Merge branch 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
[pandora-kernel.git] / drivers / net / e1000 / e1000_main.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2006 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include "e1000.h"
30 #include <net/ip6_checksum.h>
31
32 char e1000_driver_name[] = "e1000";
33 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
34 #ifndef CONFIG_E1000_NAPI
35 #define DRIVERNAPI
36 #else
37 #define DRIVERNAPI "-NAPI"
38 #endif
39 #define DRV_VERSION "7.3.20-k2"DRIVERNAPI
40 char e1000_driver_version[] = DRV_VERSION;
41 static char e1000_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
42
43 /* e1000_pci_tbl - PCI Device ID Table
44  *
45  * Last entry must be all 0s
46  *
47  * Macro expands to...
48  *   {PCI_DEVICE(PCI_VENDOR_ID_INTEL, device_id)}
49  */
50 static struct pci_device_id e1000_pci_tbl[] = {
51         INTEL_E1000_ETHERNET_DEVICE(0x1000),
52         INTEL_E1000_ETHERNET_DEVICE(0x1001),
53         INTEL_E1000_ETHERNET_DEVICE(0x1004),
54         INTEL_E1000_ETHERNET_DEVICE(0x1008),
55         INTEL_E1000_ETHERNET_DEVICE(0x1009),
56         INTEL_E1000_ETHERNET_DEVICE(0x100C),
57         INTEL_E1000_ETHERNET_DEVICE(0x100D),
58         INTEL_E1000_ETHERNET_DEVICE(0x100E),
59         INTEL_E1000_ETHERNET_DEVICE(0x100F),
60         INTEL_E1000_ETHERNET_DEVICE(0x1010),
61         INTEL_E1000_ETHERNET_DEVICE(0x1011),
62         INTEL_E1000_ETHERNET_DEVICE(0x1012),
63         INTEL_E1000_ETHERNET_DEVICE(0x1013),
64         INTEL_E1000_ETHERNET_DEVICE(0x1014),
65         INTEL_E1000_ETHERNET_DEVICE(0x1015),
66         INTEL_E1000_ETHERNET_DEVICE(0x1016),
67         INTEL_E1000_ETHERNET_DEVICE(0x1017),
68         INTEL_E1000_ETHERNET_DEVICE(0x1018),
69         INTEL_E1000_ETHERNET_DEVICE(0x1019),
70         INTEL_E1000_ETHERNET_DEVICE(0x101A),
71         INTEL_E1000_ETHERNET_DEVICE(0x101D),
72         INTEL_E1000_ETHERNET_DEVICE(0x101E),
73         INTEL_E1000_ETHERNET_DEVICE(0x1026),
74         INTEL_E1000_ETHERNET_DEVICE(0x1027),
75         INTEL_E1000_ETHERNET_DEVICE(0x1028),
76         INTEL_E1000_ETHERNET_DEVICE(0x1049),
77         INTEL_E1000_ETHERNET_DEVICE(0x104A),
78         INTEL_E1000_ETHERNET_DEVICE(0x104B),
79         INTEL_E1000_ETHERNET_DEVICE(0x104C),
80         INTEL_E1000_ETHERNET_DEVICE(0x104D),
81         INTEL_E1000_ETHERNET_DEVICE(0x105E),
82         INTEL_E1000_ETHERNET_DEVICE(0x105F),
83         INTEL_E1000_ETHERNET_DEVICE(0x1060),
84         INTEL_E1000_ETHERNET_DEVICE(0x1075),
85         INTEL_E1000_ETHERNET_DEVICE(0x1076),
86         INTEL_E1000_ETHERNET_DEVICE(0x1077),
87         INTEL_E1000_ETHERNET_DEVICE(0x1078),
88         INTEL_E1000_ETHERNET_DEVICE(0x1079),
89         INTEL_E1000_ETHERNET_DEVICE(0x107A),
90         INTEL_E1000_ETHERNET_DEVICE(0x107B),
91         INTEL_E1000_ETHERNET_DEVICE(0x107C),
92         INTEL_E1000_ETHERNET_DEVICE(0x107D),
93         INTEL_E1000_ETHERNET_DEVICE(0x107E),
94         INTEL_E1000_ETHERNET_DEVICE(0x107F),
95         INTEL_E1000_ETHERNET_DEVICE(0x108A),
96         INTEL_E1000_ETHERNET_DEVICE(0x108B),
97         INTEL_E1000_ETHERNET_DEVICE(0x108C),
98         INTEL_E1000_ETHERNET_DEVICE(0x1096),
99         INTEL_E1000_ETHERNET_DEVICE(0x1098),
100         INTEL_E1000_ETHERNET_DEVICE(0x1099),
101         INTEL_E1000_ETHERNET_DEVICE(0x109A),
102         INTEL_E1000_ETHERNET_DEVICE(0x10A4),
103         INTEL_E1000_ETHERNET_DEVICE(0x10B5),
104         INTEL_E1000_ETHERNET_DEVICE(0x10B9),
105         INTEL_E1000_ETHERNET_DEVICE(0x10BA),
106         INTEL_E1000_ETHERNET_DEVICE(0x10BB),
107         INTEL_E1000_ETHERNET_DEVICE(0x10BC),
108         INTEL_E1000_ETHERNET_DEVICE(0x10C4),
109         INTEL_E1000_ETHERNET_DEVICE(0x10C5),
110         /* required last entry */
111         {0,}
112 };
113
114 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
115
116 int e1000_up(struct e1000_adapter *adapter);
117 void e1000_down(struct e1000_adapter *adapter);
118 void e1000_reinit_locked(struct e1000_adapter *adapter);
119 void e1000_reset(struct e1000_adapter *adapter);
120 int e1000_set_spd_dplx(struct e1000_adapter *adapter, uint16_t spddplx);
121 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
122 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
123 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
124 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
125 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
126                              struct e1000_tx_ring *txdr);
127 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
128                              struct e1000_rx_ring *rxdr);
129 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
130                              struct e1000_tx_ring *tx_ring);
131 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
132                              struct e1000_rx_ring *rx_ring);
133 void e1000_update_stats(struct e1000_adapter *adapter);
134
135 static int e1000_init_module(void);
136 static void e1000_exit_module(void);
137 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
138 static void __devexit e1000_remove(struct pci_dev *pdev);
139 static int e1000_alloc_queues(struct e1000_adapter *adapter);
140 static int e1000_sw_init(struct e1000_adapter *adapter);
141 static int e1000_open(struct net_device *netdev);
142 static int e1000_close(struct net_device *netdev);
143 static void e1000_configure_tx(struct e1000_adapter *adapter);
144 static void e1000_configure_rx(struct e1000_adapter *adapter);
145 static void e1000_setup_rctl(struct e1000_adapter *adapter);
146 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
147 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
148 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
149                                 struct e1000_tx_ring *tx_ring);
150 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
151                                 struct e1000_rx_ring *rx_ring);
152 static void e1000_set_multi(struct net_device *netdev);
153 static void e1000_update_phy_info(unsigned long data);
154 static void e1000_watchdog(unsigned long data);
155 static void e1000_82547_tx_fifo_stall(unsigned long data);
156 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
157 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
158 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
159 static int e1000_set_mac(struct net_device *netdev, void *p);
160 static irqreturn_t e1000_intr(int irq, void *data);
161 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         strcpy(netdev->name, "eth%d");
1146         if ((err = register_netdev(netdev)))
1147                 goto err_register;
1148
1149         /* tell the stack to leave us alone until e1000_open() is called */
1150         netif_carrier_off(netdev);
1151         netif_stop_queue(netdev);
1152
1153         DPRINTK(PROBE, INFO, "Intel(R) PRO/1000 Network Connection\n");
1154
1155         cards_found++;
1156         return 0;
1157
1158 err_register:
1159         e1000_release_hw_control(adapter);
1160 err_eeprom:
1161         if (!e1000_check_phy_reset_block(&adapter->hw))
1162                 e1000_phy_hw_reset(&adapter->hw);
1163
1164         if (adapter->hw.flash_address)
1165                 iounmap(adapter->hw.flash_address);
1166 err_flashmap:
1167 #ifdef CONFIG_E1000_NAPI
1168         for (i = 0; i < adapter->num_rx_queues; i++)
1169                 dev_put(&adapter->polling_netdev[i]);
1170 #endif
1171
1172         kfree(adapter->tx_ring);
1173         kfree(adapter->rx_ring);
1174 #ifdef CONFIG_E1000_NAPI
1175         kfree(adapter->polling_netdev);
1176 #endif
1177 err_sw_init:
1178         iounmap(adapter->hw.hw_addr);
1179 err_ioremap:
1180         free_netdev(netdev);
1181 err_alloc_etherdev:
1182         pci_release_regions(pdev);
1183 err_pci_reg:
1184 err_dma:
1185         pci_disable_device(pdev);
1186         return err;
1187 }
1188
1189 /**
1190  * e1000_remove - Device Removal Routine
1191  * @pdev: PCI device information struct
1192  *
1193  * e1000_remove is called by the PCI subsystem to alert the driver
1194  * that it should release a PCI device.  The could be caused by a
1195  * Hot-Plug event, or because the driver is going to be removed from
1196  * memory.
1197  **/
1198
1199 static void __devexit
1200 e1000_remove(struct pci_dev *pdev)
1201 {
1202         struct net_device *netdev = pci_get_drvdata(pdev);
1203         struct e1000_adapter *adapter = netdev_priv(netdev);
1204 #ifdef CONFIG_E1000_NAPI
1205         int i;
1206 #endif
1207
1208         cancel_work_sync(&adapter->reset_task);
1209
1210         e1000_release_manageability(adapter);
1211
1212         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
1213          * would have already happened in close and is redundant. */
1214         e1000_release_hw_control(adapter);
1215
1216         unregister_netdev(netdev);
1217 #ifdef CONFIG_E1000_NAPI
1218         for (i = 0; i < adapter->num_rx_queues; i++)
1219                 dev_put(&adapter->polling_netdev[i]);
1220 #endif
1221
1222         if (!e1000_check_phy_reset_block(&adapter->hw))
1223                 e1000_phy_hw_reset(&adapter->hw);
1224
1225         kfree(adapter->tx_ring);
1226         kfree(adapter->rx_ring);
1227 #ifdef CONFIG_E1000_NAPI
1228         kfree(adapter->polling_netdev);
1229 #endif
1230
1231         iounmap(adapter->hw.hw_addr);
1232         if (adapter->hw.flash_address)
1233                 iounmap(adapter->hw.flash_address);
1234         pci_release_regions(pdev);
1235
1236         free_netdev(netdev);
1237
1238         pci_disable_device(pdev);
1239 }
1240
1241 /**
1242  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1243  * @adapter: board private structure to initialize
1244  *
1245  * e1000_sw_init initializes the Adapter private data structure.
1246  * Fields are initialized based on PCI device information and
1247  * OS network device settings (MTU size).
1248  **/
1249
1250 static int __devinit
1251 e1000_sw_init(struct e1000_adapter *adapter)
1252 {
1253         struct e1000_hw *hw = &adapter->hw;
1254         struct net_device *netdev = adapter->netdev;
1255         struct pci_dev *pdev = adapter->pdev;
1256 #ifdef CONFIG_E1000_NAPI
1257         int i;
1258 #endif
1259
1260         /* PCI config space info */
1261
1262         hw->vendor_id = pdev->vendor;
1263         hw->device_id = pdev->device;
1264         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1265         hw->subsystem_id = pdev->subsystem_device;
1266
1267         pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
1268
1269         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
1270
1271         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1272         adapter->rx_ps_bsize0 = E1000_RXBUFFER_128;
1273         hw->max_frame_size = netdev->mtu +
1274                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
1275         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
1276
1277         /* identify the MAC */
1278
1279         if (e1000_set_mac_type(hw)) {
1280                 DPRINTK(PROBE, ERR, "Unknown MAC Type\n");
1281                 return -EIO;
1282         }
1283
1284         switch (hw->mac_type) {
1285         default:
1286                 break;
1287         case e1000_82541:
1288         case e1000_82547:
1289         case e1000_82541_rev_2:
1290         case e1000_82547_rev_2:
1291                 hw->phy_init_script = 1;
1292                 break;
1293         }
1294
1295         e1000_set_media_type(hw);
1296
1297         hw->wait_autoneg_complete = FALSE;
1298         hw->tbi_compatibility_en = TRUE;
1299         hw->adaptive_ifs = TRUE;
1300
1301         /* Copper options */
1302
1303         if (hw->media_type == e1000_media_type_copper) {
1304                 hw->mdix = AUTO_ALL_MODES;
1305                 hw->disable_polarity_correction = FALSE;
1306                 hw->master_slave = E1000_MASTER_SLAVE;
1307         }
1308
1309         adapter->num_tx_queues = 1;
1310         adapter->num_rx_queues = 1;
1311
1312         if (e1000_alloc_queues(adapter)) {
1313                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
1314                 return -ENOMEM;
1315         }
1316
1317 #ifdef CONFIG_E1000_NAPI
1318         for (i = 0; i < adapter->num_rx_queues; i++) {
1319                 adapter->polling_netdev[i].priv = adapter;
1320                 adapter->polling_netdev[i].poll = &e1000_clean;
1321                 adapter->polling_netdev[i].weight = 64;
1322                 dev_hold(&adapter->polling_netdev[i]);
1323                 set_bit(__LINK_STATE_START, &adapter->polling_netdev[i].state);
1324         }
1325         spin_lock_init(&adapter->tx_queue_lock);
1326 #endif
1327
1328         /* Explicitly disable IRQ since the NIC can be in any state. */
1329         atomic_set(&adapter->irq_sem, 0);
1330         e1000_irq_disable(adapter);
1331
1332         spin_lock_init(&adapter->stats_lock);
1333
1334         set_bit(__E1000_DOWN, &adapter->flags);
1335
1336         return 0;
1337 }
1338
1339 /**
1340  * e1000_alloc_queues - Allocate memory for all rings
1341  * @adapter: board private structure to initialize
1342  *
1343  * We allocate one ring per queue at run-time since we don't know the
1344  * number of queues at compile-time.  The polling_netdev array is
1345  * intended for Multiqueue, but should work fine with a single queue.
1346  **/
1347
1348 static int __devinit
1349 e1000_alloc_queues(struct e1000_adapter *adapter)
1350 {
1351         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1352                                    sizeof(struct e1000_tx_ring), GFP_KERNEL);
1353         if (!adapter->tx_ring)
1354                 return -ENOMEM;
1355
1356         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1357                                    sizeof(struct e1000_rx_ring), GFP_KERNEL);
1358         if (!adapter->rx_ring) {
1359                 kfree(adapter->tx_ring);
1360                 return -ENOMEM;
1361         }
1362
1363 #ifdef CONFIG_E1000_NAPI
1364         adapter->polling_netdev = kcalloc(adapter->num_rx_queues,
1365                                           sizeof(struct net_device),
1366                                           GFP_KERNEL);
1367         if (!adapter->polling_netdev) {
1368                 kfree(adapter->tx_ring);
1369                 kfree(adapter->rx_ring);
1370                 return -ENOMEM;
1371         }
1372 #endif
1373
1374         return E1000_SUCCESS;
1375 }
1376
1377 /**
1378  * e1000_open - Called when a network interface is made active
1379  * @netdev: network interface device structure
1380  *
1381  * Returns 0 on success, negative value on failure
1382  *
1383  * The open entry point is called when a network interface is made
1384  * active by the system (IFF_UP).  At this point all resources needed
1385  * for transmit and receive operations are allocated, the interrupt
1386  * handler is registered with the OS, the watchdog timer is started,
1387  * and the stack is notified that the interface is ready.
1388  **/
1389
1390 static int
1391 e1000_open(struct net_device *netdev)
1392 {
1393         struct e1000_adapter *adapter = netdev_priv(netdev);
1394         int err;
1395
1396         /* disallow open during test */
1397         if (test_bit(__E1000_TESTING, &adapter->flags))
1398                 return -EBUSY;
1399
1400         /* allocate transmit descriptors */
1401         err = e1000_setup_all_tx_resources(adapter);
1402         if (err)
1403                 goto err_setup_tx;
1404
1405         /* allocate receive descriptors */
1406         err = e1000_setup_all_rx_resources(adapter);
1407         if (err)
1408                 goto err_setup_rx;
1409
1410         e1000_power_up_phy(adapter);
1411
1412         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1413         if ((adapter->hw.mng_cookie.status &
1414                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1415                 e1000_update_mng_vlan(adapter);
1416         }
1417
1418         /* If AMT is enabled, let the firmware know that the network
1419          * interface is now open */
1420         if (adapter->hw.mac_type == e1000_82573 &&
1421             e1000_check_mng_mode(&adapter->hw))
1422                 e1000_get_hw_control(adapter);
1423
1424         /* before we allocate an interrupt, we must be ready to handle it.
1425          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1426          * as soon as we call pci_request_irq, so we have to setup our
1427          * clean_rx handler before we do so.  */
1428         e1000_configure(adapter);
1429
1430         err = e1000_request_irq(adapter);
1431         if (err)
1432                 goto err_req_irq;
1433
1434         /* From here on the code is the same as e1000_up() */
1435         clear_bit(__E1000_DOWN, &adapter->flags);
1436
1437 #ifdef CONFIG_E1000_NAPI
1438         netif_poll_enable(netdev);
1439 #endif
1440
1441         e1000_irq_enable(adapter);
1442
1443         /* fire a link status change interrupt to start the watchdog */
1444         E1000_WRITE_REG(&adapter->hw, ICS, E1000_ICS_LSC);
1445
1446         return E1000_SUCCESS;
1447
1448 err_req_irq:
1449         e1000_release_hw_control(adapter);
1450         e1000_power_down_phy(adapter);
1451         e1000_free_all_rx_resources(adapter);
1452 err_setup_rx:
1453         e1000_free_all_tx_resources(adapter);
1454 err_setup_tx:
1455         e1000_reset(adapter);
1456
1457         return err;
1458 }
1459
1460 /**
1461  * e1000_close - Disables a network interface
1462  * @netdev: network interface device structure
1463  *
1464  * Returns 0, this is not allowed to fail
1465  *
1466  * The close entry point is called when an interface is de-activated
1467  * by the OS.  The hardware is still under the drivers control, but
1468  * needs to be disabled.  A global MAC reset is issued to stop the
1469  * hardware, and all transmit and receive resources are freed.
1470  **/
1471
1472 static int
1473 e1000_close(struct net_device *netdev)
1474 {
1475         struct e1000_adapter *adapter = netdev_priv(netdev);
1476
1477         WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1478         e1000_down(adapter);
1479         e1000_power_down_phy(adapter);
1480         e1000_free_irq(adapter);
1481
1482         e1000_free_all_tx_resources(adapter);
1483         e1000_free_all_rx_resources(adapter);
1484
1485         /* kill manageability vlan ID if supported, but not if a vlan with
1486          * the same ID is registered on the host OS (let 8021q kill it) */
1487         if ((adapter->hw.mng_cookie.status &
1488                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1489              !(adapter->vlgrp &&
1490                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id))) {
1491                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1492         }
1493
1494         /* If AMT is enabled, let the firmware know that the network
1495          * interface is now closed */
1496         if (adapter->hw.mac_type == e1000_82573 &&
1497             e1000_check_mng_mode(&adapter->hw))
1498                 e1000_release_hw_control(adapter);
1499
1500         return 0;
1501 }
1502
1503 /**
1504  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1505  * @adapter: address of board private structure
1506  * @start: address of beginning of memory
1507  * @len: length of memory
1508  **/
1509 static boolean_t
1510 e1000_check_64k_bound(struct e1000_adapter *adapter,
1511                       void *start, unsigned long len)
1512 {
1513         unsigned long begin = (unsigned long) start;
1514         unsigned long end = begin + len;
1515
1516         /* First rev 82545 and 82546 need to not allow any memory
1517          * write location to cross 64k boundary due to errata 23 */
1518         if (adapter->hw.mac_type == e1000_82545 ||
1519             adapter->hw.mac_type == e1000_82546) {
1520                 return ((begin ^ (end - 1)) >> 16) != 0 ? FALSE : TRUE;
1521         }
1522
1523         return TRUE;
1524 }
1525
1526 /**
1527  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1528  * @adapter: board private structure
1529  * @txdr:    tx descriptor ring (for a specific queue) to setup
1530  *
1531  * Return 0 on success, negative on failure
1532  **/
1533
1534 static int
1535 e1000_setup_tx_resources(struct e1000_adapter *adapter,
1536                          struct e1000_tx_ring *txdr)
1537 {
1538         struct pci_dev *pdev = adapter->pdev;
1539         int size;
1540
1541         size = sizeof(struct e1000_buffer) * txdr->count;
1542         txdr->buffer_info = vmalloc(size);
1543         if (!txdr->buffer_info) {
1544                 DPRINTK(PROBE, ERR,
1545                 "Unable to allocate memory for the transmit descriptor ring\n");
1546                 return -ENOMEM;
1547         }
1548         memset(txdr->buffer_info, 0, size);
1549
1550         /* round up to nearest 4K */
1551
1552         txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1553         txdr->size = ALIGN(txdr->size, 4096);
1554
1555         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1556         if (!txdr->desc) {
1557 setup_tx_desc_die:
1558                 vfree(txdr->buffer_info);
1559                 DPRINTK(PROBE, ERR,
1560                 "Unable to allocate memory for the transmit descriptor ring\n");
1561                 return -ENOMEM;
1562         }
1563
1564         /* Fix for errata 23, can't cross 64kB boundary */
1565         if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1566                 void *olddesc = txdr->desc;
1567                 dma_addr_t olddma = txdr->dma;
1568                 DPRINTK(TX_ERR, ERR, "txdr align check failed: %u bytes "
1569                                      "at %p\n", txdr->size, txdr->desc);
1570                 /* Try again, without freeing the previous */
1571                 txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1572                 /* Failed allocation, critical failure */
1573                 if (!txdr->desc) {
1574                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1575                         goto setup_tx_desc_die;
1576                 }
1577
1578                 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1579                         /* give up */
1580                         pci_free_consistent(pdev, txdr->size, txdr->desc,
1581                                             txdr->dma);
1582                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1583                         DPRINTK(PROBE, ERR,
1584                                 "Unable to allocate aligned memory "
1585                                 "for the transmit descriptor ring\n");
1586                         vfree(txdr->buffer_info);
1587                         return -ENOMEM;
1588                 } else {
1589                         /* Free old allocation, new allocation was successful */
1590                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1591                 }
1592         }
1593         memset(txdr->desc, 0, txdr->size);
1594
1595         txdr->next_to_use = 0;
1596         txdr->next_to_clean = 0;
1597         spin_lock_init(&txdr->tx_lock);
1598
1599         return 0;
1600 }
1601
1602 /**
1603  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1604  *                                (Descriptors) for all queues
1605  * @adapter: board private structure
1606  *
1607  * Return 0 on success, negative on failure
1608  **/
1609
1610 int
1611 e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1612 {
1613         int i, err = 0;
1614
1615         for (i = 0; i < adapter->num_tx_queues; i++) {
1616                 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1617                 if (err) {
1618                         DPRINTK(PROBE, ERR,
1619                                 "Allocation for Tx Queue %u failed\n", i);
1620                         for (i-- ; i >= 0; i--)
1621                                 e1000_free_tx_resources(adapter,
1622                                                         &adapter->tx_ring[i]);
1623                         break;
1624                 }
1625         }
1626
1627         return err;
1628 }
1629
1630 /**
1631  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1632  * @adapter: board private structure
1633  *
1634  * Configure the Tx unit of the MAC after a reset.
1635  **/
1636
1637 static void
1638 e1000_configure_tx(struct e1000_adapter *adapter)
1639 {
1640         uint64_t tdba;
1641         struct e1000_hw *hw = &adapter->hw;
1642         uint32_t tdlen, tctl, tipg, tarc;
1643         uint32_t ipgr1, ipgr2;
1644
1645         /* Setup the HW Tx Head and Tail descriptor pointers */
1646
1647         switch (adapter->num_tx_queues) {
1648         case 1:
1649         default:
1650                 tdba = adapter->tx_ring[0].dma;
1651                 tdlen = adapter->tx_ring[0].count *
1652                         sizeof(struct e1000_tx_desc);
1653                 E1000_WRITE_REG(hw, TDLEN, tdlen);
1654                 E1000_WRITE_REG(hw, TDBAH, (tdba >> 32));
1655                 E1000_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
1656                 E1000_WRITE_REG(hw, TDT, 0);
1657                 E1000_WRITE_REG(hw, TDH, 0);
1658                 adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ? E1000_TDH : E1000_82542_TDH);
1659                 adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ? E1000_TDT : E1000_82542_TDT);
1660                 break;
1661         }
1662
1663         /* Set the default values for the Tx Inter Packet Gap timer */
1664         if (adapter->hw.mac_type <= e1000_82547_rev_2 &&
1665             (hw->media_type == e1000_media_type_fiber ||
1666              hw->media_type == e1000_media_type_internal_serdes))
1667                 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1668         else
1669                 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1670
1671         switch (hw->mac_type) {
1672         case e1000_82542_rev2_0:
1673         case e1000_82542_rev2_1:
1674                 tipg = DEFAULT_82542_TIPG_IPGT;
1675                 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1676                 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1677                 break;
1678         case e1000_80003es2lan:
1679                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1680                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2;
1681                 break;
1682         default:
1683                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1684                 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1685                 break;
1686         }
1687         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1688         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1689         E1000_WRITE_REG(hw, TIPG, tipg);
1690
1691         /* Set the Tx Interrupt Delay register */
1692
1693         E1000_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
1694         if (hw->mac_type >= e1000_82540)
1695                 E1000_WRITE_REG(hw, TADV, adapter->tx_abs_int_delay);
1696
1697         /* Program the Transmit Control Register */
1698
1699         tctl = E1000_READ_REG(hw, TCTL);
1700         tctl &= ~E1000_TCTL_CT;
1701         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1702                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1703
1704         if (hw->mac_type == e1000_82571 || hw->mac_type == e1000_82572) {
1705                 tarc = E1000_READ_REG(hw, TARC0);
1706                 /* set the speed mode bit, we'll clear it if we're not at
1707                  * gigabit link later */
1708                 tarc |= (1 << 21);
1709                 E1000_WRITE_REG(hw, TARC0, tarc);
1710         } else if (hw->mac_type == e1000_80003es2lan) {
1711                 tarc = E1000_READ_REG(hw, TARC0);
1712                 tarc |= 1;
1713                 E1000_WRITE_REG(hw, TARC0, tarc);
1714                 tarc = E1000_READ_REG(hw, TARC1);
1715                 tarc |= 1;
1716                 E1000_WRITE_REG(hw, TARC1, tarc);
1717         }
1718
1719         e1000_config_collision_dist(hw);
1720
1721         /* Setup Transmit Descriptor Settings for eop descriptor */
1722         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1723
1724         /* only set IDE if we are delaying interrupts using the timers */
1725         if (adapter->tx_int_delay)
1726                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1727
1728         if (hw->mac_type < e1000_82543)
1729                 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1730         else
1731                 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1732
1733         /* Cache if we're 82544 running in PCI-X because we'll
1734          * need this to apply a workaround later in the send path. */
1735         if (hw->mac_type == e1000_82544 &&
1736             hw->bus_type == e1000_bus_type_pcix)
1737                 adapter->pcix_82544 = 1;
1738
1739         E1000_WRITE_REG(hw, TCTL, tctl);
1740
1741 }
1742
1743 /**
1744  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1745  * @adapter: board private structure
1746  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1747  *
1748  * Returns 0 on success, negative on failure
1749  **/
1750
1751 static int
1752 e1000_setup_rx_resources(struct e1000_adapter *adapter,
1753                          struct e1000_rx_ring *rxdr)
1754 {
1755         struct pci_dev *pdev = adapter->pdev;
1756         int size, desc_len;
1757
1758         size = sizeof(struct e1000_buffer) * rxdr->count;
1759         rxdr->buffer_info = vmalloc(size);
1760         if (!rxdr->buffer_info) {
1761                 DPRINTK(PROBE, ERR,
1762                 "Unable to allocate memory for the receive descriptor ring\n");
1763                 return -ENOMEM;
1764         }
1765         memset(rxdr->buffer_info, 0, size);
1766
1767         rxdr->ps_page = kcalloc(rxdr->count, sizeof(struct e1000_ps_page),
1768                                 GFP_KERNEL);
1769         if (!rxdr->ps_page) {
1770                 vfree(rxdr->buffer_info);
1771                 DPRINTK(PROBE, ERR,
1772                 "Unable to allocate memory for the receive descriptor ring\n");
1773                 return -ENOMEM;
1774         }
1775
1776         rxdr->ps_page_dma = kcalloc(rxdr->count,
1777                                     sizeof(struct e1000_ps_page_dma),
1778                                     GFP_KERNEL);
1779         if (!rxdr->ps_page_dma) {
1780                 vfree(rxdr->buffer_info);
1781                 kfree(rxdr->ps_page);
1782                 DPRINTK(PROBE, ERR,
1783                 "Unable to allocate memory for the receive descriptor ring\n");
1784                 return -ENOMEM;
1785         }
1786
1787         if (adapter->hw.mac_type <= e1000_82547_rev_2)
1788                 desc_len = sizeof(struct e1000_rx_desc);
1789         else
1790                 desc_len = sizeof(union e1000_rx_desc_packet_split);
1791
1792         /* Round up to nearest 4K */
1793
1794         rxdr->size = rxdr->count * desc_len;
1795         rxdr->size = ALIGN(rxdr->size, 4096);
1796
1797         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1798
1799         if (!rxdr->desc) {
1800                 DPRINTK(PROBE, ERR,
1801                 "Unable to allocate memory for the receive descriptor ring\n");
1802 setup_rx_desc_die:
1803                 vfree(rxdr->buffer_info);
1804                 kfree(rxdr->ps_page);
1805                 kfree(rxdr->ps_page_dma);
1806                 return -ENOMEM;
1807         }
1808
1809         /* Fix for errata 23, can't cross 64kB boundary */
1810         if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1811                 void *olddesc = rxdr->desc;
1812                 dma_addr_t olddma = rxdr->dma;
1813                 DPRINTK(RX_ERR, ERR, "rxdr align check failed: %u bytes "
1814                                      "at %p\n", rxdr->size, rxdr->desc);
1815                 /* Try again, without freeing the previous */
1816                 rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1817                 /* Failed allocation, critical failure */
1818                 if (!rxdr->desc) {
1819                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1820                         DPRINTK(PROBE, ERR,
1821                                 "Unable to allocate memory "
1822                                 "for the receive descriptor ring\n");
1823                         goto setup_rx_desc_die;
1824                 }
1825
1826                 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1827                         /* give up */
1828                         pci_free_consistent(pdev, rxdr->size, rxdr->desc,
1829                                             rxdr->dma);
1830                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1831                         DPRINTK(PROBE, ERR,
1832                                 "Unable to allocate aligned memory "
1833                                 "for the receive descriptor ring\n");
1834                         goto setup_rx_desc_die;
1835                 } else {
1836                         /* Free old allocation, new allocation was successful */
1837                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1838                 }
1839         }
1840         memset(rxdr->desc, 0, rxdr->size);
1841
1842         rxdr->next_to_clean = 0;
1843         rxdr->next_to_use = 0;
1844
1845         return 0;
1846 }
1847
1848 /**
1849  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1850  *                                (Descriptors) for all queues
1851  * @adapter: board private structure
1852  *
1853  * Return 0 on success, negative on failure
1854  **/
1855
1856 int
1857 e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1858 {
1859         int i, err = 0;
1860
1861         for (i = 0; i < adapter->num_rx_queues; i++) {
1862                 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1863                 if (err) {
1864                         DPRINTK(PROBE, ERR,
1865                                 "Allocation for Rx Queue %u failed\n", i);
1866                         for (i-- ; i >= 0; i--)
1867                                 e1000_free_rx_resources(adapter,
1868                                                         &adapter->rx_ring[i]);
1869                         break;
1870                 }
1871         }
1872
1873         return err;
1874 }
1875
1876 /**
1877  * e1000_setup_rctl - configure the receive control registers
1878  * @adapter: Board private structure
1879  **/
1880 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1881                         (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1882 static void
1883 e1000_setup_rctl(struct e1000_adapter *adapter)
1884 {
1885         uint32_t rctl, rfctl;
1886         uint32_t psrctl = 0;
1887 #ifndef CONFIG_E1000_DISABLE_PACKET_SPLIT
1888         uint32_t pages = 0;
1889 #endif
1890
1891         rctl = E1000_READ_REG(&adapter->hw, RCTL);
1892
1893         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1894
1895         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1896                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1897                 (adapter->hw.mc_filter_type << E1000_RCTL_MO_SHIFT);
1898
1899         if (adapter->hw.tbi_compatibility_on == 1)
1900                 rctl |= E1000_RCTL_SBP;
1901         else
1902                 rctl &= ~E1000_RCTL_SBP;
1903
1904         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1905                 rctl &= ~E1000_RCTL_LPE;
1906         else
1907                 rctl |= E1000_RCTL_LPE;
1908
1909         /* Setup buffer sizes */
1910         rctl &= ~E1000_RCTL_SZ_4096;
1911         rctl |= E1000_RCTL_BSEX;
1912         switch (adapter->rx_buffer_len) {
1913                 case E1000_RXBUFFER_256:
1914                         rctl |= E1000_RCTL_SZ_256;
1915                         rctl &= ~E1000_RCTL_BSEX;
1916                         break;
1917                 case E1000_RXBUFFER_512:
1918                         rctl |= E1000_RCTL_SZ_512;
1919                         rctl &= ~E1000_RCTL_BSEX;
1920                         break;
1921                 case E1000_RXBUFFER_1024:
1922                         rctl |= E1000_RCTL_SZ_1024;
1923                         rctl &= ~E1000_RCTL_BSEX;
1924                         break;
1925                 case E1000_RXBUFFER_2048:
1926                 default:
1927                         rctl |= E1000_RCTL_SZ_2048;
1928                         rctl &= ~E1000_RCTL_BSEX;
1929                         break;
1930                 case E1000_RXBUFFER_4096:
1931                         rctl |= E1000_RCTL_SZ_4096;
1932                         break;
1933                 case E1000_RXBUFFER_8192:
1934                         rctl |= E1000_RCTL_SZ_8192;
1935                         break;
1936                 case E1000_RXBUFFER_16384:
1937                         rctl |= E1000_RCTL_SZ_16384;
1938                         break;
1939         }
1940
1941 #ifndef CONFIG_E1000_DISABLE_PACKET_SPLIT
1942         /* 82571 and greater support packet-split where the protocol
1943          * header is placed in skb->data and the packet data is
1944          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
1945          * In the case of a non-split, skb->data is linearly filled,
1946          * followed by the page buffers.  Therefore, skb->data is
1947          * sized to hold the largest protocol header.
1948          */
1949         /* allocations using alloc_page take too long for regular MTU
1950          * so only enable packet split for jumbo frames */
1951         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
1952         if ((adapter->hw.mac_type >= e1000_82571) && (pages <= 3) &&
1953             PAGE_SIZE <= 16384 && (rctl & E1000_RCTL_LPE))
1954                 adapter->rx_ps_pages = pages;
1955         else
1956                 adapter->rx_ps_pages = 0;
1957 #endif
1958         if (adapter->rx_ps_pages) {
1959                 /* Configure extra packet-split registers */
1960                 rfctl = E1000_READ_REG(&adapter->hw, RFCTL);
1961                 rfctl |= E1000_RFCTL_EXTEN;
1962                 /* disable packet split support for IPv6 extension headers,
1963                  * because some malformed IPv6 headers can hang the RX */
1964                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
1965                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
1966
1967                 E1000_WRITE_REG(&adapter->hw, RFCTL, rfctl);
1968
1969                 rctl |= E1000_RCTL_DTYP_PS;
1970
1971                 psrctl |= adapter->rx_ps_bsize0 >>
1972                         E1000_PSRCTL_BSIZE0_SHIFT;
1973
1974                 switch (adapter->rx_ps_pages) {
1975                 case 3:
1976                         psrctl |= PAGE_SIZE <<
1977                                 E1000_PSRCTL_BSIZE3_SHIFT;
1978                 case 2:
1979                         psrctl |= PAGE_SIZE <<
1980                                 E1000_PSRCTL_BSIZE2_SHIFT;
1981                 case 1:
1982                         psrctl |= PAGE_SIZE >>
1983                                 E1000_PSRCTL_BSIZE1_SHIFT;
1984                         break;
1985                 }
1986
1987                 E1000_WRITE_REG(&adapter->hw, PSRCTL, psrctl);
1988         }
1989
1990         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
1991 }
1992
1993 /**
1994  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1995  * @adapter: board private structure
1996  *
1997  * Configure the Rx unit of the MAC after a reset.
1998  **/
1999
2000 static void
2001 e1000_configure_rx(struct e1000_adapter *adapter)
2002 {
2003         uint64_t rdba;
2004         struct e1000_hw *hw = &adapter->hw;
2005         uint32_t rdlen, rctl, rxcsum, ctrl_ext;
2006
2007         if (adapter->rx_ps_pages) {
2008                 /* this is a 32 byte descriptor */
2009                 rdlen = adapter->rx_ring[0].count *
2010                         sizeof(union e1000_rx_desc_packet_split);
2011                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2012                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2013         } else {
2014                 rdlen = adapter->rx_ring[0].count *
2015                         sizeof(struct e1000_rx_desc);
2016                 adapter->clean_rx = e1000_clean_rx_irq;
2017                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2018         }
2019
2020         /* disable receives while setting up the descriptors */
2021         rctl = E1000_READ_REG(hw, RCTL);
2022         E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
2023
2024         /* set the Receive Delay Timer Register */
2025         E1000_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
2026
2027         if (hw->mac_type >= e1000_82540) {
2028                 E1000_WRITE_REG(hw, RADV, adapter->rx_abs_int_delay);
2029                 if (adapter->itr_setting != 0)
2030                         E1000_WRITE_REG(hw, ITR,
2031                                 1000000000 / (adapter->itr * 256));
2032         }
2033
2034         if (hw->mac_type >= e1000_82571) {
2035                 ctrl_ext = E1000_READ_REG(hw, CTRL_EXT);
2036                 /* Reset delay timers after every interrupt */
2037                 ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2038 #ifdef CONFIG_E1000_NAPI
2039                 /* Auto-Mask interrupts upon ICR access */
2040                 ctrl_ext |= E1000_CTRL_EXT_IAME;
2041                 E1000_WRITE_REG(hw, IAM, 0xffffffff);
2042 #endif
2043                 E1000_WRITE_REG(hw, CTRL_EXT, ctrl_ext);
2044                 E1000_WRITE_FLUSH(hw);
2045         }
2046
2047         /* Setup the HW Rx Head and Tail Descriptor Pointers and
2048          * the Base and Length of the Rx Descriptor Ring */
2049         switch (adapter->num_rx_queues) {
2050         case 1:
2051         default:
2052                 rdba = adapter->rx_ring[0].dma;
2053                 E1000_WRITE_REG(hw, RDLEN, rdlen);
2054                 E1000_WRITE_REG(hw, RDBAH, (rdba >> 32));
2055                 E1000_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
2056                 E1000_WRITE_REG(hw, RDT, 0);
2057                 E1000_WRITE_REG(hw, RDH, 0);
2058                 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ? E1000_RDH : E1000_82542_RDH);
2059                 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ? E1000_RDT : E1000_82542_RDT);
2060                 break;
2061         }
2062
2063         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
2064         if (hw->mac_type >= e1000_82543) {
2065                 rxcsum = E1000_READ_REG(hw, RXCSUM);
2066                 if (adapter->rx_csum == TRUE) {
2067                         rxcsum |= E1000_RXCSUM_TUOFL;
2068
2069                         /* Enable 82571 IPv4 payload checksum for UDP fragments
2070                          * Must be used in conjunction with packet-split. */
2071                         if ((hw->mac_type >= e1000_82571) &&
2072                             (adapter->rx_ps_pages)) {
2073                                 rxcsum |= E1000_RXCSUM_IPPCSE;
2074                         }
2075                 } else {
2076                         rxcsum &= ~E1000_RXCSUM_TUOFL;
2077                         /* don't need to clear IPPCSE as it defaults to 0 */
2078                 }
2079                 E1000_WRITE_REG(hw, RXCSUM, rxcsum);
2080         }
2081
2082         /* enable early receives on 82573, only takes effect if using > 2048
2083          * byte total frame size.  for example only for jumbo frames */
2084 #define E1000_ERT_2048 0x100
2085         if (hw->mac_type == e1000_82573)
2086                 E1000_WRITE_REG(hw, ERT, E1000_ERT_2048);
2087
2088         /* Enable Receives */
2089         E1000_WRITE_REG(hw, RCTL, rctl);
2090 }
2091
2092 /**
2093  * e1000_free_tx_resources - Free Tx Resources per Queue
2094  * @adapter: board private structure
2095  * @tx_ring: Tx descriptor ring for a specific queue
2096  *
2097  * Free all transmit software resources
2098  **/
2099
2100 static void
2101 e1000_free_tx_resources(struct e1000_adapter *adapter,
2102                         struct e1000_tx_ring *tx_ring)
2103 {
2104         struct pci_dev *pdev = adapter->pdev;
2105
2106         e1000_clean_tx_ring(adapter, tx_ring);
2107
2108         vfree(tx_ring->buffer_info);
2109         tx_ring->buffer_info = NULL;
2110
2111         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
2112
2113         tx_ring->desc = NULL;
2114 }
2115
2116 /**
2117  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
2118  * @adapter: board private structure
2119  *
2120  * Free all transmit software resources
2121  **/
2122
2123 void
2124 e1000_free_all_tx_resources(struct e1000_adapter *adapter)
2125 {
2126         int i;
2127
2128         for (i = 0; i < adapter->num_tx_queues; i++)
2129                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
2130 }
2131
2132 static void
2133 e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
2134                         struct e1000_buffer *buffer_info)
2135 {
2136         if (buffer_info->dma) {
2137                 pci_unmap_page(adapter->pdev,
2138                                 buffer_info->dma,
2139                                 buffer_info->length,
2140                                 PCI_DMA_TODEVICE);
2141                 buffer_info->dma = 0;
2142         }
2143         if (buffer_info->skb) {
2144                 dev_kfree_skb_any(buffer_info->skb);
2145                 buffer_info->skb = NULL;
2146         }
2147         /* buffer_info must be completely set up in the transmit path */
2148 }
2149
2150 /**
2151  * e1000_clean_tx_ring - Free Tx Buffers
2152  * @adapter: board private structure
2153  * @tx_ring: ring to be cleaned
2154  **/
2155
2156 static void
2157 e1000_clean_tx_ring(struct e1000_adapter *adapter,
2158                     struct e1000_tx_ring *tx_ring)
2159 {
2160         struct e1000_buffer *buffer_info;
2161         unsigned long size;
2162         unsigned int i;
2163
2164         /* Free all the Tx ring sk_buffs */
2165
2166         for (i = 0; i < tx_ring->count; i++) {
2167                 buffer_info = &tx_ring->buffer_info[i];
2168                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2169         }
2170
2171         size = sizeof(struct e1000_buffer) * tx_ring->count;
2172         memset(tx_ring->buffer_info, 0, size);
2173
2174         /* Zero out the descriptor ring */
2175
2176         memset(tx_ring->desc, 0, tx_ring->size);
2177
2178         tx_ring->next_to_use = 0;
2179         tx_ring->next_to_clean = 0;
2180         tx_ring->last_tx_tso = 0;
2181
2182         writel(0, adapter->hw.hw_addr + tx_ring->tdh);
2183         writel(0, adapter->hw.hw_addr + tx_ring->tdt);
2184 }
2185
2186 /**
2187  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
2188  * @adapter: board private structure
2189  **/
2190
2191 static void
2192 e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
2193 {
2194         int i;
2195
2196         for (i = 0; i < adapter->num_tx_queues; i++)
2197                 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2198 }
2199
2200 /**
2201  * e1000_free_rx_resources - Free Rx Resources
2202  * @adapter: board private structure
2203  * @rx_ring: ring to clean the resources from
2204  *
2205  * Free all receive software resources
2206  **/
2207
2208 static void
2209 e1000_free_rx_resources(struct e1000_adapter *adapter,
2210                         struct e1000_rx_ring *rx_ring)
2211 {
2212         struct pci_dev *pdev = adapter->pdev;
2213
2214         e1000_clean_rx_ring(adapter, rx_ring);
2215
2216         vfree(rx_ring->buffer_info);
2217         rx_ring->buffer_info = NULL;
2218         kfree(rx_ring->ps_page);
2219         rx_ring->ps_page = NULL;
2220         kfree(rx_ring->ps_page_dma);
2221         rx_ring->ps_page_dma = NULL;
2222
2223         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2224
2225         rx_ring->desc = NULL;
2226 }
2227
2228 /**
2229  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
2230  * @adapter: board private structure
2231  *
2232  * Free all receive software resources
2233  **/
2234
2235 void
2236 e1000_free_all_rx_resources(struct e1000_adapter *adapter)
2237 {
2238         int i;
2239
2240         for (i = 0; i < adapter->num_rx_queues; i++)
2241                 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
2242 }
2243
2244 /**
2245  * e1000_clean_rx_ring - Free Rx Buffers per Queue
2246  * @adapter: board private structure
2247  * @rx_ring: ring to free buffers from
2248  **/
2249
2250 static void
2251 e1000_clean_rx_ring(struct e1000_adapter *adapter,
2252                     struct e1000_rx_ring *rx_ring)
2253 {
2254         struct e1000_buffer *buffer_info;
2255         struct e1000_ps_page *ps_page;
2256         struct e1000_ps_page_dma *ps_page_dma;
2257         struct pci_dev *pdev = adapter->pdev;
2258         unsigned long size;
2259         unsigned int i, j;
2260
2261         /* Free all the Rx ring sk_buffs */
2262         for (i = 0; i < rx_ring->count; i++) {
2263                 buffer_info = &rx_ring->buffer_info[i];
2264                 if (buffer_info->skb) {
2265                         pci_unmap_single(pdev,
2266                                          buffer_info->dma,
2267                                          buffer_info->length,
2268                                          PCI_DMA_FROMDEVICE);
2269
2270                         dev_kfree_skb(buffer_info->skb);
2271                         buffer_info->skb = NULL;
2272                 }
2273                 ps_page = &rx_ring->ps_page[i];
2274                 ps_page_dma = &rx_ring->ps_page_dma[i];
2275                 for (j = 0; j < adapter->rx_ps_pages; j++) {
2276                         if (!ps_page->ps_page[j]) break;
2277                         pci_unmap_page(pdev,
2278                                        ps_page_dma->ps_page_dma[j],
2279                                        PAGE_SIZE, PCI_DMA_FROMDEVICE);
2280                         ps_page_dma->ps_page_dma[j] = 0;
2281                         put_page(ps_page->ps_page[j]);
2282                         ps_page->ps_page[j] = NULL;
2283                 }
2284         }
2285
2286         size = sizeof(struct e1000_buffer) * rx_ring->count;
2287         memset(rx_ring->buffer_info, 0, size);
2288         size = sizeof(struct e1000_ps_page) * rx_ring->count;
2289         memset(rx_ring->ps_page, 0, size);
2290         size = sizeof(struct e1000_ps_page_dma) * rx_ring->count;
2291         memset(rx_ring->ps_page_dma, 0, size);
2292
2293         /* Zero out the descriptor ring */
2294
2295         memset(rx_ring->desc, 0, rx_ring->size);
2296
2297         rx_ring->next_to_clean = 0;
2298         rx_ring->next_to_use = 0;
2299
2300         writel(0, adapter->hw.hw_addr + rx_ring->rdh);
2301         writel(0, adapter->hw.hw_addr + rx_ring->rdt);
2302 }
2303
2304 /**
2305  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2306  * @adapter: board private structure
2307  **/
2308
2309 static void
2310 e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2311 {
2312         int i;
2313
2314         for (i = 0; i < adapter->num_rx_queues; i++)
2315                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2316 }
2317
2318 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2319  * and memory write and invalidate disabled for certain operations
2320  */
2321 static void
2322 e1000_enter_82542_rst(struct e1000_adapter *adapter)
2323 {
2324         struct net_device *netdev = adapter->netdev;
2325         uint32_t rctl;
2326
2327         e1000_pci_clear_mwi(&adapter->hw);
2328
2329         rctl = E1000_READ_REG(&adapter->hw, RCTL);
2330         rctl |= E1000_RCTL_RST;
2331         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
2332         E1000_WRITE_FLUSH(&adapter->hw);
2333         mdelay(5);
2334
2335         if (netif_running(netdev))
2336                 e1000_clean_all_rx_rings(adapter);
2337 }
2338
2339 static void
2340 e1000_leave_82542_rst(struct e1000_adapter *adapter)
2341 {
2342         struct net_device *netdev = adapter->netdev;
2343         uint32_t rctl;
2344
2345         rctl = E1000_READ_REG(&adapter->hw, RCTL);
2346         rctl &= ~E1000_RCTL_RST;
2347         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
2348         E1000_WRITE_FLUSH(&adapter->hw);
2349         mdelay(5);
2350
2351         if (adapter->hw.pci_cmd_word & PCI_COMMAND_INVALIDATE)
2352                 e1000_pci_set_mwi(&adapter->hw);
2353
2354         if (netif_running(netdev)) {
2355                 /* No need to loop, because 82542 supports only 1 queue */
2356                 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2357                 e1000_configure_rx(adapter);
2358                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2359         }
2360 }
2361
2362 /**
2363  * e1000_set_mac - Change the Ethernet Address of the NIC
2364  * @netdev: network interface device structure
2365  * @p: pointer to an address structure
2366  *
2367  * Returns 0 on success, negative on failure
2368  **/
2369
2370 static int
2371 e1000_set_mac(struct net_device *netdev, void *p)
2372 {
2373         struct e1000_adapter *adapter = netdev_priv(netdev);
2374         struct sockaddr *addr = p;
2375
2376         if (!is_valid_ether_addr(addr->sa_data))
2377                 return -EADDRNOTAVAIL;
2378
2379         /* 82542 2.0 needs to be in reset to write receive address registers */
2380
2381         if (adapter->hw.mac_type == e1000_82542_rev2_0)
2382                 e1000_enter_82542_rst(adapter);
2383
2384         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2385         memcpy(adapter->hw.mac_addr, addr->sa_data, netdev->addr_len);
2386
2387         e1000_rar_set(&adapter->hw, adapter->hw.mac_addr, 0);
2388
2389         /* With 82571 controllers, LAA may be overwritten (with the default)
2390          * due to controller reset from the other port. */
2391         if (adapter->hw.mac_type == e1000_82571) {
2392                 /* activate the work around */
2393                 adapter->hw.laa_is_present = 1;
2394
2395                 /* Hold a copy of the LAA in RAR[14] This is done so that
2396                  * between the time RAR[0] gets clobbered  and the time it
2397                  * gets fixed (in e1000_watchdog), the actual LAA is in one
2398                  * of the RARs and no incoming packets directed to this port
2399                  * are dropped. Eventaully the LAA will be in RAR[0] and
2400                  * RAR[14] */
2401                 e1000_rar_set(&adapter->hw, adapter->hw.mac_addr,
2402                                         E1000_RAR_ENTRIES - 1);
2403         }
2404
2405         if (adapter->hw.mac_type == e1000_82542_rev2_0)
2406                 e1000_leave_82542_rst(adapter);
2407
2408         return 0;
2409 }
2410
2411 /**
2412  * e1000_set_multi - Multicast and Promiscuous mode set
2413  * @netdev: network interface device structure
2414  *
2415  * The set_multi entry point is called whenever the multicast address
2416  * list or the network interface flags are updated.  This routine is
2417  * responsible for configuring the hardware for proper multicast,
2418  * promiscuous mode, and all-multi behavior.
2419  **/
2420
2421 static void
2422 e1000_set_multi(struct net_device *netdev)
2423 {
2424         struct e1000_adapter *adapter = netdev_priv(netdev);
2425         struct e1000_hw *hw = &adapter->hw;
2426         struct dev_mc_list *mc_ptr;
2427         uint32_t rctl;
2428         uint32_t hash_value;
2429         int i, rar_entries = E1000_RAR_ENTRIES;
2430         int mta_reg_count = (hw->mac_type == e1000_ich8lan) ?
2431                                 E1000_NUM_MTA_REGISTERS_ICH8LAN :
2432                                 E1000_NUM_MTA_REGISTERS;
2433
2434         if (adapter->hw.mac_type == e1000_ich8lan)
2435                 rar_entries = E1000_RAR_ENTRIES_ICH8LAN;
2436
2437         /* reserve RAR[14] for LAA over-write work-around */
2438         if (adapter->hw.mac_type == e1000_82571)
2439                 rar_entries--;
2440
2441         /* Check for Promiscuous and All Multicast modes */
2442
2443         rctl = E1000_READ_REG(hw, RCTL);
2444
2445         if (netdev->flags & IFF_PROMISC) {
2446                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2447         } else if (netdev->flags & IFF_ALLMULTI) {
2448                 rctl |= E1000_RCTL_MPE;
2449                 rctl &= ~E1000_RCTL_UPE;
2450         } else {
2451                 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2452         }
2453
2454         E1000_WRITE_REG(hw, RCTL, rctl);
2455
2456         /* 82542 2.0 needs to be in reset to write receive address registers */
2457
2458         if (hw->mac_type == e1000_82542_rev2_0)
2459                 e1000_enter_82542_rst(adapter);
2460
2461         /* load the first 14 multicast address into the exact filters 1-14
2462          * RAR 0 is used for the station MAC adddress
2463          * if there are not 14 addresses, go ahead and clear the filters
2464          * -- with 82571 controllers only 0-13 entries are filled here
2465          */
2466         mc_ptr = netdev->mc_list;
2467
2468         for (i = 1; i < rar_entries; i++) {
2469                 if (mc_ptr) {
2470                         e1000_rar_set(hw, mc_ptr->dmi_addr, i);
2471                         mc_ptr = mc_ptr->next;
2472                 } else {
2473                         E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2474                         E1000_WRITE_FLUSH(hw);
2475                         E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2476                         E1000_WRITE_FLUSH(hw);
2477                 }
2478         }
2479
2480         /* clear the old settings from the multicast hash table */
2481
2482         for (i = 0; i < mta_reg_count; i++) {
2483                 E1000_WRITE_REG_ARRAY(hw, MTA, i, 0);
2484                 E1000_WRITE_FLUSH(hw);
2485         }
2486
2487         /* load any remaining addresses into the hash table */
2488
2489         for (; mc_ptr; mc_ptr = mc_ptr->next) {
2490                 hash_value = e1000_hash_mc_addr(hw, mc_ptr->dmi_addr);
2491                 e1000_mta_set(hw, hash_value);
2492         }
2493
2494         if (hw->mac_type == e1000_82542_rev2_0)
2495                 e1000_leave_82542_rst(adapter);
2496 }
2497
2498 /* Need to wait a few seconds after link up to get diagnostic information from
2499  * the phy */
2500
2501 static void
2502 e1000_update_phy_info(unsigned long data)
2503 {
2504         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2505         e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
2506 }
2507
2508 /**
2509  * e1000_82547_tx_fifo_stall - Timer Call-back
2510  * @data: pointer to adapter cast into an unsigned long
2511  **/
2512
2513 static void
2514 e1000_82547_tx_fifo_stall(unsigned long data)
2515 {
2516         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2517         struct net_device *netdev = adapter->netdev;
2518         uint32_t tctl;
2519
2520         if (atomic_read(&adapter->tx_fifo_stall)) {
2521                 if ((E1000_READ_REG(&adapter->hw, TDT) ==
2522                     E1000_READ_REG(&adapter->hw, TDH)) &&
2523                    (E1000_READ_REG(&adapter->hw, TDFT) ==
2524                     E1000_READ_REG(&adapter->hw, TDFH)) &&
2525                    (E1000_READ_REG(&adapter->hw, TDFTS) ==
2526                     E1000_READ_REG(&adapter->hw, TDFHS))) {
2527                         tctl = E1000_READ_REG(&adapter->hw, TCTL);
2528                         E1000_WRITE_REG(&adapter->hw, TCTL,
2529                                         tctl & ~E1000_TCTL_EN);
2530                         E1000_WRITE_REG(&adapter->hw, TDFT,
2531                                         adapter->tx_head_addr);
2532                         E1000_WRITE_REG(&adapter->hw, TDFH,
2533                                         adapter->tx_head_addr);
2534                         E1000_WRITE_REG(&adapter->hw, TDFTS,
2535                                         adapter->tx_head_addr);
2536                         E1000_WRITE_REG(&adapter->hw, TDFHS,
2537                                         adapter->tx_head_addr);
2538                         E1000_WRITE_REG(&adapter->hw, TCTL, tctl);
2539                         E1000_WRITE_FLUSH(&adapter->hw);
2540
2541                         adapter->tx_fifo_head = 0;
2542                         atomic_set(&adapter->tx_fifo_stall, 0);
2543                         netif_wake_queue(netdev);
2544                 } else {
2545                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2546                 }
2547         }
2548 }
2549
2550 /**
2551  * e1000_watchdog - Timer Call-back
2552  * @data: pointer to adapter cast into an unsigned long
2553  **/
2554 static void
2555 e1000_watchdog(unsigned long data)
2556 {
2557         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2558         struct net_device *netdev = adapter->netdev;
2559         struct e1000_tx_ring *txdr = adapter->tx_ring;
2560         uint32_t link, tctl;
2561         int32_t ret_val;
2562
2563         ret_val = e1000_check_for_link(&adapter->hw);
2564         if ((ret_val == E1000_ERR_PHY) &&
2565             (adapter->hw.phy_type == e1000_phy_igp_3) &&
2566             (E1000_READ_REG(&adapter->hw, CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
2567                 /* See e1000_kumeran_lock_loss_workaround() */
2568                 DPRINTK(LINK, INFO,
2569                         "Gigabit has been disabled, downgrading speed\n");
2570         }
2571
2572         if (adapter->hw.mac_type == e1000_82573) {
2573                 e1000_enable_tx_pkt_filtering(&adapter->hw);
2574                 if (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id)
2575                         e1000_update_mng_vlan(adapter);
2576         }
2577
2578         if ((adapter->hw.media_type == e1000_media_type_internal_serdes) &&
2579            !(E1000_READ_REG(&adapter->hw, TXCW) & E1000_TXCW_ANE))
2580                 link = !adapter->hw.serdes_link_down;
2581         else
2582                 link = E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_LU;
2583
2584         if (link) {
2585                 if (!netif_carrier_ok(netdev)) {
2586                         uint32_t ctrl;
2587                         boolean_t txb2b = 1;
2588                         e1000_get_speed_and_duplex(&adapter->hw,
2589                                                    &adapter->link_speed,
2590                                                    &adapter->link_duplex);
2591
2592                         ctrl = E1000_READ_REG(&adapter->hw, CTRL);
2593                         DPRINTK(LINK, INFO, "NIC Link is Up %d Mbps %s, "
2594                                 "Flow Control: %s\n",
2595                                 adapter->link_speed,
2596                                 adapter->link_duplex == FULL_DUPLEX ?
2597                                 "Full Duplex" : "Half Duplex",
2598                                 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2599                                 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2600                                 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2601                                 E1000_CTRL_TFCE) ? "TX" : "None" )));
2602
2603                         /* tweak tx_queue_len according to speed/duplex
2604                          * and adjust the timeout factor */
2605                         netdev->tx_queue_len = adapter->tx_queue_len;
2606                         adapter->tx_timeout_factor = 1;
2607                         switch (adapter->link_speed) {
2608                         case SPEED_10:
2609                                 txb2b = 0;
2610                                 netdev->tx_queue_len = 10;
2611                                 adapter->tx_timeout_factor = 8;
2612                                 break;
2613                         case SPEED_100:
2614                                 txb2b = 0;
2615                                 netdev->tx_queue_len = 100;
2616                                 /* maybe add some timeout factor ? */
2617                                 break;
2618                         }
2619
2620                         if ((adapter->hw.mac_type == e1000_82571 ||
2621                              adapter->hw.mac_type == e1000_82572) &&
2622                             txb2b == 0) {
2623                                 uint32_t tarc0;
2624                                 tarc0 = E1000_READ_REG(&adapter->hw, TARC0);
2625                                 tarc0 &= ~(1 << 21);
2626                                 E1000_WRITE_REG(&adapter->hw, TARC0, tarc0);
2627                         }
2628
2629                         /* disable TSO for pcie and 10/100 speeds, to avoid
2630                          * some hardware issues */
2631                         if (!adapter->tso_force &&
2632                             adapter->hw.bus_type == e1000_bus_type_pci_express){
2633                                 switch (adapter->link_speed) {
2634                                 case SPEED_10:
2635                                 case SPEED_100:
2636                                         DPRINTK(PROBE,INFO,
2637                                         "10/100 speed: disabling TSO\n");
2638                                         netdev->features &= ~NETIF_F_TSO;
2639                                         netdev->features &= ~NETIF_F_TSO6;
2640                                         break;
2641                                 case SPEED_1000:
2642                                         netdev->features |= NETIF_F_TSO;
2643                                         netdev->features |= NETIF_F_TSO6;
2644                                         break;
2645                                 default:
2646                                         /* oops */
2647                                         break;
2648                                 }
2649                         }
2650
2651                         /* enable transmits in the hardware, need to do this
2652                          * after setting TARC0 */
2653                         tctl = E1000_READ_REG(&adapter->hw, TCTL);
2654                         tctl |= E1000_TCTL_EN;
2655                         E1000_WRITE_REG(&adapter->hw, TCTL, tctl);
2656
2657                         netif_carrier_on(netdev);
2658                         netif_wake_queue(netdev);
2659                         mod_timer(&adapter->phy_info_timer, round_jiffies(jiffies + 2 * HZ));
2660                         adapter->smartspeed = 0;
2661                 } else {
2662                         /* make sure the receive unit is started */
2663                         if (adapter->hw.rx_needs_kicking) {
2664                                 struct e1000_hw *hw = &adapter->hw;
2665                                 uint32_t rctl = E1000_READ_REG(hw, RCTL);
2666                                 E1000_WRITE_REG(hw, RCTL, rctl | E1000_RCTL_EN);
2667                         }
2668                 }
2669         } else {
2670                 if (netif_carrier_ok(netdev)) {
2671                         adapter->link_speed = 0;
2672                         adapter->link_duplex = 0;
2673                         DPRINTK(LINK, INFO, "NIC Link is Down\n");
2674                         netif_carrier_off(netdev);
2675                         netif_stop_queue(netdev);
2676                         mod_timer(&adapter->phy_info_timer, round_jiffies(jiffies + 2 * HZ));
2677
2678                         /* 80003ES2LAN workaround--
2679                          * For packet buffer work-around on link down event;
2680                          * disable receives in the ISR and
2681                          * reset device here in the watchdog
2682                          */
2683                         if (adapter->hw.mac_type == e1000_80003es2lan)
2684                                 /* reset device */
2685                                 schedule_work(&adapter->reset_task);
2686                 }
2687
2688                 e1000_smartspeed(adapter);
2689         }
2690
2691         e1000_update_stats(adapter);
2692
2693         adapter->hw.tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2694         adapter->tpt_old = adapter->stats.tpt;
2695         adapter->hw.collision_delta = adapter->stats.colc - adapter->colc_old;
2696         adapter->colc_old = adapter->stats.colc;
2697
2698         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2699         adapter->gorcl_old = adapter->stats.gorcl;
2700         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2701         adapter->gotcl_old = adapter->stats.gotcl;
2702
2703         e1000_update_adaptive(&adapter->hw);
2704
2705         if (!netif_carrier_ok(netdev)) {
2706                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2707                         /* We've lost link, so the controller stops DMA,
2708                          * but we've got queued Tx work that's never going
2709                          * to get done, so reset controller to flush Tx.
2710                          * (Do the reset outside of interrupt context). */
2711                         adapter->tx_timeout_count++;
2712                         schedule_work(&adapter->reset_task);
2713                 }
2714         }
2715
2716         /* Cause software interrupt to ensure rx ring is cleaned */
2717         E1000_WRITE_REG(&adapter->hw, ICS, E1000_ICS_RXDMT0);
2718
2719         /* Force detection of hung controller every watchdog period */
2720         adapter->detect_tx_hung = TRUE;
2721
2722         /* With 82571 controllers, LAA may be overwritten due to controller
2723          * reset from the other port. Set the appropriate LAA in RAR[0] */
2724         if (adapter->hw.mac_type == e1000_82571 && adapter->hw.laa_is_present)
2725                 e1000_rar_set(&adapter->hw, adapter->hw.mac_addr, 0);
2726
2727         /* Reset the timer */
2728         mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
2729 }
2730
2731 enum latency_range {
2732         lowest_latency = 0,
2733         low_latency = 1,
2734         bulk_latency = 2,
2735         latency_invalid = 255
2736 };
2737
2738 /**
2739  * e1000_update_itr - update the dynamic ITR value based on statistics
2740  *      Stores a new ITR value based on packets and byte
2741  *      counts during the last interrupt.  The advantage of per interrupt
2742  *      computation is faster updates and more accurate ITR for the current
2743  *      traffic pattern.  Constants in this function were computed
2744  *      based on theoretical maximum wire speed and thresholds were set based
2745  *      on testing data as well as attempting to minimize response time
2746  *      while increasing bulk throughput.
2747  *      this functionality is controlled by the InterruptThrottleRate module
2748  *      parameter (see e1000_param.c)
2749  * @adapter: pointer to adapter
2750  * @itr_setting: current adapter->itr
2751  * @packets: the number of packets during this measurement interval
2752  * @bytes: the number of bytes during this measurement interval
2753  **/
2754 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2755                                    uint16_t itr_setting,
2756                                    int packets,
2757                                    int bytes)
2758 {
2759         unsigned int retval = itr_setting;
2760         struct e1000_hw *hw = &adapter->hw;
2761
2762         if (unlikely(hw->mac_type < e1000_82540))
2763                 goto update_itr_done;
2764
2765         if (packets == 0)
2766                 goto update_itr_done;
2767
2768         switch (itr_setting) {
2769         case lowest_latency:
2770                 /* jumbo frames get bulk treatment*/
2771                 if (bytes/packets > 8000)
2772                         retval = bulk_latency;
2773                 else if ((packets < 5) && (bytes > 512))
2774                         retval = low_latency;
2775                 break;
2776         case low_latency:  /* 50 usec aka 20000 ints/s */
2777                 if (bytes > 10000) {
2778                         /* jumbo frames need bulk latency setting */
2779                         if (bytes/packets > 8000)
2780                                 retval = bulk_latency;
2781                         else if ((packets < 10) || ((bytes/packets) > 1200))
2782                                 retval = bulk_latency;
2783                         else if ((packets > 35))
2784                                 retval = lowest_latency;
2785                 } else if (bytes/packets > 2000)
2786                         retval = bulk_latency;
2787                 else if (packets <= 2 && bytes < 512)
2788                         retval = lowest_latency;
2789                 break;
2790         case bulk_latency: /* 250 usec aka 4000 ints/s */
2791                 if (bytes > 25000) {
2792                         if (packets > 35)
2793                                 retval = low_latency;
2794                 } else if (bytes < 6000) {
2795                         retval = low_latency;
2796                 }
2797                 break;
2798         }
2799
2800 update_itr_done:
2801         return retval;
2802 }
2803
2804 static void e1000_set_itr(struct e1000_adapter *adapter)
2805 {
2806         struct e1000_hw *hw = &adapter->hw;
2807         uint16_t current_itr;
2808         uint32_t new_itr = adapter->itr;
2809
2810         if (unlikely(hw->mac_type < e1000_82540))
2811                 return;
2812
2813         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2814         if (unlikely(adapter->link_speed != SPEED_1000)) {
2815                 current_itr = 0;
2816                 new_itr = 4000;
2817                 goto set_itr_now;
2818         }
2819
2820         adapter->tx_itr = e1000_update_itr(adapter,
2821                                     adapter->tx_itr,
2822                                     adapter->total_tx_packets,
2823                                     adapter->total_tx_bytes);
2824         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2825         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2826                 adapter->tx_itr = low_latency;
2827
2828         adapter->rx_itr = e1000_update_itr(adapter,
2829                                     adapter->rx_itr,
2830                                     adapter->total_rx_packets,
2831                                     adapter->total_rx_bytes);
2832         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2833         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2834                 adapter->rx_itr = low_latency;
2835
2836         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2837
2838         switch (current_itr) {
2839         /* counts and packets in update_itr are dependent on these numbers */
2840         case lowest_latency:
2841                 new_itr = 70000;
2842                 break;
2843         case low_latency:
2844                 new_itr = 20000; /* aka hwitr = ~200 */
2845                 break;
2846         case bulk_latency:
2847                 new_itr = 4000;
2848                 break;
2849         default:
2850                 break;
2851         }
2852
2853 set_itr_now:
2854         if (new_itr != adapter->itr) {
2855                 /* this attempts to bias the interrupt rate towards Bulk
2856                  * by adding intermediate steps when interrupt rate is
2857                  * increasing */
2858                 new_itr = new_itr > adapter->itr ?
2859                              min(adapter->itr + (new_itr >> 2), new_itr) :
2860                              new_itr;
2861                 adapter->itr = new_itr;
2862                 E1000_WRITE_REG(hw, ITR, 1000000000 / (new_itr * 256));
2863         }
2864
2865         return;
2866 }
2867
2868 #define E1000_TX_FLAGS_CSUM             0x00000001
2869 #define E1000_TX_FLAGS_VLAN             0x00000002
2870 #define E1000_TX_FLAGS_TSO              0x00000004
2871 #define E1000_TX_FLAGS_IPV4             0x00000008
2872 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2873 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2874
2875 static int
2876 e1000_tso(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2877           struct sk_buff *skb)
2878 {
2879         struct e1000_context_desc *context_desc;
2880         struct e1000_buffer *buffer_info;
2881         unsigned int i;
2882         uint32_t cmd_length = 0;
2883         uint16_t ipcse = 0, tucse, mss;
2884         uint8_t ipcss, ipcso, tucss, tucso, hdr_len;
2885         int err;
2886
2887         if (skb_is_gso(skb)) {
2888                 if (skb_header_cloned(skb)) {
2889                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2890                         if (err)
2891                                 return err;
2892                 }
2893
2894                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2895                 mss = skb_shinfo(skb)->gso_size;
2896                 if (skb->protocol == htons(ETH_P_IP)) {
2897                         struct iphdr *iph = ip_hdr(skb);
2898                         iph->tot_len = 0;
2899                         iph->check = 0;
2900                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2901                                                                  iph->daddr, 0,
2902                                                                  IPPROTO_TCP,
2903                                                                  0);
2904                         cmd_length = E1000_TXD_CMD_IP;
2905                         ipcse = skb_transport_offset(skb) - 1;
2906                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2907                         ipv6_hdr(skb)->payload_len = 0;
2908                         tcp_hdr(skb)->check =
2909                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2910                                                  &ipv6_hdr(skb)->daddr,
2911                                                  0, IPPROTO_TCP, 0);
2912                         ipcse = 0;
2913                 }
2914                 ipcss = skb_network_offset(skb);
2915                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2916                 tucss = skb_transport_offset(skb);
2917                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2918                 tucse = 0;
2919
2920                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2921                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2922
2923                 i = tx_ring->next_to_use;
2924                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2925                 buffer_info = &tx_ring->buffer_info[i];
2926
2927                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2928                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2929                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2930                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2931                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2932                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2933                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2934                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2935                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2936
2937                 buffer_info->time_stamp = jiffies;
2938                 buffer_info->next_to_watch = i;
2939
2940                 if (++i == tx_ring->count) i = 0;
2941                 tx_ring->next_to_use = i;
2942
2943                 return TRUE;
2944         }
2945         return FALSE;
2946 }
2947
2948 static boolean_t
2949 e1000_tx_csum(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2950               struct sk_buff *skb)
2951 {
2952         struct e1000_context_desc *context_desc;
2953         struct e1000_buffer *buffer_info;
2954         unsigned int i;
2955         uint8_t css;
2956
2957         if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
2958                 css = skb_transport_offset(skb);
2959
2960                 i = tx_ring->next_to_use;
2961                 buffer_info = &tx_ring->buffer_info[i];
2962                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2963
2964                 context_desc->lower_setup.ip_config = 0;
2965                 context_desc->upper_setup.tcp_fields.tucss = css;
2966                 context_desc->upper_setup.tcp_fields.tucso =
2967                         css + skb->csum_offset;
2968                 context_desc->upper_setup.tcp_fields.tucse = 0;
2969                 context_desc->tcp_seg_setup.data = 0;
2970                 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
2971
2972                 buffer_info->time_stamp = jiffies;
2973                 buffer_info->next_to_watch = i;
2974
2975                 if (unlikely(++i == tx_ring->count)) i = 0;
2976                 tx_ring->next_to_use = i;
2977
2978                 return TRUE;
2979         }
2980
2981         return FALSE;
2982 }
2983
2984 #define E1000_MAX_TXD_PWR       12
2985 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2986
2987 static int
2988 e1000_tx_map(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2989              struct sk_buff *skb, unsigned int first, unsigned int max_per_txd,
2990              unsigned int nr_frags, unsigned int mss)
2991 {
2992         struct e1000_buffer *buffer_info;
2993         unsigned int len = skb->len;
2994         unsigned int offset = 0, size, count = 0, i;
2995         unsigned int f;
2996         len -= skb->data_len;
2997
2998         i = tx_ring->next_to_use;
2999
3000         while (len) {
3001                 buffer_info = &tx_ring->buffer_info[i];
3002                 size = min(len, max_per_txd);
3003                 /* Workaround for Controller erratum --
3004                  * descriptor for non-tso packet in a linear SKB that follows a
3005                  * tso gets written back prematurely before the data is fully
3006                  * DMA'd to the controller */
3007                 if (!skb->data_len && tx_ring->last_tx_tso &&
3008                     !skb_is_gso(skb)) {
3009                         tx_ring->last_tx_tso = 0;
3010                         size -= 4;
3011                 }
3012
3013                 /* Workaround for premature desc write-backs
3014                  * in TSO mode.  Append 4-byte sentinel desc */
3015                 if (unlikely(mss && !nr_frags && size == len && size > 8))
3016                         size -= 4;
3017                 /* work-around for errata 10 and it applies
3018                  * to all controllers in PCI-X mode
3019                  * The fix is to make sure that the first descriptor of a
3020                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
3021                  */
3022                 if (unlikely((adapter->hw.bus_type == e1000_bus_type_pcix) &&
3023                                 (size > 2015) && count == 0))
3024                         size = 2015;
3025
3026                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
3027                  * terminating buffers within evenly-aligned dwords. */
3028                 if (unlikely(adapter->pcix_82544 &&
3029                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
3030                    size > 4))
3031                         size -= 4;
3032
3033                 buffer_info->length = size;
3034                 buffer_info->dma =
3035                         pci_map_single(adapter->pdev,
3036                                 skb->data + offset,
3037                                 size,
3038                                 PCI_DMA_TODEVICE);
3039                 buffer_info->time_stamp = jiffies;
3040                 buffer_info->next_to_watch = i;
3041
3042                 len -= size;
3043                 offset += size;
3044                 count++;
3045                 if (unlikely(++i == tx_ring->count)) i = 0;
3046         }
3047
3048         for (f = 0; f < nr_frags; f++) {
3049                 struct skb_frag_struct *frag;
3050
3051                 frag = &skb_shinfo(skb)->frags[f];
3052                 len = frag->size;
3053                 offset = frag->page_offset;
3054
3055                 while (len) {
3056                         buffer_info = &tx_ring->buffer_info[i];
3057                         size = min(len, max_per_txd);
3058                         /* Workaround for premature desc write-backs
3059                          * in TSO mode.  Append 4-byte sentinel desc */
3060                         if (unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
3061                                 size -= 4;
3062                         /* Workaround for potential 82544 hang in PCI-X.
3063                          * Avoid terminating buffers within evenly-aligned
3064                          * dwords. */
3065                         if (unlikely(adapter->pcix_82544 &&
3066                            !((unsigned long)(frag->page+offset+size-1) & 4) &&
3067                            size > 4))
3068                                 size -= 4;
3069
3070                         buffer_info->length = size;
3071                         buffer_info->dma =
3072                                 pci_map_page(adapter->pdev,
3073                                         frag->page,
3074                                         offset,
3075                                         size,
3076                                         PCI_DMA_TODEVICE);
3077                         buffer_info->time_stamp = jiffies;
3078                         buffer_info->next_to_watch = i;
3079
3080                         len -= size;
3081                         offset += size;
3082                         count++;
3083                         if (unlikely(++i == tx_ring->count)) i = 0;
3084                 }
3085         }
3086
3087         i = (i == 0) ? tx_ring->count - 1 : i - 1;
3088         tx_ring->buffer_info[i].skb = skb;
3089         tx_ring->buffer_info[first].next_to_watch = i;
3090
3091         return count;
3092 }
3093
3094 static void
3095 e1000_tx_queue(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
3096                int tx_flags, int count)
3097 {
3098         struct e1000_tx_desc *tx_desc = NULL;
3099         struct e1000_buffer *buffer_info;
3100         uint32_t txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3101         unsigned int i;
3102
3103         if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
3104                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3105                              E1000_TXD_CMD_TSE;
3106                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3107
3108                 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
3109                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3110         }
3111
3112         if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
3113                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3114                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3115         }
3116
3117         if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
3118                 txd_lower |= E1000_TXD_CMD_VLE;
3119                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3120         }
3121
3122         i = tx_ring->next_to_use;
3123
3124         while (count--) {
3125                 buffer_info = &tx_ring->buffer_info[i];
3126                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3127                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3128                 tx_desc->lower.data =
3129                         cpu_to_le32(txd_lower | buffer_info->length);
3130                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3131                 if (unlikely(++i == tx_ring->count)) i = 0;
3132         }
3133
3134         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3135
3136         /* Force memory writes to complete before letting h/w
3137          * know there are new descriptors to fetch.  (Only
3138          * applicable for weak-ordered memory model archs,
3139          * such as IA-64). */
3140         wmb();
3141
3142         tx_ring->next_to_use = i;
3143         writel(i, adapter->hw.hw_addr + tx_ring->tdt);
3144         /* we need this if more than one processor can write to our tail
3145          * at a time, it syncronizes IO on IA64/Altix systems */
3146         mmiowb();
3147 }
3148
3149 /**
3150  * 82547 workaround to avoid controller hang in half-duplex environment.
3151  * The workaround is to avoid queuing a large packet that would span
3152  * the internal Tx FIFO ring boundary by notifying the stack to resend
3153  * the packet at a later time.  This gives the Tx FIFO an opportunity to
3154  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
3155  * to the beginning of the Tx FIFO.
3156  **/
3157
3158 #define E1000_FIFO_HDR                  0x10
3159 #define E1000_82547_PAD_LEN             0x3E0
3160
3161 static int
3162 e1000_82547_fifo_workaround(struct e1000_adapter *adapter, struct sk_buff *skb)
3163 {
3164         uint32_t fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
3165         uint32_t skb_fifo_len = skb->len + E1000_FIFO_HDR;
3166
3167         skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
3168
3169         if (adapter->link_duplex != HALF_DUPLEX)
3170                 goto no_fifo_stall_required;
3171
3172         if (atomic_read(&adapter->tx_fifo_stall))
3173                 return 1;
3174
3175         if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
3176                 atomic_set(&adapter->tx_fifo_stall, 1);
3177                 return 1;
3178         }
3179
3180 no_fifo_stall_required:
3181         adapter->tx_fifo_head += skb_fifo_len;
3182         if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
3183                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
3184         return 0;
3185 }
3186
3187 #define MINIMUM_DHCP_PACKET_SIZE 282
3188 static int
3189 e1000_transfer_dhcp_info(struct e1000_adapter *adapter, struct sk_buff *skb)
3190 {
3191         struct e1000_hw *hw =  &adapter->hw;
3192         uint16_t length, offset;
3193         if (vlan_tx_tag_present(skb)) {
3194                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
3195                         ( adapter->hw.mng_cookie.status &
3196                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) )
3197                         return 0;
3198         }
3199         if (skb->len > MINIMUM_DHCP_PACKET_SIZE) {
3200                 struct ethhdr *eth = (struct ethhdr *) skb->data;
3201                 if ((htons(ETH_P_IP) == eth->h_proto)) {
3202                         const struct iphdr *ip =
3203                                 (struct iphdr *)((uint8_t *)skb->data+14);
3204                         if (IPPROTO_UDP == ip->protocol) {
3205                                 struct udphdr *udp =
3206                                         (struct udphdr *)((uint8_t *)ip +
3207                                                 (ip->ihl << 2));
3208                                 if (ntohs(udp->dest) == 67) {
3209                                         offset = (uint8_t *)udp + 8 - skb->data;
3210                                         length = skb->len - offset;
3211
3212                                         return e1000_mng_write_dhcp_info(hw,
3213                                                         (uint8_t *)udp + 8,
3214                                                         length);
3215                                 }
3216                         }
3217                 }
3218         }
3219         return 0;
3220 }
3221
3222 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3223 {
3224         struct e1000_adapter *adapter = netdev_priv(netdev);
3225         struct e1000_tx_ring *tx_ring = adapter->tx_ring;
3226
3227         netif_stop_queue(netdev);
3228         /* Herbert's original patch had:
3229          *  smp_mb__after_netif_stop_queue();
3230          * but since that doesn't exist yet, just open code it. */
3231         smp_mb();
3232
3233         /* We need to check again in a case another CPU has just
3234          * made room available. */
3235         if (likely(E1000_DESC_UNUSED(tx_ring) < size))
3236                 return -EBUSY;
3237
3238         /* A reprieve! */
3239         netif_start_queue(netdev);
3240         ++adapter->restart_queue;
3241         return 0;
3242 }
3243
3244 static int e1000_maybe_stop_tx(struct net_device *netdev,
3245                                struct e1000_tx_ring *tx_ring, int size)
3246 {
3247         if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
3248                 return 0;
3249         return __e1000_maybe_stop_tx(netdev, size);
3250 }
3251
3252 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3253 static int
3254 e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3255 {
3256         struct e1000_adapter *adapter = netdev_priv(netdev);
3257         struct e1000_tx_ring *tx_ring;
3258         unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
3259         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3260         unsigned int tx_flags = 0;
3261         unsigned int len = skb->len;
3262         unsigned long flags;
3263         unsigned int nr_frags = 0;
3264         unsigned int mss = 0;
3265         int count = 0;
3266         int tso;
3267         unsigned int f;
3268         len -= skb->data_len;
3269
3270         /* This goes back to the question of how to logically map a tx queue
3271          * to a flow.  Right now, performance is impacted slightly negatively
3272          * if using multiple tx queues.  If the stack breaks away from a
3273          * single qdisc implementation, we can look at this again. */
3274         tx_ring = adapter->tx_ring;
3275
3276         if (unlikely(skb->len <= 0)) {
3277                 dev_kfree_skb_any(skb);
3278                 return NETDEV_TX_OK;
3279         }
3280
3281         /* 82571 and newer doesn't need the workaround that limited descriptor
3282          * length to 4kB */
3283         if (adapter->hw.mac_type >= e1000_82571)
3284                 max_per_txd = 8192;
3285
3286         mss = skb_shinfo(skb)->gso_size;
3287         /* The controller does a simple calculation to
3288          * make sure there is enough room in the FIFO before
3289          * initiating the DMA for each buffer.  The calc is:
3290          * 4 = ceil(buffer len/mss).  To make sure we don't
3291          * overrun the FIFO, adjust the max buffer len if mss
3292          * drops. */
3293         if (mss) {
3294                 uint8_t hdr_len;
3295                 max_per_txd = min(mss << 2, max_per_txd);
3296                 max_txd_pwr = fls(max_per_txd) - 1;
3297
3298                 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
3299                 * points to just header, pull a few bytes of payload from
3300                 * frags into skb->data */
3301                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3302                 if (skb->data_len && (hdr_len == (skb->len - skb->data_len))) {
3303                         switch (adapter->hw.mac_type) {
3304                                 unsigned int pull_size;
3305                         case e1000_82544:
3306                                 /* Make sure we have room to chop off 4 bytes,
3307                                  * and that the end alignment will work out to
3308                                  * this hardware's requirements
3309                                  * NOTE: this is a TSO only workaround
3310                                  * if end byte alignment not correct move us
3311                                  * into the next dword */
3312                                 if ((unsigned long)(skb_tail_pointer(skb) - 1) & 4)
3313                                         break;
3314                                 /* fall through */
3315                         case e1000_82571:
3316                         case e1000_82572:
3317                         case e1000_82573:
3318                         case e1000_ich8lan:
3319                                 pull_size = min((unsigned int)4, skb->data_len);
3320                                 if (!__pskb_pull_tail(skb, pull_size)) {
3321                                         DPRINTK(DRV, ERR,
3322                                                 "__pskb_pull_tail failed.\n");
3323                                         dev_kfree_skb_any(skb);
3324                                         return NETDEV_TX_OK;
3325                                 }
3326                                 len = skb->len - skb->data_len;
3327                                 break;
3328                         default:
3329                                 /* do nothing */
3330                                 break;
3331                         }
3332                 }
3333         }
3334
3335         /* reserve a descriptor for the offload context */
3336         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3337                 count++;
3338         count++;
3339
3340         /* Controller Erratum workaround */
3341         if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3342                 count++;
3343
3344         count += TXD_USE_COUNT(len, max_txd_pwr);
3345
3346         if (adapter->pcix_82544)
3347                 count++;
3348
3349         /* work-around for errata 10 and it applies to all controllers
3350          * in PCI-X mode, so add one more descriptor to the count
3351          */
3352         if (unlikely((adapter->hw.bus_type == e1000_bus_type_pcix) &&
3353                         (len > 2015)))
3354                 count++;
3355
3356         nr_frags = skb_shinfo(skb)->nr_frags;
3357         for (f = 0; f < nr_frags; f++)
3358                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3359                                        max_txd_pwr);
3360         if (adapter->pcix_82544)
3361                 count += nr_frags;
3362
3363
3364         if (adapter->hw.tx_pkt_filtering &&
3365             (adapter->hw.mac_type == e1000_82573))
3366                 e1000_transfer_dhcp_info(adapter, skb);
3367
3368         if (!spin_trylock_irqsave(&tx_ring->tx_lock, flags))
3369                 /* Collision - tell upper layer to requeue */
3370                 return NETDEV_TX_LOCKED;
3371
3372         /* need: count + 2 desc gap to keep tail from touching
3373          * head, otherwise try next time */
3374         if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2))) {
3375                 spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3376                 return NETDEV_TX_BUSY;
3377         }
3378
3379         if (unlikely(adapter->hw.mac_type == e1000_82547)) {
3380                 if (unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
3381                         netif_stop_queue(netdev);
3382                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
3383                         spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3384                         return NETDEV_TX_BUSY;
3385                 }
3386         }
3387
3388         if (unlikely(adapter->vlgrp && vlan_tx_tag_present(skb))) {
3389                 tx_flags |= E1000_TX_FLAGS_VLAN;
3390                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3391         }
3392
3393         first = tx_ring->next_to_use;
3394
3395         tso = e1000_tso(adapter, tx_ring, skb);
3396         if (tso < 0) {
3397                 dev_kfree_skb_any(skb);
3398                 spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3399                 return NETDEV_TX_OK;
3400         }
3401
3402         if (likely(tso)) {
3403                 tx_ring->last_tx_tso = 1;
3404                 tx_flags |= E1000_TX_FLAGS_TSO;
3405         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3406                 tx_flags |= E1000_TX_FLAGS_CSUM;
3407
3408         /* Old method was to assume IPv4 packet by default if TSO was enabled.
3409          * 82571 hardware supports TSO capabilities for IPv6 as well...
3410          * no longer assume, we must. */
3411         if (likely(skb->protocol == htons(ETH_P_IP)))
3412                 tx_flags |= E1000_TX_FLAGS_IPV4;
3413
3414         e1000_tx_queue(adapter, tx_ring, tx_flags,
3415                        e1000_tx_map(adapter, tx_ring, skb, first,
3416                                     max_per_txd, nr_frags, mss));
3417
3418         netdev->trans_start = jiffies;
3419
3420         /* Make sure there is space in the ring for the next send. */
3421         e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3422
3423         spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3424         return NETDEV_TX_OK;
3425 }
3426
3427 /**
3428  * e1000_tx_timeout - Respond to a Tx Hang
3429  * @netdev: network interface device structure
3430  **/
3431
3432 static void
3433 e1000_tx_timeout(struct net_device *netdev)
3434 {
3435         struct e1000_adapter *adapter = netdev_priv(netdev);
3436
3437         /* Do the reset outside of interrupt context */
3438         adapter->tx_timeout_count++;
3439         schedule_work(&adapter->reset_task);
3440 }
3441
3442 static void
3443 e1000_reset_task(struct work_struct *work)
3444 {
3445         struct e1000_adapter *adapter =
3446                 container_of(work, struct e1000_adapter, reset_task);
3447
3448         e1000_reinit_locked(adapter);
3449 }
3450
3451 /**
3452  * e1000_get_stats - Get System Network Statistics
3453  * @netdev: network interface device structure
3454  *
3455  * Returns the address of the device statistics structure.
3456  * The statistics are actually updated from the timer callback.
3457  **/
3458
3459 static struct net_device_stats *
3460 e1000_get_stats(struct net_device *netdev)
3461 {
3462         struct e1000_adapter *adapter = netdev_priv(netdev);
3463
3464         /* only return the current stats */
3465         return &adapter->net_stats;
3466 }
3467
3468 /**
3469  * e1000_change_mtu - Change the Maximum Transfer Unit
3470  * @netdev: network interface device structure
3471  * @new_mtu: new value for maximum frame size
3472  *
3473  * Returns 0 on success, negative on failure
3474  **/
3475
3476 static int
3477 e1000_change_mtu(struct net_device *netdev, int new_mtu)
3478 {
3479         struct e1000_adapter *adapter = netdev_priv(netdev);
3480         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3481         uint16_t eeprom_data = 0;
3482
3483         if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3484             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3485                 DPRINTK(PROBE, ERR, "Invalid MTU setting\n");
3486                 return -EINVAL;
3487         }
3488
3489         /* Adapter-specific max frame size limits. */
3490         switch (adapter->hw.mac_type) {
3491         case e1000_undefined ... e1000_82542_rev2_1:
3492         case e1000_ich8lan:
3493                 if (max_frame > MAXIMUM_ETHERNET_FRAME_SIZE) {
3494                         DPRINTK(PROBE, ERR, "Jumbo Frames not supported.\n");
3495                         return -EINVAL;
3496                 }
3497                 break;
3498         case e1000_82573:
3499                 /* Jumbo Frames not supported if:
3500                  * - this is not an 82573L device
3501                  * - ASPM is enabled in any way (0x1A bits 3:2) */
3502                 e1000_read_eeprom(&adapter->hw, EEPROM_INIT_3GIO_3, 1,
3503                                   &eeprom_data);
3504                 if ((adapter->hw.device_id != E1000_DEV_ID_82573L) ||
3505                     (eeprom_data & EEPROM_WORD1A_ASPM_MASK)) {
3506                         if (max_frame > MAXIMUM_ETHERNET_FRAME_SIZE) {
3507                                 DPRINTK(PROBE, ERR,
3508                                         "Jumbo Frames not supported.\n");
3509                                 return -EINVAL;
3510                         }
3511                         break;
3512                 }
3513                 /* ERT will be enabled later to enable wire speed receives */
3514
3515                 /* fall through to get support */
3516         case e1000_82571:
3517         case e1000_82572:
3518         case e1000_80003es2lan:
3519 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3520                 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3521                         DPRINTK(PROBE, ERR, "MTU > 9216 not supported.\n");
3522                         return -EINVAL;
3523                 }
3524                 break;
3525         default:
3526                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3527                 break;
3528         }
3529
3530         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3531          * means we reserve 2 more, this pushes us to allocate from the next
3532          * larger slab size
3533          * i.e. RXBUFFER_2048 --> size-4096 slab */
3534
3535         if (max_frame <= E1000_RXBUFFER_256)
3536                 adapter->rx_buffer_len = E1000_RXBUFFER_256;
3537         else if (max_frame <= E1000_RXBUFFER_512)
3538                 adapter->rx_buffer_len = E1000_RXBUFFER_512;
3539         else if (max_frame <= E1000_RXBUFFER_1024)
3540                 adapter->rx_buffer_len = E1000_RXBUFFER_1024;
3541         else if (max_frame <= E1000_RXBUFFER_2048)
3542                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3543         else if (max_frame <= E1000_RXBUFFER_4096)
3544                 adapter->rx_buffer_len = E1000_RXBUFFER_4096;
3545         else if (max_frame <= E1000_RXBUFFER_8192)
3546                 adapter->rx_buffer_len = E1000_RXBUFFER_8192;
3547         else if (max_frame <= E1000_RXBUFFER_16384)
3548                 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3549
3550         /* adjust allocation if LPE protects us, and we aren't using SBP */
3551         if (!adapter->hw.tbi_compatibility_on &&
3552             ((max_frame == MAXIMUM_ETHERNET_FRAME_SIZE) ||
3553              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3554                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3555
3556         netdev->mtu = new_mtu;
3557         adapter->hw.max_frame_size = max_frame;
3558
3559         if (netif_running(netdev))
3560                 e1000_reinit_locked(adapter);
3561
3562         return 0;
3563 }
3564
3565 /**
3566  * e1000_update_stats - Update the board statistics counters
3567  * @adapter: board private structure
3568  **/
3569
3570 void
3571 e1000_update_stats(struct e1000_adapter *adapter)
3572 {
3573         struct e1000_hw *hw = &adapter->hw;
3574         struct pci_dev *pdev = adapter->pdev;
3575         unsigned long flags;
3576         uint16_t phy_tmp;
3577
3578 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3579
3580         /*
3581          * Prevent stats update while adapter is being reset, or if the pci
3582          * connection is down.
3583          */
3584         if (adapter->link_speed == 0)
3585                 return;
3586         if (pci_channel_offline(pdev))
3587                 return;
3588
3589         spin_lock_irqsave(&adapter->stats_lock, flags);
3590
3591         /* these counters are modified from e1000_adjust_tbi_stats,
3592          * called from the interrupt context, so they must only
3593          * be written while holding adapter->stats_lock
3594          */
3595
3596         adapter->stats.crcerrs += E1000_READ_REG(hw, CRCERRS);
3597         adapter->stats.gprc += E1000_READ_REG(hw, GPRC);
3598         adapter->stats.gorcl += E1000_READ_REG(hw, GORCL);
3599         adapter->stats.gorch += E1000_READ_REG(hw, GORCH);
3600         adapter->stats.bprc += E1000_READ_REG(hw, BPRC);
3601         adapter->stats.mprc += E1000_READ_REG(hw, MPRC);
3602         adapter->stats.roc += E1000_READ_REG(hw, ROC);
3603
3604         if (adapter->hw.mac_type != e1000_ich8lan) {
3605                 adapter->stats.prc64 += E1000_READ_REG(hw, PRC64);
3606                 adapter->stats.prc127 += E1000_READ_REG(hw, PRC127);
3607                 adapter->stats.prc255 += E1000_READ_REG(hw, PRC255);
3608                 adapter->stats.prc511 += E1000_READ_REG(hw, PRC511);
3609                 adapter->stats.prc1023 += E1000_READ_REG(hw, PRC1023);
3610                 adapter->stats.prc1522 += E1000_READ_REG(hw, PRC1522);
3611         }
3612
3613         adapter->stats.symerrs += E1000_READ_REG(hw, SYMERRS);
3614         adapter->stats.mpc += E1000_READ_REG(hw, MPC);
3615         adapter->stats.scc += E1000_READ_REG(hw, SCC);
3616         adapter->stats.ecol += E1000_READ_REG(hw, ECOL);
3617         adapter->stats.mcc += E1000_READ_REG(hw, MCC);
3618         adapter->stats.latecol += E1000_READ_REG(hw, LATECOL);
3619         adapter->stats.dc += E1000_READ_REG(hw, DC);
3620         adapter->stats.sec += E1000_READ_REG(hw, SEC);
3621         adapter->stats.rlec += E1000_READ_REG(hw, RLEC);
3622         adapter->stats.xonrxc += E1000_READ_REG(hw, XONRXC);
3623         adapter->stats.xontxc += E1000_READ_REG(hw, XONTXC);
3624         adapter->stats.xoffrxc += E1000_READ_REG(hw, XOFFRXC);
3625         adapter->stats.xofftxc += E1000_READ_REG(hw, XOFFTXC);
3626         adapter->stats.fcruc += E1000_READ_REG(hw, FCRUC);
3627         adapter->stats.gptc += E1000_READ_REG(hw, GPTC);
3628         adapter->stats.gotcl += E1000_READ_REG(hw, GOTCL);
3629         adapter->stats.gotch += E1000_READ_REG(hw, GOTCH);
3630         adapter->stats.rnbc += E1000_READ_REG(hw, RNBC);
3631         adapter->stats.ruc += E1000_READ_REG(hw, RUC);
3632         adapter->stats.rfc += E1000_READ_REG(hw, RFC);
3633         adapter->stats.rjc += E1000_READ_REG(hw, RJC);
3634         adapter->stats.torl += E1000_READ_REG(hw, TORL);
3635         adapter->stats.torh += E1000_READ_REG(hw, TORH);
3636         adapter->stats.totl += E1000_READ_REG(hw, TOTL);
3637         adapter->stats.toth += E1000_READ_REG(hw, TOTH);
3638         adapter->stats.tpr += E1000_READ_REG(hw, TPR);
3639
3640         if (adapter->hw.mac_type != e1000_ich8lan) {
3641                 adapter->stats.ptc64 += E1000_READ_REG(hw, PTC64);
3642                 adapter->stats.ptc127 += E1000_READ_REG(hw, PTC127);
3643                 adapter->stats.ptc255 += E1000_READ_REG(hw, PTC255);
3644                 adapter->stats.ptc511 += E1000_READ_REG(hw, PTC511);
3645                 adapter->stats.ptc1023 += E1000_READ_REG(hw, PTC1023);
3646                 adapter->stats.ptc1522 += E1000_READ_REG(hw, PTC1522);
3647         }
3648
3649         adapter->stats.mptc += E1000_READ_REG(hw, MPTC);
3650         adapter->stats.bptc += E1000_READ_REG(hw, BPTC);
3651
3652         /* used for adaptive IFS */
3653
3654         hw->tx_packet_delta = E1000_READ_REG(hw, TPT);
3655         adapter->stats.tpt += hw->tx_packet_delta;
3656         hw->collision_delta = E1000_READ_REG(hw, COLC);
3657         adapter->stats.colc += hw->collision_delta;
3658
3659         if (hw->mac_type >= e1000_82543) {
3660                 adapter->stats.algnerrc += E1000_READ_REG(hw, ALGNERRC);
3661                 adapter->stats.rxerrc += E1000_READ_REG(hw, RXERRC);
3662                 adapter->stats.tncrs += E1000_READ_REG(hw, TNCRS);
3663                 adapter->stats.cexterr += E1000_READ_REG(hw, CEXTERR);
3664                 adapter->stats.tsctc += E1000_READ_REG(hw, TSCTC);
3665                 adapter->stats.tsctfc += E1000_READ_REG(hw, TSCTFC);
3666         }
3667         if (hw->mac_type > e1000_82547_rev_2) {
3668                 adapter->stats.iac += E1000_READ_REG(hw, IAC);
3669                 adapter->stats.icrxoc += E1000_READ_REG(hw, ICRXOC);
3670
3671                 if (adapter->hw.mac_type != e1000_ich8lan) {
3672                         adapter->stats.icrxptc += E1000_READ_REG(hw, ICRXPTC);
3673                         adapter->stats.icrxatc += E1000_READ_REG(hw, ICRXATC);
3674                         adapter->stats.ictxptc += E1000_READ_REG(hw, ICTXPTC);
3675                         adapter->stats.ictxatc += E1000_READ_REG(hw, ICTXATC);
3676                         adapter->stats.ictxqec += E1000_READ_REG(hw, ICTXQEC);
3677                         adapter->stats.ictxqmtc += E1000_READ_REG(hw, ICTXQMTC);
3678                         adapter->stats.icrxdmtc += E1000_READ_REG(hw, ICRXDMTC);
3679                 }
3680         }
3681
3682         /* Fill out the OS statistics structure */
3683         adapter->net_stats.rx_packets = adapter->stats.gprc;
3684         adapter->net_stats.tx_packets = adapter->stats.gptc;
3685         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
3686         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
3687         adapter->net_stats.multicast = adapter->stats.mprc;
3688         adapter->net_stats.collisions = adapter->stats.colc;
3689
3690         /* Rx Errors */
3691
3692         /* RLEC on some newer hardware can be incorrect so build
3693         * our own version based on RUC and ROC */
3694         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3695                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3696                 adapter->stats.ruc + adapter->stats.roc +
3697                 adapter->stats.cexterr;
3698         adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3699         adapter->net_stats.rx_length_errors = adapter->stats.rlerrc;
3700         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3701         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3702         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3703
3704         /* Tx Errors */
3705         adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3706         adapter->net_stats.tx_errors = adapter->stats.txerrc;
3707         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3708         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3709         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3710         if (adapter->hw.bad_tx_carr_stats_fd &&
3711             adapter->link_duplex == FULL_DUPLEX) {
3712                 adapter->net_stats.tx_carrier_errors = 0;
3713                 adapter->stats.tncrs = 0;
3714         }
3715
3716         /* Tx Dropped needs to be maintained elsewhere */
3717
3718         /* Phy Stats */
3719         if (hw->media_type == e1000_media_type_copper) {
3720                 if ((adapter->link_speed == SPEED_1000) &&
3721                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3722                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3723                         adapter->phy_stats.idle_errors += phy_tmp;
3724                 }
3725
3726                 if ((hw->mac_type <= e1000_82546) &&
3727                    (hw->phy_type == e1000_phy_m88) &&
3728                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3729                         adapter->phy_stats.receive_errors += phy_tmp;
3730         }
3731
3732         /* Management Stats */
3733         if (adapter->hw.has_smbus) {
3734                 adapter->stats.mgptc += E1000_READ_REG(hw, MGTPTC);
3735                 adapter->stats.mgprc += E1000_READ_REG(hw, MGTPRC);
3736                 adapter->stats.mgpdc += E1000_READ_REG(hw, MGTPDC);
3737         }
3738
3739         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3740 }
3741
3742 /**
3743  * e1000_intr_msi - Interrupt Handler
3744  * @irq: interrupt number
3745  * @data: pointer to a network interface device structure
3746  **/
3747
3748 static irqreturn_t
3749 e1000_intr_msi(int irq, void *data)
3750 {
3751         struct net_device *netdev = data;
3752         struct e1000_adapter *adapter = netdev_priv(netdev);
3753         struct e1000_hw *hw = &adapter->hw;
3754 #ifndef CONFIG_E1000_NAPI
3755         int i;
3756 #endif
3757         uint32_t icr = E1000_READ_REG(hw, ICR);
3758
3759 #ifdef CONFIG_E1000_NAPI
3760         /* read ICR disables interrupts using IAM, so keep up with our
3761          * enable/disable accounting */
3762         atomic_inc(&adapter->irq_sem);
3763 #endif
3764         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3765                 hw->get_link_status = 1;
3766                 /* 80003ES2LAN workaround-- For packet buffer work-around on
3767                  * link down event; disable receives here in the ISR and reset
3768                  * adapter in watchdog */
3769                 if (netif_carrier_ok(netdev) &&
3770                     (adapter->hw.mac_type == e1000_80003es2lan)) {
3771                         /* disable receives */
3772                         uint32_t rctl = E1000_READ_REG(hw, RCTL);
3773                         E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
3774                 }
3775                 /* guard against interrupt when we're going down */
3776                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3777                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3778         }
3779
3780 #ifdef CONFIG_E1000_NAPI
3781         if (likely(netif_rx_schedule_prep(netdev))) {
3782                 adapter->total_tx_bytes = 0;
3783                 adapter->total_tx_packets = 0;
3784                 adapter->total_rx_bytes = 0;
3785                 adapter->total_rx_packets = 0;
3786                 __netif_rx_schedule(netdev);
3787         } else
3788                 e1000_irq_enable(adapter);
3789 #else
3790         adapter->total_tx_bytes = 0;
3791         adapter->total_rx_bytes = 0;
3792         adapter->total_tx_packets = 0;
3793         adapter->total_rx_packets = 0;
3794
3795         for (i = 0; i < E1000_MAX_INTR; i++)
3796                 if (unlikely(!adapter->clean_rx(adapter, adapter->rx_ring) &
3797                    !e1000_clean_tx_irq(adapter, adapter->tx_ring)))
3798                         break;
3799
3800         if (likely(adapter->itr_setting & 3))
3801                 e1000_set_itr(adapter);
3802 #endif
3803
3804         return IRQ_HANDLED;
3805 }
3806
3807 /**
3808  * e1000_intr - Interrupt Handler
3809  * @irq: interrupt number
3810  * @data: pointer to a network interface device structure
3811  **/
3812
3813 static irqreturn_t
3814 e1000_intr(int irq, void *data)
3815 {
3816         struct net_device *netdev = data;
3817         struct e1000_adapter *adapter = netdev_priv(netdev);
3818         struct e1000_hw *hw = &adapter->hw;
3819         uint32_t rctl, icr = E1000_READ_REG(hw, ICR);
3820 #ifndef CONFIG_E1000_NAPI
3821         int i;
3822 #endif
3823         if (unlikely(!icr))
3824                 return IRQ_NONE;  /* Not our interrupt */
3825
3826 #ifdef CONFIG_E1000_NAPI
3827         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
3828          * not set, then the adapter didn't send an interrupt */
3829         if (unlikely(hw->mac_type >= e1000_82571 &&
3830                      !(icr & E1000_ICR_INT_ASSERTED)))
3831                 return IRQ_NONE;
3832
3833         /* Interrupt Auto-Mask...upon reading ICR,
3834          * interrupts are masked.  No need for the
3835          * IMC write, but it does mean we should
3836          * account for it ASAP. */
3837         if (likely(hw->mac_type >= e1000_82571))
3838                 atomic_inc(&adapter->irq_sem);
3839 #endif
3840
3841         if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3842                 hw->get_link_status = 1;
3843                 /* 80003ES2LAN workaround--
3844                  * For packet buffer work-around on link down event;
3845                  * disable receives here in the ISR and
3846                  * reset adapter in watchdog
3847                  */
3848                 if (netif_carrier_ok(netdev) &&
3849                     (adapter->hw.mac_type == e1000_80003es2lan)) {
3850                         /* disable receives */
3851                         rctl = E1000_READ_REG(hw, RCTL);
3852                         E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
3853                 }
3854                 /* guard against interrupt when we're going down */
3855                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3856                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3857         }
3858
3859 #ifdef CONFIG_E1000_NAPI
3860         if (unlikely(hw->mac_type < e1000_82571)) {
3861                 /* disable interrupts, without the synchronize_irq bit */
3862                 atomic_inc(&adapter->irq_sem);
3863                 E1000_WRITE_REG(hw, IMC, ~0);
3864                 E1000_WRITE_FLUSH(hw);
3865         }
3866         if (likely(netif_rx_schedule_prep(netdev))) {
3867                 adapter->total_tx_bytes = 0;
3868                 adapter->total_tx_packets = 0;
3869                 adapter->total_rx_bytes = 0;
3870                 adapter->total_rx_packets = 0;
3871                 __netif_rx_schedule(netdev);
3872         } else
3873                 /* this really should not happen! if it does it is basically a
3874                  * bug, but not a hard error, so enable ints and continue */
3875                 e1000_irq_enable(adapter);
3876 #else
3877         /* Writing IMC and IMS is needed for 82547.
3878          * Due to Hub Link bus being occupied, an interrupt
3879          * de-assertion message is not able to be sent.
3880          * When an interrupt assertion message is generated later,
3881          * two messages are re-ordered and sent out.
3882          * That causes APIC to think 82547 is in de-assertion
3883          * state, while 82547 is in assertion state, resulting
3884          * in dead lock. Writing IMC forces 82547 into
3885          * de-assertion state.
3886          */
3887         if (hw->mac_type == e1000_82547 || hw->mac_type == e1000_82547_rev_2) {
3888                 atomic_inc(&adapter->irq_sem);
3889                 E1000_WRITE_REG(hw, IMC, ~0);
3890         }
3891
3892         adapter->total_tx_bytes = 0;
3893         adapter->total_rx_bytes = 0;
3894         adapter->total_tx_packets = 0;
3895         adapter->total_rx_packets = 0;
3896
3897         for (i = 0; i < E1000_MAX_INTR; i++)
3898                 if (unlikely(!adapter->clean_rx(adapter, adapter->rx_ring) &
3899                    !e1000_clean_tx_irq(adapter, adapter->tx_ring)))
3900                         break;
3901
3902         if (likely(adapter->itr_setting & 3))
3903                 e1000_set_itr(adapter);
3904
3905         if (hw->mac_type == e1000_82547 || hw->mac_type == e1000_82547_rev_2)
3906                 e1000_irq_enable(adapter);
3907
3908 #endif
3909         return IRQ_HANDLED;
3910 }
3911
3912 #ifdef CONFIG_E1000_NAPI
3913 /**
3914  * e1000_clean - NAPI Rx polling callback
3915  * @adapter: board private structure
3916  **/
3917
3918 static int
3919 e1000_clean(struct net_device *poll_dev, int *budget)
3920 {
3921         struct e1000_adapter *adapter;
3922         int work_to_do = min(*budget, poll_dev->quota);
3923         int tx_cleaned = 0, work_done = 0;
3924
3925         /* Must NOT use netdev_priv macro here. */
3926         adapter = poll_dev->priv;
3927
3928         /* Keep link state information with original netdev */
3929         if (!netif_carrier_ok(poll_dev))
3930                 goto quit_polling;
3931
3932         /* e1000_clean is called per-cpu.  This lock protects
3933          * tx_ring[0] from being cleaned by multiple cpus
3934          * simultaneously.  A failure obtaining the lock means
3935          * tx_ring[0] is currently being cleaned anyway. */
3936         if (spin_trylock(&adapter->tx_queue_lock)) {
3937                 tx_cleaned = e1000_clean_tx_irq(adapter,
3938                                                 &adapter->tx_ring[0]);
3939                 spin_unlock(&adapter->tx_queue_lock);
3940         }
3941
3942         adapter->clean_rx(adapter, &adapter->rx_ring[0],
3943                           &work_done, work_to_do);
3944
3945         *budget -= work_done;
3946         poll_dev->quota -= work_done;
3947
3948         /* If no Tx and not enough Rx work done, exit the polling mode */
3949         if ((!tx_cleaned && (work_done == 0)) ||
3950            !netif_running(poll_dev)) {
3951 quit_polling:
3952                 if (likely(adapter->itr_setting & 3))
3953                         e1000_set_itr(adapter);
3954                 netif_rx_complete(poll_dev);
3955                 e1000_irq_enable(adapter);
3956                 return 0;
3957         }
3958
3959         return 1;
3960 }
3961
3962 #endif
3963 /**
3964  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3965  * @adapter: board private structure
3966  **/
3967
3968 static boolean_t
3969 e1000_clean_tx_irq(struct e1000_adapter *adapter,
3970                    struct e1000_tx_ring *tx_ring)
3971 {
3972         struct net_device *netdev = adapter->netdev;
3973         struct e1000_tx_desc *tx_desc, *eop_desc;
3974         struct e1000_buffer *buffer_info;
3975         unsigned int i, eop;
3976 #ifdef CONFIG_E1000_NAPI
3977         unsigned int count = 0;
3978 #endif
3979         boolean_t cleaned = FALSE;
3980         unsigned int total_tx_bytes=0, total_tx_packets=0;
3981
3982         i = tx_ring->next_to_clean;
3983         eop = tx_ring->buffer_info[i].next_to_watch;
3984         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3985
3986         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
3987                 for (cleaned = FALSE; !cleaned; ) {
3988                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3989                         buffer_info = &tx_ring->buffer_info[i];
3990                         cleaned = (i == eop);
3991
3992                         if (cleaned) {
3993                                 struct sk_buff *skb = buffer_info->skb;
3994                                 unsigned int segs, bytecount;
3995                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3996                                 /* multiply data chunks by size of headers */
3997                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3998                                             skb->len;
3999                                 total_tx_packets += segs;
4000                                 total_tx_bytes += bytecount;
4001                         }
4002                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
4003                         tx_desc->upper.data = 0;
4004
4005                         if (unlikely(++i == tx_ring->count)) i = 0;
4006                 }
4007
4008                 eop = tx_ring->buffer_info[i].next_to_watch;
4009                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
4010 #ifdef CONFIG_E1000_NAPI
4011 #define E1000_TX_WEIGHT 64
4012                 /* weight of a sort for tx, to avoid endless transmit cleanup */
4013                 if (count++ == E1000_TX_WEIGHT) break;
4014 #endif
4015         }
4016
4017         tx_ring->next_to_clean = i;
4018
4019 #define TX_WAKE_THRESHOLD 32
4020         if (unlikely(cleaned && netif_carrier_ok(netdev) &&
4021                      E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
4022                 /* Make sure that anybody stopping the queue after this
4023                  * sees the new next_to_clean.
4024                  */
4025                 smp_mb();
4026                 if (netif_queue_stopped(netdev)) {
4027                         netif_wake_queue(netdev);
4028                         ++adapter->restart_queue;
4029                 }
4030         }
4031
4032         if (adapter->detect_tx_hung) {
4033                 /* Detect a transmit hang in hardware, this serializes the
4034                  * check with the clearing of time_stamp and movement of i */
4035                 adapter->detect_tx_hung = FALSE;
4036                 if (tx_ring->buffer_info[eop].dma &&
4037                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
4038                                (adapter->tx_timeout_factor * HZ))
4039                     && !(E1000_READ_REG(&adapter->hw, STATUS) &
4040                          E1000_STATUS_TXOFF)) {
4041
4042                         /* detected Tx unit hang */
4043                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
4044                                         "  Tx Queue             <%lu>\n"
4045                                         "  TDH                  <%x>\n"
4046                                         "  TDT                  <%x>\n"
4047                                         "  next_to_use          <%x>\n"
4048                                         "  next_to_clean        <%x>\n"
4049                                         "buffer_info[next_to_clean]\n"
4050                                         "  time_stamp           <%lx>\n"
4051                                         "  next_to_watch        <%x>\n"
4052                                         "  jiffies              <%lx>\n"
4053                                         "  next_to_watch.status <%x>\n",
4054                                 (unsigned long)((tx_ring - adapter->tx_ring) /
4055                                         sizeof(struct e1000_tx_ring)),
4056                                 readl(adapter->hw.hw_addr + tx_ring->tdh),
4057                                 readl(adapter->hw.hw_addr + tx_ring->tdt),
4058                                 tx_ring->next_to_use,
4059                                 tx_ring->next_to_clean,
4060                                 tx_ring->buffer_info[eop].time_stamp,
4061                                 eop,
4062                                 jiffies,
4063                                 eop_desc->upper.fields.status);
4064                         netif_stop_queue(netdev);
4065                 }
4066         }
4067         adapter->total_tx_bytes += total_tx_bytes;
4068         adapter->total_tx_packets += total_tx_packets;
4069         return cleaned;
4070 }
4071
4072 /**
4073  * e1000_rx_checksum - Receive Checksum Offload for 82543
4074  * @adapter:     board private structure
4075  * @status_err:  receive descriptor status and error fields
4076  * @csum:        receive descriptor csum field
4077  * @sk_buff:     socket buffer with received data
4078  **/
4079
4080 static void
4081 e1000_rx_checksum(struct e1000_adapter *adapter,
4082                   uint32_t status_err, uint32_t csum,
4083                   struct sk_buff *skb)
4084 {
4085         uint16_t status = (uint16_t)status_err;
4086         uint8_t errors = (uint8_t)(status_err >> 24);
4087         skb->ip_summed = CHECKSUM_NONE;
4088
4089         /* 82543 or newer only */
4090         if (unlikely(adapter->hw.mac_type < e1000_82543)) return;
4091         /* Ignore Checksum bit is set */
4092         if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
4093         /* TCP/UDP checksum error bit is set */
4094         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
4095                 /* let the stack verify checksum errors */
4096                 adapter->hw_csum_err++;
4097                 return;
4098         }
4099         /* TCP/UDP Checksum has not been calculated */
4100         if (adapter->hw.mac_type <= e1000_82547_rev_2) {
4101                 if (!(status & E1000_RXD_STAT_TCPCS))
4102                         return;
4103         } else {
4104                 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
4105                         return;
4106         }
4107         /* It must be a TCP or UDP packet with a valid checksum */
4108         if (likely(status & E1000_RXD_STAT_TCPCS)) {
4109                 /* TCP checksum is good */
4110                 skb->ip_summed = CHECKSUM_UNNECESSARY;
4111         } else if (adapter->hw.mac_type > e1000_82547_rev_2) {
4112                 /* IP fragment with UDP payload */
4113                 /* Hardware complements the payload checksum, so we undo it
4114                  * and then put the value in host order for further stack use.
4115                  */
4116                 csum = ntohl(csum ^ 0xFFFF);
4117                 skb->csum = csum;
4118                 skb->ip_summed = CHECKSUM_COMPLETE;
4119         }
4120         adapter->hw_csum_good++;
4121 }
4122
4123 /**
4124  * e1000_clean_rx_irq - Send received data up the network stack; legacy
4125  * @adapter: board private structure
4126  **/
4127
4128 static boolean_t
4129 #ifdef CONFIG_E1000_NAPI
4130 e1000_clean_rx_irq(struct e1000_adapter *adapter,
4131                    struct e1000_rx_ring *rx_ring,
4132                    int *work_done, int work_to_do)
4133 #else
4134 e1000_clean_rx_irq(struct e1000_adapter *adapter,
4135                    struct e1000_rx_ring *rx_ring)
4136 #endif
4137 {
4138         struct net_device *netdev = adapter->netdev;
4139         struct pci_dev *pdev = adapter->pdev;
4140         struct e1000_rx_desc *rx_desc, *next_rxd;
4141         struct e1000_buffer *buffer_info, *next_buffer;
4142         unsigned long flags;
4143         uint32_t length;
4144         uint8_t last_byte;
4145         unsigned int i;
4146         int cleaned_count = 0;
4147         boolean_t cleaned = FALSE;
4148         unsigned int total_rx_bytes=0, total_rx_packets=0;
4149
4150         i = rx_ring->next_to_clean;
4151         rx_desc = E1000_RX_DESC(*rx_ring, i);
4152         buffer_info = &rx_ring->buffer_info[i];
4153
4154         while (rx_desc->status & E1000_RXD_STAT_DD) {
4155                 struct sk_buff *skb;
4156                 u8 status;
4157
4158 #ifdef CONFIG_E1000_NAPI
4159                 if (*work_done >= work_to_do)
4160                         break;
4161                 (*work_done)++;
4162 #endif
4163                 status = rx_desc->status;
4164                 skb = buffer_info->skb;
4165                 buffer_info->skb = NULL;
4166
4167                 prefetch(skb->data - NET_IP_ALIGN);
4168
4169                 if (++i == rx_ring->count) i = 0;
4170                 next_rxd = E1000_RX_DESC(*rx_ring, i);
4171                 prefetch(next_rxd);
4172
4173                 next_buffer = &rx_ring->buffer_info[i];
4174
4175                 cleaned = TRUE;
4176                 cleaned_count++;
4177                 pci_unmap_single(pdev,
4178                                  buffer_info->dma,
4179                                  buffer_info->length,
4180                                  PCI_DMA_FROMDEVICE);
4181
4182                 length = le16_to_cpu(rx_desc->length);
4183
4184                 if (unlikely(!(status & E1000_RXD_STAT_EOP))) {
4185                         /* All receives must fit into a single buffer */
4186                         E1000_DBG("%s: Receive packet consumed multiple"
4187                                   " buffers\n", netdev->name);
4188                         /* recycle */
4189                         buffer_info->skb = skb;
4190                         goto next_desc;
4191                 }
4192
4193                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
4194                         last_byte = *(skb->data + length - 1);
4195                         if (TBI_ACCEPT(&adapter->hw, status,
4196                                       rx_desc->errors, length, last_byte)) {
4197                                 spin_lock_irqsave(&adapter->stats_lock, flags);
4198                                 e1000_tbi_adjust_stats(&adapter->hw,
4199                                                        &adapter->stats,
4200                                                        length, skb->data);
4201                                 spin_unlock_irqrestore(&adapter->stats_lock,
4202                                                        flags);
4203                                 length--;
4204                         } else {
4205                                 /* recycle */
4206                                 buffer_info->skb = skb;
4207                                 goto next_desc;
4208                         }
4209                 }
4210
4211                 /* adjust length to remove Ethernet CRC, this must be
4212                  * done after the TBI_ACCEPT workaround above */
4213                 length -= 4;
4214
4215                 /* probably a little skewed due to removing CRC */
4216                 total_rx_bytes += length;
4217                 total_rx_packets++;
4218
4219                 /* code added for copybreak, this should improve
4220                  * performance for small packets with large amounts
4221                  * of reassembly being done in the stack */
4222                 if (length < copybreak) {
4223                         struct sk_buff *new_skb =
4224                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
4225                         if (new_skb) {
4226                                 skb_reserve(new_skb, NET_IP_ALIGN);
4227                                 skb_copy_to_linear_data_offset(new_skb,
4228                                                                -NET_IP_ALIGN,
4229                                                                (skb->data -
4230                                                                 NET_IP_ALIGN),
4231                                                                (length +
4232                                                                 NET_IP_ALIGN));
4233                                 /* save the skb in buffer_info as good */
4234                                 buffer_info->skb = skb;
4235                                 skb = new_skb;
4236                         }
4237                         /* else just continue with the old one */
4238                 }
4239                 /* end copybreak code */
4240                 skb_put(skb, length);
4241
4242                 /* Receive Checksum Offload */
4243                 e1000_rx_checksum(adapter,
4244                                   (uint32_t)(status) |
4245                                   ((uint32_t)(rx_desc->errors) << 24),
4246                                   le16_to_cpu(rx_desc->csum), skb);
4247
4248                 skb->protocol = eth_type_trans(skb, netdev);
4249 #ifdef CONFIG_E1000_NAPI
4250                 if (unlikely(adapter->vlgrp &&
4251                             (status & E1000_RXD_STAT_VP))) {
4252                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
4253                                                  le16_to_cpu(rx_desc->special) &
4254                                                  E1000_RXD_SPC_VLAN_MASK);
4255                 } else {
4256                         netif_receive_skb(skb);
4257                 }
4258 #else /* CONFIG_E1000_NAPI */
4259                 if (unlikely(adapter->vlgrp &&
4260                             (status & E1000_RXD_STAT_VP))) {
4261                         vlan_hwaccel_rx(skb, adapter->vlgrp,
4262                                         le16_to_cpu(rx_desc->special) &
4263                                         E1000_RXD_SPC_VLAN_MASK);
4264                 } else {
4265                         netif_rx(skb);
4266                 }
4267 #endif /* CONFIG_E1000_NAPI */
4268                 netdev->last_rx = jiffies;
4269
4270 next_desc:
4271                 rx_desc->status = 0;
4272
4273                 /* return some buffers to hardware, one at a time is too slow */
4274                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4275                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4276                         cleaned_count = 0;
4277                 }
4278
4279                 /* use prefetched values */
4280                 rx_desc = next_rxd;
4281                 buffer_info = next_buffer;
4282         }
4283         rx_ring->next_to_clean = i;
4284
4285         cleaned_count = E1000_DESC_UNUSED(rx_ring);
4286         if (cleaned_count)
4287                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4288
4289         adapter->total_rx_packets += total_rx_packets;
4290         adapter->total_rx_bytes += total_rx_bytes;
4291         return cleaned;
4292 }
4293
4294 /**
4295  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
4296  * @adapter: board private structure
4297  **/
4298
4299 static boolean_t
4300 #ifdef CONFIG_E1000_NAPI
4301 e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
4302                       struct e1000_rx_ring *rx_ring,
4303                       int *work_done, int work_to_do)
4304 #else
4305 e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
4306                       struct e1000_rx_ring *rx_ring)
4307 #endif
4308 {
4309         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
4310         struct net_device *netdev = adapter->netdev;
4311         struct pci_dev *pdev = adapter->pdev;
4312         struct e1000_buffer *buffer_info, *next_buffer;
4313         struct e1000_ps_page *ps_page;
4314         struct e1000_ps_page_dma *ps_page_dma;
4315         struct sk_buff *skb;
4316         unsigned int i, j;
4317         uint32_t length, staterr;
4318         int cleaned_count = 0;
4319         boolean_t cleaned = FALSE;
4320         unsigned int total_rx_bytes=0, total_rx_packets=0;
4321
4322         i = rx_ring->next_to_clean;
4323         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
4324         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
4325         buffer_info = &rx_ring->buffer_info[i];
4326
4327         while (staterr & E1000_RXD_STAT_DD) {
4328                 ps_page = &rx_ring->ps_page[i];
4329                 ps_page_dma = &rx_ring->ps_page_dma[i];
4330 #ifdef CONFIG_E1000_NAPI
4331                 if (unlikely(*work_done >= work_to_do))
4332                         break;
4333                 (*work_done)++;
4334 #endif
4335                 skb = buffer_info->skb;
4336
4337                 /* in the packet split case this is header only */
4338                 prefetch(skb->data - NET_IP_ALIGN);
4339
4340                 if (++i == rx_ring->count) i = 0;
4341                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
4342                 prefetch(next_rxd);
4343
4344                 next_buffer = &rx_ring->buffer_info[i];
4345
4346                 cleaned = TRUE;
4347                 cleaned_count++;
4348                 pci_unmap_single(pdev, buffer_info->dma,
4349                                  buffer_info->length,
4350                                  PCI_DMA_FROMDEVICE);
4351
4352                 if (unlikely(!(staterr & E1000_RXD_STAT_EOP))) {
4353                         E1000_DBG("%s: Packet Split buffers didn't pick up"
4354                                   " the full packet\n", netdev->name);
4355                         dev_kfree_skb_irq(skb);
4356                         goto next_desc;
4357                 }
4358
4359                 if (unlikely(staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK)) {
4360                         dev_kfree_skb_irq(skb);
4361                         goto next_desc;
4362                 }
4363
4364                 length = le16_to_cpu(rx_desc->wb.middle.length0);
4365
4366                 if (unlikely(!length)) {
4367                         E1000_DBG("%s: Last part of the packet spanning"
4368                                   " multiple descriptors\n", netdev->name);
4369                         dev_kfree_skb_irq(skb);
4370                         goto next_desc;
4371                 }
4372
4373                 /* Good Receive */
4374                 skb_put(skb, length);
4375
4376                 {
4377                 /* this looks ugly, but it seems compiler issues make it
4378                    more efficient than reusing j */
4379                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
4380
4381                 /* page alloc/put takes too long and effects small packet
4382                  * throughput, so unsplit small packets and save the alloc/put*/
4383                 if (l1 && (l1 <= copybreak) && ((length + l1) <= adapter->rx_ps_bsize0)) {
4384                         u8 *vaddr;
4385                         /* there is no documentation about how to call
4386                          * kmap_atomic, so we can't hold the mapping
4387                          * very long */
4388                         pci_dma_sync_single_for_cpu(pdev,
4389                                 ps_page_dma->ps_page_dma[0],
4390                                 PAGE_SIZE,
4391                                 PCI_DMA_FROMDEVICE);
4392                         vaddr = kmap_atomic(ps_page->ps_page[0],
4393                                             KM_SKB_DATA_SOFTIRQ);
4394                         memcpy(skb_tail_pointer(skb), vaddr, l1);
4395                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
4396                         pci_dma_sync_single_for_device(pdev,
4397                                 ps_page_dma->ps_page_dma[0],
4398                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
4399                         /* remove the CRC */
4400                         l1 -= 4;
4401                         skb_put(skb, l1);
4402                         goto copydone;
4403                 } /* if */
4404                 }
4405
4406                 for (j = 0; j < adapter->rx_ps_pages; j++) {
4407                         if (!(length= le16_to_cpu(rx_desc->wb.upper.length[j])))
4408                                 break;
4409                         pci_unmap_page(pdev, ps_page_dma->ps_page_dma[j],
4410                                         PAGE_SIZE, PCI_DMA_FROMDEVICE);
4411                         ps_page_dma->ps_page_dma[j] = 0;
4412                         skb_fill_page_desc(skb, j, ps_page->ps_page[j], 0,
4413                                            length);
4414                         ps_page->ps_page[j] = NULL;
4415                         skb->len += length;
4416                         skb->data_len += length;
4417                         skb->truesize += length;
4418                 }
4419
4420                 /* strip the ethernet crc, problem is we're using pages now so
4421                  * this whole operation can get a little cpu intensive */
4422                 pskb_trim(skb, skb->len - 4);
4423
4424 copydone:
4425                 total_rx_bytes += skb->len;
4426                 total_rx_packets++;
4427
4428                 e1000_rx_checksum(adapter, staterr,
4429                                   le16_to_cpu(rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
4430                 skb->protocol = eth_type_trans(skb, netdev);
4431
4432                 if (likely(rx_desc->wb.upper.header_status &
4433                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP)))
4434                         adapter->rx_hdr_split++;
4435 #ifdef CONFIG_E1000_NAPI
4436                 if (unlikely(adapter->vlgrp && (staterr & E1000_RXD_STAT_VP))) {
4437                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
4438                                 le16_to_cpu(rx_desc->wb.middle.vlan) &
4439                                 E1000_RXD_SPC_VLAN_MASK);
4440                 } else {
4441                         netif_receive_skb(skb);
4442                 }
4443 #else /* CONFIG_E1000_NAPI */
4444                 if (unlikely(adapter->vlgrp && (staterr & E1000_RXD_STAT_VP))) {
4445                         vlan_hwaccel_rx(skb, adapter->vlgrp,
4446                                 le16_to_cpu(rx_desc->wb.middle.vlan) &
4447                                 E1000_RXD_SPC_VLAN_MASK);
4448                 } else {
4449                         netif_rx(skb);
4450                 }
4451 #endif /* CONFIG_E1000_NAPI */
4452                 netdev->last_rx = jiffies;
4453
4454 next_desc:
4455                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
4456                 buffer_info->skb = NULL;
4457
4458                 /* return some buffers to hardware, one at a time is too slow */
4459                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4460                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4461                         cleaned_count = 0;
4462                 }
4463
4464                 /* use prefetched values */
4465                 rx_desc = next_rxd;
4466                 buffer_info = next_buffer;
4467
4468                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
4469         }
4470         rx_ring->next_to_clean = i;
4471
4472         cleaned_count = E1000_DESC_UNUSED(rx_ring);
4473         if (cleaned_count)
4474                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4475
4476         adapter->total_rx_packets += total_rx_packets;
4477         adapter->total_rx_bytes += total_rx_bytes;
4478         return cleaned;
4479 }
4480
4481 /**
4482  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4483  * @adapter: address of board private structure
4484  **/
4485
4486 static void
4487 e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4488                        struct e1000_rx_ring *rx_ring,
4489                        int cleaned_count)
4490 {
4491         struct net_device *netdev = adapter->netdev;
4492         struct pci_dev *pdev = adapter->pdev;
4493         struct e1000_rx_desc *rx_desc;
4494         struct e1000_buffer *buffer_info;
4495         struct sk_buff *skb;
4496         unsigned int i;
4497         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
4498
4499         i = rx_ring->next_to_use;
4500         buffer_info = &rx_ring->buffer_info[i];
4501
4502         while (cleaned_count--) {
4503                 skb = buffer_info->skb;
4504                 if (skb) {
4505                         skb_trim(skb, 0);
4506                         goto map_skb;
4507                 }
4508
4509                 skb = netdev_alloc_skb(netdev, bufsz);
4510                 if (unlikely(!skb)) {
4511                         /* Better luck next round */
4512                         adapter->alloc_rx_buff_failed++;
4513                         break;
4514                 }
4515
4516                 /* Fix for errata 23, can't cross 64kB boundary */
4517                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4518                         struct sk_buff *oldskb = skb;
4519                         DPRINTK(RX_ERR, ERR, "skb align check failed: %u bytes "
4520                                              "at %p\n", bufsz, skb->data);
4521                         /* Try again, without freeing the previous */
4522                         skb = netdev_alloc_skb(netdev, bufsz);
4523                         /* Failed allocation, critical failure */
4524                         if (!skb) {
4525                                 dev_kfree_skb(oldskb);
4526                                 break;
4527                         }
4528
4529                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4530                                 /* give up */
4531                                 dev_kfree_skb(skb);
4532                                 dev_kfree_skb(oldskb);
4533                                 break; /* while !buffer_info->skb */
4534                         }
4535
4536                         /* Use new allocation */
4537                         dev_kfree_skb(oldskb);
4538                 }
4539                 /* Make buffer alignment 2 beyond a 16 byte boundary
4540                  * this will result in a 16 byte aligned IP header after
4541                  * the 14 byte MAC header is removed
4542                  */
4543                 skb_reserve(skb, NET_IP_ALIGN);
4544
4545                 buffer_info->skb = skb;
4546                 buffer_info->length = adapter->rx_buffer_len;
4547 map_skb:
4548                 buffer_info->dma = pci_map_single(pdev,
4549                                                   skb->data,
4550                                                   adapter->rx_buffer_len,
4551                                                   PCI_DMA_FROMDEVICE);
4552
4553                 /* Fix for errata 23, can't cross 64kB boundary */
4554                 if (!e1000_check_64k_bound(adapter,
4555                                         (void *)(unsigned long)buffer_info->dma,
4556                                         adapter->rx_buffer_len)) {
4557                         DPRINTK(RX_ERR, ERR,
4558                                 "dma align check failed: %u bytes at %p\n",
4559                                 adapter->rx_buffer_len,
4560                                 (void *)(unsigned long)buffer_info->dma);
4561                         dev_kfree_skb(skb);
4562                         buffer_info->skb = NULL;
4563
4564                         pci_unmap_single(pdev, buffer_info->dma,
4565                                          adapter->rx_buffer_len,
4566                                          PCI_DMA_FROMDEVICE);
4567
4568                         break; /* while !buffer_info->skb */
4569                 }
4570                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4571                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4572
4573                 if (unlikely(++i == rx_ring->count))
4574                         i = 0;
4575                 buffer_info = &rx_ring->buffer_info[i];
4576         }
4577
4578         if (likely(rx_ring->next_to_use != i)) {
4579                 rx_ring->next_to_use = i;
4580                 if (unlikely(i-- == 0))
4581                         i = (rx_ring->count - 1);
4582
4583                 /* Force memory writes to complete before letting h/w
4584                  * know there are new descriptors to fetch.  (Only
4585                  * applicable for weak-ordered memory model archs,
4586                  * such as IA-64). */
4587                 wmb();
4588                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4589         }
4590 }
4591
4592 /**
4593  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
4594  * @adapter: address of board private structure
4595  **/
4596
4597 static void
4598 e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
4599                           struct e1000_rx_ring *rx_ring,
4600                           int cleaned_count)
4601 {
4602         struct net_device *netdev = adapter->netdev;
4603         struct pci_dev *pdev = adapter->pdev;
4604         union e1000_rx_desc_packet_split *rx_desc;
4605         struct e1000_buffer *buffer_info;
4606         struct e1000_ps_page *ps_page;
4607         struct e1000_ps_page_dma *ps_page_dma;
4608         struct sk_buff *skb;
4609         unsigned int i, j;
4610
4611         i = rx_ring->next_to_use;
4612         buffer_info = &rx_ring->buffer_info[i];
4613         ps_page = &rx_ring->ps_page[i];
4614         ps_page_dma = &rx_ring->ps_page_dma[i];
4615
4616         while (cleaned_count--) {
4617                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
4618
4619                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
4620                         if (j < adapter->rx_ps_pages) {
4621                                 if (likely(!ps_page->ps_page[j])) {
4622                                         ps_page->ps_page[j] =
4623                                                 alloc_page(GFP_ATOMIC);
4624                                         if (unlikely(!ps_page->ps_page[j])) {
4625                                                 adapter->alloc_rx_buff_failed++;
4626                                                 goto no_buffers;
4627                                         }
4628                                         ps_page_dma->ps_page_dma[j] =
4629                                                 pci_map_page(pdev,
4630                                                             ps_page->ps_page[j],
4631                                                             0, PAGE_SIZE,
4632                                                             PCI_DMA_FROMDEVICE);
4633                                 }
4634                                 /* Refresh the desc even if buffer_addrs didn't
4635                                  * change because each write-back erases
4636                                  * this info.
4637                                  */
4638                                 rx_desc->read.buffer_addr[j+1] =
4639                                      cpu_to_le64(ps_page_dma->ps_page_dma[j]);
4640                         } else
4641                                 rx_desc->read.buffer_addr[j+1] = ~0;
4642                 }
4643
4644                 skb = netdev_alloc_skb(netdev,
4645                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
4646
4647                 if (unlikely(!skb)) {
4648                         adapter->alloc_rx_buff_failed++;
4649                         break;
4650                 }
4651
4652                 /* Make buffer alignment 2 beyond a 16 byte boundary
4653                  * this will result in a 16 byte aligned IP header after
4654                  * the 14 byte MAC header is removed
4655                  */
4656                 skb_reserve(skb, NET_IP_ALIGN);
4657
4658                 buffer_info->skb = skb;
4659                 buffer_info->length = adapter->rx_ps_bsize0;
4660                 buffer_info->dma = pci_map_single(pdev, skb->data,
4661                                                   adapter->rx_ps_bsize0,
4662                                                   PCI_DMA_FROMDEVICE);
4663
4664                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
4665
4666                 if (unlikely(++i == rx_ring->count)) i = 0;
4667                 buffer_info = &rx_ring->buffer_info[i];
4668                 ps_page = &rx_ring->ps_page[i];
4669                 ps_page_dma = &rx_ring->ps_page_dma[i];
4670         }
4671
4672 no_buffers:
4673         if (likely(rx_ring->next_to_use != i)) {
4674                 rx_ring->next_to_use = i;
4675                 if (unlikely(i-- == 0)) i = (rx_ring->count - 1);
4676
4677                 /* Force memory writes to complete before letting h/w
4678                  * know there are new descriptors to fetch.  (Only
4679                  * applicable for weak-ordered memory model archs,
4680                  * such as IA-64). */
4681                 wmb();
4682                 /* Hardware increments by 16 bytes, but packet split
4683                  * descriptors are 32 bytes...so we increment tail
4684                  * twice as much.
4685                  */
4686                 writel(i<<1, adapter->hw.hw_addr + rx_ring->rdt);
4687         }
4688 }
4689
4690 /**
4691  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4692  * @adapter:
4693  **/
4694
4695 static void
4696 e1000_smartspeed(struct e1000_adapter *adapter)
4697 {
4698         uint16_t phy_status;
4699         uint16_t phy_ctrl;
4700
4701         if ((adapter->hw.phy_type != e1000_phy_igp) || !adapter->hw.autoneg ||
4702            !(adapter->hw.autoneg_advertised & ADVERTISE_1000_FULL))
4703                 return;
4704
4705         if (adapter->smartspeed == 0) {
4706                 /* If Master/Slave config fault is asserted twice,
4707                  * we assume back-to-back */
4708                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_status);
4709                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
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_CTRL, &phy_ctrl);
4713                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4714                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4715                         e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL,
4716                                             phy_ctrl);
4717                         adapter->smartspeed++;
4718                         if (!e1000_phy_setup_autoneg(&adapter->hw) &&
4719                            !e1000_read_phy_reg(&adapter->hw, PHY_CTRL,
4720                                                &phy_ctrl)) {
4721                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4722                                              MII_CR_RESTART_AUTO_NEG);
4723                                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL,
4724                                                     phy_ctrl);
4725                         }
4726                 }
4727                 return;
4728         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4729                 /* If still no link, perhaps using 2/3 pair cable */
4730                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_ctrl);
4731                 phy_ctrl |= CR_1000T_MS_ENABLE;
4732                 e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL, phy_ctrl);
4733                 if (!e1000_phy_setup_autoneg(&adapter->hw) &&
4734                    !e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &phy_ctrl)) {
4735                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4736                                      MII_CR_RESTART_AUTO_NEG);
4737                         e1000_write_phy_reg(&adapter->hw, PHY_CTRL, phy_ctrl);
4738                 }
4739         }
4740         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4741         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4742                 adapter->smartspeed = 0;
4743 }
4744
4745 /**
4746  * e1000_ioctl -
4747  * @netdev:
4748  * @ifreq:
4749  * @cmd:
4750  **/
4751
4752 static int
4753 e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4754 {
4755         switch (cmd) {
4756         case SIOCGMIIPHY:
4757         case SIOCGMIIREG:
4758         case SIOCSMIIREG:
4759                 return e1000_mii_ioctl(netdev, ifr, cmd);
4760         default:
4761                 return -EOPNOTSUPP;
4762         }
4763 }
4764
4765 /**
4766  * e1000_mii_ioctl -
4767  * @netdev:
4768  * @ifreq:
4769  * @cmd:
4770  **/
4771
4772 static int
4773 e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4774 {
4775         struct e1000_adapter *adapter = netdev_priv(netdev);
4776         struct mii_ioctl_data *data = if_mii(ifr);
4777         int retval;
4778         uint16_t mii_reg;
4779         uint16_t spddplx;
4780         unsigned long flags;
4781
4782         if (adapter->hw.media_type != e1000_media_type_copper)
4783                 return -EOPNOTSUPP;
4784
4785         switch (cmd) {
4786         case SIOCGMIIPHY:
4787                 data->phy_id = adapter->hw.phy_addr;
4788                 break;
4789         case SIOCGMIIREG:
4790                 if (!capable(CAP_NET_ADMIN))
4791                         return -EPERM;
4792                 spin_lock_irqsave(&adapter->stats_lock, flags);
4793                 if (e1000_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
4794                                    &data->val_out)) {
4795                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4796                         return -EIO;
4797                 }
4798                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4799                 break;
4800         case SIOCSMIIREG:
4801                 if (!capable(CAP_NET_ADMIN))
4802                         return -EPERM;
4803                 if (data->reg_num & ~(0x1F))
4804                         return -EFAULT;
4805                 mii_reg = data->val_in;
4806                 spin_lock_irqsave(&adapter->stats_lock, flags);
4807                 if (e1000_write_phy_reg(&adapter->hw, data->reg_num,
4808                                         mii_reg)) {
4809                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4810                         return -EIO;
4811                 }
4812                 if (adapter->hw.media_type == e1000_media_type_copper) {
4813                         switch (data->reg_num) {
4814                         case PHY_CTRL:
4815                                 if (mii_reg & MII_CR_POWER_DOWN)
4816                                         break;
4817                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4818                                         adapter->hw.autoneg = 1;
4819                                         adapter->hw.autoneg_advertised = 0x2F;
4820                                 } else {
4821                                         if (mii_reg & 0x40)
4822                                                 spddplx = SPEED_1000;
4823                                         else if (mii_reg & 0x2000)
4824                                                 spddplx = SPEED_100;
4825                                         else
4826                                                 spddplx = SPEED_10;
4827                                         spddplx += (mii_reg & 0x100)
4828                                                    ? DUPLEX_FULL :
4829                                                    DUPLEX_HALF;
4830                                         retval = e1000_set_spd_dplx(adapter,
4831                                                                     spddplx);
4832                                         if (retval) {
4833                                                 spin_unlock_irqrestore(
4834                                                         &adapter->stats_lock,
4835                                                         flags);
4836                                                 return retval;
4837                                         }
4838                                 }
4839                                 if (netif_running(adapter->netdev))
4840                                         e1000_reinit_locked(adapter);
4841                                 else
4842                                         e1000_reset(adapter);
4843                                 break;
4844                         case M88E1000_PHY_SPEC_CTRL:
4845                         case M88E1000_EXT_PHY_SPEC_CTRL:
4846                                 if (e1000_phy_reset(&adapter->hw)) {
4847                                         spin_unlock_irqrestore(
4848                                                 &adapter->stats_lock, flags);
4849                                         return -EIO;
4850                                 }
4851                                 break;
4852                         }
4853                 } else {
4854                         switch (data->reg_num) {
4855                         case PHY_CTRL:
4856                                 if (mii_reg & MII_CR_POWER_DOWN)
4857                                         break;
4858                                 if (netif_running(adapter->netdev))
4859                                         e1000_reinit_locked(adapter);
4860                                 else
4861                                         e1000_reset(adapter);
4862                                 break;
4863                         }
4864                 }
4865                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4866                 break;
4867         default:
4868                 return -EOPNOTSUPP;
4869         }
4870         return E1000_SUCCESS;
4871 }
4872
4873 void
4874 e1000_pci_set_mwi(struct e1000_hw *hw)
4875 {
4876         struct e1000_adapter *adapter = hw->back;
4877         int ret_val = pci_set_mwi(adapter->pdev);
4878
4879         if (ret_val)
4880                 DPRINTK(PROBE, ERR, "Error in setting MWI\n");
4881 }
4882
4883 void
4884 e1000_pci_clear_mwi(struct e1000_hw *hw)
4885 {
4886         struct e1000_adapter *adapter = hw->back;
4887
4888         pci_clear_mwi(adapter->pdev);
4889 }
4890
4891 void
4892 e1000_read_pci_cfg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4893 {
4894         struct e1000_adapter *adapter = hw->back;
4895
4896         pci_read_config_word(adapter->pdev, reg, value);
4897 }
4898
4899 void
4900 e1000_write_pci_cfg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4901 {
4902         struct e1000_adapter *adapter = hw->back;
4903
4904         pci_write_config_word(adapter->pdev, reg, *value);
4905 }
4906
4907 int32_t
4908 e1000_read_pcie_cap_reg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4909 {
4910     struct e1000_adapter *adapter = hw->back;
4911     uint16_t cap_offset;
4912
4913     cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
4914     if (!cap_offset)
4915         return -E1000_ERR_CONFIG;
4916
4917     pci_read_config_word(adapter->pdev, cap_offset + reg, value);
4918
4919     return E1000_SUCCESS;
4920 }
4921
4922 void
4923 e1000_io_write(struct e1000_hw *hw, unsigned long port, uint32_t value)
4924 {
4925         outl(value, port);
4926 }
4927
4928 static void
4929 e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
4930 {
4931         struct e1000_adapter *adapter = netdev_priv(netdev);
4932         uint32_t ctrl, rctl;
4933
4934         e1000_irq_disable(adapter);
4935         adapter->vlgrp = grp;
4936
4937         if (grp) {
4938                 /* enable VLAN tag insert/strip */
4939                 ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4940                 ctrl |= E1000_CTRL_VME;
4941                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4942
4943                 if (adapter->hw.mac_type != e1000_ich8lan) {
4944                         /* enable VLAN receive filtering */
4945                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
4946                         rctl |= E1000_RCTL_VFE;
4947                         rctl &= ~E1000_RCTL_CFIEN;
4948                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4949                         e1000_update_mng_vlan(adapter);
4950                 }
4951         } else {
4952                 /* disable VLAN tag insert/strip */
4953                 ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4954                 ctrl &= ~E1000_CTRL_VME;
4955                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4956
4957                 if (adapter->hw.mac_type != e1000_ich8lan) {
4958                         /* disable VLAN filtering */
4959                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
4960                         rctl &= ~E1000_RCTL_VFE;
4961                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4962                         if (adapter->mng_vlan_id !=
4963                             (uint16_t)E1000_MNG_VLAN_NONE) {
4964                                 e1000_vlan_rx_kill_vid(netdev,
4965                                                        adapter->mng_vlan_id);
4966                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4967                         }
4968                 }
4969         }
4970
4971         e1000_irq_enable(adapter);
4972 }
4973
4974 static void
4975 e1000_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid)
4976 {
4977         struct e1000_adapter *adapter = netdev_priv(netdev);
4978         uint32_t vfta, index;
4979
4980         if ((adapter->hw.mng_cookie.status &
4981              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4982             (vid == adapter->mng_vlan_id))
4983                 return;
4984         /* add VID to filter table */
4985         index = (vid >> 5) & 0x7F;
4986         vfta = E1000_READ_REG_ARRAY(&adapter->hw, VFTA, index);
4987         vfta |= (1 << (vid & 0x1F));
4988         e1000_write_vfta(&adapter->hw, index, vfta);
4989 }
4990
4991 static void
4992 e1000_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid)
4993 {
4994         struct e1000_adapter *adapter = netdev_priv(netdev);
4995         uint32_t vfta, index;
4996
4997         e1000_irq_disable(adapter);
4998         vlan_group_set_device(adapter->vlgrp, vid, NULL);
4999         e1000_irq_enable(adapter);
5000
5001         if ((adapter->hw.mng_cookie.status &
5002              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
5003             (vid == adapter->mng_vlan_id)) {
5004                 /* release control to f/w */
5005                 e1000_release_hw_control(adapter);
5006                 return;
5007         }
5008
5009         /* remove VID from filter table */
5010         index = (vid >> 5) & 0x7F;
5011         vfta = E1000_READ_REG_ARRAY(&adapter->hw, VFTA, index);
5012         vfta &= ~(1 << (vid & 0x1F));
5013         e1000_write_vfta(&adapter->hw, index, vfta);
5014 }
5015
5016 static void
5017 e1000_restore_vlan(struct e1000_adapter *adapter)
5018 {
5019         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
5020
5021         if (adapter->vlgrp) {
5022                 uint16_t vid;
5023                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
5024                         if (!vlan_group_get_device(adapter->vlgrp, vid))
5025                                 continue;
5026                         e1000_vlan_rx_add_vid(adapter->netdev, vid);
5027                 }
5028         }
5029 }
5030
5031 int
5032 e1000_set_spd_dplx(struct e1000_adapter *adapter, uint16_t spddplx)
5033 {
5034         adapter->hw.autoneg = 0;
5035
5036         /* Fiber NICs only allow 1000 gbps Full duplex */
5037         if ((adapter->hw.media_type == e1000_media_type_fiber) &&
5038                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
5039                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
5040                 return -EINVAL;
5041         }
5042
5043         switch (spddplx) {
5044         case SPEED_10 + DUPLEX_HALF:
5045                 adapter->hw.forced_speed_duplex = e1000_10_half;
5046                 break;
5047         case SPEED_10 + DUPLEX_FULL:
5048                 adapter->hw.forced_speed_duplex = e1000_10_full;
5049                 break;
5050         case SPEED_100 + DUPLEX_HALF:
5051                 adapter->hw.forced_speed_duplex = e1000_100_half;
5052                 break;
5053         case SPEED_100 + DUPLEX_FULL:
5054                 adapter->hw.forced_speed_duplex = e1000_100_full;
5055                 break;
5056         case SPEED_1000 + DUPLEX_FULL:
5057                 adapter->hw.autoneg = 1;
5058                 adapter->hw.autoneg_advertised = ADVERTISE_1000_FULL;
5059                 break;
5060         case SPEED_1000 + DUPLEX_HALF: /* not supported */
5061         default:
5062                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
5063                 return -EINVAL;
5064         }
5065         return 0;
5066 }
5067
5068 static int
5069 e1000_suspend(struct pci_dev *pdev, pm_message_t state)
5070 {
5071         struct net_device *netdev = pci_get_drvdata(pdev);
5072         struct e1000_adapter *adapter = netdev_priv(netdev);
5073         uint32_t ctrl, ctrl_ext, rctl, status;
5074         uint32_t wufc = adapter->wol;
5075 #ifdef CONFIG_PM
5076         int retval = 0;
5077 #endif
5078
5079         netif_device_detach(netdev);
5080
5081         if (netif_running(netdev)) {
5082                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
5083                 e1000_down(adapter);
5084         }
5085
5086 #ifdef CONFIG_PM
5087         retval = pci_save_state(pdev);
5088         if (retval)
5089                 return retval;
5090 #endif
5091
5092         status = E1000_READ_REG(&adapter->hw, STATUS);
5093         if (status & E1000_STATUS_LU)
5094                 wufc &= ~E1000_WUFC_LNKC;
5095
5096         if (wufc) {
5097                 e1000_setup_rctl(adapter);
5098                 e1000_set_multi(netdev);
5099
5100                 /* turn on all-multi mode if wake on multicast is enabled */
5101                 if (wufc & E1000_WUFC_MC) {
5102                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
5103                         rctl |= E1000_RCTL_MPE;
5104                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
5105                 }
5106
5107                 if (adapter->hw.mac_type >= e1000_82540) {
5108                         ctrl = E1000_READ_REG(&adapter->hw, CTRL);
5109                         /* advertise wake from D3Cold */
5110                         #define E1000_CTRL_ADVD3WUC 0x00100000
5111                         /* phy power management enable */
5112                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5113                         ctrl |= E1000_CTRL_ADVD3WUC |
5114                                 E1000_CTRL_EN_PHY_PWR_MGMT;
5115                         E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
5116                 }
5117
5118                 if (adapter->hw.media_type == e1000_media_type_fiber ||
5119                    adapter->hw.media_type == e1000_media_type_internal_serdes) {
5120                         /* keep the laser running in D3 */
5121                         ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
5122                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
5123                         E1000_WRITE_REG(&adapter->hw, CTRL_EXT, ctrl_ext);
5124                 }
5125
5126                 /* Allow time for pending master requests to run */
5127                 e1000_disable_pciex_master(&adapter->hw);
5128
5129                 E1000_WRITE_REG(&adapter->hw, WUC, E1000_WUC_PME_EN);
5130                 E1000_WRITE_REG(&adapter->hw, WUFC, wufc);
5131                 pci_enable_wake(pdev, PCI_D3hot, 1);
5132                 pci_enable_wake(pdev, PCI_D3cold, 1);
5133         } else {
5134                 E1000_WRITE_REG(&adapter->hw, WUC, 0);
5135                 E1000_WRITE_REG(&adapter->hw, WUFC, 0);
5136                 pci_enable_wake(pdev, PCI_D3hot, 0);
5137                 pci_enable_wake(pdev, PCI_D3cold, 0);
5138         }
5139
5140         e1000_release_manageability(adapter);
5141
5142         /* make sure adapter isn't asleep if manageability is enabled */
5143         if (adapter->en_mng_pt) {
5144                 pci_enable_wake(pdev, PCI_D3hot, 1);
5145                 pci_enable_wake(pdev, PCI_D3cold, 1);
5146         }
5147
5148         if (adapter->hw.phy_type == e1000_phy_igp_3)
5149                 e1000_phy_powerdown_workaround(&adapter->hw);
5150
5151         if (netif_running(netdev))
5152                 e1000_free_irq(adapter);
5153
5154         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
5155          * would have already happened in close and is redundant. */
5156         e1000_release_hw_control(adapter);
5157
5158         pci_disable_device(pdev);
5159
5160         pci_set_power_state(pdev, pci_choose_state(pdev, state));
5161
5162         return 0;
5163 }
5164
5165 #ifdef CONFIG_PM
5166 static int
5167 e1000_resume(struct pci_dev *pdev)
5168 {
5169         struct net_device *netdev = pci_get_drvdata(pdev);
5170         struct e1000_adapter *adapter = netdev_priv(netdev);
5171         uint32_t err;
5172
5173         pci_set_power_state(pdev, PCI_D0);
5174         pci_restore_state(pdev);
5175         if ((err = pci_enable_device(pdev))) {
5176                 printk(KERN_ERR "e1000: Cannot enable PCI device from suspend\n");
5177                 return err;
5178         }
5179         pci_set_master(pdev);
5180
5181         pci_enable_wake(pdev, PCI_D3hot, 0);
5182         pci_enable_wake(pdev, PCI_D3cold, 0);
5183
5184         if (netif_running(netdev) && (err = e1000_request_irq(adapter)))
5185                 return err;
5186
5187         e1000_power_up_phy(adapter);
5188         e1000_reset(adapter);
5189         E1000_WRITE_REG(&adapter->hw, WUS, ~0);
5190
5191         e1000_init_manageability(adapter);
5192
5193         if (netif_running(netdev))
5194                 e1000_up(adapter);
5195
5196         netif_device_attach(netdev);
5197
5198         /* If the controller is 82573 and f/w is AMT, do not set
5199          * DRV_LOAD until the interface is up.  For all other cases,
5200          * let the f/w know that the h/w is now under the control
5201          * of the driver. */
5202         if (adapter->hw.mac_type != e1000_82573 ||
5203             !e1000_check_mng_mode(&adapter->hw))
5204                 e1000_get_hw_control(adapter);
5205
5206         return 0;
5207 }
5208 #endif
5209
5210 static void e1000_shutdown(struct pci_dev *pdev)
5211 {
5212         e1000_suspend(pdev, PMSG_SUSPEND);
5213 }
5214
5215 #ifdef CONFIG_NET_POLL_CONTROLLER
5216 /*
5217  * Polling 'interrupt' - used by things like netconsole to send skbs
5218  * without having to re-enable interrupts. It's not called while
5219  * the interrupt routine is executing.
5220  */
5221 static void
5222 e1000_netpoll(struct net_device *netdev)
5223 {
5224         struct e1000_adapter *adapter = netdev_priv(netdev);
5225
5226         disable_irq(adapter->pdev->irq);
5227         e1000_intr(adapter->pdev->irq, netdev);
5228         e1000_clean_tx_irq(adapter, adapter->tx_ring);
5229 #ifndef CONFIG_E1000_NAPI
5230         adapter->clean_rx(adapter, adapter->rx_ring);
5231 #endif
5232         enable_irq(adapter->pdev->irq);
5233 }
5234 #endif
5235
5236 /**
5237  * e1000_io_error_detected - called when PCI error is detected
5238  * @pdev: Pointer to PCI device
5239  * @state: The current pci conneection state
5240  *
5241  * This function is called after a PCI bus error affecting
5242  * this device has been detected.
5243  */
5244 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
5245 {
5246         struct net_device *netdev = pci_get_drvdata(pdev);
5247         struct e1000_adapter *adapter = netdev->priv;
5248
5249         netif_device_detach(netdev);
5250
5251         if (netif_running(netdev))
5252                 e1000_down(adapter);
5253         pci_disable_device(pdev);
5254
5255         /* Request a slot slot reset. */
5256         return PCI_ERS_RESULT_NEED_RESET;
5257 }
5258
5259 /**
5260  * e1000_io_slot_reset - called after the pci bus has been reset.
5261  * @pdev: Pointer to PCI device
5262  *
5263  * Restart the card from scratch, as if from a cold-boot. Implementation
5264  * resembles the first-half of the e1000_resume routine.
5265  */
5266 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5267 {
5268         struct net_device *netdev = pci_get_drvdata(pdev);
5269         struct e1000_adapter *adapter = netdev->priv;
5270
5271         if (pci_enable_device(pdev)) {
5272                 printk(KERN_ERR "e1000: Cannot re-enable PCI device after reset.\n");
5273                 return PCI_ERS_RESULT_DISCONNECT;
5274         }
5275         pci_set_master(pdev);
5276
5277         pci_enable_wake(pdev, PCI_D3hot, 0);
5278         pci_enable_wake(pdev, PCI_D3cold, 0);
5279
5280         e1000_reset(adapter);
5281         E1000_WRITE_REG(&adapter->hw, WUS, ~0);
5282
5283         return PCI_ERS_RESULT_RECOVERED;
5284 }
5285
5286 /**
5287  * e1000_io_resume - called when traffic can start flowing again.
5288  * @pdev: Pointer to PCI device
5289  *
5290  * This callback is called when the error recovery driver tells us that
5291  * its OK to resume normal operation. Implementation resembles the
5292  * second-half of the e1000_resume routine.
5293  */
5294 static void e1000_io_resume(struct pci_dev *pdev)
5295 {
5296         struct net_device *netdev = pci_get_drvdata(pdev);
5297         struct e1000_adapter *adapter = netdev->priv;
5298
5299         e1000_init_manageability(adapter);
5300
5301         if (netif_running(netdev)) {
5302                 if (e1000_up(adapter)) {
5303                         printk("e1000: can't bring device back up after reset\n");
5304                         return;
5305                 }
5306         }
5307
5308         netif_device_attach(netdev);
5309
5310         /* If the controller is 82573 and f/w is AMT, do not set
5311          * DRV_LOAD until the interface is up.  For all other cases,
5312          * let the f/w know that the h/w is now under the control
5313          * of the driver. */
5314         if (adapter->hw.mac_type != e1000_82573 ||
5315             !e1000_check_mng_mode(&adapter->hw))
5316                 e1000_get_hw_control(adapter);
5317
5318 }
5319
5320 /* e1000_main.c */