @adapter: board private structure
2191  **/
2192
2193 static void
2194 e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
2195 {
2196         int i;
2197
2198         for (i = 0; i < adapter->num_tx_queues; i++)
2199                 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2200 }
2201
2202 /**
2203  * e1000_free_rx_resources - Free Rx Resources
2204  * @adapter: board private structure
2205  * @rx_ring: ring to clean the resources from
2206  *
2207  * Free all receive software resources
2208  **/
2209
2210 static void
2211 e1000_free_rx_resources(struct e1000_adapter *adapter,
2212                         struct e1000_rx_ring *rx_ring)
2213 {
2214         struct pci_dev *pdev = adapter->pdev;
2215
2216         e1000_clean_rx_ring(adapter, rx_ring);
2217
2218         vfree(rx_ring->buffer_info);
2219         rx_ring->buffer_info = NULL;
2220         kfree(rx_ring->ps_page);
2221         rx_ring->ps_page = NULL;
2222         kfree(rx_ring->ps_page_dma);
2223         rx_ring->ps_page_dma = NULL;
2224
2225         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2226
2227         rx_ring->desc = NULL;
2228 }
2229
2230 /**
2231  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
2232  * @adapter: board private structure
2233  *
2234  * Free all receive software resources
2235  **/
2236
2237 void
2238 e1000_free_all_rx_resources(struct e1000_adapter *adapter)
2239 {
2240         int i;
2241
2242         for (i = 0; i < adapter->num_rx_queues; i++)
2243                 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
2244 }
2245
2246 /**
2247  * e1000_clean_rx_ring - Free Rx Buffers per Queue
2248  * @adapter: board private structure
2249  * @rx_ring: ring to free buffers from
2250  **/
2251
2252 static void
2253 e1000_clean_rx_ring(struct e1000_adapter *adapter,
2254                     struct e1000_rx_ring *rx_ring)
2255 {
2256         struct e1000_buffer *buffer_info;
2257         struct e1000_ps_page *ps_page;
2258         struct e1000_ps_page_dma *ps_page_dma;
2259         struct pci_dev *pdev = adapter->pdev;
2260         unsigned long size;
2261         unsigned int i, j;
2262
2263         /* Free all the Rx ring sk_buffs */
2264         for (i = 0; i < rx_ring->count; i++) {
2265                 buffer_info = &rx_ring->buffer_info[i];
2266                 if (buffer_info->skb) {
2267                         pci_unmap_single(pdev,
2268                                          buffer_info->dma,
2269                                          buffer_info->length,
2270                                          PCI_DMA_FROMDEVICE);
2271
2272                         dev_kfree_skb(buffer_info->skb);
2273                         buffer_info->skb = NULL;
2274                 }
2275                 ps_page = &rx_ring->ps_page[i];
2276                 ps_page_dma = &rx_ring->ps_page_dma[i];
2277                 for (j = 0; j < adapter->rx_ps_pages; j++) {
2278                         if (!ps_page->ps_page[j]) break;
2279                         pci_unmap_page(pdev,
2280                                        ps_page_dma->ps_page_dma[j],
2281                                        PAGE_SIZE, PCI_DMA_FROMDEVICE);
2282                         ps_page_dma->ps_page_dma[j] = 0;
2283                         put_page(ps_page->ps_page[j]);
2284                         ps_page->ps_page[j] = NULL;
2285                 }
2286         }
2287
2288         size = sizeof(struct e1000_buffer) * rx_ring->count;
2289         memset(rx_ring->buffer_info, 0, size);
2290         size = sizeof(struct e1000_ps_page) * rx_ring->count;
2291         memset(rx_ring->ps_page, 0, size);
2292         size = sizeof(struct e1000_ps_page_dma) * rx_ring->count;
2293         memset(rx_ring->ps_page_dma, 0, size);
2294
2295         /* Zero out the descriptor ring */
2296
2297         memset(rx_ring->desc, 0, rx_ring->size);
2298
2299         rx_ring->next_to_clean = 0;
2300         rx_ring->next_to_use = 0;
2301
2302         writel(0, adapter->hw.hw_addr + rx_ring->rdh);
2303         writel(0, adapter->hw.hw_addr + rx_ring->rdt);
2304 }
2305
2306 /**
2307  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2308  * @adapter: board private structure
2309  **/
2310
2311 static void
2312 e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2313 {
2314         int i;
2315
2316         for (i = 0; i < adapter->num_rx_queues; i++)
2317                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2318 }
2319
2320 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2321  * and memory write and invalidate disabled for certain operations
2322  */
2323 static void
2324 e1000_enter_82542_rst(struct e1000_adapter *adapter)
2325 {
2326         struct net_device *netdev = adapter->netdev;
2327         uint32_t rctl;
2328
2329         e1000_pci_clear_mwi(&adapter->hw);
2330
2331         rctl = E1000_READ_REG(&adapter->hw, RCTL);
2332         rctl |= E1000_RCTL_RST;
2333         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
2334         E1000_WRITE_FLUSH(&adapter->hw);
2335         mdelay(5);
2336
2337         if (netif_running(netdev))
2338                 e1000_clean_all_rx_rings(adapter);
2339 }
2340
2341 static void
2342 e1000_leave_82542_rst(struct e1000_adapter *adapter)
2343 {
2344         struct net_device *netdev = adapter->netdev;
2345         uint32_t rctl;
2346
2347         rctl = E1000_READ_REG(&adapter->hw, RCTL);
2348         rctl &= ~E1000_RCTL_RST;
2349         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
2350         E1000_WRITE_FLUSH(&adapter->hw);
2351         mdelay(5);
2352
2353         if (adapter->hw.pci_cmd_word & PCI_COMMAND_INVALIDATE)
2354                 e1000_pci_set_mwi(&adapter->hw);
2355
2356         if (netif_running(netdev)) {
2357                 /* No need to loop, because 82542 supports only 1 queue */
2358                 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2359                 e1000_configure_rx(adapter);
2360                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2361         }
2362 }
2363
2364 /**
2365  * e1000_set_mac - Change the Ethernet Address of the NIC
2366  * @netdev: network interface device structure
2367  * @p: pointer to an address structure
2368  *
2369  * Returns 0 on success, negative on failure
2370  **/
2371
2372 static int
2373 e1000_set_mac(struct net_device *netdev, void *p)
2374 {
2375         struct e1000_adapter *adapter = netdev_priv(netdev);
2376         struct sockaddr *addr = p;
2377
2378         if (!is_valid_ether_addr(addr->sa_data))
2379                 return -EADDRNOTAVAIL;
2380
2381         /* 82542 2.0 needs to be in reset to write receive address registers */
2382
2383         if (adapter->hw.mac_type == e1000_82542_rev2_0)
2384                 e1000_enter_82542_rst(adapter);
2385
2386         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2387         memcpy(adapter->hw.mac_addr, addr->sa_data, netdev->addr_len);
2388
2389         e1000_rar_set(&adapter->hw, adapter->hw.mac_addr, 0);
2390
2391         /* With 82571 controllers, LAA may be overwritten (with the default)
2392          * due to controller reset from the other port. */
2393         if (adapter->hw.mac_type == e1000_82571) {
2394                 /* activate the work around */
2395                 adapter->hw.laa_is_present = 1;
2396
2397                 /* Hold a copy of the LAA in RAR[14] This is done so that
2398                  * between the time RAR[0] gets clobbered  and the time it
2399                  * gets fixed (in e1000_watchdog), the actual LAA is in one
2400                  * of the RARs and no incoming packets directed to this port
2401                  * are dropped. Eventaully the LAA will be in RAR[0] and
2402                  * RAR[14] */
2403                 e1000_rar_set(&adapter->hw, adapter->hw.mac_addr,
2404                                         E1000_RAR_ENTRIES - 1);
2405         }
2406
2407         if (adapter->hw.mac_type == e1000_82542_rev2_0)
2408                 e1000_leave_82542_rst(adapter);
2409
2410         return 0;
2411 }
2412
2413 /**
2414  * e1000_set_multi - Multicast and Promiscuous mode set
2415  * @netdev: network interface device structure
2416  *
2417  * The set_multi entry point is called whenever the multicast address
2418  * list or the network interface flags are updated.  This routine is
2419  * responsible for configuring the hardware for proper multicast,
2420  * promiscuous mode, and all-multi behavior.
2421  **/
2422
2423 static void
2424 e1000_set_multi(struct net_device *netdev)
2425 {
2426         struct e1000_adapter *adapter = netdev_priv(netdev);
2427         struct e1000_hw *hw = &adapter->hw;
2428         struct dev_mc_list *mc_ptr;
2429         uint32_t rctl;
2430         uint32_t hash_value;
2431         int i, rar_entries = E1000_RAR_ENTRIES;
2432         int mta_reg_count = (hw->mac_type == e1000_ich8lan) ?
2433                                 E1000_NUM_MTA_REGISTERS_ICH8LAN :
2434                                 E1000_NUM_MTA_REGISTERS;
2435
2436         if (adapter->hw.mac_type == e1000_ich8lan)
2437                 rar_entries = E1000_RAR_ENTRIES_ICH8LAN;
2438
2439         /* reserve RAR[14] for LAA over-write work-around */
2440         if (adapter->hw.mac_type == e1000_82571)
2441                 rar_entries--;
2442
2443         /* Check for Promiscuous and All Multicast modes */
2444
2445         rctl = E1000_READ_REG(hw, RCTL);
2446
2447         if (netdev->flags & IFF_PROMISC) {
2448                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2449         } else if (netdev->flags & IFF_ALLMULTI) {
2450                 rctl |= E1000_RCTL_MPE;
2451                 rctl &= ~E1000_RCTL_UPE;
2452         } else {
2453                 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2454         }
2455
2456         E1000_WRITE_REG(hw, RCTL, rctl);
2457
2458         /* 82542 2.0 needs to be in reset to write receive address registers */
2459
2460         if (hw->mac_type == e1000_82542_rev2_0)
2461                 e1000_enter_82542_rst(adapter);
2462
2463         /* load the first 14 multicast address into the exact filters 1-14
2464          * RAR 0 is used for the station MAC adddress
2465          * if there are not 14 addresses, go ahead and clear the filters
2466          * -- with 82571 controllers only 0-13 entries are filled here
2467          */
2468         mc_ptr = netdev->mc_list;
2469
2470         for (i = 1; i < rar_entries; i++) {
2471                 if (mc_ptr) {
2472                         e1000_rar_set(hw, mc_ptr->dmi_addr, i);
2473                         mc_ptr = mc_ptr->next;
2474                 } else {
2475                         E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2476                         E1000_WRITE_FLUSH(hw);
2477                         E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2478                         E1000_WRITE_FLUSH(hw);
2479                 }
2480         }
2481
2482         /* clear the old settings from the multicast hash table */
2483
2484         for (i = 0; i < mta_reg_count; i++) {
2485                 E1000_WRITE_REG_ARRAY(hw, MTA, i, 0);
2486                 E1000_WRITE_FLUSH(hw);
2487         }
2488
2489         /* load any remaining addresses into the hash table */
2490
2491         for (; mc_ptr; mc_ptr = mc_ptr->next) {
2492                 hash_value = e1000_hash_mc_addr(hw, mc_ptr->dmi_addr);
2493                 e1000_mta_set(hw, hash_value);
2494         }
2495
2496         if (hw->mac_type == e1000_82542_rev2_0)
2497                 e1000_leave_82542_rst(adapter);
2498 }
2499
2500 /* Need to wait a few seconds after link up to get diagnostic information from
2501  * the phy */
2502
2503 static void
2504 e1000_update_phy_info(unsigned long data)
2505 {
2506         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2507         e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
2508 }
2509
2510 /**
2511  * e1000_82547_tx_fifo_stall - Timer Call-back
2512  * @data: pointer to adapter cast into an unsigned long
2513  **/
2514
2515 static void
2516 e1000_82547_tx_fifo_stall(unsigned long data)
2517 {
2518         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2519         struct net_device *netdev = adapter->netdev;
2520         uint32_t tctl;
2521
2522         if (atomic_read(&adapter->tx_fifo_stall)) {
2523                 if ((E1000_READ_REG(&adapter->hw, TDT) ==
2524                     E1000_READ_REG(&adapter->hw, TDH)) &&
2525                    (E1000_READ_REG(&adapter->hw, TDFT) ==
2526                     E1000_READ_REG(&adapter->hw, TDFH)) &&
2527                    (E1000_READ_REG(&adapter->hw, TDFTS) ==
2528                     E1000_READ_REG(&adapter->hw, TDFHS))) {
2529                         tctl = E1000_READ_REG(&adapter->hw, TCTL);
2530                         E1000_WRITE_REG(&adapter->hw, TCTL,
2531                                         tctl & ~E1000_TCTL_EN);
2532                         E1000_WRITE_REG(&adapter->hw, TDFT,
2533                                         adapter->tx_head_addr);
2534                         E1000_WRITE_REG(&adapter->hw, TDFH,
2535                                         adapter->tx_head_addr);
2536                         E1000_WRITE_REG(&adapter->hw, TDFTS,
2537                                         adapter->tx_head_addr);
2538                         E1000_WRITE_REG(&adapter->hw, TDFHS,
2539                                         adapter->tx_head_addr);
2540                         E1000_WRITE_REG(&adapter->hw, TCTL, tctl);
2541                         E1000_WRITE_FLUSH(&adapter->hw);
2542
2543                         adapter->tx_fifo_head = 0;
2544                         atomic_set(&adapter->tx_fifo_stall, 0);
2545                         netif_wake_queue(netdev);
2546                 } else {
2547                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2548                 }
2549         }
2550 }
2551
2552 /**
2553  * e1000_watchdog - Timer Call-back
2554  * @data: pointer to adapter cast into an unsigned long
2555  **/
2556 static void
2557 e1000_watchdog(unsigned long data)
2558 {
2559         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2560         struct net_device *netdev = adapter->netdev;
2561         struct e1000_tx_ring *txdr = adapter->tx_ring;
2562         uint32_t link, tctl;
2563         int32_t ret_val;
2564
2565         ret_val = e1000_check_for_link(&adapter->hw);
2566         if ((ret_val == E1000_ERR_PHY) &&
2567             (adapter->hw.phy_type == e1000_phy_igp_3) &&
2568             (E1000_READ_REG(&adapter->hw, CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
2569                 /* See e1000_kumeran_lock_loss_workaround() */
2570                 DPRINTK(LINK, INFO,
2571                         "Gigabit has been disabled, downgrading speed\n");
2572         }
2573
2574         if (adapter->hw.mac_type == e1000_82573) {
2575                 e1000_enable_tx_pkt_filtering(&adapter->hw);
2576                 if (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id)
2577                         e1000_update_mng_vlan(adapter);
2578         }
2579
2580         if ((adapter->hw.media_type == e1000_media_type_internal_serdes) &&
2581            !(E1000_READ_REG(&adapter->hw, TXCW) & E1000_TXCW_ANE))
2582                 link = !adapter->hw.serdes_link_down;
2583         else
2584                 link = E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_LU;
2585
2586         if (link) {
2587                 if (!netif_carrier_ok(netdev)) {
2588                         uint32_t ctrl;
2589                         boolean_t txb2b = 1;
2590                         e1000_get_speed_and_duplex(&adapter->hw,
2591                                                    &adapter->link_speed,
2592                                                    &adapter->link_duplex);
2593
2594                         ctrl = E1000_READ_REG(&adapter->hw, CTRL);
2595                         DPRINTK(LINK, INFO, "NIC Link is Up %d Mbps %s, "
2596                                 "Flow Control: %s\n",
2597                                 adapter->link_speed,
2598                                 adapter->link_duplex == FULL_DUPLEX ?
2599                                 "Full Duplex" : "Half Duplex",
2600                                 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2601                                 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2602                                 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2603                                 E1000_CTRL_TFCE) ? "TX" : "None" )));
2604
2605                         /* tweak tx_queue_len according to speed/duplex
2606                          * and adjust the timeout factor */
2607                         netdev->tx_queue_len = adapter->tx_queue_len;
2608                         adapter->tx_timeout_factor = 1;
2609                         switch (adapter->link_speed) {
2610                         case SPEED_10:
2611                                 txb2b = 0;
2612                                 netdev->tx_queue_len = 10;
2613                                 adapter->tx_timeout_factor = 8;
2614                                 break;
2615                         case SPEED_100:
2616                                 txb2b = 0;
2617                                 netdev->tx_queue_len = 100;
2618                                 /* maybe add some timeout factor ? */
2619                                 break;
2620                         }
2621
2622                         if ((adapter->hw.mac_type == e1000_82571 ||
2623                              adapter->hw.mac_type == e1000_82572) &&
2624                             txb2b == 0) {
2625                                 uint32_t tarc0;
2626                                 tarc0 = E1000_READ_REG(&adapter->hw, TARC0);
2627                                 tarc0 &= ~(1 << 21);
2628                                 E1000_WRITE_REG(&adapter->hw, TARC0, tarc0);
2629                         }
2630
2631                         /* disable TSO for pcie and 10/100 speeds, to avoid
2632                          * some hardware issues */
2633                         if (!adapter->tso_force &&
2634                             adapter->hw.bus_type == e1000_bus_type_pci_express){
2635                                 switch (adapter->link_speed) {
2636                                 case SPEED_10:
2637                                 case SPEED_100:
2638                                         DPRINTK(PROBE,INFO,
2639                                         "10/100 speed: disabling TSO\n");
2640                                         netdev->features &= ~NETIF_F_TSO;
2641                                         netdev->features &= ~NETIF_F_TSO6;
2642                                         break;
2643                                 case SPEED_1000:
2644                                         netdev->features |= NETIF_F_TSO;
2645                                         netdev->features |= NETIF_F_TSO6;
2646                                         break;
2647                                 default:
2648                                         /* oops */
2649                                         break;
2650                                 }
2651                         }
2652
2653                         /* enable transmits in the hardware, need to do this
2654                          * after setting TARC0 */
2655                         tctl = E1000_READ_REG(&adapter->hw, TCTL);
2656                         tctl |= E1000_TCTL_EN;
2657                         E1000_WRITE_REG(&adapter->hw, TCTL, tctl);
2658
2659                         netif_carrier_on(netdev);
2660                         netif_wake_queue(netdev);
2661                         mod_timer(&adapter->phy_info_timer, round_jiffies(jiffies + 2 * HZ));
2662                         adapter->smartspeed = 0;
2663                 } else {
2664                         /* make sure the receive unit is started */
2665                         if (adapter->hw.rx_needs_kicking) {
2666                                 struct e1000_hw *hw = &adapter->hw;
2667                                 uint32_t rctl = E1000_READ_REG(hw, RCTL);
2668                                 E1000_WRITE_REG(hw, RCTL, rctl | E1000_RCTL_EN);
2669                         }
2670                 }
2671         } else {
2672                 if (netif_carrier_ok(netdev)) {
2673                         adapter->link_speed = 0;
2674                         adapter->link_duplex = 0;
2675                         DPRINTK(LINK, INFO, "NIC Link is Down\n");
2676                         netif_carrier_off(netdev);
2677                         netif_stop_queue(netdev);
2678                         mod_timer(&adapter->phy_info_timer, round_jiffies(jiffies + 2 * HZ));
2679
2680                         /* 80003ES2LAN workaround--
2681                          * For packet buffer work-around on link down event;
2682                          * disable receives in the ISR and
2683                          * reset device here in the watchdog
2684                          */
2685                         if (adapter->hw.mac_type == e1000_80003es2lan)
2686                                 /* reset device */
2687                                 schedule_work(&adapter->reset_task);
2688                 }
2689
2690                 e1000_smartspeed(adapter);
2691         }
2692
2693         e1000_update_stats(adapter);
2694
2695         adapter->hw.tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2696         adapter->tpt_old = adapter->stats.tpt;
2697         adapter->hw.collision_delta = adapter->stats.colc - adapter->colc_old;
2698         adapter->colc_old = adapter->stats.colc;
2699
2700         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2701         adapter->gorcl_old = adapter->stats.gorcl;
2702         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2703         adapter->gotcl_old = adapter->stats.gotcl;
2704
2705         e1000_update_adaptive(&adapter->hw);
2706
2707         if (!netif_carrier_ok(netdev)) {
2708                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2709                         /* We've lost link, so the controller stops DMA,
2710                          * but we've got queued Tx work that's never going
2711                          * to get done, so reset controller to flush Tx.
2712                          * (Do the reset outside of interrupt context). */
2713                         adapter->tx_timeout_count++;
2714                         schedule_work(&adapter->reset_task);
2715                 }
2716         }
2717
2718         /* Cause software interrupt to ensure rx ring is cleaned */
2719         E1000_WRITE_REG(&adapter->hw, ICS, E1000_ICS_RXDMT0);
2720
2721         /* Force detection of hung controller every watchdog period */
2722         adapter->detect_tx_hung = TRUE;
2723
2724         /* With 82571 controllers, LAA may be overwritten due to controller
2725          * reset from the other port. Set the appropriate LAA in RAR[0] */
2726         if (adapter->hw.mac_type == e1000_82571 && adapter->hw.laa_is_present)
2727                 e1000_rar_set(&adapter->hw, adapter->hw.mac_addr, 0);
2728
2729         /* Reset the timer */
2730         mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
2731 }
2732
2733 enum latency_range {
2734         lowest_latency = 0,
2735         low_latency = 1,
2736         bulk_latency = 2,
2737         latency_invalid = 255
2738 };
2739
2740 /**
2741  * e1000_update_itr - update the dynamic ITR value based on statistics
2742  *      Stores a new ITR value based on packets and byte
2743  *      counts during the last interrupt.  The advantage of per interrupt
2744  *      computation is faster updates and more accurate ITR for the current
2745  *      traffic pattern.  Constants in this function were computed
2746  *      based on theoretical maximum wire speed and thresholds were set based
2747  *      on testing data as well as attempting to minimize response time
2748  *      while increasing bulk throughput.
2749  *      this functionality is controlled by the InterruptThrottleRate module
2750  *      parameter (see e1000_param.c)
2751  * @adapter: pointer to adapter
2752  * @itr_setting: current adapter->itr
2753  * @packets: the number of packets during this measurement interval
2754  * @bytes: the number of bytes during this measurement interval
2755  **/
2756 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2757                                    uint16_t itr_setting,
2758                                    int packets,
2759                                    int bytes)
2760 {
2761         unsigned int retval = itr_setting;
2762         struct e1000_hw *hw = &adapter->hw;
2763
2764         if (unlikely(hw->mac_type < e1000_82540))
2765                 goto update_itr_done;
2766
2767         if (packets == 0)
2768                 goto update_itr_done;
2769
2770         switch (itr_setting) {
2771         case lowest_latency:
2772                 /* jumbo frames get bulk treatment*/
2773                 if (bytes/packets > 8000)
2774                         retval = bulk_latency;
2775                 else if ((packets < 5) && (bytes > 512))
2776                         retval = low_latency;
2777                 break;
2778         case low_latency:  /* 50 usec aka 20000 ints/s */
2779                 if (bytes > 10000) {
2780                         /* jumbo frames need bulk latency setting */
2781                         if (bytes/packets > 8000)
2782                                 retval = bulk_latency;
2783                         else if ((packets < 10) || ((bytes/packets) > 1200))
2784                                 retval = bulk_latency;
2785                         else if ((packets > 35))
2786                                 retval = lowest_latency;
2787                 } else if (bytes/packets > 2000)
2788                         retval = bulk_latency;
2789                 else if (packets <= 2 && bytes < 512)
2790                         retval = lowest_latency;
2791                 break;
2792         case bulk_latency: /* 250 usec aka 4000 ints/s */
2793                 if (bytes > 25000) {
2794                         if (packets > 35)
2795                                 retval = low_latency;
2796                 } else if (bytes < 6000) {
2797                         retval = low_latency;
2798                 }
2799                 break;
2800         }
2801
2802 update_itr_done:
2803         return retval;
2804 }
2805
2806 static void e1000_set_itr(struct e1000_adapter *adapter)
2807 {
2808         struct e1000_hw *hw = &adapter->hw;
2809         uint16_t current_itr;
2810         uint32_t new_itr = adapter->itr;
2811
2812         if (unlikely(hw->mac_type < e1000_82540))
2813                 return;
2814
2815         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2816         if (unlikely(adapter->link_speed != SPEED_1000)) {
2817                 current_itr = 0;
2818                 new_itr = 4000;
2819                 goto set_itr_now;
2820         }
2821
2822         adapter->tx_itr = e1000_update_itr(adapter,
2823                                     adapter->tx_itr,
2824                                     adapter->total_tx_packets,
2825                                     adapter->total_tx_bytes);
2826         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2827         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2828                 adapter->tx_itr = low_latency;
2829
2830         adapter->rx_itr = e1000_update_itr(adapter,
2831                                     adapter->rx_itr,
2832                                     adapter->total_rx_packets,
2833                                     adapter->total_rx_bytes);
2834         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2835         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2836                 adapter->rx_itr = low_latency;
2837
2838         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2839
2840         switch (current_itr) {
2841         /* counts and packets in update_itr are dependent on these numbers */
2842         case lowest_latency:
2843                 new_itr = 70000;
2844                 break;
2845         case low_latency:
2846                 new_itr = 20000; /* aka hwitr = ~200 */
2847                 break;
2848         case bulk_latency:
2849                 new_itr = 4000;
2850                 break;
2851         default:
2852                 break;
2853         }
2854
2855 set_itr_now:
2856         if (new_itr != adapter->itr) {
2857                 /* this attempts to bias the interrupt rate towards Bulk
2858                  * by adding intermediate steps when interrupt rate is
2859                  * increasing */
2860                 new_itr = new_itr > adapter->itr ?
2861                              min(adapter->itr + (new_itr >> 2), new_itr) :
2862                              new_itr;
2863                 adapter->itr = new_itr;
2864                 E1000_WRITE_REG(hw, ITR, 1000000000 / (new_itr * 256));
2865         }
2866
2867         return;
2868 }
2869
2870 #define E1000_TX_FLAGS_CSUM             0x00000001
2871 #define E1000_TX_FLAGS_VLAN             0x00000002
2872 #define E1000_TX_FLAGS_TSO              0x00000004
2873 #define E1000_TX_FLAGS_IPV4             0x00000008
2874 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2875 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2876
2877 static int
2878 e1000_tso(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2879           struct sk_buff *skb)
2880 {
2881         struct e1000_context_desc *context_desc;
2882         struct e1000_buffer *buffer_info;
2883         unsigned int i;
2884         uint32_t cmd_length = 0;
2885         uint16_t ipcse = 0, tucse, mss;
2886         uint8_t ipcss, ipcso, tucss, tucso, hdr_len;
2887         int err;
2888
2889         if (skb_is_gso(skb)) {
2890                 if (skb_header_cloned(skb)) {
2891                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2892                         if (err)
2893                                 return err;
2894                 }
2895
2896                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2897                 mss = skb_shinfo(skb)->gso_size;
2898                 if (skb->protocol == htons(ETH_P_IP)) {
2899                         struct iphdr *iph = ip_hdr(skb);
2900                         iph->tot_len = 0;
2901                         iph->check = 0;
2902                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2903                                                                  iph->daddr, 0,
2904                                                                  IPPROTO_TCP,
2905                                                                  0);
2906                         cmd_length = E1000_TXD_CMD_IP;
2907                         ipcse = skb_transport_offset(skb) - 1;
2908                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2909                         ipv6_hdr(skb)->payload_len = 0;
2910                         tcp_hdr(skb)->check =
2911                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2912                                                  &ipv6_hdr(skb)->daddr,
2913                                                  0, IPPROTO_TCP, 0);
2914                         ipcse = 0;
2915                 }
2916                 ipcss = skb_network_offset(skb);
2917                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2918                 tucss = skb_transport_offset(skb);
2919                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2920                 tucse = 0;
2921
2922                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2923                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2924
2925                 i = tx_ring->next_to_use;
2926                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2927                 buffer_info = &tx_ring->buffer_info[i];
2928
2929                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2930                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2931                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2932                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2933                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2934                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2935                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2936                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2937                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2938
2939                 buffer_info->time_stamp = jiffies;
2940                 buffer_info->next_to_watch = i;
2941
2942                 if (++i == tx_ring->count) i = 0;
2943                 tx_ring->next_to_use = i;
2944
2945                 return TRUE;
2946         }
2947         return FALSE;
2948 }
2949
2950 static boolean_t
2951 e1000_tx_csum(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2952               struct sk_buff *skb)
2953 {
2954         struct e1000_context_desc *context_desc;
2955         struct e1000_buffer *buffer_info;
2956         unsigned int i;
2957         uint8_t css;
2958
2959         if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
2960                 css = skb_transport_offset(skb);
2961
2962                 i = tx_ring->next_to_use;
2963                 buffer_info = &tx_ring->buffer_info[i];
2964                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2965
2966                 context_desc->lower_setup.ip_config = 0;
2967                 context_desc->upper_setup.tcp_fields.tucss = css;
2968                 context_desc->upper_setup.tcp_fields.tucso =
2969                         css + skb->csum_offset;
2970                 context_desc->upper_setup.tcp_fields.tucse = 0;
2971                 context_desc->tcp_seg_setup.data = 0;
2972                 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
2973
2974                 buffer_info->time_stamp = jiffies;
2975                 buffer_info->next_to_watch = i;
2976
2977                 if (unlikely(++i == tx_ring->count)) i = 0;
2978                 tx_ring->next_to_use = i;
2979
2980                 return TRUE;
2981         }
2982
2983         return FALSE;
2984 }
2985
2986 #define E1000_MAX_TXD_PWR       12
2987 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2988
2989 static int
2990 e1000_tx_map(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2991              struct sk_buff *skb, unsigned int first, unsigned int max_per_txd,
2992              unsigned int nr_frags, unsigned int mss)
2993 {
2994         struct e1000_buffer *buffer_info;
2995         unsigned int len = skb->len;
2996         unsigned int offset = 0, size, count = 0, i;
2997         unsigned int f;
2998         len -= skb->data_len;
2999
3000         i = tx_ring->next_to_use;
3001
3002         while (len) {
3003                 buffer_info = &tx_ring->buffer_info[i];
3004                 size = min(len, max_per_txd);
3005                 /* Workaround for Controller erratum --
3006                  * descriptor for non-tso packet in a linear SKB that follows a
3007                  * tso gets written back prematurely before the data is fully
3008                  * DMA'd to the controller */
3009                 if (!skb->data_len && tx_ring->last_tx_tso &&
3010                     !skb_is_gso(skb)) {
3011                         tx_ring->last_tx_tso = 0;
3012                         size -= 4;
3013                 }
3014
3015                 /* Workaround for premature desc write-backs
3016                  * in TSO mode.  Append 4-byte sentinel desc */
3017                 if (unlikely(mss && !nr_frags && size == len && size > 8))
3018                         size -= 4;
3019                 /* work-around for errata 10 and it applies
3020                  * to all controllers in PCI-X mode
3021                  * The fix is to make sure that the first descriptor of a
3022                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
3023                  */
3024                 if (unlikely((adapter->hw.bus_type == e1000_bus_type_pcix) &&
3025                                 (size > 2015) && count == 0))
3026                         size = 2015;
3027
3028                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
3029                  * terminating buffers within evenly-aligned dwords. */
3030                 if (unlikely(adapter->pcix_82544 &&
3031                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
3032                    size > 4))
3033                         size -= 4;
3034
3035                 buffer_info->length = size;
3036                 buffer_info->dma =
3037                         pci_map_single(adapter->pdev,
3038                                 skb->data + offset,
3039                                 size,
3040                                 PCI_DMA_TODEVICE);
3041                 buffer_info->time_stamp = jiffies;
3042                 buffer_info->next_to_watch = i;
3043
3044                 len -= size;
3045                 offset += size;
3046                 count++;
3047                 if (unlikely(++i == tx_ring->count)) i = 0;
3048         }
3049
3050         for (f = 0; f < nr_frags; f++) {
3051                 struct skb_frag_struct *frag;
3052
3053                 frag = &skb_shinfo(skb)->frags[f];
3054                 len = frag->size;
3055                 offset = frag->page_offset;
3056
3057                 while (len) {
3058                         buffer_info = &tx_ring->buffer_info[i];
3059                         size = min(len, max_per_txd);
3060                         /* Workaround for premature desc write-backs
3061                          * in TSO mode.  Append 4-byte sentinel desc */
3062                         if (unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
3063                                 size -= 4;
3064                         /* Workaround for potential 82544 hang in PCI-X.
3065                          * Avoid terminating buffers within evenly-aligned
3066                          * dwords. */
3067                         if (unlikely(adapter->pcix_82544 &&
3068                            !((unsigned long)(frag->page+offset+size-1) & 4) &&
3069                            size > 4))
3070                                 size -= 4;
3071
3072                         buffer_info->length = size;
3073                         buffer_info->dma =
3074                                 pci_map_page(adapter->pdev,
3075                                         frag->page,
3076                                         offset,
3077                                         size,
3078                                         PCI_DMA_TODEVICE);
3079                         buffer_info->time_stamp = jiffies;
3080                         buffer_info->next_to_watch = i;
3081
3082                         len -= size;
3083                         offset += size;
3084                         count++;
3085                         if (unlikely(++i == tx_ring->count)) i = 0;
3086                 }
3087         }
3088
3089         i = (i == 0) ? tx_ring->count - 1 : i - 1;
3090         tx_ring->buffer_info[i].skb = skb;
3091         tx_ring->buffer_info[first].next_to_watch = i;
3092
3093         return count;
3094 }
3095
3096 static void
3097 e1000_tx_queue(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
3098                int tx_flags, int count)
3099 {
3100         struct e1000_tx_desc *tx_desc = NULL;
3101         struct e1000_buffer *buffer_info;
3102         uint32_t txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3103         unsigned int i;
3104
3105         if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
3106                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3107                              E1000_TXD_CMD_TSE;
3108                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3109
3110                 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
3111                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3112         }
3113
3114         if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
3115                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3116                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3117         }
3118
3119         if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
3120                 txd_lower |= E1000_TXD_CMD_VLE;
3121                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3122         }
3123
3124         i = tx_ring->next_to_use;
3125
3126         while (count--) {
3127                 buffer_info = &tx_ring->buffer_info[i];
3128                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3129                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3130                 tx_desc->lower.data =
3131                         cpu_to_le32(txd_lower | buffer_info->length);
3132                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3133                 if (unlikely(++i == tx_ring->count)) i = 0;
3134         }
3135
3136         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3137
3138         /* Force memory writes to complete before letting h/w
3139          * know there are new descriptors to fetch.  (Only
3140          * applicable for weak-ordered memory model archs,
3141          * such as IA-64). */
3142         wmb();
3143
3144         tx_ring->next_to_use = i;
3145         writel(i, adapter->hw.hw_addr + tx_ring->tdt);
3146         /* we need this if more than one processor can write to our tail
3147          * at a time, it syncronizes IO on IA64/Altix systems */
3148         mmiowb();
3149 }
3150
3151 /**
3152  * 82547 workaround to avoid controller hang in half-duplex environment.
3153  * The workaround is to avoid queuing a large packet that would span
3154  * the internal Tx FIFO ring boundary by notifying the stack to resend
3155  * the packet at a later time.  This gives the Tx FIFO an opportunity to
3156  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
3157  * to the beginning of the Tx FIFO.
3158  **/
3159
3160 #define E1000_FIFO_HDR                  0x10
3161 #define E1000_82547_PAD_LEN             0x3E0
3162
3163 static int
3164 e1000_82547_fifo_workaround(struct e1000_adapter *adapter, struct sk_buff *skb)
3165 {
3166         uint32_t fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
3167         uint32_t skb_fifo_len = skb->len + E1000_FIFO_HDR;
3168
3169         skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
3170
3171         if (adapter->link_duplex != HALF_DUPLEX)
3172                 goto no_fifo_stall_required;
3173
3174         if (atomic_read(&adapter->tx_fifo_stall))
3175                 return 1;
3176
3177         if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
3178                 atomic_set(&adapter->tx_fifo_stall, 1);
3179                 return 1;
3180         }
3181
3182 no_fifo_stall_required:
3183         adapter->tx_fifo_head += skb_fifo_len;
3184         if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
3185                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
3186         return 0;
3187 }
3188
3189 #define MINIMUM_DHCP_PACKET_SIZE 282
3190 static int
3191 e1000_transfer_dhcp_info(struct e1000_adapter *adapter, struct sk_buff *skb)
3192 {
3193         struct e1000_hw *hw =  &adapter->hw;
3194         uint16_t length, offset;
3195         if (vlan_tx_tag_present(skb)) {
3196                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
3197                         ( adapter->hw.mng_cookie.status &
3198                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) )
3199                         return 0;
3200         }
3201         if (skb->len > MINIMUM_DHCP_PACKET_SIZE) {
3202                 struct ethhdr *eth = (struct ethhdr *) skb->data;
3203                 if ((htons(ETH_P_IP) == eth->h_proto)) {
3204                         const struct iphdr *ip =
3205                                 (struct iphdr *)((uint8_t *)skb->data+14);
3206                         if (IPPROTO_UDP == ip->protocol) {
3207                                 struct udphdr *udp =
3208                                         (struct udphdr *)((uint8_t *)ip +
3209                                                 (ip->ihl << 2));
3210                                 if (ntohs(udp->dest) == 67) {
3211                                         offset = (uint8_t *)udp + 8 - skb->data;
3212                                         length = skb->len - offset;
3213
3214                                         return e1000_mng_write_dhcp_info(hw,
3215                                                         (uint8_t *)udp + 8,
3216                                                         length);
3217                                 }
3218                         }
3219                 }
3220         }
3221         return 0;
3222 }
3223
3224 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3225 {
3226         struct e1000_adapter *adapter = netdev_priv(netdev);
3227         struct e1000_tx_ring *tx_ring = adapter->tx_ring;
3228
3229         netif_stop_queue(netdev);
3230         /* Herbert's original patch had:
3231          *  smp_mb__after_netif_stop_queue();
3232          * but since that doesn't exist yet, just open code it. */
3233         smp_mb();
3234
3235         /* We need to check again in a case another CPU has just
3236          * made room available. */
3237         if (likely(E1000_DESC_UNUSED(tx_ring) < size))
3238                 return -EBUSY;
3239
3240         /* A reprieve! */
3241         netif_start_queue(netdev);
3242         ++adapter->restart_queue;
3243         return 0;
3244 }
3245
3246 static int e1000_maybe_stop_tx(struct net_device *netdev,
3247                                struct e1000_tx_ring *tx_ring, int size)
3248 {
3249         if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
3250                 return 0;
3251         return __e1000_maybe_stop_tx(netdev, size);
3252 }
3253
3254 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3255 static int
3256 e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3257 {
3258         struct e1000_adapter *adapter = netdev_priv(netdev);
3259         struct e1000_tx_ring *tx_ring;
3260         unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
3261         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3262         unsigned int tx_flags = 0;
3263         unsigned int len = skb->len;
3264         unsigned long flags;
3265         unsigned int nr_frags = 0;
3266         unsigned int mss = 0;
3267         int count = 0;
3268         int tso;
3269         unsigned int f;
3270         len -= skb->data_len;
3271
3272         /* This goes back to the question of how to logically map a tx queue
3273          * to a flow.  Right now, performance is impacted slightly negatively
3274          * if using multiple tx queues.  If the stack breaks away from a
3275          * single qdisc implementation, we can look at this again. */
3276         tx_ring = adapter->tx_ring;
3277
3278         if (unlikely(skb->len <= 0)) {
3279                 dev_kfree_skb_any(skb);
3280                 return NETDEV_TX_OK;
3281         }
3282
3283         /* 82571 and newer doesn't need the workaround that limited descriptor
3284          * length to 4kB */
3285         if (adapter->hw.mac_type >= e1000_82571)
3286                 max_per_txd = 8192;
3287
3288         mss = skb_shinfo(skb)->gso_size;
3289         /* The controller does a simple calculation to
3290          * make sure there is enough room in the FIFO before
3291          * initiating the DMA for each buffer.  The calc is:
3292          * 4 = ceil(buffer len/mss).  To make sure we don't
3293          * overrun the FIFO, adjust the max buffer len if mss
3294          * drops. */
3295         if (mss) {
3296                 uint8_t hdr_len;
3297                 max_per_txd = min(mss << 2, max_per_txd);
3298                 max_txd_pwr = fls(max_per_txd) - 1;
3299
3300                 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
3301                 * points to just header, pull a few bytes of payload from
3302                 * frags into skb->data */
3303                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3304                 if (skb->data_len && (hdr_len == (skb->len - skb->data_len))) {
3305                         switch (adapter->hw.mac_type) {
3306                                 unsigned int pull_size;
3307                         case e1000_82544:
3308                                 /* Make sure we have room to chop off 4 bytes,
3309                                  * and that the end alignment will work out to
3310                                  * this hardware's requirements
3311                                  * NOTE: this is a TSO only workaround
3312                                  * if end byte alignment not correct move us
3313                                  * into the next dword */
3314                                 if ((unsigned long)(skb_tail_pointer(skb) - 1) & 4)
3315                                         break;
3316                                 /* fall through */
3317                         case e1000_82571:
3318                         case e1000_82572:
3319                         case e1000_82573:
3320                         case e1000_ich8lan:
3321                                 pull_size = min((unsigned int)4, skb->data_len);
3322                                 if (!__pskb_pull_tail(skb, pull_size)) {
3323                                         DPRINTK(DRV, ERR,
3324                                                 "__pskb_pull_tail failed.\n");
3325                                         dev_kfree_skb_any(skb);
3326                                         return NETDEV_TX_OK;
3327                                 }
3328                                 len = skb->len - skb->data_len;
3329                                 break;
3330                         default:
3331                                 /* do nothing */
3332                                 break;
3333                         }
3334                 }
3335         }
3336
3337         /* reserve a descriptor for the offload context */
3338         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3339                 count++;
3340         count++;
3341
3342         /* Controller Erratum workaround */
3343         if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3344                 count++;
3345
3346         count += TXD_USE_COUNT(len, max_txd_pwr);
3347
3348         if (adapter->pcix_82544)
3349                 count++;
3350
3351         /* work-around for errata 10 and it applies to all controllers
3352          * in PCI-X mode, so add one more descriptor to the count
3353          */
3354         if (unlikely((adapter->hw.bus_type == e1000_bus_type_pcix) &&
3355                         (len > 2015)))
3356                 count++;
3357
3358         nr_frags = skb_shinfo(skb)->nr_frags;
3359         for (f = 0; f < nr_frags; f++)
3360                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3361                                        max_txd_pwr);
3362         if (adapter->pcix_82544)
3363                 count += nr_frags;
3364
3365
3366         if (adapter->hw.tx_pkt_filtering &&
3367             (adapter->hw.mac_type == e1000_82573))
3368                 e1000_transfer_dhcp_info(adapter, skb);
3369
3370         if (!spin_trylock_irqsave(&tx_ring->tx_lock, flags))
3371                 /* Collision - tell upper layer to requeue */
3372                 return NETDEV_TX_LOCKED;
3373
3374         /* need: count + 2 desc gap to keep tail from touching
3375          * head, otherwise try next time */
3376         if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2))) {
3377                 spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3378                 return NETDEV_TX_BUSY;
3379         }
3380
3381         if (unlikely(adapter->hw.mac_type == e1000_82547)) {
3382                 if (unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
3383                         netif_stop_queue(netdev);
3384                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
3385                         spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3386                         return NETDEV_TX_BUSY;
3387                 }
3388         }
3389
3390         if (unlikely(adapter->vlgrp && vlan_tx_tag_present(skb))) {
3391                 tx_flags |= E1000_TX_FLAGS_VLAN;
3392                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3393         }
3394
3395         first = tx_ring->next_to_use;
3396
3397         tso = e1000_tso(adapter, tx_ring, skb);
3398         if (tso < 0) {
3399                 dev_kfree_skb_any(skb);
3400                 spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3401                 return NETDEV_TX_OK;
3402         }
3403
3404         if (likely(tso)) {
3405                 tx_ring->last_tx_tso = 1;
3406                 tx_flags |= E1000_TX_FLAGS_TSO;
3407         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3408                 tx_flags |= E1000_TX_FLAGS_CSUM;
3409
3410         /* Old method was to assume IPv4 packet by default if TSO was enabled.
3411          * 82571 hardware supports TSO capabilities for IPv6 as well...
3412          * no longer assume, we must. */
3413         if (likely(skb->protocol == htons(ETH_P_IP)))
3414                 tx_flags |= E1000_TX_FLAGS_IPV4;
3415
3416         e1000_tx_queue(adapter, tx_ring, tx_flags,
3417                        e1000_tx_map(adapter, tx_ring, skb, first,
3418                                     max_per_txd, nr_frags, mss));
3419
3420         netdev->trans_start = jiffies;
3421
3422         /* Make sure there is space in the ring for the next send. */
3423         e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3424
3425         spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3426         return NETDEV_TX_OK;
3427 }
3428
3429 /**
3430  * e1000_tx_timeout - Respond to a Tx Hang
3431  * @netdev: network interface device structure
3432  **/
3433
3434 static void
3435 e1000_tx_timeout(struct net_device *netdev)
3436 {
3437         struct e1000_adapter *adapter = netdev_priv(netdev);
3438
3439         /* Do the reset outside of interrupt context */
3440         adapter->tx_timeout_count++;
3441         schedule_work(&adapter->reset_task);
3442 }
3443
3444 static void
3445 e1000_reset_task(struct work_struct *work)
3446 {
3447         struct e1000_adapter *adapter =
3448                 container_of(work, struct e1000_adapter, reset_task);
3449
3450         e1000_reinit_locked(adapter);
3451 }
3452
3453 /**
3454  * e1000_get_stats - Get System Network Statistics
3455  * @netdev: network interface device structure
3456  *
3457  * Returns the address of the device statistics structure.
3458  * The statistics are actually updated from the timer callback.
3459  **/
3460
3461 static struct net_device_stats *
3462 e1000_get_stats(struct net_device *netdev)
3463 {
3464         struct e1000_adapter *adapter = netdev_priv(netdev);
3465
3466         /* only return the current stats */
3467         return &adapter->net_stats;
3468 }
3469
3470 /**
3471  * e1000_change_mtu - Change the Maximum Transfer Unit
3472  * @netdev: network interface device structure
3473  * @new_mtu: new value for maximum frame size
3474  *
3475  * Returns 0 on success, negative on failure
3476  **/
3477
3478 static int
3479 e1000_change_mtu(struct net_device *netdev, int new_mtu)
3480 {
3481         struct e1000_adapter *adapter = netdev_priv(netdev);
3482         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3483         uint16_t eeprom_data = 0;
3484
3485         if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3486             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3487                 DPRINTK(PROBE, ERR, "Invalid MTU setting\n");
3488                 return -EINVAL;
3489         }
3490
3491         /* Adapter-specific max frame size limits. */
3492         switch (adapter->hw.mac_type) {
3493         case e1000_undefined ... e1000_82542_rev2_1:
3494         case e1000_ich8lan:
3495                 if (max_frame > MAXIMUM_ETHERNET_FRAME_SIZE) {
3496                         DPRINTK(PROBE, ERR, "Jumbo Frames not supported.\n");
3497                         return -EINVAL;
3498                 }
3499                 break;
3500         case e1000_82573:
3501                 /* Jumbo Frames not supported if:
3502                  * - this is not an 82573L device
3503                  * - ASPM is enabled in any way (0x1A bits 3:2) */
3504                 e1000_read_eeprom(&adapter->hw, EEPROM_INIT_3GIO_3, 1,
3505                                   &eeprom_data);
3506                 if ((adapter->hw.device_id != E1000_DEV_ID_82573L) ||
3507                     (eeprom_data & EEPROM_WORD1A_ASPM_MASK)) {
3508                         if (max_frame > MAXIMUM_ETHERNET_FRAME_SIZE) {
3509                                 DPRINTK(PROBE, ERR,
3510                                         "Jumbo Frames not supported.\n");
3511                                 return -EINVAL;
3512                         }
3513                         break;
3514                 }
3515                 /* ERT will be enabled later to enable wire speed receives */
3516
3517                 /* fall through to get support */
3518         case e1000_82571:
3519         case e1000_82572:
3520         case e1000_80003es2lan:
3521 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3522                 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3523                         DPRINTK(PROBE, ERR, "MTU > 9216 not supported.\n");
3524                         return -EINVAL;
3525                 }
3526                 break;
3527         default:
3528                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3529                 break;
3530         }
3531
3532         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3533          * means we reserve 2 more, this pushes us to allocate from the next
3534          * larger slab size
3535          * i.e. RXBUFFER_2048 --> size-4096 slab */
3536
3537         if (max_frame <= E1000_RXBUFFER_256)
3538                 adapter->rx_buffer_len = E1000_RXBUFFER_256;
3539         else if (max_frame <= E1000_RXBUFFER_512)
3540                 adapter->rx_buffer_len = E1000_RXBUFFER_512;
3541         else if (max_frame <= E1000_RXBUFFER_1024)
3542                 adapter->rx_buffer_len = E1000_RXBUFFER_1024;
3543         else if (max_frame <= E1000_RXBUFFER_2048)
3544                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3545         else if (max_frame <= E1000_RXBUFFER_4096)
3546                 adapter->rx_buffer_len = E1000_RXBUFFER_4096;
3547         else if (max_frame <= E1000_RXBUFFER_8192)
3548                 adapter->rx_buffer_len = E1000_RXBUFFER_8192;
3549         else if (max_frame <= E1000_RXBUFFER_16384)
3550                 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3551
3552         /* adjust allocation if LPE protects us, and we aren't using SBP */
3553         if (!adapter->hw.tbi_compatibility_on &&
3554             ((max_frame == MAXIMUM_ETHERNET_FRAME_SIZE) ||
3555              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3556                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3557
3558         netdev->mtu = new_mtu;
3559         adapter->hw.max_frame_size = max_frame;
3560
3561         if (netif_running(netdev))
3562                 e1000_reinit_locked(adapter);
3563
3564         return 0;
3565 }
3566
3567 /**
3568  * e1000_update_stats - Update the board statistics counters
3569  * @adapter: board private structure
3570  **/
3571
3572 void
3573 e1000_update_stats(struct e1000_adapter *adapter)
3574 {
3575         struct e1000_hw *hw = &adapter->hw;
3576         struct pci_dev *pdev = adapter->pdev;
3577         unsigned long flags;
3578         uint16_t phy_tmp;
3579
3580 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3581
3582         /*
3583          * Prevent stats update while adapter is being reset, or if the pci
3584          * connection is down.
3585          */
3586         if (adapter->link_speed == 0)
3587                 return;
3588         if (pci_channel_offline(pdev))
3589                 return;
3590
3591         spin_lock_irqsave(&adapter->stats_lock, flags);
3592
3593         /* these counters are modified from e1000_adjust_tbi_stats,
3594          * called from the interrupt context, so they must only
3595          * be written while holding adapter->stats_lock
3596          */
3597
3598         adapter->stats.crcerrs += E1000_READ_REG(hw, CRCERRS);
3599         adapter->stats.gprc += E1000_READ_REG(hw, GPRC);
3600         adapter->stats.gorcl += E1000_READ_REG(hw, GORCL);
3601         adapter->stats.gorch += E1000_READ_REG(hw, GORCH);
3602         adapter->stats.bprc += E1000_READ_REG(hw, BPRC);
3603         adapter->stats.mprc += E1000_READ_REG(hw, MPRC);
3604         adapter->stats.roc += E1000_READ_REG(hw, ROC);
3605
3606         if (adapter->hw.mac_type != e1000_ich8lan) {
3607                 adapter->stats.prc64 += E1000_READ_REG(hw, PRC64);
3608                 adapter->stats.prc127 += E1000_READ_REG(hw, PRC127);
3609                 adapter->stats.prc255 += E1000_READ_REG(hw, PRC255);
3610                 adapter->stats.prc511 += E1000_READ_REG(hw, PRC511);
3611                 adapter->stats.prc1023 += E1000_READ_REG(hw, PRC1023);
3612                 adapter->stats.prc1522 += E1000_READ_REG(hw, PRC1522);
3613         }
3614
3615         adapter->stats.symerrs += E1000_READ_REG(hw, SYMERRS);
3616         adapter->stats.mpc += E1000_READ_REG(hw, MPC);
3617         adapter->stats.scc += E1000_READ_REG(hw, SCC);
3618         adapter->stats.ecol += E1000_READ_REG(hw, ECOL);
3619         adapter->stats.mcc += E1000_READ_REG(hw, MCC);
3620         adapter->stats.latecol += E1000_READ_REG(hw, LATECOL);
3621         adapter->stats.dc += E1000_READ_REG(hw, DC);
3622         adapter->stats.sec += E1000_READ_REG(hw, SEC);
3623         adapter->stats.rlec += E1000_READ_REG(hw, RLEC);
3624         adapter->stats.xonrxc += E1000_READ_REG(hw, XONRXC);
3625         adapter->stats.xontxc += E1000_READ_REG(hw, XONTXC);
3626         adapter->stats.xoffrxc += E1000_READ_REG(hw, XOFFRXC);
3627         adapter->stats.xofftxc += E1000_READ_REG(hw, XOFFTXC);
3628         adapter->stats.fcruc += E1000_READ_REG(hw, FCRUC);
3629         adapter->stats.gptc += E1000_READ_REG(hw, GPTC);
3630         adapter->stats.gotcl += E1000_READ_REG(hw, GOTCL);
3631         adapter->stats.gotch += E1000_READ_REG(hw, GOTCH);
3632         adapter->stats.rnbc += E1000_READ_REG(hw, RNBC);
3633         adapter->stats.ruc += E1000_READ_REG(hw, RUC);
3634         adapter->stats.rfc += E1000_READ_REG(hw, RFC);
3635         adapter->stats.rjc += E1000_READ_REG(hw, RJC);
3636         adapter->stats.torl += E1000_READ_REG(hw, TORL);
3637         adapter->stats.torh += E1000_READ_REG(hw, TORH);
3638         adapter->stats.totl += E1000_READ_REG(hw, TOTL);
3639         adapter->stats.toth += E1000_READ_REG(hw, TOTH);
3640         adapter->stats.tpr += E1000_READ_REG(hw, TPR);
3641
3642         if (adapter->hw.mac_type != e1000_ich8lan) {
3643                 adapter->stats.ptc64 += E1000_READ_REG(hw, PTC64);
3644                 adapter->stats.ptc127 += E1000_READ_REG(hw, PTC127);
3645                 adapter->stats.ptc255 += E1000_READ_REG(hw, PTC255);
3646                 adapter->stats.ptc511 += E1000_READ_REG(hw, PTC511);
3647                 adapter->stats.ptc1023 += E1000_READ_REG(hw, PTC1023);
3648                 adapter->stats.ptc1522 += E1000_READ_REG(hw, PTC1522);
3649         }
3650
3651         adapter->stats.mptc += E1000_READ_REG(hw, MPTC);
3652         adapter->stats.bptc += E1000_READ_REG(hw, BPTC);
3653
3654         /* used for adaptive IFS */
3655
3656         hw->tx_packet_delta = E1000_READ_REG(hw, TPT);
3657         adapter->stats.tpt += hw->tx_packet_delta;
3658         hw->collision_delta = E1000_READ_REG(hw, COLC);
3659         adapter->stats.colc += hw->collision_delta;
3660
3661         if (hw->mac_type >= e1000_82543) {
3662                 adapter->stats.algnerrc += E1000_READ_REG(hw, ALGNERRC);
3663                 adapter->stats.rxerrc += E1000_READ_REG(hw, RXERRC);
3664                 adapter->stats.tncrs += E1000_READ_REG(hw, TNCRS);
3665                 adapter->stats.cexterr += E1000_READ_REG(hw, CEXTERR);
3666                 adapter->stats.tsctc += E1000_READ_REG(hw, TSCTC);
3667                 adapter->stats.tsctfc += E1000_READ_REG(hw, TSCTFC);
3668         }
3669         if (hw->mac_type > e1000_82547_rev_2) {
3670                 adapter->stats.iac += E1000_READ_REG(hw, IAC);
3671                 adapter->stats.icrxoc += E1000_READ_REG(hw, ICRXOC);
3672
3673                 if (adapter->hw.mac_type != e1000_ich8lan) {
3674                         adapter->stats.icrxptc += E1000_READ_REG(hw, ICRXPTC);
3675                         adapter->stats.icrxatc += E1000_READ_REG(hw, ICRXATC);
3676                         adapter->stats.ictxptc += E1000_READ_REG(hw, ICTXPTC);
3677                         adapter->stats.ictxatc += E1000_READ_REG(hw, ICTXATC);
3678                         adapter->stats.ictxqec += E1000_READ_REG(hw, ICTXQEC);
3679                         adapter->stats.ictxqmtc += E1000_READ_REG(hw, ICTXQMTC);
3680                         adapter->stats.icrxdmtc += E1000_READ_REG(hw, ICRXDMTC);
3681                 }
3682         }
3683
3684         /* Fill out the OS statistics structure */
3685         adapter->net_stats.rx_packets = adapter->stats.gprc;
3686         adapter->net_stats.tx_packets = adapter->stats.gptc;
3687         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
3688         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
3689         adapter->net_stats.multicast = adapter->stats.mprc;
3690         adapter->net_stats.collisions = adapter->stats.colc;
3691
3692         /* Rx Errors */
3693
3694         /* RLEC on some newer hardware can be incorrect so build
3695         * our own version based on RUC and ROC */
3696         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3697                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3698                 adapter->stats.ruc + adapter->stats.roc +
3699                 adapter->stats.cexterr;
3700         adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3701         adapter->net_stats.rx_length_errors = adapter->stats.rlerrc;
3702         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3703         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3704         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3705
3706         /* Tx Errors */
3707         adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3708         adapter->net_stats.tx_errors = adapter->stats.txerrc;
3709         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3710         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3711         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3712         if (adapter->hw.bad_tx_carr_stats_fd &&
3713             adapter->link_duplex == FULL_DUPLEX) {
3714                 adapter->net_stats.tx_carrier_errors = 0;
3715                 adapter->stats.tncrs = 0;
3716         }
3717
3718         /* Tx Dropped needs to be maintained elsewhere */
3719
3720         /* Phy Stats */
3721         if (hw->media_type == e1000_media_type_copper) {
3722                 if ((adapter->link_speed == SPEED_1000) &&
3723                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3724                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3725                         adapter->phy_stats.idle_errors += phy_tmp;
3726                 }
3727
3728                 if ((hw->mac_type <= e1000_82546) &&
3729                    (hw->phy_type == e1000_phy_m88) &&
3730                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3731                         adapter->phy_stats.receive_errors += phy_tmp;
3732         }
3733
3734         /* Management Stats */
3735         if (adapter->hw.has_smbus) {
3736                 adapter->stats.mgptc += E1000_READ_REG(hw, MGTPTC);
3737                 adapter->stats.mgprc += E1000_READ_REG(hw, MGTPRC);
3738                 adapter->stats.mgpdc += E1000_READ_REG(hw, MGTPDC);
3739         }
3740
3741         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3742 }
3743
3744 /**
3745  * e1000_intr_msi - Interrupt Handler
3746  * @irq: interrupt number
3747  * @data: pointer to a network interface device structure
3748  **/
3749
3750 static irqreturn_t
3751 e1000_intr_msi(int irq, void *data)
3752 {
3753         struct net_device *netdev = data;
3754         struct e1000_adapter *adapter = netdev_priv(netdev);
3755         struct e1000_hw *hw = &adapter->hw;
3756 #ifndef CONFIG_E1000_NAPI
3757         int i;
3758 #endif
3759         uint32_t icr = E1000_READ_REG(hw, ICR);
3760
3761 #ifdef CONFIG_E1000_NAPI
3762         /* read ICR disables interrupts using IAM, so keep up with our
3763          * enable/disable accounting */
3764         atomic_inc(&adapter->irq_sem);
3765 #endif
3766         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3767                 hw->get_link_status = 1;
3768                 /* 80003ES2LAN workaround-- For packet buffer work-around on
3769                  * link down event; disable receives here in the ISR and reset
3770                  * adapter in watchdog */
3771                 if (netif_carrier_ok(netdev) &&
3772                     (adapter->hw.mac_type == e1000_80003es2lan)) {
3773                         /* disable receives */
3774                         uint32_t rctl = E1000_READ_REG(hw, RCTL);
3775                         E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
3776                 }
3777                 /* guard against interrupt when we're going down */
3778                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3779                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3780         }
3781
3782 #ifdef CONFIG_E1000_NAPI
3783         if (likely(netif_rx_schedule_prep(netdev))) {
3784                 adapter->total_tx_bytes = 0;
3785                 adapter->total_tx_packets = 0;
3786                 adapter->total_rx_bytes = 0;
3787                 adapter->total_rx_packets = 0;
3788                 __netif_rx_schedule(netdev);
3789         } else
3790                 e1000_irq_enable(adapter);
3791 #else
3792         adapter->total_tx_bytes = 0;
3793         adapter->total_rx_bytes = 0;
3794         adapter->total_tx_packets = 0;
3795         adapter->total_rx_packets = 0;
3796
3797         for (i = 0; i < E1000_MAX_INTR; i++)
3798                 if (unlikely(!adapter->clean_rx(adapter, adapter->rx_ring) &
3799                    !e1000_clean_tx_irq(adapter, adapter->tx_ring)))
3800                         break;
3801
3802         if (likely(adapter->itr_setting & 3))
3803                 e1000_set_itr(adapter);
3804 #endif
3805
3806         return IRQ_HANDLED;
3807 }
3808
3809 /**
3810  * e1000_intr - Interrupt Handler
3811  * @irq: interrupt number
3812  * @data: pointer to a network interface device structure
3813  **/
3814
3815 static irqreturn_t
3816 e1000_intr(int irq, void *data)
3817 {
3818         struct net_device *netdev = data;
3819         struct e1000_adapter *adapter = netdev_priv(netdev);
3820         struct e1000_hw *hw = &adapter->hw;
3821         uint32_t rctl, icr = E1000_READ_REG(hw, ICR);
3822 #ifndef CONFIG_E1000_NAPI
3823         int i;
3824 #endif
3825         if (unlikely(!icr))
3826                 return IRQ_NONE;  /* Not our interrupt */
3827
3828 #ifdef CONFIG_E1000_NAPI
3829         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
3830          * not set, then the adapter didn't send an interrupt */
3831         if (unlikely(hw->mac_type >= e1000_82571 &&
3832                      !(icr & E1000_ICR_INT_ASSERTED)))
3833                 return IRQ_NONE;
3834
3835         /* Interrupt Auto-Mask...upon reading ICR,
3836          * interrupts are masked.  No need for the
3837          * IMC write, but it does mean we should
3838          * account for it ASAP. */
3839         if (likely(hw->mac_type >= e1000_82571))
3840                 atomic_inc(&adapter->irq_sem);
3841 #endif
3842
3843         if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3844                 hw->get_link_status = 1;
3845                 /* 80003ES2LAN workaround--
3846                  * For packet buffer work-around on link down event;
3847                  * disable receives here in the ISR and
3848                  * reset adapter in watchdog
3849                  */
3850                 if (netif_carrier_ok(netdev) &&
3851                     (adapter->hw.mac_type == e1000_80003es2lan)) {
3852                         /* disable receives */
3853                         rctl = E1000_READ_REG(hw, RCTL);
3854                         E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
3855                 }
3856                 /* guard against interrupt when we're going down */
3857                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3858                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3859         }
3860
3861 #ifdef CONFIG_E1000_NAPI
3862         if (unlikely(hw->mac_type < e1000_82571)) {
3863                 /* disable interrupts, without the synchronize_irq bit */
3864                 atomic_inc(&adapter->irq_sem);
3865                 E1000_WRITE_REG(hw, IMC, ~0);
3866                 E1000_WRITE_FLUSH(hw);
3867         }
3868         if (likely(netif_rx_schedule_prep(netdev))) {
3869                 adapter->total_tx_bytes = 0;
3870                 adapter->total_tx_packets = 0;
3871                 adapter->total_rx_bytes = 0;
3872                 adapter->total_rx_packets = 0;
3873                 __netif_rx_schedule(netdev);
3874         } else
3875                 /* this really should not happen! if it does it is basically a
3876                  * bug, but not a hard error, so enable ints and continue */
3877                 e1000_irq_enable(adapter);
3878 #else
3879         /* Writing IMC and IMS is needed for 82547.
3880          * Due to Hub Link bus being occupied, an interrupt
3881          * de-assertion message is not able to be sent.
3882          * When an interrupt assertion message is generated later,
3883          * two messages are re-ordered and sent out.
3884          * That causes APIC to think 82547 is in de-assertion
3885          * state, while 82547 is in assertion state, resulting
3886          * in dead lock. Writing IMC forces 82547 into
3887          * de-assertion state.
3888          */
3889         if (hw->mac_type == e1000_82547 || hw->mac_type == e1000_82547_rev_2) {
3890                 atomic_inc(&adapter->irq_sem);
3891                 E1000_WRITE_REG(hw, IMC, ~0);
3892         }
3893
3894         adapter->total_tx_bytes = 0;
3895         adapter->total_rx_bytes = 0;
3896         adapter->total_tx_packets = 0;
3897         adapter->total_rx_packets = 0;
3898
3899         for (i = 0; i < E1000_MAX_INTR; i++)
3900                 if (unlikely(!adapter->clean_rx(adapter, adapter->rx_ring) &
3901                    !e1000_clean_tx_irq(adapter, adapter->tx_ring)))
3902                         break;
3903
3904         if (likely(adapter->itr_setting & 3))
3905                 e1000_set_itr(adapter);
3906
3907         if (hw->mac_type == e1000_82547 || hw->mac_type == e1000_82547_rev_2)
3908                 e1000_irq_enable(adapter);
3909
3910 #endif
3911         return IRQ_HANDLED;
3912 }
3913
3914 #ifdef CONFIG_E1000_NAPI
3915 /**
3916  * e1000_clean - NAPI Rx polling callback
3917  * @adapter: board private structure
3918  **/
3919
3920 static int
3921 e1000_clean(struct net_device *poll_dev, int *budget)
3922 {
3923         struct e1000_adapter *adapter;
3924         int work_to_do = min(*budget, poll_dev->quota);
3925         int tx_cleaned = 0, work_done = 0;
3926
3927         /* Must NOT use netdev_priv macro here. */
3928         adapter = poll_dev->priv;
3929
3930         /* Keep link state information with original netdev */
3931         if (!netif_carrier_ok(poll_dev))
3932                 goto quit_polling;
3933
3934         /* e1000_clean is called per-cpu.  This lock protects
3935          * tx_ring[0] from being cleaned by multiple cpus
3936          * simultaneously.  A failure obtaining the lock means
3937          * tx_ring[0] is currently being cleaned anyway. */
3938         if (spin_trylock(&adapter->tx_queue_lock)) {
3939                 tx_cleaned = e1000_clean_tx_irq(adapter,
3940                                                 &adapter->tx_ring[0]);
3941                 spin_unlock(&adapter->tx_queue_lock);
3942         }
3943
3944         adapter->clean_rx(adapter, &adapter->rx_ring[0],
3945                           &work_done, work_to_do);
3946
3947         *budget -= work_done;
3948         poll_dev->quota -= work_done;
3949
3950         /* If no Tx and not enough Rx work done, exit the polling mode */
3951         if ((!tx_cleaned && (work_done == 0)) ||
3952            !netif_running(poll_dev)) {
3953 quit_polling:
3954                 if (likely(adapter->itr_setting & 3))
3955                         e1000_set_itr(adapter);
3956                 netif_rx_complete(poll_dev);
3957                 e1000_irq_enable(adapter);
3958                 return 0;
3959         }
3960
3961         return 1;
3962 }
3963
3964 #endif
3965 /**
3966  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3967  * @adapter: board private structure
3968  **/
3969
3970 static boolean_t
3971 e1000_clean_tx_irq(struct e1000_adapter *adapter,
3972                    struct e1000_tx_ring *tx_ring)
3973 {
3974         struct net_device *netdev = adapter->netdev;
3975         struct e1000_tx_desc *tx_desc, *eop_desc;
3976         struct e1000_buffer *buffer_info;
3977         unsigned int i, eop;
3978 #ifdef CONFIG_E1000_NAPI
3979         unsigned int count = 0;
3980 #endif
3981         boolean_t cleaned = FALSE;
3982         unsigned int total_tx_bytes=0, total_tx_packets=0;
3983
3984         i = tx_ring->next_to_clean;
3985         eop = tx_ring->buffer_info[i].next_to_watch;
3986         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3987
3988         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
3989                 for (cleaned = FALSE; !cleaned; ) {
3990                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3991                         buffer_info = &tx_ring->buffer_info[i];
3992                         cleaned = (i == eop);
3993
3994                         if (cleaned) {
3995                                 struct sk_buff *skb = buffer_info->skb;
3996                                 unsigned int segs, bytecount;
3997                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3998                                 /* multiply data chunks by size of headers */
3999                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
4000                                             skb->len;
4001                                 total_tx_packets += segs;
4002                                 total_tx_bytes += bytecount;
4003                         }
4004                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
4005                         tx_desc->upper.data = 0;
4006
4007                         if (unlikely(++i == tx_ring->count)) i = 0;
4008                 }
4009
4010                 eop = tx_ring->buffer_info[i].next_to_watch;
4011                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
4012 #ifdef CONFIG_E1000_NAPI
4013 #define E1000_TX_WEIGHT 64
4014                 /* weight of a sort for tx, to avoid endless transmit cleanup */
4015                 if (count++ == E1000_TX_WEIGHT) break;
4016 #endif
4017         }
4018
4019         tx_ring->next_to_clean = i;
4020
4021 #define TX_WAKE_THRESHOLD 32
4022         if (unlikely(cleaned && netif_carrier_ok(netdev) &&
4023                      E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
4024                 /* Make sure that anybody stopping the queue after this
4025                  * sees the new next_to_clean.
4026                  */
4027                 smp_mb();
4028                 if (netif_queue_stopped(netdev)) {
4029                         netif_wake_queue(netdev);
4030                         ++adapter->restart_queue;
4031                 }
4032         }
4033
4034         if (adapter->detect_tx_hung) {
4035                 /* Detect a transmit hang in hardware, this serializes the
4036                  * check with the clearing of time_stamp and movement of i */
4037                 adapter->detect_tx_hung = FALSE;
4038                 if (tx_ring->buffer_info[eop].dma &&
4039                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
4040                                (adapter->tx_timeout_factor * HZ))
4041                     && !(E1000_READ_REG(&adapter->hw, STATUS) &
4042                          E1000_STATUS_TXOFF)) {
4043
4044                         /* detected Tx unit hang */
4045                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
4046                                         "  Tx Queue             <%lu>\n"
4047                                         "  TDH                  <%x>\n"
4048                                         "  TDT                  <%x>\n"
4049                                         "  next_to_use          <%x>\n"
4050                                         "  next_to_clean        <%x>\n"
4051                                         "buffer_info[next_to_clean]\n"
4052                                         "  time_stamp           <%lx>\n"
4053                                         "  next_to_watch        <%x>\n"
4054                                         "  jiffies              <%lx>\n"
4055                                         "  next_to_watch.status <%x>\n",
4056                                 (unsigned long)((tx_ring - adapter->tx_ring) /
4057                                         sizeof(struct e1000_tx_ring)),
4058                                 readl(adapter->hw.hw_addr + tx_ring->tdh),
4059                                 readl(adapter->hw.hw_addr + tx_ring->tdt),
4060                                 tx_ring->next_to_use,
4061                                 tx_ring->next_to_clean,
4062                                 tx_ring->buffer_info[eop].time_stamp,
4063                                 eop,
4064                                 jiffies,
4065                                 eop_desc->upper.fields.status);
4066                         netif_stop_queue(netdev);
4067                 }
4068         }
4069         adapter->total_tx_bytes += total_tx_bytes;
4070         adapter->total_tx_packets += total_tx_packets;
4071         return cleaned;
4072 }
4073
4074 /**
4075  * e1000_rx_checksum - Receive Checksum Offload for 82543
4076  * @adapter:     board private structure
4077  * @status_err:  receive descriptor status and error fields
4078  * @csum:        receive descriptor csum field
4079  * @sk_buff:     socket buffer with received data
4080  **/
4081
4082 static void
4083 e1000_rx_checksum(struct e1000_adapter *adapter,
4084                   uint32_t status_err, uint32_t csum,
4085                   struct sk_buff *skb)
4086 {
4087         uint16_t status = (uint16_t)status_err;
4088         uint8_t errors = (uint8_t)(status_err >> 24);
4089         skb->ip_summed = CHECKSUM_NONE;
4090
4091         /* 82543 or newer only */
4092         if (unlikely(adapter->hw.mac_type < e1000_82543)) return;
4093         /* Ignore Checksum bit is set */
4094         if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
4095         /* TCP/UDP checksum error bit is set */
4096         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
4097                 /* let the stack verify checksum errors */
4098                 adapter->hw_csum_err++;
4099                 return;
4100         }
4101         /* TCP/UDP Checksum has not been calculated */
4102         if (adapter->hw.mac_type <= e1000_82547_rev_2) {
4103                 if (!(status & E1000_RXD_STAT_TCPCS))
4104                         return;
4105         } else {
4106                 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
4107                         return;
4108         }
4109         /* It must be a TCP or UDP packet with a valid checksum */
4110         if (likely(status & E1000_RXD_STAT_TCPCS)) {
4111                 /* TCP checksum is good */
4112                 skb->ip_summed = CHECKSUM_UNNECESSARY;
4113         } else if (adapter->hw.mac_type > e1000_82547_rev_2) {
4114                 /* IP fragment with UDP payload */
4115                 /* Hardware complements the payload checksum, so we undo it
4116                  * and then put the value in host order for further stack use.
4117                  */
4118                 csum = ntohl(csum ^ 0xFFFF);
4119                 skb->csum = csum;
4120                 skb->ip_summed = CHECKSUM_COMPLETE;
4121         }
4122         adapter->hw_csum_good++;
4123 }
4124
4125 /**
4126  * e1000_clean_rx_irq - Send received data up the network stack; legacy
4127  * @adapter: board private structure
4128  **/
4129
4130 static boolean_t
4131 #ifdef CONFIG_E1000_NAPI
4132 e1000_clean_rx_irq(struct e1000_adapter *adapter,
4133                    struct e1000_rx_ring *rx_ring,
4134                    int *work_done, int work_to_do)
4135 #else
4136 e1000_clean_rx_irq(struct e1000_adapter *adapter,
4137                    struct e1000_rx_ring *rx_ring)
4138 #endif
4139 {
4140         struct net_device *netdev = adapter->netdev;
4141         struct pci_dev *pdev = adapter->pdev;
4142         struct e1000_rx_desc *rx_desc, *next_rxd;
4143         struct e1000_buffer *buffer_info, *next_buffer;
4144         unsigned long flags;
4145         uint32_t length;
4146         uint8_t last_byte;
4147         unsigned int i;
4148         int cleaned_count = 0;
4149         boolean_t cleaned = FALSE;
4150         unsigned int total_rx_bytes=0, total_rx_packets=0;
4151
4152         i = rx_ring->next_to_clean;
4153         rx_desc = E1000_RX_DESC(*rx_ring, i);
4154         buffer_info = &rx_ring->buffer_info[i];
4155
4156         while (rx_desc->status & E1000_RXD_STAT_DD) {
4157                 struct sk_buff *skb;
4158                 u8 status;
4159
4160 #ifdef CONFIG_E1000_NAPI
4161                 if (*work_done >= work_to_do)
4162                         break;
4163                 (*work_done)++;
4164 #endif
4165                 status = rx_desc->status;
4166                 skb = buffer_info->skb;
4167                 buffer_info->skb = NULL;
4168
4169                 prefetch(skb->data - NET_IP_ALIGN);
4170
4171                 if (++i == rx_ring->count) i = 0;
4172                 next_rxd = E1000_RX_DESC(*rx_ring, i);
4173                 prefetch(next_rxd);
4174
4175                 next_buffer = &rx_ring->buffer_info[i];
4176
4177                 cleaned = TRUE;
4178                 cleaned_count++;
4179                 pci_unmap_single(pdev,
4180                                  buffer_info->dma,
4181                                  buffer_info->length,
4182                                  PCI_DMA_FROMDEVICE);
4183
4184                 length = le16_to_cpu(rx_desc->length);
4185
4186                 if (unlikely(!(status & E1000_RXD_STAT_EOP))) {
4187                         /* All receives must fit into a single buffer */
4188                         E1000_DBG("%s: Receive packet consumed multiple"
4189                                   " buffers\n", netdev->name);
4190                         /* recycle */
4191                         buffer_info->skb = skb;
4192                         goto next_desc;
4193                 }
4194
4195                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
4196                         last_byte = *(skb->data + length - 1);
4197                         if (TBI_ACCEPT(&adapter->hw, status,
4198                                       rx_desc->errors, length, last_byte)) {
4199                                 spin_lock_irqsave(&adapter->stats_lock, flags);
4200                                 e1000_tbi_adjust_stats(&adapter->hw,
4201                                                        &adapter->stats,
4202                                                        length, skb->data);
4203                                 spin_unlock_irqrestore(&adapter->stats_lock,
4204                                                        flags);
4205                                 length--;
4206                         } else {
4207                                 /* recycle */
4208                                 buffer_info->skb = skb;
4209                                 goto next_desc;
4210                         }
4211                 }
4212
4213                 /* adjust length to remove Ethernet CRC, this must be
4214                  * done after the TBI_ACCEPT workaround above */
4215                 length -= 4;
4216
4217                 /* probably a little skewed due to removing CRC */
4218                 total_rx_bytes += length;
4219                 total_rx_packets++;
4220
4221                 /* code added for copybreak, this should improve
4222                  * performance for small packets with large amounts
4223                  * of reassembly being done in the stack */
4224                 if (length < copybreak) {
4225                         struct sk_buff *new_skb =
4226                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
4227                         if (new_skb) {
4228                                 skb_reserve(new_skb, NET_IP_ALIGN);
4229                                 skb_copy_to_linear_data_offset(new_skb,
4230                                                                -NET_IP_ALIGN,
4231                                                                (skb->data -
4232                                                                 NET_IP_ALIGN),
4233                                                                (length +
4234                                                                 NET_IP_ALIGN));
4235                                 /* save the skb in buffer_info as good */
4236                                 buffer_info->skb = skb;
4237                                 skb = new_skb;
4238                         }
4239                         /* else just continue with the old one */
4240                 }
4241                 /* end copybreak code */
4242                 skb_put(skb, length);
4243
4244                 /* Receive Checksum Offload */
4245                 e1000_rx_checksum(adapter,
4246                                   (uint32_t)(status) |
4247                                   ((uint32_t)(rx_desc->errors) << 24),
4248                                   le16_to_cpu(rx_desc->csum), skb);
4249
4250                 skb->protocol = eth_type_trans(skb, netdev);
4251 #ifdef CONFIG_E1000_NAPI
4252                 if (unlikely(adapter->vlgrp &&
4253                             (status & E1000_RXD_STAT_VP))) {
4254                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
4255                                                  le16_to_cpu(rx_desc->special) &
4256                                                  E1000_RXD_SPC_VLAN_MASK);
4257                 } else {
4258                         netif_receive_skb(skb);
4259                 }
4260 #else /* CONFIG_E1000_NAPI */
4261                 if (unlikely(adapter->vlgrp &&
4262                             (status & E1000_RXD_STAT_VP))) {
4263                         vlan_hwaccel_rx(skb, adapter->vlgrp,
4264                                         le16_to_cpu(rx_desc->special) &
4265                                         E1000_RXD_SPC_VLAN_MASK);
4266                 } else {
4267                         netif_rx(skb);
4268                 }
4269 #endif /* CONFIG_E1000_NAPI */
4270                 netdev->last_rx = jiffies;
4271
4272 next_desc:
4273                 rx_desc->status = 0;
4274
4275                 /* return some buffers to hardware, one at a time is too slow */
4276                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4277                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4278                         cleaned_count = 0;
4279                 }
4280
4281                 /* use prefetched values */
4282                 rx_desc = next_rxd;
4283                 buffer_info = next_buffer;
4284         }
4285         rx_ring->next_to_clean = i;
4286
4287         cleaned_count = E1000_DESC_UNUSED(rx_ring);
4288         if (cleaned_count)
4289                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4290
4291         adapter->total_rx_packets += total_rx_packets;
4292         adapter->total_rx_bytes += total_rx_bytes;
4293         return cleaned;
4294 }
4295
4296 /**
4297  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
4298  * @adapter: board private structure
4299  **/
4300
4301 static boolean_t
4302 #ifdef CONFIG_E1000_NAPI
4303 e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
4304                       struct e1000_rx_ring *rx_ring,
4305                       int *work_done, int work_to_do)
4306 #else
4307 e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
4308                       struct e1000_rx_ring *rx_ring)
4309 #endif
4310 {
4311         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
4312         struct net_device *netdev = adapter->netdev;
4313         struct pci_dev *pdev = adapter->pdev;
4314         struct e1000_buffer *buffer_info, *next_buffer;
4315         struct e1000_ps_page *ps_page;
4316         struct e1000_ps_page_dma *ps_page_dma;
4317         struct sk_buff *skb;
4318         unsigned int i, j;
4319         uint32_t length, staterr;
4320         int cleaned_count = 0;
4321         boolean_t cleaned = FALSE;
4322         unsigned int total_rx_bytes=0, total_rx_packets=0;
4323
4324         i = rx_ring->next_to_clean;
4325         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
4326         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
4327         buffer_info = &rx_ring->buffer_info[i];
4328
4329         while (staterr & E1000_RXD_STAT_DD) {
4330                 ps_page = &rx_ring->ps_page[i];
4331                 ps_page_dma = &rx_ring->ps_page_dma[i];
4332 #ifdef CONFIG_E1000_NAPI
4333                 if (unlikely(*work_done >= work_to_do))
4334                         break;
4335                 (*work_done)++;
4336 #endif
4337                 skb = buffer_info->skb;
4338
4339                 /* in the packet split case this is header only */
4340                 prefetch(skb->data - NET_IP_ALIGN);
4341
4342                 if (++i == rx_ring->count) i = 0;
4343                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
4344                 prefetch(next_rxd);
4345
4346                 next_buffer = &rx_ring->buffer_info[i];
4347
4348                 cleaned = TRUE;
4349                 cleaned_count++;
4350                 pci_unmap_single(pdev, buffer_info->dma,
4351                                  buffer_info->length,
4352                                  PCI_DMA_FROMDEVICE);
4353
4354                 if (unlikely(!(staterr & E1000_RXD_STAT_EOP))) {
4355                         E1000_DBG("%s: Packet Split buffers didn't pick up"
4356                                   " the full packet\n", netdev->name);
4357                         dev_kfree_skb_irq(skb);
4358                         goto next_desc;
4359                 }
4360
4361                 if (unlikely(staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK)) {
4362                         dev_kfree_skb_irq(skb);
4363                         goto next_desc;
4364                 }
4365
4366                 length = le16_to_cpu(rx_desc->wb.middle.length0);
4367
4368                 if (unlikely(!length)) {
4369                         E1000_DBG("%s: Last part of the packet spanning"
4370                                   " multiple descriptors\n", netdev->name);
4371                         dev_kfree_skb_irq(skb);
4372                         goto next_desc;
4373                 }
4374
4375                 /* Good Receive */
4376                 skb_put(skb, length);
4377
4378                 {
4379                 /* this looks ugly, but it seems compiler issues make it
4380                    more efficient than reusing j */
4381                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
4382
4383                 /* page alloc/put takes too long and effects small packet
4384                  * throughput, so unsplit small packets and save the alloc/put*/
4385                 if (l1 && (l1 <= copybreak) && ((length + l1) <= adapter->rx_ps_bsize0)) {
4386                         u8 *vaddr;
4387                         /* there is no documentation about how to call
4388                          * kmap_atomic, so we can't hold the mapping
4389                          * very long */
4390                         pci_dma_sync_single_for_cpu(pdev,
4391                                 ps_page_dma->ps_page_dma[0],
4392                                 PAGE_SIZE,
4393                                 PCI_DMA_FROMDEVICE);
4394                         vaddr = kmap_atomic(ps_page->ps_page[0],
4395                                             KM_SKB_DATA_SOFTIRQ);
4396                         memcpy(skb_tail_pointer(skb), vaddr, l1);
4397                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
4398                         pci_dma_sync_single_for_device(pdev,
4399                                 ps_page_dma->ps_page_dma[0],
4400                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
4401                         /* remove the CRC */
4402                         l1 -= 4;
4403                         skb_put(skb, l1);
4404                         goto copydone;
4405                 } /* if */
4406                 }
4407
4408                 for (j = 0; j < adapter->rx_ps_pages; j++) {
4409                         if (!(length= le16_to_cpu(rx_desc->wb.upper.length[j])))
4410                                 break;
4411                         pci_unmap_page(pdev, ps_page_dma->ps_page_dma[j],
4412                                         PAGE_SIZE, PCI_DMA_FROMDEVICE);
4413                         ps_page_dma->ps_page_dma[j] = 0;
4414                         skb_fill_page_desc(skb, j, ps_page->ps_page[j], 0,
4415                                            length);
4416                         ps_page->ps_page[j] = NULL;
4417                         skb->len += length;
4418                         skb->data_len += length;
4419                         skb->truesize += length;
4420                 }
4421
4422                 /* strip the ethernet crc, problem is we're using pages now so
4423                  * this whole operation can get a little cpu intensive */
4424                 pskb_trim(skb, skb->len - 4);
4425
4426 copydone:
4427                 total_rx_bytes += skb->len;
4428                 total_rx_packets++;
4429
4430                 e1000_rx_checksum(adapter, staterr,
4431                                   le16_to_cpu(rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
4432                 skb->protocol = eth_type_trans(skb, netdev);
4433
4434                 if (likely(rx_desc->wb.upper.header_status &
4435                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP)))
4436                         adapter->rx_hdr_split++;
4437 #ifdef CONFIG_E1000_NAPI
4438                 if (unlikely(adapter->vlgrp && (staterr & E1000_RXD_STAT_VP))) {
4439                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
4440                                 le16_to_cpu(rx_desc->wb.middle.vlan) &
4441                                 E1000_RXD_SPC_VLAN_MASK);
4442                 } else {
4443                         netif_receive_skb(skb);
4444                 }
4445 #else /* CONFIG_E1000_NAPI */
4446                 if (unlikely(adapter->vlgrp && (staterr & E1000_RXD_STAT_VP))) {
4447                         vlan_hwaccel_rx(skb, adapter->vlgrp,
4448                                 le16_to_cpu(rx_desc->wb.middle.vlan) &
4449                                 E1000_RXD_SPC_VLAN_MASK);
4450                 } else {
4451                         netif_rx(skb);
4452                 }
4453 #endif /* CONFIG_E1000_NAPI */
4454                 netdev->last_rx = jiffies;
4455
4456 next_desc:
4457                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
4458                 buffer_info->skb = NULL;
4459
4460                 /* return some buffers to hardware, one at a time is too slow */
4461                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4462                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4463                         cleaned_count = 0;
4464                 }
4465
4466                 /* use prefetched values */
4467                 rx_desc = next_rxd;
4468                 buffer_info = next_buffer;
4469
4470                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
4471         }
4472         rx_ring->next_to_clean = i;
4473
4474         cleaned_count = E1000_DESC_UNUSED(rx_ring);
4475         if (cleaned_count)
4476                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4477
4478         adapter->total_rx_packets += total_rx_packets;
4479         adapter->total_rx_bytes += total_rx_bytes;
4480         return cleaned;
4481 }
4482
4483 /**
4484  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4485  * @adapter: address of board private structure
4486  **/
4487
4488 static void
4489 e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4490                        struct e1000_rx_ring *rx_ring,
4491                        int cleaned_count)
4492 {
4493         struct net_device *netdev = adapter->netdev;
4494         struct pci_dev *pdev = adapter->pdev;
4495         struct e1000_rx_desc *rx_desc;
4496         struct e1000_buffer *buffer_info;
4497         struct sk_buff *skb;
4498         unsigned int i;
4499         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
4500
4501         i = rx_ring->next_to_use;
4502         buffer_info = &rx_ring->buffer_info[i];
4503
4504         while (cleaned_count--) {
4505                 skb = buffer_info->skb;
4506                 if (skb) {
4507                         skb_trim(skb, 0);
4508                         goto map_skb;
4509                 }
4510
4511                 skb = netdev_alloc_skb(netdev, bufsz);
4512                 if (unlikely(!skb)) {
4513                         /* Better luck next round */
4514                         adapter->alloc_rx_buff_failed++;
4515                         break;
4516                 }
4517
4518                 /* Fix for errata 23, can't cross 64kB boundary */
4519                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4520                         struct sk_buff *oldskb = skb;
4521                         DPRINTK(RX_ERR, ERR, "skb align check failed: %u bytes "
4522                                              "at %p\n", bufsz, skb->data);
4523                         /* Try again, without freeing the previous */
4524                         skb = netdev_alloc_skb(netdev, bufsz);
4525                         /* Failed allocation, critical failure */
4526                         if (!skb) {
4527                                 dev_kfree_skb(oldskb);
4528                                 break;
4529                         }
4530
4531                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4532                                 /* give up */
4533                                 dev_kfree_skb(skb);
4534                                 dev_kfree_skb(oldskb);
4535                                 break; /* while !buffer_info->skb */
4536                         }
4537
4538                         /* Use new allocation */
4539                         dev_kfree_skb(oldskb);
4540                 }
4541                 /* Make buffer alignment 2 beyond a 16 byte boundary
4542                  * this will result in a 16 byte aligned IP header after
4543                  * the 14 byte MAC header is removed
4544                  */
4545                 skb_reserve(skb, NET_IP_ALIGN);
4546
4547                 buffer_info->skb = skb;
4548                 buffer_info->length = adapter->rx_buffer_len;
4549 map_skb:
4550                 buffer_info->dma = pci_map_single(pdev,
4551                                                   skb->data,
4552                                                   adapter->rx_buffer_len,
4553                                                   PCI_DMA_FROMDEVICE);
4554
4555                 /* Fix for errata 23, can't cross 64kB boundary */
4556                 if (!e1000_check_64k_bound(adapter,
4557                                         (void *)(unsigned long)buffer_info->dma,
4558                                         adapter->rx_buffer_len)) {
4559                         DPRINTK(RX_ERR, ERR,
4560                                 "dma align check failed: %u bytes at %p\n",
4561                                 adapter->rx_buffer_len,
4562                                 (void *)(unsigned long)buffer_info->dma);
4563                         dev_kfree_skb(skb);
4564                         buffer_info->skb = NULL;
4565
4566                         pci_unmap_single(pdev, buffer_info->dma,
4567                                          adapter->rx_buffer_len,
4568                                          PCI_DMA_FROMDEVICE);
4569
4570                         break; /* while !buffer_info->skb */
4571                 }
4572                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4573                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4574
4575                 if (unlikely(++i == rx_ring->count))
4576                         i = 0;
4577                 buffer_info = &rx_ring->buffer_info[i];
4578         }
4579
4580         if (likely(rx_ring->next_to_use != i)) {
4581                 rx_ring->next_to_use = i;
4582                 if (unlikely(i-- == 0))
4583                         i = (rx_ring->count - 1);
4584
4585                 /* Force memory writes to complete before letting h/w
4586                  * know there are new descriptors to fetch.  (Only
4587                  * applicable for weak-ordered memory model archs,
4588                  * such as IA-64). */
4589                 wmb();
4590                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4591         }
4592 }
4593
4594 /**
4595  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
4596  * @adapter: address of board private structure
4597  **/
4598
4599 static void
4600 e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
4601                           struct e1000_rx_ring *rx_ring,
4602                           int cleaned_count)
4603 {
4604         struct net_device *netdev = adapter->netdev;
4605         struct pci_dev *pdev = adapter->pdev;
4606         union e1000_rx_desc_packet_split *rx_desc;
4607         struct e1000_buffer *buffer_info;
4608         struct e1000_ps_page *ps_page;
4609         struct e1000_ps_page_dma *ps_page_dma;
4610         struct sk_buff *skb;
4611         unsigned int i, j;
4612
4613         i = rx_ring->next_to_use;
4614         buffer_info = &rx_ring->buffer_info[i];
4615         ps_page = &rx_ring->ps_page[i];
4616         ps_page_dma = &rx_ring->ps_page_dma[i];
4617
4618         while (cleaned_count--) {
4619                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
4620
4621                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
4622                         if (j < adapter->rx_ps_pages) {
4623                                 if (likely(!ps_page->ps_page[j])) {
4624                                         ps_page->ps_page[j] =
4625                                                 alloc_page(GFP_ATOMIC);
4626                                         if (unlikely(!ps_page->ps_page[j])) {
4627                                                 adapter->alloc_rx_buff_failed++;
4628                                                 goto no_buffers;
4629                                         }
4630                                         ps_page_dma->ps_page_dma[j] =
4631                                                 pci_map_page(pdev,
4632                                                             ps_page->ps_page[j],
4633                                                             0, PAGE_SIZE,
4634                                                             PCI_DMA_FROMDEVICE);
4635                                 }
4636                                 /* Refresh the desc even if buffer_addrs didn't
4637                                  * change because each write-back erases
4638                                  * this info.
4639                                  */
4640                                 rx_desc->read.buffer_addr[j+1] =
4641                                      cpu_to_le64(ps_page_dma->ps_page_dma[j]);
4642                         } else
4643                                 rx_desc->read.buffer_addr[j+1] = ~0;
4644                 }
4645
4646                 skb = netdev_alloc_skb(netdev,
4647                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
4648
4649                 if (unlikely(!skb)) {
4650                         adapter->alloc_rx_buff_failed++;
4651                         break;
4652                 }
4653
4654                 /* Make buffer alignment 2 beyond a 16 byte boundary
4655                  * this will result in a 16 byte aligned IP header after
4656                  * the 14 byte MAC header is removed
4657                  */
4658                 skb_reserve(skb, NET_IP_ALIGN);
4659
4660                 buffer_info->skb = skb;
4661                 buffer_info->length = adapter->rx_ps_bsize0;
4662                 buffer_info->dma = pci_map_single(pdev, skb->data,
4663                                                   adapter->rx_ps_bsize0,
4664                                                   PCI_DMA_FROMDEVICE);
4665
4666                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
4667
4668                 if (unlikely(++i == rx_ring->count)) i = 0;
4669                 buffer_info = &rx_ring->buffer_info[i];
4670                 ps_page = &rx_ring->ps_page[i];
4671                 ps_page_dma = &rx_ring->ps_page_dma[i];
4672         }
4673
4674 no_buffers:
4675         if (likely(rx_ring->next_to_use != i)) {
4676                 rx_ring->next_to_use = i;
4677                 if (unlikely(i-- == 0)) i = (rx_ring->count - 1);
4678
4679                 /* Force memory writes to complete before letting h/w
4680                  * know there are new descriptors to fetch.  (Only
4681                  * applicable for weak-ordered memory model archs,
4682                  * such as IA-64). */
4683                 wmb();
4684                 /* Hardware increments by 16 bytes, but packet split
4685                  * descriptors are 32 bytes...so we increment tail
4686                  * twice as much.
4687                  */
4688                 writel(i<<1, adapter->hw.hw_addr + rx_ring->rdt);
4689         }
4690 }
4691
4692 /**
4693  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4694  * @adapter:
4695  **/
4696
4697 static void
4698 e1000_smartspeed(struct e1000_adapter *adapter)
4699 {
4700         uint16_t phy_status;
4701         uint16_t phy_ctrl;
4702
4703         if ((adapter->hw.phy_type != e1000_phy_igp) || !adapter->hw.autoneg ||
4704            !(adapter->hw.autoneg_advertised & ADVERTISE_1000_FULL))
4705                 return;
4706
4707         if (adapter->smartspeed == 0) {
4708                 /* If Master/Slave config fault is asserted twice,
4709                  * we assume back-to-back */
4710                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_status);
4711                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4712                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_status);
4713                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4714                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_ctrl);
4715                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4716                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4717                         e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL,
4718                                             phy_ctrl);
4719                         adapter->smartspeed++;
4720                         if (!e1000_phy_setup_autoneg(&adapter->hw) &&
4721                            !e1000_read_phy_reg(&adapter->hw, PHY_CTRL,
4722                                                &phy_ctrl)) {
4723                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4724                                              MII_CR_RESTART_AUTO_NEG);
4725                                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL,
4726                                                     phy_ctrl);
4727                         }
4728                 }
4729                 return;
4730         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4731                 /* If still no link, perhaps using 2/3 pair cable */
4732                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_ctrl);
4733                 phy_ctrl |= CR_1000T_MS_ENABLE;
4734                 e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL, phy_ctrl);
4735                 if (!e1000_phy_setup_autoneg(&adapter->hw) &&
4736                    !e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &phy_ctrl)) {
4737                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4738                                      MII_CR_RESTART_AUTO_NEG);
4739                         e1000_write_phy_reg(&adapter->hw, PHY_CTRL, phy_ctrl);
4740                 }
4741         }
4742         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4743         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4744                 adapter->smartspeed = 0;
4745 }
4746
4747 /**
4748  * e1000_ioctl -
4749  * @netdev:
4750  * @ifreq:
4751  * @cmd:
4752  **/
4753
4754 static int
4755 e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4756 {
4757         switch (cmd) {
4758         case SIOCGMIIPHY:
4759         case SIOCGMIIREG:
4760         case SIOCSMIIREG:
4761                 return e1000_mii_ioctl(netdev, ifr, cmd);
4762         default:
4763                 return -EOPNOTSUPP;
4764         }
4765 }
4766
4767 /**
4768  * e1000_mii_ioctl -
4769  * @netdev:
4770  * @ifreq:
4771  * @cmd:
4772  **/
4773
4774 static int
4775 e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4776 {
4777         struct e1000_adapter *adapter = netdev_priv(netdev);
4778         struct mii_ioctl_data *data = if_mii(ifr);
4779         int retval;
4780         uint16_t mii_reg;
4781         uint16_t spddplx;
4782         unsigned long flags;
4783
4784         if (adapter->hw.media_type != e1000_media_type_copper)
4785                 return -EOPNOTSUPP;
4786
4787         switch (cmd) {
4788         case SIOCGMIIPHY:
4789                 data->phy_id = adapter->hw.phy_addr;
4790                 break;
4791         case SIOCGMIIREG:
4792                 if (!capable(CAP_NET_ADMIN))
4793                         return -EPERM;
4794                 spin_lock_irqsave(&adapter->stats_lock, flags);
4795                 if (e1000_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
4796                                    &data->val_out)) {
4797                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4798                         return -EIO;
4799                 }
4800                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4801                 break;
4802         case SIOCSMIIREG:
4803                 if (!capable(CAP_NET_ADMIN))
4804                         return -EPERM;
4805                 if (data->reg_num & ~(0x1F))
4806                         return -EFAULT;
4807                 mii_reg = data->val_in;
4808                 spin_lock_irqsave(&adapter->stats_lock, flags);
4809                 if (e1000_write_phy_reg(&adapter->hw, data->reg_num,
4810                                         mii_reg)) {
4811                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4812                         return -EIO;
4813                 }
4814                 if (adapter->hw.media_type == e1000_media_type_copper) {
4815                         switch (data->reg_num) {
4816                         case PHY_CTRL:
4817                                 if (mii_reg & MII_CR_POWER_DOWN)
4818                                         break;
4819                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4820                                         adapter->hw.autoneg = 1;
4821                                         adapter->hw.autoneg_advertised = 0x2F;
4822                                 } else {
4823                                         if (mii_reg & 0x40)
4824                                                 spddplx = SPEED_1000;
4825                                         else if (mii_reg & 0x2000)
4826                                                 spddplx = SPEED_100;
4827                                         else
4828                                                 spddplx = SPEED_10;
4829                                         spddplx += (mii_reg & 0x100)
4830                                                    ? DUPLEX_FULL :
4831                                                    DUPLEX_HALF;
4832                                         retval = e1000_set_spd_dplx(adapter,
4833                                                                     spddplx);
4834                                         if (retval) {
4835                                                 spin_unlock_irqrestore(
4836                                                         &adapter->stats_lock,
4837                                                         flags);
4838                                                 return retval;
4839                                         }
4840                                 }
4841                                 if (netif_running(adapter->netdev))
4842                                         e1000_reinit_locked(adapter);
4843                                 else
4844                                         e1000_reset(adapter);
4845                                 break;
4846                         case M88E1000_PHY_SPEC_CTRL:
4847                         case M88E1000_EXT_PHY_SPEC_CTRL:
4848                                 if (e1000_phy_reset(&adapter->hw)) {
4849                                         spin_unlock_irqrestore(
4850                                                 &adapter->stats_lock, flags);
4851                                         return -EIO;
4852                                 }
4853                                 break;
4854                         }
4855                 } else {
4856                         switch (data->reg_num) {
4857                         case PHY_CTRL:
4858                                 if (mii_reg & MII_CR_POWER_DOWN)
4859                                         break;
4860                                 if (netif_running(adapter->netdev))
4861                                         e1000_reinit_locked(adapter);
4862                                 else
4863                                         e1000_reset(adapter);
4864                                 break;
4865                         }
4866                 }
4867                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4868                 break;
4869         default:
4870                 return -EOPNOTSUPP;
4871         }
4872         return E1000_SUCCESS;
4873 }
4874
4875 void
4876 e1000_pci_set_mwi(struct e1000_hw *hw)
4877 {
4878         struct e1000_adapter *adapter = hw->back;
4879         int ret_val = pci_set_mwi(adapter->pdev);
4880
4881         if (ret_val)
4882                 DPRINTK(PROBE, ERR, "Error in setting MWI\n");
4883 }
4884
4885 void
4886 e1000_pci_clear_mwi(struct e1000_hw *hw)
4887 {
4888         struct e1000_adapter *adapter = hw->back;
4889
4890         pci_clear_mwi(adapter->pdev);
4891 }
4892
4893 void
4894 e1000_read_pci_cfg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4895 {
4896         struct e1000_adapter *adapter = hw->back;
4897
4898         pci_read_config_word(adapter->pdev, reg, value);
4899 }
4900
4901 void
4902 e1000_write_pci_cfg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4903 {
4904         struct e1000_adapter *adapter = hw->back;
4905
4906         pci_write_config_word(adapter->pdev, reg, *value);
4907 }
4908
4909 int32_t
4910 e1000_read_pcie_cap_reg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4911 {
4912     struct e1000_adapter *adapter = hw->back;
4913     uint16_t cap_offset;
4914
4915     cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
4916     if (!cap_offset)
4917         return -E1000_ERR_CONFIG;
4918
4919     pci_read_config_word(adapter->pdev, cap_offset + reg, value);
4920
4921     return E1000_SUCCESS;
4922 }
4923
4924 void
4925 e1000_io_write(struct e1000_hw *hw, unsigned long port, uint32_t value)
4926 {
4927         outl(value, port);
4928 }
4929
4930 static void
4931 e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
4932 {
4933         struct e1000_adapter *adapter = netdev_priv(netdev);
4934         uint32_t ctrl, rctl;
4935
4936         e1000_irq_disable(adapter);
4937         adapter->vlgrp = grp;
4938
4939         if (grp) {
4940                 /* enable VLAN tag insert/strip */
4941                 ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4942                 ctrl |= E1000_CTRL_VME;
4943                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4944
4945                 if (adapter->hw.mac_type != e1000_ich8lan) {
4946                         /* enable VLAN receive filtering */
4947                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
4948                         rctl |= E1000_RCTL_VFE;
4949                         rctl &= ~E1000_RCTL_CFIEN;
4950                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4951                         e1000_update_mng_vlan(adapter);
4952                 }
4953         } else {
4954                 /* disable VLAN tag insert/strip */
4955                 ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4956                 ctrl &= ~E1000_CTRL_VME;
4957                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4958
4959                 if (adapter->hw.mac_type != e1000_ich8lan) {
4960                         /* disable VLAN filtering */
4961                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
4962                         rctl &= ~E1000_RCTL_VFE;
4963                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4964                         if (adapter->mng_vlan_id !=
4965                             (uint16_t)E1000_MNG_VLAN_NONE) {
4966                                 e1000_vlan_rx_kill_vid(netdev,
4967                                                        adapter->mng_vlan_id);
4968                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4969                         }
4970                 }
4971         }
4972
4973         e1000_irq_enable(adapter);
4974 }
4975
4976 static void
4977 e1000_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid)
4978 {
4979         struct e1000_adapter *adapter = netdev_priv(netdev);
4980         uint32_t vfta, index;
4981
4982         if ((adapter->hw.mng_cookie.status &
4983              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4984             (vid == adapter->mng_vlan_id))
4985                 return;
4986         /* add VID to filter table */
4987         index = (vid >> 5) & 0x7F;
4988         vfta = E1000_READ_REG_ARRAY(&adapter->hw, VFTA, index);
4989         vfta |= (1 << (vid & 0x1F));
4990         e1000_write_vfta(&adapter->hw, index, vfta);
4991 }
4992
4993 static void
4994 e1000_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid)
4995 {
4996         struct e1000_adapter *adapter = netdev_priv(netdev);
4997         uint32_t vfta, index;
4998
4999         e1000_irq_disable(adapter);
5000         vlan_group_set_device(adapter->vlgrp, vid, NULL);
5001         e1000_irq_enable(adapter);
5002
5003         if ((adapter->hw.mng_cookie.status &
5004              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
5005             (vid == adapter->mng_vlan_id)) {
5006                 /* release control to f/w */
5007                 e1000_release_hw_control(adapter);
5008                 return;
5009         }
5010
5011         /* remove VID from filter table */
5012         index = (vid >> 5) & 0x7F;
5013         vfta = E1000_READ_REG_ARRAY(&adapter->hw, VFTA, index);
5014         vfta &= ~(1 << (vid & 0x1F));
5015         e1000_write_vfta(&adapter->hw, index, vfta);
5016 }
5017
5018 static void
5019 e1000_restore_vlan(struct e1000_adapter *adapter)
5020 {
5021         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
5022
5023         if (adapter->vlgrp) {
5024                 uint16_t vid;
5025                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
5026                         if (!vlan_group_get_device(adapter->vlgrp, vid))
5027                                 continue;
5028                         e1000_vlan_rx_add_vid(adapter->netdev, vid);
5029                 }
5030         }
5031 }
5032
5033 int
5034 e1000_set_spd_dplx(struct e1000_adapter *adapter, uint16_t spddplx)
5035 {
5036         adapter->hw.autoneg = 0;
5037
5038         /* Fiber NICs only allow 1000 gbps Full duplex */
5039         if ((adapter->hw.media_type == e1000_media_type_fiber) &&
5040                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
5041                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
5042                 return -EINVAL;
5043         }
5044
5045         switch (spddplx) {
5046         case SPEED_10 + DUPLEX_HALF:
5047                 adapter->hw.forced_speed_duplex = e1000_10_half;
5048                 break;
5049         case SPEED_10 + DUPLEX_FULL:
5050                 adapter->hw.forced_speed_duplex = e1000_10_full;
5051                 break;
5052         case SPEED_100 + DUPLEX_HALF:
5053                 adapter->hw.forced_speed_duplex = e1000_100_half;
5054                 break;
5055         case SPEED_100 + DUPLEX_FULL:
5056                 adapter->hw.forced_speed_duplex = e1000_100_full;
5057                 break;
5058         case SPEED_1000 + DUPLEX_FULL:
5059                 adapter->hw.autoneg = 1;
5060                 adapter->hw.autoneg_advertised = ADVERTISE_1000_FULL;
5061                 break;
5062         case SPEED_1000 + DUPLEX_HALF: /* not supported */
5063         default:
5064                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
5065                 return -EINVAL;
5066         }
5067         return 0;
5068 }
5069
5070 static int
5071 e1000_suspend(struct pci_dev *pdev, pm_message_t state)
5072 {
5073         struct net_device *netdev = pci_get_drvdata(pdev);
5074         struct e1000_adapter *adapter = netdev_priv(netdev);
5075         uint32_t ctrl, ctrl_ext, rctl, status;
5076         uint32_t wufc = adapter->wol;
5077 #ifdef CONFIG_PM
5078         int retval = 0;
5079 #endif
5080
5081         netif_device_detach(netdev);
5082
5083         if (netif_running(netdev)) {
5084                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
5085                 e1000_down(adapter);
5086         }
5087
5088 #ifdef CONFIG_PM
5089         retval = pci_save_state(pdev);
5090         if (retval)
5091                 return retval;
5092 #endif
5093
5094         status = E1000_READ_REG(&adapter->hw, STATUS);
5095         if (status & E1000_STATUS_LU)
5096                 wufc &= ~E1000_WUFC_LNKC;
5097
5098         if (wufc) {
5099                 e1000_setup_rctl(adapter);
5100                 e1000_set_multi(netdev);
5101
5102                 /* turn on all-multi mode if wake on multicast is enabled */
5103                 if (wufc & E1000_WUFC_MC) {
5104                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
5105                         rctl |= E1000_RCTL_MPE;
5106                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
5107                 }
5108
5109                 if (adapter->hw.mac_type >= e1000_82540) {
5110                         ctrl = E1000_READ_REG(&adapter->hw, CTRL);
5111                         /* advertise wake from D3Cold */
5112                         #define E1000_CTRL_ADVD3WUC 0x00100000
5113                         /* phy power management enable */
5114                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5115                         ctrl |= E1000_CTRL_ADVD3WUC |
5116                                 E1000_CTRL_EN_PHY_PWR_MGMT;
5117                         E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
5118                 }
5119
5120                 if (adapter->hw.media_type == e1000_media_type_fiber ||
5121                    adapter->hw.media_type == e1000_media_type_internal_serdes) {
5122                         /* keep the laser running in D3 */
5123                         ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
5124                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
5125                         E1000_WRITE_REG(&adapter->hw, CTRL_EXT, ctrl_ext);
5126                 }
5127
5128                 /* Allow time for pending master requests to run */
5129                 e1000_disable_pciex_master(&adapter->hw);
5130
5131                 E1000_WRITE_REG(&adapter->hw, WUC, E1000_WUC_PME_EN);
5132                 E1000_WRITE_REG(&adapter->hw, WUFC, wufc);
5133                 pci_enable_wake(pdev, PCI_D3hot, 1);
5134                 pci_enable_wake(pdev, PCI_D3cold, 1);
5135         } else {
5136                 E1000_WRITE_REG(&adapter->hw, WUC, 0);
5137                 E1000_WRITE_REG(&adapter->hw, WUFC, 0);
5138                 pci_enable_wake(pdev, PCI_D3hot, 0);
5139                 pci_enable_wake(pdev, PCI_D3cold, 0);
5140         }
5141
5142         e1000_release_manageability(adapter);
5143
5144         /* make sure adapter isn't asleep if manageability is enabled */
5145         if (adapter->en_mng_pt) {
5146                 pci_enable_wake(pdev, PCI_D3hot, 1);
5147                 pci_enable_wake(pdev, PCI_D3cold, 1);
5148         }
5149
5150         if (adapter->hw.phy_type == e1000_phy_igp_3)
5151                 e1000_phy_powerdown_workaround(&adapter->hw);
5152
5153         if (netif_running(netdev))
5154                 e1000_free_irq(adapter);
5155
5156         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
5157          * would have already happened in close and is redundant. */
5158         e1000_release_hw_control(adapter);
5159
5160         pci_disable_device(pdev);
5161
5162         pci_set_power_state(pdev, pci_choose_state(pdev, state));
5163
5164         return 0;
5165 }
5166
5167 #ifdef CONFIG_PM
5168 static int
5169 e1000_resume(struct pci_dev *pdev)
5170 {
5171         struct net_device *netdev = pci_get_drvdata(pdev);
5172         struct e1000_adapter *adapter = netdev_priv(netdev);
5173         uint32_t err;
5174
5175         pci_set_power_state(pdev, PCI_D0);
5176         pci_restore_state(pdev);
5177         if ((err = pci_enable_device(pdev))) {
5178                 printk(KERN_ERR "e1000: Cannot enable PCI device from suspend\n");
5179                 return err;
5180         }
5181         pci_set_master(pdev);
5182
5183         pci_enable_wake(pdev, PCI_D3hot, 0);
5184         pci_enable_wake(pdev, PCI_D3cold, 0);
5185
5186         if (netif_running(netdev) && (err = e1000_request_irq(adapter)))
5187                 return err;
5188
5189         e1000_power_up_phy(adapter);
5190         e1000_reset(adapter);
5191         E1000_WRITE_REG(&adapter->hw, WUS, ~0);
5192
5193         e1000_init_manageability(adapter);
5194
5195         if (netif_running(netdev))
5196                 e1000_up(adapter);
5197
5198         netif_device_attach(netdev);
5199
5200         /* If the controller is 82573 and f/w is AMT, do not set
5201          * DRV_LOAD until the interface is up.  For all other cases,
5202          * let the f/w know that the h/w is now under the control
5203          * of the driver. */
5204         if (adapter->hw.mac_type != e1000_82573 ||
5205             !e1000_check_mng_mode(&adapter->hw))
5206                 e1000_get_hw_control(adapter);
5207
5208         return 0;
5209 }
5210 #endif
5211
5212 static void e1000_shutdown(struct pci_dev *pdev)
5213 {
5214         e1000_suspend(pdev, PMSG_SUSPEND);
5215 }
5216
5217 #ifdef CONFIG_NET_POLL_CONTROLLER
5218 /*
5219  * Polling 'interrupt' - used by things like netconsole to send skbs
5220  * without having to re-enable interrupts. It's not called while
5221  * the interrupt routine is executing.
5222  */
5223 static void
5224 e1000_netpoll(struct net_device *netdev)
5225 {
5226         struct e1000_adapter *adapter = netdev_priv(netdev);
5227
5228         disable_irq(adapter->pdev->irq);
5229         e1000_intr(adapter->pdev->irq, netdev);
5230         e1000_clean_tx_irq(adapter, adapter->tx_ring);
5231 #ifndef CONFIG_E1000_NAPI
5232         adapter->clean_rx(adapter, adapter->rx_ring);
5233 #endif
5234         enable_irq(adapter->pdev->irq);
5235 }
5236 #endif
5237
5238 /**
5239  * e1000_io_error_detected - called when PCI error is detected
5240  * @pdev: Pointer to PCI device
5241  * @state: The current pci conneection state
5242  *
5243  * This function is called after a PCI bus error affecting
5244  * this device has been detected.
5245  */
5246 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
5247 {
5248         struct net_device *netdev = pci_get_drvdata(pdev);
5249         struct e1000_adapter *adapter = netdev->priv;
5250
5251         netif_device_detach(netdev);
5252
5253         if (netif_running(netdev))
5254                 e1000_down(adapter);
5255         pci_disable_device(pdev);
5256
5257         /* Request a slot slot reset. */
5258         return PCI_ERS_RESULT_NEED_RESET;
5259 }
5260
5261 /**
5262  * e1000_io_slot_reset - called after the pci bus has been reset.
5263  * @pdev: Pointer to PCI device
5264  *
5265  * Restart the card from scratch, as if from a cold-boot. Implementation
5266  * resembles the first-half of the e1000_resume routine.
5267  */
5268 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5269 {
5270         struct net_device *netdev = pci_get_drvdata(pdev);
5271         struct e1000_adapter *adapter = netdev->priv;
5272
5273         if (pci_enable_device(pdev)) {
5274                 printk(KERN_ERR "e1000: Cannot re-enable PCI device after reset.\n");
5275                 return PCI_ERS_RESULT_DISCONNECT;
5276         }
5277         pci_set_master(pdev);
5278
5279         pci_enable_wake(pdev, PCI_D3hot, 0);
5280         pci_enable_wake(pdev, PCI_D3cold, 0);
5281
5282         e1000_reset(adapter);
5283         E1000_WRITE_REG(&adapter->hw, WUS, ~0);
5284
5285         return PCI_ERS_RESULT_RECOVERED;
5286 }
5287
5288 /**
5289  * e1000_io_resume - called when traffic can start flowing again.
5290  * @pdev: Pointer to PCI device
5291  *
5292  * This callback is called when the error recovery driver tells us that
5293  * its OK to resume normal operation. Implementation resembles the
5294  * second-half of the e1000_resume routine.
5295  */
5296 static void e1000_io_resume(struct pci_dev *pdev)
5297 {
5298         struct net_device *netdev = pci_get_drvdata(pdev);
5299         struct e1000_adapter *adapter = netdev->priv;
5300
5301         e1000_init_manageability(adapter);
5302
5303         if (netif_running(netdev)) {
5304                 if (e1000_up(adapter)) {
5305                         printk("e1000: can't bring device back up after reset\n");
5306                         return;
5307                 }
5308         }
5309
5310         netif_device_attach(netdev);
5311
5312         /* If the controller is 82573 and f/w is AMT, do not set
5313          * DRV_LOAD until the interface is up.  For all other cases,
5314          * let the f/w know that the h/w is now under the control
5315          * of the driver. */
5316         if (adapter->hw.mac_type != e1000_82573 ||
5317             !e1000_check_mng_mode(&adapter->hw))
5318                 e1000_get_hw_control(adapter);
5319
5320 }
5321
5322 /* e1000_main.c */