e1000: Fix msi enable leak on error, don't print error message, cleanup
[linux-2.6.git] / drivers / net / e1000 / e1000_main.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2006 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include "e1000.h"
30 #include <net/ip6_checksum.h>
31
32 char e1000_driver_name[] = "e1000";
33 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
34 #ifndef CONFIG_E1000_NAPI
35 #define DRIVERNAPI
36 #else
37 #define DRIVERNAPI "-NAPI"
38 #endif
39 #define DRV_VERSION "7.3.20-k2"DRIVERNAPI
40 char e1000_driver_version[] = DRV_VERSION;
41 static char e1000_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
42
43 /* e1000_pci_tbl - PCI Device ID Table
44  *
45  * Last entry must be all 0s
46  *
47  * Macro expands to...
48  *   {PCI_DEVICE(PCI_VENDOR_ID_INTEL, device_id)}
49  */
50 static struct pci_device_id e1000_pci_tbl[] = {
51         INTEL_E1000_ETHERNET_DEVICE(0x1000),
52         INTEL_E1000_ETHERNET_DEVICE(0x1001),
53         INTEL_E1000_ETHERNET_DEVICE(0x1004),
54         INTEL_E1000_ETHERNET_DEVICE(0x1008),
55         INTEL_E1000_ETHERNET_DEVICE(0x1009),
56         INTEL_E1000_ETHERNET_DEVICE(0x100C),
57         INTEL_E1000_ETHERNET_DEVICE(0x100D),
58         INTEL_E1000_ETHERNET_DEVICE(0x100E),
59         INTEL_E1000_ETHERNET_DEVICE(0x100F),
60         INTEL_E1000_ETHERNET_DEVICE(0x1010),
61         INTEL_E1000_ETHERNET_DEVICE(0x1011),
62         INTEL_E1000_ETHERNET_DEVICE(0x1012),
63         INTEL_E1000_ETHERNET_DEVICE(0x1013),
64         INTEL_E1000_ETHERNET_DEVICE(0x1014),
65         INTEL_E1000_ETHERNET_DEVICE(0x1015),
66         INTEL_E1000_ETHERNET_DEVICE(0x1016),
67         INTEL_E1000_ETHERNET_DEVICE(0x1017),
68         INTEL_E1000_ETHERNET_DEVICE(0x1018),
69         INTEL_E1000_ETHERNET_DEVICE(0x1019),
70         INTEL_E1000_ETHERNET_DEVICE(0x101A),
71         INTEL_E1000_ETHERNET_DEVICE(0x101D),
72         INTEL_E1000_ETHERNET_DEVICE(0x101E),
73         INTEL_E1000_ETHERNET_DEVICE(0x1026),
74         INTEL_E1000_ETHERNET_DEVICE(0x1027),
75         INTEL_E1000_ETHERNET_DEVICE(0x1028),
76         INTEL_E1000_ETHERNET_DEVICE(0x1049),
77         INTEL_E1000_ETHERNET_DEVICE(0x104A),
78         INTEL_E1000_ETHERNET_DEVICE(0x104B),
79         INTEL_E1000_ETHERNET_DEVICE(0x104C),
80         INTEL_E1000_ETHERNET_DEVICE(0x104D),
81         INTEL_E1000_ETHERNET_DEVICE(0x105E),
82         INTEL_E1000_ETHERNET_DEVICE(0x105F),
83         INTEL_E1000_ETHERNET_DEVICE(0x1060),
84         INTEL_E1000_ETHERNET_DEVICE(0x1075),
85         INTEL_E1000_ETHERNET_DEVICE(0x1076),
86         INTEL_E1000_ETHERNET_DEVICE(0x1077),
87         INTEL_E1000_ETHERNET_DEVICE(0x1078),
88         INTEL_E1000_ETHERNET_DEVICE(0x1079),
89         INTEL_E1000_ETHERNET_DEVICE(0x107A),
90         INTEL_E1000_ETHERNET_DEVICE(0x107B),
91         INTEL_E1000_ETHERNET_DEVICE(0x107C),
92         INTEL_E1000_ETHERNET_DEVICE(0x107D),
93         INTEL_E1000_ETHERNET_DEVICE(0x107E),
94         INTEL_E1000_ETHERNET_DEVICE(0x107F),
95         INTEL_E1000_ETHERNET_DEVICE(0x108A),
96         INTEL_E1000_ETHERNET_DEVICE(0x108B),
97         INTEL_E1000_ETHERNET_DEVICE(0x108C),
98         INTEL_E1000_ETHERNET_DEVICE(0x1096),
99         INTEL_E1000_ETHERNET_DEVICE(0x1098),
100         INTEL_E1000_ETHERNET_DEVICE(0x1099),
101         INTEL_E1000_ETHERNET_DEVICE(0x109A),
102         INTEL_E1000_ETHERNET_DEVICE(0x10A4),
103         INTEL_E1000_ETHERNET_DEVICE(0x10B5),
104         INTEL_E1000_ETHERNET_DEVICE(0x10B9),
105         INTEL_E1000_ETHERNET_DEVICE(0x10BA),
106         INTEL_E1000_ETHERNET_DEVICE(0x10BB),
107         INTEL_E1000_ETHERNET_DEVICE(0x10BC),
108         INTEL_E1000_ETHERNET_DEVICE(0x10C4),
109         INTEL_E1000_ETHERNET_DEVICE(0x10C5),
110         /* required last entry */
111         {0,}
112 };
113
114 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
115
116 int e1000_up(struct e1000_adapter *adapter);
117 void e1000_down(struct e1000_adapter *adapter);
118 void e1000_reinit_locked(struct e1000_adapter *adapter);
119 void e1000_reset(struct e1000_adapter *adapter);
120 int e1000_set_spd_dplx(struct e1000_adapter *adapter, uint16_t spddplx);
121 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
122 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
123 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
124 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
125 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
126                              struct e1000_tx_ring *txdr);
127 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
128                              struct e1000_rx_ring *rxdr);
129 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
130                              struct e1000_tx_ring *tx_ring);
131 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
132                              struct e1000_rx_ring *rx_ring);
133 void e1000_update_stats(struct e1000_adapter *adapter);
134
135 static int e1000_init_module(void);
136 static void e1000_exit_module(void);
137 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
138 static void __devexit e1000_remove(struct pci_dev *pdev);
139 static int e1000_alloc_queues(struct e1000_adapter *adapter);
140 static int e1000_sw_init(struct e1000_adapter *adapter);
141 static int e1000_open(struct net_device *netdev);
142 static int e1000_close(struct net_device *netdev);
143 static void e1000_configure_tx(struct e1000_adapter *adapter);
144 static void e1000_configure_rx(struct e1000_adapter *adapter);
145 static void e1000_setup_rctl(struct e1000_adapter *adapter);
146 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
147 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
148 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
149                                 struct e1000_tx_ring *tx_ring);
150 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
151                                 struct e1000_rx_ring *rx_ring);
152 static void e1000_set_multi(struct net_device *netdev);
153 static void e1000_update_phy_info(unsigned long data);
154 static void e1000_watchdog(unsigned long data);
155 static void e1000_82547_tx_fifo_stall(unsigned long data);
156 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
157 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
158 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
159 static int e1000_set_mac(struct net_device *netdev, void *p);
160 static irqreturn_t e1000_intr(int irq, void *data);
161 static irqreturn_t e1000_intr_msi(int irq, void *data);
162 static boolean_t e1000_clean_tx_irq(struct e1000_adapter *adapter,
163                                     struct e1000_tx_ring *tx_ring);
164 #ifdef CONFIG_E1000_NAPI
165 static int e1000_clean(struct net_device *poll_dev, int *budget);
166 static boolean_t e1000_clean_rx_irq(struct e1000_adapter *adapter,
167                                     struct e1000_rx_ring *rx_ring,
168                                     int *work_done, int work_to_do);
169 static boolean_t e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
170                                        struct e1000_rx_ring *rx_ring,
171                                        int *work_done, int work_to_do);
172 #else
173 static boolean_t e1000_clean_rx_irq(struct e1000_adapter *adapter,
174                                     struct e1000_rx_ring *rx_ring);
175 static boolean_t e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
176                                        struct e1000_rx_ring *rx_ring);
177 #endif
178 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
179                                    struct e1000_rx_ring *rx_ring,
180                                    int cleaned_count);
181 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
182                                       struct e1000_rx_ring *rx_ring,
183                                       int cleaned_count);
184 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
185 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
186                            int cmd);
187 void e1000_set_ethtool_ops(struct net_device *netdev);
188 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
189 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
190 static void e1000_tx_timeout(struct net_device *dev);
191 static void e1000_reset_task(struct work_struct *work);
192 static void e1000_smartspeed(struct e1000_adapter *adapter);
193 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
194                                        struct sk_buff *skb);
195
196 static void e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp);
197 static void e1000_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid);
198 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid);
199 static void e1000_restore_vlan(struct e1000_adapter *adapter);
200
201 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
202 #ifdef CONFIG_PM
203 static int e1000_resume(struct pci_dev *pdev);
204 #endif
205 static void e1000_shutdown(struct pci_dev *pdev);
206
207 #ifdef CONFIG_NET_POLL_CONTROLLER
208 /* for netdump / net console */
209 static void e1000_netpoll (struct net_device *netdev);
210 #endif
211
212 extern void e1000_check_options(struct e1000_adapter *adapter);
213
214 #define COPYBREAK_DEFAULT 256
215 static unsigned int copybreak __read_mostly = COPYBREAK_DEFAULT;
216 module_param(copybreak, uint, 0644);
217 MODULE_PARM_DESC(copybreak,
218         "Maximum size of packet that is copied to a new buffer on receive");
219
220 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
221                      pci_channel_state_t state);
222 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
223 static void e1000_io_resume(struct pci_dev *pdev);
224
225 static struct pci_error_handlers e1000_err_handler = {
226         .error_detected = e1000_io_error_detected,
227         .slot_reset = e1000_io_slot_reset,
228         .resume = e1000_io_resume,
229 };
230
231 static struct pci_driver e1000_driver = {
232         .name     = e1000_driver_name,
233         .id_table = e1000_pci_tbl,
234         .probe    = e1000_probe,
235         .remove   = __devexit_p(e1000_remove),
236 #ifdef CONFIG_PM
237         /* Power Managment Hooks */
238         .suspend  = e1000_suspend,
239         .resume   = e1000_resume,
240 #endif
241         .shutdown = e1000_shutdown,
242         .err_handler = &e1000_err_handler
243 };
244
245 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
246 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
247 MODULE_LICENSE("GPL");
248 MODULE_VERSION(DRV_VERSION);
249
250 static int debug = NETIF_MSG_DRV | NETIF_MSG_PROBE;
251 module_param(debug, int, 0);
252 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
253
254 /**
255  * e1000_init_module - Driver Registration Routine
256  *
257  * e1000_init_module is the first routine called when the driver is
258  * loaded. All it does is register with the PCI subsystem.
259  **/
260
261 static int __init
262 e1000_init_module(void)
263 {
264         int ret;
265         printk(KERN_INFO "%s - version %s\n",
266                e1000_driver_string, e1000_driver_version);
267
268         printk(KERN_INFO "%s\n", e1000_copyright);
269
270         ret = pci_register_driver(&e1000_driver);
271         if (copybreak != COPYBREAK_DEFAULT) {
272                 if (copybreak == 0)
273                         printk(KERN_INFO "e1000: copybreak disabled\n");
274                 else
275                         printk(KERN_INFO "e1000: copybreak enabled for "
276                                "packets <= %u bytes\n", copybreak);
277         }
278         return ret;
279 }
280
281 module_init(e1000_init_module);
282
283 /**
284  * e1000_exit_module - Driver Exit Cleanup Routine
285  *
286  * e1000_exit_module is called just before the driver is removed
287  * from memory.
288  **/
289
290 static void __exit
291 e1000_exit_module(void)
292 {
293         pci_unregister_driver(&e1000_driver);
294 }
295
296 module_exit(e1000_exit_module);
297
298 static int e1000_request_irq(struct e1000_adapter *adapter)
299 {
300         struct net_device *netdev = adapter->netdev;
301         void (*handler) = &e1000_intr;
302         int irq_flags = IRQF_SHARED;
303         int err;
304
305         if (adapter->hw.mac_type >= e1000_82571) {
306                 adapter->have_msi = !pci_enable_msi(adapter->pdev);
307                 if (adapter->have_msi) {
308                         handler = &e1000_intr_msi;
309                         irq_flags = 0;
310                 }
311         }
312
313         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
314                           netdev);
315         if (err) {
316                 if (adapter->have_msi)
317                         pci_disable_msi(adapter->pdev);
318                 DPRINTK(PROBE, ERR,
319                         "Unable to allocate interrupt Error: %d\n", err);
320         }
321
322         return err;
323 }
324
325 static void e1000_free_irq(struct e1000_adapter *adapter)
326 {
327         struct net_device *netdev = adapter->netdev;
328
329         free_irq(adapter->pdev->irq, netdev);
330
331         if (adapter->have_msi)
332                 pci_disable_msi(adapter->pdev);
333 }
334
335 /**
336  * e1000_irq_disable - Mask off interrupt generation on the NIC
337  * @adapter: board private structure
338  **/
339
340 static void
341 e1000_irq_disable(struct e1000_adapter *adapter)
342 {
343         atomic_inc(&adapter->irq_sem);
344         E1000_WRITE_REG(&adapter->hw, IMC, ~0);
345         E1000_WRITE_FLUSH(&adapter->hw);
346         synchronize_irq(adapter->pdev->irq);
347 }
348
349 /**
350  * e1000_irq_enable - Enable default interrupt generation settings
351  * @adapter: board private structure
352  **/
353
354 static void
355 e1000_irq_enable(struct e1000_adapter *adapter)
356 {
357         if (likely(atomic_dec_and_test(&adapter->irq_sem))) {
358                 E1000_WRITE_REG(&adapter->hw, IMS, IMS_ENABLE_MASK);
359                 E1000_WRITE_FLUSH(&adapter->hw);
360         }
361 }
362
363 static void
364 e1000_update_mng_vlan(struct e1000_adapter *adapter)
365 {
366         struct net_device *netdev = adapter->netdev;
367         uint16_t vid = adapter->hw.mng_cookie.vlan_id;
368         uint16_t old_vid = adapter->mng_vlan_id;
369         if (adapter->vlgrp) {
370                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
371                         if (adapter->hw.mng_cookie.status &
372                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
373                                 e1000_vlan_rx_add_vid(netdev, vid);
374                                 adapter->mng_vlan_id = vid;
375                         } else
376                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
377
378                         if ((old_vid != (uint16_t)E1000_MNG_VLAN_NONE) &&
379                                         (vid != old_vid) &&
380                             !vlan_group_get_device(adapter->vlgrp, old_vid))
381                                 e1000_vlan_rx_kill_vid(netdev, old_vid);
382                 } else
383                         adapter->mng_vlan_id = vid;
384         }
385 }
386
387 /**
388  * e1000_release_hw_control - release control of the h/w to f/w
389  * @adapter: address of board private structure
390  *
391  * e1000_release_hw_control resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
392  * For ASF and Pass Through versions of f/w this means that the
393  * driver is no longer loaded. For AMT version (only with 82573) i
394  * of the f/w this means that the network i/f is closed.
395  *
396  **/
397
398 static void
399 e1000_release_hw_control(struct e1000_adapter *adapter)
400 {
401         uint32_t ctrl_ext;
402         uint32_t swsm;
403
404         /* Let firmware taken over control of h/w */
405         switch (adapter->hw.mac_type) {
406         case e1000_82573:
407                 swsm = E1000_READ_REG(&adapter->hw, SWSM);
408                 E1000_WRITE_REG(&adapter->hw, SWSM,
409                                 swsm & ~E1000_SWSM_DRV_LOAD);
410                 break;
411         case e1000_82571:
412         case e1000_82572:
413         case e1000_80003es2lan:
414         case e1000_ich8lan:
415                 ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
416                 E1000_WRITE_REG(&adapter->hw, CTRL_EXT,
417                                 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
418                 break;
419         default:
420                 break;
421         }
422 }
423
424 /**
425  * e1000_get_hw_control - get control of the h/w from f/w
426  * @adapter: address of board private structure
427  *
428  * e1000_get_hw_control sets {CTRL_EXT|FWSM}:DRV_LOAD bit.
429  * For ASF and Pass Through versions of f/w this means that
430  * the driver is loaded. For AMT version (only with 82573)
431  * of the f/w this means that the network i/f is open.
432  *
433  **/
434
435 static void
436 e1000_get_hw_control(struct e1000_adapter *adapter)
437 {
438         uint32_t ctrl_ext;
439         uint32_t swsm;
440
441         /* Let firmware know the driver has taken over */
442         switch (adapter->hw.mac_type) {
443         case e1000_82573:
444                 swsm = E1000_READ_REG(&adapter->hw, SWSM);
445                 E1000_WRITE_REG(&adapter->hw, SWSM,
446                                 swsm | E1000_SWSM_DRV_LOAD);
447                 break;
448         case e1000_82571:
449         case e1000_82572:
450         case e1000_80003es2lan:
451         case e1000_ich8lan:
452                 ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
453                 E1000_WRITE_REG(&adapter->hw, CTRL_EXT,
454                                 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
455                 break;
456         default:
457                 break;
458         }
459 }
460
461 static void
462 e1000_init_manageability(struct e1000_adapter *adapter)
463 {
464         if (adapter->en_mng_pt) {
465                 uint32_t manc = E1000_READ_REG(&adapter->hw, MANC);
466
467                 /* disable hardware interception of ARP */
468                 manc &= ~(E1000_MANC_ARP_EN);
469
470                 /* enable receiving management packets to the host */
471                 /* this will probably generate destination unreachable messages
472                  * from the host OS, but the packets will be handled on SMBUS */
473                 if (adapter->hw.has_manc2h) {
474                         uint32_t manc2h = E1000_READ_REG(&adapter->hw, MANC2H);
475
476                         manc |= E1000_MANC_EN_MNG2HOST;
477 #define E1000_MNG2HOST_PORT_623 (1 << 5)
478 #define E1000_MNG2HOST_PORT_664 (1 << 6)
479                         manc2h |= E1000_MNG2HOST_PORT_623;
480                         manc2h |= E1000_MNG2HOST_PORT_664;
481                         E1000_WRITE_REG(&adapter->hw, MANC2H, manc2h);
482                 }
483
484                 E1000_WRITE_REG(&adapter->hw, MANC, manc);
485         }
486 }
487
488 static void
489 e1000_release_manageability(struct e1000_adapter *adapter)
490 {
491         if (adapter->en_mng_pt) {
492                 uint32_t manc = E1000_READ_REG(&adapter->hw, MANC);
493
494                 /* re-enable hardware interception of ARP */
495                 manc |= E1000_MANC_ARP_EN;
496
497                 if (adapter->hw.has_manc2h)
498                         manc &= ~E1000_MANC_EN_MNG2HOST;
499
500                 /* don't explicitly have to mess with MANC2H since
501                  * MANC has an enable disable that gates MANC2H */
502
503                 E1000_WRITE_REG(&adapter->hw, MANC, manc);
504         }
505 }
506
507 /**
508  * e1000_configure - configure the hardware for RX and TX
509  * @adapter = private board structure
510  **/
511 static void e1000_configure(struct e1000_adapter *adapter)
512 {
513         struct net_device *netdev = adapter->netdev;
514         int i;
515
516         e1000_set_multi(netdev);
517
518         e1000_restore_vlan(adapter);
519         e1000_init_manageability(adapter);
520
521         e1000_configure_tx(adapter);
522         e1000_setup_rctl(adapter);
523         e1000_configure_rx(adapter);
524         /* call E1000_DESC_UNUSED which always leaves
525          * at least 1 descriptor unused to make sure
526          * next_to_use != next_to_clean */
527         for (i = 0; i < adapter->num_rx_queues; i++) {
528                 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
529                 adapter->alloc_rx_buf(adapter, ring,
530                                       E1000_DESC_UNUSED(ring));
531         }
532
533         adapter->tx_queue_len = netdev->tx_queue_len;
534 }
535
536 int e1000_up(struct e1000_adapter *adapter)
537 {
538         /* hardware has been reset, we need to reload some things */
539         e1000_configure(adapter);
540
541         clear_bit(__E1000_DOWN, &adapter->flags);
542
543 #ifdef CONFIG_E1000_NAPI
544         netif_poll_enable(adapter->netdev);
545 #endif
546         e1000_irq_enable(adapter);
547
548         /* fire a link change interrupt to start the watchdog */
549         E1000_WRITE_REG(&adapter->hw, ICS, E1000_ICS_LSC);
550         return 0;
551 }
552
553 /**
554  * e1000_power_up_phy - restore link in case the phy was powered down
555  * @adapter: address of board private structure
556  *
557  * The phy may be powered down to save power and turn off link when the
558  * driver is unloaded and wake on lan is not enabled (among others)
559  * *** this routine MUST be followed by a call to e1000_reset ***
560  *
561  **/
562
563 void e1000_power_up_phy(struct e1000_adapter *adapter)
564 {
565         uint16_t mii_reg = 0;
566
567         /* Just clear the power down bit to wake the phy back up */
568         if (adapter->hw.media_type == e1000_media_type_copper) {
569                 /* according to the manual, the phy will retain its
570                  * settings across a power-down/up cycle */
571                 e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &mii_reg);
572                 mii_reg &= ~MII_CR_POWER_DOWN;
573                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL, mii_reg);
574         }
575 }
576
577 static void e1000_power_down_phy(struct e1000_adapter *adapter)
578 {
579         /* Power down the PHY so no link is implied when interface is down *
580          * The PHY cannot be powered down if any of the following is TRUE *
581          * (a) WoL is enabled
582          * (b) AMT is active
583          * (c) SoL/IDER session is active */
584         if (!adapter->wol && adapter->hw.mac_type >= e1000_82540 &&
585            adapter->hw.media_type == e1000_media_type_copper) {
586                 uint16_t mii_reg = 0;
587
588                 switch (adapter->hw.mac_type) {
589                 case e1000_82540:
590                 case e1000_82545:
591                 case e1000_82545_rev_3:
592                 case e1000_82546:
593                 case e1000_82546_rev_3:
594                 case e1000_82541:
595                 case e1000_82541_rev_2:
596                 case e1000_82547:
597                 case e1000_82547_rev_2:
598                         if (E1000_READ_REG(&adapter->hw, MANC) &
599                             E1000_MANC_SMBUS_EN)
600                                 goto out;
601                         break;
602                 case e1000_82571:
603                 case e1000_82572:
604                 case e1000_82573:
605                 case e1000_80003es2lan:
606                 case e1000_ich8lan:
607                         if (e1000_check_mng_mode(&adapter->hw) ||
608                             e1000_check_phy_reset_block(&adapter->hw))
609                                 goto out;
610                         break;
611                 default:
612                         goto out;
613                 }
614                 e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &mii_reg);
615                 mii_reg |= MII_CR_POWER_DOWN;
616                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL, mii_reg);
617                 mdelay(1);
618         }
619 out:
620         return;
621 }
622
623 void
624 e1000_down(struct e1000_adapter *adapter)
625 {
626         struct net_device *netdev = adapter->netdev;
627
628         /* signal that we're down so the interrupt handler does not
629          * reschedule our watchdog timer */
630         set_bit(__E1000_DOWN, &adapter->flags);
631
632 #ifdef CONFIG_E1000_NAPI
633         netif_poll_disable(netdev);
634 #endif
635         e1000_irq_disable(adapter);
636
637         del_timer_sync(&adapter->tx_fifo_stall_timer);
638         del_timer_sync(&adapter->watchdog_timer);
639         del_timer_sync(&adapter->phy_info_timer);
640
641         netdev->tx_queue_len = adapter->tx_queue_len;
642         adapter->link_speed = 0;
643         adapter->link_duplex = 0;
644         netif_carrier_off(netdev);
645         netif_stop_queue(netdev);
646
647         e1000_reset(adapter);
648         e1000_clean_all_tx_rings(adapter);
649         e1000_clean_all_rx_rings(adapter);
650 }
651
652 void
653 e1000_reinit_locked(struct e1000_adapter *adapter)
654 {
655         WARN_ON(in_interrupt());
656         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
657                 msleep(1);
658         e1000_down(adapter);
659         e1000_up(adapter);
660         clear_bit(__E1000_RESETTING, &adapter->flags);
661 }
662
663 void
664 e1000_reset(struct e1000_adapter *adapter)
665 {
666         uint32_t pba = 0, tx_space, min_tx_space, min_rx_space;
667         uint16_t fc_high_water_mark = E1000_FC_HIGH_DIFF;
668         boolean_t legacy_pba_adjust = FALSE;
669
670         /* Repartition Pba for greater than 9k mtu
671          * To take effect CTRL.RST is required.
672          */
673
674         switch (adapter->hw.mac_type) {
675         case e1000_82542_rev2_0:
676         case e1000_82542_rev2_1:
677         case e1000_82543:
678         case e1000_82544:
679         case e1000_82540:
680         case e1000_82541:
681         case e1000_82541_rev_2:
682                 legacy_pba_adjust = TRUE;
683                 pba = E1000_PBA_48K;
684                 break;
685         case e1000_82545:
686         case e1000_82545_rev_3:
687         case e1000_82546:
688         case e1000_82546_rev_3:
689                 pba = E1000_PBA_48K;
690                 break;
691         case e1000_82547:
692         case e1000_82547_rev_2:
693                 legacy_pba_adjust = TRUE;
694                 pba = E1000_PBA_30K;
695                 break;
696         case e1000_82571:
697         case e1000_82572:
698         case e1000_80003es2lan:
699                 pba = E1000_PBA_38K;
700                 break;
701         case e1000_82573:
702                 pba = E1000_PBA_20K;
703                 break;
704         case e1000_ich8lan:
705                 pba = E1000_PBA_8K;
706         case e1000_undefined:
707         case e1000_num_macs:
708                 break;
709         }
710
711         if (legacy_pba_adjust == TRUE) {
712                 if (adapter->netdev->mtu > E1000_RXBUFFER_8192)
713                         pba -= 8; /* allocate more FIFO for Tx */
714
715                 if (adapter->hw.mac_type == e1000_82547) {
716                         adapter->tx_fifo_head = 0;
717                         adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
718                         adapter->tx_fifo_size =
719                                 (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
720                         atomic_set(&adapter->tx_fifo_stall, 0);
721                 }
722         } else if (adapter->hw.max_frame_size > MAXIMUM_ETHERNET_FRAME_SIZE) {
723                 /* adjust PBA for jumbo frames */
724                 E1000_WRITE_REG(&adapter->hw, PBA, pba);
725
726                 /* To maintain wire speed transmits, the Tx FIFO should be
727                  * large enough to accomodate two full transmit packets,
728                  * rounded up to the next 1KB and expressed in KB.  Likewise,
729                  * the Rx FIFO should be large enough to accomodate at least
730                  * one full receive packet and is similarly rounded up and
731                  * expressed in KB. */
732                 pba = E1000_READ_REG(&adapter->hw, PBA);
733                 /* upper 16 bits has Tx packet buffer allocation size in KB */
734                 tx_space = pba >> 16;
735                 /* lower 16 bits has Rx packet buffer allocation size in KB */
736                 pba &= 0xffff;
737                 /* don't include ethernet FCS because hardware appends/strips */
738                 min_rx_space = adapter->netdev->mtu + ENET_HEADER_SIZE +
739                                VLAN_TAG_SIZE;
740                 min_tx_space = min_rx_space;
741                 min_tx_space *= 2;
742                 min_tx_space = ALIGN(min_tx_space, 1024);
743                 min_tx_space >>= 10;
744                 min_rx_space = ALIGN(min_rx_space, 1024);
745                 min_rx_space >>= 10;
746
747                 /* If current Tx allocation is less than the min Tx FIFO size,
748                  * and the min Tx FIFO size is less than the current Rx FIFO
749                  * allocation, take space away from current Rx allocation */
750                 if (tx_space < min_tx_space &&
751                     ((min_tx_space - tx_space) < pba)) {
752                         pba = pba - (min_tx_space - tx_space);
753
754                         /* PCI/PCIx hardware has PBA alignment constraints */
755                         switch (adapter->hw.mac_type) {
756                         case e1000_82545 ... e1000_82546_rev_3:
757                                 pba &= ~(E1000_PBA_8K - 1);
758                                 break;
759                         default:
760                                 break;
761                         }
762
763                         /* if short on rx space, rx wins and must trump tx
764                          * adjustment or use Early Receive if available */
765                         if (pba < min_rx_space) {
766                                 switch (adapter->hw.mac_type) {
767                                 case e1000_82573:
768                                         /* ERT enabled in e1000_configure_rx */
769                                         break;
770                                 default:
771                                         pba = min_rx_space;
772                                         break;
773                                 }
774                         }
775                 }
776         }
777
778         E1000_WRITE_REG(&adapter->hw, PBA, pba);
779
780         /* flow control settings */
781         /* Set the FC high water mark to 90% of the FIFO size.
782          * Required to clear last 3 LSB */
783         fc_high_water_mark = ((pba * 9216)/10) & 0xFFF8;
784         /* We can't use 90% on small FIFOs because the remainder
785          * would be less than 1 full frame.  In this case, we size
786          * it to allow at least a full frame above the high water
787          *  mark. */
788         if (pba < E1000_PBA_16K)
789                 fc_high_water_mark = (pba * 1024) - 1600;
790
791         adapter->hw.fc_high_water = fc_high_water_mark;
792         adapter->hw.fc_low_water = fc_high_water_mark - 8;
793         if (adapter->hw.mac_type == e1000_80003es2lan)
794                 adapter->hw.fc_pause_time = 0xFFFF;
795         else
796                 adapter->hw.fc_pause_time = E1000_FC_PAUSE_TIME;
797         adapter->hw.fc_send_xon = 1;
798         adapter->hw.fc = adapter->hw.original_fc;
799
800         /* Allow time for pending master requests to run */
801         e1000_reset_hw(&adapter->hw);
802         if (adapter->hw.mac_type >= e1000_82544)
803                 E1000_WRITE_REG(&adapter->hw, WUC, 0);
804
805         if (e1000_init_hw(&adapter->hw))
806                 DPRINTK(PROBE, ERR, "Hardware Error\n");
807         e1000_update_mng_vlan(adapter);
808
809         /* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
810         if (adapter->hw.mac_type >= e1000_82544 &&
811             adapter->hw.mac_type <= e1000_82547_rev_2 &&
812             adapter->hw.autoneg == 1 &&
813             adapter->hw.autoneg_advertised == ADVERTISE_1000_FULL) {
814                 uint32_t ctrl = E1000_READ_REG(&adapter->hw, CTRL);
815                 /* clear phy power management bit if we are in gig only mode,
816                  * which if enabled will attempt negotiation to 100Mb, which
817                  * can cause a loss of link at power off or driver unload */
818                 ctrl &= ~E1000_CTRL_SWDPIN3;
819                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
820         }
821
822         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
823         E1000_WRITE_REG(&adapter->hw, VET, ETHERNET_IEEE_VLAN_TYPE);
824
825         e1000_reset_adaptive(&adapter->hw);
826         e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
827
828         if (!adapter->smart_power_down &&
829             (adapter->hw.mac_type == e1000_82571 ||
830              adapter->hw.mac_type == e1000_82572)) {
831                 uint16_t phy_data = 0;
832                 /* speed up time to link by disabling smart power down, ignore
833                  * the return value of this function because there is nothing
834                  * different we would do if it failed */
835                 e1000_read_phy_reg(&adapter->hw, IGP02E1000_PHY_POWER_MGMT,
836                                    &phy_data);
837                 phy_data &= ~IGP02E1000_PM_SPD;
838                 e1000_write_phy_reg(&adapter->hw, IGP02E1000_PHY_POWER_MGMT,
839                                     phy_data);
840         }
841
842         e1000_release_manageability(adapter);
843 }
844
845 /**
846  * e1000_probe - Device Initialization Routine
847  * @pdev: PCI device information struct
848  * @ent: entry in e1000_pci_tbl
849  *
850  * Returns 0 on success, negative on failure
851  *
852  * e1000_probe initializes an adapter identified by a pci_dev structure.
853  * The OS initialization, configuring of the adapter private structure,
854  * and a hardware reset occur.
855  **/
856
857 static int __devinit
858 e1000_probe(struct pci_dev *pdev,
859             const struct pci_device_id *ent)
860 {
861         struct net_device *netdev;
862         struct e1000_adapter *adapter;
863         unsigned long mmio_start, mmio_len;
864         unsigned long flash_start, flash_len;
865
866         static int cards_found = 0;
867         static int global_quad_port_a = 0; /* global ksp3 port a indication */
868         int i, err, pci_using_dac;
869         uint16_t eeprom_data = 0;
870         uint16_t eeprom_apme_mask = E1000_EEPROM_APME;
871         if ((err = pci_enable_device(pdev)))
872                 return err;
873
874         if (!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK)) &&
875             !(err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK))) {
876                 pci_using_dac = 1;
877         } else {
878                 if ((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK)) &&
879                     (err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK))) {
880                         E1000_ERR("No usable DMA configuration, aborting\n");
881                         goto err_dma;
882                 }
883                 pci_using_dac = 0;
884         }
885
886         if ((err = pci_request_regions(pdev, e1000_driver_name)))
887                 goto err_pci_reg;
888
889         pci_set_master(pdev);
890
891         err = -ENOMEM;
892         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
893         if (!netdev)
894                 goto err_alloc_etherdev;
895
896         SET_MODULE_OWNER(netdev);
897         SET_NETDEV_DEV(netdev, &pdev->dev);
898
899         pci_set_drvdata(pdev, netdev);
900         adapter = netdev_priv(netdev);
901         adapter->netdev = netdev;
902         adapter->pdev = pdev;
903         adapter->hw.back = adapter;
904         adapter->msg_enable = (1 << debug) - 1;
905
906         mmio_start = pci_resource_start(pdev, BAR_0);
907         mmio_len = pci_resource_len(pdev, BAR_0);
908
909         err = -EIO;
910         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
911         if (!adapter->hw.hw_addr)
912                 goto err_ioremap;
913
914         for (i = BAR_1; i <= BAR_5; i++) {
915                 if (pci_resource_len(pdev, i) == 0)
916                         continue;
917                 if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
918                         adapter->hw.io_base = pci_resource_start(pdev, i);
919                         break;
920                 }
921         }
922
923         netdev->open = &e1000_open;
924         netdev->stop = &e1000_close;
925         netdev->hard_start_xmit = &e1000_xmit_frame;
926         netdev->get_stats = &e1000_get_stats;
927         netdev->set_multicast_list = &e1000_set_multi;
928         netdev->set_mac_address = &e1000_set_mac;
929         netdev->change_mtu = &e1000_change_mtu;
930         netdev->do_ioctl = &e1000_ioctl;
931         e1000_set_ethtool_ops(netdev);
932         netdev->tx_timeout = &e1000_tx_timeout;
933         netdev->watchdog_timeo = 5 * HZ;
934 #ifdef CONFIG_E1000_NAPI
935         netdev->poll = &e1000_clean;
936         netdev->weight = 64;
937 #endif
938         netdev->vlan_rx_register = e1000_vlan_rx_register;
939         netdev->vlan_rx_add_vid = e1000_vlan_rx_add_vid;
940         netdev->vlan_rx_kill_vid = e1000_vlan_rx_kill_vid;
941 #ifdef CONFIG_NET_POLL_CONTROLLER
942         netdev->poll_controller = e1000_netpoll;
943 #endif
944         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
945
946         netdev->mem_start = mmio_start;
947         netdev->mem_end = mmio_start + mmio_len;
948         netdev->base_addr = adapter->hw.io_base;
949
950         adapter->bd_number = cards_found;
951
952         /* setup the private structure */
953
954         if ((err = e1000_sw_init(adapter)))
955                 goto err_sw_init;
956
957         err = -EIO;
958         /* Flash BAR mapping must happen after e1000_sw_init
959          * because it depends on mac_type */
960         if ((adapter->hw.mac_type == e1000_ich8lan) &&
961            (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
962                 flash_start = pci_resource_start(pdev, 1);
963                 flash_len = pci_resource_len(pdev, 1);
964                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
965                 if (!adapter->hw.flash_address)
966                         goto err_flashmap;
967         }
968
969         if (e1000_check_phy_reset_block(&adapter->hw))
970                 DPRINTK(PROBE, INFO, "PHY reset is blocked due to SOL/IDER session.\n");
971
972         if (adapter->hw.mac_type >= e1000_82543) {
973                 netdev->features = NETIF_F_SG |
974                                    NETIF_F_HW_CSUM |
975                                    NETIF_F_HW_VLAN_TX |
976                                    NETIF_F_HW_VLAN_RX |
977                                    NETIF_F_HW_VLAN_FILTER;
978                 if (adapter->hw.mac_type == e1000_ich8lan)
979                         netdev->features &= ~NETIF_F_HW_VLAN_FILTER;
980         }
981
982         if ((adapter->hw.mac_type >= e1000_82544) &&
983            (adapter->hw.mac_type != e1000_82547))
984                 netdev->features |= NETIF_F_TSO;
985
986         if (adapter->hw.mac_type > e1000_82547_rev_2)
987                 netdev->features |= NETIF_F_TSO6;
988         if (pci_using_dac)
989                 netdev->features |= NETIF_F_HIGHDMA;
990
991         netdev->features |= NETIF_F_LLTX;
992
993         adapter->en_mng_pt = e1000_enable_mng_pass_thru(&adapter->hw);
994
995         /* initialize eeprom parameters */
996
997         if (e1000_init_eeprom_params(&adapter->hw)) {
998                 E1000_ERR("EEPROM initialization failed\n");
999                 goto err_eeprom;
1000         }
1001
1002         /* before reading the EEPROM, reset the controller to
1003          * put the device in a known good starting state */
1004
1005         e1000_reset_hw(&adapter->hw);
1006
1007         /* make sure the EEPROM is good */
1008
1009         if (e1000_validate_eeprom_checksum(&adapter->hw) < 0) {
1010                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
1011                 goto err_eeprom;
1012         }
1013
1014         /* copy the MAC address out of the EEPROM */
1015
1016         if (e1000_read_mac_addr(&adapter->hw))
1017                 DPRINTK(PROBE, ERR, "EEPROM Read Error\n");
1018         memcpy(netdev->dev_addr, adapter->hw.mac_addr, netdev->addr_len);
1019         memcpy(netdev->perm_addr, adapter->hw.mac_addr, netdev->addr_len);
1020
1021         if (!is_valid_ether_addr(netdev->perm_addr)) {
1022                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
1023                 goto err_eeprom;
1024         }
1025
1026         e1000_get_bus_info(&adapter->hw);
1027
1028         init_timer(&adapter->tx_fifo_stall_timer);
1029         adapter->tx_fifo_stall_timer.function = &e1000_82547_tx_fifo_stall;
1030         adapter->tx_fifo_stall_timer.data = (unsigned long) adapter;
1031
1032         init_timer(&adapter->watchdog_timer);
1033         adapter->watchdog_timer.function = &e1000_watchdog;
1034         adapter->watchdog_timer.data = (unsigned long) adapter;
1035
1036         init_timer(&adapter->phy_info_timer);
1037         adapter->phy_info_timer.function = &e1000_update_phy_info;
1038         adapter->phy_info_timer.data = (unsigned long) adapter;
1039
1040         INIT_WORK(&adapter->reset_task, e1000_reset_task);
1041
1042         e1000_check_options(adapter);
1043
1044         /* Initial Wake on LAN setting
1045          * If APM wake is enabled in the EEPROM,
1046          * enable the ACPI Magic Packet filter
1047          */
1048
1049         switch (adapter->hw.mac_type) {
1050         case e1000_82542_rev2_0:
1051         case e1000_82542_rev2_1:
1052         case e1000_82543:
1053                 break;
1054         case e1000_82544:
1055                 e1000_read_eeprom(&adapter->hw,
1056                         EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
1057                 eeprom_apme_mask = E1000_EEPROM_82544_APM;
1058                 break;
1059         case e1000_ich8lan:
1060                 e1000_read_eeprom(&adapter->hw,
1061                         EEPROM_INIT_CONTROL1_REG, 1, &eeprom_data);
1062                 eeprom_apme_mask = E1000_EEPROM_ICH8_APME;
1063                 break;
1064         case e1000_82546:
1065         case e1000_82546_rev_3:
1066         case e1000_82571:
1067         case e1000_80003es2lan:
1068                 if (E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_FUNC_1){
1069                         e1000_read_eeprom(&adapter->hw,
1070                                 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1071                         break;
1072                 }
1073                 /* Fall Through */
1074         default:
1075                 e1000_read_eeprom(&adapter->hw,
1076                         EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1077                 break;
1078         }
1079         if (eeprom_data & eeprom_apme_mask)
1080                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1081
1082         /* now that we have the eeprom settings, apply the special cases
1083          * where the eeprom may be wrong or the board simply won't support
1084          * wake on lan on a particular port */
1085         switch (pdev->device) {
1086         case E1000_DEV_ID_82546GB_PCIE:
1087                 adapter->eeprom_wol = 0;
1088                 break;
1089         case E1000_DEV_ID_82546EB_FIBER:
1090         case E1000_DEV_ID_82546GB_FIBER:
1091         case E1000_DEV_ID_82571EB_FIBER:
1092                 /* Wake events only supported on port A for dual fiber
1093                  * regardless of eeprom setting */
1094                 if (E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_FUNC_1)
1095                         adapter->eeprom_wol = 0;
1096                 break;
1097         case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
1098         case E1000_DEV_ID_82571EB_QUAD_COPPER:
1099         case E1000_DEV_ID_82571EB_QUAD_COPPER_LOWPROFILE:
1100                 /* if quad port adapter, disable WoL on all but port A */
1101                 if (global_quad_port_a != 0)
1102                         adapter->eeprom_wol = 0;
1103                 else
1104                         adapter->quad_port_a = 1;
1105                 /* Reset for multiple quad port adapters */
1106                 if (++global_quad_port_a == 4)
1107                         global_quad_port_a = 0;
1108                 break;
1109         }
1110
1111         /* initialize the wol settings based on the eeprom settings */
1112         adapter->wol = adapter->eeprom_wol;
1113
1114         /* print bus type/speed/width info */
1115         {
1116         struct e1000_hw *hw = &adapter->hw;
1117         DPRINTK(PROBE, INFO, "(PCI%s:%s:%s) ",
1118                 ((hw->bus_type == e1000_bus_type_pcix) ? "-X" :
1119                  (hw->bus_type == e1000_bus_type_pci_express ? " Express":"")),
1120                 ((hw->bus_speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
1121                  (hw->bus_speed == e1000_bus_speed_133) ? "133MHz" :
1122                  (hw->bus_speed == e1000_bus_speed_120) ? "120MHz" :
1123                  (hw->bus_speed == e1000_bus_speed_100) ? "100MHz" :
1124                  (hw->bus_speed == e1000_bus_speed_66) ? "66MHz" : "33MHz"),
1125                 ((hw->bus_width == e1000_bus_width_64) ? "64-bit" :
1126                  (hw->bus_width == e1000_bus_width_pciex_4) ? "Width x4" :
1127                  (hw->bus_width == e1000_bus_width_pciex_1) ? "Width x1" :
1128                  "32-bit"));
1129         }
1130
1131         for (i = 0; i < 6; i++)
1132                 printk("%2.2x%c", netdev->dev_addr[i], i == 5 ? '\n' : ':');
1133
1134         /* reset the hardware with the new settings */
1135         e1000_reset(adapter);
1136
1137         /* If the controller is 82573 and f/w is AMT, do not set
1138          * DRV_LOAD until the interface is up.  For all other cases,
1139          * let the f/w know that the h/w is now under the control
1140          * of the driver. */
1141         if (adapter->hw.mac_type != e1000_82573 ||
1142             !e1000_check_mng_mode(&adapter->hw))
1143                 e1000_get_hw_control(adapter);
1144
1145         strcpy(netdev->name, "eth%d");
1146         if ((err = register_netdev(netdev)))
1147                 goto err_register;
1148
1149         /* tell the stack to leave us alone until e1000_open() is called */
1150         netif_carrier_off(netdev);
1151         netif_stop_queue(netdev);
1152
1153         DPRINTK(PROBE, INFO, "Intel(R) PRO/1000 Network Connection\n");
1154
1155         cards_found++;
1156         return 0;
1157
1158 err_register:
1159         e1000_release_hw_control(adapter);
1160 err_eeprom:
1161         if (!e1000_check_phy_reset_block(&adapter->hw))
1162                 e1000_phy_hw_reset(&adapter->hw);
1163
1164         if (adapter->hw.flash_address)
1165                 iounmap(adapter->hw.flash_address);
1166 err_flashmap:
1167 #ifdef CONFIG_E1000_NAPI
1168         for (i = 0; i < adapter->num_rx_queues; i++)
1169                 dev_put(&adapter->polling_netdev[i]);
1170 #endif
1171
1172         kfree(adapter->tx_ring);
1173         kfree(adapter->rx_ring);
1174 #ifdef CONFIG_E1000_NAPI
1175         kfree(adapter->polling_netdev);
1176 #endif
1177 err_sw_init:
1178         iounmap(adapter->hw.hw_addr);
1179 err_ioremap:
1180         free_netdev(netdev);
1181 err_alloc_etherdev:
1182         pci_release_regions(pdev);
1183 err_pci_reg:
1184 err_dma:
1185         pci_disable_device(pdev);
1186         return err;
1187 }
1188
1189 /**
1190  * e1000_remove - Device Removal Routine
1191  * @pdev: PCI device information struct
1192  *
1193  * e1000_remove is called by the PCI subsystem to alert the driver
1194  * that it should release a PCI device.  The could be caused by a
1195  * Hot-Plug event, or because the driver is going to be removed from
1196  * memory.
1197  **/
1198
1199 static void __devexit
1200 e1000_remove(struct pci_dev *pdev)
1201 {
1202         struct net_device *netdev = pci_get_drvdata(pdev);
1203         struct e1000_adapter *adapter = netdev_priv(netdev);
1204 #ifdef CONFIG_E1000_NAPI
1205         int i;
1206 #endif
1207
1208         cancel_work_sync(&adapter->reset_task);
1209
1210         e1000_release_manageability(adapter);
1211
1212         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
1213          * would have already happened in close and is redundant. */
1214         e1000_release_hw_control(adapter);
1215
1216         unregister_netdev(netdev);
1217 #ifdef CONFIG_E1000_NAPI
1218         for (i = 0; i < adapter->num_rx_queues; i++)
1219                 dev_put(&adapter->polling_netdev[i]);
1220 #endif
1221
1222         if (!e1000_check_phy_reset_block(&adapter->hw))
1223                 e1000_phy_hw_reset(&adapter->hw);
1224
1225         kfree(adapter->tx_ring);
1226         kfree(adapter->rx_ring);
1227 #ifdef CONFIG_E1000_NAPI
1228         kfree(adapter->polling_netdev);
1229 #endif
1230
1231         iounmap(adapter->hw.hw_addr);
1232         if (adapter->hw.flash_address)
1233                 iounmap(adapter->hw.flash_address);
1234         pci_release_regions(pdev);
1235
1236         free_netdev(netdev);
1237
1238         pci_disable_device(pdev);
1239 }
1240
1241 /**
1242  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1243  * @adapter: board private structure to initialize
1244  *
1245  * e1000_sw_init initializes the Adapter private data structure.
1246  * Fields are initialized based on PCI device information and
1247  * OS network device settings (MTU size).
1248  **/
1249
1250 static int __devinit
1251 e1000_sw_init(struct e1000_adapter *adapter)
1252 {
1253         struct e1000_hw *hw = &adapter->hw;
1254         struct net_device *netdev = adapter->netdev;
1255         struct pci_dev *pdev = adapter->pdev;
1256 #ifdef CONFIG_E1000_NAPI
1257         int i;
1258 #endif
1259
1260         /* PCI config space info */
1261
1262         hw->vendor_id = pdev->vendor;
1263         hw->device_id = pdev->device;
1264         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1265         hw->subsystem_id = pdev->subsystem_device;
1266
1267         pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
1268
1269         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
1270
1271         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1272         adapter->rx_ps_bsize0 = E1000_RXBUFFER_128;
1273         hw->max_frame_size = netdev->mtu +
1274                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
1275         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
1276
1277         /* identify the MAC */
1278
1279         if (e1000_set_mac_type(hw)) {
1280                 DPRINTK(PROBE, ERR, "Unknown MAC Type\n");
1281                 return -EIO;
1282         }
1283
1284         switch (hw->mac_type) {
1285         default:
1286                 break;
1287         case e1000_82541:
1288         case e1000_82547:
1289         case e1000_82541_rev_2:
1290         case e1000_82547_rev_2:
1291                 hw->phy_init_script = 1;
1292                 break;
1293         }
1294
1295         e1000_set_media_type(hw);
1296
1297         hw->wait_autoneg_complete = FALSE;
1298         hw->tbi_compatibility_en = TRUE;
1299         hw->adaptive_ifs = TRUE;
1300
1301         /* Copper options */
1302
1303         if (hw->media_type == e1000_media_type_copper) {
1304                 hw->mdix = AUTO_ALL_MODES;
1305                 hw->disable_polarity_correction = FALSE;
1306                 hw->master_slave = E1000_MASTER_SLAVE;
1307         }
1308
1309         adapter->num_tx_queues = 1;
1310         adapter->num_rx_queues = 1;
1311
1312         if (e1000_alloc_queues(adapter)) {
1313                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
1314                 return -ENOMEM;
1315         }
1316
1317 #ifdef CONFIG_E1000_NAPI
1318         for (i = 0; i < adapter->num_rx_queues; i++) {
1319                 adapter->polling_netdev[i].priv = adapter;
1320                 adapter->polling_netdev[i].poll = &e1000_clean;
1321                 adapter->polling_netdev[i].weight = 64;
1322                 dev_hold(&adapter->polling_netdev[i]);
1323                 set_bit(__LINK_STATE_START, &adapter->polling_netdev[i].state);
1324         }
1325         spin_lock_init(&adapter->tx_queue_lock);
1326 #endif
1327
1328         atomic_set(&adapter->irq_sem, 1);
1329         spin_lock_init(&adapter->stats_lock);
1330
1331         set_bit(__E1000_DOWN, &adapter->flags);
1332
1333         return 0;
1334 }
1335
1336 /**
1337  * e1000_alloc_queues - Allocate memory for all rings
1338  * @adapter: board private structure to initialize
1339  *
1340  * We allocate one ring per queue at run-time since we don't know the
1341  * number of queues at compile-time.  The polling_netdev array is
1342  * intended for Multiqueue, but should work fine with a single queue.
1343  **/
1344
1345 static int __devinit
1346 e1000_alloc_queues(struct e1000_adapter *adapter)
1347 {
1348         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1349                                    sizeof(struct e1000_tx_ring), GFP_KERNEL);
1350         if (!adapter->tx_ring)
1351                 return -ENOMEM;
1352
1353         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1354                                    sizeof(struct e1000_rx_ring), GFP_KERNEL);
1355         if (!adapter->rx_ring) {
1356                 kfree(adapter->tx_ring);
1357                 return -ENOMEM;
1358         }
1359
1360 #ifdef CONFIG_E1000_NAPI
1361         adapter->polling_netdev = kcalloc(adapter->num_rx_queues,
1362                                           sizeof(struct net_device),
1363                                           GFP_KERNEL);
1364         if (!adapter->polling_netdev) {
1365                 kfree(adapter->tx_ring);
1366                 kfree(adapter->rx_ring);
1367                 return -ENOMEM;
1368         }
1369 #endif
1370
1371         return E1000_SUCCESS;
1372 }
1373
1374 /**
1375  * e1000_open - Called when a network interface is made active
1376  * @netdev: network interface device structure
1377  *
1378  * Returns 0 on success, negative value on failure
1379  *
1380  * The open entry point is called when a network interface is made
1381  * active by the system (IFF_UP).  At this point all resources needed
1382  * for transmit and receive operations are allocated, the interrupt
1383  * handler is registered with the OS, the watchdog timer is started,
1384  * and the stack is notified that the interface is ready.
1385  **/
1386
1387 static int
1388 e1000_open(struct net_device *netdev)
1389 {
1390         struct e1000_adapter *adapter = netdev_priv(netdev);
1391         int err;
1392
1393         /* disallow open during test */
1394         if (test_bit(__E1000_TESTING, &adapter->flags))
1395                 return -EBUSY;
1396
1397         /* allocate transmit descriptors */
1398         err = e1000_setup_all_tx_resources(adapter);
1399         if (err)
1400                 goto err_setup_tx;
1401
1402         /* allocate receive descriptors */
1403         err = e1000_setup_all_rx_resources(adapter);
1404         if (err)
1405                 goto err_setup_rx;
1406
1407         e1000_power_up_phy(adapter);
1408
1409         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1410         if ((adapter->hw.mng_cookie.status &
1411                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1412                 e1000_update_mng_vlan(adapter);
1413         }
1414
1415         /* If AMT is enabled, let the firmware know that the network
1416          * interface is now open */
1417         if (adapter->hw.mac_type == e1000_82573 &&
1418             e1000_check_mng_mode(&adapter->hw))
1419                 e1000_get_hw_control(adapter);
1420
1421         /* before we allocate an interrupt, we must be ready to handle it.
1422          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1423          * as soon as we call pci_request_irq, so we have to setup our
1424          * clean_rx handler before we do so.  */
1425         e1000_configure(adapter);
1426
1427         err = e1000_request_irq(adapter);
1428         if (err)
1429                 goto err_req_irq;
1430
1431         /* From here on the code is the same as e1000_up() */
1432         clear_bit(__E1000_DOWN, &adapter->flags);
1433
1434 #ifdef CONFIG_E1000_NAPI
1435         netif_poll_enable(netdev);
1436 #endif
1437
1438         e1000_irq_enable(adapter);
1439
1440         /* fire a link status change interrupt to start the watchdog */
1441         E1000_WRITE_REG(&adapter->hw, ICS, E1000_ICS_LSC);
1442
1443         return E1000_SUCCESS;
1444
1445 err_req_irq:
1446         e1000_release_hw_control(adapter);
1447         e1000_power_down_phy(adapter);
1448         e1000_free_all_rx_resources(adapter);
1449 err_setup_rx:
1450         e1000_free_all_tx_resources(adapter);
1451 err_setup_tx:
1452         e1000_reset(adapter);
1453
1454         return err;
1455 }
1456
1457 /**
1458  * e1000_close - Disables a network interface
1459  * @netdev: network interface device structure
1460  *
1461  * Returns 0, this is not allowed to fail
1462  *
1463  * The close entry point is called when an interface is de-activated
1464  * by the OS.  The hardware is still under the drivers control, but
1465  * needs to be disabled.  A global MAC reset is issued to stop the
1466  * hardware, and all transmit and receive resources are freed.
1467  **/
1468
1469 static int
1470 e1000_close(struct net_device *netdev)
1471 {
1472         struct e1000_adapter *adapter = netdev_priv(netdev);
1473
1474         WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1475         e1000_down(adapter);
1476         e1000_power_down_phy(adapter);
1477         e1000_free_irq(adapter);
1478
1479         e1000_free_all_tx_resources(adapter);
1480         e1000_free_all_rx_resources(adapter);
1481
1482         /* kill manageability vlan ID if supported, but not if a vlan with
1483          * the same ID is registered on the host OS (let 8021q kill it) */
1484         if ((adapter->hw.mng_cookie.status &
1485                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1486              !(adapter->vlgrp &&
1487                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id))) {
1488                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1489         }
1490
1491         /* If AMT is enabled, let the firmware know that the network
1492          * interface is now closed */
1493         if (adapter->hw.mac_type == e1000_82573 &&
1494             e1000_check_mng_mode(&adapter->hw))
1495                 e1000_release_hw_control(adapter);
1496
1497         return 0;
1498 }
1499
1500 /**
1501  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1502  * @adapter: address of board private structure
1503  * @start: address of beginning of memory
1504  * @len: length of memory
1505  **/
1506 static boolean_t
1507 e1000_check_64k_bound(struct e1000_adapter *adapter,
1508                       void *start, unsigned long len)
1509 {
1510         unsigned long begin = (unsigned long) start;
1511         unsigned long end = begin + len;
1512
1513         /* First rev 82545 and 82546 need to not allow any memory
1514          * write location to cross 64k boundary due to errata 23 */
1515         if (adapter->hw.mac_type == e1000_82545 ||
1516             adapter->hw.mac_type == e1000_82546) {
1517                 return ((begin ^ (end - 1)) >> 16) != 0 ? FALSE : TRUE;
1518         }
1519
1520         return TRUE;
1521 }
1522
1523 /**
1524  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1525  * @adapter: board private structure
1526  * @txdr:    tx descriptor ring (for a specific queue) to setup
1527  *
1528  * Return 0 on success, negative on failure
1529  **/
1530
1531 static int
1532 e1000_setup_tx_resources(struct e1000_adapter *adapter,
1533                          struct e1000_tx_ring *txdr)
1534 {
1535         struct pci_dev *pdev = adapter->pdev;
1536         int size;
1537
1538         size = sizeof(struct e1000_buffer) * txdr->count;
1539         txdr->buffer_info = vmalloc(size);
1540         if (!txdr->buffer_info) {
1541                 DPRINTK(PROBE, ERR,
1542                 "Unable to allocate memory for the transmit descriptor ring\n");
1543                 return -ENOMEM;
1544         }
1545         memset(txdr->buffer_info, 0, size);
1546
1547         /* round up to nearest 4K */
1548
1549         txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1550         txdr->size = ALIGN(txdr->size, 4096);
1551
1552         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1553         if (!txdr->desc) {
1554 setup_tx_desc_die:
1555                 vfree(txdr->buffer_info);
1556                 DPRINTK(PROBE, ERR,
1557                 "Unable to allocate memory for the transmit descriptor ring\n");
1558                 return -ENOMEM;
1559         }
1560
1561         /* Fix for errata 23, can't cross 64kB boundary */
1562         if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1563                 void *olddesc = txdr->desc;
1564                 dma_addr_t olddma = txdr->dma;
1565                 DPRINTK(TX_ERR, ERR, "txdr align check failed: %u bytes "
1566                                      "at %p\n", txdr->size, txdr->desc);
1567                 /* Try again, without freeing the previous */
1568                 txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1569                 /* Failed allocation, critical failure */
1570                 if (!txdr->desc) {
1571                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1572                         goto setup_tx_desc_die;
1573                 }
1574
1575                 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1576                         /* give up */
1577                         pci_free_consistent(pdev, txdr->size, txdr->desc,
1578                                             txdr->dma);
1579                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1580                         DPRINTK(PROBE, ERR,
1581                                 "Unable to allocate aligned memory "
1582                                 "for the transmit descriptor ring\n");
1583                         vfree(txdr->buffer_info);
1584                         return -ENOMEM;
1585                 } else {
1586                         /* Free old allocation, new allocation was successful */
1587                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1588                 }
1589         }
1590         memset(txdr->desc, 0, txdr->size);
1591
1592         txdr->next_to_use = 0;
1593         txdr->next_to_clean = 0;
1594         spin_lock_init(&txdr->tx_lock);
1595
1596         return 0;
1597 }
1598
1599 /**
1600  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1601  *                                (Descriptors) for all queues
1602  * @adapter: board private structure
1603  *
1604  * Return 0 on success, negative on failure
1605  **/
1606
1607 int
1608 e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1609 {
1610         int i, err = 0;
1611
1612         for (i = 0; i < adapter->num_tx_queues; i++) {
1613                 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1614                 if (err) {
1615                         DPRINTK(PROBE, ERR,
1616                                 "Allocation for Tx Queue %u failed\n", i);
1617                         for (i-- ; i >= 0; i--)
1618                                 e1000_free_tx_resources(adapter,
1619                                                         &adapter->tx_ring[i]);
1620                         break;
1621                 }
1622         }
1623
1624         return err;
1625 }
1626
1627 /**
1628  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1629  * @adapter: board private structure
1630  *
1631  * Configure the Tx unit of the MAC after a reset.
1632  **/
1633
1634 static void
1635 e1000_configure_tx(struct e1000_adapter *adapter)
1636 {
1637         uint64_t tdba;
1638         struct e1000_hw *hw = &adapter->hw;
1639         uint32_t tdlen, tctl, tipg, tarc;
1640         uint32_t ipgr1, ipgr2;
1641
1642         /* Setup the HW Tx Head and Tail descriptor pointers */
1643
1644         switch (adapter->num_tx_queues) {
1645         case 1:
1646         default:
1647                 tdba = adapter->tx_ring[0].dma;
1648                 tdlen = adapter->tx_ring[0].count *
1649                         sizeof(struct e1000_tx_desc);
1650                 E1000_WRITE_REG(hw, TDLEN, tdlen);
1651                 E1000_WRITE_REG(hw, TDBAH, (tdba >> 32));
1652                 E1000_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
1653                 E1000_WRITE_REG(hw, TDT, 0);
1654                 E1000_WRITE_REG(hw, TDH, 0);
1655                 adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ? E1000_TDH : E1000_82542_TDH);
1656                 adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ? E1000_TDT : E1000_82542_TDT);
1657                 break;
1658         }
1659
1660         /* Set the default values for the Tx Inter Packet Gap timer */
1661         if (adapter->hw.mac_type <= e1000_82547_rev_2 &&
1662             (hw->media_type == e1000_media_type_fiber ||
1663              hw->media_type == e1000_media_type_internal_serdes))
1664                 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1665         else
1666                 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1667
1668         switch (hw->mac_type) {
1669         case e1000_82542_rev2_0:
1670         case e1000_82542_rev2_1:
1671                 tipg = DEFAULT_82542_TIPG_IPGT;
1672                 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1673                 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1674                 break;
1675         case e1000_80003es2lan:
1676                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1677                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2;
1678                 break;
1679         default:
1680                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1681                 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1682                 break;
1683         }
1684         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1685         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1686         E1000_WRITE_REG(hw, TIPG, tipg);
1687
1688         /* Set the Tx Interrupt Delay register */
1689
1690         E1000_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
1691         if (hw->mac_type >= e1000_82540)
1692                 E1000_WRITE_REG(hw, TADV, adapter->tx_abs_int_delay);
1693
1694         /* Program the Transmit Control Register */
1695
1696         tctl = E1000_READ_REG(hw, TCTL);
1697         tctl &= ~E1000_TCTL_CT;
1698         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1699                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1700
1701         if (hw->mac_type == e1000_82571 || hw->mac_type == e1000_82572) {
1702                 tarc = E1000_READ_REG(hw, TARC0);
1703                 /* set the speed mode bit, we'll clear it if we're not at
1704                  * gigabit link later */
1705                 tarc |= (1 << 21);
1706                 E1000_WRITE_REG(hw, TARC0, tarc);
1707         } else if (hw->mac_type == e1000_80003es2lan) {
1708                 tarc = E1000_READ_REG(hw, TARC0);
1709                 tarc |= 1;
1710                 E1000_WRITE_REG(hw, TARC0, tarc);
1711                 tarc = E1000_READ_REG(hw, TARC1);
1712                 tarc |= 1;
1713                 E1000_WRITE_REG(hw, TARC1, tarc);
1714         }
1715
1716         e1000_config_collision_dist(hw);
1717
1718         /* Setup Transmit Descriptor Settings for eop descriptor */
1719         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1720
1721         /* only set IDE if we are delaying interrupts using the timers */
1722         if (adapter->tx_int_delay)
1723                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1724
1725         if (hw->mac_type < e1000_82543)
1726                 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1727         else
1728                 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1729
1730         /* Cache if we're 82544 running in PCI-X because we'll
1731          * need this to apply a workaround later in the send path. */
1732         if (hw->mac_type == e1000_82544 &&
1733             hw->bus_type == e1000_bus_type_pcix)
1734                 adapter->pcix_82544 = 1;
1735
1736         E1000_WRITE_REG(hw, TCTL, tctl);
1737
1738 }
1739
1740 /**
1741  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1742  * @adapter: board private structure
1743  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1744  *
1745  * Returns 0 on success, negative on failure
1746  **/
1747
1748 static int
1749 e1000_setup_rx_resources(struct e1000_adapter *adapter,
1750                          struct e1000_rx_ring *rxdr)
1751 {
1752         struct pci_dev *pdev = adapter->pdev;
1753         int size, desc_len;
1754
1755         size = sizeof(struct e1000_buffer) * rxdr->count;
1756         rxdr->buffer_info = vmalloc(size);
1757         if (!rxdr->buffer_info) {
1758                 DPRINTK(PROBE, ERR,
1759                 "Unable to allocate memory for the receive descriptor ring\n");
1760                 return -ENOMEM;
1761         }
1762         memset(rxdr->buffer_info, 0, size);
1763
1764         rxdr->ps_page = kcalloc(rxdr->count, sizeof(struct e1000_ps_page),
1765                                 GFP_KERNEL);
1766         if (!rxdr->ps_page) {
1767                 vfree(rxdr->buffer_info);
1768                 DPRINTK(PROBE, ERR,
1769                 "Unable to allocate memory for the receive descriptor ring\n");
1770                 return -ENOMEM;
1771         }
1772
1773         rxdr->ps_page_dma = kcalloc(rxdr->count,
1774                                     sizeof(struct e1000_ps_page_dma),
1775                                     GFP_KERNEL);
1776         if (!rxdr->ps_page_dma) {
1777                 vfree(rxdr->buffer_info);
1778                 kfree(rxdr->ps_page);
1779                 DPRINTK(PROBE, ERR,
1780                 "Unable to allocate memory for the receive descriptor ring\n");
1781                 return -ENOMEM;
1782         }
1783
1784         if (adapter->hw.mac_type <= e1000_82547_rev_2)
1785                 desc_len = sizeof(struct e1000_rx_desc);
1786         else
1787                 desc_len = sizeof(union e1000_rx_desc_packet_split);
1788
1789         /* Round up to nearest 4K */
1790
1791         rxdr->size = rxdr->count * desc_len;
1792         rxdr->size = ALIGN(rxdr->size, 4096);
1793
1794         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1795
1796         if (!rxdr->desc) {
1797                 DPRINTK(PROBE, ERR,
1798                 "Unable to allocate memory for the receive descriptor ring\n");
1799 setup_rx_desc_die:
1800                 vfree(rxdr->buffer_info);
1801                 kfree(rxdr->ps_page);
1802                 kfree(rxdr->ps_page_dma);
1803                 return -ENOMEM;
1804         }
1805
1806         /* Fix for errata 23, can't cross 64kB boundary */
1807         if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1808                 void *olddesc = rxdr->desc;
1809                 dma_addr_t olddma = rxdr->dma;
1810                 DPRINTK(RX_ERR, ERR, "rxdr align check failed: %u bytes "
1811                                      "at %p\n", rxdr->size, rxdr->desc);
1812                 /* Try again, without freeing the previous */
1813                 rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1814                 /* Failed allocation, critical failure */
1815                 if (!rxdr->desc) {
1816                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1817                         DPRINTK(PROBE, ERR,
1818                                 "Unable to allocate memory "
1819                                 "for the receive descriptor ring\n");
1820                         goto setup_rx_desc_die;
1821                 }
1822
1823                 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1824                         /* give up */
1825                         pci_free_consistent(pdev, rxdr->size, rxdr->desc,
1826                                             rxdr->dma);
1827                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1828                         DPRINTK(PROBE, ERR,
1829                                 "Unable to allocate aligned memory "
1830                                 "for the receive descriptor ring\n");
1831                         goto setup_rx_desc_die;
1832                 } else {
1833                         /* Free old allocation, new allocation was successful */
1834                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1835                 }
1836         }
1837         memset(rxdr->desc, 0, rxdr->size);
1838
1839         rxdr->next_to_clean = 0;
1840         rxdr->next_to_use = 0;
1841
1842         return 0;
1843 }
1844
1845 /**
1846  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1847  *                                (Descriptors) for all queues
1848  * @adapter: board private structure
1849  *
1850  * Return 0 on success, negative on failure
1851  **/
1852
1853 int
1854 e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1855 {
1856         int i, err = 0;
1857
1858         for (i = 0; i < adapter->num_rx_queues; i++) {
1859                 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1860                 if (err) {
1861                         DPRINTK(PROBE, ERR,
1862                                 "Allocation for Rx Queue %u failed\n", i);
1863                         for (i-- ; i >= 0; i--)
1864                                 e1000_free_rx_resources(adapter,
1865                                                         &adapter->rx_ring[i]);
1866                         break;
1867                 }
1868         }
1869
1870         return err;
1871 }
1872
1873 /**
1874  * e1000_setup_rctl - configure the receive control registers
1875  * @adapter: Board private structure
1876  **/
1877 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1878                         (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1879 static void
1880 e1000_setup_rctl(struct e1000_adapter *adapter)
1881 {
1882         uint32_t rctl, rfctl;
1883         uint32_t psrctl = 0;
1884 #ifndef CONFIG_E1000_DISABLE_PACKET_SPLIT
1885         uint32_t pages = 0;
1886 #endif
1887
1888         rctl = E1000_READ_REG(&adapter->hw, RCTL);
1889
1890         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1891
1892         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1893                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1894                 (adapter->hw.mc_filter_type << E1000_RCTL_MO_SHIFT);
1895
1896         if (adapter->hw.tbi_compatibility_on == 1)
1897                 rctl |= E1000_RCTL_SBP;
1898         else
1899                 rctl &= ~E1000_RCTL_SBP;
1900
1901         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1902                 rctl &= ~E1000_RCTL_LPE;
1903         else
1904                 rctl |= E1000_RCTL_LPE;
1905
1906         /* Setup buffer sizes */
1907         rctl &= ~E1000_RCTL_SZ_4096;
1908         rctl |= E1000_RCTL_BSEX;
1909         switch (adapter->rx_buffer_len) {
1910                 case E1000_RXBUFFER_256:
1911                         rctl |= E1000_RCTL_SZ_256;
1912                         rctl &= ~E1000_RCTL_BSEX;
1913                         break;
1914                 case E1000_RXBUFFER_512:
1915                         rctl |= E1000_RCTL_SZ_512;
1916                         rctl &= ~E1000_RCTL_BSEX;
1917                         break;
1918                 case E1000_RXBUFFER_1024:
1919                         rctl |= E1000_RCTL_SZ_1024;
1920                         rctl &= ~E1000_RCTL_BSEX;
1921                         break;
1922                 case E1000_RXBUFFER_2048:
1923                 default:
1924                         rctl |= E1000_RCTL_SZ_2048;
1925                         rctl &= ~E1000_RCTL_BSEX;
1926                         break;
1927                 case E1000_RXBUFFER_4096:
1928                         rctl |= E1000_RCTL_SZ_4096;
1929                         break;
1930                 case E1000_RXBUFFER_8192:
1931                         rctl |= E1000_RCTL_SZ_8192;
1932                         break;
1933                 case E1000_RXBUFFER_16384:
1934                         rctl |= E1000_RCTL_SZ_16384;
1935                         break;
1936         }
1937
1938 #ifndef CONFIG_E1000_DISABLE_PACKET_SPLIT
1939         /* 82571 and greater support packet-split where the protocol
1940          * header is placed in skb->data and the packet data is
1941          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
1942          * In the case of a non-split, skb->data is linearly filled,
1943          * followed by the page buffers.  Therefore, skb->data is
1944          * sized to hold the largest protocol header.
1945          */
1946         /* allocations using alloc_page take too long for regular MTU
1947          * so only enable packet split for jumbo frames */
1948         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
1949         if ((adapter->hw.mac_type >= e1000_82571) && (pages <= 3) &&
1950             PAGE_SIZE <= 16384 && (rctl & E1000_RCTL_LPE))
1951                 adapter->rx_ps_pages = pages;
1952         else
1953                 adapter->rx_ps_pages = 0;
1954 #endif
1955         if (adapter->rx_ps_pages) {
1956                 /* Configure extra packet-split registers */
1957                 rfctl = E1000_READ_REG(&adapter->hw, RFCTL);
1958                 rfctl |= E1000_RFCTL_EXTEN;
1959                 /* disable packet split support for IPv6 extension headers,
1960                  * because some malformed IPv6 headers can hang the RX */
1961                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
1962                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
1963
1964                 E1000_WRITE_REG(&adapter->hw, RFCTL, rfctl);
1965
1966                 rctl |= E1000_RCTL_DTYP_PS;
1967
1968                 psrctl |= adapter->rx_ps_bsize0 >>
1969                         E1000_PSRCTL_BSIZE0_SHIFT;
1970
1971                 switch (adapter->rx_ps_pages) {
1972                 case 3:
1973                         psrctl |= PAGE_SIZE <<
1974                                 E1000_PSRCTL_BSIZE3_SHIFT;
1975                 case 2:
1976                         psrctl |= PAGE_SIZE <<
1977                                 E1000_PSRCTL_BSIZE2_SHIFT;
1978                 case 1:
1979                         psrctl |= PAGE_SIZE >>
1980                                 E1000_PSRCTL_BSIZE1_SHIFT;
1981                         break;
1982                 }
1983
1984                 E1000_WRITE_REG(&adapter->hw, PSRCTL, psrctl);
1985         }
1986
1987         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
1988 }
1989
1990 /**
1991  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1992  * @adapter: board private structure
1993  *
1994  * Configure the Rx unit of the MAC after a reset.
1995  **/
1996
1997 static void
1998 e1000_configure_rx(struct e1000_adapter *adapter)
1999 {
2000         uint64_t rdba;
2001         struct e1000_hw *hw = &adapter->hw;
2002         uint32_t rdlen, rctl, rxcsum, ctrl_ext;
2003
2004         if (adapter->rx_ps_pages) {
2005                 /* this is a 32 byte descriptor */
2006                 rdlen = adapter->rx_ring[0].count *
2007                         sizeof(union e1000_rx_desc_packet_split);
2008                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2009                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2010         } else {
2011                 rdlen = adapter->rx_ring[0].count *
2012                         sizeof(struct e1000_rx_desc);
2013                 adapter->clean_rx = e1000_clean_rx_irq;
2014                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2015         }
2016
2017         /* disable receives while setting up the descriptors */
2018         rctl = E1000_READ_REG(hw, RCTL);
2019         E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
2020
2021         /* set the Receive Delay Timer Register */
2022         E1000_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
2023
2024         if (hw->mac_type >= e1000_82540) {
2025                 E1000_WRITE_REG(hw, RADV, adapter->rx_abs_int_delay);
2026                 if (adapter->itr_setting != 0)
2027                         E1000_WRITE_REG(hw, ITR,
2028                                 1000000000 / (adapter->itr * 256));
2029         }
2030
2031         if (hw->mac_type >= e1000_82571) {
2032                 ctrl_ext = E1000_READ_REG(hw, CTRL_EXT);
2033                 /* Reset delay timers after every interrupt */
2034                 ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2035 #ifdef CONFIG_E1000_NAPI
2036                 /* Auto-Mask interrupts upon ICR access */
2037                 ctrl_ext |= E1000_CTRL_EXT_IAME;
2038                 E1000_WRITE_REG(hw, IAM, 0xffffffff);
2039 #endif
2040                 E1000_WRITE_REG(hw, CTRL_EXT, ctrl_ext);
2041                 E1000_WRITE_FLUSH(hw);
2042         }
2043
2044         /* Setup the HW Rx Head and Tail Descriptor Pointers and
2045          * the Base and Length of the Rx Descriptor Ring */
2046         switch (adapter->num_rx_queues) {
2047         case 1:
2048         default:
2049                 rdba = adapter->rx_ring[0].dma;
2050                 E1000_WRITE_REG(hw, RDLEN, rdlen);
2051                 E1000_WRITE_REG(hw, RDBAH, (rdba >> 32));
2052                 E1000_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
2053                 E1000_WRITE_REG(hw, RDT, 0);
2054                 E1000_WRITE_REG(hw, RDH, 0);
2055                 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ? E1000_RDH : E1000_82542_RDH);
2056                 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ? E1000_RDT : E1000_82542_RDT);
2057                 break;
2058         }
2059
2060         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
2061         if (hw->mac_type >= e1000_82543) {
2062                 rxcsum = E1000_READ_REG(hw, RXCSUM);
2063                 if (adapter->rx_csum == TRUE) {
2064                         rxcsum |= E1000_RXCSUM_TUOFL;
2065
2066                         /* Enable 82571 IPv4 payload checksum for UDP fragments
2067                          * Must be used in conjunction with packet-split. */
2068                         if ((hw->mac_type >= e1000_82571) &&
2069                             (adapter->rx_ps_pages)) {
2070                                 rxcsum |= E1000_RXCSUM_IPPCSE;
2071                         }
2072                 } else {
2073                         rxcsum &= ~E1000_RXCSUM_TUOFL;
2074                         /* don't need to clear IPPCSE as it defaults to 0 */
2075                 }
2076                 E1000_WRITE_REG(hw, RXCSUM, rxcsum);
2077         }
2078
2079         /* enable early receives on 82573, only takes effect if using > 2048
2080          * byte total frame size.  for example only for jumbo frames */
2081 #define E1000_ERT_2048 0x100
2082         if (hw->mac_type == e1000_82573)
2083                 E1000_WRITE_REG(hw, ERT, E1000_ERT_2048);
2084
2085         /* Enable Receives */
2086         E1000_WRITE_REG(hw, RCTL, rctl);
2087 }
2088
2089 /**
2090  * e1000_free_tx_resources - Free Tx Resources per Queue
2091  * @adapter: board private structure
2092  * @tx_ring: Tx descriptor ring for a specific queue
2093  *
2094  * Free all transmit software resources
2095  **/
2096
2097 static void
2098 e1000_free_tx_resources(struct e1000_adapter *adapter,
2099                         struct e1000_tx_ring *tx_ring)
2100 {
2101         struct pci_dev *pdev = adapter->pdev;
2102
2103         e1000_clean_tx_ring(adapter, tx_ring);
2104
2105         vfree(tx_ring->buffer_info);
2106         tx_ring->buffer_info = NULL;
2107
2108         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
2109
2110         tx_ring->desc = NULL;
2111 }
2112
2113 /**
2114  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
2115  * @adapter: board private structure
2116  *
2117  * Free all transmit software resources
2118  **/
2119
2120 void
2121 e1000_free_all_tx_resources(struct e1000_adapter *adapter)
2122 {
2123         int i;
2124
2125         for (i = 0; i < adapter->num_tx_queues; i++)
2126                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
2127 }
2128
2129 static void
2130 e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
2131                         struct e1000_buffer *buffer_info)
2132 {
2133         if (buffer_info->dma) {
2134                 pci_unmap_page(adapter->pdev,
2135                                 buffer_info->dma,
2136                                 buffer_info->length,
2137                                 PCI_DMA_TODEVICE);
2138                 buffer_info->dma = 0;
2139         }
2140         if (buffer_info->skb) {
2141                 dev_kfree_skb_any(buffer_info->skb);
2142                 buffer_info->skb = NULL;
2143         }
2144         /* buffer_info must be completely set up in the transmit path */
2145 }
2146
2147 /**
2148  * e1000_clean_tx_ring - Free Tx Buffers
2149  * @adapter: board private structure
2150  * @tx_ring: ring to be cleaned
2151  **/
2152
2153 static void
2154 e1000_clean_tx_ring(struct e1000_adapter *adapter,
2155                     struct e1000_tx_ring *tx_ring)
2156 {
2157         struct e1000_buffer *buffer_info;
2158         unsigned long size;
2159         unsigned int i;
2160
2161         /* Free all the Tx ring sk_buffs */
2162
2163         for (i = 0; i < tx_ring->count; i++) {
2164                 buffer_info = &tx_ring->buffer_info[i];
2165                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2166         }
2167
2168         size = sizeof(struct e1000_buffer) * tx_ring->count;
2169         memset(tx_ring->buffer_info, 0, size);
2170
2171         /* Zero out the descriptor ring */
2172
2173         memset(tx_ring->desc, 0, tx_ring->size);
2174
2175         tx_ring->next_to_use = 0;
2176         tx_ring->next_to_clean = 0;
2177         tx_ring->last_tx_tso = 0;
2178
2179         writel(0, adapter->hw.hw_addr + tx_ring->tdh);
2180         writel(0, adapter->hw.hw_addr + tx_ring->tdt);
2181 }
2182
2183 /**
2184  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
2185  * @adapter: board private structure
2186  **/
2187
2188 static void
2189 e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
2190 {
2191         int i;
2192
2193         for (i = 0; i < adapter->num_tx_queues; i++)
2194                 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2195 }
2196
2197 /**
2198  * e1000_free_rx_resources - Free Rx Resources
2199  * @adapter: board private structure
2200  * @rx_ring: ring to clean the resources from
2201  *
2202  * Free all receive software resources
2203  **/
2204
2205 static void
2206 e1000_free_rx_resources(struct e1000_adapter *adapter,
2207                         struct e1000_rx_ring *rx_ring)
2208 {
2209         struct pci_dev *pdev = adapter->pdev;
2210
2211         e1000_clean_rx_ring(adapter, rx_ring);
2212
2213         vfree(rx_ring->buffer_info);
2214         rx_ring->buffer_info = NULL;
2215         kfree(rx_ring->ps_page);
2216         rx_ring->ps_page = NULL;
2217         kfree(rx_ring->ps_page_dma);
2218         rx_ring->ps_page_dma = NULL;
2219
2220         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2221
2222         rx_ring->desc = NULL;
2223 }
2224
2225 /**
2226  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
2227  * @adapter: board private structure
2228  *
2229  * Free all receive software resources
2230  **/
2231
2232 void
2233 e1000_free_all_rx_resources(struct e1000_adapter *adapter)
2234 {
2235         int i;
2236
2237         for (i = 0; i < adapter->num_rx_queues; i++)
2238                 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
2239 }
2240
2241 /**
2242  * e1000_clean_rx_ring - Free Rx Buffers per Queue
2243  * @adapter: board private structure
2244  * @rx_ring: ring to free buffers from
2245  **/
2246
2247 static void
2248 e1000_clean_rx_ring(struct e1000_adapter *adapter,
2249                     struct e1000_rx_ring *rx_ring)
2250 {
2251         struct e1000_buffer *buffer_info;
2252         struct e1000_ps_page *ps_page;
2253         struct e1000_ps_page_dma *ps_page_dma;
2254         struct pci_dev *pdev = adapter->pdev;
2255         unsigned long size;
2256         unsigned int i, j;
2257
2258         /* Free all the Rx ring sk_buffs */
2259         for (i = 0; i < rx_ring->count; i++) {
2260                 buffer_info = &rx_ring->buffer_info[i];
2261                 if (buffer_info->skb) {
2262                         pci_unmap_single(pdev,
2263                                          buffer_info->dma,
2264                                          buffer_info->length,
2265                                          PCI_DMA_FROMDEVICE);
2266
2267                         dev_kfree_skb(buffer_info->skb);
2268                         buffer_info->skb = NULL;
2269                 }
2270                 ps_page = &rx_ring->ps_page[i];
2271                 ps_page_dma = &rx_ring->ps_page_dma[i];
2272                 for (j = 0; j < adapter->rx_ps_pages; j++) {
2273                         if (!ps_page->ps_page[j]) break;
2274                         pci_unmap_page(pdev,
2275                                        ps_page_dma->ps_page_dma[j],
2276                                        PAGE_SIZE, PCI_DMA_FROMDEVICE);
2277                         ps_page_dma->ps_page_dma[j] = 0;
2278                         put_page(ps_page->ps_page[j]);
2279                         ps_page->ps_page[j] = NULL;
2280                 }
2281         }
2282
2283         size = sizeof(struct e1000_buffer) * rx_ring->count;
2284         memset(rx_ring->buffer_info, 0, size);
2285         size = sizeof(struct e1000_ps_page) * rx_ring->count;
2286         memset(rx_ring->ps_page, 0, size);
2287         size = sizeof(struct e1000_ps_page_dma) * rx_ring->count;
2288         memset(rx_ring->ps_page_dma, 0, size);
2289
2290         /* Zero out the descriptor ring */
2291
2292         memset(rx_ring->desc, 0, rx_ring->size);
2293
2294         rx_ring->next_to_clean = 0;
2295         rx_ring->next_to_use = 0;
2296
2297         writel(0, adapter->hw.hw_addr + rx_ring->rdh);
2298         writel(0, adapter->hw.hw_addr + rx_ring->rdt);
2299 }
2300
2301 /**
2302  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2303  * @adapter: board private structure
2304  **/
2305
2306 static void
2307 e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2308 {
2309         int i;
2310
2311         for (i = 0; i < adapter->num_rx_queues; i++)
2312                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2313 }
2314
2315 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2316  * and memory write and invalidate disabled for certain operations
2317  */
2318 static void
2319 e1000_enter_82542_rst(struct e1000_adapter *adapter)
2320 {
2321         struct net_device *netdev = adapter->netdev;
2322         uint32_t rctl;
2323
2324         e1000_pci_clear_mwi(&adapter->hw);
2325
2326         rctl = E1000_READ_REG(&adapter->hw, RCTL);
2327         rctl |= E1000_RCTL_RST;
2328         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
2329         E1000_WRITE_FLUSH(&adapter->hw);
2330         mdelay(5);
2331
2332         if (netif_running(netdev))
2333                 e1000_clean_all_rx_rings(adapter);
2334 }
2335
2336 static void
2337 e1000_leave_82542_rst(struct e1000_adapter *adapter)
2338 {
2339         struct net_device *netdev = adapter->netdev;
2340         uint32_t rctl;
2341
2342         rctl = E1000_READ_REG(&adapter->hw, RCTL);
2343         rctl &= ~E1000_RCTL_RST;
2344         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
2345         E1000_WRITE_FLUSH(&adapter->hw);
2346         mdelay(5);
2347
2348         if (adapter->hw.pci_cmd_word & PCI_COMMAND_INVALIDATE)
2349                 e1000_pci_set_mwi(&adapter->hw);
2350
2351         if (netif_running(netdev)) {
2352                 /* No need to loop, because 82542 supports only 1 queue */
2353                 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2354                 e1000_configure_rx(adapter);
2355                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2356         }
2357 }
2358
2359 /**
2360  * e1000_set_mac - Change the Ethernet Address of the NIC
2361  * @netdev: network interface device structure
2362  * @p: pointer to an address structure
2363  *
2364  * Returns 0 on success, negative on failure
2365  **/
2366
2367 static int
2368 e1000_set_mac(struct net_device *netdev, void *p)
2369 {
2370         struct e1000_adapter *adapter = netdev_priv(netdev);
2371         struct sockaddr *addr = p;
2372
2373         if (!is_valid_ether_addr(addr->sa_data))
2374                 return -EADDRNOTAVAIL;
2375
2376         /* 82542 2.0 needs to be in reset to write receive address registers */
2377
2378         if (adapter->hw.mac_type == e1000_82542_rev2_0)
2379                 e1000_enter_82542_rst(adapter);
2380
2381         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2382         memcpy(adapter->hw.mac_addr, addr->sa_data, netdev->addr_len);
2383
2384         e1000_rar_set(&adapter->hw, adapter->hw.mac_addr, 0);
2385
2386         /* With 82571 controllers, LAA may be overwritten (with the default)
2387          * due to controller reset from the other port. */
2388         if (adapter->hw.mac_type == e1000_82571) {
2389                 /* activate the work around */
2390                 adapter->hw.laa_is_present = 1;
2391
2392                 /* Hold a copy of the LAA in RAR[14] This is done so that
2393                  * between the time RAR[0] gets clobbered  and the time it
2394                  * gets fixed (in e1000_watchdog), the actual LAA is in one
2395                  * of the RARs and no incoming packets directed to this port
2396                  * are dropped. Eventaully the LAA will be in RAR[0] and
2397                  * RAR[14] */
2398                 e1000_rar_set(&adapter->hw, adapter->hw.mac_addr,
2399                                         E1000_RAR_ENTRIES - 1);
2400         }
2401
2402         if (adapter->hw.mac_type == e1000_82542_rev2_0)
2403                 e1000_leave_82542_rst(adapter);
2404
2405         return 0;
2406 }
2407
2408 /**
2409  * e1000_set_multi - Multicast and Promiscuous mode set
2410  * @netdev: network interface device structure
2411  *
2412  * The set_multi entry point is called whenever the multicast address
2413  * list or the network interface flags are updated.  This routine is
2414  * responsible for configuring the hardware for proper multicast,
2415  * promiscuous mode, and all-multi behavior.
2416  **/
2417
2418 static void
2419 e1000_set_multi(struct net_device *netdev)
2420 {
2421         struct e1000_adapter *adapter = netdev_priv(netdev);
2422         struct e1000_hw *hw = &adapter->hw;
2423         struct dev_mc_list *mc_ptr;
2424         uint32_t rctl;
2425         uint32_t hash_value;
2426         int i, rar_entries = E1000_RAR_ENTRIES;
2427         int mta_reg_count = (hw->mac_type == e1000_ich8lan) ?
2428                                 E1000_NUM_MTA_REGISTERS_ICH8LAN :
2429                                 E1000_NUM_MTA_REGISTERS;
2430
2431         if (adapter->hw.mac_type == e1000_ich8lan)
2432                 rar_entries = E1000_RAR_ENTRIES_ICH8LAN;
2433
2434         /* reserve RAR[14] for LAA over-write work-around */
2435         if (adapter->hw.mac_type == e1000_82571)
2436                 rar_entries--;
2437
2438         /* Check for Promiscuous and All Multicast modes */
2439
2440         rctl = E1000_READ_REG(hw, RCTL);
2441
2442         if (netdev->flags & IFF_PROMISC) {
2443                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2444         } else if (netdev->flags & IFF_ALLMULTI) {
2445                 rctl |= E1000_RCTL_MPE;
2446                 rctl &= ~E1000_RCTL_UPE;
2447         } else {
2448                 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2449         }
2450
2451         E1000_WRITE_REG(hw, RCTL, rctl);
2452
2453         /* 82542 2.0 needs to be in reset to write receive address registers */
2454
2455         if (hw->mac_type == e1000_82542_rev2_0)
2456                 e1000_enter_82542_rst(adapter);
2457
2458         /* load the first 14 multicast address into the exact filters 1-14
2459          * RAR 0 is used for the station MAC adddress
2460          * if there are not 14 addresses, go ahead and clear the filters
2461          * -- with 82571 controllers only 0-13 entries are filled here
2462          */
2463         mc_ptr = netdev->mc_list;
2464
2465         for (i = 1; i < rar_entries; i++) {
2466                 if (mc_ptr) {
2467                         e1000_rar_set(hw, mc_ptr->dmi_addr, i);
2468                         mc_ptr = mc_ptr->next;
2469                 } else {
2470                         E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2471                         E1000_WRITE_FLUSH(hw);
2472                         E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2473                         E1000_WRITE_FLUSH(hw);
2474                 }
2475         }
2476
2477         /* clear the old settings from the multicast hash table */
2478
2479         for (i = 0; i < mta_reg_count; i++) {
2480                 E1000_WRITE_REG_ARRAY(hw, MTA, i, 0);
2481                 E1000_WRITE_FLUSH(hw);
2482         }
2483
2484         /* load any remaining addresses into the hash table */
2485
2486         for (; mc_ptr; mc_ptr = mc_ptr->next) {
2487                 hash_value = e1000_hash_mc_addr(hw, mc_ptr->dmi_addr);
2488                 e1000_mta_set(hw, hash_value);
2489         }
2490
2491         if (hw->mac_type == e1000_82542_rev2_0)
2492                 e1000_leave_82542_rst(adapter);
2493 }
2494
2495 /* Need to wait a few seconds after link up to get diagnostic information from
2496  * the phy */
2497
2498 static void
2499 e1000_update_phy_info(unsigned long data)
2500 {
2501         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2502         e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
2503 }
2504
2505 /**
2506  * e1000_82547_tx_fifo_stall - Timer Call-back
2507  * @data: pointer to adapter cast into an unsigned long
2508  **/
2509
2510 static void
2511 e1000_82547_tx_fifo_stall(unsigned long data)
2512 {
2513         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2514         struct net_device *netdev = adapter->netdev;
2515         uint32_t tctl;
2516
2517         if (atomic_read(&adapter->tx_fifo_stall)) {
2518                 if ((E1000_READ_REG(&adapter->hw, TDT) ==
2519                     E1000_READ_REG(&adapter->hw, TDH)) &&
2520                    (E1000_READ_REG(&adapter->hw, TDFT) ==
2521                     E1000_READ_REG(&adapter->hw, TDFH)) &&
2522                    (E1000_READ_REG(&adapter->hw, TDFTS) ==
2523                     E1000_READ_REG(&adapter->hw, TDFHS))) {
2524                         tctl = E1000_READ_REG(&adapter->hw, TCTL);
2525                         E1000_WRITE_REG(&adapter->hw, TCTL,
2526                                         tctl & ~E1000_TCTL_EN);
2527                         E1000_WRITE_REG(&adapter->hw, TDFT,
2528                                         adapter->tx_head_addr);
2529                         E1000_WRITE_REG(&adapter->hw, TDFH,
2530                                         adapter->tx_head_addr);
2531                         E1000_WRITE_REG(&adapter->hw, TDFTS,
2532                                         adapter->tx_head_addr);
2533                         E1000_WRITE_REG(&adapter->hw, TDFHS,
2534                                         adapter->tx_head_addr);
2535                         E1000_WRITE_REG(&adapter->hw, TCTL, tctl);
2536                         E1000_WRITE_FLUSH(&adapter->hw);
2537
2538                         adapter->tx_fifo_head = 0;
2539                         atomic_set(&adapter->tx_fifo_stall, 0);
2540                         netif_wake_queue(netdev);
2541                 } else {
2542                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2543                 }
2544         }
2545 }
2546
2547 /**
2548  * e1000_watchdog - Timer Call-back
2549  * @data: pointer to adapter cast into an unsigned long
2550  **/
2551 static void
2552 e1000_watchdog(unsigned long data)
2553 {
2554         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2555         struct net_device *netdev = adapter->netdev;
2556         struct e1000_tx_ring *txdr = adapter->tx_ring;
2557         uint32_t link, tctl;
2558         int32_t ret_val;
2559
2560         ret_val = e1000_check_for_link(&adapter->hw);
2561         if ((ret_val == E1000_ERR_PHY) &&
2562             (adapter->hw.phy_type == e1000_phy_igp_3) &&
2563             (E1000_READ_REG(&adapter->hw, CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
2564                 /* See e1000_kumeran_lock_loss_workaround() */
2565                 DPRINTK(LINK, INFO,
2566                         "Gigabit has been disabled, downgrading speed\n");
2567         }
2568
2569         if (adapter->hw.mac_type == e1000_82573) {
2570                 e1000_enable_tx_pkt_filtering(&adapter->hw);
2571                 if (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id)
2572                         e1000_update_mng_vlan(adapter);
2573         }
2574
2575         if ((adapter->hw.media_type == e1000_media_type_internal_serdes) &&
2576            !(E1000_READ_REG(&adapter->hw, TXCW) & E1000_TXCW_ANE))
2577                 link = !adapter->hw.serdes_link_down;
2578         else
2579                 link = E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_LU;
2580
2581         if (link) {
2582                 if (!netif_carrier_ok(netdev)) {
2583                         uint32_t ctrl;
2584                         boolean_t txb2b = 1;
2585                         e1000_get_speed_and_duplex(&adapter->hw,
2586                                                    &adapter->link_speed,
2587                                                    &adapter->link_duplex);
2588
2589                         ctrl = E1000_READ_REG(&adapter->hw, CTRL);
2590                         DPRINTK(LINK, INFO, "NIC Link is Up %d Mbps %s, "
2591                                 "Flow Control: %s\n",
2592                                 adapter->link_speed,
2593                                 adapter->link_duplex == FULL_DUPLEX ?
2594                                 "Full Duplex" : "Half Duplex",
2595                                 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2596                                 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2597                                 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2598                                 E1000_CTRL_TFCE) ? "TX" : "None" )));
2599
2600                         /* tweak tx_queue_len according to speed/duplex
2601                          * and adjust the timeout factor */
2602                         netdev->tx_queue_len = adapter->tx_queue_len;
2603                         adapter->tx_timeout_factor = 1;
2604                         switch (adapter->link_speed) {
2605                         case SPEED_10:
2606                                 txb2b = 0;
2607                                 netdev->tx_queue_len = 10;
2608                                 adapter->tx_timeout_factor = 8;
2609                                 break;
2610                         case SPEED_100:
2611                                 txb2b = 0;
2612                                 netdev->tx_queue_len = 100;
2613                                 /* maybe add some timeout factor ? */
2614                                 break;
2615                         }
2616
2617                         if ((adapter->hw.mac_type == e1000_82571 ||
2618                              adapter->hw.mac_type == e1000_82572) &&
2619                             txb2b == 0) {
2620                                 uint32_t tarc0;
2621                                 tarc0 = E1000_READ_REG(&adapter->hw, TARC0);
2622                                 tarc0 &= ~(1 << 21);
2623                                 E1000_WRITE_REG(&adapter->hw, TARC0, tarc0);
2624                         }
2625
2626                         /* disable TSO for pcie and 10/100 speeds, to avoid
2627                          * some hardware issues */
2628                         if (!adapter->tso_force &&
2629                             adapter->hw.bus_type == e1000_bus_type_pci_express){
2630                                 switch (adapter->link_speed) {
2631                                 case SPEED_10:
2632                                 case SPEED_100:
2633                                         DPRINTK(PROBE,INFO,
2634                                         "10/100 speed: disabling TSO\n");
2635                                         netdev->features &= ~NETIF_F_TSO;
2636                                         netdev->features &= ~NETIF_F_TSO6;
2637                                         break;
2638                                 case SPEED_1000:
2639                                         netdev->features |= NETIF_F_TSO;
2640                                         netdev->features |= NETIF_F_TSO6;
2641                                         break;
2642                                 default:
2643                                         /* oops */
2644                                         break;
2645                                 }
2646                         }
2647
2648                         /* enable transmits in the hardware, need to do this
2649                          * after setting TARC0 */
2650                         tctl = E1000_READ_REG(&adapter->hw, TCTL);
2651                         tctl |= E1000_TCTL_EN;
2652                         E1000_WRITE_REG(&adapter->hw, TCTL, tctl);
2653
2654                         netif_carrier_on(netdev);
2655                         netif_wake_queue(netdev);
2656                         mod_timer(&adapter->phy_info_timer, round_jiffies(jiffies + 2 * HZ));
2657                         adapter->smartspeed = 0;
2658                 } else {
2659                         /* make sure the receive unit is started */
2660                         if (adapter->hw.rx_needs_kicking) {
2661                                 struct e1000_hw *hw = &adapter->hw;
2662                                 uint32_t rctl = E1000_READ_REG(hw, RCTL);
2663                                 E1000_WRITE_REG(hw, RCTL, rctl | E1000_RCTL_EN);
2664                         }
2665                 }
2666         } else {
2667                 if (netif_carrier_ok(netdev)) {
2668                         adapter->link_speed = 0;
2669                         adapter->link_duplex = 0;
2670                         DPRINTK(LINK, INFO, "NIC Link is Down\n");
2671                         netif_carrier_off(netdev);
2672                         netif_stop_queue(netdev);
2673                         mod_timer(&adapter->phy_info_timer, round_jiffies(jiffies + 2 * HZ));
2674
2675                         /* 80003ES2LAN workaround--
2676                          * For packet buffer work-around on link down event;
2677                          * disable receives in the ISR and
2678                          * reset device here in the watchdog
2679                          */
2680                         if (adapter->hw.mac_type == e1000_80003es2lan)
2681                                 /* reset device */
2682                                 schedule_work(&adapter->reset_task);
2683                 }
2684
2685                 e1000_smartspeed(adapter);
2686         }
2687
2688         e1000_update_stats(adapter);
2689
2690         adapter->hw.tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2691         adapter->tpt_old = adapter->stats.tpt;
2692         adapter->hw.collision_delta = adapter->stats.colc - adapter->colc_old;
2693         adapter->colc_old = adapter->stats.colc;
2694
2695         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2696         adapter->gorcl_old = adapter->stats.gorcl;
2697         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2698         adapter->gotcl_old = adapter->stats.gotcl;
2699
2700         e1000_update_adaptive(&adapter->hw);
2701
2702         if (!netif_carrier_ok(netdev)) {
2703                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2704                         /* We've lost link, so the controller stops DMA,
2705                          * but we've got queued Tx work that's never going
2706                          * to get done, so reset controller to flush Tx.
2707                          * (Do the reset outside of interrupt context). */
2708                         adapter->tx_timeout_count++;
2709                         schedule_work(&adapter->reset_task);
2710                 }
2711         }
2712
2713         /* Cause software interrupt to ensure rx ring is cleaned */
2714         E1000_WRITE_REG(&adapter->hw, ICS, E1000_ICS_RXDMT0);
2715
2716         /* Force detection of hung controller every watchdog period */
2717         adapter->detect_tx_hung = TRUE;
2718
2719         /* With 82571 controllers, LAA may be overwritten due to controller
2720          * reset from the other port. Set the appropriate LAA in RAR[0] */
2721         if (adapter->hw.mac_type == e1000_82571 && adapter->hw.laa_is_present)
2722                 e1000_rar_set(&adapter->hw, adapter->hw.mac_addr, 0);
2723
2724         /* Reset the timer */
2725         mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
2726 }
2727
2728 enum latency_range {
2729         lowest_latency = 0,
2730         low_latency = 1,
2731         bulk_latency = 2,
2732         latency_invalid = 255
2733 };
2734
2735 /**
2736  * e1000_update_itr - update the dynamic ITR value based on statistics
2737  *      Stores a new ITR value based on packets and byte
2738  *      counts during the last interrupt.  The advantage of per interrupt
2739  *      computation is faster updates and more accurate ITR for the current
2740  *      traffic pattern.  Constants in this function were computed
2741  *      based on theoretical maximum wire speed and thresholds were set based
2742  *      on testing data as well as attempting to minimize response time
2743  *      while increasing bulk throughput.
2744  *      this functionality is controlled by the InterruptThrottleRate module
2745  *      parameter (see e1000_param.c)
2746  * @adapter: pointer to adapter
2747  * @itr_setting: current adapter->itr
2748  * @packets: the number of packets during this measurement interval
2749  * @bytes: the number of bytes during this measurement interval
2750  **/
2751 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2752                                    uint16_t itr_setting,
2753                                    int packets,
2754                                    int bytes)
2755 {
2756         unsigned int retval = itr_setting;
2757         struct e1000_hw *hw = &adapter->hw;
2758
2759         if (unlikely(hw->mac_type < e1000_82540))
2760                 goto update_itr_done;
2761
2762         if (packets == 0)
2763                 goto update_itr_done;
2764
2765         switch (itr_setting) {
2766         case lowest_latency:
2767                 /* jumbo frames get bulk treatment*/
2768                 if (bytes/packets > 8000)
2769                         retval = bulk_latency;
2770                 else if ((packets < 5) && (bytes > 512))
2771                         retval = low_latency;
2772                 break;
2773         case low_latency:  /* 50 usec aka 20000 ints/s */
2774                 if (bytes > 10000) {
2775                         /* jumbo frames need bulk latency setting */
2776                         if (bytes/packets > 8000)
2777                                 retval = bulk_latency;
2778                         else if ((packets < 10) || ((bytes/packets) > 1200))
2779                                 retval = bulk_latency;
2780                         else if ((packets > 35))
2781                                 retval = lowest_latency;
2782                 } else if (bytes/packets > 2000)
2783                         retval = bulk_latency;
2784                 else if (packets <= 2 && bytes < 512)
2785                         retval = lowest_latency;
2786                 break;
2787         case bulk_latency: /* 250 usec aka 4000 ints/s */
2788                 if (bytes > 25000) {
2789                         if (packets > 35)
2790                                 retval = low_latency;
2791                 } else if (bytes < 6000) {
2792                         retval = low_latency;
2793                 }
2794                 break;
2795         }
2796
2797 update_itr_done:
2798         return retval;
2799 }
2800
2801 static void e1000_set_itr(struct e1000_adapter *adapter)
2802 {
2803         struct e1000_hw *hw = &adapter->hw;
2804         uint16_t current_itr;
2805         uint32_t new_itr = adapter->itr;
2806
2807         if (unlikely(hw->mac_type < e1000_82540))
2808                 return;
2809
2810         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2811         if (unlikely(adapter->link_speed != SPEED_1000)) {
2812                 current_itr = 0;
2813                 new_itr = 4000;
2814                 goto set_itr_now;
2815         }
2816
2817         adapter->tx_itr = e1000_update_itr(adapter,
2818                                     adapter->tx_itr,
2819                                     adapter->total_tx_packets,
2820                                     adapter->total_tx_bytes);
2821         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2822         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2823                 adapter->tx_itr = low_latency;
2824
2825         adapter->rx_itr = e1000_update_itr(adapter,
2826                                     adapter->rx_itr,
2827                                     adapter->total_rx_packets,
2828                                     adapter->total_rx_bytes);
2829         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2830         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2831                 adapter->rx_itr = low_latency;
2832
2833         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2834
2835         switch (current_itr) {
2836         /* counts and packets in update_itr are dependent on these numbers */
2837         case lowest_latency:
2838                 new_itr = 70000;
2839                 break;
2840         case low_latency:
2841                 new_itr = 20000; /* aka hwitr = ~200 */
2842                 break;
2843         case bulk_latency:
2844                 new_itr = 4000;
2845                 break;
2846         default:
2847                 break;
2848         }
2849
2850 set_itr_now:
2851         if (new_itr != adapter->itr) {
2852                 /* this attempts to bias the interrupt rate towards Bulk
2853                  * by adding intermediate steps when interrupt rate is
2854                  * increasing */
2855                 new_itr = new_itr > adapter->itr ?
2856                              min(adapter->itr + (new_itr >> 2), new_itr) :
2857                              new_itr;
2858                 adapter->itr = new_itr;
2859                 E1000_WRITE_REG(hw, ITR, 1000000000 / (new_itr * 256));
2860         }
2861
2862         return;
2863 }
2864
2865 #define E1000_TX_FLAGS_CSUM             0x00000001
2866 #define E1000_TX_FLAGS_VLAN             0x00000002
2867 #define E1000_TX_FLAGS_TSO              0x00000004
2868 #define E1000_TX_FLAGS_IPV4             0x00000008
2869 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2870 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2871
2872 static int
2873 e1000_tso(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2874           struct sk_buff *skb)
2875 {
2876         struct e1000_context_desc *context_desc;
2877         struct e1000_buffer *buffer_info;
2878         unsigned int i;
2879         uint32_t cmd_length = 0;
2880         uint16_t ipcse = 0, tucse, mss;
2881         uint8_t ipcss, ipcso, tucss, tucso, hdr_len;
2882         int err;
2883
2884         if (skb_is_gso(skb)) {
2885                 if (skb_header_cloned(skb)) {
2886                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2887                         if (err)
2888                                 return err;
2889                 }
2890
2891                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2892                 mss = skb_shinfo(skb)->gso_size;
2893                 if (skb->protocol == htons(ETH_P_IP)) {
2894                         struct iphdr *iph = ip_hdr(skb);
2895                         iph->tot_len = 0;
2896                         iph->check = 0;
2897                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2898                                                                  iph->daddr, 0,
2899                                                                  IPPROTO_TCP,
2900                                                                  0);
2901                         cmd_length = E1000_TXD_CMD_IP;
2902                         ipcse = skb_transport_offset(skb) - 1;
2903                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2904                         ipv6_hdr(skb)->payload_len = 0;
2905                         tcp_hdr(skb)->check =
2906                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2907                                                  &ipv6_hdr(skb)->daddr,
2908                                                  0, IPPROTO_TCP, 0);
2909                         ipcse = 0;
2910                 }
2911                 ipcss = skb_network_offset(skb);
2912                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2913                 tucss = skb_transport_offset(skb);
2914                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2915                 tucse = 0;
2916
2917                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2918                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2919
2920                 i = tx_ring->next_to_use;
2921                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2922                 buffer_info = &tx_ring->buffer_info[i];
2923
2924                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2925                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2926                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2927                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2928                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2929                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2930                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2931                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2932                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2933
2934                 buffer_info->time_stamp = jiffies;
2935                 buffer_info->next_to_watch = i;
2936
2937                 if (++i == tx_ring->count) i = 0;
2938                 tx_ring->next_to_use = i;
2939
2940                 return TRUE;
2941         }
2942         return FALSE;
2943 }
2944
2945 static boolean_t
2946 e1000_tx_csum(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2947               struct sk_buff *skb)
2948 {
2949         struct e1000_context_desc *context_desc;
2950         struct e1000_buffer *buffer_info;
2951         unsigned int i;
2952         uint8_t css;
2953
2954         if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
2955                 css = skb_transport_offset(skb);
2956
2957                 i = tx_ring->next_to_use;
2958                 buffer_info = &tx_ring->buffer_info[i];
2959                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2960
2961                 context_desc->lower_setup.ip_config = 0;
2962                 context_desc->upper_setup.tcp_fields.tucss = css;
2963                 context_desc->upper_setup.tcp_fields.tucso =
2964                         css + skb->csum_offset;
2965                 context_desc->upper_setup.tcp_fields.tucse = 0;
2966                 context_desc->tcp_seg_setup.data = 0;
2967                 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
2968
2969                 buffer_info->time_stamp = jiffies;
2970                 buffer_info->next_to_watch = i;
2971
2972                 if (unlikely(++i == tx_ring->count)) i = 0;
2973                 tx_ring->next_to_use = i;
2974
2975                 return TRUE;
2976         }
2977
2978         return FALSE;
2979 }
2980
2981 #define E1000_MAX_TXD_PWR       12
2982 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2983
2984 static int
2985 e1000_tx_map(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2986              struct sk_buff *skb, unsigned int first, unsigned int max_per_txd,
2987              unsigned int nr_frags, unsigned int mss)
2988 {
2989         struct e1000_buffer *buffer_info;
2990         unsigned int len = skb->len;
2991         unsigned int offset = 0, size, count = 0, i;
2992         unsigned int f;
2993         len -= skb->data_len;
2994
2995         i = tx_ring->next_to_use;
2996
2997         while (len) {
2998                 buffer_info = &tx_ring->buffer_info[i];
2999                 size = min(len, max_per_txd);
3000                 /* Workaround for Controller erratum --
3001                  * descriptor for non-tso packet in a linear SKB that follows a
3002                  * tso gets written back prematurely before the data is fully
3003                  * DMA'd to the controller */
3004                 if (!skb->data_len && tx_ring->last_tx_tso &&
3005                     !skb_is_gso(skb)) {
3006                         tx_ring->last_tx_tso = 0;
3007                         size -= 4;
3008                 }
3009
3010                 /* Workaround for premature desc write-backs
3011                  * in TSO mode.  Append 4-byte sentinel desc */
3012                 if (unlikely(mss && !nr_frags && size == len && size > 8))
3013                         size -= 4;
3014                 /* work-around for errata 10 and it applies
3015                  * to all controllers in PCI-X mode
3016                  * The fix is to make sure that the first descriptor of a
3017                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
3018                  */
3019                 if (unlikely((adapter->hw.bus_type == e1000_bus_type_pcix) &&
3020                                 (size > 2015) && count == 0))
3021                         size = 2015;
3022
3023                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
3024                  * terminating buffers within evenly-aligned dwords. */
3025                 if (unlikely(adapter->pcix_82544 &&
3026                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
3027                    size > 4))
3028                         size -= 4;
3029
3030                 buffer_info->length = size;
3031                 buffer_info->dma =
3032                         pci_map_single(adapter->pdev,
3033                                 skb->data + offset,
3034                                 size,
3035                                 PCI_DMA_TODEVICE);
3036                 buffer_info->time_stamp = jiffies;
3037                 buffer_info->next_to_watch = i;
3038
3039                 len -= size;
3040                 offset += size;
3041                 count++;
3042                 if (unlikely(++i == tx_ring->count)) i = 0;
3043         }
3044
3045         for (f = 0; f < nr_frags; f++) {
3046                 struct skb_frag_struct *frag;
3047
3048                 frag = &skb_shinfo(skb)->frags[f];
3049                 len = frag->size;
3050                 offset = frag->page_offset;
3051
3052                 while (len) {
3053                         buffer_info = &tx_ring->buffer_info[i];
3054                         size = min(len, max_per_txd);
3055                         /* Workaround for premature desc write-backs
3056                          * in TSO mode.  Append 4-byte sentinel desc */
3057                         if (unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
3058                                 size -= 4;
3059                         /* Workaround for potential 82544 hang in PCI-X.
3060                          * Avoid terminating buffers within evenly-aligned
3061                          * dwords. */
3062                         if (unlikely(adapter->pcix_82544 &&
3063                            !((unsigned long)(frag->page+offset+size-1) & 4) &&
3064                            size > 4))
3065                                 size -= 4;
3066
3067                         buffer_info->length = size;
3068                         buffer_info->dma =
3069                                 pci_map_page(adapter->pdev,
3070                                         frag->page,
3071                                         offset,
3072                                         size,
3073                                         PCI_DMA_TODEVICE);
3074                         buffer_info->time_stamp = jiffies;
3075                         buffer_info->next_to_watch = i;
3076
3077                         len -= size;
3078                         offset += size;
3079                         count++;
3080                         if (unlikely(++i == tx_ring->count)) i = 0;
3081                 }
3082         }
3083
3084         i = (i == 0) ? tx_ring->count - 1 : i - 1;
3085         tx_ring->buffer_info[i].skb = skb;
3086         tx_ring->buffer_info[first].next_to_watch = i;
3087
3088         return count;
3089 }
3090
3091 static void
3092 e1000_tx_queue(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
3093                int tx_flags, int count)
3094 {
3095         struct e1000_tx_desc *tx_desc = NULL;
3096         struct e1000_buffer *buffer_info;
3097         uint32_t txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3098         unsigned int i;
3099
3100         if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
3101                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3102                              E1000_TXD_CMD_TSE;
3103                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3104
3105                 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
3106                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3107         }
3108
3109         if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
3110                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3111                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3112         }
3113
3114         if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
3115                 txd_lower |= E1000_TXD_CMD_VLE;
3116                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3117         }
3118
3119         i = tx_ring->next_to_use;
3120
3121         while (count--) {
3122                 buffer_info = &tx_ring->buffer_info[i];
3123                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3124                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3125                 tx_desc->lower.data =
3126                         cpu_to_le32(txd_lower | buffer_info->length);
3127                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3128                 if (unlikely(++i == tx_ring->count)) i = 0;
3129         }
3130
3131         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3132
3133         /* Force memory writes to complete before letting h/w
3134          * know there are new descriptors to fetch.  (Only
3135          * applicable for weak-ordered memory model archs,
3136          * such as IA-64). */
3137         wmb();
3138
3139         tx_ring->next_to_use = i;
3140         writel(i, adapter->hw.hw_addr + tx_ring->tdt);
3141         /* we need this if more than one processor can write to our tail
3142          * at a time, it syncronizes IO on IA64/Altix systems */
3143         mmiowb();
3144 }
3145
3146 /**
3147  * 82547 workaround to avoid controller hang in half-duplex environment.
3148  * The workaround is to avoid queuing a large packet that would span
3149  * the internal Tx FIFO ring boundary by notifying the stack to resend
3150  * the packet at a later time.  This gives the Tx FIFO an opportunity to
3151  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
3152  * to the beginning of the Tx FIFO.
3153  **/
3154
3155 #define E1000_FIFO_HDR                  0x10
3156 #define E1000_82547_PAD_LEN             0x3E0
3157
3158 static int
3159 e1000_82547_fifo_workaround(struct e1000_adapter *adapter, struct sk_buff *skb)
3160 {
3161         uint32_t fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
3162         uint32_t skb_fifo_len = skb->len + E1000_FIFO_HDR;
3163
3164         skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
3165
3166         if (adapter->link_duplex != HALF_DUPLEX)
3167                 goto no_fifo_stall_required;
3168
3169         if (atomic_read(&adapter->tx_fifo_stall))
3170                 return 1;
3171
3172         if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
3173                 atomic_set(&adapter->tx_fifo_stall, 1);
3174                 return 1;
3175         }
3176
3177 no_fifo_stall_required:
3178         adapter->tx_fifo_head += skb_fifo_len;
3179         if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
3180                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
3181         return 0;
3182 }
3183
3184 #define MINIMUM_DHCP_PACKET_SIZE 282
3185 static int
3186 e1000_transfer_dhcp_info(struct e1000_adapter *adapter, struct sk_buff *skb)
3187 {
3188         struct e1000_hw *hw =  &adapter->hw;
3189         uint16_t length, offset;
3190         if (vlan_tx_tag_present(skb)) {
3191                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
3192                         ( adapter->hw.mng_cookie.status &
3193                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) )
3194                         return 0;
3195         }
3196         if (skb->len > MINIMUM_DHCP_PACKET_SIZE) {
3197                 struct ethhdr *eth = (struct ethhdr *) skb->data;
3198                 if ((htons(ETH_P_IP) == eth->h_proto)) {
3199                         const struct iphdr *ip =
3200                                 (struct iphdr *)((uint8_t *)skb->data+14);
3201                         if (IPPROTO_UDP == ip->protocol) {
3202                                 struct udphdr *udp =
3203                                         (struct udphdr *)((uint8_t *)ip +
3204                                                 (ip->ihl << 2));
3205                                 if (ntohs(udp->dest) == 67) {
3206                                         offset = (uint8_t *)udp + 8 - skb->data;
3207                                         length = skb->len - offset;
3208
3209                                         return e1000_mng_write_dhcp_info(hw,
3210                                                         (uint8_t *)udp + 8,
3211                                                         length);
3212                                 }
3213                         }
3214                 }
3215         }
3216         return 0;
3217 }
3218
3219 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3220 {
3221         struct e1000_adapter *adapter = netdev_priv(netdev);
3222         struct e1000_tx_ring *tx_ring = adapter->tx_ring;
3223
3224         netif_stop_queue(netdev);
3225         /* Herbert's original patch had:
3226          *  smp_mb__after_netif_stop_queue();
3227          * but since that doesn't exist yet, just open code it. */
3228         smp_mb();
3229
3230         /* We need to check again in a case another CPU has just
3231          * made room available. */
3232         if (likely(E1000_DESC_UNUSED(tx_ring) < size))
3233                 return -EBUSY;
3234
3235         /* A reprieve! */
3236         netif_start_queue(netdev);
3237         ++adapter->restart_queue;
3238         return 0;
3239 }
3240
3241 static int e1000_maybe_stop_tx(struct net_device *netdev,
3242                                struct e1000_tx_ring *tx_ring, int size)
3243 {
3244         if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
3245                 return 0;
3246         return __e1000_maybe_stop_tx(netdev, size);
3247 }
3248
3249 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3250 static int
3251 e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3252 {
3253         struct e1000_adapter *adapter = netdev_priv(netdev);
3254         struct e1000_tx_ring *tx_ring;
3255         unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
3256         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3257         unsigned int tx_flags = 0;
3258         unsigned int len = skb->len;
3259         unsigned long flags;
3260         unsigned int nr_frags = 0;
3261         unsigned int mss = 0;
3262         int count = 0;
3263         int tso;
3264         unsigned int f;
3265         len -= skb->data_len;
3266
3267         /* This goes back to the question of how to logically map a tx queue
3268          * to a flow.  Right now, performance is impacted slightly negatively
3269          * if using multiple tx queues.  If the stack breaks away from a
3270          * single qdisc implementation, we can look at this again. */
3271         tx_ring = adapter->tx_ring;
3272
3273         if (unlikely(skb->len <= 0)) {
3274                 dev_kfree_skb_any(skb);
3275                 return NETDEV_TX_OK;
3276         }
3277
3278         /* 82571 and newer doesn't need the workaround that limited descriptor
3279          * length to 4kB */
3280         if (adapter->hw.mac_type >= e1000_82571)
3281                 max_per_txd = 8192;
3282
3283         mss = skb_shinfo(skb)->gso_size;
3284         /* The controller does a simple calculation to
3285          * make sure there is enough room in the FIFO before
3286          * initiating the DMA for each buffer.  The calc is:
3287          * 4 = ceil(buffer len/mss).  To make sure we don't
3288          * overrun the FIFO, adjust the max buffer len if mss
3289          * drops. */
3290         if (mss) {
3291                 uint8_t hdr_len;
3292                 max_per_txd = min(mss << 2, max_per_txd);
3293                 max_txd_pwr = fls(max_per_txd) - 1;
3294
3295                 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
3296                 * points to just header, pull a few bytes of payload from
3297                 * frags into skb->data */
3298                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3299                 if (skb->data_len && (hdr_len == (skb->len - skb->data_len))) {
3300                         switch (adapter->hw.mac_type) {
3301                                 unsigned int pull_size;
3302                         case e1000_82544:
3303                                 /* Make sure we have room to chop off 4 bytes,
3304                                  * and that the end alignment will work out to
3305                                  * this hardware's requirements
3306                                  * NOTE: this is a TSO only workaround
3307                                  * if end byte alignment not correct move us
3308                                  * into the next dword */
3309                                 if ((unsigned long)(skb_tail_pointer(skb) - 1) & 4)
3310                                         break;
3311                                 /* fall through */
3312                         case e1000_82571:
3313                         case e1000_82572:
3314                         case e1000_82573:
3315                         case e1000_ich8lan:
3316                                 pull_size = min((unsigned int)4, skb->data_len);
3317                                 if (!__pskb_pull_tail(skb, pull_size)) {
3318                                         DPRINTK(DRV, ERR,
3319                                                 "__pskb_pull_tail failed.\n");
3320                                         dev_kfree_skb_any(skb);
3321                                         return NETDEV_TX_OK;
3322                                 }
3323                                 len = skb->len - skb->data_len;
3324                                 break;
3325                         default:
3326                                 /* do nothing */
3327                                 break;
3328                         }
3329                 }
3330         }
3331
3332         /* reserve a descriptor for the offload context */
3333         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3334                 count++;
3335         count++;
3336
3337         /* Controller Erratum workaround */
3338         if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3339                 count++;
3340
3341         count += TXD_USE_COUNT(len, max_txd_pwr);
3342
3343         if (adapter->pcix_82544)
3344                 count++;
3345
3346         /* work-around for errata 10 and it applies to all controllers
3347          * in PCI-X mode, so add one more descriptor to the count
3348          */
3349         if (unlikely((adapter->hw.bus_type == e1000_bus_type_pcix) &&
3350                         (len > 2015)))
3351                 count++;
3352
3353         nr_frags = skb_shinfo(skb)->nr_frags;
3354         for (f = 0; f < nr_frags; f++)
3355                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3356                                        max_txd_pwr);
3357         if (adapter->pcix_82544)
3358                 count += nr_frags;
3359
3360
3361         if (adapter->hw.tx_pkt_filtering &&
3362             (adapter->hw.mac_type == e1000_82573))
3363                 e1000_transfer_dhcp_info(adapter, skb);
3364
3365         if (!spin_trylock_irqsave(&tx_ring->tx_lock, flags))
3366                 /* Collision - tell upper layer to requeue */
3367                 return NETDEV_TX_LOCKED;
3368
3369         /* need: count + 2 desc gap to keep tail from touching
3370          * head, otherwise try next time */
3371         if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2))) {
3372                 spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3373                 return NETDEV_TX_BUSY;
3374         }
3375
3376         if (unlikely(adapter->hw.mac_type == e1000_82547)) {
3377                 if (unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
3378                         netif_stop_queue(netdev);
3379                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
3380                         spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3381                         return NETDEV_TX_BUSY;
3382                 }
3383         }
3384
3385         if (unlikely(adapter->vlgrp && vlan_tx_tag_present(skb))) {
3386                 tx_flags |= E1000_TX_FLAGS_VLAN;
3387                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3388         }
3389
3390         first = tx_ring->next_to_use;
3391
3392         tso = e1000_tso(adapter, tx_ring, skb);
3393         if (tso < 0) {
3394                 dev_kfree_skb_any(skb);
3395                 spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3396                 return NETDEV_TX_OK;
3397         }
3398
3399         if (likely(tso)) {
3400                 tx_ring->last_tx_tso = 1;
3401                 tx_flags |= E1000_TX_FLAGS_TSO;
3402         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3403                 tx_flags |= E1000_TX_FLAGS_CSUM;
3404
3405         /* Old method was to assume IPv4 packet by default if TSO was enabled.
3406          * 82571 hardware supports TSO capabilities for IPv6 as well...
3407          * no longer assume, we must. */
3408         if (likely(skb->protocol == htons(ETH_P_IP)))
3409                 tx_flags |= E1000_TX_FLAGS_IPV4;
3410
3411         e1000_tx_queue(adapter, tx_ring, tx_flags,
3412                        e1000_tx_map(adapter, tx_ring, skb, first,
3413                                     max_per_txd, nr_frags, mss));
3414
3415         netdev->trans_start = jiffies;
3416
3417         /* Make sure there is space in the ring for the next send. */
3418         e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3419
3420         spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3421         return NETDEV_TX_OK;
3422 }
3423
3424 /**
3425  * e1000_tx_timeout - Respond to a Tx Hang
3426  * @netdev: network interface device structure
3427  **/
3428
3429 static void
3430 e1000_tx_timeout(struct net_device *netdev)
3431 {
3432         struct e1000_adapter *adapter = netdev_priv(netdev);
3433
3434         /* Do the reset outside of interrupt context */
3435         adapter->tx_timeout_count++;
3436         schedule_work(&adapter->reset_task);
3437 }
3438
3439 static void
3440 e1000_reset_task(struct work_struct *work)
3441 {
3442         struct e1000_adapter *adapter =
3443                 container_of(work, struct e1000_adapter, reset_task);
3444
3445         e1000_reinit_locked(adapter);
3446 }
3447
3448 /**
3449  * e1000_get_stats - Get System Network Statistics
3450  * @netdev: network interface device structure
3451  *
3452  * Returns the address of the device statistics structure.
3453  * The statistics are actually updated from the timer callback.
3454  **/
3455
3456 static struct net_device_stats *
3457 e1000_get_stats(struct net_device *netdev)
3458 {
3459         struct e1000_adapter *adapter = netdev_priv(netdev);
3460
3461         /* only return the current stats */
3462         return &adapter->net_stats;
3463 }
3464
3465 /**
3466  * e1000_change_mtu - Change the Maximum Transfer Unit
3467  * @netdev: network interface device structure
3468  * @new_mtu: new value for maximum frame size
3469  *
3470  * Returns 0 on success, negative on failure
3471  **/
3472
3473 static int
3474 e1000_change_mtu(struct net_device *netdev, int new_mtu)
3475 {
3476         struct e1000_adapter *adapter = netdev_priv(netdev);
3477         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3478         uint16_t eeprom_data = 0;
3479
3480         if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3481             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3482                 DPRINTK(PROBE, ERR, "Invalid MTU setting\n");
3483                 return -EINVAL;
3484         }
3485
3486         /* Adapter-specific max frame size limits. */
3487         switch (adapter->hw.mac_type) {
3488         case e1000_undefined ... e1000_82542_rev2_1:
3489         case e1000_ich8lan:
3490                 if (max_frame > MAXIMUM_ETHERNET_FRAME_SIZE) {
3491                         DPRINTK(PROBE, ERR, "Jumbo Frames not supported.\n");
3492                         return -EINVAL;
3493                 }
3494                 break;
3495         case e1000_82573:
3496                 /* Jumbo Frames not supported if:
3497                  * - this is not an 82573L device
3498                  * - ASPM is enabled in any way (0x1A bits 3:2) */
3499                 e1000_read_eeprom(&adapter->hw, EEPROM_INIT_3GIO_3, 1,
3500                                   &eeprom_data);
3501                 if ((adapter->hw.device_id != E1000_DEV_ID_82573L) ||
3502                     (eeprom_data & EEPROM_WORD1A_ASPM_MASK)) {
3503                         if (max_frame > MAXIMUM_ETHERNET_FRAME_SIZE) {
3504                                 DPRINTK(PROBE, ERR,
3505                                         "Jumbo Frames not supported.\n");
3506                                 return -EINVAL;
3507                         }
3508                         break;
3509                 }
3510                 /* ERT will be enabled later to enable wire speed receives */
3511
3512                 /* fall through to get support */
3513         case e1000_82571:
3514         case e1000_82572:
3515         case e1000_80003es2lan:
3516 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3517                 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3518                         DPRINTK(PROBE, ERR, "MTU > 9216 not supported.\n");
3519                         return -EINVAL;
3520                 }
3521                 break;
3522         default:
3523                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3524                 break;
3525         }
3526
3527         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3528          * means we reserve 2 more, this pushes us to allocate from the next
3529          * larger slab size
3530          * i.e. RXBUFFER_2048 --> size-4096 slab */
3531
3532         if (max_frame <= E1000_RXBUFFER_256)
3533                 adapter->rx_buffer_len = E1000_RXBUFFER_256;
3534         else if (max_frame <= E1000_RXBUFFER_512)
3535                 adapter->rx_buffer_len = E1000_RXBUFFER_512;
3536         else if (max_frame <= E1000_RXBUFFER_1024)
3537                 adapter->rx_buffer_len = E1000_RXBUFFER_1024;
3538         else if (max_frame <= E1000_RXBUFFER_2048)
3539                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3540         else if (max_frame <= E1000_RXBUFFER_4096)
3541                 adapter->rx_buffer_len = E1000_RXBUFFER_4096;
3542         else if (max_frame <= E1000_RXBUFFER_8192)
3543                 adapter->rx_buffer_len = E1000_RXBUFFER_8192;
3544         else if (max_frame <= E1000_RXBUFFER_16384)
3545                 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3546
3547         /* adjust allocation if LPE protects us, and we aren't using SBP */
3548         if (!adapter->hw.tbi_compatibility_on &&
3549             ((max_frame == MAXIMUM_ETHERNET_FRAME_SIZE) ||
3550              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3551                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3552
3553         netdev->mtu = new_mtu;
3554         adapter->hw.max_frame_size = max_frame;
3555
3556         if (netif_running(netdev))
3557                 e1000_reinit_locked(adapter);
3558
3559         return 0;
3560 }
3561
3562 /**
3563  * e1000_update_stats - Update the board statistics counters
3564  * @adapter: board private structure
3565  **/
3566
3567 void
3568 e1000_update_stats(struct e1000_adapter *adapter)
3569 {
3570         struct e1000_hw *hw = &adapter->hw;
3571         struct pci_dev *pdev = adapter->pdev;
3572         unsigned long flags;
3573         uint16_t phy_tmp;
3574
3575 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3576
3577         /*
3578          * Prevent stats update while adapter is being reset, or if the pci
3579          * connection is down.
3580          */
3581         if (adapter->link_speed == 0)
3582                 return;
3583         if (pci_channel_offline(pdev))
3584                 return;
3585
3586         spin_lock_irqsave(&adapter->stats_lock, flags);
3587
3588         /* these counters are modified from e1000_adjust_tbi_stats,
3589          * called from the interrupt context, so they must only
3590          * be written while holding adapter->stats_lock
3591          */
3592
3593         adapter->stats.crcerrs += E1000_READ_REG(hw, CRCERRS);
3594         adapter->stats.gprc += E1000_READ_REG(hw, GPRC);
3595         adapter->stats.gorcl += E1000_READ_REG(hw, GORCL);
3596         adapter->stats.gorch += E1000_READ_REG(hw, GORCH);
3597         adapter->stats.bprc += E1000_READ_REG(hw, BPRC);
3598         adapter->stats.mprc += E1000_READ_REG(hw, MPRC);
3599         adapter->stats.roc += E1000_READ_REG(hw, ROC);
3600
3601         if (adapter->hw.mac_type != e1000_ich8lan) {
3602                 adapter->stats.prc64 += E1000_READ_REG(hw, PRC64);
3603                 adapter->stats.prc127 += E1000_READ_REG(hw, PRC127);
3604                 adapter->stats.prc255 += E1000_READ_REG(hw, PRC255);
3605                 adapter->stats.prc511 += E1000_READ_REG(hw, PRC511);
3606                 adapter->stats.prc1023 += E1000_READ_REG(hw, PRC1023);
3607                 adapter->stats.prc1522 += E1000_READ_REG(hw, PRC1522);
3608         }
3609
3610         adapter->stats.symerrs += E1000_READ_REG(hw, SYMERRS);
3611         adapter->stats.mpc += E1000_READ_REG(hw, MPC);
3612         adapter->stats.scc += E1000_READ_REG(hw, SCC);
3613         adapter->stats.ecol += E1000_READ_REG(hw, ECOL);
3614         adapter->stats.mcc += E1000_READ_REG(hw, MCC);
3615         adapter->stats.latecol += E1000_READ_REG(hw, LATECOL);
3616         adapter->stats.dc += E1000_READ_REG(hw, DC);
3617         adapter->stats.sec += E1000_READ_REG(hw, SEC);
3618         adapter->stats.rlec += E1000_READ_REG(hw, RLEC);
3619         adapter->stats.xonrxc += E1000_READ_REG(hw, XONRXC);
3620         adapter->stats.xontxc += E1000_READ_REG(hw, XONTXC);
3621         adapter->stats.xoffrxc += E1000_READ_REG(hw, XOFFRXC);
3622         adapter->stats.xofftxc += E1000_READ_REG(hw, XOFFTXC);
3623         adapter->stats.fcruc += E1000_READ_REG(hw, FCRUC);
3624         adapter->stats.gptc += E1000_READ_REG(hw, GPTC);
3625         adapter->stats.gotcl += E1000_READ_REG(hw, GOTCL);
3626         adapter->stats.gotch += E1000_READ_REG(hw, GOTCH);
3627         adapter->stats.rnbc += E1000_READ_REG(hw, RNBC);
3628         adapter->stats.ruc += E1000_READ_REG(hw, RUC);
3629         adapter->stats.rfc += E1000_READ_REG(hw, RFC);
3630         adapter->stats.rjc += E1000_READ_REG(hw, RJC);
3631         adapter->stats.torl += E1000_READ_REG(hw, TORL);
3632         adapter->stats.torh += E1000_READ_REG(hw, TORH);
3633         adapter->stats.totl += E1000_READ_REG(hw, TOTL);
3634         adapter->stats.toth += E1000_READ_REG(hw, TOTH);
3635         adapter->stats.tpr += E1000_READ_REG(hw, TPR);
3636
3637         if (adapter->hw.mac_type != e1000_ich8lan) {
3638                 adapter->stats.ptc64 += E1000_READ_REG(hw, PTC64);
3639                 adapter->stats.ptc127 += E1000_READ_REG(hw, PTC127);
3640                 adapter->stats.ptc255 += E1000_READ_REG(hw, PTC255);
3641                 adapter->stats.ptc511 += E1000_READ_REG(hw, PTC511);
3642                 adapter->stats.ptc1023 += E1000_READ_REG(hw, PTC1023);
3643                 adapter->stats.ptc1522 += E1000_READ_REG(hw, PTC1522);
3644         }
3645
3646         adapter->stats.mptc += E1000_READ_REG(hw, MPTC);
3647         adapter->stats.bptc += E1000_READ_REG(hw, BPTC);
3648
3649         /* used for adaptive IFS */
3650
3651         hw->tx_packet_delta = E1000_READ_REG(hw, TPT);
3652         adapter->stats.tpt += hw->tx_packet_delta;
3653         hw->collision_delta = E1000_READ_REG(hw, COLC);
3654         adapter->stats.colc += hw->collision_delta;
3655
3656         if (hw->mac_type >= e1000_82543) {
3657                 adapter->stats.algnerrc += E1000_READ_REG(hw, ALGNERRC);
3658                 adapter->stats.rxerrc += E1000_READ_REG(hw, RXERRC);
3659                 adapter->stats.tncrs += E1000_READ_REG(hw, TNCRS);
3660                 adapter->stats.cexterr += E1000_READ_REG(hw, CEXTERR);
3661                 adapter->stats.tsctc += E1000_READ_REG(hw, TSCTC);
3662                 adapter->stats.tsctfc += E1000_READ_REG(hw, TSCTFC);
3663         }
3664         if (hw->mac_type > e1000_82547_rev_2) {
3665                 adapter->stats.iac += E1000_READ_REG(hw, IAC);
3666                 adapter->stats.icrxoc += E1000_READ_REG(hw, ICRXOC);
3667
3668                 if (adapter->hw.mac_type != e1000_ich8lan) {
3669                         adapter->stats.icrxptc += E1000_READ_REG(hw, ICRXPTC);
3670                         adapter->stats.icrxatc += E1000_READ_REG(hw, ICRXATC);
3671                         adapter->stats.ictxptc += E1000_READ_REG(hw, ICTXPTC);
3672                         adapter->stats.ictxatc += E1000_READ_REG(hw, ICTXATC);
3673                         adapter->stats.ictxqec += E1000_READ_REG(hw, ICTXQEC);
3674                         adapter->stats.ictxqmtc += E1000_READ_REG(hw, ICTXQMTC);
3675                         adapter->stats.icrxdmtc += E1000_READ_REG(hw, ICRXDMTC);
3676                 }
3677         }
3678
3679         /* Fill out the OS statistics structure */
3680         adapter->net_stats.rx_packets = adapter->stats.gprc;
3681         adapter->net_stats.tx_packets = adapter->stats.gptc;
3682         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
3683         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
3684         adapter->net_stats.multicast = adapter->stats.mprc;
3685         adapter->net_stats.collisions = adapter->stats.colc;
3686
3687         /* Rx Errors */
3688
3689         /* RLEC on some newer hardware can be incorrect so build
3690         * our own version based on RUC and ROC */
3691         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3692                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3693                 adapter->stats.ruc + adapter->stats.roc +
3694                 adapter->stats.cexterr;
3695         adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3696         adapter->net_stats.rx_length_errors = adapter->stats.rlerrc;
3697         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3698         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3699         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3700
3701         /* Tx Errors */
3702         adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3703         adapter->net_stats.tx_errors = adapter->stats.txerrc;
3704         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3705         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3706         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3707         if (adapter->hw.bad_tx_carr_stats_fd &&
3708             adapter->link_duplex == FULL_DUPLEX) {
3709                 adapter->net_stats.tx_carrier_errors = 0;
3710                 adapter->stats.tncrs = 0;
3711         }
3712
3713         /* Tx Dropped needs to be maintained elsewhere */
3714
3715         /* Phy Stats */
3716         if (hw->media_type == e1000_media_type_copper) {
3717                 if ((adapter->link_speed == SPEED_1000) &&
3718                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3719                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3720                         adapter->phy_stats.idle_errors += phy_tmp;
3721                 }
3722
3723                 if ((hw->mac_type <= e1000_82546) &&
3724                    (hw->phy_type == e1000_phy_m88) &&
3725                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3726                         adapter->phy_stats.receive_errors += phy_tmp;
3727         }
3728
3729         /* Management Stats */
3730         if (adapter->hw.has_smbus) {
3731                 adapter->stats.mgptc += E1000_READ_REG(hw, MGTPTC);
3732                 adapter->stats.mgprc += E1000_READ_REG(hw, MGTPRC);
3733                 adapter->stats.mgpdc += E1000_READ_REG(hw, MGTPDC);
3734         }
3735
3736         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3737 }
3738
3739 /**
3740  * e1000_intr_msi - Interrupt Handler
3741  * @irq: interrupt number
3742  * @data: pointer to a network interface device structure
3743  **/
3744
3745 static irqreturn_t
3746 e1000_intr_msi(int irq, void *data)
3747 {
3748         struct net_device *netdev = data;
3749         struct e1000_adapter *adapter = netdev_priv(netdev);
3750         struct e1000_hw *hw = &adapter->hw;
3751 #ifndef CONFIG_E1000_NAPI
3752         int i;
3753 #endif
3754         uint32_t icr = E1000_READ_REG(hw, ICR);
3755
3756 #ifdef CONFIG_E1000_NAPI
3757         /* read ICR disables interrupts using IAM, so keep up with our
3758          * enable/disable accounting */
3759         atomic_inc(&adapter->irq_sem);
3760 #endif
3761         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3762                 hw->get_link_status = 1;
3763                 /* 80003ES2LAN workaround-- For packet buffer work-around on
3764                  * link down event; disable receives here in the ISR and reset
3765                  * adapter in watchdog */
3766                 if (netif_carrier_ok(netdev) &&
3767                     (adapter->hw.mac_type == e1000_80003es2lan)) {
3768                         /* disable receives */
3769                         uint32_t rctl = E1000_READ_REG(hw, RCTL);
3770                         E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
3771                 }
3772                 /* guard against interrupt when we're going down */
3773                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3774                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3775         }
3776
3777 #ifdef CONFIG_E1000_NAPI
3778         if (likely(netif_rx_schedule_prep(netdev))) {
3779                 adapter->total_tx_bytes = 0;
3780                 adapter->total_tx_packets = 0;
3781                 adapter->total_rx_bytes = 0;
3782                 adapter->total_rx_packets = 0;
3783                 __netif_rx_schedule(netdev);
3784         } else
3785                 e1000_irq_enable(adapter);
3786 #else
3787         adapter->total_tx_bytes = 0;
3788         adapter->total_rx_bytes = 0;
3789         adapter->total_tx_packets = 0;
3790         adapter->total_rx_packets = 0;
3791
3792         for (i = 0; i < E1000_MAX_INTR; i++)
3793                 if (unlikely(!adapter->clean_rx(adapter, adapter->rx_ring) &
3794                    !e1000_clean_tx_irq(adapter, adapter->tx_ring)))
3795                         break;
3796
3797         if (likely(adapter->itr_setting & 3))
3798                 e1000_set_itr(adapter);
3799 #endif
3800
3801         return IRQ_HANDLED;
3802 }
3803
3804 /**
3805  * e1000_intr - Interrupt Handler
3806  * @irq: interrupt number
3807  * @data: pointer to a network interface device structure
3808  **/
3809
3810 static irqreturn_t
3811 e1000_intr(int irq, void *data)
3812 {
3813         struct net_device *netdev = data;
3814         struct e1000_adapter *adapter = netdev_priv(netdev);
3815         struct e1000_hw *hw = &adapter->hw;
3816         uint32_t rctl, icr = E1000_READ_REG(hw, ICR);
3817 #ifndef CONFIG_E1000_NAPI
3818         int i;
3819 #endif
3820         if (unlikely(!icr))
3821                 return IRQ_NONE;  /* Not our interrupt */
3822
3823 #ifdef CONFIG_E1000_NAPI
3824         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
3825          * not set, then the adapter didn't send an interrupt */
3826         if (unlikely(hw->mac_type >= e1000_82571 &&
3827                      !(icr & E1000_ICR_INT_ASSERTED)))
3828                 return IRQ_NONE;
3829
3830         /* Interrupt Auto-Mask...upon reading ICR,
3831          * interrupts are masked.  No need for the
3832          * IMC write, but it does mean we should
3833          * account for it ASAP. */
3834         if (likely(hw->mac_type >= e1000_82571))
3835                 atomic_inc(&adapter->irq_sem);
3836 #endif
3837
3838         if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3839                 hw->get_link_status = 1;
3840                 /* 80003ES2LAN workaround--
3841                  * For packet buffer work-around on link down event;
3842                  * disable receives here in the ISR and
3843                  * reset adapter in watchdog
3844                  */
3845                 if (netif_carrier_ok(netdev) &&
3846                     (adapter->hw.mac_type == e1000_80003es2lan)) {
3847                         /* disable receives */
3848                         rctl = E1000_READ_REG(hw, RCTL);
3849                         E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
3850                 }
3851                 /* guard against interrupt when we're going down */
3852                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3853                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3854         }
3855
3856 #ifdef CONFIG_E1000_NAPI
3857         if (unlikely(hw->mac_type < e1000_82571)) {
3858                 /* disable interrupts, without the synchronize_irq bit */
3859                 atomic_inc(&adapter->irq_sem);
3860                 E1000_WRITE_REG(hw, IMC, ~0);
3861                 E1000_WRITE_FLUSH(hw);
3862         }
3863         if (likely(netif_rx_schedule_prep(netdev))) {
3864                 adapter->total_tx_bytes = 0;
3865                 adapter->total_tx_packets = 0;
3866                 adapter->total_rx_bytes = 0;
3867                 adapter->total_rx_packets = 0;
3868                 __netif_rx_schedule(netdev);
3869         } else
3870                 /* this really should not happen! if it does it is basically a
3871                  * bug, but not a hard error, so enable ints and continue */
3872                 e1000_irq_enable(adapter);
3873 #else
3874         /* Writing IMC and IMS is needed for 82547.
3875          * Due to Hub Link bus being occupied, an interrupt
3876          * de-assertion message is not able to be sent.
3877          * When an interrupt assertion message is generated later,
3878          * two messages are re-ordered and sent out.
3879          * That causes APIC to think 82547 is in de-assertion
3880          * state, while 82547 is in assertion state, resulting
3881          * in dead lock. Writing IMC forces 82547 into
3882          * de-assertion state.
3883          */
3884         if (hw->mac_type == e1000_82547 || hw->mac_type == e1000_82547_rev_2) {
3885                 atomic_inc(&adapter->irq_sem);
3886                 E1000_WRITE_REG(hw, IMC, ~0);
3887         }
3888
3889         adapter->total_tx_bytes = 0;
3890         adapter->total_rx_bytes = 0;
3891         adapter->total_tx_packets = 0;
3892         adapter->total_rx_packets = 0;
3893
3894         for (i = 0; i < E1000_MAX_INTR; i++)
3895                 if (unlikely(!adapter->clean_rx(adapter, adapter->rx_ring) &
3896                    !e1000_clean_tx_irq(adapter, adapter->tx_ring)))
3897                         break;
3898
3899         if (likely(adapter->itr_setting & 3))
3900                 e1000_set_itr(adapter);
3901
3902         if (hw->mac_type == e1000_82547 || hw->mac_type == e1000_82547_rev_2)
3903                 e1000_irq_enable(adapter);
3904
3905 #endif
3906         return IRQ_HANDLED;
3907 }
3908
3909 #ifdef CONFIG_E1000_NAPI
3910 /**
3911  * e1000_clean - NAPI Rx polling callback
3912  * @adapter: board private structure
3913  **/
3914
3915 static int
3916 e1000_clean(struct net_device *poll_dev, int *budget)
3917 {
3918         struct e1000_adapter *adapter;
3919         int work_to_do = min(*budget, poll_dev->quota);
3920         int tx_cleaned = 0, work_done = 0;
3921
3922         /* Must NOT use netdev_priv macro here. */
3923         adapter = poll_dev->priv;
3924
3925         /* Keep link state information with original netdev */
3926         if (!netif_carrier_ok(poll_dev))
3927                 goto quit_polling;
3928
3929         /* e1000_clean is called per-cpu.  This lock protects
3930          * tx_ring[0] from being cleaned by multiple cpus
3931          * simultaneously.  A failure obtaining the lock means
3932          * tx_ring[0] is currently being cleaned anyway. */
3933         if (spin_trylock(&adapter->tx_queue_lock)) {
3934                 tx_cleaned = e1000_clean_tx_irq(adapter,
3935                                                 &adapter->tx_ring[0]);
3936                 spin_unlock(&adapter->tx_queue_lock);
3937         }
3938
3939         adapter->clean_rx(adapter, &adapter->rx_ring[0],
3940                           &work_done, work_to_do);
3941
3942         *budget -= work_done;
3943         poll_dev->quota -= work_done;
3944
3945         /* If no Tx and not enough Rx work done, exit the polling mode */
3946         if ((!tx_cleaned && (work_done == 0)) ||
3947            !netif_running(poll_dev)) {
3948 quit_polling:
3949                 if (likely(adapter->itr_setting & 3))
3950                         e1000_set_itr(adapter);
3951                 netif_rx_complete(poll_dev);
3952                 e1000_irq_enable(adapter);
3953                 return 0;
3954         }
3955
3956         return 1;
3957 }
3958
3959 #endif
3960 /**
3961  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3962  * @adapter: board private structure
3963  **/
3964
3965 static boolean_t
3966 e1000_clean_tx_irq(struct e1000_adapter *adapter,
3967                    struct e1000_tx_ring *tx_ring)
3968 {
3969         struct net_device *netdev = adapter->netdev;
3970         struct e1000_tx_desc *tx_desc, *eop_desc;
3971         struct e1000_buffer *buffer_info;
3972         unsigned int i, eop;
3973 #ifdef CONFIG_E1000_NAPI
3974         unsigned int count = 0;
3975 #endif
3976         boolean_t cleaned = FALSE;
3977         unsigned int total_tx_bytes=0, total_tx_packets=0;
3978
3979         i = tx_ring->next_to_clean;
3980         eop = tx_ring->buffer_info[i].next_to_watch;
3981         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3982
3983         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
3984                 for (cleaned = FALSE; !cleaned; ) {
3985                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3986                         buffer_info = &tx_ring->buffer_info[i];
3987                         cleaned = (i == eop);
3988
3989                         if (cleaned) {
3990                                 struct sk_buff *skb = buffer_info->skb;
3991                                 unsigned int segs, bytecount;
3992                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3993                                 /* multiply data chunks by size of headers */
3994                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3995                                             skb->len;
3996                                 total_tx_packets += segs;
3997                                 total_tx_bytes += bytecount;
3998                         }
3999                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
4000                         tx_desc->upper.data = 0;
4001
4002                         if (unlikely(++i == tx_ring->count)) i = 0;
4003                 }
4004
4005                 eop = tx_ring->buffer_info[i].next_to_watch;
4006                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
4007 #ifdef CONFIG_E1000_NAPI
4008 #define E1000_TX_WEIGHT 64
4009                 /* weight of a sort for tx, to avoid endless transmit cleanup */
4010                 if (count++ == E1000_TX_WEIGHT) break;
4011 #endif
4012         }
4013
4014         tx_ring->next_to_clean = i;
4015
4016 #define TX_WAKE_THRESHOLD 32
4017         if (unlikely(cleaned && netif_carrier_ok(netdev) &&
4018                      E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
4019                 /* Make sure that anybody stopping the queue after this
4020                  * sees the new next_to_clean.
4021                  */
4022                 smp_mb();
4023                 if (netif_queue_stopped(netdev)) {
4024                         netif_wake_queue(netdev);
4025                         ++adapter->restart_queue;
4026                 }
4027         }
4028
4029         if (adapter->detect_tx_hung) {
4030                 /* Detect a transmit hang in hardware, this serializes the
4031                  * check with the clearing of time_stamp and movement of i */
4032                 adapter->detect_tx_hung = FALSE;
4033                 if (tx_ring->buffer_info[eop].dma &&
4034                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
4035                                (adapter->tx_timeout_factor * HZ))
4036                     && !(E1000_READ_REG(&adapter->hw, STATUS) &
4037                          E1000_STATUS_TXOFF)) {
4038
4039                         /* detected Tx unit hang */
4040                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
4041                                         "  Tx Queue             <%lu>\n"
4042                                         "  TDH                  <%x>\n"
4043                                         "  TDT                  <%x>\n"
4044                                         "  next_to_use          <%x>\n"
4045                                         "  next_to_clean        <%x>\n"
4046                                         "buffer_info[next_to_clean]\n"
4047                                         "  time_stamp           <%lx>\n"
4048                                         "  next_to_watch        <%x>\n"
4049                                         "  jiffies              <%lx>\n"
4050                                         "  next_to_watch.status <%x>\n",
4051                                 (unsigned long)((tx_ring - adapter->tx_ring) /
4052                                         sizeof(struct e1000_tx_ring)),
4053                                 readl(adapter->hw.hw_addr + tx_ring->tdh),
4054                                 readl(adapter->hw.hw_addr + tx_ring->tdt),
4055                                 tx_ring->next_to_use,
4056                                 tx_ring->next_to_clean,
4057                                 tx_ring->buffer_info[eop].time_stamp,
4058                                 eop,
4059                                 jiffies,
4060                                 eop_desc->upper.fields.status);
4061                         netif_stop_queue(netdev);
4062                 }
4063         }
4064         adapter->total_tx_bytes += total_tx_bytes;
4065         adapter->total_tx_packets += total_tx_packets;
4066         return cleaned;
4067 }
4068
4069 /**
4070  * e1000_rx_checksum - Receive Checksum Offload for 82543
4071  * @adapter:     board private structure
4072  * @status_err:  receive descriptor status and error fields
4073  * @csum:        receive descriptor csum field
4074  * @sk_buff:     socket buffer with received data
4075  **/
4076
4077 static void
4078 e1000_rx_checksum(struct e1000_adapter *adapter,
4079                   uint32_t status_err, uint32_t csum,
4080                   struct sk_buff *skb)
4081 {
4082         uint16_t status = (uint16_t)status_err;
4083         uint8_t errors = (uint8_t)(status_err >> 24);
4084         skb->ip_summed = CHECKSUM_NONE;
4085
4086         /* 82543 or newer only */
4087         if (unlikely(adapter->hw.mac_type < e1000_82543)) return;
4088         /* Ignore Checksum bit is set */
4089         if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
4090         /* TCP/UDP checksum error bit is set */
4091         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
4092                 /* let the stack verify checksum errors */
4093                 adapter->hw_csum_err++;
4094                 return;
4095         }
4096         /* TCP/UDP Checksum has not been calculated */
4097         if (adapter->hw.mac_type <= e1000_82547_rev_2) {
4098                 if (!(status & E1000_RXD_STAT_TCPCS))
4099                         return;
4100         } else {
4101                 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
4102                         return;
4103         }
4104         /* It must be a TCP or UDP packet with a valid checksum */
4105         if (likely(status & E1000_RXD_STAT_TCPCS)) {
4106                 /* TCP checksum is good */
4107                 skb->ip_summed = CHECKSUM_UNNECESSARY;
4108         } else if (adapter->hw.mac_type > e1000_82547_rev_2) {
4109                 /* IP fragment with UDP payload */
4110                 /* Hardware complements the payload checksum, so we undo it
4111                  * and then put the value in host order for further stack use.
4112                  */
4113                 csum = ntohl(csum ^ 0xFFFF);
4114                 skb->csum = csum;
4115                 skb->ip_summed = CHECKSUM_COMPLETE;
4116         }
4117         adapter->hw_csum_good++;
4118 }
4119
4120 /**
4121  * e1000_clean_rx_irq - Send received data up the network stack; legacy
4122  * @adapter: board private structure
4123  **/
4124
4125 static boolean_t
4126 #ifdef CONFIG_E1000_NAPI
4127 e1000_clean_rx_irq(struct e1000_adapter *adapter,
4128                    struct e1000_rx_ring *rx_ring,
4129                    int *work_done, int work_to_do)
4130 #else
4131 e1000_clean_rx_irq(struct e1000_adapter *adapter,
4132                    struct e1000_rx_ring *rx_ring)
4133 #endif
4134 {
4135         struct net_device *netdev = adapter->netdev;
4136         struct pci_dev *pdev = adapter->pdev;
4137         struct e1000_rx_desc *rx_desc, *next_rxd;
4138         struct e1000_buffer *buffer_info, *next_buffer;
4139         unsigned long flags;
4140         uint32_t length;
4141         uint8_t last_byte;
4142         unsigned int i;
4143         int cleaned_count = 0;
4144         boolean_t cleaned = FALSE;
4145         unsigned int total_rx_bytes=0, total_rx_packets=0;
4146
4147         i = rx_ring->next_to_clean;
4148         rx_desc = E1000_RX_DESC(*rx_ring, i);
4149         buffer_info = &rx_ring->buffer_info[i];
4150
4151         while (rx_desc->status & E1000_RXD_STAT_DD) {
4152                 struct sk_buff *skb;
4153                 u8 status;
4154
4155 #ifdef CONFIG_E1000_NAPI
4156                 if (*work_done >= work_to_do)
4157                         break;
4158                 (*work_done)++;
4159 #endif
4160                 status = rx_desc->status;
4161                 skb = buffer_info->skb;
4162                 buffer_info->skb = NULL;
4163
4164                 prefetch(skb->data - NET_IP_ALIGN);
4165
4166                 if (++i == rx_ring->count) i = 0;
4167                 next_rxd = E1000_RX_DESC(*rx_ring, i);
4168                 prefetch(next_rxd);
4169
4170                 next_buffer = &rx_ring->buffer_info[i];
4171
4172                 cleaned = TRUE;
4173                 cleaned_count++;
4174                 pci_unmap_single(pdev,
4175                                  buffer_info->dma,
4176                                  buffer_info->length,
4177                                  PCI_DMA_FROMDEVICE);
4178
4179                 length = le16_to_cpu(rx_desc->length);
4180
4181                 if (unlikely(!(status & E1000_RXD_STAT_EOP))) {
4182                         /* All receives must fit into a single buffer */
4183                         E1000_DBG("%s: Receive packet consumed multiple"
4184                                   " buffers\n", netdev->name);
4185                         /* recycle */
4186                         buffer_info->skb = skb;
4187                         goto next_desc;
4188                 }
4189
4190                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
4191                         last_byte = *(skb->data + length - 1);
4192                         if (TBI_ACCEPT(&adapter->hw, status,
4193                                       rx_desc->errors, length, last_byte)) {
4194                                 spin_lock_irqsave(&adapter->stats_lock, flags);
4195                                 e1000_tbi_adjust_stats(&adapter->hw,
4196                                                        &adapter->stats,
4197                                                        length, skb->data);
4198                                 spin_unlock_irqrestore(&adapter->stats_lock,
4199                                                        flags);
4200                                 length--;
4201                         } else {
4202                                 /* recycle */
4203                                 buffer_info->skb = skb;
4204                                 goto next_desc;
4205                         }
4206                 }
4207
4208                 /* adjust length to remove Ethernet CRC, this must be
4209                  * done after the TBI_ACCEPT workaround above */
4210                 length -= 4;
4211
4212                 /* probably a little skewed due to removing CRC */
4213                 total_rx_bytes += length;
4214                 total_rx_packets++;
4215
4216                 /* code added for copybreak, this should improve
4217                  * performance for small packets with large amounts
4218                  * of reassembly being done in the stack */
4219                 if (length < copybreak) {
4220                         struct sk_buff *new_skb =
4221                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
4222                         if (new_skb) {
4223                                 skb_reserve(new_skb, NET_IP_ALIGN);
4224                                 skb_copy_to_linear_data_offset(new_skb,
4225                                                                -NET_IP_ALIGN,
4226                                                                (skb->data -
4227                                                                 NET_IP_ALIGN),
4228                                                                (length +
4229                                                                 NET_IP_ALIGN));
4230                                 /* save the skb in buffer_info as good */
4231                                 buffer_info->skb = skb;
4232                                 skb = new_skb;
4233                         }
4234                         /* else just continue with the old one */
4235                 }
4236                 /* end copybreak code */
4237                 skb_put(skb, length);
4238
4239                 /* Receive Checksum Offload */
4240                 e1000_rx_checksum(adapter,
4241                                   (uint32_t)(status) |
4242                                   ((uint32_t)(rx_desc->errors) << 24),
4243                                   le16_to_cpu(rx_desc->csum), skb);
4244
4245                 skb->protocol = eth_type_trans(skb, netdev);
4246 #ifdef CONFIG_E1000_NAPI
4247                 if (unlikely(adapter->vlgrp &&
4248                             (status & E1000_RXD_STAT_VP))) {
4249                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
4250                                                  le16_to_cpu(rx_desc->special) &
4251                                                  E1000_RXD_SPC_VLAN_MASK);
4252                 } else {
4253                         netif_receive_skb(skb);
4254                 }
4255 #else /* CONFIG_E1000_NAPI */
4256                 if (unlikely(adapter->vlgrp &&
4257                             (status & E1000_RXD_STAT_VP))) {
4258                         vlan_hwaccel_rx(skb, adapter->vlgrp,
4259                                         le16_to_cpu(rx_desc->special) &
4260                                         E1000_RXD_SPC_VLAN_MASK);
4261                 } else {
4262                         netif_rx(skb);
4263                 }
4264 #endif /* CONFIG_E1000_NAPI */
4265                 netdev->last_rx = jiffies;
4266
4267 next_desc:
4268                 rx_desc->status = 0;
4269
4270                 /* return some buffers to hardware, one at a time is too slow */
4271                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4272                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4273                         cleaned_count = 0;
4274                 }
4275
4276                 /* use prefetched values */
4277                 rx_desc = next_rxd;
4278                 buffer_info = next_buffer;
4279         }
4280         rx_ring->next_to_clean = i;
4281
4282         cleaned_count = E1000_DESC_UNUSED(rx_ring);
4283         if (cleaned_count)
4284                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4285
4286         adapter->total_rx_packets += total_rx_packets;
4287         adapter->total_rx_bytes += total_rx_bytes;
4288         return cleaned;
4289 }
4290
4291 /**
4292  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
4293  * @adapter: board private structure
4294  **/
4295
4296 static boolean_t
4297 #ifdef CONFIG_E1000_NAPI
4298 e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
4299                       struct e1000_rx_ring *rx_ring,
4300                       int *work_done, int work_to_do)
4301 #else
4302 e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
4303                       struct e1000_rx_ring *rx_ring)
4304 #endif
4305 {
4306         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
4307         struct net_device *netdev = adapter->netdev;
4308         struct pci_dev *pdev = adapter->pdev;
4309         struct e1000_buffer *buffer_info, *next_buffer;
4310         struct e1000_ps_page *ps_page;
4311         struct e1000_ps_page_dma *ps_page_dma;
4312         struct sk_buff *skb;
4313         unsigned int i, j;
4314         uint32_t length, staterr;
4315         int cleaned_count = 0;
4316         boolean_t cleaned = FALSE;
4317         unsigned int total_rx_bytes=0, total_rx_packets=0;
4318
4319         i = rx_ring->next_to_clean;
4320         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
4321         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
4322         buffer_info = &rx_ring->buffer_info[i];
4323
4324         while (staterr & E1000_RXD_STAT_DD) {
4325                 ps_page = &rx_ring->ps_page[i];
4326                 ps_page_dma = &rx_ring->ps_page_dma[i];
4327 #ifdef CONFIG_E1000_NAPI
4328                 if (unlikely(*work_done >= work_to_do))
4329                         break;
4330                 (*work_done)++;
4331 #endif
4332                 skb = buffer_info->skb;
4333
4334                 /* in the packet split case this is header only */
4335                 prefetch(skb->data - NET_IP_ALIGN);
4336
4337                 if (++i == rx_ring->count) i = 0;
4338                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
4339                 prefetch(next_rxd);
4340
4341                 next_buffer = &rx_ring->buffer_info[i];
4342
4343                 cleaned = TRUE;
4344                 cleaned_count++;
4345                 pci_unmap_single(pdev, buffer_info->dma,
4346                                  buffer_info->length,
4347                                  PCI_DMA_FROMDEVICE);
4348
4349                 if (unlikely(!(staterr & E1000_RXD_STAT_EOP))) {
4350                         E1000_DBG("%s: Packet Split buffers didn't pick up"
4351                                   " the full packet\n", netdev->name);
4352                         dev_kfree_skb_irq(skb);
4353                         goto next_desc;
4354                 }
4355
4356                 if (unlikely(staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK)) {
4357                         dev_kfree_skb_irq(skb);
4358                         goto next_desc;
4359                 }
4360
4361                 length = le16_to_cpu(rx_desc->wb.middle.length0);
4362
4363                 if (unlikely(!length)) {
4364                         E1000_DBG("%s: Last part of the packet spanning"
4365                                   " multiple descriptors\n", netdev->name);
4366                         dev_kfree_skb_irq(skb);
4367                         goto next_desc;
4368                 }
4369
4370                 /* Good Receive */
4371                 skb_put(skb, length);
4372
4373                 {
4374                 /* this looks ugly, but it seems compiler issues make it
4375                    more efficient than reusing j */
4376                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
4377
4378                 /* page alloc/put takes too long and effects small packet
4379                  * throughput, so unsplit small packets and save the alloc/put*/
4380                 if (l1 && (l1 <= copybreak) && ((length + l1) <= adapter->rx_ps_bsize0)) {
4381                         u8 *vaddr;
4382                         /* there is no documentation about how to call
4383                          * kmap_atomic, so we can't hold the mapping
4384                          * very long */
4385                         pci_dma_sync_single_for_cpu(pdev,
4386                                 ps_page_dma->ps_page_dma[0],
4387                                 PAGE_SIZE,
4388                                 PCI_DMA_FROMDEVICE);
4389                         vaddr = kmap_atomic(ps_page->ps_page[0],
4390                                             KM_SKB_DATA_SOFTIRQ);
4391                         memcpy(skb_tail_pointer(skb), vaddr, l1);
4392                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
4393                         pci_dma_sync_single_for_device(pdev,
4394                                 ps_page_dma->ps_page_dma[0],
4395                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
4396                         /* remove the CRC */
4397                         l1 -= 4;
4398                         skb_put(skb, l1);
4399                         goto copydone;
4400                 } /* if */
4401                 }
4402
4403                 for (j = 0; j < adapter->rx_ps_pages; j++) {
4404                         if (!(length= le16_to_cpu(rx_desc->wb.upper.length[j])))
4405                                 break;
4406                         pci_unmap_page(pdev, ps_page_dma->ps_page_dma[j],
4407                                         PAGE_SIZE, PCI_DMA_FROMDEVICE);
4408                         ps_page_dma->ps_page_dma[j] = 0;
4409                         skb_fill_page_desc(skb, j, ps_page->ps_page[j], 0,
4410                                            length);
4411                         ps_page->ps_page[j] = NULL;
4412                         skb->len += length;
4413                         skb->data_len += length;
4414                         skb->truesize += length;
4415                 }
4416
4417                 /* strip the ethernet crc, problem is we're using pages now so
4418                  * this whole operation can get a little cpu intensive */
4419                 pskb_trim(skb, skb->len - 4);
4420
4421 copydone:
4422                 total_rx_bytes += skb->len;
4423                 total_rx_packets++;
4424
4425                 e1000_rx_checksum(adapter, staterr,
4426                                   le16_to_cpu(rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
4427                 skb->protocol = eth_type_trans(skb, netdev);
4428
4429                 if (likely(rx_desc->wb.upper.header_status &
4430                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP)))
4431                         adapter->rx_hdr_split++;
4432 #ifdef CONFIG_E1000_NAPI
4433                 if (unlikely(adapter->vlgrp && (staterr & E1000_RXD_STAT_VP))) {
4434                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
4435                                 le16_to_cpu(rx_desc->wb.middle.vlan) &
4436                                 E1000_RXD_SPC_VLAN_MASK);
4437                 } else {
4438                         netif_receive_skb(skb);
4439                 }
4440 #else /* CONFIG_E1000_NAPI */
4441                 if (unlikely(adapter->vlgrp && (staterr & E1000_RXD_STAT_VP))) {
4442                         vlan_hwaccel_rx(skb, adapter->vlgrp,
4443                                 le16_to_cpu(rx_desc->wb.middle.vlan) &
4444                                 E1000_RXD_SPC_VLAN_MASK);
4445                 } else {
4446                         netif_rx(skb);
4447                 }
4448 #endif /* CONFIG_E1000_NAPI */
4449                 netdev->last_rx = jiffies;
4450
4451 next_desc:
4452                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
4453                 buffer_info->skb = NULL;
4454
4455                 /* return some buffers to hardware, one at a time is too slow */
4456                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4457                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4458                         cleaned_count = 0;
4459                 }
4460
4461                 /* use prefetched values */
4462                 rx_desc = next_rxd;
4463                 buffer_info = next_buffer;
4464
4465                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
4466         }
4467         rx_ring->next_to_clean = i;
4468
4469         cleaned_count = E1000_DESC_UNUSED(rx_ring);
4470         if (cleaned_count)
4471                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4472
4473         adapter->total_rx_packets += total_rx_packets;
4474         adapter->total_rx_bytes += total_rx_bytes;
4475         return cleaned;
4476 }
4477
4478 /**
4479  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4480  * @adapter: address of board private structure
4481  **/
4482
4483 static void
4484 e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4485                        struct e1000_rx_ring *rx_ring,
4486                        int cleaned_count)
4487 {
4488         struct net_device *netdev = adapter->netdev;
4489         struct pci_dev *pdev = adapter->pdev;
4490         struct e1000_rx_desc *rx_desc;
4491         struct e1000_buffer *buffer_info;
4492         struct sk_buff *skb;
4493         unsigned int i;
4494         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
4495
4496         i = rx_ring->next_to_use;
4497         buffer_info = &rx_ring->buffer_info[i];
4498
4499         while (cleaned_count--) {
4500                 skb = buffer_info->skb;
4501                 if (skb) {
4502                         skb_trim(skb, 0);
4503                         goto map_skb;
4504                 }
4505
4506                 skb = netdev_alloc_skb(netdev, bufsz);
4507                 if (unlikely(!skb)) {
4508                         /* Better luck next round */
4509                         adapter->alloc_rx_buff_failed++;
4510                         break;
4511                 }
4512
4513                 /* Fix for errata 23, can't cross 64kB boundary */
4514                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4515                         struct sk_buff *oldskb = skb;
4516                         DPRINTK(RX_ERR, ERR, "skb align check failed: %u bytes "
4517                                              "at %p\n", bufsz, skb->data);
4518                         /* Try again, without freeing the previous */
4519                         skb = netdev_alloc_skb(netdev, bufsz);
4520                         /* Failed allocation, critical failure */
4521                         if (!skb) {
4522                                 dev_kfree_skb(oldskb);
4523                                 break;
4524                         }
4525
4526                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4527                                 /* give up */
4528                                 dev_kfree_skb(skb);
4529                                 dev_kfree_skb(oldskb);
4530                                 break; /* while !buffer_info->skb */
4531                         }
4532
4533                         /* Use new allocation */
4534                         dev_kfree_skb(oldskb);
4535                 }
4536                 /* Make buffer alignment 2 beyond a 16 byte boundary
4537                  * this will result in a 16 byte aligned IP header after
4538                  * the 14 byte MAC header is removed
4539                  */
4540                 skb_reserve(skb, NET_IP_ALIGN);
4541
4542                 buffer_info->skb = skb;
4543                 buffer_info->length = adapter->rx_buffer_len;
4544 map_skb:
4545                 buffer_info->dma = pci_map_single(pdev,
4546                                                   skb->data,
4547                                                   adapter->rx_buffer_len,
4548                                                   PCI_DMA_FROMDEVICE);
4549
4550                 /* Fix for errata 23, can't cross 64kB boundary */
4551                 if (!e1000_check_64k_bound(adapter,
4552                                         (void *)(unsigned long)buffer_info->dma,
4553                                         adapter->rx_buffer_len)) {
4554                         DPRINTK(RX_ERR, ERR,
4555                                 "dma align check failed: %u bytes at %p\n",
4556                                 adapter->rx_buffer_len,
4557                                 (void *)(unsigned long)buffer_info->dma);
4558                         dev_kfree_skb(skb);
4559                         buffer_info->skb = NULL;
4560
4561                         pci_unmap_single(pdev, buffer_info->dma,
4562                                          adapter->rx_buffer_len,
4563                                          PCI_DMA_FROMDEVICE);
4564
4565                         break; /* while !buffer_info->skb */
4566                 }
4567                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4568                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4569
4570                 if (unlikely(++i == rx_ring->count))
4571                         i = 0;
4572                 buffer_info = &rx_ring->buffer_info[i];
4573         }
4574
4575         if (likely(rx_ring->next_to_use != i)) {
4576                 rx_ring->next_to_use = i;
4577                 if (unlikely(i-- == 0))
4578                         i = (rx_ring->count - 1);
4579
4580                 /* Force memory writes to complete before letting h/w
4581                  * know there are new descriptors to fetch.  (Only
4582                  * applicable for weak-ordered memory model archs,
4583                  * such as IA-64). */
4584                 wmb();
4585                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4586         }
4587 }
4588
4589 /**
4590  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
4591  * @adapter: address of board private structure
4592  **/
4593
4594 static void
4595 e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
4596                           struct e1000_rx_ring *rx_ring,
4597                           int cleaned_count)
4598 {
4599         struct net_device *netdev = adapter->netdev;
4600         struct pci_dev *pdev = adapter->pdev;
4601         union e1000_rx_desc_packet_split *rx_desc;
4602         struct e1000_buffer *buffer_info;
4603         struct e1000_ps_page *ps_page;
4604         struct e1000_ps_page_dma *ps_page_dma;
4605         struct sk_buff *skb;
4606         unsigned int i, j;
4607
4608         i = rx_ring->next_to_use;
4609         buffer_info = &rx_ring->buffer_info[i];
4610         ps_page = &rx_ring->ps_page[i];
4611         ps_page_dma = &rx_ring->ps_page_dma[i];
4612
4613         while (cleaned_count--) {
4614                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
4615
4616                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
4617                         if (j < adapter->rx_ps_pages) {
4618                                 if (likely(!ps_page->ps_page[j])) {
4619                                         ps_page->ps_page[j] =
4620                                                 alloc_page(GFP_ATOMIC);
4621                                         if (unlikely(!ps_page->ps_page[j])) {
4622                                                 adapter->alloc_rx_buff_failed++;
4623                                                 goto no_buffers;
4624                                         }
4625                                         ps_page_dma->ps_page_dma[j] =
4626                                                 pci_map_page(pdev,
4627                                                             ps_page->ps_page[j],
4628                                                             0, PAGE_SIZE,
4629                                                             PCI_DMA_FROMDEVICE);
4630                                 }
4631                                 /* Refresh the desc even if buffer_addrs didn't
4632                                  * change because each write-back erases
4633                                  * this info.
4634                                  */
4635                                 rx_desc->read.buffer_addr[j+1] =
4636                                      cpu_to_le64(ps_page_dma->ps_page_dma[j]);
4637                         } else
4638                                 rx_desc->read.buffer_addr[j+1] = ~0;
4639                 }
4640
4641                 skb = netdev_alloc_skb(netdev,
4642                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
4643
4644                 if (unlikely(!skb)) {
4645                         adapter->alloc_rx_buff_failed++;
4646                         break;
4647                 }
4648
4649                 /* Make buffer alignment 2 beyond a 16 byte boundary
4650                  * this will result in a 16 byte aligned IP header after
4651                  * the 14 byte MAC header is removed
4652                  */
4653                 skb_reserve(skb, NET_IP_ALIGN);
4654
4655                 buffer_info->skb = skb;
4656                 buffer_info->length = adapter->rx_ps_bsize0;
4657                 buffer_info->dma = pci_map_single(pdev, skb->data,
4658                                                   adapter->rx_ps_bsize0,
4659                                                   PCI_DMA_FROMDEVICE);
4660
4661                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
4662
4663                 if (unlikely(++i == rx_ring->count)) i = 0;
4664                 buffer_info = &rx_ring->buffer_info[i];
4665                 ps_page = &rx_ring->ps_page[i];
4666                 ps_page_dma = &rx_ring->ps_page_dma[i];
4667         }
4668
4669 no_buffers:
4670         if (likely(rx_ring->next_to_use != i)) {
4671                 rx_ring->next_to_use = i;
4672                 if (unlikely(i-- == 0)) i = (rx_ring->count - 1);
4673
4674                 /* Force memory writes to complete before letting h/w
4675                  * know there are new descriptors to fetch.  (Only
4676                  * applicable for weak-ordered memory model archs,
4677                  * such as IA-64). */
4678                 wmb();
4679                 /* Hardware increments by 16 bytes, but packet split
4680                  * descriptors are 32 bytes...so we increment tail
4681                  * twice as much.
4682                  */
4683                 writel(i<<1, adapter->hw.hw_addr + rx_ring->rdt);
4684         }
4685 }
4686
4687 /**
4688  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4689  * @adapter:
4690  **/
4691
4692 static void
4693 e1000_smartspeed(struct e1000_adapter *adapter)
4694 {
4695         uint16_t phy_status;
4696         uint16_t phy_ctrl;
4697
4698         if ((adapter->hw.phy_type != e1000_phy_igp) || !adapter->hw.autoneg ||
4699            !(adapter->hw.autoneg_advertised & ADVERTISE_1000_FULL))
4700                 return;
4701
4702         if (adapter->smartspeed == 0) {
4703                 /* If Master/Slave config fault is asserted twice,
4704                  * we assume back-to-back */
4705                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_status);
4706                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4707                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_status);
4708                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4709                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_ctrl);
4710                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4711                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4712                         e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL,
4713                                             phy_ctrl);
4714                         adapter->smartspeed++;
4715                         if (!e1000_phy_setup_autoneg(&adapter->hw) &&
4716                            !e1000_read_phy_reg(&adapter->hw, PHY_CTRL,
4717                                                &phy_ctrl)) {
4718                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4719                                              MII_CR_RESTART_AUTO_NEG);
4720                                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL,
4721                                                     phy_ctrl);
4722                         }
4723                 }
4724                 return;
4725         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4726                 /* If still no link, perhaps using 2/3 pair cable */
4727                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_ctrl);
4728                 phy_ctrl |= CR_1000T_MS_ENABLE;
4729                 e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL, phy_ctrl);
4730                 if (!e1000_phy_setup_autoneg(&adapter->hw) &&
4731                    !e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &phy_ctrl)) {
4732                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4733                                      MII_CR_RESTART_AUTO_NEG);
4734                         e1000_write_phy_reg(&adapter->hw, PHY_CTRL, phy_ctrl);
4735                 }
4736         }
4737         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4738         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4739                 adapter->smartspeed = 0;
4740 }
4741
4742 /**
4743  * e1000_ioctl -
4744  * @netdev:
4745  * @ifreq:
4746  * @cmd:
4747  **/
4748
4749 static int
4750 e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4751 {
4752         switch (cmd) {
4753         case SIOCGMIIPHY:
4754         case SIOCGMIIREG:
4755         case SIOCSMIIREG:
4756                 return e1000_mii_ioctl(netdev, ifr, cmd);
4757         default:
4758                 return -EOPNOTSUPP;
4759         }
4760 }
4761
4762 /**
4763  * e1000_mii_ioctl -
4764  * @netdev:
4765  * @ifreq:
4766  * @cmd:
4767  **/
4768
4769 static int
4770 e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4771 {
4772         struct e1000_adapter *adapter = netdev_priv(netdev);
4773         struct mii_ioctl_data *data = if_mii(ifr);
4774         int retval;
4775         uint16_t mii_reg;
4776         uint16_t spddplx;
4777         unsigned long flags;
4778
4779         if (adapter->hw.media_type != e1000_media_type_copper)
4780                 return -EOPNOTSUPP;
4781
4782         switch (cmd) {
4783         case SIOCGMIIPHY:
4784                 data->phy_id = adapter->hw.phy_addr;
4785                 break;
4786         case SIOCGMIIREG:
4787                 if (!capable(CAP_NET_ADMIN))
4788                         return -EPERM;
4789                 spin_lock_irqsave(&adapter->stats_lock, flags);
4790                 if (e1000_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
4791                                    &data->val_out)) {
4792                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4793                         return -EIO;
4794                 }
4795                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4796                 break;
4797         case SIOCSMIIREG:
4798                 if (!capable(CAP_NET_ADMIN))
4799                         return -EPERM;
4800                 if (data->reg_num & ~(0x1F))
4801                         return -EFAULT;
4802                 mii_reg = data->val_in;
4803                 spin_lock_irqsave(&adapter->stats_lock, flags);
4804                 if (e1000_write_phy_reg(&adapter->hw, data->reg_num,
4805                                         mii_reg)) {
4806                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4807                         return -EIO;
4808                 }
4809                 if (adapter->hw.media_type == e1000_media_type_copper) {
4810                         switch (data->reg_num) {
4811                         case PHY_CTRL:
4812                                 if (mii_reg & MII_CR_POWER_DOWN)
4813                                         break;
4814                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4815                                         adapter->hw.autoneg = 1;
4816                                         adapter->hw.autoneg_advertised = 0x2F;
4817                                 } else {
4818                                         if (mii_reg & 0x40)
4819                                                 spddplx = SPEED_1000;
4820                                         else if (mii_reg & 0x2000)
4821                                                 spddplx = SPEED_100;
4822                                         else
4823                                                 spddplx = SPEED_10;
4824                                         spddplx += (mii_reg & 0x100)
4825                                                    ? DUPLEX_FULL :
4826                                                    DUPLEX_HALF;
4827                                         retval = e1000_set_spd_dplx(adapter,
4828                                                                     spddplx);
4829                                         if (retval) {
4830                                                 spin_unlock_irqrestore(
4831                                                         &adapter->stats_lock,
4832                                                         flags);
4833                                                 return retval;
4834                                         }
4835                                 }
4836                                 if (netif_running(adapter->netdev))
4837                                         e1000_reinit_locked(adapter);
4838                                 else
4839                                         e1000_reset(adapter);
4840                                 break;
4841                         case M88E1000_PHY_SPEC_CTRL:
4842                         case M88E1000_EXT_PHY_SPEC_CTRL:
4843                                 if (e1000_phy_reset(&adapter->hw)) {
4844                                         spin_unlock_irqrestore(
4845                                                 &adapter->stats_lock, flags);
4846                                         return -EIO;
4847                                 }
4848                                 break;
4849                         }
4850                 } else {
4851                         switch (data->reg_num) {
4852                         case PHY_CTRL:
4853                                 if (mii_reg & MII_CR_POWER_DOWN)
4854                                         break;
4855                                 if (netif_running(adapter->netdev))
4856                                         e1000_reinit_locked(adapter);
4857                                 else
4858                                         e1000_reset(adapter);
4859                                 break;
4860                         }
4861                 }
4862                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4863                 break;
4864         default:
4865                 return -EOPNOTSUPP;
4866         }
4867         return E1000_SUCCESS;
4868 }
4869
4870 void
4871 e1000_pci_set_mwi(struct e1000_hw *hw)
4872 {
4873         struct e1000_adapter *adapter = hw->back;
4874         int ret_val = pci_set_mwi(adapter->pdev);
4875
4876         if (ret_val)
4877                 DPRINTK(PROBE, ERR, "Error in setting MWI\n");
4878 }
4879
4880 void
4881 e1000_pci_clear_mwi(struct e1000_hw *hw)
4882 {
4883         struct e1000_adapter *adapter = hw->back;
4884
4885         pci_clear_mwi(adapter->pdev);
4886 }
4887
4888 void
4889 e1000_read_pci_cfg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4890 {
4891         struct e1000_adapter *adapter = hw->back;
4892
4893         pci_read_config_word(adapter->pdev, reg, value);
4894 }
4895
4896 void
4897 e1000_write_pci_cfg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4898 {
4899         struct e1000_adapter *adapter = hw->back;
4900
4901         pci_write_config_word(adapter->pdev, reg, *value);
4902 }
4903
4904 int32_t
4905 e1000_read_pcie_cap_reg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4906 {
4907     struct e1000_adapter *adapter = hw->back;
4908     uint16_t cap_offset;
4909
4910     cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
4911     if (!cap_offset)
4912         return -E1000_ERR_CONFIG;
4913
4914     pci_read_config_word(adapter->pdev, cap_offset + reg, value);
4915
4916     return E1000_SUCCESS;
4917 }
4918
4919 void
4920 e1000_io_write(struct e1000_hw *hw, unsigned long port, uint32_t value)
4921 {
4922         outl(value, port);
4923 }
4924
4925 static void
4926 e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
4927 {
4928         struct e1000_adapter *adapter = netdev_priv(netdev);
4929         uint32_t ctrl, rctl;
4930
4931         e1000_irq_disable(adapter);
4932         adapter->vlgrp = grp;
4933
4934         if (grp) {
4935                 /* enable VLAN tag insert/strip */
4936                 ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4937                 ctrl |= E1000_CTRL_VME;
4938                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4939
4940                 if (adapter->hw.mac_type != e1000_ich8lan) {
4941                         /* enable VLAN receive filtering */
4942                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
4943                         rctl |= E1000_RCTL_VFE;
4944                         rctl &= ~E1000_RCTL_CFIEN;
4945                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4946                         e1000_update_mng_vlan(adapter);
4947                 }
4948         } else {
4949                 /* disable VLAN tag insert/strip */
4950                 ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4951                 ctrl &= ~E1000_CTRL_VME;
4952                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4953
4954                 if (adapter->hw.mac_type != e1000_ich8lan) {
4955                         /* disable VLAN filtering */
4956                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
4957                         rctl &= ~E1000_RCTL_VFE;
4958                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4959                         if (adapter->mng_vlan_id !=
4960                             (uint16_t)E1000_MNG_VLAN_NONE) {
4961                                 e1000_vlan_rx_kill_vid(netdev,
4962                                                        adapter->mng_vlan_id);
4963                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4964                         }
4965                 }
4966         }
4967
4968         e1000_irq_enable(adapter);
4969 }
4970
4971 static void
4972 e1000_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid)
4973 {
4974         struct e1000_adapter *adapter = netdev_priv(netdev);
4975         uint32_t vfta, index;
4976
4977         if ((adapter->hw.mng_cookie.status &
4978              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4979             (vid == adapter->mng_vlan_id))
4980                 return;
4981         /* add VID to filter table */
4982         index = (vid >> 5) & 0x7F;
4983         vfta = E1000_READ_REG_ARRAY(&adapter->hw, VFTA, index);
4984         vfta |= (1 << (vid & 0x1F));
4985         e1000_write_vfta(&adapter->hw, index, vfta);
4986 }
4987
4988 static void
4989 e1000_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid)
4990 {
4991         struct e1000_adapter *adapter = netdev_priv(netdev);
4992         uint32_t vfta, index;
4993
4994         e1000_irq_disable(adapter);
4995         vlan_group_set_device(adapter->vlgrp, vid, NULL);
4996         e1000_irq_enable(adapter);
4997
4998         if ((adapter->hw.mng_cookie.status &
4999              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
5000             (vid == adapter->mng_vlan_id)) {
5001                 /* release control to f/w */
5002                 e1000_release_hw_control(adapter);
5003                 return;
5004         }
5005
5006         /* remove VID from filter table */
5007         index = (vid >> 5) & 0x7F;
5008         vfta = E1000_READ_REG_ARRAY(&adapter->hw, VFTA, index);
5009         vfta &= ~(1 << (vid & 0x1F));
5010         e1000_write_vfta(&adapter->hw, index, vfta);
5011 }
5012
5013 static void
5014 e1000_restore_vlan(struct e1000_adapter *adapter)
5015 {
5016         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
5017
5018         if (adapter->vlgrp) {
5019                 uint16_t vid;
5020                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
5021                         if (!vlan_group_get_device(adapter->vlgrp, vid))
5022                                 continue;
5023                         e1000_vlan_rx_add_vid(adapter->netdev, vid);
5024                 }
5025         }
5026 }
5027
5028 int
5029 e1000_set_spd_dplx(struct e1000_adapter *adapter, uint16_t spddplx)
5030 {
5031         adapter->hw.autoneg = 0;
5032
5033         /* Fiber NICs only allow 1000 gbps Full duplex */
5034         if ((adapter->hw.media_type == e1000_media_type_fiber) &&
5035                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
5036                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
5037                 return -EINVAL;
5038         }
5039
5040         switch (spddplx) {
5041         case SPEED_10 + DUPLEX_HALF:
5042                 adapter->hw.forced_speed_duplex = e1000_10_half;
5043                 break;
5044         case SPEED_10 + DUPLEX_FULL:
5045                 adapter->hw.forced_speed_duplex = e1000_10_full;
5046                 break;
5047         case SPEED_100 + DUPLEX_HALF:
5048                 adapter->hw.forced_speed_duplex = e1000_100_half;
5049                 break;
5050         case SPEED_100 + DUPLEX_FULL:
5051                 adapter->hw.forced_speed_duplex = e1000_100_full;
5052                 break;
5053         case SPEED_1000 + DUPLEX_FULL:
5054                 adapter->hw.autoneg = 1;
5055                 adapter->hw.autoneg_advertised = ADVERTISE_1000_FULL;
5056                 break;
5057         case SPEED_1000 + DUPLEX_HALF: /* not supported */
5058         default:
5059                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
5060                 return -EINVAL;
5061         }
5062         return 0;
5063 }
5064
5065 static int
5066 e1000_suspend(struct pci_dev *pdev, pm_message_t state)
5067 {
5068         struct net_device *netdev = pci_get_drvdata(pdev);
5069         struct e1000_adapter *adapter = netdev_priv(netdev);
5070         uint32_t ctrl, ctrl_ext, rctl, status;
5071         uint32_t wufc = adapter->wol;
5072 #ifdef CONFIG_PM
5073         int retval = 0;
5074 #endif
5075
5076         netif_device_detach(netdev);
5077
5078         if (netif_running(netdev)) {
5079                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
5080                 e1000_down(adapter);
5081         }
5082
5083 #ifdef CONFIG_PM
5084         retval = pci_save_state(pdev);
5085         if (retval)
5086                 return retval;
5087 #endif
5088
5089         status = E1000_READ_REG(&adapter->hw, STATUS);
5090         if (status & E1000_STATUS_LU)
5091                 wufc &= ~E1000_WUFC_LNKC;
5092
5093         if (wufc) {
5094                 e1000_setup_rctl(adapter);
5095                 e1000_set_multi(netdev);
5096
5097                 /* turn on all-multi mode if wake on multicast is enabled */
5098                 if (wufc & E1000_WUFC_MC) {
5099                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
5100                         rctl |= E1000_RCTL_MPE;
5101                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
5102                 }
5103
5104                 if (adapter->hw.mac_type >= e1000_82540) {
5105                         ctrl = E1000_READ_REG(&adapter->hw, CTRL);
5106                         /* advertise wake from D3Cold */
5107                         #define E1000_CTRL_ADVD3WUC 0x00100000
5108                         /* phy power management enable */
5109                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5110                         ctrl |= E1000_CTRL_ADVD3WUC |
5111                                 E1000_CTRL_EN_PHY_PWR_MGMT;
5112                         E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
5113                 }
5114
5115                 if (adapter->hw.media_type == e1000_media_type_fiber ||
5116                    adapter->hw.media_type == e1000_media_type_internal_serdes) {
5117                         /* keep the laser running in D3 */
5118                         ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
5119                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
5120                         E1000_WRITE_REG(&adapter->hw, CTRL_EXT, ctrl_ext);
5121                 }
5122
5123                 /* Allow time for pending master requests to run */
5124                 e1000_disable_pciex_master(&adapter->hw);
5125
5126                 E1000_WRITE_REG(&adapter->hw, WUC, E1000_WUC_PME_EN);
5127                 E1000_WRITE_REG(&adapter->hw, WUFC, wufc);
5128                 pci_enable_wake(pdev, PCI_D3hot, 1);
5129                 pci_enable_wake(pdev, PCI_D3cold, 1);
5130         } else {
5131                 E1000_WRITE_REG(&adapter->hw, WUC, 0);
5132                 E1000_WRITE_REG(&adapter->hw, WUFC, 0);
5133                 pci_enable_wake(pdev, PCI_D3hot, 0);
5134                 pci_enable_wake(pdev, PCI_D3cold, 0);
5135         }
5136
5137         e1000_release_manageability(adapter);
5138
5139         /* make sure adapter isn't asleep if manageability is enabled */
5140         if (adapter->en_mng_pt) {
5141                 pci_enable_wake(pdev, PCI_D3hot, 1);
5142                 pci_enable_wake(pdev, PCI_D3cold, 1);
5143         }
5144
5145         if (adapter->hw.phy_type == e1000_phy_igp_3)
5146                 e1000_phy_powerdown_workaround(&adapter->hw);
5147
5148         if (netif_running(netdev))
5149                 e1000_free_irq(adapter);
5150
5151         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
5152          * would have already happened in close and is redundant. */
5153         e1000_release_hw_control(adapter);
5154
5155         pci_disable_device(pdev);
5156
5157         pci_set_power_state(pdev, pci_choose_state(pdev, state));
5158
5159         return 0;
5160 }
5161
5162 #ifdef CONFIG_PM
5163 static int
5164 e1000_resume(struct pci_dev *pdev)
5165 {
5166         struct net_device *netdev = pci_get_drvdata(pdev);
5167         struct e1000_adapter *adapter = netdev_priv(netdev);
5168         uint32_t err;
5169
5170         pci_set_power_state(pdev, PCI_D0);
5171         pci_restore_state(pdev);
5172         if ((err = pci_enable_device(pdev))) {
5173                 printk(KERN_ERR "e1000: Cannot enable PCI device from suspend\n");
5174                 return err;
5175         }
5176         pci_set_master(pdev);
5177
5178         pci_enable_wake(pdev, PCI_D3hot, 0);
5179         pci_enable_wake(pdev, PCI_D3cold, 0);
5180
5181         if (netif_running(netdev) && (err = e1000_request_irq(adapter)))
5182                 return err;
5183
5184         e1000_power_up_phy(adapter);
5185         e1000_reset(adapter);
5186         E1000_WRITE_REG(&adapter->hw, WUS, ~0);
5187
5188         e1000_init_manageability(adapter);
5189
5190         if (netif_running(netdev))
5191                 e1000_up(adapter);
5192
5193         netif_device_attach(netdev);
5194
5195         /* If the controller is 82573 and f/w is AMT, do not set
5196          * DRV_LOAD until the interface is up.  For all other cases,
5197          * let the f/w know that the h/w is now under the control
5198          * of the driver. */
5199         if (adapter->hw.mac_type != e1000_82573 ||
5200             !e1000_check_mng_mode(&adapter->hw))
5201                 e1000_get_hw_control(adapter);
5202
5203         return 0;
5204 }
5205 #endif
5206
5207 static void e1000_shutdown(struct pci_dev *pdev)
5208 {
5209         e1000_suspend(pdev, PMSG_SUSPEND);
5210 }
5211
5212 #ifdef CONFIG_NET_POLL_CONTROLLER
5213 /*
5214  * Polling 'interrupt' - used by things like netconsole to send skbs
5215  * without having to re-enable interrupts. It's not called while
5216  * the interrupt routine is executing.
5217  */
5218 static void
5219 e1000_netpoll(struct net_device *netdev)
5220 {
5221         struct e1000_adapter *adapter = netdev_priv(netdev);
5222
5223         disable_irq(adapter->pdev->irq);
5224         e1000_intr(adapter->pdev->irq, netdev);
5225         e1000_clean_tx_irq(adapter, adapter->tx_ring);
5226 #ifndef CONFIG_E1000_NAPI
5227         adapter->clean_rx(adapter, adapter->rx_ring);
5228 #endif
5229         enable_irq(adapter->pdev->irq);
5230 }
5231 #endif
5232
5233 /**
5234  * e1000_io_error_detected - called when PCI error is detected
5235  * @pdev: Pointer to PCI device
5236  * @state: The current pci conneection state
5237  *
5238  * This function is called after a PCI bus error affecting
5239  * this device has been detected.
5240  */
5241 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
5242 {
5243         struct net_device *netdev = pci_get_drvdata(pdev);
5244         struct e1000_adapter *adapter = netdev->priv;
5245
5246         netif_device_detach(netdev);
5247
5248         if (netif_running(netdev))
5249                 e1000_down(adapter);
5250         pci_disable_device(pdev);
5251
5252         /* Request a slot slot reset. */
5253         return PCI_ERS_RESULT_NEED_RESET;
5254 }
5255
5256 /**
5257  * e1000_io_slot_reset - called after the pci bus has been reset.
5258  * @pdev: Pointer to PCI device
5259  *
5260  * Restart the card from scratch, as if from a cold-boot. Implementation
5261  * resembles the first-half of the e1000_resume routine.
5262  */
5263 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5264 {
5265         struct net_device *netdev = pci_get_drvdata(pdev);
5266         struct e1000_adapter *adapter = netdev->priv;
5267
5268         if (pci_enable_device(pdev)) {
5269                 printk(KERN_ERR "e1000: Cannot re-enable PCI device after reset.\n");
5270                 return PCI_ERS_RESULT_DISCONNECT;
5271         }
5272         pci_set_master(pdev);
5273
5274         pci_enable_wake(pdev, PCI_D3hot, 0);
5275         pci_enable_wake(pdev, PCI_D3cold, 0);
5276
5277         e1000_reset(adapter);
5278         E1000_WRITE_REG(&adapter->hw, WUS, ~0);
5279
5280         return PCI_ERS_RESULT_RECOVERED;
5281 }
5282
5283 /**
5284  * e1000_io_resume - called when traffic can start flowing again.
5285  * @pdev: Pointer to PCI device
5286  *
5287  * This callback is called when the error recovery driver tells us that
5288  * its OK to resume normal operation. Implementation resembles the
5289  * second-half of the e1000_resume routine.
5290  */
5291 static void e1000_io_resume(struct pci_dev *pdev)
5292 {
5293         struct net_device *netdev = pci_get_drvdata(pdev);
5294         struct e1000_adapter *adapter = netdev->priv;
5295
5296         e1000_init_manageability(adapter);
5297
5298         if (netif_running(netdev)) {
5299                 if (e1000_up(adapter)) {
5300                         printk("e1000: can't bring device back up after reset\n");
5301                         return;
5302                 }
5303         }
5304
5305         netif_device_attach(netdev);
5306
5307         /* If the controller is 82573 and f/w is AMT, do not set
5308          * DRV_LOAD until the interface is up.  For all other cases,
5309          * let the f/w know that the h/w is now under the control
5310          * of the driver. */
5311         if (adapter->hw.mac_type != e1000_82573 ||
5312             !e1000_check_mng_mode(&adapter->hw))
5313                 e1000_get_hw_control(adapter);
5314
5315 }
5316
5317 /* e1000_main.c */