[NET]: Fix TX timeout regression in Intel drivers.
[linux-2.6.git] / drivers / net / e1000 / e1000_main.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2006 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include "e1000.h"
30 #include <net/ip6_checksum.h>
31
32 char e1000_driver_name[] = "e1000";
33 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
34 #ifndef CONFIG_E1000_NAPI
35 #define DRIVERNAPI
36 #else
37 #define DRIVERNAPI "-NAPI"
38 #endif
39 #define DRV_VERSION "7.3.20-k2"DRIVERNAPI
40 const char e1000_driver_version[] = DRV_VERSION;
41 static const 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(0x10A5),
104         INTEL_E1000_ETHERNET_DEVICE(0x10B5),
105         INTEL_E1000_ETHERNET_DEVICE(0x10B9),
106         INTEL_E1000_ETHERNET_DEVICE(0x10BA),
107         INTEL_E1000_ETHERNET_DEVICE(0x10BB),
108         INTEL_E1000_ETHERNET_DEVICE(0x10BC),
109         INTEL_E1000_ETHERNET_DEVICE(0x10C4),
110         INTEL_E1000_ETHERNET_DEVICE(0x10C5),
111         INTEL_E1000_ETHERNET_DEVICE(0x10D5),
112         INTEL_E1000_ETHERNET_DEVICE(0x10D9),
113         INTEL_E1000_ETHERNET_DEVICE(0x10DA),
114         /* required last entry */
115         {0,}
116 };
117
118 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
119
120 int e1000_up(struct e1000_adapter *adapter);
121 void e1000_down(struct e1000_adapter *adapter);
122 void e1000_reinit_locked(struct e1000_adapter *adapter);
123 void e1000_reset(struct e1000_adapter *adapter);
124 int e1000_set_spd_dplx(struct e1000_adapter *adapter, uint16_t spddplx);
125 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
126 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
127 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
128 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
129 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
130                              struct e1000_tx_ring *txdr);
131 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
132                              struct e1000_rx_ring *rxdr);
133 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
134                              struct e1000_tx_ring *tx_ring);
135 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
136                              struct e1000_rx_ring *rx_ring);
137 void e1000_update_stats(struct e1000_adapter *adapter);
138
139 static int e1000_init_module(void);
140 static void e1000_exit_module(void);
141 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
142 static void __devexit e1000_remove(struct pci_dev *pdev);
143 static int e1000_alloc_queues(struct e1000_adapter *adapter);
144 static int e1000_sw_init(struct e1000_adapter *adapter);
145 static int e1000_open(struct net_device *netdev);
146 static int e1000_close(struct net_device *netdev);
147 static void e1000_configure_tx(struct e1000_adapter *adapter);
148 static void e1000_configure_rx(struct e1000_adapter *adapter);
149 static void e1000_setup_rctl(struct e1000_adapter *adapter);
150 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
151 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
152 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
153                                 struct e1000_tx_ring *tx_ring);
154 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
155                                 struct e1000_rx_ring *rx_ring);
156 static void e1000_set_multi(struct net_device *netdev);
157 static void e1000_update_phy_info(unsigned long data);
158 static void e1000_watchdog(unsigned long data);
159 static void e1000_82547_tx_fifo_stall(unsigned long data);
160 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
161 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
162 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
163 static int e1000_set_mac(struct net_device *netdev, void *p);
164 static irqreturn_t e1000_intr(int irq, void *data);
165 static irqreturn_t e1000_intr_msi(int irq, void *data);
166 static boolean_t e1000_clean_tx_irq(struct e1000_adapter *adapter,
167                                     struct e1000_tx_ring *tx_ring);
168 #ifdef CONFIG_E1000_NAPI
169 static int e1000_clean(struct napi_struct *napi, int budget);
170 static boolean_t e1000_clean_rx_irq(struct e1000_adapter *adapter,
171                                     struct e1000_rx_ring *rx_ring,
172                                     int *work_done, int work_to_do);
173 static boolean_t e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
174                                        struct e1000_rx_ring *rx_ring,
175                                        int *work_done, int work_to_do);
176 #else
177 static boolean_t e1000_clean_rx_irq(struct e1000_adapter *adapter,
178                                     struct e1000_rx_ring *rx_ring);
179 static boolean_t e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
180                                        struct e1000_rx_ring *rx_ring);
181 #endif
182 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
183                                    struct e1000_rx_ring *rx_ring,
184                                    int cleaned_count);
185 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
186                                       struct e1000_rx_ring *rx_ring,
187                                       int cleaned_count);
188 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
189 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
190                            int cmd);
191 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
192 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
193 static void e1000_tx_timeout(struct net_device *dev);
194 static void e1000_reset_task(struct work_struct *work);
195 static void e1000_smartspeed(struct e1000_adapter *adapter);
196 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
197                                        struct sk_buff *skb);
198
199 static void e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp);
200 static void e1000_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid);
201 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid);
202 static void e1000_restore_vlan(struct e1000_adapter *adapter);
203
204 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
205 #ifdef CONFIG_PM
206 static int e1000_resume(struct pci_dev *pdev);
207 #endif
208 static void e1000_shutdown(struct pci_dev *pdev);
209
210 #ifdef CONFIG_NET_POLL_CONTROLLER
211 /* for netdump / net console */
212 static void e1000_netpoll (struct net_device *netdev);
213 #endif
214
215 #define COPYBREAK_DEFAULT 256
216 static unsigned int copybreak __read_mostly = COPYBREAK_DEFAULT;
217 module_param(copybreak, uint, 0644);
218 MODULE_PARM_DESC(copybreak,
219         "Maximum size of packet that is copied to a new buffer on receive");
220
221 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
222                      pci_channel_state_t state);
223 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
224 static void e1000_io_resume(struct pci_dev *pdev);
225
226 static struct pci_error_handlers e1000_err_handler = {
227         .error_detected = e1000_io_error_detected,
228         .slot_reset = e1000_io_slot_reset,
229         .resume = e1000_io_resume,
230 };
231
232 static struct pci_driver e1000_driver = {
233         .name     = e1000_driver_name,
234         .id_table = e1000_pci_tbl,
235         .probe    = e1000_probe,
236         .remove   = __devexit_p(e1000_remove),
237 #ifdef CONFIG_PM
238         /* Power Managment Hooks */
239         .suspend  = e1000_suspend,
240         .resume   = e1000_resume,
241 #endif
242         .shutdown = e1000_shutdown,
243         .err_handler = &e1000_err_handler
244 };
245
246 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
247 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
248 MODULE_LICENSE("GPL");
249 MODULE_VERSION(DRV_VERSION);
250
251 static int debug = NETIF_MSG_DRV | NETIF_MSG_PROBE;
252 module_param(debug, int, 0);
253 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
254
255 /**
256  * e1000_init_module - Driver Registration Routine
257  *
258  * e1000_init_module is the first routine called when the driver is
259  * loaded. All it does is register with the PCI subsystem.
260  **/
261
262 static int __init
263 e1000_init_module(void)
264 {
265         int ret;
266         printk(KERN_INFO "%s - version %s\n",
267                e1000_driver_string, e1000_driver_version);
268
269         printk(KERN_INFO "%s\n", e1000_copyright);
270
271         ret = pci_register_driver(&e1000_driver);
272         if (copybreak != COPYBREAK_DEFAULT) {
273                 if (copybreak == 0)
274                         printk(KERN_INFO "e1000: copybreak disabled\n");
275                 else
276                         printk(KERN_INFO "e1000: copybreak enabled for "
277                                "packets <= %u bytes\n", copybreak);
278         }
279         return ret;
280 }
281
282 module_init(e1000_init_module);
283
284 /**
285  * e1000_exit_module - Driver Exit Cleanup Routine
286  *
287  * e1000_exit_module is called just before the driver is removed
288  * from memory.
289  **/
290
291 static void __exit
292 e1000_exit_module(void)
293 {
294         pci_unregister_driver(&e1000_driver);
295 }
296
297 module_exit(e1000_exit_module);
298
299 static int e1000_request_irq(struct e1000_adapter *adapter)
300 {
301         struct net_device *netdev = adapter->netdev;
302         void (*handler) = &e1000_intr;
303         int irq_flags = IRQF_SHARED;
304         int err;
305
306         if (adapter->hw.mac_type >= e1000_82571) {
307                 adapter->have_msi = !pci_enable_msi(adapter->pdev);
308                 if (adapter->have_msi) {
309                         handler = &e1000_intr_msi;
310                         irq_flags = 0;
311                 }
312         }
313
314         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
315                           netdev);
316         if (err) {
317                 if (adapter->have_msi)
318                         pci_disable_msi(adapter->pdev);
319                 DPRINTK(PROBE, ERR,
320                         "Unable to allocate interrupt Error: %d\n", err);
321         }
322
323         return err;
324 }
325
326 static void e1000_free_irq(struct e1000_adapter *adapter)
327 {
328         struct net_device *netdev = adapter->netdev;
329
330         free_irq(adapter->pdev->irq, netdev);
331
332         if (adapter->have_msi)
333                 pci_disable_msi(adapter->pdev);
334 }
335
336 /**
337  * e1000_irq_disable - Mask off interrupt generation on the NIC
338  * @adapter: board private structure
339  **/
340
341 static void
342 e1000_irq_disable(struct e1000_adapter *adapter)
343 {
344         atomic_inc(&adapter->irq_sem);
345         E1000_WRITE_REG(&adapter->hw, IMC, ~0);
346         E1000_WRITE_FLUSH(&adapter->hw);
347         synchronize_irq(adapter->pdev->irq);
348 }
349
350 /**
351  * e1000_irq_enable - Enable default interrupt generation settings
352  * @adapter: board private structure
353  **/
354
355 static void
356 e1000_irq_enable(struct e1000_adapter *adapter)
357 {
358         if (likely(atomic_dec_and_test(&adapter->irq_sem))) {
359                 E1000_WRITE_REG(&adapter->hw, IMS, IMS_ENABLE_MASK);
360                 E1000_WRITE_FLUSH(&adapter->hw);
361         }
362 }
363
364 static void
365 e1000_update_mng_vlan(struct e1000_adapter *adapter)
366 {
367         struct net_device *netdev = adapter->netdev;
368         uint16_t vid = adapter->hw.mng_cookie.vlan_id;
369         uint16_t old_vid = adapter->mng_vlan_id;
370         if (adapter->vlgrp) {
371                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
372                         if (adapter->hw.mng_cookie.status &
373                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
374                                 e1000_vlan_rx_add_vid(netdev, vid);
375                                 adapter->mng_vlan_id = vid;
376                         } else
377                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
378
379                         if ((old_vid != (uint16_t)E1000_MNG_VLAN_NONE) &&
380                                         (vid != old_vid) &&
381                             !vlan_group_get_device(adapter->vlgrp, old_vid))
382                                 e1000_vlan_rx_kill_vid(netdev, old_vid);
383                 } else
384                         adapter->mng_vlan_id = vid;
385         }
386 }
387
388 /**
389  * e1000_release_hw_control - release control of the h/w to f/w
390  * @adapter: address of board private structure
391  *
392  * e1000_release_hw_control resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
393  * For ASF and Pass Through versions of f/w this means that the
394  * driver is no longer loaded. For AMT version (only with 82573) i
395  * of the f/w this means that the network i/f is closed.
396  *
397  **/
398
399 static void
400 e1000_release_hw_control(struct e1000_adapter *adapter)
401 {
402         uint32_t ctrl_ext;
403         uint32_t swsm;
404
405         /* Let firmware taken over control of h/w */
406         switch (adapter->hw.mac_type) {
407         case e1000_82573:
408                 swsm = E1000_READ_REG(&adapter->hw, SWSM);
409                 E1000_WRITE_REG(&adapter->hw, SWSM,
410                                 swsm & ~E1000_SWSM_DRV_LOAD);
411                 break;
412         case e1000_82571:
413         case e1000_82572:
414         case e1000_80003es2lan:
415         case e1000_ich8lan:
416                 ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
417                 E1000_WRITE_REG(&adapter->hw, CTRL_EXT,
418                                 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
419                 break;
420         default:
421                 break;
422         }
423 }
424
425 /**
426  * e1000_get_hw_control - get control of the h/w from f/w
427  * @adapter: address of board private structure
428  *
429  * e1000_get_hw_control sets {CTRL_EXT|FWSM}:DRV_LOAD bit.
430  * For ASF and Pass Through versions of f/w this means that
431  * the driver is loaded. For AMT version (only with 82573)
432  * of the f/w this means that the network i/f is open.
433  *
434  **/
435
436 static void
437 e1000_get_hw_control(struct e1000_adapter *adapter)
438 {
439         uint32_t ctrl_ext;
440         uint32_t swsm;
441
442         /* Let firmware know the driver has taken over */
443         switch (adapter->hw.mac_type) {
444         case e1000_82573:
445                 swsm = E1000_READ_REG(&adapter->hw, SWSM);
446                 E1000_WRITE_REG(&adapter->hw, SWSM,
447                                 swsm | E1000_SWSM_DRV_LOAD);
448                 break;
449         case e1000_82571:
450         case e1000_82572:
451         case e1000_80003es2lan:
452         case e1000_ich8lan:
453                 ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
454                 E1000_WRITE_REG(&adapter->hw, CTRL_EXT,
455                                 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
456                 break;
457         default:
458                 break;
459         }
460 }
461
462 static void
463 e1000_init_manageability(struct e1000_adapter *adapter)
464 {
465         if (adapter->en_mng_pt) {
466                 uint32_t manc = E1000_READ_REG(&adapter->hw, MANC);
467
468                 /* disable hardware interception of ARP */
469                 manc &= ~(E1000_MANC_ARP_EN);
470
471                 /* enable receiving management packets to the host */
472                 /* this will probably generate destination unreachable messages
473                  * from the host OS, but the packets will be handled on SMBUS */
474                 if (adapter->hw.has_manc2h) {
475                         uint32_t manc2h = E1000_READ_REG(&adapter->hw, MANC2H);
476
477                         manc |= E1000_MANC_EN_MNG2HOST;
478 #define E1000_MNG2HOST_PORT_623 (1 << 5)
479 #define E1000_MNG2HOST_PORT_664 (1 << 6)
480                         manc2h |= E1000_MNG2HOST_PORT_623;
481                         manc2h |= E1000_MNG2HOST_PORT_664;
482                         E1000_WRITE_REG(&adapter->hw, MANC2H, manc2h);
483                 }
484
485                 E1000_WRITE_REG(&adapter->hw, MANC, manc);
486         }
487 }
488
489 static void
490 e1000_release_manageability(struct e1000_adapter *adapter)
491 {
492         if (adapter->en_mng_pt) {
493                 uint32_t manc = E1000_READ_REG(&adapter->hw, MANC);
494
495                 /* re-enable hardware interception of ARP */
496                 manc |= E1000_MANC_ARP_EN;
497
498                 if (adapter->hw.has_manc2h)
499                         manc &= ~E1000_MANC_EN_MNG2HOST;
500
501                 /* don't explicitly have to mess with MANC2H since
502                  * MANC has an enable disable that gates MANC2H */
503
504                 E1000_WRITE_REG(&adapter->hw, MANC, manc);
505         }
506 }
507
508 /**
509  * e1000_configure - configure the hardware for RX and TX
510  * @adapter = private board structure
511  **/
512 static void e1000_configure(struct e1000_adapter *adapter)
513 {
514         struct net_device *netdev = adapter->netdev;
515         int i;
516
517         e1000_set_multi(netdev);
518
519         e1000_restore_vlan(adapter);
520         e1000_init_manageability(adapter);
521
522         e1000_configure_tx(adapter);
523         e1000_setup_rctl(adapter);
524         e1000_configure_rx(adapter);
525         /* call E1000_DESC_UNUSED which always leaves
526          * at least 1 descriptor unused to make sure
527          * next_to_use != next_to_clean */
528         for (i = 0; i < adapter->num_rx_queues; i++) {
529                 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
530                 adapter->alloc_rx_buf(adapter, ring,
531                                       E1000_DESC_UNUSED(ring));
532         }
533
534         adapter->tx_queue_len = netdev->tx_queue_len;
535 }
536
537 int e1000_up(struct e1000_adapter *adapter)
538 {
539         /* hardware has been reset, we need to reload some things */
540         e1000_configure(adapter);
541
542         clear_bit(__E1000_DOWN, &adapter->flags);
543
544 #ifdef CONFIG_E1000_NAPI
545         napi_enable(&adapter->napi);
546 #endif
547         e1000_irq_enable(adapter);
548
549         /* fire a link change interrupt to start the watchdog */
550         E1000_WRITE_REG(&adapter->hw, ICS, E1000_ICS_LSC);
551         return 0;
552 }
553
554 /**
555  * e1000_power_up_phy - restore link in case the phy was powered down
556  * @adapter: address of board private structure
557  *
558  * The phy may be powered down to save power and turn off link when the
559  * driver is unloaded and wake on lan is not enabled (among others)
560  * *** this routine MUST be followed by a call to e1000_reset ***
561  *
562  **/
563
564 void e1000_power_up_phy(struct e1000_adapter *adapter)
565 {
566         uint16_t mii_reg = 0;
567
568         /* Just clear the power down bit to wake the phy back up */
569         if (adapter->hw.media_type == e1000_media_type_copper) {
570                 /* according to the manual, the phy will retain its
571                  * settings across a power-down/up cycle */
572                 e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &mii_reg);
573                 mii_reg &= ~MII_CR_POWER_DOWN;
574                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL, mii_reg);
575         }
576 }
577
578 static void e1000_power_down_phy(struct e1000_adapter *adapter)
579 {
580         /* Power down the PHY so no link is implied when interface is down *
581          * The PHY cannot be powered down if any of the following is TRUE *
582          * (a) WoL is enabled
583          * (b) AMT is active
584          * (c) SoL/IDER session is active */
585         if (!adapter->wol && adapter->hw.mac_type >= e1000_82540 &&
586            adapter->hw.media_type == e1000_media_type_copper) {
587                 uint16_t mii_reg = 0;
588
589                 switch (adapter->hw.mac_type) {
590                 case e1000_82540:
591                 case e1000_82545:
592                 case e1000_82545_rev_3:
593                 case e1000_82546:
594                 case e1000_82546_rev_3:
595                 case e1000_82541:
596                 case e1000_82541_rev_2:
597                 case e1000_82547:
598                 case e1000_82547_rev_2:
599                         if (E1000_READ_REG(&adapter->hw, MANC) &
600                             E1000_MANC_SMBUS_EN)
601                                 goto out;
602                         break;
603                 case e1000_82571:
604                 case e1000_82572:
605                 case e1000_82573:
606                 case e1000_80003es2lan:
607                 case e1000_ich8lan:
608                         if (e1000_check_mng_mode(&adapter->hw) ||
609                             e1000_check_phy_reset_block(&adapter->hw))
610                                 goto out;
611                         break;
612                 default:
613                         goto out;
614                 }
615                 e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &mii_reg);
616                 mii_reg |= MII_CR_POWER_DOWN;
617                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL, mii_reg);
618                 mdelay(1);
619         }
620 out:
621         return;
622 }
623
624 void
625 e1000_down(struct e1000_adapter *adapter)
626 {
627         struct net_device *netdev = adapter->netdev;
628
629         /* signal that we're down so the interrupt handler does not
630          * reschedule our watchdog timer */
631         set_bit(__E1000_DOWN, &adapter->flags);
632
633 #ifdef CONFIG_E1000_NAPI
634         napi_disable(&adapter->napi);
635 #endif
636         e1000_irq_disable(adapter);
637
638         del_timer_sync(&adapter->tx_fifo_stall_timer);
639         del_timer_sync(&adapter->watchdog_timer);
640         del_timer_sync(&adapter->phy_info_timer);
641
642         netdev->tx_queue_len = adapter->tx_queue_len;
643         adapter->link_speed = 0;
644         adapter->link_duplex = 0;
645         netif_carrier_off(netdev);
646         netif_stop_queue(netdev);
647
648         e1000_reset(adapter);
649         e1000_clean_all_tx_rings(adapter);
650         e1000_clean_all_rx_rings(adapter);
651 }
652
653 void
654 e1000_reinit_locked(struct e1000_adapter *adapter)
655 {
656         WARN_ON(in_interrupt());
657         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
658                 msleep(1);
659         e1000_down(adapter);
660         e1000_up(adapter);
661         clear_bit(__E1000_RESETTING, &adapter->flags);
662 }
663
664 void
665 e1000_reset(struct e1000_adapter *adapter)
666 {
667         uint32_t pba = 0, tx_space, min_tx_space, min_rx_space;
668         uint16_t fc_high_water_mark = E1000_FC_HIGH_DIFF;
669         boolean_t legacy_pba_adjust = FALSE;
670
671         /* Repartition Pba for greater than 9k mtu
672          * To take effect CTRL.RST is required.
673          */
674
675         switch (adapter->hw.mac_type) {
676         case e1000_82542_rev2_0:
677         case e1000_82542_rev2_1:
678         case e1000_82543:
679         case e1000_82544:
680         case e1000_82540:
681         case e1000_82541:
682         case e1000_82541_rev_2:
683                 legacy_pba_adjust = TRUE;
684                 pba = E1000_PBA_48K;
685                 break;
686         case e1000_82545:
687         case e1000_82545_rev_3:
688         case e1000_82546:
689         case e1000_82546_rev_3:
690                 pba = E1000_PBA_48K;
691                 break;
692         case e1000_82547:
693         case e1000_82547_rev_2:
694                 legacy_pba_adjust = TRUE;
695                 pba = E1000_PBA_30K;
696                 break;
697         case e1000_82571:
698         case e1000_82572:
699         case e1000_80003es2lan:
700                 pba = E1000_PBA_38K;
701                 break;
702         case e1000_82573:
703                 pba = E1000_PBA_20K;
704                 break;
705         case e1000_ich8lan:
706                 pba = E1000_PBA_8K;
707         case e1000_undefined:
708         case e1000_num_macs:
709                 break;
710         }
711
712         if (legacy_pba_adjust == TRUE) {
713                 if (adapter->netdev->mtu > E1000_RXBUFFER_8192)
714                         pba -= 8; /* allocate more FIFO for Tx */
715
716                 if (adapter->hw.mac_type == e1000_82547) {
717                         adapter->tx_fifo_head = 0;
718                         adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
719                         adapter->tx_fifo_size =
720                                 (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
721                         atomic_set(&adapter->tx_fifo_stall, 0);
722                 }
723         } else if (adapter->hw.max_frame_size > MAXIMUM_ETHERNET_FRAME_SIZE) {
724                 /* adjust PBA for jumbo frames */
725                 E1000_WRITE_REG(&adapter->hw, PBA, pba);
726
727                 /* To maintain wire speed transmits, the Tx FIFO should be
728                  * large enough to accomodate two full transmit packets,
729                  * rounded up to the next 1KB and expressed in KB.  Likewise,
730                  * the Rx FIFO should be large enough to accomodate at least
731                  * one full receive packet and is similarly rounded up and
732                  * expressed in KB. */
733                 pba = E1000_READ_REG(&adapter->hw, PBA);
734                 /* upper 16 bits has Tx packet buffer allocation size in KB */
735                 tx_space = pba >> 16;
736                 /* lower 16 bits has Rx packet buffer allocation size in KB */
737                 pba &= 0xffff;
738                 /* don't include ethernet FCS because hardware appends/strips */
739                 min_rx_space = adapter->netdev->mtu + ENET_HEADER_SIZE +
740                                VLAN_TAG_SIZE;
741                 min_tx_space = min_rx_space;
742                 min_tx_space *= 2;
743                 min_tx_space = ALIGN(min_tx_space, 1024);
744                 min_tx_space >>= 10;
745                 min_rx_space = ALIGN(min_rx_space, 1024);
746                 min_rx_space >>= 10;
747
748                 /* If current Tx allocation is less than the min Tx FIFO size,
749                  * and the min Tx FIFO size is less than the current Rx FIFO
750                  * allocation, take space away from current Rx allocation */
751                 if (tx_space < min_tx_space &&
752                     ((min_tx_space - tx_space) < pba)) {
753                         pba = pba - (min_tx_space - tx_space);
754
755                         /* PCI/PCIx hardware has PBA alignment constraints */
756                         switch (adapter->hw.mac_type) {
757                         case e1000_82545 ... e1000_82546_rev_3:
758                                 pba &= ~(E1000_PBA_8K - 1);
759                                 break;
760                         default:
761                                 break;
762                         }
763
764                         /* if short on rx space, rx wins and must trump tx
765                          * adjustment or use Early Receive if available */
766                         if (pba < min_rx_space) {
767                                 switch (adapter->hw.mac_type) {
768                                 case e1000_82573:
769                                         /* ERT enabled in e1000_configure_rx */
770                                         break;
771                                 default:
772                                         pba = min_rx_space;
773                                         break;
774                                 }
775                         }
776                 }
777         }
778
779         E1000_WRITE_REG(&adapter->hw, PBA, pba);
780
781         /* flow control settings */
782         /* Set the FC high water mark to 90% of the FIFO size.
783          * Required to clear last 3 LSB */
784         fc_high_water_mark = ((pba * 9216)/10) & 0xFFF8;
785         /* We can't use 90% on small FIFOs because the remainder
786          * would be less than 1 full frame.  In this case, we size
787          * it to allow at least a full frame above the high water
788          *  mark. */
789         if (pba < E1000_PBA_16K)
790                 fc_high_water_mark = (pba * 1024) - 1600;
791
792         adapter->hw.fc_high_water = fc_high_water_mark;
793         adapter->hw.fc_low_water = fc_high_water_mark - 8;
794         if (adapter->hw.mac_type == e1000_80003es2lan)
795                 adapter->hw.fc_pause_time = 0xFFFF;
796         else
797                 adapter->hw.fc_pause_time = E1000_FC_PAUSE_TIME;
798         adapter->hw.fc_send_xon = 1;
799         adapter->hw.fc = adapter->hw.original_fc;
800
801         /* Allow time for pending master requests to run */
802         e1000_reset_hw(&adapter->hw);
803         if (adapter->hw.mac_type >= e1000_82544)
804                 E1000_WRITE_REG(&adapter->hw, WUC, 0);
805
806         if (e1000_init_hw(&adapter->hw))
807                 DPRINTK(PROBE, ERR, "Hardware Error\n");
808         e1000_update_mng_vlan(adapter);
809
810         /* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
811         if (adapter->hw.mac_type >= e1000_82544 &&
812             adapter->hw.mac_type <= e1000_82547_rev_2 &&
813             adapter->hw.autoneg == 1 &&
814             adapter->hw.autoneg_advertised == ADVERTISE_1000_FULL) {
815                 uint32_t ctrl = E1000_READ_REG(&adapter->hw, CTRL);
816                 /* clear phy power management bit if we are in gig only mode,
817                  * which if enabled will attempt negotiation to 100Mb, which
818                  * can cause a loss of link at power off or driver unload */
819                 ctrl &= ~E1000_CTRL_SWDPIN3;
820                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
821         }
822
823         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
824         E1000_WRITE_REG(&adapter->hw, VET, ETHERNET_IEEE_VLAN_TYPE);
825
826         e1000_reset_adaptive(&adapter->hw);
827         e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
828
829         if (!adapter->smart_power_down &&
830             (adapter->hw.mac_type == e1000_82571 ||
831              adapter->hw.mac_type == e1000_82572)) {
832                 uint16_t phy_data = 0;
833                 /* speed up time to link by disabling smart power down, ignore
834                  * the return value of this function because there is nothing
835                  * different we would do if it failed */
836                 e1000_read_phy_reg(&adapter->hw, IGP02E1000_PHY_POWER_MGMT,
837                                    &phy_data);
838                 phy_data &= ~IGP02E1000_PM_SPD;
839                 e1000_write_phy_reg(&adapter->hw, IGP02E1000_PHY_POWER_MGMT,
840                                     phy_data);
841         }
842
843         e1000_release_manageability(adapter);
844 }
845
846 /**
847  * e1000_probe - Device Initialization Routine
848  * @pdev: PCI device information struct
849  * @ent: entry in e1000_pci_tbl
850  *
851  * Returns 0 on success, negative on failure
852  *
853  * e1000_probe initializes an adapter identified by a pci_dev structure.
854  * The OS initialization, configuring of the adapter private structure,
855  * and a hardware reset occur.
856  **/
857
858 static int __devinit
859 e1000_probe(struct pci_dev *pdev,
860             const struct pci_device_id *ent)
861 {
862         struct net_device *netdev;
863         struct e1000_adapter *adapter;
864         unsigned long mmio_start, mmio_len;
865         unsigned long flash_start, flash_len;
866
867         static int cards_found = 0;
868         static int global_quad_port_a = 0; /* global ksp3 port a indication */
869         int i, err, pci_using_dac;
870         uint16_t eeprom_data = 0;
871         uint16_t eeprom_apme_mask = E1000_EEPROM_APME;
872         DECLARE_MAC_BUF(mac);
873
874         if ((err = pci_enable_device(pdev)))
875                 return err;
876
877         if (!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK)) &&
878             !(err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK))) {
879                 pci_using_dac = 1;
880         } else {
881                 if ((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK)) &&
882                     (err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK))) {
883                         E1000_ERR("No usable DMA configuration, aborting\n");
884                         goto err_dma;
885                 }
886                 pci_using_dac = 0;
887         }
888
889         if ((err = pci_request_regions(pdev, e1000_driver_name)))
890                 goto err_pci_reg;
891
892         pci_set_master(pdev);
893
894         err = -ENOMEM;
895         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
896         if (!netdev)
897                 goto err_alloc_etherdev;
898
899         SET_NETDEV_DEV(netdev, &pdev->dev);
900
901         pci_set_drvdata(pdev, netdev);
902         adapter = netdev_priv(netdev);
903         adapter->netdev = netdev;
904         adapter->pdev = pdev;
905         adapter->hw.back = adapter;
906         adapter->msg_enable = (1 << debug) - 1;
907
908         mmio_start = pci_resource_start(pdev, BAR_0);
909         mmio_len = pci_resource_len(pdev, BAR_0);
910
911         err = -EIO;
912         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
913         if (!adapter->hw.hw_addr)
914                 goto err_ioremap;
915
916         for (i = BAR_1; i <= BAR_5; i++) {
917                 if (pci_resource_len(pdev, i) == 0)
918                         continue;
919                 if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
920                         adapter->hw.io_base = pci_resource_start(pdev, i);
921                         break;
922                 }
923         }
924
925         netdev->open = &e1000_open;
926         netdev->stop = &e1000_close;
927         netdev->hard_start_xmit = &e1000_xmit_frame;
928         netdev->get_stats = &e1000_get_stats;
929         netdev->set_multicast_list = &e1000_set_multi;
930         netdev->set_mac_address = &e1000_set_mac;
931         netdev->change_mtu = &e1000_change_mtu;
932         netdev->do_ioctl = &e1000_ioctl;
933         e1000_set_ethtool_ops(netdev);
934         netdev->tx_timeout = &e1000_tx_timeout;
935         netdev->watchdog_timeo = 5 * HZ;
936 #ifdef CONFIG_E1000_NAPI
937         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
938 #endif
939         netdev->vlan_rx_register = e1000_vlan_rx_register;
940         netdev->vlan_rx_add_vid = e1000_vlan_rx_add_vid;
941         netdev->vlan_rx_kill_vid = e1000_vlan_rx_kill_vid;
942 #ifdef CONFIG_NET_POLL_CONTROLLER
943         netdev->poll_controller = e1000_netpoll;
944 #endif
945         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
946
947         netdev->mem_start = mmio_start;
948         netdev->mem_end = mmio_start + mmio_len;
949         netdev->base_addr = adapter->hw.io_base;
950
951         adapter->bd_number = cards_found;
952
953         /* setup the private structure */
954
955         if ((err = e1000_sw_init(adapter)))
956                 goto err_sw_init;
957
958         err = -EIO;
959         /* Flash BAR mapping must happen after e1000_sw_init
960          * because it depends on mac_type */
961         if ((adapter->hw.mac_type == e1000_ich8lan) &&
962            (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
963                 flash_start = pci_resource_start(pdev, 1);
964                 flash_len = pci_resource_len(pdev, 1);
965                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
966                 if (!adapter->hw.flash_address)
967                         goto err_flashmap;
968         }
969
970         if (e1000_check_phy_reset_block(&adapter->hw))
971                 DPRINTK(PROBE, INFO, "PHY reset is blocked due to SOL/IDER session.\n");
972
973         if (adapter->hw.mac_type >= e1000_82543) {
974                 netdev->features = NETIF_F_SG |
975                                    NETIF_F_HW_CSUM |
976                                    NETIF_F_HW_VLAN_TX |
977                                    NETIF_F_HW_VLAN_RX |
978                                    NETIF_F_HW_VLAN_FILTER;
979                 if (adapter->hw.mac_type == e1000_ich8lan)
980                         netdev->features &= ~NETIF_F_HW_VLAN_FILTER;
981         }
982
983         if ((adapter->hw.mac_type >= e1000_82544) &&
984            (adapter->hw.mac_type != e1000_82547))
985                 netdev->features |= NETIF_F_TSO;
986
987         if (adapter->hw.mac_type > e1000_82547_rev_2)
988                 netdev->features |= NETIF_F_TSO6;
989         if (pci_using_dac)
990                 netdev->features |= NETIF_F_HIGHDMA;
991
992         netdev->features |= NETIF_F_LLTX;
993
994         adapter->en_mng_pt = e1000_enable_mng_pass_thru(&adapter->hw);
995
996         /* initialize eeprom parameters */
997
998         if (e1000_init_eeprom_params(&adapter->hw)) {
999                 E1000_ERR("EEPROM initialization failed\n");
1000                 goto err_eeprom;
1001         }
1002
1003         /* before reading the EEPROM, reset the controller to
1004          * put the device in a known good starting state */
1005
1006         e1000_reset_hw(&adapter->hw);
1007
1008         /* make sure the EEPROM is good */
1009
1010         if (e1000_validate_eeprom_checksum(&adapter->hw) < 0) {
1011                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
1012                 goto err_eeprom;
1013         }
1014
1015         /* copy the MAC address out of the EEPROM */
1016
1017         if (e1000_read_mac_addr(&adapter->hw))
1018                 DPRINTK(PROBE, ERR, "EEPROM Read Error\n");
1019         memcpy(netdev->dev_addr, adapter->hw.mac_addr, netdev->addr_len);
1020         memcpy(netdev->perm_addr, adapter->hw.mac_addr, netdev->addr_len);
1021
1022         if (!is_valid_ether_addr(netdev->perm_addr)) {
1023                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
1024                 goto err_eeprom;
1025         }
1026
1027         e1000_get_bus_info(&adapter->hw);
1028
1029         init_timer(&adapter->tx_fifo_stall_timer);
1030         adapter->tx_fifo_stall_timer.function = &e1000_82547_tx_fifo_stall;
1031         adapter->tx_fifo_stall_timer.data = (unsigned long) adapter;
1032
1033         init_timer(&adapter->watchdog_timer);
1034         adapter->watchdog_timer.function = &e1000_watchdog;
1035         adapter->watchdog_timer.data = (unsigned long) adapter;
1036
1037         init_timer(&adapter->phy_info_timer);
1038         adapter->phy_info_timer.function = &e1000_update_phy_info;
1039         adapter->phy_info_timer.data = (unsigned long) adapter;
1040
1041         INIT_WORK(&adapter->reset_task, e1000_reset_task);
1042
1043         e1000_check_options(adapter);
1044
1045         /* Initial Wake on LAN setting
1046          * If APM wake is enabled in the EEPROM,
1047          * enable the ACPI Magic Packet filter
1048          */
1049
1050         switch (adapter->hw.mac_type) {
1051         case e1000_82542_rev2_0:
1052         case e1000_82542_rev2_1:
1053         case e1000_82543:
1054                 break;
1055         case e1000_82544:
1056                 e1000_read_eeprom(&adapter->hw,
1057                         EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
1058                 eeprom_apme_mask = E1000_EEPROM_82544_APM;
1059                 break;
1060         case e1000_ich8lan:
1061                 e1000_read_eeprom(&adapter->hw,
1062                         EEPROM_INIT_CONTROL1_REG, 1, &eeprom_data);
1063                 eeprom_apme_mask = E1000_EEPROM_ICH8_APME;
1064                 break;
1065         case e1000_82546:
1066         case e1000_82546_rev_3:
1067         case e1000_82571:
1068         case e1000_80003es2lan:
1069                 if (E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_FUNC_1){
1070                         e1000_read_eeprom(&adapter->hw,
1071                                 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1072                         break;
1073                 }
1074                 /* Fall Through */
1075         default:
1076                 e1000_read_eeprom(&adapter->hw,
1077                         EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1078                 break;
1079         }
1080         if (eeprom_data & eeprom_apme_mask)
1081                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1082
1083         /* now that we have the eeprom settings, apply the special cases
1084          * where the eeprom may be wrong or the board simply won't support
1085          * wake on lan on a particular port */
1086         switch (pdev->device) {
1087         case E1000_DEV_ID_82546GB_PCIE:
1088                 adapter->eeprom_wol = 0;
1089                 break;
1090         case E1000_DEV_ID_82546EB_FIBER:
1091         case E1000_DEV_ID_82546GB_FIBER:
1092         case E1000_DEV_ID_82571EB_FIBER:
1093                 /* Wake events only supported on port A for dual fiber
1094                  * regardless of eeprom setting */
1095                 if (E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_FUNC_1)
1096                         adapter->eeprom_wol = 0;
1097                 break;
1098         case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
1099         case E1000_DEV_ID_82571EB_QUAD_COPPER:
1100         case E1000_DEV_ID_82571EB_QUAD_FIBER:
1101         case E1000_DEV_ID_82571EB_QUAD_COPPER_LOWPROFILE:
1102         case E1000_DEV_ID_82571PT_QUAD_COPPER:
1103                 /* if quad port adapter, disable WoL on all but port A */
1104                 if (global_quad_port_a != 0)
1105                         adapter->eeprom_wol = 0;
1106                 else
1107                         adapter->quad_port_a = 1;
1108                 /* Reset for multiple quad port adapters */
1109                 if (++global_quad_port_a == 4)
1110                         global_quad_port_a = 0;
1111                 break;
1112         }
1113
1114         /* initialize the wol settings based on the eeprom settings */
1115         adapter->wol = adapter->eeprom_wol;
1116
1117         /* print bus type/speed/width info */
1118         {
1119         struct e1000_hw *hw = &adapter->hw;
1120         DPRINTK(PROBE, INFO, "(PCI%s:%s:%s) ",
1121                 ((hw->bus_type == e1000_bus_type_pcix) ? "-X" :
1122                  (hw->bus_type == e1000_bus_type_pci_express ? " Express":"")),
1123                 ((hw->bus_speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
1124                  (hw->bus_speed == e1000_bus_speed_133) ? "133MHz" :
1125                  (hw->bus_speed == e1000_bus_speed_120) ? "120MHz" :
1126                  (hw->bus_speed == e1000_bus_speed_100) ? "100MHz" :
1127                  (hw->bus_speed == e1000_bus_speed_66) ? "66MHz" : "33MHz"),
1128                 ((hw->bus_width == e1000_bus_width_64) ? "64-bit" :
1129                  (hw->bus_width == e1000_bus_width_pciex_4) ? "Width x4" :
1130                  (hw->bus_width == e1000_bus_width_pciex_1) ? "Width x1" :
1131                  "32-bit"));
1132         }
1133
1134         printk("%s\n", print_mac(mac, netdev->dev_addr));
1135
1136         /* reset the hardware with the new settings */
1137         e1000_reset(adapter);
1138
1139         /* If the controller is 82573 and f/w is AMT, do not set
1140          * DRV_LOAD until the interface is up.  For all other cases,
1141          * let the f/w know that the h/w is now under the control
1142          * of the driver. */
1143         if (adapter->hw.mac_type != e1000_82573 ||
1144             !e1000_check_mng_mode(&adapter->hw))
1145                 e1000_get_hw_control(adapter);
1146
1147         /* tell the stack to leave us alone until e1000_open() is called */
1148         netif_carrier_off(netdev);
1149         netif_stop_queue(netdev);
1150
1151         strcpy(netdev->name, "eth%d");
1152         if ((err = register_netdev(netdev)))
1153                 goto err_register;
1154
1155         DPRINTK(PROBE, INFO, "Intel(R) PRO/1000 Network Connection\n");
1156
1157         cards_found++;
1158         return 0;
1159
1160 err_register:
1161         e1000_release_hw_control(adapter);
1162 err_eeprom:
1163         if (!e1000_check_phy_reset_block(&adapter->hw))
1164                 e1000_phy_hw_reset(&adapter->hw);
1165
1166         if (adapter->hw.flash_address)
1167                 iounmap(adapter->hw.flash_address);
1168 err_flashmap:
1169 #ifdef CONFIG_E1000_NAPI
1170         for (i = 0; i < adapter->num_rx_queues; i++)
1171                 dev_put(&adapter->polling_netdev[i]);
1172 #endif
1173
1174         kfree(adapter->tx_ring);
1175         kfree(adapter->rx_ring);
1176 #ifdef CONFIG_E1000_NAPI
1177         kfree(adapter->polling_netdev);
1178 #endif
1179 err_sw_init:
1180         iounmap(adapter->hw.hw_addr);
1181 err_ioremap:
1182         free_netdev(netdev);
1183 err_alloc_etherdev:
1184         pci_release_regions(pdev);
1185 err_pci_reg:
1186 err_dma:
1187         pci_disable_device(pdev);
1188         return err;
1189 }
1190
1191 /**
1192  * e1000_remove - Device Removal Routine
1193  * @pdev: PCI device information struct
1194  *
1195  * e1000_remove is called by the PCI subsystem to alert the driver
1196  * that it should release a PCI device.  The could be caused by a
1197  * Hot-Plug event, or because the driver is going to be removed from
1198  * memory.
1199  **/
1200
1201 static void __devexit
1202 e1000_remove(struct pci_dev *pdev)
1203 {
1204         struct net_device *netdev = pci_get_drvdata(pdev);
1205         struct e1000_adapter *adapter = netdev_priv(netdev);
1206 #ifdef CONFIG_E1000_NAPI
1207         int i;
1208 #endif
1209
1210         cancel_work_sync(&adapter->reset_task);
1211
1212         e1000_release_manageability(adapter);
1213
1214         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
1215          * would have already happened in close and is redundant. */
1216         e1000_release_hw_control(adapter);
1217
1218 #ifdef CONFIG_E1000_NAPI
1219         for (i = 0; i < adapter->num_rx_queues; i++)
1220                 dev_put(&adapter->polling_netdev[i]);
1221 #endif
1222
1223         unregister_netdev(netdev);
1224
1225         if (!e1000_check_phy_reset_block(&adapter->hw))
1226                 e1000_phy_hw_reset(&adapter->hw);
1227
1228         kfree(adapter->tx_ring);
1229         kfree(adapter->rx_ring);
1230 #ifdef CONFIG_E1000_NAPI
1231         kfree(adapter->polling_netdev);
1232 #endif
1233
1234         iounmap(adapter->hw.hw_addr);
1235         if (adapter->hw.flash_address)
1236                 iounmap(adapter->hw.flash_address);
1237         pci_release_regions(pdev);
1238
1239         free_netdev(netdev);
1240
1241         pci_disable_device(pdev);
1242 }
1243
1244 /**
1245  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1246  * @adapter: board private structure to initialize
1247  *
1248  * e1000_sw_init initializes the Adapter private data structure.
1249  * Fields are initialized based on PCI device information and
1250  * OS network device settings (MTU size).
1251  **/
1252
1253 static int __devinit
1254 e1000_sw_init(struct e1000_adapter *adapter)
1255 {
1256         struct e1000_hw *hw = &adapter->hw;
1257         struct net_device *netdev = adapter->netdev;
1258         struct pci_dev *pdev = adapter->pdev;
1259 #ifdef CONFIG_E1000_NAPI
1260         int i;
1261 #endif
1262
1263         /* PCI config space info */
1264
1265         hw->vendor_id = pdev->vendor;
1266         hw->device_id = pdev->device;
1267         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1268         hw->subsystem_id = pdev->subsystem_device;
1269         hw->revision_id = pdev->revision;
1270
1271         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
1272
1273         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1274         adapter->rx_ps_bsize0 = E1000_RXBUFFER_128;
1275         hw->max_frame_size = netdev->mtu +
1276                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
1277         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
1278
1279         /* identify the MAC */
1280
1281         if (e1000_set_mac_type(hw)) {
1282                 DPRINTK(PROBE, ERR, "Unknown MAC Type\n");
1283                 return -EIO;
1284         }
1285
1286         switch (hw->mac_type) {
1287         default:
1288                 break;
1289         case e1000_82541:
1290         case e1000_82547:
1291         case e1000_82541_rev_2:
1292         case e1000_82547_rev_2:
1293                 hw->phy_init_script = 1;
1294                 break;
1295         }
1296
1297         e1000_set_media_type(hw);
1298
1299         hw->wait_autoneg_complete = FALSE;
1300         hw->tbi_compatibility_en = TRUE;
1301         hw->adaptive_ifs = TRUE;
1302
1303         /* Copper options */
1304
1305         if (hw->media_type == e1000_media_type_copper) {
1306                 hw->mdix = AUTO_ALL_MODES;
1307                 hw->disable_polarity_correction = FALSE;
1308                 hw->master_slave = E1000_MASTER_SLAVE;
1309         }
1310
1311         adapter->num_tx_queues = 1;
1312         adapter->num_rx_queues = 1;
1313
1314         if (e1000_alloc_queues(adapter)) {
1315                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
1316                 return -ENOMEM;
1317         }
1318
1319 #ifdef CONFIG_E1000_NAPI
1320         for (i = 0; i < adapter->num_rx_queues; i++) {
1321                 adapter->polling_netdev[i].priv = adapter;
1322                 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         napi_enable(&adapter->napi);
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 - skb->data_len;
3262         unsigned long flags;
3263         unsigned int nr_frags;
3264         unsigned int mss;
3265         int count = 0;
3266         int tso;
3267         unsigned int f;
3268
3269         /* This goes back to the question of how to logically map a tx queue
3270          * to a flow.  Right now, performance is impacted slightly negatively
3271          * if using multiple tx queues.  If the stack breaks away from a
3272          * single qdisc implementation, we can look at this again. */
3273         tx_ring = adapter->tx_ring;
3274
3275         if (unlikely(skb->len <= 0)) {
3276                 dev_kfree_skb_any(skb);
3277                 return NETDEV_TX_OK;
3278         }
3279
3280         /* 82571 and newer doesn't need the workaround that limited descriptor
3281          * length to 4kB */
3282         if (adapter->hw.mac_type >= e1000_82571)
3283                 max_per_txd = 8192;
3284
3285         mss = skb_shinfo(skb)->gso_size;
3286         /* The controller does a simple calculation to
3287          * make sure there is enough room in the FIFO before
3288          * initiating the DMA for each buffer.  The calc is:
3289          * 4 = ceil(buffer len/mss).  To make sure we don't
3290          * overrun the FIFO, adjust the max buffer len if mss
3291          * drops. */
3292         if (mss) {
3293                 uint8_t hdr_len;
3294                 max_per_txd = min(mss << 2, max_per_txd);
3295                 max_txd_pwr = fls(max_per_txd) - 1;
3296
3297                 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
3298                 * points to just header, pull a few bytes of payload from
3299                 * frags into skb->data */
3300                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3301                 if (skb->data_len && hdr_len == len) {
3302                         switch (adapter->hw.mac_type) {
3303                                 unsigned int pull_size;
3304                         case e1000_82544:
3305                                 /* Make sure we have room to chop off 4 bytes,
3306                                  * and that the end alignment will work out to
3307                                  * this hardware's requirements
3308                                  * NOTE: this is a TSO only workaround
3309                                  * if end byte alignment not correct move us
3310                                  * into the next dword */
3311                                 if ((unsigned long)(skb_tail_pointer(skb) - 1) & 4)
3312                                         break;
3313                                 /* fall through */
3314                         case e1000_82571:
3315                         case e1000_82572:
3316                         case e1000_82573:
3317                         case e1000_ich8lan:
3318                                 pull_size = min((unsigned int)4, skb->data_len);
3319                                 if (!__pskb_pull_tail(skb, pull_size)) {
3320                                         DPRINTK(DRV, ERR,
3321                                                 "__pskb_pull_tail failed.\n");
3322                                         dev_kfree_skb_any(skb);
3323                                         return NETDEV_TX_OK;
3324                                 }
3325                                 len = skb->len - skb->data_len;
3326                                 break;
3327                         default:
3328                                 /* do nothing */
3329                                 break;
3330                         }
3331                 }
3332         }
3333
3334         /* reserve a descriptor for the offload context */
3335         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3336                 count++;
3337         count++;
3338
3339         /* Controller Erratum workaround */
3340         if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3341                 count++;
3342
3343         count += TXD_USE_COUNT(len, max_txd_pwr);
3344
3345         if (adapter->pcix_82544)
3346                 count++;
3347
3348         /* work-around for errata 10 and it applies to all controllers
3349          * in PCI-X mode, so add one more descriptor to the count
3350          */
3351         if (unlikely((adapter->hw.bus_type == e1000_bus_type_pcix) &&
3352                         (len > 2015)))
3353                 count++;
3354
3355         nr_frags = skb_shinfo(skb)->nr_frags;
3356         for (f = 0; f < nr_frags; f++)
3357                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3358                                        max_txd_pwr);
3359         if (adapter->pcix_82544)
3360                 count += nr_frags;
3361
3362
3363         if (adapter->hw.tx_pkt_filtering &&
3364             (adapter->hw.mac_type == e1000_82573))
3365                 e1000_transfer_dhcp_info(adapter, skb);
3366
3367         if (!spin_trylock_irqsave(&tx_ring->tx_lock, flags))
3368                 /* Collision - tell upper layer to requeue */
3369                 return NETDEV_TX_LOCKED;
3370
3371         /* need: count + 2 desc gap to keep tail from touching
3372          * head, otherwise try next time */
3373         if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2))) {
3374                 spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3375                 return NETDEV_TX_BUSY;
3376         }
3377
3378         if (unlikely(adapter->hw.mac_type == e1000_82547)) {
3379                 if (unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
3380                         netif_stop_queue(netdev);
3381                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
3382                         spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3383                         return NETDEV_TX_BUSY;
3384                 }
3385         }
3386
3387         if (unlikely(adapter->vlgrp && vlan_tx_tag_present(skb))) {
3388                 tx_flags |= E1000_TX_FLAGS_VLAN;
3389                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3390         }
3391
3392         first = tx_ring->next_to_use;
3393
3394         tso = e1000_tso(adapter, tx_ring, skb);
3395         if (tso < 0) {
3396                 dev_kfree_skb_any(skb);
3397                 spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3398                 return NETDEV_TX_OK;
3399         }
3400
3401         if (likely(tso)) {
3402                 tx_ring->last_tx_tso = 1;
3403                 tx_flags |= E1000_TX_FLAGS_TSO;
3404         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3405                 tx_flags |= E1000_TX_FLAGS_CSUM;
3406
3407         /* Old method was to assume IPv4 packet by default if TSO was enabled.
3408          * 82571 hardware supports TSO capabilities for IPv6 as well...
3409          * no longer assume, we must. */
3410         if (likely(skb->protocol == htons(ETH_P_IP)))
3411                 tx_flags |= E1000_TX_FLAGS_IPV4;
3412
3413         e1000_tx_queue(adapter, tx_ring, tx_flags,
3414                        e1000_tx_map(adapter, tx_ring, skb, first,
3415                                     max_per_txd, nr_frags, mss));
3416
3417         netdev->trans_start = jiffies;
3418
3419         /* Make sure there is space in the ring for the next send. */
3420         e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3421
3422         spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3423         return NETDEV_TX_OK;
3424 }
3425
3426 /**
3427  * e1000_tx_timeout - Respond to a Tx Hang
3428  * @netdev: network interface device structure
3429  **/
3430
3431 static void
3432 e1000_tx_timeout(struct net_device *netdev)
3433 {
3434         struct e1000_adapter *adapter = netdev_priv(netdev);
3435
3436         /* Do the reset outside of interrupt context */
3437         adapter->tx_timeout_count++;
3438         schedule_work(&adapter->reset_task);
3439 }
3440
3441 static void
3442 e1000_reset_task(struct work_struct *work)
3443 {
3444         struct e1000_adapter *adapter =
3445                 container_of(work, struct e1000_adapter, reset_task);
3446
3447         e1000_reinit_locked(adapter);
3448 }
3449
3450 /**
3451  * e1000_get_stats - Get System Network Statistics
3452  * @netdev: network interface device structure
3453  *
3454  * Returns the address of the device statistics structure.
3455  * The statistics are actually updated from the timer callback.
3456  **/
3457
3458 static struct net_device_stats *
3459 e1000_get_stats(struct net_device *netdev)
3460 {
3461         struct e1000_adapter *adapter = netdev_priv(netdev);
3462
3463         /* only return the current stats */
3464         return &adapter->net_stats;
3465 }
3466
3467 /**
3468  * e1000_change_mtu - Change the Maximum Transfer Unit
3469  * @netdev: network interface device structure
3470  * @new_mtu: new value for maximum frame size
3471  *
3472  * Returns 0 on success, negative on failure
3473  **/
3474
3475 static int
3476 e1000_change_mtu(struct net_device *netdev, int new_mtu)
3477 {
3478         struct e1000_adapter *adapter = netdev_priv(netdev);
3479         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3480         uint16_t eeprom_data = 0;
3481
3482         if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3483             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3484                 DPRINTK(PROBE, ERR, "Invalid MTU setting\n");
3485                 return -EINVAL;
3486         }
3487
3488         /* Adapter-specific max frame size limits. */
3489         switch (adapter->hw.mac_type) {
3490         case e1000_undefined ... e1000_82542_rev2_1:
3491         case e1000_ich8lan:
3492                 if (max_frame > MAXIMUM_ETHERNET_FRAME_SIZE) {
3493                         DPRINTK(PROBE, ERR, "Jumbo Frames not supported.\n");
3494                         return -EINVAL;
3495                 }
3496                 break;
3497         case e1000_82573:
3498                 /* Jumbo Frames not supported if:
3499                  * - this is not an 82573L device
3500                  * - ASPM is enabled in any way (0x1A bits 3:2) */
3501                 e1000_read_eeprom(&adapter->hw, EEPROM_INIT_3GIO_3, 1,
3502                                   &eeprom_data);
3503                 if ((adapter->hw.device_id != E1000_DEV_ID_82573L) ||
3504                     (eeprom_data & EEPROM_WORD1A_ASPM_MASK)) {
3505                         if (max_frame > MAXIMUM_ETHERNET_FRAME_SIZE) {
3506                                 DPRINTK(PROBE, ERR,
3507                                         "Jumbo Frames not supported.\n");
3508                                 return -EINVAL;
3509                         }
3510                         break;
3511                 }
3512                 /* ERT will be enabled later to enable wire speed receives */
3513
3514                 /* fall through to get support */
3515         case e1000_82571:
3516         case e1000_82572:
3517         case e1000_80003es2lan:
3518 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3519                 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3520                         DPRINTK(PROBE, ERR, "MTU > 9216 not supported.\n");
3521                         return -EINVAL;
3522                 }
3523                 break;
3524         default:
3525                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3526                 break;
3527         }
3528
3529         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3530          * means we reserve 2 more, this pushes us to allocate from the next
3531          * larger slab size
3532          * i.e. RXBUFFER_2048 --> size-4096 slab */
3533
3534         if (max_frame <= E1000_RXBUFFER_256)
3535                 adapter->rx_buffer_len = E1000_RXBUFFER_256;
3536         else if (max_frame <= E1000_RXBUFFER_512)
3537                 adapter->rx_buffer_len = E1000_RXBUFFER_512;
3538         else if (max_frame <= E1000_RXBUFFER_1024)
3539                 adapter->rx_buffer_len = E1000_RXBUFFER_1024;
3540         else if (max_frame <= E1000_RXBUFFER_2048)
3541                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3542         else if (max_frame <= E1000_RXBUFFER_4096)
3543                 adapter->rx_buffer_len = E1000_RXBUFFER_4096;
3544         else if (max_frame <= E1000_RXBUFFER_8192)
3545                 adapter->rx_buffer_len = E1000_RXBUFFER_8192;
3546         else if (max_frame <= E1000_RXBUFFER_16384)
3547                 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3548
3549         /* adjust allocation if LPE protects us, and we aren't using SBP */
3550         if (!adapter->hw.tbi_compatibility_on &&
3551             ((max_frame == MAXIMUM_ETHERNET_FRAME_SIZE) ||
3552              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3553                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3554
3555         netdev->mtu = new_mtu;
3556         adapter->hw.max_frame_size = max_frame;
3557
3558         if (netif_running(netdev))
3559                 e1000_reinit_locked(adapter);
3560
3561         return 0;
3562 }
3563
3564 /**
3565  * e1000_update_stats - Update the board statistics counters
3566  * @adapter: board private structure
3567  **/
3568
3569 void
3570 e1000_update_stats(struct e1000_adapter *adapter)
3571 {
3572         struct e1000_hw *hw = &adapter->hw;
3573         struct pci_dev *pdev = adapter->pdev;
3574         unsigned long flags;
3575         uint16_t phy_tmp;
3576
3577 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3578
3579         /*
3580          * Prevent stats update while adapter is being reset, or if the pci
3581          * connection is down.
3582          */
3583         if (adapter->link_speed == 0)
3584                 return;
3585         if (pci_channel_offline(pdev))
3586                 return;
3587
3588         spin_lock_irqsave(&adapter->stats_lock, flags);
3589
3590         /* these counters are modified from e1000_tbi_adjust_stats,
3591          * called from the interrupt context, so they must only
3592          * be written while holding adapter->stats_lock
3593          */
3594
3595         adapter->stats.crcerrs += E1000_READ_REG(hw, CRCERRS);
3596         adapter->stats.gprc += E1000_READ_REG(hw, GPRC);
3597         adapter->stats.gorcl += E1000_READ_REG(hw, GORCL);
3598         adapter->stats.gorch += E1000_READ_REG(hw, GORCH);
3599         adapter->stats.bprc += E1000_READ_REG(hw, BPRC);
3600         adapter->stats.mprc += E1000_READ_REG(hw, MPRC);
3601         adapter->stats.roc += E1000_READ_REG(hw, ROC);
3602
3603         if (adapter->hw.mac_type != e1000_ich8lan) {
3604                 adapter->stats.prc64 += E1000_READ_REG(hw, PRC64);
3605                 adapter->stats.prc127 += E1000_READ_REG(hw, PRC127);
3606                 adapter->stats.prc255 += E1000_READ_REG(hw, PRC255);
3607                 adapter->stats.prc511 += E1000_READ_REG(hw, PRC511);
3608                 adapter->stats.prc1023 += E1000_READ_REG(hw, PRC1023);
3609                 adapter->stats.prc1522 += E1000_READ_REG(hw, PRC1522);
3610         }
3611
3612         adapter->stats.symerrs += E1000_READ_REG(hw, SYMERRS);
3613         adapter->stats.mpc += E1000_READ_REG(hw, MPC);
3614         adapter->stats.scc += E1000_READ_REG(hw, SCC);
3615         adapter->stats.ecol += E1000_READ_REG(hw, ECOL);
3616         adapter->stats.mcc += E1000_READ_REG(hw, MCC);
3617         adapter->stats.latecol += E1000_READ_REG(hw, LATECOL);
3618         adapter->stats.dc += E1000_READ_REG(hw, DC);
3619         adapter->stats.sec += E1000_READ_REG(hw, SEC);
3620         adapter->stats.rlec += E1000_READ_REG(hw, RLEC);
3621         adapter->stats.xonrxc += E1000_READ_REG(hw, XONRXC);
3622         adapter->stats.xontxc += E1000_READ_REG(hw, XONTXC);
3623         adapter->stats.xoffrxc += E1000_READ_REG(hw, XOFFRXC);
3624         adapter->stats.xofftxc += E1000_READ_REG(hw, XOFFTXC);
3625         adapter->stats.fcruc += E1000_READ_REG(hw, FCRUC);
3626         adapter->stats.gptc += E1000_READ_REG(hw, GPTC);
3627         adapter->stats.gotcl += E1000_READ_REG(hw, GOTCL);
3628         adapter->stats.gotch += E1000_READ_REG(hw, GOTCH);
3629         adapter->stats.rnbc += E1000_READ_REG(hw, RNBC);
3630         adapter->stats.ruc += E1000_READ_REG(hw, RUC);
3631         adapter->stats.rfc += E1000_READ_REG(hw, RFC);
3632         adapter->stats.rjc += E1000_READ_REG(hw, RJC);
3633         adapter->stats.torl += E1000_READ_REG(hw, TORL);
3634         adapter->stats.torh += E1000_READ_REG(hw, TORH);
3635         adapter->stats.totl += E1000_READ_REG(hw, TOTL);
3636         adapter->stats.toth += E1000_READ_REG(hw, TOTH);
3637         adapter->stats.tpr += E1000_READ_REG(hw, TPR);
3638
3639         if (adapter->hw.mac_type != e1000_ich8lan) {
3640                 adapter->stats.ptc64 += E1000_READ_REG(hw, PTC64);
3641                 adapter->stats.ptc127 += E1000_READ_REG(hw, PTC127);
3642                 adapter->stats.ptc255 += E1000_READ_REG(hw, PTC255);
3643                 adapter->stats.ptc511 += E1000_READ_REG(hw, PTC511);
3644                 adapter->stats.ptc1023 += E1000_READ_REG(hw, PTC1023);
3645                 adapter->stats.ptc1522 += E1000_READ_REG(hw, PTC1522);
3646         }
3647
3648         adapter->stats.mptc += E1000_READ_REG(hw, MPTC);
3649         adapter->stats.bptc += E1000_READ_REG(hw, BPTC);
3650
3651         /* used for adaptive IFS */
3652
3653         hw->tx_packet_delta = E1000_READ_REG(hw, TPT);
3654         adapter->stats.tpt += hw->tx_packet_delta;
3655         hw->collision_delta = E1000_READ_REG(hw, COLC);
3656         adapter->stats.colc += hw->collision_delta;
3657
3658         if (hw->mac_type >= e1000_82543) {
3659                 adapter->stats.algnerrc += E1000_READ_REG(hw, ALGNERRC);
3660                 adapter->stats.rxerrc += E1000_READ_REG(hw, RXERRC);
3661                 adapter->stats.tncrs += E1000_READ_REG(hw, TNCRS);
3662                 adapter->stats.cexterr += E1000_READ_REG(hw, CEXTERR);
3663                 adapter->stats.tsctc += E1000_READ_REG(hw, TSCTC);
3664                 adapter->stats.tsctfc += E1000_READ_REG(hw, TSCTFC);
3665         }
3666         if (hw->mac_type > e1000_82547_rev_2) {
3667                 adapter->stats.iac += E1000_READ_REG(hw, IAC);
3668                 adapter->stats.icrxoc += E1000_READ_REG(hw, ICRXOC);
3669
3670                 if (adapter->hw.mac_type != e1000_ich8lan) {
3671                         adapter->stats.icrxptc += E1000_READ_REG(hw, ICRXPTC);
3672                         adapter->stats.icrxatc += E1000_READ_REG(hw, ICRXATC);
3673                         adapter->stats.ictxptc += E1000_READ_REG(hw, ICTXPTC);
3674                         adapter->stats.ictxatc += E1000_READ_REG(hw, ICTXATC);
3675                         adapter->stats.ictxqec += E1000_READ_REG(hw, ICTXQEC);
3676                         adapter->stats.ictxqmtc += E1000_READ_REG(hw, ICTXQMTC);
3677                         adapter->stats.icrxdmtc += E1000_READ_REG(hw, ICRXDMTC);
3678                 }
3679         }
3680
3681         /* Fill out the OS statistics structure */
3682         adapter->net_stats.rx_packets = adapter->stats.gprc;
3683         adapter->net_stats.tx_packets = adapter->stats.gptc;
3684         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
3685         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
3686         adapter->net_stats.multicast = adapter->stats.mprc;
3687         adapter->net_stats.collisions = adapter->stats.colc;
3688
3689         /* Rx Errors */
3690
3691         /* RLEC on some newer hardware can be incorrect so build
3692         * our own version based on RUC and ROC */
3693         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3694                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3695                 adapter->stats.ruc + adapter->stats.roc +
3696                 adapter->stats.cexterr;
3697         adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3698         adapter->net_stats.rx_length_errors = adapter->stats.rlerrc;
3699         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3700         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3701         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3702
3703         /* Tx Errors */
3704         adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3705         adapter->net_stats.tx_errors = adapter->stats.txerrc;
3706         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3707         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3708         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3709         if (adapter->hw.bad_tx_carr_stats_fd &&
3710             adapter->link_duplex == FULL_DUPLEX) {
3711                 adapter->net_stats.tx_carrier_errors = 0;
3712                 adapter->stats.tncrs = 0;
3713         }
3714
3715         /* Tx Dropped needs to be maintained elsewhere */
3716
3717         /* Phy Stats */
3718         if (hw->media_type == e1000_media_type_copper) {
3719                 if ((adapter->link_speed == SPEED_1000) &&
3720                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3721                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3722                         adapter->phy_stats.idle_errors += phy_tmp;
3723                 }
3724
3725                 if ((hw->mac_type <= e1000_82546) &&
3726                    (hw->phy_type == e1000_phy_m88) &&
3727                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3728                         adapter->phy_stats.receive_errors += phy_tmp;
3729         }
3730
3731         /* Management Stats */
3732         if (adapter->hw.has_smbus) {
3733                 adapter->stats.mgptc += E1000_READ_REG(hw, MGTPTC);
3734                 adapter->stats.mgprc += E1000_READ_REG(hw, MGTPRC);
3735                 adapter->stats.mgpdc += E1000_READ_REG(hw, MGTPDC);
3736         }
3737
3738         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3739 }
3740
3741 /**
3742  * e1000_intr_msi - Interrupt Handler
3743  * @irq: interrupt number
3744  * @data: pointer to a network interface device structure
3745  **/
3746
3747 static irqreturn_t
3748 e1000_intr_msi(int irq, void *data)
3749 {
3750         struct net_device *netdev = data;
3751         struct e1000_adapter *adapter = netdev_priv(netdev);
3752         struct e1000_hw *hw = &adapter->hw;
3753 #ifndef CONFIG_E1000_NAPI
3754         int i;
3755 #endif
3756         uint32_t icr = E1000_READ_REG(hw, ICR);
3757
3758 #ifdef CONFIG_E1000_NAPI
3759         /* read ICR disables interrupts using IAM, so keep up with our
3760          * enable/disable accounting */
3761         atomic_inc(&adapter->irq_sem);
3762 #endif
3763         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3764                 hw->get_link_status = 1;
3765                 /* 80003ES2LAN workaround-- For packet buffer work-around on
3766                  * link down event; disable receives here in the ISR and reset
3767                  * adapter in watchdog */
3768                 if (netif_carrier_ok(netdev) &&
3769                     (adapter->hw.mac_type == e1000_80003es2lan)) {
3770                         /* disable receives */
3771                         uint32_t rctl = E1000_READ_REG(hw, RCTL);
3772                         E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
3773                 }
3774                 /* guard against interrupt when we're going down */
3775                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3776                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3777         }
3778
3779 #ifdef CONFIG_E1000_NAPI
3780         if (likely(netif_rx_schedule_prep(netdev, &adapter->napi))) {
3781                 adapter->total_tx_bytes = 0;
3782                 adapter->total_tx_packets = 0;
3783                 adapter->total_rx_bytes = 0;
3784                 adapter->total_rx_packets = 0;
3785                 __netif_rx_schedule(netdev, &adapter->napi);
3786         } else
3787                 e1000_irq_enable(adapter);
3788 #else
3789         adapter->total_tx_bytes = 0;
3790         adapter->total_rx_bytes = 0;
3791         adapter->total_tx_packets = 0;
3792         adapter->total_rx_packets = 0;
3793
3794         for (i = 0; i < E1000_MAX_INTR; i++)
3795                 if (unlikely(!adapter->clean_rx(adapter, adapter->rx_ring) &
3796                    !e1000_clean_tx_irq(adapter, adapter->tx_ring)))
3797                         break;
3798
3799         if (likely(adapter->itr_setting & 3))
3800                 e1000_set_itr(adapter);
3801 #endif
3802
3803         return IRQ_HANDLED;
3804 }
3805
3806 /**
3807  * e1000_intr - Interrupt Handler
3808  * @irq: interrupt number
3809  * @data: pointer to a network interface device structure
3810  **/
3811
3812 static irqreturn_t
3813 e1000_intr(int irq, void *data)
3814 {
3815         struct net_device *netdev = data;
3816         struct e1000_adapter *adapter = netdev_priv(netdev);
3817         struct e1000_hw *hw = &adapter->hw;
3818         uint32_t rctl, icr = E1000_READ_REG(hw, ICR);
3819 #ifndef CONFIG_E1000_NAPI
3820         int i;
3821 #endif
3822         if (unlikely(!icr))
3823                 return IRQ_NONE;  /* Not our interrupt */
3824
3825 #ifdef CONFIG_E1000_NAPI
3826         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
3827          * not set, then the adapter didn't send an interrupt */
3828         if (unlikely(hw->mac_type >= e1000_82571 &&
3829                      !(icr & E1000_ICR_INT_ASSERTED)))
3830                 return IRQ_NONE;
3831
3832         /* Interrupt Auto-Mask...upon reading ICR,
3833          * interrupts are masked.  No need for the
3834          * IMC write, but it does mean we should
3835          * account for it ASAP. */
3836         if (likely(hw->mac_type >= e1000_82571))
3837                 atomic_inc(&adapter->irq_sem);
3838 #endif
3839
3840         if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3841                 hw->get_link_status = 1;
3842                 /* 80003ES2LAN workaround--
3843                  * For packet buffer work-around on link down event;
3844                  * disable receives here in the ISR and
3845                  * reset adapter in watchdog
3846                  */
3847                 if (netif_carrier_ok(netdev) &&
3848                     (adapter->hw.mac_type == e1000_80003es2lan)) {
3849                         /* disable receives */
3850                         rctl = E1000_READ_REG(hw, RCTL);
3851                         E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
3852                 }
3853                 /* guard against interrupt when we're going down */
3854                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3855                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3856         }
3857
3858 #ifdef CONFIG_E1000_NAPI
3859         if (unlikely(hw->mac_type < e1000_82571)) {
3860                 /* disable interrupts, without the synchronize_irq bit */
3861                 atomic_inc(&adapter->irq_sem);
3862                 E1000_WRITE_REG(hw, IMC, ~0);
3863                 E1000_WRITE_FLUSH(hw);
3864         }
3865         if (likely(netif_rx_schedule_prep(netdev, &adapter->napi))) {
3866                 adapter->total_tx_bytes = 0;
3867                 adapter->total_tx_packets = 0;
3868                 adapter->total_rx_bytes = 0;
3869                 adapter->total_rx_packets = 0;
3870                 __netif_rx_schedule(netdev, &adapter->napi);
3871         } else
3872                 /* this really should not happen! if it does it is basically a
3873                  * bug, but not a hard error, so enable ints and continue */
3874                 e1000_irq_enable(adapter);
3875 #else
3876         /* Writing IMC and IMS is needed for 82547.
3877          * Due to Hub Link bus being occupied, an interrupt
3878          * de-assertion message is not able to be sent.
3879          * When an interrupt assertion message is generated later,
3880          * two messages are re-ordered and sent out.
3881          * That causes APIC to think 82547 is in de-assertion
3882          * state, while 82547 is in assertion state, resulting
3883          * in dead lock. Writing IMC forces 82547 into
3884          * de-assertion state.
3885          */
3886         if (hw->mac_type == e1000_82547 || hw->mac_type == e1000_82547_rev_2) {
3887                 atomic_inc(&adapter->irq_sem);
3888                 E1000_WRITE_REG(hw, IMC, ~0);
3889         }
3890
3891         adapter->total_tx_bytes = 0;
3892         adapter->total_rx_bytes = 0;
3893         adapter->total_tx_packets = 0;
3894         adapter->total_rx_packets = 0;
3895
3896         for (i = 0; i < E1000_MAX_INTR; i++)
3897                 if (unlikely(!adapter->clean_rx(adapter, adapter->rx_ring) &
3898                    !e1000_clean_tx_irq(adapter, adapter->tx_ring)))
3899                         break;
3900
3901         if (likely(adapter->itr_setting & 3))
3902                 e1000_set_itr(adapter);
3903
3904         if (hw->mac_type == e1000_82547 || hw->mac_type == e1000_82547_rev_2)
3905                 e1000_irq_enable(adapter);
3906
3907 #endif
3908         return IRQ_HANDLED;
3909 }
3910
3911 #ifdef CONFIG_E1000_NAPI
3912 /**
3913  * e1000_clean - NAPI Rx polling callback
3914  * @adapter: board private structure
3915  **/
3916
3917 static int
3918 e1000_clean(struct napi_struct *napi, int budget)
3919 {
3920         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
3921         struct net_device *poll_dev = adapter->netdev;
3922         int tx_cleaned = 0, work_done = 0;
3923
3924         /* Must NOT use netdev_priv macro here. */
3925         adapter = poll_dev->priv;
3926
3927         /* e1000_clean is called per-cpu.  This lock protects
3928          * tx_ring[0] from being cleaned by multiple cpus
3929          * simultaneously.  A failure obtaining the lock means
3930          * tx_ring[0] is currently being cleaned anyway. */
3931         if (spin_trylock(&adapter->tx_queue_lock)) {
3932                 tx_cleaned = e1000_clean_tx_irq(adapter,
3933                                                 &adapter->tx_ring[0]);
3934                 spin_unlock(&adapter->tx_queue_lock);
3935         }
3936
3937         adapter->clean_rx(adapter, &adapter->rx_ring[0],
3938                           &work_done, budget);
3939
3940         if (tx_cleaned)
3941                 work_done = budget;
3942
3943         /* If budget not fully consumed, exit the polling mode */
3944         if (work_done < budget) {
3945                 if (likely(adapter->itr_setting & 3))
3946                         e1000_set_itr(adapter);
3947                 netif_rx_complete(poll_dev, napi);
3948                 e1000_irq_enable(adapter);
3949         }
3950
3951         return work_done;
3952 }
3953
3954 #endif
3955 /**
3956  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3957  * @adapter: board private structure
3958  **/
3959
3960 static boolean_t
3961 e1000_clean_tx_irq(struct e1000_adapter *adapter,
3962                    struct e1000_tx_ring *tx_ring)
3963 {
3964         struct net_device *netdev = adapter->netdev;
3965         struct e1000_tx_desc *tx_desc, *eop_desc;
3966         struct e1000_buffer *buffer_info;
3967         unsigned int i, eop;
3968 #ifdef CONFIG_E1000_NAPI
3969         unsigned int count = 0;
3970 #endif
3971         boolean_t cleaned = FALSE;
3972         unsigned int total_tx_bytes=0, total_tx_packets=0;
3973
3974         i = tx_ring->next_to_clean;
3975         eop = tx_ring->buffer_info[i].next_to_watch;
3976         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3977
3978         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
3979                 for (cleaned = FALSE; !cleaned; ) {
3980                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3981                         buffer_info = &tx_ring->buffer_info[i];
3982                         cleaned = (i == eop);
3983
3984                         if (cleaned) {
3985                                 struct sk_buff *skb = buffer_info->skb;
3986                                 unsigned int segs, bytecount;
3987                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3988                                 /* multiply data chunks by size of headers */
3989                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3990                                             skb->len;
3991                                 total_tx_packets += segs;
3992                                 total_tx_bytes += bytecount;
3993                         }
3994                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3995                         tx_desc->upper.data = 0;
3996
3997                         if (unlikely(++i == tx_ring->count)) i = 0;
3998                 }
3999
4000                 eop = tx_ring->buffer_info[i].next_to_watch;
4001                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
4002 #ifdef CONFIG_E1000_NAPI
4003 #define E1000_TX_WEIGHT 64
4004                 /* weight of a sort for tx, to avoid endless transmit cleanup */
4005                 if (count++ == E1000_TX_WEIGHT) break;
4006 #endif
4007         }
4008
4009         tx_ring->next_to_clean = i;
4010
4011 #define TX_WAKE_THRESHOLD 32
4012         if (unlikely(cleaned && netif_carrier_ok(netdev) &&
4013                      E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
4014                 /* Make sure that anybody stopping the queue after this
4015                  * sees the new next_to_clean.
4016                  */
4017                 smp_mb();
4018                 if (netif_queue_stopped(netdev)) {
4019                         netif_wake_queue(netdev);
4020                         ++adapter->restart_queue;
4021                 }
4022         }
4023
4024         if (adapter->detect_tx_hung) {
4025                 /* Detect a transmit hang in hardware, this serializes the
4026                  * check with the clearing of time_stamp and movement of i */
4027                 adapter->detect_tx_hung = FALSE;
4028                 if (tx_ring->buffer_info[eop].dma &&
4029                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
4030                                (adapter->tx_timeout_factor * HZ))
4031                     && !(E1000_READ_REG(&adapter->hw, STATUS) &
4032                          E1000_STATUS_TXOFF)) {
4033
4034                         /* detected Tx unit hang */
4035                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
4036                                         "  Tx Queue             <%lu>\n"
4037                                         "  TDH                  <%x>\n"
4038                                         "  TDT                  <%x>\n"
4039                                         "  next_to_use          <%x>\n"
4040                                         "  next_to_clean        <%x>\n"
4041                                         "buffer_info[next_to_clean]\n"
4042                                         "  time_stamp           <%lx>\n"
4043                                         "  next_to_watch        <%x>\n"
4044                                         "  jiffies              <%lx>\n"
4045                                         "  next_to_watch.status <%x>\n",
4046                                 (unsigned long)((tx_ring - adapter->tx_ring) /
4047                                         sizeof(struct e1000_tx_ring)),
4048                                 readl(adapter->hw.hw_addr + tx_ring->tdh),
4049                                 readl(adapter->hw.hw_addr + tx_ring->tdt),
4050                                 tx_ring->next_to_use,
4051                                 tx_ring->next_to_clean,
4052                                 tx_ring->buffer_info[eop].time_stamp,
4053                                 eop,
4054                                 jiffies,
4055                                 eop_desc->upper.fields.status);
4056                         netif_stop_queue(netdev);
4057                 }
4058         }
4059         adapter->total_tx_bytes += total_tx_bytes;
4060         adapter->total_tx_packets += total_tx_packets;
4061         return cleaned;
4062 }
4063
4064 /**
4065  * e1000_rx_checksum - Receive Checksum Offload for 82543
4066  * @adapter:     board private structure
4067  * @status_err:  receive descriptor status and error fields
4068  * @csum:        receive descriptor csum field
4069  * @sk_buff:     socket buffer with received data
4070  **/
4071
4072 static void
4073 e1000_rx_checksum(struct e1000_adapter *adapter,
4074                   uint32_t status_err, uint32_t csum,
4075                   struct sk_buff *skb)
4076 {
4077         uint16_t status = (uint16_t)status_err;
4078         uint8_t errors = (uint8_t)(status_err >> 24);
4079         skb->ip_summed = CHECKSUM_NONE;
4080
4081         /* 82543 or newer only */
4082         if (unlikely(adapter->hw.mac_type < e1000_82543)) return;
4083         /* Ignore Checksum bit is set */
4084         if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
4085         /* TCP/UDP checksum error bit is set */
4086         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
4087                 /* let the stack verify checksum errors */
4088                 adapter->hw_csum_err++;
4089                 return;
4090         }
4091         /* TCP/UDP Checksum has not been calculated */
4092         if (adapter->hw.mac_type <= e1000_82547_rev_2) {
4093                 if (!(status & E1000_RXD_STAT_TCPCS))
4094                         return;
4095         } else {
4096                 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
4097                         return;
4098         }
4099         /* It must be a TCP or UDP packet with a valid checksum */
4100         if (likely(status & E1000_RXD_STAT_TCPCS)) {
4101                 /* TCP checksum is good */
4102                 skb->ip_summed = CHECKSUM_UNNECESSARY;
4103         } else if (adapter->hw.mac_type > e1000_82547_rev_2) {
4104                 /* IP fragment with UDP payload */
4105                 /* Hardware complements the payload checksum, so we undo it
4106                  * and then put the value in host order for further stack use.
4107                  */
4108                 csum = ntohl(csum ^ 0xFFFF);
4109                 skb->csum = csum;
4110                 skb->ip_summed = CHECKSUM_COMPLETE;
4111         }
4112         adapter->hw_csum_good++;
4113 }
4114
4115 /**
4116  * e1000_clean_rx_irq - Send received data up the network stack; legacy
4117  * @adapter: board private structure
4118  **/
4119
4120 static boolean_t
4121 #ifdef CONFIG_E1000_NAPI
4122 e1000_clean_rx_irq(struct e1000_adapter *adapter,
4123                    struct e1000_rx_ring *rx_ring,
4124                    int *work_done, int work_to_do)
4125 #else
4126 e1000_clean_rx_irq(struct e1000_adapter *adapter,
4127                    struct e1000_rx_ring *rx_ring)
4128 #endif
4129 {
4130         struct net_device *netdev = adapter->netdev;
4131         struct pci_dev *pdev = adapter->pdev;
4132         struct e1000_rx_desc *rx_desc, *next_rxd;
4133         struct e1000_buffer *buffer_info, *next_buffer;
4134         unsigned long flags;
4135         uint32_t length;
4136         uint8_t last_byte;
4137         unsigned int i;
4138         int cleaned_count = 0;
4139         boolean_t cleaned = FALSE;
4140         unsigned int total_rx_bytes=0, total_rx_packets=0;
4141
4142         i = rx_ring->next_to_clean;
4143         rx_desc = E1000_RX_DESC(*rx_ring, i);
4144         buffer_info = &rx_ring->buffer_info[i];
4145
4146         while (rx_desc->status & E1000_RXD_STAT_DD) {
4147                 struct sk_buff *skb;
4148                 u8 status;
4149
4150 #ifdef CONFIG_E1000_NAPI
4151                 if (*work_done >= work_to_do)
4152                         break;
4153                 (*work_done)++;
4154 #endif
4155                 status = rx_desc->status;
4156                 skb = buffer_info->skb;
4157                 buffer_info->skb = NULL;
4158
4159                 prefetch(skb->data - NET_IP_ALIGN);
4160
4161                 if (++i == rx_ring->count) i = 0;
4162                 next_rxd = E1000_RX_DESC(*rx_ring, i);
4163                 prefetch(next_rxd);
4164
4165                 next_buffer = &rx_ring->buffer_info[i];
4166
4167                 cleaned = TRUE;
4168                 cleaned_count++;
4169                 pci_unmap_single(pdev,
4170                                  buffer_info->dma,
4171                                  buffer_info->length,
4172                                  PCI_DMA_FROMDEVICE);
4173
4174                 length = le16_to_cpu(rx_desc->length);
4175
4176                 if (unlikely(!(status & E1000_RXD_STAT_EOP))) {
4177                         /* All receives must fit into a single buffer */
4178                         E1000_DBG("%s: Receive packet consumed multiple"
4179                                   " buffers\n", netdev->name);
4180                         /* recycle */
4181                         buffer_info->skb = skb;
4182                         goto next_desc;
4183                 }
4184
4185                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
4186                         last_byte = *(skb->data + length - 1);
4187                         if (TBI_ACCEPT(&adapter->hw, status,
4188                                       rx_desc->errors, length, last_byte)) {
4189                                 spin_lock_irqsave(&adapter->stats_lock, flags);
4190                                 e1000_tbi_adjust_stats(&adapter->hw,
4191                                                        &adapter->stats,
4192                                                        length, skb->data);
4193                                 spin_unlock_irqrestore(&adapter->stats_lock,
4194                                                        flags);
4195                                 length--;
4196                         } else {
4197                                 /* recycle */
4198                                 buffer_info->skb = skb;
4199                                 goto next_desc;
4200                         }
4201                 }
4202
4203                 /* adjust length to remove Ethernet CRC, this must be
4204                  * done after the TBI_ACCEPT workaround above */
4205                 length -= 4;
4206
4207                 /* probably a little skewed due to removing CRC */
4208                 total_rx_bytes += length;
4209                 total_rx_packets++;
4210
4211                 /* code added for copybreak, this should improve
4212                  * performance for small packets with large amounts
4213                  * of reassembly being done in the stack */
4214                 if (length < copybreak) {
4215                         struct sk_buff *new_skb =
4216                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
4217                         if (new_skb) {
4218                                 skb_reserve(new_skb, NET_IP_ALIGN);
4219                                 skb_copy_to_linear_data_offset(new_skb,
4220                                                                -NET_IP_ALIGN,
4221                                                                (skb->data -
4222                                                                 NET_IP_ALIGN),
4223                                                                (length +
4224                                                                 NET_IP_ALIGN));
4225                                 /* save the skb in buffer_info as good */
4226                                 buffer_info->skb = skb;
4227                                 skb = new_skb;
4228                         }
4229                         /* else just continue with the old one */
4230                 }
4231                 /* end copybreak code */
4232                 skb_put(skb, length);
4233
4234                 /* Receive Checksum Offload */
4235                 e1000_rx_checksum(adapter,
4236                                   (uint32_t)(status) |
4237                                   ((uint32_t)(rx_desc->errors) << 24),
4238                                   le16_to_cpu(rx_desc->csum), skb);
4239
4240                 skb->protocol = eth_type_trans(skb, netdev);
4241 #ifdef CONFIG_E1000_NAPI
4242                 if (unlikely(adapter->vlgrp &&
4243                             (status & E1000_RXD_STAT_VP))) {
4244                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
4245                                                  le16_to_cpu(rx_desc->special) &
4246                                                  E1000_RXD_SPC_VLAN_MASK);
4247                 } else {
4248                         netif_receive_skb(skb);
4249                 }
4250 #else /* CONFIG_E1000_NAPI */
4251                 if (unlikely(adapter->vlgrp &&
4252                             (status & E1000_RXD_STAT_VP))) {
4253                         vlan_hwaccel_rx(skb, adapter->vlgrp,
4254                                         le16_to_cpu(rx_desc->special) &
4255                                         E1000_RXD_SPC_VLAN_MASK);
4256                 } else {
4257                         netif_rx(skb);
4258                 }
4259 #endif /* CONFIG_E1000_NAPI */
4260                 netdev->last_rx = jiffies;
4261
4262 next_desc:
4263                 rx_desc->status = 0;
4264
4265                 /* return some buffers to hardware, one at a time is too slow */
4266                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4267                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4268                         cleaned_count = 0;
4269                 }
4270
4271                 /* use prefetched values */
4272                 rx_desc = next_rxd;
4273                 buffer_info = next_buffer;
4274         }
4275         rx_ring->next_to_clean = i;
4276
4277         cleaned_count = E1000_DESC_UNUSED(rx_ring);
4278         if (cleaned_count)
4279                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4280
4281         adapter->total_rx_packets += total_rx_packets;
4282         adapter->total_rx_bytes += total_rx_bytes;
4283         return cleaned;
4284 }
4285
4286 /**
4287  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
4288  * @adapter: board private structure
4289  **/
4290
4291 static boolean_t
4292 #ifdef CONFIG_E1000_NAPI
4293 e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
4294                       struct e1000_rx_ring *rx_ring,
4295                       int *work_done, int work_to_do)
4296 #else
4297 e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
4298                       struct e1000_rx_ring *rx_ring)
4299 #endif
4300 {
4301         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
4302         struct net_device *netdev = adapter->netdev;
4303         struct pci_dev *pdev = adapter->pdev;
4304         struct e1000_buffer *buffer_info, *next_buffer;
4305         struct e1000_ps_page *ps_page;
4306         struct e1000_ps_page_dma *ps_page_dma;
4307         struct sk_buff *skb;
4308         unsigned int i, j;
4309         uint32_t length, staterr;
4310         int cleaned_count = 0;
4311         boolean_t cleaned = FALSE;
4312         unsigned int total_rx_bytes=0, total_rx_packets=0;
4313
4314         i = rx_ring->next_to_clean;
4315         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
4316         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
4317         buffer_info = &rx_ring->buffer_info[i];
4318
4319         while (staterr & E1000_RXD_STAT_DD) {
4320                 ps_page = &rx_ring->ps_page[i];
4321                 ps_page_dma = &rx_ring->ps_page_dma[i];
4322 #ifdef CONFIG_E1000_NAPI
4323                 if (unlikely(*work_done >= work_to_do))
4324                         break;
4325                 (*work_done)++;
4326 #endif
4327                 skb = buffer_info->skb;
4328
4329                 /* in the packet split case this is header only */
4330                 prefetch(skb->data - NET_IP_ALIGN);
4331
4332                 if (++i == rx_ring->count) i = 0;
4333                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
4334                 prefetch(next_rxd);
4335
4336                 next_buffer = &rx_ring->buffer_info[i];
4337
4338                 cleaned = TRUE;
4339                 cleaned_count++;
4340                 pci_unmap_single(pdev, buffer_info->dma,
4341                                  buffer_info->length,
4342                                  PCI_DMA_FROMDEVICE);
4343
4344                 if (unlikely(!(staterr & E1000_RXD_STAT_EOP))) {
4345                         E1000_DBG("%s: Packet Split buffers didn't pick up"
4346                                   " the full packet\n", netdev->name);
4347                         dev_kfree_skb_irq(skb);
4348                         goto next_desc;
4349                 }
4350
4351                 if (unlikely(staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK)) {
4352                         dev_kfree_skb_irq(skb);
4353                         goto next_desc;
4354                 }
4355
4356                 length = le16_to_cpu(rx_desc->wb.middle.length0);
4357
4358                 if (unlikely(!length)) {
4359                         E1000_DBG("%s: Last part of the packet spanning"
4360                                   " multiple descriptors\n", netdev->name);
4361                         dev_kfree_skb_irq(skb);
4362                         goto next_desc;
4363                 }
4364
4365                 /* Good Receive */
4366                 skb_put(skb, length);
4367
4368                 {
4369                 /* this looks ugly, but it seems compiler issues make it
4370                    more efficient than reusing j */
4371                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
4372
4373                 /* page alloc/put takes too long and effects small packet
4374                  * throughput, so unsplit small packets and save the alloc/put*/
4375                 if (l1 && (l1 <= copybreak) && ((length + l1) <= adapter->rx_ps_bsize0)) {
4376                         u8 *vaddr;
4377                         /* there is no documentation about how to call
4378                          * kmap_atomic, so we can't hold the mapping
4379                          * very long */
4380                         pci_dma_sync_single_for_cpu(pdev,
4381                                 ps_page_dma->ps_page_dma[0],
4382                                 PAGE_SIZE,
4383                                 PCI_DMA_FROMDEVICE);
4384                         vaddr = kmap_atomic(ps_page->ps_page[0],
4385                                             KM_SKB_DATA_SOFTIRQ);
4386                         memcpy(skb_tail_pointer(skb), vaddr, l1);
4387                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
4388                         pci_dma_sync_single_for_device(pdev,
4389                                 ps_page_dma->ps_page_dma[0],
4390                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
4391                         /* remove the CRC */
4392                         l1 -= 4;
4393                         skb_put(skb, l1);
4394                         goto copydone;
4395                 } /* if */
4396                 }
4397
4398                 for (j = 0; j < adapter->rx_ps_pages; j++) {
4399                         if (!(length= le16_to_cpu(rx_desc->wb.upper.length[j])))
4400                                 break;
4401                         pci_unmap_page(pdev, ps_page_dma->ps_page_dma[j],
4402                                         PAGE_SIZE, PCI_DMA_FROMDEVICE);
4403                         ps_page_dma->ps_page_dma[j] = 0;
4404                         skb_fill_page_desc(skb, j, ps_page->ps_page[j], 0,
4405                                            length);
4406                         ps_page->ps_page[j] = NULL;
4407                         skb->len += length;
4408                         skb->data_len += length;
4409                         skb->truesize += length;
4410                 }
4411
4412                 /* strip the ethernet crc, problem is we're using pages now so
4413                  * this whole operation can get a little cpu intensive */
4414                 pskb_trim(skb, skb->len - 4);
4415
4416 copydone:
4417                 total_rx_bytes += skb->len;
4418                 total_rx_packets++;
4419
4420                 e1000_rx_checksum(adapter, staterr,
4421                                   le16_to_cpu(rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
4422                 skb->protocol = eth_type_trans(skb, netdev);
4423
4424                 if (likely(rx_desc->wb.upper.header_status &
4425                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP)))
4426                         adapter->rx_hdr_split++;
4427 #ifdef CONFIG_E1000_NAPI
4428                 if (unlikely(adapter->vlgrp && (staterr & E1000_RXD_STAT_VP))) {
4429                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
4430                                 le16_to_cpu(rx_desc->wb.middle.vlan) &
4431                                 E1000_RXD_SPC_VLAN_MASK);
4432                 } else {
4433                         netif_receive_skb(skb);
4434                 }
4435 #else /* CONFIG_E1000_NAPI */
4436                 if (unlikely(adapter->vlgrp && (staterr & E1000_RXD_STAT_VP))) {
4437                         vlan_hwaccel_rx(skb, adapter->vlgrp,
4438                                 le16_to_cpu(rx_desc->wb.middle.vlan) &
4439                                 E1000_RXD_SPC_VLAN_MASK);
4440                 } else {
4441                         netif_rx(skb);
4442                 }
4443 #endif /* CONFIG_E1000_NAPI */
4444                 netdev->last_rx = jiffies;
4445
4446 next_desc:
4447                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
4448                 buffer_info->skb = NULL;
4449
4450                 /* return some buffers to hardware, one at a time is too slow */
4451                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4452                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4453                         cleaned_count = 0;
4454                 }
4455
4456                 /* use prefetched values */
4457                 rx_desc = next_rxd;
4458                 buffer_info = next_buffer;
4459
4460                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
4461         }
4462         rx_ring->next_to_clean = i;
4463
4464         cleaned_count = E1000_DESC_UNUSED(rx_ring);
4465         if (cleaned_count)
4466                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4467
4468         adapter->total_rx_packets += total_rx_packets;
4469         adapter->total_rx_bytes += total_rx_bytes;
4470         return cleaned;
4471 }
4472
4473 /**
4474  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4475  * @adapter: address of board private structure
4476  **/
4477
4478 static void
4479 e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4480                        struct e1000_rx_ring *rx_ring,
4481                        int cleaned_count)
4482 {
4483         struct net_device *netdev = adapter->netdev;
4484         struct pci_dev *pdev = adapter->pdev;
4485         struct e1000_rx_desc *rx_desc;
4486         struct e1000_buffer *buffer_info;
4487         struct sk_buff *skb;
4488         unsigned int i;
4489         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
4490
4491         i = rx_ring->next_to_use;
4492         buffer_info = &rx_ring->buffer_info[i];
4493
4494         while (cleaned_count--) {
4495                 skb = buffer_info->skb;
4496                 if (skb) {
4497                         skb_trim(skb, 0);
4498                         goto map_skb;
4499                 }
4500
4501                 skb = netdev_alloc_skb(netdev, bufsz);
4502                 if (unlikely(!skb)) {
4503                         /* Better luck next round */
4504                         adapter->alloc_rx_buff_failed++;
4505                         break;
4506                 }
4507
4508                 /* Fix for errata 23, can't cross 64kB boundary */
4509                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4510                         struct sk_buff *oldskb = skb;
4511                         DPRINTK(RX_ERR, ERR, "skb align check failed: %u bytes "
4512                                              "at %p\n", bufsz, skb->data);
4513                         /* Try again, without freeing the previous */
4514                         skb = netdev_alloc_skb(netdev, bufsz);
4515                         /* Failed allocation, critical failure */
4516                         if (!skb) {
4517                                 dev_kfree_skb(oldskb);
4518                                 break;
4519                         }
4520
4521                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4522                                 /* give up */
4523                                 dev_kfree_skb(skb);
4524                                 dev_kfree_skb(oldskb);
4525                                 break; /* while !buffer_info->skb */
4526                         }
4527
4528                         /* Use new allocation */
4529                         dev_kfree_skb(oldskb);
4530                 }
4531                 /* Make buffer alignment 2 beyond a 16 byte boundary
4532                  * this will result in a 16 byte aligned IP header after
4533                  * the 14 byte MAC header is removed
4534                  */
4535                 skb_reserve(skb, NET_IP_ALIGN);
4536
4537                 buffer_info->skb = skb;
4538                 buffer_info->length = adapter->rx_buffer_len;
4539 map_skb:
4540                 buffer_info->dma = pci_map_single(pdev,
4541                                                   skb->data,
4542                                                   adapter->rx_buffer_len,
4543                                                   PCI_DMA_FROMDEVICE);
4544
4545                 /* Fix for errata 23, can't cross 64kB boundary */
4546                 if (!e1000_check_64k_bound(adapter,
4547                                         (void *)(unsigned long)buffer_info->dma,
4548                                         adapter->rx_buffer_len)) {
4549                         DPRINTK(RX_ERR, ERR,
4550                                 "dma align check failed: %u bytes at %p\n",
4551                                 adapter->rx_buffer_len,
4552                                 (void *)(unsigned long)buffer_info->dma);
4553                         dev_kfree_skb(skb);
4554                         buffer_info->skb = NULL;
4555
4556                         pci_unmap_single(pdev, buffer_info->dma,
4557                                          adapter->rx_buffer_len,
4558                                          PCI_DMA_FROMDEVICE);
4559
4560                         break; /* while !buffer_info->skb */
4561                 }
4562                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4563                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4564
4565                 if (unlikely(++i == rx_ring->count))
4566                         i = 0;
4567                 buffer_info = &rx_ring->buffer_info[i];
4568         }
4569
4570         if (likely(rx_ring->next_to_use != i)) {
4571                 rx_ring->next_to_use = i;
4572                 if (unlikely(i-- == 0))
4573                         i = (rx_ring->count - 1);
4574
4575                 /* Force memory writes to complete before letting h/w
4576                  * know there are new descriptors to fetch.  (Only
4577                  * applicable for weak-ordered memory model archs,
4578                  * such as IA-64). */
4579                 wmb();
4580                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4581         }
4582 }
4583
4584 /**
4585  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
4586  * @adapter: address of board private structure
4587  **/
4588
4589 static void
4590 e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
4591                           struct e1000_rx_ring *rx_ring,
4592                           int cleaned_count)
4593 {
4594         struct net_device *netdev = adapter->netdev;
4595         struct pci_dev *pdev = adapter->pdev;
4596         union e1000_rx_desc_packet_split *rx_desc;
4597         struct e1000_buffer *buffer_info;
4598         struct e1000_ps_page *ps_page;
4599         struct e1000_ps_page_dma *ps_page_dma;
4600         struct sk_buff *skb;
4601         unsigned int i, j;
4602
4603         i = rx_ring->next_to_use;
4604         buffer_info = &rx_ring->buffer_info[i];
4605         ps_page = &rx_ring->ps_page[i];
4606         ps_page_dma = &rx_ring->ps_page_dma[i];
4607
4608         while (cleaned_count--) {
4609                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
4610
4611                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
4612                         if (j < adapter->rx_ps_pages) {
4613                                 if (likely(!ps_page->ps_page[j])) {
4614                                         ps_page->ps_page[j] =
4615                                                 alloc_page(GFP_ATOMIC);
4616                                         if (unlikely(!ps_page->ps_page[j])) {
4617                                                 adapter->alloc_rx_buff_failed++;
4618                                                 goto no_buffers;
4619                                         }
4620                                         ps_page_dma->ps_page_dma[j] =
4621                                                 pci_map_page(pdev,
4622                                                             ps_page->ps_page[j],
4623                                                             0, PAGE_SIZE,
4624                                                             PCI_DMA_FROMDEVICE);
4625                                 }
4626                                 /* Refresh the desc even if buffer_addrs didn't
4627                                  * change because each write-back erases
4628                                  * this info.
4629                                  */
4630                                 rx_desc->read.buffer_addr[j+1] =
4631                                      cpu_to_le64(ps_page_dma->ps_page_dma[j]);
4632                         } else
4633                                 rx_desc->read.buffer_addr[j+1] = ~0;
4634                 }
4635
4636                 skb = netdev_alloc_skb(netdev,
4637                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
4638
4639                 if (unlikely(!skb)) {
4640                         adapter->alloc_rx_buff_failed++;
4641                         break;
4642                 }
4643
4644                 /* Make buffer alignment 2 beyond a 16 byte boundary
4645                  * this will result in a 16 byte aligned IP header after
4646                  * the 14 byte MAC header is removed
4647                  */
4648                 skb_reserve(skb, NET_IP_ALIGN);
4649
4650                 buffer_info->skb = skb;
4651                 buffer_info->length = adapter->rx_ps_bsize0;
4652                 buffer_info->dma = pci_map_single(pdev, skb->data,
4653                                                   adapter->rx_ps_bsize0,
4654                                                   PCI_DMA_FROMDEVICE);
4655
4656                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
4657
4658                 if (unlikely(++i == rx_ring->count)) i = 0;
4659                 buffer_info = &rx_ring->buffer_info[i];
4660                 ps_page = &rx_ring->ps_page[i];
4661                 ps_page_dma = &rx_ring->ps_page_dma[i];
4662         }
4663
4664 no_buffers:
4665         if (likely(rx_ring->next_to_use != i)) {
4666                 rx_ring->next_to_use = i;
4667                 if (unlikely(i-- == 0)) i = (rx_ring->count - 1);
4668
4669                 /* Force memory writes to complete before letting h/w
4670                  * know there are new descriptors to fetch.  (Only
4671                  * applicable for weak-ordered memory model archs,
4672                  * such as IA-64). */
4673                 wmb();
4674                 /* Hardware increments by 16 bytes, but packet split
4675                  * descriptors are 32 bytes...so we increment tail
4676                  * twice as much.
4677                  */
4678                 writel(i<<1, adapter->hw.hw_addr + rx_ring->rdt);
4679         }
4680 }
4681
4682 /**
4683  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4684  * @adapter:
4685  **/
4686
4687 static void
4688 e1000_smartspeed(struct e1000_adapter *adapter)
4689 {
4690         uint16_t phy_status;
4691         uint16_t phy_ctrl;
4692
4693         if ((adapter->hw.phy_type != e1000_phy_igp) || !adapter->hw.autoneg ||
4694            !(adapter->hw.autoneg_advertised & ADVERTISE_1000_FULL))
4695                 return;
4696
4697         if (adapter->smartspeed == 0) {
4698                 /* If Master/Slave config fault is asserted twice,
4699                  * we assume back-to-back */
4700                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_status);
4701                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4702                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_status);
4703                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4704                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_ctrl);
4705                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4706                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4707                         e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL,
4708                                             phy_ctrl);
4709                         adapter->smartspeed++;
4710                         if (!e1000_phy_setup_autoneg(&adapter->hw) &&
4711                            !e1000_read_phy_reg(&adapter->hw, PHY_CTRL,
4712                                                &phy_ctrl)) {
4713                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4714                                              MII_CR_RESTART_AUTO_NEG);
4715                                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL,
4716                                                     phy_ctrl);
4717                         }
4718                 }
4719                 return;
4720         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4721                 /* If still no link, perhaps using 2/3 pair cable */
4722                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_ctrl);
4723                 phy_ctrl |= CR_1000T_MS_ENABLE;
4724                 e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL, phy_ctrl);
4725                 if (!e1000_phy_setup_autoneg(&adapter->hw) &&
4726                    !e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &phy_ctrl)) {
4727                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4728                                      MII_CR_RESTART_AUTO_NEG);
4729                         e1000_write_phy_reg(&adapter->hw, PHY_CTRL, phy_ctrl);
4730                 }
4731         }
4732         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4733         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4734                 adapter->smartspeed = 0;
4735 }
4736
4737 /**
4738  * e1000_ioctl -
4739  * @netdev:
4740  * @ifreq:
4741  * @cmd:
4742  **/
4743
4744 static int
4745 e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4746 {
4747         switch (cmd) {
4748         case SIOCGMIIPHY:
4749         case SIOCGMIIREG:
4750         case SIOCSMIIREG:
4751                 return e1000_mii_ioctl(netdev, ifr, cmd);
4752         default:
4753                 return -EOPNOTSUPP;
4754         }
4755 }
4756
4757 /**
4758  * e1000_mii_ioctl -
4759  * @netdev:
4760  * @ifreq:
4761  * @cmd:
4762  **/
4763
4764 static int
4765 e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4766 {
4767         struct e1000_adapter *adapter = netdev_priv(netdev);
4768         struct mii_ioctl_data *data = if_mii(ifr);
4769         int retval;
4770         uint16_t mii_reg;
4771         uint16_t spddplx;
4772         unsigned long flags;
4773
4774         if (adapter->hw.media_type != e1000_media_type_copper)
4775                 return -EOPNOTSUPP;
4776
4777         switch (cmd) {
4778         case SIOCGMIIPHY:
4779                 data->phy_id = adapter->hw.phy_addr;
4780                 break;
4781         case SIOCGMIIREG:
4782                 if (!capable(CAP_NET_ADMIN))
4783                         return -EPERM;
4784                 spin_lock_irqsave(&adapter->stats_lock, flags);
4785                 if (e1000_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
4786                                    &data->val_out)) {
4787                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4788                         return -EIO;
4789                 }
4790                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4791                 break;
4792         case SIOCSMIIREG:
4793                 if (!capable(CAP_NET_ADMIN))
4794                         return -EPERM;
4795                 if (data->reg_num & ~(0x1F))
4796                         return -EFAULT;
4797                 mii_reg = data->val_in;
4798                 spin_lock_irqsave(&adapter->stats_lock, flags);
4799                 if (e1000_write_phy_reg(&adapter->hw, data->reg_num,
4800                                         mii_reg)) {
4801                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4802                         return -EIO;
4803                 }
4804                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4805                 if (adapter->hw.media_type == e1000_media_type_copper) {
4806                         switch (data->reg_num) {
4807                         case PHY_CTRL:
4808                                 if (mii_reg & MII_CR_POWER_DOWN)
4809                                         break;
4810                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4811                                         adapter->hw.autoneg = 1;
4812                                         adapter->hw.autoneg_advertised = 0x2F;
4813                                 } else {
4814                                         if (mii_reg & 0x40)
4815                                                 spddplx = SPEED_1000;
4816                                         else if (mii_reg & 0x2000)
4817                                                 spddplx = SPEED_100;
4818                                         else
4819                                                 spddplx = SPEED_10;
4820                                         spddplx += (mii_reg & 0x100)
4821                                                    ? DUPLEX_FULL :
4822                                                    DUPLEX_HALF;
4823                                         retval = e1000_set_spd_dplx(adapter,
4824                                                                     spddplx);
4825                                         if (retval)
4826                                                 return retval;
4827                                 }
4828                                 if (netif_running(adapter->netdev))
4829                                         e1000_reinit_locked(adapter);
4830                                 else
4831                                         e1000_reset(adapter);
4832                                 break;
4833                         case M88E1000_PHY_SPEC_CTRL:
4834                         case M88E1000_EXT_PHY_SPEC_CTRL:
4835                                 if (e1000_phy_reset(&adapter->hw))
4836                                         return -EIO;
4837                                 break;
4838                         }
4839                 } else {
4840                         switch (data->reg_num) {
4841                         case PHY_CTRL:
4842                                 if (mii_reg & MII_CR_POWER_DOWN)
4843                                         break;
4844                                 if (netif_running(adapter->netdev))
4845                                         e1000_reinit_locked(adapter);
4846                                 else
4847                                         e1000_reset(adapter);
4848                                 break;
4849                         }
4850                 }
4851                 break;
4852         default:
4853                 return -EOPNOTSUPP;
4854         }
4855         return E1000_SUCCESS;
4856 }
4857
4858 void
4859 e1000_pci_set_mwi(struct e1000_hw *hw)
4860 {
4861         struct e1000_adapter *adapter = hw->back;
4862         int ret_val = pci_set_mwi(adapter->pdev);
4863
4864         if (ret_val)
4865                 DPRINTK(PROBE, ERR, "Error in setting MWI\n");
4866 }
4867
4868 void
4869 e1000_pci_clear_mwi(struct e1000_hw *hw)
4870 {
4871         struct e1000_adapter *adapter = hw->back;
4872
4873         pci_clear_mwi(adapter->pdev);
4874 }
4875
4876 void
4877 e1000_read_pci_cfg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4878 {
4879         struct e1000_adapter *adapter = hw->back;
4880
4881         pci_read_config_word(adapter->pdev, reg, value);
4882 }
4883
4884 void
4885 e1000_write_pci_cfg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4886 {
4887         struct e1000_adapter *adapter = hw->back;
4888
4889         pci_write_config_word(adapter->pdev, reg, *value);
4890 }
4891
4892 int
4893 e1000_pcix_get_mmrbc(struct e1000_hw *hw)
4894 {
4895         struct e1000_adapter *adapter = hw->back;
4896         return pcix_get_mmrbc(adapter->pdev);
4897 }
4898
4899 void
4900 e1000_pcix_set_mmrbc(struct e1000_hw *hw, int mmrbc)
4901 {
4902         struct e1000_adapter *adapter = hw->back;
4903         pcix_set_mmrbc(adapter->pdev, mmrbc);
4904 }
4905
4906 int32_t
4907 e1000_read_pcie_cap_reg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4908 {
4909     struct e1000_adapter *adapter = hw->back;
4910     uint16_t cap_offset;
4911
4912     cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
4913     if (!cap_offset)
4914         return -E1000_ERR_CONFIG;
4915
4916     pci_read_config_word(adapter->pdev, cap_offset + reg, value);
4917
4918     return E1000_SUCCESS;
4919 }
4920
4921 void
4922 e1000_io_write(struct e1000_hw *hw, unsigned long port, uint32_t value)
4923 {
4924         outl(value, port);
4925 }
4926
4927 static void
4928 e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
4929 {
4930         struct e1000_adapter *adapter = netdev_priv(netdev);
4931         uint32_t ctrl, rctl;
4932
4933         e1000_irq_disable(adapter);
4934         adapter->vlgrp = grp;
4935
4936         if (grp) {
4937                 /* enable VLAN tag insert/strip */
4938                 ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4939                 ctrl |= E1000_CTRL_VME;
4940                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4941
4942                 if (adapter->hw.mac_type != e1000_ich8lan) {
4943                         /* enable VLAN receive filtering */
4944                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
4945                         rctl |= E1000_RCTL_VFE;
4946                         rctl &= ~E1000_RCTL_CFIEN;
4947                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4948                         e1000_update_mng_vlan(adapter);
4949                 }
4950         } else {
4951                 /* disable VLAN tag insert/strip */
4952                 ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4953                 ctrl &= ~E1000_CTRL_VME;
4954                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4955
4956                 if (adapter->hw.mac_type != e1000_ich8lan) {
4957                         /* disable VLAN filtering */
4958                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
4959                         rctl &= ~E1000_RCTL_VFE;
4960                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4961                         if (adapter->mng_vlan_id !=
4962                             (uint16_t)E1000_MNG_VLAN_NONE) {
4963                                 e1000_vlan_rx_kill_vid(netdev,
4964                                                        adapter->mng_vlan_id);
4965                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4966                         }
4967                 }
4968         }
4969
4970         e1000_irq_enable(adapter);
4971 }
4972
4973 static void
4974 e1000_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid)
4975 {
4976         struct e1000_adapter *adapter = netdev_priv(netdev);
4977         uint32_t vfta, index;
4978
4979         if ((adapter->hw.mng_cookie.status &
4980              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4981             (vid == adapter->mng_vlan_id))
4982                 return;
4983         /* add VID to filter table */
4984         index = (vid >> 5) & 0x7F;
4985         vfta = E1000_READ_REG_ARRAY(&adapter->hw, VFTA, index);
4986         vfta |= (1 << (vid & 0x1F));
4987         e1000_write_vfta(&adapter->hw, index, vfta);
4988 }
4989
4990 static void
4991 e1000_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid)
4992 {
4993         struct e1000_adapter *adapter = netdev_priv(netdev);
4994         uint32_t vfta, index;
4995
4996         e1000_irq_disable(adapter);
4997         vlan_group_set_device(adapter->vlgrp, vid, NULL);
4998         e1000_irq_enable(adapter);
4999
5000         if ((adapter->hw.mng_cookie.status &
5001              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
5002             (vid == adapter->mng_vlan_id)) {
5003                 /* release control to f/w */
5004                 e1000_release_hw_control(adapter);
5005                 return;
5006         }
5007
5008         /* remove VID from filter table */
5009         index = (vid >> 5) & 0x7F;
5010         vfta = E1000_READ_REG_ARRAY(&adapter->hw, VFTA, index);
5011         vfta &= ~(1 << (vid & 0x1F));
5012         e1000_write_vfta(&adapter->hw, index, vfta);
5013 }
5014
5015 static void
5016 e1000_restore_vlan(struct e1000_adapter *adapter)
5017 {
5018         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
5019
5020         if (adapter->vlgrp) {
5021                 uint16_t vid;
5022                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
5023                         if (!vlan_group_get_device(adapter->vlgrp, vid))
5024                                 continue;
5025                         e1000_vlan_rx_add_vid(adapter->netdev, vid);
5026                 }
5027         }
5028 }
5029
5030 int
5031 e1000_set_spd_dplx(struct e1000_adapter *adapter, uint16_t spddplx)
5032 {
5033         adapter->hw.autoneg = 0;
5034
5035         /* Fiber NICs only allow 1000 gbps Full duplex */
5036         if ((adapter->hw.media_type == e1000_media_type_fiber) &&
5037                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
5038                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
5039                 return -EINVAL;
5040         }
5041
5042         switch (spddplx) {
5043         case SPEED_10 + DUPLEX_HALF:
5044                 adapter->hw.forced_speed_duplex = e1000_10_half;
5045                 break;
5046         case SPEED_10 + DUPLEX_FULL:
5047                 adapter->hw.forced_speed_duplex = e1000_10_full;
5048                 break;
5049         case SPEED_100 + DUPLEX_HALF:
5050                 adapter->hw.forced_speed_duplex = e1000_100_half;
5051                 break;
5052         case SPEED_100 + DUPLEX_FULL:
5053                 adapter->hw.forced_speed_duplex = e1000_100_full;
5054                 break;
5055         case SPEED_1000 + DUPLEX_FULL:
5056                 adapter->hw.autoneg = 1;
5057                 adapter->hw.autoneg_advertised = ADVERTISE_1000_FULL;
5058                 break;
5059         case SPEED_1000 + DUPLEX_HALF: /* not supported */
5060         default:
5061                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
5062                 return -EINVAL;
5063         }
5064         return 0;
5065 }
5066
5067 static int
5068 e1000_suspend(struct pci_dev *pdev, pm_message_t state)
5069 {
5070         struct net_device *netdev = pci_get_drvdata(pdev);
5071         struct e1000_adapter *adapter = netdev_priv(netdev);
5072         uint32_t ctrl, ctrl_ext, rctl, status;
5073         uint32_t wufc = adapter->wol;
5074 #ifdef CONFIG_PM
5075         int retval = 0;
5076 #endif
5077
5078         netif_device_detach(netdev);
5079
5080         if (netif_running(netdev)) {
5081                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
5082                 e1000_down(adapter);
5083         }
5084
5085 #ifdef CONFIG_PM
5086         retval = pci_save_state(pdev);
5087         if (retval)
5088                 return retval;
5089 #endif
5090
5091         status = E1000_READ_REG(&adapter->hw, STATUS);
5092         if (status & E1000_STATUS_LU)
5093                 wufc &= ~E1000_WUFC_LNKC;
5094
5095         if (wufc) {
5096                 e1000_setup_rctl(adapter);
5097                 e1000_set_multi(netdev);
5098
5099                 /* turn on all-multi mode if wake on multicast is enabled */
5100                 if (wufc & E1000_WUFC_MC) {
5101                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
5102                         rctl |= E1000_RCTL_MPE;
5103                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
5104                 }
5105
5106                 if (adapter->hw.mac_type >= e1000_82540) {
5107                         ctrl = E1000_READ_REG(&adapter->hw, CTRL);
5108                         /* advertise wake from D3Cold */
5109                         #define E1000_CTRL_ADVD3WUC 0x00100000
5110                         /* phy power management enable */
5111                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5112                         ctrl |= E1000_CTRL_ADVD3WUC |
5113                                 E1000_CTRL_EN_PHY_PWR_MGMT;
5114                         E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
5115                 }
5116
5117                 if (adapter->hw.media_type == e1000_media_type_fiber ||
5118                    adapter->hw.media_type == e1000_media_type_internal_serdes) {
5119                         /* keep the laser running in D3 */
5120                         ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
5121                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
5122                         E1000_WRITE_REG(&adapter->hw, CTRL_EXT, ctrl_ext);
5123                 }
5124
5125                 /* Allow time for pending master requests to run */
5126                 e1000_disable_pciex_master(&adapter->hw);
5127
5128                 E1000_WRITE_REG(&adapter->hw, WUC, E1000_WUC_PME_EN);
5129                 E1000_WRITE_REG(&adapter->hw, WUFC, wufc);
5130                 pci_enable_wake(pdev, PCI_D3hot, 1);
5131                 pci_enable_wake(pdev, PCI_D3cold, 1);
5132         } else {
5133                 E1000_WRITE_REG(&adapter->hw, WUC, 0);
5134                 E1000_WRITE_REG(&adapter->hw, WUFC, 0);
5135                 pci_enable_wake(pdev, PCI_D3hot, 0);
5136                 pci_enable_wake(pdev, PCI_D3cold, 0);
5137         }
5138
5139         e1000_release_manageability(adapter);
5140
5141         /* make sure adapter isn't asleep if manageability is enabled */
5142         if (adapter->en_mng_pt) {
5143                 pci_enable_wake(pdev, PCI_D3hot, 1);
5144                 pci_enable_wake(pdev, PCI_D3cold, 1);
5145         }
5146
5147         if (adapter->hw.phy_type == e1000_phy_igp_3)
5148                 e1000_phy_powerdown_workaround(&adapter->hw);
5149
5150         if (netif_running(netdev))
5151                 e1000_free_irq(adapter);
5152
5153         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
5154          * would have already happened in close and is redundant. */
5155         e1000_release_hw_control(adapter);
5156
5157         pci_disable_device(pdev);
5158
5159         pci_set_power_state(pdev, pci_choose_state(pdev, state));
5160
5161         return 0;
5162 }
5163
5164 #ifdef CONFIG_PM
5165 static int
5166 e1000_resume(struct pci_dev *pdev)
5167 {
5168         struct net_device *netdev = pci_get_drvdata(pdev);
5169         struct e1000_adapter *adapter = netdev_priv(netdev);
5170         uint32_t err;
5171
5172         pci_set_power_state(pdev, PCI_D0);
5173         pci_restore_state(pdev);
5174         if ((err = pci_enable_device(pdev))) {
5175                 printk(KERN_ERR "e1000: Cannot enable PCI device from suspend\n");
5176                 return err;
5177         }
5178         pci_set_master(pdev);
5179
5180         pci_enable_wake(pdev, PCI_D3hot, 0);
5181         pci_enable_wake(pdev, PCI_D3cold, 0);
5182
5183         if (netif_running(netdev) && (err = e1000_request_irq(adapter)))
5184                 return err;
5185
5186         e1000_power_up_phy(adapter);
5187         e1000_reset(adapter);
5188         E1000_WRITE_REG(&adapter->hw, WUS, ~0);
5189
5190         e1000_init_manageability(adapter);
5191
5192         if (netif_running(netdev))
5193                 e1000_up(adapter);
5194
5195         netif_device_attach(netdev);
5196
5197         /* If the controller is 82573 and f/w is AMT, do not set
5198          * DRV_LOAD until the interface is up.  For all other cases,
5199          * let the f/w know that the h/w is now under the control
5200          * of the driver. */
5201         if (adapter->hw.mac_type != e1000_82573 ||
5202             !e1000_check_mng_mode(&adapter->hw))
5203                 e1000_get_hw_control(adapter);
5204
5205         return 0;
5206 }
5207 #endif
5208
5209 static void e1000_shutdown(struct pci_dev *pdev)
5210 {
5211         e1000_suspend(pdev, PMSG_SUSPEND);
5212 }
5213
5214 #ifdef CONFIG_NET_POLL_CONTROLLER
5215 /*
5216  * Polling 'interrupt' - used by things like netconsole to send skbs
5217  * without having to re-enable interrupts. It's not called while
5218  * the interrupt routine is executing.
5219  */
5220 static void
5221 e1000_netpoll(struct net_device *netdev)
5222 {
5223         struct e1000_adapter *adapter = netdev_priv(netdev);
5224
5225         disable_irq(adapter->pdev->irq);
5226         e1000_intr(adapter->pdev->irq, netdev);
5227         e1000_clean_tx_irq(adapter, adapter->tx_ring);
5228 #ifndef CONFIG_E1000_NAPI
5229         adapter->clean_rx(adapter, adapter->rx_ring);
5230 #endif
5231         enable_irq(adapter->pdev->irq);
5232 }
5233 #endif
5234
5235 /**
5236  * e1000_io_error_detected - called when PCI error is detected
5237  * @pdev: Pointer to PCI device
5238  * @state: The current pci conneection state
5239  *
5240  * This function is called after a PCI bus error affecting
5241  * this device has been detected.
5242  */
5243 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
5244 {
5245         struct net_device *netdev = pci_get_drvdata(pdev);
5246         struct e1000_adapter *adapter = netdev->priv;
5247
5248         netif_device_detach(netdev);
5249
5250         if (netif_running(netdev))
5251                 e1000_down(adapter);
5252         pci_disable_device(pdev);
5253
5254         /* Request a slot slot reset. */
5255         return PCI_ERS_RESULT_NEED_RESET;
5256 }
5257
5258 /**
5259  * e1000_io_slot_reset - called after the pci bus has been reset.
5260  * @pdev: Pointer to PCI device
5261  *
5262  * Restart the card from scratch, as if from a cold-boot. Implementation
5263  * resembles the first-half of the e1000_resume routine.
5264  */
5265 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5266 {
5267         struct net_device *netdev = pci_get_drvdata(pdev);
5268         struct e1000_adapter *adapter = netdev->priv;
5269
5270         if (pci_enable_device(pdev)) {
5271                 printk(KERN_ERR "e1000: Cannot re-enable PCI device after reset.\n");
5272                 return PCI_ERS_RESULT_DISCONNECT;
5273         }
5274         pci_set_master(pdev);
5275
5276         pci_enable_wake(pdev, PCI_D3hot, 0);
5277         pci_enable_wake(pdev, PCI_D3cold, 0);
5278
5279         e1000_reset(adapter);
5280         E1000_WRITE_REG(&adapter->hw, WUS, ~0);
5281
5282         return PCI_ERS_RESULT_RECOVERED;
5283 }
5284
5285 /**
5286  * e1000_io_resume - called when traffic can start flowing again.
5287  * @pdev: Pointer to PCI device
5288  *
5289  * This callback is called when the error recovery driver tells us that
5290  * its OK to resume normal operation. Implementation resembles the
5291  * second-half of the e1000_resume routine.
5292  */
5293 static void e1000_io_resume(struct pci_dev *pdev)
5294 {
5295         struct net_device *netdev = pci_get_drvdata(pdev);
5296         struct e1000_adapter *adapter = netdev->priv;
5297
5298         e1000_init_manageability(adapter);
5299
5300         if (netif_running(netdev)) {
5301                 if (e1000_up(adapter)) {
5302                         printk("e1000: can't bring device back up after reset\n");
5303                         return;
5304                 }
5305         }
5306
5307         netif_device_attach(netdev);
5308
5309         /* If the controller is 82573 and f/w is AMT, do not set
5310          * DRV_LOAD until the interface is up.  For all other cases,
5311          * let the f/w know that the h/w is now under the control
5312          * of the driver. */
5313         if (adapter->hw.mac_type != e1000_82573 ||
5314             !e1000_check_mng_mode(&adapter->hw))
5315                 e1000_get_hw_control(adapter);
5316
5317 }
5318
5319 /* e1000_main.c */