e1000: Add device IDs of blade version of the 82571 quad port
[sfrench/cifs-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(0x10A5),
104         INTEL_E1000_ETHERNET_DEVICE(0x10B5),
105         INTEL_E1000_ETHERNET_DEVICE(0x10B9),
106         INTEL_E1000_ETHERNET_DEVICE(0x10BA),
107         INTEL_E1000_ETHERNET_DEVICE(0x10BB),
108         INTEL_E1000_ETHERNET_DEVICE(0x10BC),
109         INTEL_E1000_ETHERNET_DEVICE(0x10C4),
110         INTEL_E1000_ETHERNET_DEVICE(0x10C5),
111         INTEL_E1000_ETHERNET_DEVICE(0x10D5),
112         INTEL_E1000_ETHERNET_DEVICE(0x10D9),
113         INTEL_E1000_ETHERNET_DEVICE(0x10DA),
114         /* required last entry */
115         {0,}
116 };
117
118 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
119
120 int e1000_up(struct e1000_adapter *adapter);
121 void e1000_down(struct e1000_adapter *adapter);
122 void e1000_reinit_locked(struct e1000_adapter *adapter);
123 void e1000_reset(struct e1000_adapter *adapter);
124 int e1000_set_spd_dplx(struct e1000_adapter *adapter, uint16_t spddplx);
125 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
126 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
127 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
128 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
129 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
130                              struct e1000_tx_ring *txdr);
131 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
132                              struct e1000_rx_ring *rxdr);
133 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
134                              struct e1000_tx_ring *tx_ring);
135 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
136                              struct e1000_rx_ring *rx_ring);
137 void e1000_update_stats(struct e1000_adapter *adapter);
138
139 static int e1000_init_module(void);
140 static void e1000_exit_module(void);
141 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
142 static void __devexit e1000_remove(struct pci_dev *pdev);
143 static int e1000_alloc_queues(struct e1000_adapter *adapter);
144 static int e1000_sw_init(struct e1000_adapter *adapter);
145 static int e1000_open(struct net_device *netdev);
146 static int e1000_close(struct net_device *netdev);
147 static void e1000_configure_tx(struct e1000_adapter *adapter);
148 static void e1000_configure_rx(struct e1000_adapter *adapter);
149 static void e1000_setup_rctl(struct e1000_adapter *adapter);
150 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
151 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
152 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
153                                 struct e1000_tx_ring *tx_ring);
154 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
155                                 struct e1000_rx_ring *rx_ring);
156 static void e1000_set_multi(struct net_device *netdev);
157 static void e1000_update_phy_info(unsigned long data);
158 static void e1000_watchdog(unsigned long data);
159 static void e1000_82547_tx_fifo_stall(unsigned long data);
160 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
161 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
162 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
163 static int e1000_set_mac(struct net_device *netdev, void *p);
164 static irqreturn_t e1000_intr(int irq, void *data);
165 static irqreturn_t e1000_intr_msi(int irq, void *data);
166 static boolean_t e1000_clean_tx_irq(struct e1000_adapter *adapter,
167                                     struct e1000_tx_ring *tx_ring);
168 #ifdef CONFIG_E1000_NAPI
169 static int e1000_clean(struct net_device *poll_dev, int *budget);
170 static boolean_t e1000_clean_rx_irq(struct e1000_adapter *adapter,
171                                     struct e1000_rx_ring *rx_ring,
172                                     int *work_done, int work_to_do);
173 static boolean_t e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
174                                        struct e1000_rx_ring *rx_ring,
175                                        int *work_done, int work_to_do);
176 #else
177 static boolean_t e1000_clean_rx_irq(struct e1000_adapter *adapter,
178                                     struct e1000_rx_ring *rx_ring);
179 static boolean_t e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
180                                        struct e1000_rx_ring *rx_ring);
181 #endif
182 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
183                                    struct e1000_rx_ring *rx_ring,
184                                    int cleaned_count);
185 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
186                                       struct e1000_rx_ring *rx_ring,
187                                       int cleaned_count);
188 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
189 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
190                            int cmd);
191 void e1000_set_ethtool_ops(struct net_device *netdev);
192 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
193 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
194 static void e1000_tx_timeout(struct net_device *dev);
195 static void e1000_reset_task(struct work_struct *work);
196 static void e1000_smartspeed(struct e1000_adapter *adapter);
197 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
198                                        struct sk_buff *skb);
199
200 static void e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp);
201 static void e1000_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid);
202 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid);
203 static void e1000_restore_vlan(struct e1000_adapter *adapter);
204
205 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
206 #ifdef CONFIG_PM
207 static int e1000_resume(struct pci_dev *pdev);
208 #endif
209 static void e1000_shutdown(struct pci_dev *pdev);
210
211 #ifdef CONFIG_NET_POLL_CONTROLLER
212 /* for netdump / net console */
213 static void e1000_netpoll (struct net_device *netdev);
214 #endif
215
216 extern void e1000_check_options(struct e1000_adapter *adapter);
217
218 #define COPYBREAK_DEFAULT 256
219 static unsigned int copybreak __read_mostly = COPYBREAK_DEFAULT;
220 module_param(copybreak, uint, 0644);
221 MODULE_PARM_DESC(copybreak,
222         "Maximum size of packet that is copied to a new buffer on receive");
223
224 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
225                      pci_channel_state_t state);
226 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
227 static void e1000_io_resume(struct pci_dev *pdev);
228
229 static struct pci_error_handlers e1000_err_handler = {
230         .error_detected = e1000_io_error_detected,
231         .slot_reset = e1000_io_slot_reset,
232         .resume = e1000_io_resume,
233 };
234
235 static struct pci_driver e1000_driver = {
236         .name     = e1000_driver_name,
237         .id_table = e1000_pci_tbl,
238         .probe    = e1000_probe,
239         .remove   = __devexit_p(e1000_remove),
240 #ifdef CONFIG_PM
241         /* Power Managment Hooks */
242         .suspend  = e1000_suspend,
243         .resume   = e1000_resume,
244 #endif
245         .shutdown = e1000_shutdown,
246         .err_handler = &e1000_err_handler
247 };
248
249 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
250 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
251 MODULE_LICENSE("GPL");
252 MODULE_VERSION(DRV_VERSION);
253
254 static int debug = NETIF_MSG_DRV | NETIF_MSG_PROBE;
255 module_param(debug, int, 0);
256 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
257
258 /**
259  * e1000_init_module - Driver Registration Routine
260  *
261  * e1000_init_module is the first routine called when the driver is
262  * loaded. All it does is register with the PCI subsystem.
263  **/
264
265 static int __init
266 e1000_init_module(void)
267 {
268         int ret;
269         printk(KERN_INFO "%s - version %s\n",
270                e1000_driver_string, e1000_driver_version);
271
272         printk(KERN_INFO "%s\n", e1000_copyright);
273
274         ret = pci_register_driver(&e1000_driver);
275         if (copybreak != COPYBREAK_DEFAULT) {
276                 if (copybreak == 0)
277                         printk(KERN_INFO "e1000: copybreak disabled\n");
278                 else
279                         printk(KERN_INFO "e1000: copybreak enabled for "
280                                "packets <= %u bytes\n", copybreak);
281         }
282         return ret;
283 }
284
285 module_init(e1000_init_module);
286
287 /**
288  * e1000_exit_module - Driver Exit Cleanup Routine
289  *
290  * e1000_exit_module is called just before the driver is removed
291  * from memory.
292  **/
293
294 static void __exit
295 e1000_exit_module(void)
296 {
297         pci_unregister_driver(&e1000_driver);
298 }
299
300 module_exit(e1000_exit_module);
301
302 static int e1000_request_irq(struct e1000_adapter *adapter)
303 {
304         struct net_device *netdev = adapter->netdev;
305         void (*handler) = &e1000_intr;
306         int irq_flags = IRQF_SHARED;
307         int err;
308
309         if (adapter->hw.mac_type >= e1000_82571) {
310                 adapter->have_msi = !pci_enable_msi(adapter->pdev);
311                 if (adapter->have_msi) {
312                         handler = &e1000_intr_msi;
313                         irq_flags = 0;
314                 }
315         }
316
317         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
318                           netdev);
319         if (err) {
320                 if (adapter->have_msi)
321                         pci_disable_msi(adapter->pdev);
322                 DPRINTK(PROBE, ERR,
323                         "Unable to allocate interrupt Error: %d\n", err);
324         }
325
326         return err;
327 }
328
329 static void e1000_free_irq(struct e1000_adapter *adapter)
330 {
331         struct net_device *netdev = adapter->netdev;
332
333         free_irq(adapter->pdev->irq, netdev);
334
335         if (adapter->have_msi)
336                 pci_disable_msi(adapter->pdev);
337 }
338
339 /**
340  * e1000_irq_disable - Mask off interrupt generation on the NIC
341  * @adapter: board private structure
342  **/
343
344 static void
345 e1000_irq_disable(struct e1000_adapter *adapter)
346 {
347         atomic_inc(&adapter->irq_sem);
348         E1000_WRITE_REG(&adapter->hw, IMC, ~0);
349         E1000_WRITE_FLUSH(&adapter->hw);
350         synchronize_irq(adapter->pdev->irq);
351 }
352
353 /**
354  * e1000_irq_enable - Enable default interrupt generation settings
355  * @adapter: board private structure
356  **/
357
358 static void
359 e1000_irq_enable(struct e1000_adapter *adapter)
360 {
361         if (likely(atomic_dec_and_test(&adapter->irq_sem))) {
362                 E1000_WRITE_REG(&adapter->hw, IMS, IMS_ENABLE_MASK);
363                 E1000_WRITE_FLUSH(&adapter->hw);
364         }
365 }
366
367 static void
368 e1000_update_mng_vlan(struct e1000_adapter *adapter)
369 {
370         struct net_device *netdev = adapter->netdev;
371         uint16_t vid = adapter->hw.mng_cookie.vlan_id;
372         uint16_t old_vid = adapter->mng_vlan_id;
373         if (adapter->vlgrp) {
374                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
375                         if (adapter->hw.mng_cookie.status &
376                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
377                                 e1000_vlan_rx_add_vid(netdev, vid);
378                                 adapter->mng_vlan_id = vid;
379                         } else
380                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
381
382                         if ((old_vid != (uint16_t)E1000_MNG_VLAN_NONE) &&
383                                         (vid != old_vid) &&
384                             !vlan_group_get_device(adapter->vlgrp, old_vid))
385                                 e1000_vlan_rx_kill_vid(netdev, old_vid);
386                 } else
387                         adapter->mng_vlan_id = vid;
388         }
389 }
390
391 /**
392  * e1000_release_hw_control - release control of the h/w to f/w
393  * @adapter: address of board private structure
394  *
395  * e1000_release_hw_control resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
396  * For ASF and Pass Through versions of f/w this means that the
397  * driver is no longer loaded. For AMT version (only with 82573) i
398  * of the f/w this means that the network i/f is closed.
399  *
400  **/
401
402 static void
403 e1000_release_hw_control(struct e1000_adapter *adapter)
404 {
405         uint32_t ctrl_ext;
406         uint32_t swsm;
407
408         /* Let firmware taken over control of h/w */
409         switch (adapter->hw.mac_type) {
410         case e1000_82573:
411                 swsm = E1000_READ_REG(&adapter->hw, SWSM);
412                 E1000_WRITE_REG(&adapter->hw, SWSM,
413                                 swsm & ~E1000_SWSM_DRV_LOAD);
414                 break;
415         case e1000_82571:
416         case e1000_82572:
417         case e1000_80003es2lan:
418         case e1000_ich8lan:
419                 ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
420                 E1000_WRITE_REG(&adapter->hw, CTRL_EXT,
421                                 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
422                 break;
423         default:
424                 break;
425         }
426 }
427
428 /**
429  * e1000_get_hw_control - get control of the h/w from f/w
430  * @adapter: address of board private structure
431  *
432  * e1000_get_hw_control sets {CTRL_EXT|FWSM}:DRV_LOAD bit.
433  * For ASF and Pass Through versions of f/w this means that
434  * the driver is loaded. For AMT version (only with 82573)
435  * of the f/w this means that the network i/f is open.
436  *
437  **/
438
439 static void
440 e1000_get_hw_control(struct e1000_adapter *adapter)
441 {
442         uint32_t ctrl_ext;
443         uint32_t swsm;
444
445         /* Let firmware know the driver has taken over */
446         switch (adapter->hw.mac_type) {
447         case e1000_82573:
448                 swsm = E1000_READ_REG(&adapter->hw, SWSM);
449                 E1000_WRITE_REG(&adapter->hw, SWSM,
450                                 swsm | E1000_SWSM_DRV_LOAD);
451                 break;
452         case e1000_82571:
453         case e1000_82572:
454         case e1000_80003es2lan:
455         case e1000_ich8lan:
456                 ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
457                 E1000_WRITE_REG(&adapter->hw, CTRL_EXT,
458                                 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
459                 break;
460         default:
461                 break;
462         }
463 }
464
465 static void
466 e1000_init_manageability(struct e1000_adapter *adapter)
467 {
468         if (adapter->en_mng_pt) {
469                 uint32_t manc = E1000_READ_REG(&adapter->hw, MANC);
470
471                 /* disable hardware interception of ARP */
472                 manc &= ~(E1000_MANC_ARP_EN);
473
474                 /* enable receiving management packets to the host */
475                 /* this will probably generate destination unreachable messages
476                  * from the host OS, but the packets will be handled on SMBUS */
477                 if (adapter->hw.has_manc2h) {
478                         uint32_t manc2h = E1000_READ_REG(&adapter->hw, MANC2H);
479
480                         manc |= E1000_MANC_EN_MNG2HOST;
481 #define E1000_MNG2HOST_PORT_623 (1 << 5)
482 #define E1000_MNG2HOST_PORT_664 (1 << 6)
483                         manc2h |= E1000_MNG2HOST_PORT_623;
484                         manc2h |= E1000_MNG2HOST_PORT_664;
485                         E1000_WRITE_REG(&adapter->hw, MANC2H, manc2h);
486                 }
487
488                 E1000_WRITE_REG(&adapter->hw, MANC, manc);
489         }
490 }
491
492 static void
493 e1000_release_manageability(struct e1000_adapter *adapter)
494 {
495         if (adapter->en_mng_pt) {
496                 uint32_t manc = E1000_READ_REG(&adapter->hw, MANC);
497
498                 /* re-enable hardware interception of ARP */
499                 manc |= E1000_MANC_ARP_EN;
500
501                 if (adapter->hw.has_manc2h)
502                         manc &= ~E1000_MANC_EN_MNG2HOST;
503
504                 /* don't explicitly have to mess with MANC2H since
505                  * MANC has an enable disable that gates MANC2H */
506
507                 E1000_WRITE_REG(&adapter->hw, MANC, manc);
508         }
509 }
510
511 /**
512  * e1000_configure - configure the hardware for RX and TX
513  * @adapter = private board structure
514  **/
515 static void e1000_configure(struct e1000_adapter *adapter)
516 {
517         struct net_device *netdev = adapter->netdev;
518         int i;
519
520         e1000_set_multi(netdev);
521
522         e1000_restore_vlan(adapter);
523         e1000_init_manageability(adapter);
524
525         e1000_configure_tx(adapter);
526         e1000_setup_rctl(adapter);
527         e1000_configure_rx(adapter);
528         /* call E1000_DESC_UNUSED which always leaves
529          * at least 1 descriptor unused to make sure
530          * next_to_use != next_to_clean */
531         for (i = 0; i < adapter->num_rx_queues; i++) {
532                 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
533                 adapter->alloc_rx_buf(adapter, ring,
534                                       E1000_DESC_UNUSED(ring));
535         }
536
537         adapter->tx_queue_len = netdev->tx_queue_len;
538 }
539
540 int e1000_up(struct e1000_adapter *adapter)
541 {
542         /* hardware has been reset, we need to reload some things */
543         e1000_configure(adapter);
544
545         clear_bit(__E1000_DOWN, &adapter->flags);
546
547 #ifdef CONFIG_E1000_NAPI
548         netif_poll_enable(adapter->netdev);
549 #endif
550         e1000_irq_enable(adapter);
551
552         /* fire a link change interrupt to start the watchdog */
553         E1000_WRITE_REG(&adapter->hw, ICS, E1000_ICS_LSC);
554         return 0;
555 }
556
557 /**
558  * e1000_power_up_phy - restore link in case the phy was powered down
559  * @adapter: address of board private structure
560  *
561  * The phy may be powered down to save power and turn off link when the
562  * driver is unloaded and wake on lan is not enabled (among others)
563  * *** this routine MUST be followed by a call to e1000_reset ***
564  *
565  **/
566
567 void e1000_power_up_phy(struct e1000_adapter *adapter)
568 {
569         uint16_t mii_reg = 0;
570
571         /* Just clear the power down bit to wake the phy back up */
572         if (adapter->hw.media_type == e1000_media_type_copper) {
573                 /* according to the manual, the phy will retain its
574                  * settings across a power-down/up cycle */
575                 e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &mii_reg);
576                 mii_reg &= ~MII_CR_POWER_DOWN;
577                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL, mii_reg);
578         }
579 }
580
581 static void e1000_power_down_phy(struct e1000_adapter *adapter)
582 {
583         /* Power down the PHY so no link is implied when interface is down *
584          * The PHY cannot be powered down if any of the following is TRUE *
585          * (a) WoL is enabled
586          * (b) AMT is active
587          * (c) SoL/IDER session is active */
588         if (!adapter->wol && adapter->hw.mac_type >= e1000_82540 &&
589            adapter->hw.media_type == e1000_media_type_copper) {
590                 uint16_t mii_reg = 0;
591
592                 switch (adapter->hw.mac_type) {
593                 case e1000_82540:
594                 case e1000_82545:
595                 case e1000_82545_rev_3:
596                 case e1000_82546:
597                 case e1000_82546_rev_3:
598                 case e1000_82541:
599                 case e1000_82541_rev_2:
600                 case e1000_82547:
601                 case e1000_82547_rev_2:
602                         if (E1000_READ_REG(&adapter->hw, MANC) &
603                             E1000_MANC_SMBUS_EN)
604                                 goto out;
605                         break;
606                 case e1000_82571:
607                 case e1000_82572:
608                 case e1000_82573:
609                 case e1000_80003es2lan:
610                 case e1000_ich8lan:
611                         if (e1000_check_mng_mode(&adapter->hw) ||
612                             e1000_check_phy_reset_block(&adapter->hw))
613                                 goto out;
614                         break;
615                 default:
616                         goto out;
617                 }
618                 e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &mii_reg);
619                 mii_reg |= MII_CR_POWER_DOWN;
620                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL, mii_reg);
621                 mdelay(1);
622         }
623 out:
624         return;
625 }
626
627 void
628 e1000_down(struct e1000_adapter *adapter)
629 {
630         struct net_device *netdev = adapter->netdev;
631
632         /* signal that we're down so the interrupt handler does not
633          * reschedule our watchdog timer */
634         set_bit(__E1000_DOWN, &adapter->flags);
635
636 #ifdef CONFIG_E1000_NAPI
637         netif_poll_disable(netdev);
638 #endif
639         e1000_irq_disable(adapter);
640
641         del_timer_sync(&adapter->tx_fifo_stall_timer);
642         del_timer_sync(&adapter->watchdog_timer);
643         del_timer_sync(&adapter->phy_info_timer);
644
645         netdev->tx_queue_len = adapter->tx_queue_len;
646         adapter->link_speed = 0;
647         adapter->link_duplex = 0;
648         netif_carrier_off(netdev);
649         netif_stop_queue(netdev);
650
651         e1000_reset(adapter);
652         e1000_clean_all_tx_rings(adapter);
653         e1000_clean_all_rx_rings(adapter);
654 }
655
656 void
657 e1000_reinit_locked(struct e1000_adapter *adapter)
658 {
659         WARN_ON(in_interrupt());
660         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
661                 msleep(1);
662         e1000_down(adapter);
663         e1000_up(adapter);
664         clear_bit(__E1000_RESETTING, &adapter->flags);
665 }
666
667 void
668 e1000_reset(struct e1000_adapter *adapter)
669 {
670         uint32_t pba = 0, tx_space, min_tx_space, min_rx_space;
671         uint16_t fc_high_water_mark = E1000_FC_HIGH_DIFF;
672         boolean_t legacy_pba_adjust = FALSE;
673
674         /* Repartition Pba for greater than 9k mtu
675          * To take effect CTRL.RST is required.
676          */
677
678         switch (adapter->hw.mac_type) {
679         case e1000_82542_rev2_0:
680         case e1000_82542_rev2_1:
681         case e1000_82543:
682         case e1000_82544:
683         case e1000_82540:
684         case e1000_82541:
685         case e1000_82541_rev_2:
686                 legacy_pba_adjust = TRUE;
687                 pba = E1000_PBA_48K;
688                 break;
689         case e1000_82545:
690         case e1000_82545_rev_3:
691         case e1000_82546:
692         case e1000_82546_rev_3:
693                 pba = E1000_PBA_48K;
694                 break;
695         case e1000_82547:
696         case e1000_82547_rev_2:
697                 legacy_pba_adjust = TRUE;
698                 pba = E1000_PBA_30K;
699                 break;
700         case e1000_82571:
701         case e1000_82572:
702         case e1000_80003es2lan:
703                 pba = E1000_PBA_38K;
704                 break;
705         case e1000_82573:
706                 pba = E1000_PBA_20K;
707                 break;
708         case e1000_ich8lan:
709                 pba = E1000_PBA_8K;
710         case e1000_undefined:
711         case e1000_num_macs:
712                 break;
713         }
714
715         if (legacy_pba_adjust == TRUE) {
716                 if (adapter->netdev->mtu > E1000_RXBUFFER_8192)
717                         pba -= 8; /* allocate more FIFO for Tx */
718
719                 if (adapter->hw.mac_type == e1000_82547) {
720                         adapter->tx_fifo_head = 0;
721                         adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
722                         adapter->tx_fifo_size =
723                                 (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
724                         atomic_set(&adapter->tx_fifo_stall, 0);
725                 }
726         } else if (adapter->hw.max_frame_size > MAXIMUM_ETHERNET_FRAME_SIZE) {
727                 /* adjust PBA for jumbo frames */
728                 E1000_WRITE_REG(&adapter->hw, PBA, pba);
729
730                 /* To maintain wire speed transmits, the Tx FIFO should be
731                  * large enough to accomodate two full transmit packets,
732                  * rounded up to the next 1KB and expressed in KB.  Likewise,
733                  * the Rx FIFO should be large enough to accomodate at least
734                  * one full receive packet and is similarly rounded up and
735                  * expressed in KB. */
736                 pba = E1000_READ_REG(&adapter->hw, PBA);
737                 /* upper 16 bits has Tx packet buffer allocation size in KB */
738                 tx_space = pba >> 16;
739                 /* lower 16 bits has Rx packet buffer allocation size in KB */
740                 pba &= 0xffff;
741                 /* don't include ethernet FCS because hardware appends/strips */
742                 min_rx_space = adapter->netdev->mtu + ENET_HEADER_SIZE +
743                                VLAN_TAG_SIZE;
744                 min_tx_space = min_rx_space;
745                 min_tx_space *= 2;
746                 min_tx_space = ALIGN(min_tx_space, 1024);
747                 min_tx_space >>= 10;
748                 min_rx_space = ALIGN(min_rx_space, 1024);
749                 min_rx_space >>= 10;
750
751                 /* If current Tx allocation is less than the min Tx FIFO size,
752                  * and the min Tx FIFO size is less than the current Rx FIFO
753                  * allocation, take space away from current Rx allocation */
754                 if (tx_space < min_tx_space &&
755                     ((min_tx_space - tx_space) < pba)) {
756                         pba = pba - (min_tx_space - tx_space);
757
758                         /* PCI/PCIx hardware has PBA alignment constraints */
759                         switch (adapter->hw.mac_type) {
760                         case e1000_82545 ... e1000_82546_rev_3:
761                                 pba &= ~(E1000_PBA_8K - 1);
762                                 break;
763                         default:
764                                 break;
765                         }
766
767                         /* if short on rx space, rx wins and must trump tx
768                          * adjustment or use Early Receive if available */
769                         if (pba < min_rx_space) {
770                                 switch (adapter->hw.mac_type) {
771                                 case e1000_82573:
772                                         /* ERT enabled in e1000_configure_rx */
773                                         break;
774                                 default:
775                                         pba = min_rx_space;
776                                         break;
777                                 }
778                         }
779                 }
780         }
781
782         E1000_WRITE_REG(&adapter->hw, PBA, pba);
783
784         /* flow control settings */
785         /* Set the FC high water mark to 90% of the FIFO size.
786          * Required to clear last 3 LSB */
787         fc_high_water_mark = ((pba * 9216)/10) & 0xFFF8;
788         /* We can't use 90% on small FIFOs because the remainder
789          * would be less than 1 full frame.  In this case, we size
790          * it to allow at least a full frame above the high water
791          *  mark. */
792         if (pba < E1000_PBA_16K)
793                 fc_high_water_mark = (pba * 1024) - 1600;
794
795         adapter->hw.fc_high_water = fc_high_water_mark;
796         adapter->hw.fc_low_water = fc_high_water_mark - 8;
797         if (adapter->hw.mac_type == e1000_80003es2lan)
798                 adapter->hw.fc_pause_time = 0xFFFF;
799         else
800                 adapter->hw.fc_pause_time = E1000_FC_PAUSE_TIME;
801         adapter->hw.fc_send_xon = 1;
802         adapter->hw.fc = adapter->hw.original_fc;
803
804         /* Allow time for pending master requests to run */
805         e1000_reset_hw(&adapter->hw);
806         if (adapter->hw.mac_type >= e1000_82544)
807                 E1000_WRITE_REG(&adapter->hw, WUC, 0);
808
809         if (e1000_init_hw(&adapter->hw))
810                 DPRINTK(PROBE, ERR, "Hardware Error\n");
811         e1000_update_mng_vlan(adapter);
812
813         /* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
814         if (adapter->hw.mac_type >= e1000_82544 &&
815             adapter->hw.mac_type <= e1000_82547_rev_2 &&
816             adapter->hw.autoneg == 1 &&
817             adapter->hw.autoneg_advertised == ADVERTISE_1000_FULL) {
818                 uint32_t ctrl = E1000_READ_REG(&adapter->hw, CTRL);
819                 /* clear phy power management bit if we are in gig only mode,
820                  * which if enabled will attempt negotiation to 100Mb, which
821                  * can cause a loss of link at power off or driver unload */
822                 ctrl &= ~E1000_CTRL_SWDPIN3;
823                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
824         }
825
826         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
827         E1000_WRITE_REG(&adapter->hw, VET, ETHERNET_IEEE_VLAN_TYPE);
828
829         e1000_reset_adaptive(&adapter->hw);
830         e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
831
832         if (!adapter->smart_power_down &&
833             (adapter->hw.mac_type == e1000_82571 ||
834              adapter->hw.mac_type == e1000_82572)) {
835                 uint16_t phy_data = 0;
836                 /* speed up time to link by disabling smart power down, ignore
837                  * the return value of this function because there is nothing
838                  * different we would do if it failed */
839                 e1000_read_phy_reg(&adapter->hw, IGP02E1000_PHY_POWER_MGMT,
840                                    &phy_data);
841                 phy_data &= ~IGP02E1000_PM_SPD;
842                 e1000_write_phy_reg(&adapter->hw, IGP02E1000_PHY_POWER_MGMT,
843                                     phy_data);
844         }
845
846         e1000_release_manageability(adapter);
847 }
848
849 /**
850  * e1000_probe - Device Initialization Routine
851  * @pdev: PCI device information struct
852  * @ent: entry in e1000_pci_tbl
853  *
854  * Returns 0 on success, negative on failure
855  *
856  * e1000_probe initializes an adapter identified by a pci_dev structure.
857  * The OS initialization, configuring of the adapter private structure,
858  * and a hardware reset occur.
859  **/
860
861 static int __devinit
862 e1000_probe(struct pci_dev *pdev,
863             const struct pci_device_id *ent)
864 {
865         struct net_device *netdev;
866         struct e1000_adapter *adapter;
867         unsigned long mmio_start, mmio_len;
868         unsigned long flash_start, flash_len;
869
870         static int cards_found = 0;
871         static int global_quad_port_a = 0; /* global ksp3 port a indication */
872         int i, err, pci_using_dac;
873         uint16_t eeprom_data = 0;
874         uint16_t eeprom_apme_mask = E1000_EEPROM_APME;
875         if ((err = pci_enable_device(pdev)))
876                 return err;
877
878         if (!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK)) &&
879             !(err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK))) {
880                 pci_using_dac = 1;
881         } else {
882                 if ((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK)) &&
883                     (err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK))) {
884                         E1000_ERR("No usable DMA configuration, aborting\n");
885                         goto err_dma;
886                 }
887                 pci_using_dac = 0;
888         }
889
890         if ((err = pci_request_regions(pdev, e1000_driver_name)))
891                 goto err_pci_reg;
892
893         pci_set_master(pdev);
894
895         err = -ENOMEM;
896         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
897         if (!netdev)
898                 goto err_alloc_etherdev;
899
900         SET_MODULE_OWNER(netdev);
901         SET_NETDEV_DEV(netdev, &pdev->dev);
902
903         pci_set_drvdata(pdev, netdev);
904         adapter = netdev_priv(netdev);
905         adapter->netdev = netdev;
906         adapter->pdev = pdev;
907         adapter->hw.back = adapter;
908         adapter->msg_enable = (1 << debug) - 1;
909
910         mmio_start = pci_resource_start(pdev, BAR_0);
911         mmio_len = pci_resource_len(pdev, BAR_0);
912
913         err = -EIO;
914         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
915         if (!adapter->hw.hw_addr)
916                 goto err_ioremap;
917
918         for (i = BAR_1; i <= BAR_5; i++) {
919                 if (pci_resource_len(pdev, i) == 0)
920                         continue;
921                 if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
922                         adapter->hw.io_base = pci_resource_start(pdev, i);
923                         break;
924                 }
925         }
926
927         netdev->open = &e1000_open;
928         netdev->stop = &e1000_close;
929         netdev->hard_start_xmit = &e1000_xmit_frame;
930         netdev->get_stats = &e1000_get_stats;
931         netdev->set_multicast_list = &e1000_set_multi;
932         netdev->set_mac_address = &e1000_set_mac;
933         netdev->change_mtu = &e1000_change_mtu;
934         netdev->do_ioctl = &e1000_ioctl;
935         e1000_set_ethtool_ops(netdev);
936         netdev->tx_timeout = &e1000_tx_timeout;
937         netdev->watchdog_timeo = 5 * HZ;
938 #ifdef CONFIG_E1000_NAPI
939         netdev->poll = &e1000_clean;
940         netdev->weight = 64;
941 #endif
942         netdev->vlan_rx_register = e1000_vlan_rx_register;
943         netdev->vlan_rx_add_vid = e1000_vlan_rx_add_vid;
944         netdev->vlan_rx_kill_vid = e1000_vlan_rx_kill_vid;
945 #ifdef CONFIG_NET_POLL_CONTROLLER
946         netdev->poll_controller = e1000_netpoll;
947 #endif
948         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
949
950         netdev->mem_start = mmio_start;
951         netdev->mem_end = mmio_start + mmio_len;
952         netdev->base_addr = adapter->hw.io_base;
953
954         adapter->bd_number = cards_found;
955
956         /* setup the private structure */
957
958         if ((err = e1000_sw_init(adapter)))
959                 goto err_sw_init;
960
961         err = -EIO;
962         /* Flash BAR mapping must happen after e1000_sw_init
963          * because it depends on mac_type */
964         if ((adapter->hw.mac_type == e1000_ich8lan) &&
965            (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
966                 flash_start = pci_resource_start(pdev, 1);
967                 flash_len = pci_resource_len(pdev, 1);
968                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
969                 if (!adapter->hw.flash_address)
970                         goto err_flashmap;
971         }
972
973         if (e1000_check_phy_reset_block(&adapter->hw))
974                 DPRINTK(PROBE, INFO, "PHY reset is blocked due to SOL/IDER session.\n");
975
976         if (adapter->hw.mac_type >= e1000_82543) {
977                 netdev->features = NETIF_F_SG |
978                                    NETIF_F_HW_CSUM |
979                                    NETIF_F_HW_VLAN_TX |
980                                    NETIF_F_HW_VLAN_RX |
981                                    NETIF_F_HW_VLAN_FILTER;
982                 if (adapter->hw.mac_type == e1000_ich8lan)
983                         netdev->features &= ~NETIF_F_HW_VLAN_FILTER;
984         }
985
986         if ((adapter->hw.mac_type >= e1000_82544) &&
987            (adapter->hw.mac_type != e1000_82547))
988                 netdev->features |= NETIF_F_TSO;
989
990         if (adapter->hw.mac_type > e1000_82547_rev_2)
991                 netdev->features |= NETIF_F_TSO6;
992         if (pci_using_dac)
993                 netdev->features |= NETIF_F_HIGHDMA;
994
995         netdev->features |= NETIF_F_LLTX;
996
997         adapter->en_mng_pt = e1000_enable_mng_pass_thru(&adapter->hw);
998
999         /* initialize eeprom parameters */
1000
1001         if (e1000_init_eeprom_params(&adapter->hw)) {
1002                 E1000_ERR("EEPROM initialization failed\n");
1003                 goto err_eeprom;
1004         }
1005
1006         /* before reading the EEPROM, reset the controller to
1007          * put the device in a known good starting state */
1008
1009         e1000_reset_hw(&adapter->hw);
1010
1011         /* make sure the EEPROM is good */
1012
1013         if (e1000_validate_eeprom_checksum(&adapter->hw) < 0) {
1014                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
1015                 goto err_eeprom;
1016         }
1017
1018         /* copy the MAC address out of the EEPROM */
1019
1020         if (e1000_read_mac_addr(&adapter->hw))
1021                 DPRINTK(PROBE, ERR, "EEPROM Read Error\n");
1022         memcpy(netdev->dev_addr, adapter->hw.mac_addr, netdev->addr_len);
1023         memcpy(netdev->perm_addr, adapter->hw.mac_addr, netdev->addr_len);
1024
1025         if (!is_valid_ether_addr(netdev->perm_addr)) {
1026                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
1027                 goto err_eeprom;
1028         }
1029
1030         e1000_get_bus_info(&adapter->hw);
1031
1032         init_timer(&adapter->tx_fifo_stall_timer);
1033         adapter->tx_fifo_stall_timer.function = &e1000_82547_tx_fifo_stall;
1034         adapter->tx_fifo_stall_timer.data = (unsigned long) adapter;
1035
1036         init_timer(&adapter->watchdog_timer);
1037         adapter->watchdog_timer.function = &e1000_watchdog;
1038         adapter->watchdog_timer.data = (unsigned long) adapter;
1039
1040         init_timer(&adapter->phy_info_timer);
1041         adapter->phy_info_timer.function = &e1000_update_phy_info;
1042         adapter->phy_info_timer.data = (unsigned long) adapter;
1043
1044         INIT_WORK(&adapter->reset_task, e1000_reset_task);
1045
1046         e1000_check_options(adapter);
1047
1048         /* Initial Wake on LAN setting
1049          * If APM wake is enabled in the EEPROM,
1050          * enable the ACPI Magic Packet filter
1051          */
1052
1053         switch (adapter->hw.mac_type) {
1054         case e1000_82542_rev2_0:
1055         case e1000_82542_rev2_1:
1056         case e1000_82543:
1057                 break;
1058         case e1000_82544:
1059                 e1000_read_eeprom(&adapter->hw,
1060                         EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
1061                 eeprom_apme_mask = E1000_EEPROM_82544_APM;
1062                 break;
1063         case e1000_ich8lan:
1064                 e1000_read_eeprom(&adapter->hw,
1065                         EEPROM_INIT_CONTROL1_REG, 1, &eeprom_data);
1066                 eeprom_apme_mask = E1000_EEPROM_ICH8_APME;
1067                 break;
1068         case e1000_82546:
1069         case e1000_82546_rev_3:
1070         case e1000_82571:
1071         case e1000_80003es2lan:
1072                 if (E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_FUNC_1){
1073                         e1000_read_eeprom(&adapter->hw,
1074                                 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1075                         break;
1076                 }
1077                 /* Fall Through */
1078         default:
1079                 e1000_read_eeprom(&adapter->hw,
1080                         EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1081                 break;
1082         }
1083         if (eeprom_data & eeprom_apme_mask)
1084                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1085
1086         /* now that we have the eeprom settings, apply the special cases
1087          * where the eeprom may be wrong or the board simply won't support
1088          * wake on lan on a particular port */
1089         switch (pdev->device) {
1090         case E1000_DEV_ID_82546GB_PCIE:
1091                 adapter->eeprom_wol = 0;
1092                 break;
1093         case E1000_DEV_ID_82546EB_FIBER:
1094         case E1000_DEV_ID_82546GB_FIBER:
1095         case E1000_DEV_ID_82571EB_FIBER:
1096                 /* Wake events only supported on port A for dual fiber
1097                  * regardless of eeprom setting */
1098                 if (E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_FUNC_1)
1099                         adapter->eeprom_wol = 0;
1100                 break;
1101         case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
1102         case E1000_DEV_ID_82571EB_QUAD_COPPER:
1103         case E1000_DEV_ID_82571EB_QUAD_FIBER:
1104         case E1000_DEV_ID_82571EB_QUAD_COPPER_LOWPROFILE:
1105         case E1000_DEV_ID_82571PT_QUAD_COPPER:
1106                 /* if quad port adapter, disable WoL on all but port A */
1107                 if (global_quad_port_a != 0)
1108                         adapter->eeprom_wol = 0;
1109                 else
1110                         adapter->quad_port_a = 1;
1111                 /* Reset for multiple quad port adapters */
1112                 if (++global_quad_port_a == 4)
1113                         global_quad_port_a = 0;
1114                 break;
1115         }
1116
1117         /* initialize the wol settings based on the eeprom settings */
1118         adapter->wol = adapter->eeprom_wol;
1119
1120         /* print bus type/speed/width info */
1121         {
1122         struct e1000_hw *hw = &adapter->hw;
1123         DPRINTK(PROBE, INFO, "(PCI%s:%s:%s) ",
1124                 ((hw->bus_type == e1000_bus_type_pcix) ? "-X" :
1125                  (hw->bus_type == e1000_bus_type_pci_express ? " Express":"")),
1126                 ((hw->bus_speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
1127                  (hw->bus_speed == e1000_bus_speed_133) ? "133MHz" :
1128                  (hw->bus_speed == e1000_bus_speed_120) ? "120MHz" :
1129                  (hw->bus_speed == e1000_bus_speed_100) ? "100MHz" :
1130                  (hw->bus_speed == e1000_bus_speed_66) ? "66MHz" : "33MHz"),
1131                 ((hw->bus_width == e1000_bus_width_64) ? "64-bit" :
1132                  (hw->bus_width == e1000_bus_width_pciex_4) ? "Width x4" :
1133                  (hw->bus_width == e1000_bus_width_pciex_1) ? "Width x1" :
1134                  "32-bit"));
1135         }
1136
1137         for (i = 0; i < 6; i++)
1138                 printk("%2.2x%c", netdev->dev_addr[i], i == 5 ? '\n' : ':');
1139
1140         /* reset the hardware with the new settings */
1141         e1000_reset(adapter);
1142
1143         /* If the controller is 82573 and f/w is AMT, do not set
1144          * DRV_LOAD until the interface is up.  For all other cases,
1145          * let the f/w know that the h/w is now under the control
1146          * of the driver. */
1147         if (adapter->hw.mac_type != e1000_82573 ||
1148             !e1000_check_mng_mode(&adapter->hw))
1149                 e1000_get_hw_control(adapter);
1150
1151         /* tell the stack to leave us alone until e1000_open() is called */
1152         netif_carrier_off(netdev);
1153         netif_stop_queue(netdev);
1154 #ifdef CONFIG_E1000_NAPI
1155         netif_poll_disable(netdev);
1156 #endif
1157
1158         strcpy(netdev->name, "eth%d");
1159         if ((err = register_netdev(netdev)))
1160                 goto err_register;
1161
1162         DPRINTK(PROBE, INFO, "Intel(R) PRO/1000 Network Connection\n");
1163
1164         cards_found++;
1165         return 0;
1166
1167 err_register:
1168         e1000_release_hw_control(adapter);
1169 err_eeprom:
1170         if (!e1000_check_phy_reset_block(&adapter->hw))
1171                 e1000_phy_hw_reset(&adapter->hw);
1172
1173         if (adapter->hw.flash_address)
1174                 iounmap(adapter->hw.flash_address);
1175 err_flashmap:
1176 #ifdef CONFIG_E1000_NAPI
1177         for (i = 0; i < adapter->num_rx_queues; i++)
1178                 dev_put(&adapter->polling_netdev[i]);
1179 #endif
1180
1181         kfree(adapter->tx_ring);
1182         kfree(adapter->rx_ring);
1183 #ifdef CONFIG_E1000_NAPI
1184         kfree(adapter->polling_netdev);
1185 #endif
1186 err_sw_init:
1187         iounmap(adapter->hw.hw_addr);
1188 err_ioremap:
1189         free_netdev(netdev);
1190 err_alloc_etherdev:
1191         pci_release_regions(pdev);
1192 err_pci_reg:
1193 err_dma:
1194         pci_disable_device(pdev);
1195         return err;
1196 }
1197
1198 /**
1199  * e1000_remove - Device Removal Routine
1200  * @pdev: PCI device information struct
1201  *
1202  * e1000_remove is called by the PCI subsystem to alert the driver
1203  * that it should release a PCI device.  The could be caused by a
1204  * Hot-Plug event, or because the driver is going to be removed from
1205  * memory.
1206  **/
1207
1208 static void __devexit
1209 e1000_remove(struct pci_dev *pdev)
1210 {
1211         struct net_device *netdev = pci_get_drvdata(pdev);
1212         struct e1000_adapter *adapter = netdev_priv(netdev);
1213 #ifdef CONFIG_E1000_NAPI
1214         int i;
1215 #endif
1216
1217         cancel_work_sync(&adapter->reset_task);
1218
1219         e1000_release_manageability(adapter);
1220
1221         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
1222          * would have already happened in close and is redundant. */
1223         e1000_release_hw_control(adapter);
1224
1225         unregister_netdev(netdev);
1226 #ifdef CONFIG_E1000_NAPI
1227         for (i = 0; i < adapter->num_rx_queues; i++)
1228                 dev_put(&adapter->polling_netdev[i]);
1229 #endif
1230
1231         if (!e1000_check_phy_reset_block(&adapter->hw))
1232                 e1000_phy_hw_reset(&adapter->hw);
1233
1234         kfree(adapter->tx_ring);
1235         kfree(adapter->rx_ring);
1236 #ifdef CONFIG_E1000_NAPI
1237         kfree(adapter->polling_netdev);
1238 #endif
1239
1240         iounmap(adapter->hw.hw_addr);
1241         if (adapter->hw.flash_address)
1242                 iounmap(adapter->hw.flash_address);
1243         pci_release_regions(pdev);
1244
1245         free_netdev(netdev);
1246
1247         pci_disable_device(pdev);
1248 }
1249
1250 /**
1251  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1252  * @adapter: board private structure to initialize
1253  *
1254  * e1000_sw_init initializes the Adapter private data structure.
1255  * Fields are initialized based on PCI device information and
1256  * OS network device settings (MTU size).
1257  **/
1258
1259 static int __devinit
1260 e1000_sw_init(struct e1000_adapter *adapter)
1261 {
1262         struct e1000_hw *hw = &adapter->hw;
1263         struct net_device *netdev = adapter->netdev;
1264         struct pci_dev *pdev = adapter->pdev;
1265 #ifdef CONFIG_E1000_NAPI
1266         int i;
1267 #endif
1268
1269         /* PCI config space info */
1270
1271         hw->vendor_id = pdev->vendor;
1272         hw->device_id = pdev->device;
1273         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1274         hw->subsystem_id = pdev->subsystem_device;
1275         hw->revision_id = pdev->revision;
1276
1277         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
1278
1279         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1280         adapter->rx_ps_bsize0 = E1000_RXBUFFER_128;
1281         hw->max_frame_size = netdev->mtu +
1282                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
1283         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
1284
1285         /* identify the MAC */
1286
1287         if (e1000_set_mac_type(hw)) {
1288                 DPRINTK(PROBE, ERR, "Unknown MAC Type\n");
1289                 return -EIO;
1290         }
1291
1292         switch (hw->mac_type) {
1293         default:
1294                 break;
1295         case e1000_82541:
1296         case e1000_82547:
1297         case e1000_82541_rev_2:
1298         case e1000_82547_rev_2:
1299                 hw->phy_init_script = 1;
1300                 break;
1301         }
1302
1303         e1000_set_media_type(hw);
1304
1305         hw->wait_autoneg_complete = FALSE;
1306         hw->tbi_compatibility_en = TRUE;
1307         hw->adaptive_ifs = TRUE;
1308
1309         /* Copper options */
1310
1311         if (hw->media_type == e1000_media_type_copper) {
1312                 hw->mdix = AUTO_ALL_MODES;
1313                 hw->disable_polarity_correction = FALSE;
1314                 hw->master_slave = E1000_MASTER_SLAVE;
1315         }
1316
1317         adapter->num_tx_queues = 1;
1318         adapter->num_rx_queues = 1;
1319
1320         if (e1000_alloc_queues(adapter)) {
1321                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
1322                 return -ENOMEM;
1323         }
1324
1325 #ifdef CONFIG_E1000_NAPI
1326         for (i = 0; i < adapter->num_rx_queues; i++) {
1327                 adapter->polling_netdev[i].priv = adapter;
1328                 adapter->polling_netdev[i].poll = &e1000_clean;
1329                 adapter->polling_netdev[i].weight = 64;
1330                 dev_hold(&adapter->polling_netdev[i]);
1331                 set_bit(__LINK_STATE_START, &adapter->polling_netdev[i].state);
1332         }
1333         spin_lock_init(&adapter->tx_queue_lock);
1334 #endif
1335
1336         /* Explicitly disable IRQ since the NIC can be in any state. */
1337         atomic_set(&adapter->irq_sem, 0);
1338         e1000_irq_disable(adapter);
1339
1340         spin_lock_init(&adapter->stats_lock);
1341
1342         set_bit(__E1000_DOWN, &adapter->flags);
1343
1344         return 0;
1345 }
1346
1347 /**
1348  * e1000_alloc_queues - Allocate memory for all rings
1349  * @adapter: board private structure to initialize
1350  *
1351  * We allocate one ring per queue at run-time since we don't know the
1352  * number of queues at compile-time.  The polling_netdev array is
1353  * intended for Multiqueue, but should work fine with a single queue.
1354  **/
1355
1356 static int __devinit
1357 e1000_alloc_queues(struct e1000_adapter *adapter)
1358 {
1359         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1360                                    sizeof(struct e1000_tx_ring), GFP_KERNEL);
1361         if (!adapter->tx_ring)
1362                 return -ENOMEM;
1363
1364         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1365                                    sizeof(struct e1000_rx_ring), GFP_KERNEL);
1366         if (!adapter->rx_ring) {
1367                 kfree(adapter->tx_ring);
1368                 return -ENOMEM;
1369         }
1370
1371 #ifdef CONFIG_E1000_NAPI
1372         adapter->polling_netdev = kcalloc(adapter->num_rx_queues,
1373                                           sizeof(struct net_device),
1374                                           GFP_KERNEL);
1375         if (!adapter->polling_netdev) {
1376                 kfree(adapter->tx_ring);
1377                 kfree(adapter->rx_ring);
1378                 return -ENOMEM;
1379         }
1380 #endif
1381
1382         return E1000_SUCCESS;
1383 }
1384
1385 /**
1386  * e1000_open - Called when a network interface is made active
1387  * @netdev: network interface device structure
1388  *
1389  * Returns 0 on success, negative value on failure
1390  *
1391  * The open entry point is called when a network interface is made
1392  * active by the system (IFF_UP).  At this point all resources needed
1393  * for transmit and receive operations are allocated, the interrupt
1394  * handler is registered with the OS, the watchdog timer is started,
1395  * and the stack is notified that the interface is ready.
1396  **/
1397
1398 static int
1399 e1000_open(struct net_device *netdev)
1400 {
1401         struct e1000_adapter *adapter = netdev_priv(netdev);
1402         int err;
1403
1404         /* disallow open during test */
1405         if (test_bit(__E1000_TESTING, &adapter->flags))
1406                 return -EBUSY;
1407
1408         /* allocate transmit descriptors */
1409         err = e1000_setup_all_tx_resources(adapter);
1410         if (err)
1411                 goto err_setup_tx;
1412
1413         /* allocate receive descriptors */
1414         err = e1000_setup_all_rx_resources(adapter);
1415         if (err)
1416                 goto err_setup_rx;
1417
1418         e1000_power_up_phy(adapter);
1419
1420         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1421         if ((adapter->hw.mng_cookie.status &
1422                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1423                 e1000_update_mng_vlan(adapter);
1424         }
1425
1426         /* If AMT is enabled, let the firmware know that the network
1427          * interface is now open */
1428         if (adapter->hw.mac_type == e1000_82573 &&
1429             e1000_check_mng_mode(&adapter->hw))
1430                 e1000_get_hw_control(adapter);
1431
1432         /* before we allocate an interrupt, we must be ready to handle it.
1433          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1434          * as soon as we call pci_request_irq, so we have to setup our
1435          * clean_rx handler before we do so.  */
1436         e1000_configure(adapter);
1437
1438         err = e1000_request_irq(adapter);
1439         if (err)
1440                 goto err_req_irq;
1441
1442         /* From here on the code is the same as e1000_up() */
1443         clear_bit(__E1000_DOWN, &adapter->flags);
1444
1445 #ifdef CONFIG_E1000_NAPI
1446         netif_poll_enable(netdev);
1447 #endif
1448
1449         e1000_irq_enable(adapter);
1450
1451         /* fire a link status change interrupt to start the watchdog */
1452         E1000_WRITE_REG(&adapter->hw, ICS, E1000_ICS_LSC);
1453
1454         return E1000_SUCCESS;
1455
1456 err_req_irq:
1457         e1000_release_hw_control(adapter);
1458         e1000_power_down_phy(adapter);
1459         e1000_free_all_rx_resources(adapter);
1460 err_setup_rx:
1461         e1000_free_all_tx_resources(adapter);
1462 err_setup_tx:
1463         e1000_reset(adapter);
1464
1465         return err;
1466 }
1467
1468 /**
1469  * e1000_close - Disables a network interface
1470  * @netdev: network interface device structure
1471  *
1472  * Returns 0, this is not allowed to fail
1473  *
1474  * The close entry point is called when an interface is de-activated
1475  * by the OS.  The hardware is still under the drivers control, but
1476  * needs to be disabled.  A global MAC reset is issued to stop the
1477  * hardware, and all transmit and receive resources are freed.
1478  **/
1479
1480 static int
1481 e1000_close(struct net_device *netdev)
1482 {
1483         struct e1000_adapter *adapter = netdev_priv(netdev);
1484
1485         WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1486         e1000_down(adapter);
1487         e1000_power_down_phy(adapter);
1488         e1000_free_irq(adapter);
1489
1490         e1000_free_all_tx_resources(adapter);
1491         e1000_free_all_rx_resources(adapter);
1492
1493         /* kill manageability vlan ID if supported, but not if a vlan with
1494          * the same ID is registered on the host OS (let 8021q kill it) */
1495         if ((adapter->hw.mng_cookie.status &
1496                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1497              !(adapter->vlgrp &&
1498                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id))) {
1499                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1500         }
1501
1502         /* If AMT is enabled, let the firmware know that the network
1503          * interface is now closed */
1504         if (adapter->hw.mac_type == e1000_82573 &&
1505             e1000_check_mng_mode(&adapter->hw))
1506                 e1000_release_hw_control(adapter);
1507
1508         return 0;
1509 }
1510
1511 /**
1512  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1513  * @adapter: address of board private structure
1514  * @start: address of beginning of memory
1515  * @len: length of memory
1516  **/
1517 static boolean_t
1518 e1000_check_64k_bound(struct e1000_adapter *adapter,
1519                       void *start, unsigned long len)
1520 {
1521         unsigned long begin = (unsigned long) start;
1522         unsigned long end = begin + len;
1523
1524         /* First rev 82545 and 82546 need to not allow any memory
1525          * write location to cross 64k boundary due to errata 23 */
1526         if (adapter->hw.mac_type == e1000_82545 ||
1527             adapter->hw.mac_type == e1000_82546) {
1528                 return ((begin ^ (end - 1)) >> 16) != 0 ? FALSE : TRUE;
1529         }
1530
1531         return TRUE;
1532 }
1533
1534 /**
1535  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1536  * @adapter: board private structure
1537  * @txdr:    tx descriptor ring (for a specific queue) to setup
1538  *
1539  * Return 0 on success, negative on failure
1540  **/
1541
1542 static int
1543 e1000_setup_tx_resources(struct e1000_adapter *adapter,
1544                          struct e1000_tx_ring *txdr)
1545 {
1546         struct pci_dev *pdev = adapter->pdev;
1547         int size;
1548
1549         size = sizeof(struct e1000_buffer) * txdr->count;
1550         txdr->buffer_info = vmalloc(size);
1551         if (!txdr->buffer_info) {
1552                 DPRINTK(PROBE, ERR,
1553                 "Unable to allocate memory for the transmit descriptor ring\n");
1554                 return -ENOMEM;
1555         }
1556         memset(txdr->buffer_info, 0, size);
1557
1558         /* round up to nearest 4K */
1559
1560         txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1561         txdr->size = ALIGN(txdr->size, 4096);
1562
1563         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1564         if (!txdr->desc) {
1565 setup_tx_desc_die:
1566                 vfree(txdr->buffer_info);
1567                 DPRINTK(PROBE, ERR,
1568                 "Unable to allocate memory for the transmit descriptor ring\n");
1569                 return -ENOMEM;
1570         }
1571
1572         /* Fix for errata 23, can't cross 64kB boundary */
1573         if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1574                 void *olddesc = txdr->desc;
1575                 dma_addr_t olddma = txdr->dma;
1576                 DPRINTK(TX_ERR, ERR, "txdr align check failed: %u bytes "
1577                                      "at %p\n", txdr->size, txdr->desc);
1578                 /* Try again, without freeing the previous */
1579                 txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1580                 /* Failed allocation, critical failure */
1581                 if (!txdr->desc) {
1582                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1583                         goto setup_tx_desc_die;
1584                 }
1585
1586                 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1587                         /* give up */
1588                         pci_free_consistent(pdev, txdr->size, txdr->desc,
1589                                             txdr->dma);
1590                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1591                         DPRINTK(PROBE, ERR,
1592                                 "Unable to allocate aligned memory "
1593                                 "for the transmit descriptor ring\n");
1594                         vfree(txdr->buffer_info);
1595                         return -ENOMEM;
1596                 } else {
1597                         /* Free old allocation, new allocation was successful */
1598                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1599                 }
1600         }
1601         memset(txdr->desc, 0, txdr->size);
1602
1603         txdr->next_to_use = 0;
1604         txdr->next_to_clean = 0;
1605         spin_lock_init(&txdr->tx_lock);
1606
1607         return 0;
1608 }
1609
1610 /**
1611  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1612  *                                (Descriptors) for all queues
1613  * @adapter: board private structure
1614  *
1615  * Return 0 on success, negative on failure
1616  **/
1617
1618 int
1619 e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1620 {
1621         int i, err = 0;
1622
1623         for (i = 0; i < adapter->num_tx_queues; i++) {
1624                 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1625                 if (err) {
1626                         DPRINTK(PROBE, ERR,
1627                                 "Allocation for Tx Queue %u failed\n", i);
1628                         for (i-- ; i >= 0; i--)
1629                                 e1000_free_tx_resources(adapter,
1630                                                         &adapter->tx_ring[i]);
1631                         break;
1632                 }
1633         }
1634
1635         return err;
1636 }
1637
1638 /**
1639  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1640  * @adapter: board private structure
1641  *
1642  * Configure the Tx unit of the MAC after a reset.
1643  **/
1644
1645 static void
1646 e1000_configure_tx(struct e1000_adapter *adapter)
1647 {
1648         uint64_t tdba;
1649         struct e1000_hw *hw = &adapter->hw;
1650         uint32_t tdlen, tctl, tipg, tarc;
1651         uint32_t ipgr1, ipgr2;
1652
1653         /* Setup the HW Tx Head and Tail descriptor pointers */
1654
1655         switch (adapter->num_tx_queues) {
1656         case 1:
1657         default:
1658                 tdba = adapter->tx_ring[0].dma;
1659                 tdlen = adapter->tx_ring[0].count *
1660                         sizeof(struct e1000_tx_desc);
1661                 E1000_WRITE_REG(hw, TDLEN, tdlen);
1662                 E1000_WRITE_REG(hw, TDBAH, (tdba >> 32));
1663                 E1000_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
1664                 E1000_WRITE_REG(hw, TDT, 0);
1665                 E1000_WRITE_REG(hw, TDH, 0);
1666                 adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ? E1000_TDH : E1000_82542_TDH);
1667                 adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ? E1000_TDT : E1000_82542_TDT);
1668                 break;
1669         }
1670
1671         /* Set the default values for the Tx Inter Packet Gap timer */
1672         if (adapter->hw.mac_type <= e1000_82547_rev_2 &&
1673             (hw->media_type == e1000_media_type_fiber ||
1674              hw->media_type == e1000_media_type_internal_serdes))
1675                 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1676         else
1677                 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1678
1679         switch (hw->mac_type) {
1680         case e1000_82542_rev2_0:
1681         case e1000_82542_rev2_1:
1682                 tipg = DEFAULT_82542_TIPG_IPGT;
1683                 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1684                 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1685                 break;
1686         case e1000_80003es2lan:
1687                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1688                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2;
1689                 break;
1690         default:
1691                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1692                 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1693                 break;
1694         }
1695         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1696         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1697         E1000_WRITE_REG(hw, TIPG, tipg);
1698
1699         /* Set the Tx Interrupt Delay register */
1700
1701         E1000_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
1702         if (hw->mac_type >= e1000_82540)
1703                 E1000_WRITE_REG(hw, TADV, adapter->tx_abs_int_delay);
1704
1705         /* Program the Transmit Control Register */
1706
1707         tctl = E1000_READ_REG(hw, TCTL);
1708         tctl &= ~E1000_TCTL_CT;
1709         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1710                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1711
1712         if (hw->mac_type == e1000_82571 || hw->mac_type == e1000_82572) {
1713                 tarc = E1000_READ_REG(hw, TARC0);
1714                 /* set the speed mode bit, we'll clear it if we're not at
1715                  * gigabit link later */
1716                 tarc |= (1 << 21);
1717                 E1000_WRITE_REG(hw, TARC0, tarc);
1718         } else if (hw->mac_type == e1000_80003es2lan) {
1719                 tarc = E1000_READ_REG(hw, TARC0);
1720                 tarc |= 1;
1721                 E1000_WRITE_REG(hw, TARC0, tarc);
1722                 tarc = E1000_READ_REG(hw, TARC1);
1723                 tarc |= 1;
1724                 E1000_WRITE_REG(hw, TARC1, tarc);
1725         }
1726
1727         e1000_config_collision_dist(hw);
1728
1729         /* Setup Transmit Descriptor Settings for eop descriptor */
1730         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1731
1732         /* only set IDE if we are delaying interrupts using the timers */
1733         if (adapter->tx_int_delay)
1734                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1735
1736         if (hw->mac_type < e1000_82543)
1737                 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1738         else
1739                 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1740
1741         /* Cache if we're 82544 running in PCI-X because we'll
1742          * need this to apply a workaround later in the send path. */
1743         if (hw->mac_type == e1000_82544 &&
1744             hw->bus_type == e1000_bus_type_pcix)
1745                 adapter->pcix_82544 = 1;
1746
1747         E1000_WRITE_REG(hw, TCTL, tctl);
1748
1749 }
1750
1751 /**
1752  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1753  * @adapter: board private structure
1754  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1755  *
1756  * Returns 0 on success, negative on failure
1757  **/
1758
1759 static int
1760 e1000_setup_rx_resources(struct e1000_adapter *adapter,
1761                          struct e1000_rx_ring *rxdr)
1762 {
1763         struct pci_dev *pdev = adapter->pdev;
1764         int size, desc_len;
1765
1766         size = sizeof(struct e1000_buffer) * rxdr->count;
1767         rxdr->buffer_info = vmalloc(size);
1768         if (!rxdr->buffer_info) {
1769                 DPRINTK(PROBE, ERR,
1770                 "Unable to allocate memory for the receive descriptor ring\n");
1771                 return -ENOMEM;
1772         }
1773         memset(rxdr->buffer_info, 0, size);
1774
1775         rxdr->ps_page = kcalloc(rxdr->count, sizeof(struct e1000_ps_page),
1776                                 GFP_KERNEL);
1777         if (!rxdr->ps_page) {
1778                 vfree(rxdr->buffer_info);
1779                 DPRINTK(PROBE, ERR,
1780                 "Unable to allocate memory for the receive descriptor ring\n");
1781                 return -ENOMEM;
1782         }
1783
1784         rxdr->ps_page_dma = kcalloc(rxdr->count,
1785                                     sizeof(struct e1000_ps_page_dma),
1786                                     GFP_KERNEL);
1787         if (!rxdr->ps_page_dma) {
1788                 vfree(rxdr->buffer_info);
1789                 kfree(rxdr->ps_page);
1790                 DPRINTK(PROBE, ERR,
1791                 "Unable to allocate memory for the receive descriptor ring\n");
1792                 return -ENOMEM;
1793         }
1794
1795         if (adapter->hw.mac_type <= e1000_82547_rev_2)
1796                 desc_len = sizeof(struct e1000_rx_desc);
1797         else
1798                 desc_len = sizeof(union e1000_rx_desc_packet_split);
1799
1800         /* Round up to nearest 4K */
1801
1802         rxdr->size = rxdr->count * desc_len;
1803         rxdr->size = ALIGN(rxdr->size, 4096);
1804
1805         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1806
1807         if (!rxdr->desc) {
1808                 DPRINTK(PROBE, ERR,
1809                 "Unable to allocate memory for the receive descriptor ring\n");
1810 setup_rx_desc_die:
1811                 vfree(rxdr->buffer_info);
1812                 kfree(rxdr->ps_page);
1813                 kfree(rxdr->ps_page_dma);
1814                 return -ENOMEM;
1815         }
1816
1817         /* Fix for errata 23, can't cross 64kB boundary */
1818         if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1819                 void *olddesc = rxdr->desc;
1820                 dma_addr_t olddma = rxdr->dma;
1821                 DPRINTK(RX_ERR, ERR, "rxdr align check failed: %u bytes "
1822                                      "at %p\n", rxdr->size, rxdr->desc);
1823                 /* Try again, without freeing the previous */
1824                 rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1825                 /* Failed allocation, critical failure */
1826                 if (!rxdr->desc) {
1827                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1828                         DPRINTK(PROBE, ERR,
1829                                 "Unable to allocate memory "
1830                                 "for the receive descriptor ring\n");
1831                         goto setup_rx_desc_die;
1832                 }
1833
1834                 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1835                         /* give up */
1836                         pci_free_consistent(pdev, rxdr->size, rxdr->desc,
1837                                             rxdr->dma);
1838                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1839                         DPRINTK(PROBE, ERR,
1840                                 "Unable to allocate aligned memory "
1841                                 "for the receive descriptor ring\n");
1842                         goto setup_rx_desc_die;
1843                 } else {
1844                         /* Free old allocation, new allocation was successful */
1845                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1846                 }
1847         }
1848         memset(rxdr->desc, 0, rxdr->size);
1849
1850         rxdr->next_to_clean = 0;
1851         rxdr->next_to_use = 0;
1852
1853         return 0;
1854 }
1855
1856 /**
1857  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1858  *                                (Descriptors) for all queues
1859  * @adapter: board private structure
1860  *
1861  * Return 0 on success, negative on failure
1862  **/
1863
1864 int
1865 e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1866 {
1867         int i, err = 0;
1868
1869         for (i = 0; i < adapter->num_rx_queues; i++) {
1870                 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1871                 if (err) {
1872                         DPRINTK(PROBE, ERR,
1873                                 "Allocation for Rx Queue %u failed\n", i);
1874                         for (i-- ; i >= 0; i--)
1875                                 e1000_free_rx_resources(adapter,
1876                                                         &adapter->rx_ring[i]);
1877                         break;
1878                 }
1879         }
1880
1881         return err;
1882 }
1883
1884 /**
1885  * e1000_setup_rctl - configure the receive control registers
1886  * @adapter: Board private structure
1887  **/
1888 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1889                         (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1890 static void
1891 e1000_setup_rctl(struct e1000_adapter *adapter)
1892 {
1893         uint32_t rctl, rfctl;
1894         uint32_t psrctl = 0;
1895 #ifndef CONFIG_E1000_DISABLE_PACKET_SPLIT
1896         uint32_t pages = 0;
1897 #endif
1898
1899         rctl = E1000_READ_REG(&adapter->hw, RCTL);
1900
1901         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1902
1903         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1904                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1905                 (adapter->hw.mc_filter_type << E1000_RCTL_MO_SHIFT);
1906
1907         if (adapter->hw.tbi_compatibility_on == 1)
1908                 rctl |= E1000_RCTL_SBP;
1909         else
1910                 rctl &= ~E1000_RCTL_SBP;
1911
1912         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1913                 rctl &= ~E1000_RCTL_LPE;
1914         else
1915                 rctl |= E1000_RCTL_LPE;
1916
1917         /* Setup buffer sizes */
1918         rctl &= ~E1000_RCTL_SZ_4096;
1919         rctl |= E1000_RCTL_BSEX;
1920         switch (adapter->rx_buffer_len) {
1921                 case E1000_RXBUFFER_256:
1922                         rctl |= E1000_RCTL_SZ_256;
1923                         rctl &= ~E1000_RCTL_BSEX;
1924                         break;
1925                 case E1000_RXBUFFER_512:
1926                         rctl |= E1000_RCTL_SZ_512;
1927                         rctl &= ~E1000_RCTL_BSEX;
1928                         break;
1929                 case E1000_RXBUFFER_1024:
1930                         rctl |= E1000_RCTL_SZ_1024;
1931                         rctl &= ~E1000_RCTL_BSEX;
1932                         break;
1933                 case E1000_RXBUFFER_2048:
1934                 default:
1935                         rctl |= E1000_RCTL_SZ_2048;
1936                         rctl &= ~E1000_RCTL_BSEX;
1937                         break;
1938                 case E1000_RXBUFFER_4096:
1939                         rctl |= E1000_RCTL_SZ_4096;
1940                         break;
1941                 case E1000_RXBUFFER_8192:
1942                         rctl |= E1000_RCTL_SZ_8192;
1943                         break;
1944                 case E1000_RXBUFFER_16384:
1945                         rctl |= E1000_RCTL_SZ_16384;
1946                         break;
1947         }
1948
1949 #ifndef CONFIG_E1000_DISABLE_PACKET_SPLIT
1950         /* 82571 and greater support packet-split where the protocol
1951          * header is placed in skb->data and the packet data is
1952          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
1953          * In the case of a non-split, skb->data is linearly filled,
1954          * followed by the page buffers.  Therefore, skb->data is
1955          * sized to hold the largest protocol header.
1956          */
1957         /* allocations using alloc_page take too long for regular MTU
1958          * so only enable packet split for jumbo frames */
1959         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
1960         if ((adapter->hw.mac_type >= e1000_82571) && (pages <= 3) &&
1961             PAGE_SIZE <= 16384 && (rctl & E1000_RCTL_LPE))
1962                 adapter->rx_ps_pages = pages;
1963         else
1964                 adapter->rx_ps_pages = 0;
1965 #endif
1966         if (adapter->rx_ps_pages) {
1967                 /* Configure extra packet-split registers */
1968                 rfctl = E1000_READ_REG(&adapter->hw, RFCTL);
1969                 rfctl |= E1000_RFCTL_EXTEN;
1970                 /* disable packet split support for IPv6 extension headers,
1971                  * because some malformed IPv6 headers can hang the RX */
1972                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
1973                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
1974
1975                 E1000_WRITE_REG(&adapter->hw, RFCTL, rfctl);
1976
1977                 rctl |= E1000_RCTL_DTYP_PS;
1978
1979                 psrctl |= adapter->rx_ps_bsize0 >>
1980                         E1000_PSRCTL_BSIZE0_SHIFT;
1981
1982                 switch (adapter->rx_ps_pages) {
1983                 case 3:
1984                         psrctl |= PAGE_SIZE <<
1985                                 E1000_PSRCTL_BSIZE3_SHIFT;
1986                 case 2:
1987                         psrctl |= PAGE_SIZE <<
1988                                 E1000_PSRCTL_BSIZE2_SHIFT;
1989                 case 1:
1990                         psrctl |= PAGE_SIZE >>
1991                                 E1000_PSRCTL_BSIZE1_SHIFT;
1992                         break;
1993                 }
1994
1995                 E1000_WRITE_REG(&adapter->hw, PSRCTL, psrctl);
1996         }
1997
1998         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
1999 }
2000
2001 /**
2002  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
2003  * @adapter: board private structure
2004  *
2005  * Configure the Rx unit of the MAC after a reset.
2006  **/
2007
2008 static void
2009 e1000_configure_rx(struct e1000_adapter *adapter)
2010 {
2011         uint64_t rdba;
2012         struct e1000_hw *hw = &adapter->hw;
2013         uint32_t rdlen, rctl, rxcsum, ctrl_ext;
2014
2015         if (adapter->rx_ps_pages) {
2016                 /* this is a 32 byte descriptor */
2017                 rdlen = adapter->rx_ring[0].count *
2018                         sizeof(union e1000_rx_desc_packet_split);
2019                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2020                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2021         } else {
2022                 rdlen = adapter->rx_ring[0].count *
2023                         sizeof(struct e1000_rx_desc);
2024                 adapter->clean_rx = e1000_clean_rx_irq;
2025                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2026         }
2027
2028         /* disable receives while setting up the descriptors */
2029         rctl = E1000_READ_REG(hw, RCTL);
2030         E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
2031
2032         /* set the Receive Delay Timer Register */
2033         E1000_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
2034
2035         if (hw->mac_type >= e1000_82540) {
2036                 E1000_WRITE_REG(hw, RADV, adapter->rx_abs_int_delay);
2037                 if (adapter->itr_setting != 0)
2038                         E1000_WRITE_REG(hw, ITR,
2039                                 1000000000 / (adapter->itr * 256));
2040         }
2041
2042         if (hw->mac_type >= e1000_82571) {
2043                 ctrl_ext = E1000_READ_REG(hw, CTRL_EXT);
2044                 /* Reset delay timers after every interrupt */
2045                 ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2046 #ifdef CONFIG_E1000_NAPI
2047                 /* Auto-Mask interrupts upon ICR access */
2048                 ctrl_ext |= E1000_CTRL_EXT_IAME;
2049                 E1000_WRITE_REG(hw, IAM, 0xffffffff);
2050 #endif
2051                 E1000_WRITE_REG(hw, CTRL_EXT, ctrl_ext);
2052                 E1000_WRITE_FLUSH(hw);
2053         }
2054
2055         /* Setup the HW Rx Head and Tail Descriptor Pointers and
2056          * the Base and Length of the Rx Descriptor Ring */
2057         switch (adapter->num_rx_queues) {
2058         case 1:
2059         default:
2060                 rdba = adapter->rx_ring[0].dma;
2061                 E1000_WRITE_REG(hw, RDLEN, rdlen);
2062                 E1000_WRITE_REG(hw, RDBAH, (rdba >> 32));
2063                 E1000_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
2064                 E1000_WRITE_REG(hw, RDT, 0);
2065                 E1000_WRITE_REG(hw, RDH, 0);
2066                 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ? E1000_RDH : E1000_82542_RDH);
2067                 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ? E1000_RDT : E1000_82542_RDT);
2068                 break;
2069         }
2070
2071         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
2072         if (hw->mac_type >= e1000_82543) {
2073                 rxcsum = E1000_READ_REG(hw, RXCSUM);
2074                 if (adapter->rx_csum == TRUE) {
2075                         rxcsum |= E1000_RXCSUM_TUOFL;
2076
2077                         /* Enable 82571 IPv4 payload checksum for UDP fragments
2078                          * Must be used in conjunction with packet-split. */
2079                         if ((hw->mac_type >= e1000_82571) &&
2080                             (adapter->rx_ps_pages)) {
2081                                 rxcsum |= E1000_RXCSUM_IPPCSE;
2082                         }
2083                 } else {
2084                         rxcsum &= ~E1000_RXCSUM_TUOFL;
2085                         /* don't need to clear IPPCSE as it defaults to 0 */
2086                 }
2087                 E1000_WRITE_REG(hw, RXCSUM, rxcsum);
2088         }
2089
2090         /* enable early receives on 82573, only takes effect if using > 2048
2091          * byte total frame size.  for example only for jumbo frames */
2092 #define E1000_ERT_2048 0x100
2093         if (hw->mac_type == e1000_82573)
2094                 E1000_WRITE_REG(hw, ERT, E1000_ERT_2048);
2095
2096         /* Enable Receives */
2097         E1000_WRITE_REG(hw, RCTL, rctl);
2098 }
2099
2100 /**
2101  * e1000_free_tx_resources - Free Tx Resources per Queue
2102  * @adapter: board private structure
2103  * @tx_ring: Tx descriptor ring for a specific queue
2104  *
2105  * Free all transmit software resources
2106  **/
2107
2108 static void
2109 e1000_free_tx_resources(struct e1000_adapter *adapter,
2110                         struct e1000_tx_ring *tx_ring)
2111 {
2112         struct pci_dev *pdev = adapter->pdev;
2113
2114         e1000_clean_tx_ring(adapter, tx_ring);
2115
2116         vfree(tx_ring->buffer_info);
2117         tx_ring->buffer_info = NULL;
2118
2119         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
2120
2121         tx_ring->desc = NULL;
2122 }
2123
2124 /**
2125  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
2126  * @adapter: board private structure
2127  *
2128  * Free all transmit software resources
2129  **/
2130
2131 void
2132 e1000_free_all_tx_resources(struct e1000_adapter *adapter)
2133 {
2134         int i;
2135
2136         for (i = 0; i < adapter->num_tx_queues; i++)
2137                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
2138 }
2139
2140 static void
2141 e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
2142                         struct e1000_buffer *buffer_info)
2143 {
2144         if (buffer_info->dma) {
2145                 pci_unmap_page(adapter->pdev,
2146                                 buffer_info->dma,
2147                                 buffer_info->length,
2148                                 PCI_DMA_TODEVICE);
2149                 buffer_info->dma = 0;
2150         }
2151         if (buffer_info->skb) {
2152                 dev_kfree_skb_any(buffer_info->skb);
2153                 buffer_info->skb = NULL;
2154         }
2155         /* buffer_info must be completely set up in the transmit path */
2156 }
2157
2158 /**
2159  * e1000_clean_tx_ring - Free Tx Buffers
2160  * @adapter: board private structure
2161  * @tx_ring: ring to be cleaned
2162  **/
2163
2164 static void
2165 e1000_clean_tx_ring(struct e1000_adapter *adapter,
2166                     struct e1000_tx_ring *tx_ring)
2167 {
2168         struct e1000_buffer *buffer_info;
2169         unsigned long size;
2170         unsigned int i;
2171
2172         /* Free all the Tx ring sk_buffs */
2173
2174         for (i = 0; i < tx_ring->count; i++) {
2175                 buffer_info = &tx_ring->buffer_info[i];
2176                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2177         }
2178
2179         size = sizeof(struct e1000_buffer) * tx_ring->count;
2180         memset(tx_ring->buffer_info, 0, size);
2181
2182         /* Zero out the descriptor ring */
2183
2184         memset(tx_ring->desc, 0, tx_ring->size);
2185
2186         tx_ring->next_to_use = 0;
2187         tx_ring->next_to_clean = 0;
2188         tx_ring->last_tx_tso = 0;
2189
2190         writel(0, adapter->hw.hw_addr + tx_ring->tdh);
2191         writel(0, adapter->hw.hw_addr + tx_ring->tdt);
2192 }
2193
2194 /**
2195  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
2196  * @adapter: board private structure
2197  **/
2198
2199 static void
2200 e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
2201 {
2202         int i;
2203
2204         for (i = 0; i < adapter->num_tx_queues; i++)
2205                 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2206 }
2207
2208 /**
2209  * e1000_free_rx_resources - Free Rx Resources
2210  * @adapter: board private structure
2211  * @rx_ring: ring to clean the resources from
2212  *
2213  * Free all receive software resources
2214  **/
2215
2216 static void
2217 e1000_free_rx_resources(struct e1000_adapter *adapter,
2218                         struct e1000_rx_ring *rx_ring)
2219 {
2220         struct pci_dev *pdev = adapter->pdev;
2221
2222         e1000_clean_rx_ring(adapter, rx_ring);
2223
2224         vfree(rx_ring->buffer_info);
2225         rx_ring->buffer_info = NULL;
2226         kfree(rx_ring->ps_page);
2227         rx_ring->ps_page = NULL;
2228         kfree(rx_ring->ps_page_dma);
2229         rx_ring->ps_page_dma = NULL;
2230
2231         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2232
2233         rx_ring->desc = NULL;
2234 }
2235
2236 /**
2237  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
2238  * @adapter: board private structure
2239  *
2240  * Free all receive software resources
2241  **/
2242
2243 void
2244 e1000_free_all_rx_resources(struct e1000_adapter *adapter)
2245 {
2246         int i;
2247
2248         for (i = 0; i < adapter->num_rx_queues; i++)
2249                 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
2250 }
2251
2252 /**
2253  * e1000_clean_rx_ring - Free Rx Buffers per Queue
2254  * @adapter: board private structure
2255  * @rx_ring: ring to free buffers from
2256  **/
2257
2258 static void
2259 e1000_clean_rx_ring(struct e1000_adapter *adapter,
2260                     struct e1000_rx_ring *rx_ring)
2261 {
2262         struct e1000_buffer *buffer_info;
2263         struct e1000_ps_page *ps_page;
2264         struct e1000_ps_page_dma *ps_page_dma;
2265         struct pci_dev *pdev = adapter->pdev;
2266         unsigned long size;
2267         unsigned int i, j;
2268
2269         /* Free all the Rx ring sk_buffs */
2270         for (i = 0; i < rx_ring->count; i++) {
2271                 buffer_info = &rx_ring->buffer_info[i];
2272                 if (buffer_info->skb) {
2273                         pci_unmap_single(pdev,
2274                                          buffer_info->dma,
2275                                          buffer_info->length,
2276                                          PCI_DMA_FROMDEVICE);
2277
2278                         dev_kfree_skb(buffer_info->skb);
2279                         buffer_info->skb = NULL;
2280                 }
2281                 ps_page = &rx_ring->ps_page[i];
2282                 ps_page_dma = &rx_ring->ps_page_dma[i];
2283                 for (j = 0; j < adapter->rx_ps_pages; j++) {
2284                         if (!ps_page->ps_page[j]) break;
2285                         pci_unmap_page(pdev,
2286                                        ps_page_dma->ps_page_dma[j],
2287                                        PAGE_SIZE, PCI_DMA_FROMDEVICE);
2288                         ps_page_dma->ps_page_dma[j] = 0;
2289                         put_page(ps_page->ps_page[j]);
2290                         ps_page->ps_page[j] = NULL;
2291                 }
2292         }
2293
2294         size = sizeof(struct e1000_buffer) * rx_ring->count;
2295         memset(rx_ring->buffer_info, 0, size);
2296         size = sizeof(struct e1000_ps_page) * rx_ring->count;
2297         memset(rx_ring->ps_page, 0, size);
2298         size = sizeof(struct e1000_ps_page_dma) * rx_ring->count;
2299         memset(rx_ring->ps_page_dma, 0, size);
2300
2301         /* Zero out the descriptor ring */
2302
2303         memset(rx_ring->desc, 0, rx_ring->size);
2304
2305         rx_ring->next_to_clean = 0;
2306         rx_ring->next_to_use = 0;
2307
2308         writel(0, adapter->hw.hw_addr + rx_ring->rdh);
2309         writel(0, adapter->hw.hw_addr + rx_ring->rdt);
2310 }
2311
2312 /**
2313  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2314  * @adapter: board private structure
2315  **/
2316
2317 static void
2318 e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2319 {
2320         int i;
2321
2322         for (i = 0; i < adapter->num_rx_queues; i++)
2323                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2324 }
2325
2326 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2327  * and memory write and invalidate disabled for certain operations
2328  */
2329 static void
2330 e1000_enter_82542_rst(struct e1000_adapter *adapter)
2331 {
2332         struct net_device *netdev = adapter->netdev;
2333         uint32_t rctl;
2334
2335         e1000_pci_clear_mwi(&adapter->hw);
2336
2337         rctl = E1000_READ_REG(&adapter->hw, RCTL);
2338         rctl |= E1000_RCTL_RST;
2339         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
2340         E1000_WRITE_FLUSH(&adapter->hw);
2341         mdelay(5);
2342
2343         if (netif_running(netdev))
2344                 e1000_clean_all_rx_rings(adapter);
2345 }
2346
2347 static void
2348 e1000_leave_82542_rst(struct e1000_adapter *adapter)
2349 {
2350         struct net_device *netdev = adapter->netdev;
2351         uint32_t rctl;
2352
2353         rctl = E1000_READ_REG(&adapter->hw, RCTL);
2354         rctl &= ~E1000_RCTL_RST;
2355         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
2356         E1000_WRITE_FLUSH(&adapter->hw);
2357         mdelay(5);
2358
2359         if (adapter->hw.pci_cmd_word & PCI_COMMAND_INVALIDATE)
2360                 e1000_pci_set_mwi(&adapter->hw);
2361
2362         if (netif_running(netdev)) {
2363                 /* No need to loop, because 82542 supports only 1 queue */
2364                 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2365                 e1000_configure_rx(adapter);
2366                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2367         }
2368 }
2369
2370 /**
2371  * e1000_set_mac - Change the Ethernet Address of the NIC
2372  * @netdev: network interface device structure
2373  * @p: pointer to an address structure
2374  *
2375  * Returns 0 on success, negative on failure
2376  **/
2377
2378 static int
2379 e1000_set_mac(struct net_device *netdev, void *p)
2380 {
2381         struct e1000_adapter *adapter = netdev_priv(netdev);
2382         struct sockaddr *addr = p;
2383
2384         if (!is_valid_ether_addr(addr->sa_data))
2385                 return -EADDRNOTAVAIL;
2386
2387         /* 82542 2.0 needs to be in reset to write receive address registers */
2388
2389         if (adapter->hw.mac_type == e1000_82542_rev2_0)
2390                 e1000_enter_82542_rst(adapter);
2391
2392         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2393         memcpy(adapter->hw.mac_addr, addr->sa_data, netdev->addr_len);
2394
2395         e1000_rar_set(&adapter->hw, adapter->hw.mac_addr, 0);
2396
2397         /* With 82571 controllers, LAA may be overwritten (with the default)
2398          * due to controller reset from the other port. */
2399         if (adapter->hw.mac_type == e1000_82571) {
2400                 /* activate the work around */
2401                 adapter->hw.laa_is_present = 1;
2402
2403                 /* Hold a copy of the LAA in RAR[14] This is done so that
2404                  * between the time RAR[0] gets clobbered  and the time it
2405                  * gets fixed (in e1000_watchdog), the actual LAA is in one
2406                  * of the RARs and no incoming packets directed to this port
2407                  * are dropped. Eventaully the LAA will be in RAR[0] and
2408                  * RAR[14] */
2409                 e1000_rar_set(&adapter->hw, adapter->hw.mac_addr,
2410                                         E1000_RAR_ENTRIES - 1);
2411         }
2412
2413         if (adapter->hw.mac_type == e1000_82542_rev2_0)
2414                 e1000_leave_82542_rst(adapter);
2415
2416         return 0;
2417 }
2418
2419 /**
2420  * e1000_set_multi - Multicast and Promiscuous mode set
2421  * @netdev: network interface device structure
2422  *
2423  * The set_multi entry point is called whenever the multicast address
2424  * list or the network interface flags are updated.  This routine is
2425  * responsible for configuring the hardware for proper multicast,
2426  * promiscuous mode, and all-multi behavior.
2427  **/
2428
2429 static void
2430 e1000_set_multi(struct net_device *netdev)
2431 {
2432         struct e1000_adapter *adapter = netdev_priv(netdev);
2433         struct e1000_hw *hw = &adapter->hw;
2434         struct dev_mc_list *mc_ptr;
2435         uint32_t rctl;
2436         uint32_t hash_value;
2437         int i, rar_entries = E1000_RAR_ENTRIES;
2438         int mta_reg_count = (hw->mac_type == e1000_ich8lan) ?
2439                                 E1000_NUM_MTA_REGISTERS_ICH8LAN :
2440                                 E1000_NUM_MTA_REGISTERS;
2441
2442         if (adapter->hw.mac_type == e1000_ich8lan)
2443                 rar_entries = E1000_RAR_ENTRIES_ICH8LAN;
2444
2445         /* reserve RAR[14] for LAA over-write work-around */
2446         if (adapter->hw.mac_type == e1000_82571)
2447                 rar_entries--;
2448
2449         /* Check for Promiscuous and All Multicast modes */
2450
2451         rctl = E1000_READ_REG(hw, RCTL);
2452
2453         if (netdev->flags & IFF_PROMISC) {
2454                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2455         } else if (netdev->flags & IFF_ALLMULTI) {
2456                 rctl |= E1000_RCTL_MPE;
2457                 rctl &= ~E1000_RCTL_UPE;
2458         } else {
2459                 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2460         }
2461
2462         E1000_WRITE_REG(hw, RCTL, rctl);
2463
2464         /* 82542 2.0 needs to be in reset to write receive address registers */
2465
2466         if (hw->mac_type == e1000_82542_rev2_0)
2467                 e1000_enter_82542_rst(adapter);
2468
2469         /* load the first 14 multicast address into the exact filters 1-14
2470          * RAR 0 is used for the station MAC adddress
2471          * if there are not 14 addresses, go ahead and clear the filters
2472          * -- with 82571 controllers only 0-13 entries are filled here
2473          */
2474         mc_ptr = netdev->mc_list;
2475
2476         for (i = 1; i < rar_entries; i++) {
2477                 if (mc_ptr) {
2478                         e1000_rar_set(hw, mc_ptr->dmi_addr, i);
2479                         mc_ptr = mc_ptr->next;
2480                 } else {
2481                         E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2482                         E1000_WRITE_FLUSH(hw);
2483                         E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2484                         E1000_WRITE_FLUSH(hw);
2485                 }
2486         }
2487
2488         /* clear the old settings from the multicast hash table */
2489
2490         for (i = 0; i < mta_reg_count; i++) {
2491                 E1000_WRITE_REG_ARRAY(hw, MTA, i, 0);
2492                 E1000_WRITE_FLUSH(hw);
2493         }
2494
2495         /* load any remaining addresses into the hash table */
2496
2497         for (; mc_ptr; mc_ptr = mc_ptr->next) {
2498                 hash_value = e1000_hash_mc_addr(hw, mc_ptr->dmi_addr);
2499                 e1000_mta_set(hw, hash_value);
2500         }
2501
2502         if (hw->mac_type == e1000_82542_rev2_0)
2503                 e1000_leave_82542_rst(adapter);
2504 }
2505
2506 /* Need to wait a few seconds after link up to get diagnostic information from
2507  * the phy */
2508
2509 static void
2510 e1000_update_phy_info(unsigned long data)
2511 {
2512         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2513         e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
2514 }
2515
2516 /**
2517  * e1000_82547_tx_fifo_stall - Timer Call-back
2518  * @data: pointer to adapter cast into an unsigned long
2519  **/
2520
2521 static void
2522 e1000_82547_tx_fifo_stall(unsigned long data)
2523 {
2524         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2525         struct net_device *netdev = adapter->netdev;
2526         uint32_t tctl;
2527
2528         if (atomic_read(&adapter->tx_fifo_stall)) {
2529                 if ((E1000_READ_REG(&adapter->hw, TDT) ==
2530                     E1000_READ_REG(&adapter->hw, TDH)) &&
2531                    (E1000_READ_REG(&adapter->hw, TDFT) ==
2532                     E1000_READ_REG(&adapter->hw, TDFH)) &&
2533                    (E1000_READ_REG(&adapter->hw, TDFTS) ==
2534                     E1000_READ_REG(&adapter->hw, TDFHS))) {
2535                         tctl = E1000_READ_REG(&adapter->hw, TCTL);
2536                         E1000_WRITE_REG(&adapter->hw, TCTL,
2537                                         tctl & ~E1000_TCTL_EN);
2538                         E1000_WRITE_REG(&adapter->hw, TDFT,
2539                                         adapter->tx_head_addr);
2540                         E1000_WRITE_REG(&adapter->hw, TDFH,
2541                                         adapter->tx_head_addr);
2542                         E1000_WRITE_REG(&adapter->hw, TDFTS,
2543                                         adapter->tx_head_addr);
2544                         E1000_WRITE_REG(&adapter->hw, TDFHS,
2545                                         adapter->tx_head_addr);
2546                         E1000_WRITE_REG(&adapter->hw, TCTL, tctl);
2547                         E1000_WRITE_FLUSH(&adapter->hw);
2548
2549                         adapter->tx_fifo_head = 0;
2550                         atomic_set(&adapter->tx_fifo_stall, 0);
2551                         netif_wake_queue(netdev);
2552                 } else {
2553                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2554                 }
2555         }
2556 }
2557
2558 /**
2559  * e1000_watchdog - Timer Call-back
2560  * @data: pointer to adapter cast into an unsigned long
2561  **/
2562 static void
2563 e1000_watchdog(unsigned long data)
2564 {
2565         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2566         struct net_device *netdev = adapter->netdev;
2567         struct e1000_tx_ring *txdr = adapter->tx_ring;
2568         uint32_t link, tctl;
2569         int32_t ret_val;
2570
2571         ret_val = e1000_check_for_link(&adapter->hw);
2572         if ((ret_val == E1000_ERR_PHY) &&
2573             (adapter->hw.phy_type == e1000_phy_igp_3) &&
2574             (E1000_READ_REG(&adapter->hw, CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
2575                 /* See e1000_kumeran_lock_loss_workaround() */
2576                 DPRINTK(LINK, INFO,
2577                         "Gigabit has been disabled, downgrading speed\n");
2578         }
2579
2580         if (adapter->hw.mac_type == e1000_82573) {
2581                 e1000_enable_tx_pkt_filtering(&adapter->hw);
2582                 if (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id)
2583                         e1000_update_mng_vlan(adapter);
2584         }
2585
2586         if ((adapter->hw.media_type == e1000_media_type_internal_serdes) &&
2587            !(E1000_READ_REG(&adapter->hw, TXCW) & E1000_TXCW_ANE))
2588                 link = !adapter->hw.serdes_link_down;
2589         else
2590                 link = E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_LU;
2591
2592         if (link) {
2593                 if (!netif_carrier_ok(netdev)) {
2594                         uint32_t ctrl;
2595                         boolean_t txb2b = 1;
2596                         e1000_get_speed_and_duplex(&adapter->hw,
2597                                                    &adapter->link_speed,
2598                                                    &adapter->link_duplex);
2599
2600                         ctrl = E1000_READ_REG(&adapter->hw, CTRL);
2601                         DPRINTK(LINK, INFO, "NIC Link is Up %d Mbps %s, "
2602                                 "Flow Control: %s\n",
2603                                 adapter->link_speed,
2604                                 adapter->link_duplex == FULL_DUPLEX ?
2605                                 "Full Duplex" : "Half Duplex",
2606                                 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2607                                 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2608                                 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2609                                 E1000_CTRL_TFCE) ? "TX" : "None" )));
2610
2611                         /* tweak tx_queue_len according to speed/duplex
2612                          * and adjust the timeout factor */
2613                         netdev->tx_queue_len = adapter->tx_queue_len;
2614                         adapter->tx_timeout_factor = 1;
2615                         switch (adapter->link_speed) {
2616                         case SPEED_10:
2617                                 txb2b = 0;
2618                                 netdev->tx_queue_len = 10;
2619                                 adapter->tx_timeout_factor = 8;
2620                                 break;
2621                         case SPEED_100:
2622                                 txb2b = 0;
2623                                 netdev->tx_queue_len = 100;
2624                                 /* maybe add some timeout factor ? */
2625                                 break;
2626                         }
2627
2628                         if ((adapter->hw.mac_type == e1000_82571 ||
2629                              adapter->hw.mac_type == e1000_82572) &&
2630                             txb2b == 0) {
2631                                 uint32_t tarc0;
2632                                 tarc0 = E1000_READ_REG(&adapter->hw, TARC0);
2633                                 tarc0 &= ~(1 << 21);
2634                                 E1000_WRITE_REG(&adapter->hw, TARC0, tarc0);
2635                         }
2636
2637                         /* disable TSO for pcie and 10/100 speeds, to avoid
2638                          * some hardware issues */
2639                         if (!adapter->tso_force &&
2640                             adapter->hw.bus_type == e1000_bus_type_pci_express){
2641                                 switch (adapter->link_speed) {
2642                                 case SPEED_10:
2643                                 case SPEED_100:
2644                                         DPRINTK(PROBE,INFO,
2645                                         "10/100 speed: disabling TSO\n");
2646                                         netdev->features &= ~NETIF_F_TSO;
2647                                         netdev->features &= ~NETIF_F_TSO6;
2648                                         break;
2649                                 case SPEED_1000:
2650                                         netdev->features |= NETIF_F_TSO;
2651                                         netdev->features |= NETIF_F_TSO6;
2652                                         break;
2653                                 default:
2654                                         /* oops */
2655                                         break;
2656                                 }
2657                         }
2658
2659                         /* enable transmits in the hardware, need to do this
2660                          * after setting TARC0 */
2661                         tctl = E1000_READ_REG(&adapter->hw, TCTL);
2662                         tctl |= E1000_TCTL_EN;
2663                         E1000_WRITE_REG(&adapter->hw, TCTL, tctl);
2664
2665                         netif_carrier_on(netdev);
2666                         netif_wake_queue(netdev);
2667                         mod_timer(&adapter->phy_info_timer, round_jiffies(jiffies + 2 * HZ));
2668                         adapter->smartspeed = 0;
2669                 } else {
2670                         /* make sure the receive unit is started */
2671                         if (adapter->hw.rx_needs_kicking) {
2672                                 struct e1000_hw *hw = &adapter->hw;
2673                                 uint32_t rctl = E1000_READ_REG(hw, RCTL);
2674                                 E1000_WRITE_REG(hw, RCTL, rctl | E1000_RCTL_EN);
2675                         }
2676                 }
2677         } else {
2678                 if (netif_carrier_ok(netdev)) {
2679                         adapter->link_speed = 0;
2680                         adapter->link_duplex = 0;
2681                         DPRINTK(LINK, INFO, "NIC Link is Down\n");
2682                         netif_carrier_off(netdev);
2683                         netif_stop_queue(netdev);
2684                         mod_timer(&adapter->phy_info_timer, round_jiffies(jiffies + 2 * HZ));
2685
2686                         /* 80003ES2LAN workaround--
2687                          * For packet buffer work-around on link down event;
2688                          * disable receives in the ISR and
2689                          * reset device here in the watchdog
2690                          */
2691                         if (adapter->hw.mac_type == e1000_80003es2lan)
2692                                 /* reset device */
2693                                 schedule_work(&adapter->reset_task);
2694                 }
2695
2696                 e1000_smartspeed(adapter);
2697         }
2698
2699         e1000_update_stats(adapter);
2700
2701         adapter->hw.tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2702         adapter->tpt_old = adapter->stats.tpt;
2703         adapter->hw.collision_delta = adapter->stats.colc - adapter->colc_old;
2704         adapter->colc_old = adapter->stats.colc;
2705
2706         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2707         adapter->gorcl_old = adapter->stats.gorcl;
2708         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2709         adapter->gotcl_old = adapter->stats.gotcl;
2710
2711         e1000_update_adaptive(&adapter->hw);
2712
2713         if (!netif_carrier_ok(netdev)) {
2714                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2715                         /* We've lost link, so the controller stops DMA,
2716                          * but we've got queued Tx work that's never going
2717                          * to get done, so reset controller to flush Tx.
2718                          * (Do the reset outside of interrupt context). */
2719                         adapter->tx_timeout_count++;
2720                         schedule_work(&adapter->reset_task);
2721                 }
2722         }
2723
2724         /* Cause software interrupt to ensure rx ring is cleaned */
2725         E1000_WRITE_REG(&adapter->hw, ICS, E1000_ICS_RXDMT0);
2726
2727         /* Force detection of hung controller every watchdog period */
2728         adapter->detect_tx_hung = TRUE;
2729
2730         /* With 82571 controllers, LAA may be overwritten due to controller
2731          * reset from the other port. Set the appropriate LAA in RAR[0] */
2732         if (adapter->hw.mac_type == e1000_82571 && adapter->hw.laa_is_present)
2733                 e1000_rar_set(&adapter->hw, adapter->hw.mac_addr, 0);
2734
2735         /* Reset the timer */
2736         mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
2737 }
2738
2739 enum latency_range {
2740         lowest_latency = 0,
2741         low_latency = 1,
2742         bulk_latency = 2,
2743         latency_invalid = 255
2744 };
2745
2746 /**
2747  * e1000_update_itr - update the dynamic ITR value based on statistics
2748  *      Stores a new ITR value based on packets and byte
2749  *      counts during the last interrupt.  The advantage of per interrupt
2750  *      computation is faster updates and more accurate ITR for the current
2751  *      traffic pattern.  Constants in this function were computed
2752  *      based on theoretical maximum wire speed and thresholds were set based
2753  *      on testing data as well as attempting to minimize response time
2754  *      while increasing bulk throughput.
2755  *      this functionality is controlled by the InterruptThrottleRate module
2756  *      parameter (see e1000_param.c)
2757  * @adapter: pointer to adapter
2758  * @itr_setting: current adapter->itr
2759  * @packets: the number of packets during this measurement interval
2760  * @bytes: the number of bytes during this measurement interval
2761  **/
2762 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2763                                    uint16_t itr_setting,
2764                                    int packets,
2765                                    int bytes)
2766 {
2767         unsigned int retval = itr_setting;
2768         struct e1000_hw *hw = &adapter->hw;
2769
2770         if (unlikely(hw->mac_type < e1000_82540))
2771                 goto update_itr_done;
2772
2773         if (packets == 0)
2774                 goto update_itr_done;
2775
2776         switch (itr_setting) {
2777         case lowest_latency:
2778                 /* jumbo frames get bulk treatment*/
2779                 if (bytes/packets > 8000)
2780                         retval = bulk_latency;
2781                 else if ((packets < 5) && (bytes > 512))
2782                         retval = low_latency;
2783                 break;
2784         case low_latency:  /* 50 usec aka 20000 ints/s */
2785                 if (bytes > 10000) {
2786                         /* jumbo frames need bulk latency setting */
2787                         if (bytes/packets > 8000)
2788                                 retval = bulk_latency;
2789                         else if ((packets < 10) || ((bytes/packets) > 1200))
2790                                 retval = bulk_latency;
2791                         else if ((packets > 35))
2792                                 retval = lowest_latency;
2793                 } else if (bytes/packets > 2000)
2794                         retval = bulk_latency;
2795                 else if (packets <= 2 && bytes < 512)
2796                         retval = lowest_latency;
2797                 break;
2798         case bulk_latency: /* 250 usec aka 4000 ints/s */
2799                 if (bytes > 25000) {
2800                         if (packets > 35)
2801                                 retval = low_latency;
2802                 } else if (bytes < 6000) {
2803                         retval = low_latency;
2804                 }
2805                 break;
2806         }
2807
2808 update_itr_done:
2809         return retval;
2810 }
2811
2812 static void e1000_set_itr(struct e1000_adapter *adapter)
2813 {
2814         struct e1000_hw *hw = &adapter->hw;
2815         uint16_t current_itr;
2816         uint32_t new_itr = adapter->itr;
2817
2818         if (unlikely(hw->mac_type < e1000_82540))
2819                 return;
2820
2821         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2822         if (unlikely(adapter->link_speed != SPEED_1000)) {
2823                 current_itr = 0;
2824                 new_itr = 4000;
2825                 goto set_itr_now;
2826         }
2827
2828         adapter->tx_itr = e1000_update_itr(adapter,
2829                                     adapter->tx_itr,
2830                                     adapter->total_tx_packets,
2831                                     adapter->total_tx_bytes);
2832         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2833         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2834                 adapter->tx_itr = low_latency;
2835
2836         adapter->rx_itr = e1000_update_itr(adapter,
2837                                     adapter->rx_itr,
2838                                     adapter->total_rx_packets,
2839                                     adapter->total_rx_bytes);
2840         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2841         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2842                 adapter->rx_itr = low_latency;
2843
2844         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2845
2846         switch (current_itr) {
2847         /* counts and packets in update_itr are dependent on these numbers */
2848         case lowest_latency:
2849                 new_itr = 70000;
2850                 break;
2851         case low_latency:
2852                 new_itr = 20000; /* aka hwitr = ~200 */
2853                 break;
2854         case bulk_latency:
2855                 new_itr = 4000;
2856                 break;
2857         default:
2858                 break;
2859         }
2860
2861 set_itr_now:
2862         if (new_itr != adapter->itr) {
2863                 /* this attempts to bias the interrupt rate towards Bulk
2864                  * by adding intermediate steps when interrupt rate is
2865                  * increasing */
2866                 new_itr = new_itr > adapter->itr ?
2867                              min(adapter->itr + (new_itr >> 2), new_itr) :
2868                              new_itr;
2869                 adapter->itr = new_itr;
2870                 E1000_WRITE_REG(hw, ITR, 1000000000 / (new_itr * 256));
2871         }
2872
2873         return;
2874 }
2875
2876 #define E1000_TX_FLAGS_CSUM             0x00000001
2877 #define E1000_TX_FLAGS_VLAN             0x00000002
2878 #define E1000_TX_FLAGS_TSO              0x00000004
2879 #define E1000_TX_FLAGS_IPV4             0x00000008
2880 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2881 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2882
2883 static int
2884 e1000_tso(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2885           struct sk_buff *skb)
2886 {
2887         struct e1000_context_desc *context_desc;
2888         struct e1000_buffer *buffer_info;
2889         unsigned int i;
2890         uint32_t cmd_length = 0;
2891         uint16_t ipcse = 0, tucse, mss;
2892         uint8_t ipcss, ipcso, tucss, tucso, hdr_len;
2893         int err;
2894
2895         if (skb_is_gso(skb)) {
2896                 if (skb_header_cloned(skb)) {
2897                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2898                         if (err)
2899                                 return err;
2900                 }
2901
2902                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2903                 mss = skb_shinfo(skb)->gso_size;
2904                 if (skb->protocol == htons(ETH_P_IP)) {
2905                         struct iphdr *iph = ip_hdr(skb);
2906                         iph->tot_len = 0;
2907                         iph->check = 0;
2908                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2909                                                                  iph->daddr, 0,
2910                                                                  IPPROTO_TCP,
2911                                                                  0);
2912                         cmd_length = E1000_TXD_CMD_IP;
2913                         ipcse = skb_transport_offset(skb) - 1;
2914                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2915                         ipv6_hdr(skb)->payload_len = 0;
2916                         tcp_hdr(skb)->check =
2917                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2918                                                  &ipv6_hdr(skb)->daddr,
2919                                                  0, IPPROTO_TCP, 0);
2920                         ipcse = 0;
2921                 }
2922                 ipcss = skb_network_offset(skb);
2923                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2924                 tucss = skb_transport_offset(skb);
2925                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2926                 tucse = 0;
2927
2928                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2929                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2930
2931                 i = tx_ring->next_to_use;
2932                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2933                 buffer_info = &tx_ring->buffer_info[i];
2934
2935                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2936                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2937                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2938                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2939                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2940                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2941                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2942                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2943                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2944
2945                 buffer_info->time_stamp = jiffies;
2946                 buffer_info->next_to_watch = i;
2947
2948                 if (++i == tx_ring->count) i = 0;
2949                 tx_ring->next_to_use = i;
2950
2951                 return TRUE;
2952         }
2953         return FALSE;
2954 }
2955
2956 static boolean_t
2957 e1000_tx_csum(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2958               struct sk_buff *skb)
2959 {
2960         struct e1000_context_desc *context_desc;
2961         struct e1000_buffer *buffer_info;
2962         unsigned int i;
2963         uint8_t css;
2964
2965         if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
2966                 css = skb_transport_offset(skb);
2967
2968                 i = tx_ring->next_to_use;
2969                 buffer_info = &tx_ring->buffer_info[i];
2970                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2971
2972                 context_desc->lower_setup.ip_config = 0;
2973                 context_desc->upper_setup.tcp_fields.tucss = css;
2974                 context_desc->upper_setup.tcp_fields.tucso =
2975                         css + skb->csum_offset;
2976                 context_desc->upper_setup.tcp_fields.tucse = 0;
2977                 context_desc->tcp_seg_setup.data = 0;
2978                 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
2979
2980                 buffer_info->time_stamp = jiffies;
2981                 buffer_info->next_to_watch = i;
2982
2983                 if (unlikely(++i == tx_ring->count)) i = 0;
2984                 tx_ring->next_to_use = i;
2985
2986                 return TRUE;
2987         }
2988
2989         return FALSE;
2990 }
2991
2992 #define E1000_MAX_TXD_PWR       12
2993 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2994
2995 static int
2996 e1000_tx_map(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2997              struct sk_buff *skb, unsigned int first, unsigned int max_per_txd,
2998              unsigned int nr_frags, unsigned int mss)
2999 {
3000         struct e1000_buffer *buffer_info;
3001         unsigned int len = skb->len;
3002         unsigned int offset = 0, size, count = 0, i;
3003         unsigned int f;
3004         len -= skb->data_len;
3005
3006         i = tx_ring->next_to_use;
3007
3008         while (len) {
3009                 buffer_info = &tx_ring->buffer_info[i];
3010                 size = min(len, max_per_txd);
3011                 /* Workaround for Controller erratum --
3012                  * descriptor for non-tso packet in a linear SKB that follows a
3013                  * tso gets written back prematurely before the data is fully
3014                  * DMA'd to the controller */
3015                 if (!skb->data_len && tx_ring->last_tx_tso &&
3016                     !skb_is_gso(skb)) {
3017                         tx_ring->last_tx_tso = 0;
3018                         size -= 4;
3019                 }
3020
3021                 /* Workaround for premature desc write-backs
3022                  * in TSO mode.  Append 4-byte sentinel desc */
3023                 if (unlikely(mss && !nr_frags && size == len && size > 8))
3024                         size -= 4;
3025                 /* work-around for errata 10 and it applies
3026                  * to all controllers in PCI-X mode
3027                  * The fix is to make sure that the first descriptor of a
3028                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
3029                  */
3030                 if (unlikely((adapter->hw.bus_type == e1000_bus_type_pcix) &&
3031                                 (size > 2015) && count == 0))
3032                         size = 2015;
3033
3034                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
3035                  * terminating buffers within evenly-aligned dwords. */
3036                 if (unlikely(adapter->pcix_82544 &&
3037                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
3038                    size > 4))
3039                         size -= 4;
3040
3041                 buffer_info->length = size;
3042                 buffer_info->dma =
3043                         pci_map_single(adapter->pdev,
3044                                 skb->data + offset,
3045                                 size,
3046                                 PCI_DMA_TODEVICE);
3047                 buffer_info->time_stamp = jiffies;
3048                 buffer_info->next_to_watch = i;
3049
3050                 len -= size;
3051                 offset += size;
3052                 count++;
3053                 if (unlikely(++i == tx_ring->count)) i = 0;
3054         }
3055
3056         for (f = 0; f < nr_frags; f++) {
3057                 struct skb_frag_struct *frag;
3058
3059                 frag = &skb_shinfo(skb)->frags[f];
3060                 len = frag->size;
3061                 offset = frag->page_offset;
3062
3063                 while (len) {
3064                         buffer_info = &tx_ring->buffer_info[i];
3065                         size = min(len, max_per_txd);
3066                         /* Workaround for premature desc write-backs
3067                          * in TSO mode.  Append 4-byte sentinel desc */
3068                         if (unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
3069                                 size -= 4;
3070                         /* Workaround for potential 82544 hang in PCI-X.
3071                          * Avoid terminating buffers within evenly-aligned
3072                          * dwords. */
3073                         if (unlikely(adapter->pcix_82544 &&
3074                            !((unsigned long)(frag->page+offset+size-1) & 4) &&
3075                            size > 4))
3076                                 size -= 4;
3077
3078                         buffer_info->length = size;
3079                         buffer_info->dma =
3080                                 pci_map_page(adapter->pdev,
3081                                         frag->page,
3082                                         offset,
3083                                         size,
3084                                         PCI_DMA_TODEVICE);
3085                         buffer_info->time_stamp = jiffies;
3086                         buffer_info->next_to_watch = i;
3087
3088                         len -= size;
3089                         offset += size;
3090                         count++;
3091                         if (unlikely(++i == tx_ring->count)) i = 0;
3092                 }
3093         }
3094
3095         i = (i == 0) ? tx_ring->count - 1 : i - 1;
3096         tx_ring->buffer_info[i].skb = skb;
3097         tx_ring->buffer_info[first].next_to_watch = i;
3098
3099         return count;
3100 }
3101
3102 static void
3103 e1000_tx_queue(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
3104                int tx_flags, int count)
3105 {
3106         struct e1000_tx_desc *tx_desc = NULL;
3107         struct e1000_buffer *buffer_info;
3108         uint32_t txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3109         unsigned int i;
3110
3111         if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
3112                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3113                              E1000_TXD_CMD_TSE;
3114                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3115
3116                 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
3117                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3118         }
3119
3120         if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
3121                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3122                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3123         }
3124
3125         if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
3126                 txd_lower |= E1000_TXD_CMD_VLE;
3127                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3128         }
3129
3130         i = tx_ring->next_to_use;
3131
3132         while (count--) {
3133                 buffer_info = &tx_ring->buffer_info[i];
3134                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3135                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3136                 tx_desc->lower.data =
3137                         cpu_to_le32(txd_lower | buffer_info->length);
3138                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3139                 if (unlikely(++i == tx_ring->count)) i = 0;
3140         }
3141
3142         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3143
3144         /* Force memory writes to complete before letting h/w
3145          * know there are new descriptors to fetch.  (Only
3146          * applicable for weak-ordered memory model archs,
3147          * such as IA-64). */
3148         wmb();
3149
3150         tx_ring->next_to_use = i;
3151         writel(i, adapter->hw.hw_addr + tx_ring->tdt);
3152         /* we need this if more than one processor can write to our tail
3153          * at a time, it syncronizes IO on IA64/Altix systems */
3154         mmiowb();
3155 }
3156
3157 /**
3158  * 82547 workaround to avoid controller hang in half-duplex environment.
3159  * The workaround is to avoid queuing a large packet that would span
3160  * the internal Tx FIFO ring boundary by notifying the stack to resend
3161  * the packet at a later time.  This gives the Tx FIFO an opportunity to
3162  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
3163  * to the beginning of the Tx FIFO.
3164  **/
3165
3166 #define E1000_FIFO_HDR                  0x10
3167 #define E1000_82547_PAD_LEN             0x3E0
3168
3169 static int
3170 e1000_82547_fifo_workaround(struct e1000_adapter *adapter, struct sk_buff *skb)
3171 {
3172         uint32_t fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
3173         uint32_t skb_fifo_len = skb->len + E1000_FIFO_HDR;
3174
3175         skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
3176
3177         if (adapter->link_duplex != HALF_DUPLEX)
3178                 goto no_fifo_stall_required;
3179
3180         if (atomic_read(&adapter->tx_fifo_stall))
3181                 return 1;
3182
3183         if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
3184                 atomic_set(&adapter->tx_fifo_stall, 1);
3185                 return 1;
3186         }
3187
3188 no_fifo_stall_required:
3189         adapter->tx_fifo_head += skb_fifo_len;
3190         if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
3191                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
3192         return 0;
3193 }
3194
3195 #define MINIMUM_DHCP_PACKET_SIZE 282
3196 static int
3197 e1000_transfer_dhcp_info(struct e1000_adapter *adapter, struct sk_buff *skb)
3198 {
3199         struct e1000_hw *hw =  &adapter->hw;
3200         uint16_t length, offset;
3201         if (vlan_tx_tag_present(skb)) {
3202                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
3203                         ( adapter->hw.mng_cookie.status &
3204                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) )
3205                         return 0;
3206         }
3207         if (skb->len > MINIMUM_DHCP_PACKET_SIZE) {
3208                 struct ethhdr *eth = (struct ethhdr *) skb->data;
3209                 if ((htons(ETH_P_IP) == eth->h_proto)) {
3210                         const struct iphdr *ip =
3211                                 (struct iphdr *)((uint8_t *)skb->data+14);
3212                         if (IPPROTO_UDP == ip->protocol) {
3213                                 struct udphdr *udp =
3214                                         (struct udphdr *)((uint8_t *)ip +
3215                                                 (ip->ihl << 2));
3216                                 if (ntohs(udp->dest) == 67) {
3217                                         offset = (uint8_t *)udp + 8 - skb->data;
3218                                         length = skb->len - offset;
3219
3220                                         return e1000_mng_write_dhcp_info(hw,
3221                                                         (uint8_t *)udp + 8,
3222                                                         length);
3223                                 }
3224                         }
3225                 }
3226         }
3227         return 0;
3228 }
3229
3230 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3231 {
3232         struct e1000_adapter *adapter = netdev_priv(netdev);
3233         struct e1000_tx_ring *tx_ring = adapter->tx_ring;
3234
3235         netif_stop_queue(netdev);
3236         /* Herbert's original patch had:
3237          *  smp_mb__after_netif_stop_queue();
3238          * but since that doesn't exist yet, just open code it. */
3239         smp_mb();
3240
3241         /* We need to check again in a case another CPU has just
3242          * made room available. */
3243         if (likely(E1000_DESC_UNUSED(tx_ring) < size))
3244                 return -EBUSY;
3245
3246         /* A reprieve! */
3247         netif_start_queue(netdev);
3248         ++adapter->restart_queue;
3249         return 0;
3250 }
3251
3252 static int e1000_maybe_stop_tx(struct net_device *netdev,
3253                                struct e1000_tx_ring *tx_ring, int size)
3254 {
3255         if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
3256                 return 0;
3257         return __e1000_maybe_stop_tx(netdev, size);
3258 }
3259
3260 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3261 static int
3262 e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3263 {
3264         struct e1000_adapter *adapter = netdev_priv(netdev);
3265         struct e1000_tx_ring *tx_ring;
3266         unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
3267         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3268         unsigned int tx_flags = 0;
3269         unsigned int len = skb->len;
3270         unsigned long flags;
3271         unsigned int nr_frags = 0;
3272         unsigned int mss = 0;
3273         int count = 0;
3274         int tso;
3275         unsigned int f;
3276         len -= skb->data_len;
3277
3278         /* This goes back to the question of how to logically map a tx queue
3279          * to a flow.  Right now, performance is impacted slightly negatively
3280          * if using multiple tx queues.  If the stack breaks away from a
3281          * single qdisc implementation, we can look at this again. */
3282         tx_ring = adapter->tx_ring;
3283
3284         if (unlikely(skb->len <= 0)) {
3285                 dev_kfree_skb_any(skb);
3286                 return NETDEV_TX_OK;
3287         }
3288
3289         /* 82571 and newer doesn't need the workaround that limited descriptor
3290          * length to 4kB */
3291         if (adapter->hw.mac_type >= e1000_82571)
3292                 max_per_txd = 8192;
3293
3294         mss = skb_shinfo(skb)->gso_size;
3295         /* The controller does a simple calculation to
3296          * make sure there is enough room in the FIFO before
3297          * initiating the DMA for each buffer.  The calc is:
3298          * 4 = ceil(buffer len/mss).  To make sure we don't
3299          * overrun the FIFO, adjust the max buffer len if mss
3300          * drops. */
3301         if (mss) {
3302                 uint8_t hdr_len;
3303                 max_per_txd = min(mss << 2, max_per_txd);
3304                 max_txd_pwr = fls(max_per_txd) - 1;
3305
3306                 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
3307                 * points to just header, pull a few bytes of payload from
3308                 * frags into skb->data */
3309                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3310                 if (skb->data_len && (hdr_len == (skb->len - skb->data_len))) {
3311                         switch (adapter->hw.mac_type) {
3312                                 unsigned int pull_size;
3313                         case e1000_82544:
3314                                 /* Make sure we have room to chop off 4 bytes,
3315                                  * and that the end alignment will work out to
3316                                  * this hardware's requirements
3317                                  * NOTE: this is a TSO only workaround
3318                                  * if end byte alignment not correct move us
3319                                  * into the next dword */
3320                                 if ((unsigned long)(skb_tail_pointer(skb) - 1) & 4)
3321                                         break;
3322                                 /* fall through */
3323                         case e1000_82571:
3324                         case e1000_82572:
3325                         case e1000_82573:
3326                         case e1000_ich8lan:
3327                                 pull_size = min((unsigned int)4, skb->data_len);
3328                                 if (!__pskb_pull_tail(skb, pull_size)) {
3329                                         DPRINTK(DRV, ERR,
3330                                                 "__pskb_pull_tail failed.\n");
3331                                         dev_kfree_skb_any(skb);
3332                                         return NETDEV_TX_OK;
3333                                 }
3334                                 len = skb->len - skb->data_len;
3335                                 break;
3336                         default:
3337                                 /* do nothing */
3338                                 break;
3339                         }
3340                 }
3341         }
3342
3343         /* reserve a descriptor for the offload context */
3344         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3345                 count++;
3346         count++;
3347
3348         /* Controller Erratum workaround */
3349         if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3350                 count++;
3351
3352         count += TXD_USE_COUNT(len, max_txd_pwr);
3353
3354         if (adapter->pcix_82544)
3355                 count++;
3356
3357         /* work-around for errata 10 and it applies to all controllers
3358          * in PCI-X mode, so add one more descriptor to the count
3359          */
3360         if (unlikely((adapter->hw.bus_type == e1000_bus_type_pcix) &&
3361                         (len > 2015)))
3362                 count++;
3363
3364         nr_frags = skb_shinfo(skb)->nr_frags;
3365         for (f = 0; f < nr_frags; f++)
3366                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3367                                        max_txd_pwr);
3368         if (adapter->pcix_82544)
3369                 count += nr_frags;
3370
3371
3372         if (adapter->hw.tx_pkt_filtering &&
3373             (adapter->hw.mac_type == e1000_82573))
3374                 e1000_transfer_dhcp_info(adapter, skb);
3375
3376         if (!spin_trylock_irqsave(&tx_ring->tx_lock, flags))
3377                 /* Collision - tell upper layer to requeue */
3378                 return NETDEV_TX_LOCKED;
3379
3380         /* need: count + 2 desc gap to keep tail from touching
3381          * head, otherwise try next time */
3382         if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2))) {
3383                 spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3384                 return NETDEV_TX_BUSY;
3385         }
3386
3387         if (unlikely(adapter->hw.mac_type == e1000_82547)) {
3388                 if (unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
3389                         netif_stop_queue(netdev);
3390                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
3391                         spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3392                         return NETDEV_TX_BUSY;
3393                 }
3394         }
3395
3396         if (unlikely(adapter->vlgrp && vlan_tx_tag_present(skb))) {
3397                 tx_flags |= E1000_TX_FLAGS_VLAN;
3398                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3399         }
3400
3401         first = tx_ring->next_to_use;
3402
3403         tso = e1000_tso(adapter, tx_ring, skb);
3404         if (tso < 0) {
3405                 dev_kfree_skb_any(skb);
3406                 spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3407                 return NETDEV_TX_OK;
3408         }
3409
3410         if (likely(tso)) {
3411                 tx_ring->last_tx_tso = 1;
3412                 tx_flags |= E1000_TX_FLAGS_TSO;
3413         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3414                 tx_flags |= E1000_TX_FLAGS_CSUM;
3415
3416         /* Old method was to assume IPv4 packet by default if TSO was enabled.
3417          * 82571 hardware supports TSO capabilities for IPv6 as well...
3418          * no longer assume, we must. */
3419         if (likely(skb->protocol == htons(ETH_P_IP)))
3420                 tx_flags |= E1000_TX_FLAGS_IPV4;
3421
3422         e1000_tx_queue(adapter, tx_ring, tx_flags,
3423                        e1000_tx_map(adapter, tx_ring, skb, first,
3424                                     max_per_txd, nr_frags, mss));
3425
3426         netdev->trans_start = jiffies;
3427
3428         /* Make sure there is space in the ring for the next send. */
3429         e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3430
3431         spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3432         return NETDEV_TX_OK;
3433 }
3434
3435 /**
3436  * e1000_tx_timeout - Respond to a Tx Hang
3437  * @netdev: network interface device structure
3438  **/
3439
3440 static void
3441 e1000_tx_timeout(struct net_device *netdev)
3442 {
3443         struct e1000_adapter *adapter = netdev_priv(netdev);
3444
3445         /* Do the reset outside of interrupt context */
3446         adapter->tx_timeout_count++;
3447         schedule_work(&adapter->reset_task);
3448 }
3449
3450 static void
3451 e1000_reset_task(struct work_struct *work)
3452 {
3453         struct e1000_adapter *adapter =
3454                 container_of(work, struct e1000_adapter, reset_task);
3455
3456         e1000_reinit_locked(adapter);
3457 }
3458
3459 /**
3460  * e1000_get_stats - Get System Network Statistics
3461  * @netdev: network interface device structure
3462  *
3463  * Returns the address of the device statistics structure.
3464  * The statistics are actually updated from the timer callback.
3465  **/
3466
3467 static struct net_device_stats *
3468 e1000_get_stats(struct net_device *netdev)
3469 {
3470         struct e1000_adapter *adapter = netdev_priv(netdev);
3471
3472         /* only return the current stats */
3473         return &adapter->net_stats;
3474 }
3475
3476 /**
3477  * e1000_change_mtu - Change the Maximum Transfer Unit
3478  * @netdev: network interface device structure
3479  * @new_mtu: new value for maximum frame size
3480  *
3481  * Returns 0 on success, negative on failure
3482  **/
3483
3484 static int
3485 e1000_change_mtu(struct net_device *netdev, int new_mtu)
3486 {
3487         struct e1000_adapter *adapter = netdev_priv(netdev);
3488         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3489         uint16_t eeprom_data = 0;
3490
3491         if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3492             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3493                 DPRINTK(PROBE, ERR, "Invalid MTU setting\n");
3494                 return -EINVAL;
3495         }
3496
3497         /* Adapter-specific max frame size limits. */
3498         switch (adapter->hw.mac_type) {
3499         case e1000_undefined ... e1000_82542_rev2_1:
3500         case e1000_ich8lan:
3501                 if (max_frame > MAXIMUM_ETHERNET_FRAME_SIZE) {
3502                         DPRINTK(PROBE, ERR, "Jumbo Frames not supported.\n");
3503                         return -EINVAL;
3504                 }
3505                 break;
3506         case e1000_82573:
3507                 /* Jumbo Frames not supported if:
3508                  * - this is not an 82573L device
3509                  * - ASPM is enabled in any way (0x1A bits 3:2) */
3510                 e1000_read_eeprom(&adapter->hw, EEPROM_INIT_3GIO_3, 1,
3511                                   &eeprom_data);
3512                 if ((adapter->hw.device_id != E1000_DEV_ID_82573L) ||
3513                     (eeprom_data & EEPROM_WORD1A_ASPM_MASK)) {
3514                         if (max_frame > MAXIMUM_ETHERNET_FRAME_SIZE) {
3515                                 DPRINTK(PROBE, ERR,
3516                                         "Jumbo Frames not supported.\n");
3517                                 return -EINVAL;
3518                         }
3519                         break;
3520                 }
3521                 /* ERT will be enabled later to enable wire speed receives */
3522
3523                 /* fall through to get support */
3524         case e1000_82571:
3525         case e1000_82572:
3526         case e1000_80003es2lan:
3527 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3528                 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3529                         DPRINTK(PROBE, ERR, "MTU > 9216 not supported.\n");
3530                         return -EINVAL;
3531                 }
3532                 break;
3533         default:
3534                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3535                 break;
3536         }
3537
3538         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3539          * means we reserve 2 more, this pushes us to allocate from the next
3540          * larger slab size
3541          * i.e. RXBUFFER_2048 --> size-4096 slab */
3542
3543         if (max_frame <= E1000_RXBUFFER_256)
3544                 adapter->rx_buffer_len = E1000_RXBUFFER_256;
3545         else if (max_frame <= E1000_RXBUFFER_512)
3546                 adapter->rx_buffer_len = E1000_RXBUFFER_512;
3547         else if (max_frame <= E1000_RXBUFFER_1024)
3548                 adapter->rx_buffer_len = E1000_RXBUFFER_1024;
3549         else if (max_frame <= E1000_RXBUFFER_2048)
3550                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3551         else if (max_frame <= E1000_RXBUFFER_4096)
3552                 adapter->rx_buffer_len = E1000_RXBUFFER_4096;
3553         else if (max_frame <= E1000_RXBUFFER_8192)
3554                 adapter->rx_buffer_len = E1000_RXBUFFER_8192;
3555         else if (max_frame <= E1000_RXBUFFER_16384)
3556                 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3557
3558         /* adjust allocation if LPE protects us, and we aren't using SBP */
3559         if (!adapter->hw.tbi_compatibility_on &&
3560             ((max_frame == MAXIMUM_ETHERNET_FRAME_SIZE) ||
3561              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3562                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3563
3564         netdev->mtu = new_mtu;
3565         adapter->hw.max_frame_size = max_frame;
3566
3567         if (netif_running(netdev))
3568                 e1000_reinit_locked(adapter);
3569
3570         return 0;
3571 }
3572
3573 /**
3574  * e1000_update_stats - Update the board statistics counters
3575  * @adapter: board private structure
3576  **/
3577
3578 void
3579 e1000_update_stats(struct e1000_adapter *adapter)
3580 {
3581         struct e1000_hw *hw = &adapter->hw;
3582         struct pci_dev *pdev = adapter->pdev;
3583         unsigned long flags;
3584         uint16_t phy_tmp;
3585
3586 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3587
3588         /*
3589          * Prevent stats update while adapter is being reset, or if the pci
3590          * connection is down.
3591          */
3592         if (adapter->link_speed == 0)
3593                 return;
3594         if (pci_channel_offline(pdev))
3595                 return;
3596
3597         spin_lock_irqsave(&adapter->stats_lock, flags);
3598
3599         /* these counters are modified from e1000_adjust_tbi_stats,
3600          * called from the interrupt context, so they must only
3601          * be written while holding adapter->stats_lock
3602          */
3603
3604         adapter->stats.crcerrs += E1000_READ_REG(hw, CRCERRS);
3605         adapter->stats.gprc += E1000_READ_REG(hw, GPRC);
3606         adapter->stats.gorcl += E1000_READ_REG(hw, GORCL);
3607         adapter->stats.gorch += E1000_READ_REG(hw, GORCH);
3608         adapter->stats.bprc += E1000_READ_REG(hw, BPRC);
3609         adapter->stats.mprc += E1000_READ_REG(hw, MPRC);
3610         adapter->stats.roc += E1000_READ_REG(hw, ROC);
3611
3612         if (adapter->hw.mac_type != e1000_ich8lan) {
3613                 adapter->stats.prc64 += E1000_READ_REG(hw, PRC64);
3614                 adapter->stats.prc127 += E1000_READ_REG(hw, PRC127);
3615                 adapter->stats.prc255 += E1000_READ_REG(hw, PRC255);
3616                 adapter->stats.prc511 += E1000_READ_REG(hw, PRC511);
3617                 adapter->stats.prc1023 += E1000_READ_REG(hw, PRC1023);
3618                 adapter->stats.prc1522 += E1000_READ_REG(hw, PRC1522);
3619         }
3620
3621         adapter->stats.symerrs += E1000_READ_REG(hw, SYMERRS);
3622         adapter->stats.mpc += E1000_READ_REG(hw, MPC);
3623         adapter->stats.scc += E1000_READ_REG(hw, SCC);
3624         adapter->stats.ecol += E1000_READ_REG(hw, ECOL);
3625         adapter->stats.mcc += E1000_READ_REG(hw, MCC);
3626         adapter->stats.latecol += E1000_READ_REG(hw, LATECOL);
3627         adapter->stats.dc += E1000_READ_REG(hw, DC);
3628         adapter->stats.sec += E1000_READ_REG(hw, SEC);
3629         adapter->stats.rlec += E1000_READ_REG(hw, RLEC);
3630         adapter->stats.xonrxc += E1000_READ_REG(hw, XONRXC);
3631         adapter->stats.xontxc += E1000_READ_REG(hw, XONTXC);
3632         adapter->stats.xoffrxc += E1000_READ_REG(hw, XOFFRXC);
3633         adapter->stats.xofftxc += E1000_READ_REG(hw, XOFFTXC);
3634         adapter->stats.fcruc += E1000_READ_REG(hw, FCRUC);
3635         adapter->stats.gptc += E1000_READ_REG(hw, GPTC);
3636         adapter->stats.gotcl += E1000_READ_REG(hw, GOTCL);
3637         adapter->stats.gotch += E1000_READ_REG(hw, GOTCH);
3638         adapter->stats.rnbc += E1000_READ_REG(hw, RNBC);
3639         adapter->stats.ruc += E1000_READ_REG(hw, RUC);
3640         adapter->stats.rfc += E1000_READ_REG(hw, RFC);
3641         adapter->stats.rjc += E1000_READ_REG(hw, RJC);
3642         adapter->stats.torl += E1000_READ_REG(hw, TORL);
3643         adapter->stats.torh += E1000_READ_REG(hw, TORH);
3644         adapter->stats.totl += E1000_READ_REG(hw, TOTL);
3645         adapter->stats.toth += E1000_READ_REG(hw, TOTH);
3646         adapter->stats.tpr += E1000_READ_REG(hw, TPR);
3647
3648         if (adapter->hw.mac_type != e1000_ich8lan) {
3649                 adapter->stats.ptc64 += E1000_READ_REG(hw, PTC64);
3650                 adapter->stats.ptc127 += E1000_READ_REG(hw, PTC127);
3651                 adapter->stats.ptc255 += E1000_READ_REG(hw, PTC255);
3652                 adapter->stats.ptc511 += E1000_READ_REG(hw, PTC511);
3653                 adapter->stats.ptc1023 += E1000_READ_REG(hw, PTC1023);
3654                 adapter->stats.ptc1522 += E1000_READ_REG(hw, PTC1522);
3655         }
3656
3657         adapter->stats.mptc += E1000_READ_REG(hw, MPTC);
3658         adapter->stats.bptc += E1000_READ_REG(hw, BPTC);
3659
3660         /* used for adaptive IFS */
3661
3662         hw->tx_packet_delta = E1000_READ_REG(hw, TPT);
3663         adapter->stats.tpt += hw->tx_packet_delta;
3664         hw->collision_delta = E1000_READ_REG(hw, COLC);
3665         adapter->stats.colc += hw->collision_delta;
3666
3667         if (hw->mac_type >= e1000_82543) {
3668                 adapter->stats.algnerrc += E1000_READ_REG(hw, ALGNERRC);
3669                 adapter->stats.rxerrc += E1000_READ_REG(hw, RXERRC);
3670                 adapter->stats.tncrs += E1000_READ_REG(hw, TNCRS);
3671                 adapter->stats.cexterr += E1000_READ_REG(hw, CEXTERR);
3672                 adapter->stats.tsctc += E1000_READ_REG(hw, TSCTC);
3673                 adapter->stats.tsctfc += E1000_READ_REG(hw, TSCTFC);
3674         }
3675         if (hw->mac_type > e1000_82547_rev_2) {
3676                 adapter->stats.iac += E1000_READ_REG(hw, IAC);
3677                 adapter->stats.icrxoc += E1000_READ_REG(hw, ICRXOC);
3678
3679                 if (adapter->hw.mac_type != e1000_ich8lan) {
3680                         adapter->stats.icrxptc += E1000_READ_REG(hw, ICRXPTC);
3681                         adapter->stats.icrxatc += E1000_READ_REG(hw, ICRXATC);
3682                         adapter->stats.ictxptc += E1000_READ_REG(hw, ICTXPTC);
3683                         adapter->stats.ictxatc += E1000_READ_REG(hw, ICTXATC);
3684                         adapter->stats.ictxqec += E1000_READ_REG(hw, ICTXQEC);
3685                         adapter->stats.ictxqmtc += E1000_READ_REG(hw, ICTXQMTC);
3686                         adapter->stats.icrxdmtc += E1000_READ_REG(hw, ICRXDMTC);
3687                 }
3688         }
3689
3690         /* Fill out the OS statistics structure */
3691         adapter->net_stats.rx_packets = adapter->stats.gprc;
3692         adapter->net_stats.tx_packets = adapter->stats.gptc;
3693         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
3694         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
3695         adapter->net_stats.multicast = adapter->stats.mprc;
3696         adapter->net_stats.collisions = adapter->stats.colc;
3697
3698         /* Rx Errors */
3699
3700         /* RLEC on some newer hardware can be incorrect so build
3701         * our own version based on RUC and ROC */
3702         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3703                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3704                 adapter->stats.ruc + adapter->stats.roc +
3705                 adapter->stats.cexterr;
3706         adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3707         adapter->net_stats.rx_length_errors = adapter->stats.rlerrc;
3708         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3709         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3710         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3711
3712         /* Tx Errors */
3713         adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3714         adapter->net_stats.tx_errors = adapter->stats.txerrc;
3715         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3716         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3717         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3718         if (adapter->hw.bad_tx_carr_stats_fd &&
3719             adapter->link_duplex == FULL_DUPLEX) {
3720                 adapter->net_stats.tx_carrier_errors = 0;
3721                 adapter->stats.tncrs = 0;
3722         }
3723
3724         /* Tx Dropped needs to be maintained elsewhere */
3725
3726         /* Phy Stats */
3727         if (hw->media_type == e1000_media_type_copper) {
3728                 if ((adapter->link_speed == SPEED_1000) &&
3729                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3730                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3731                         adapter->phy_stats.idle_errors += phy_tmp;
3732                 }
3733
3734                 if ((hw->mac_type <= e1000_82546) &&
3735                    (hw->phy_type == e1000_phy_m88) &&
3736                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3737                         adapter->phy_stats.receive_errors += phy_tmp;
3738         }
3739
3740         /* Management Stats */
3741         if (adapter->hw.has_smbus) {
3742                 adapter->stats.mgptc += E1000_READ_REG(hw, MGTPTC);
3743                 adapter->stats.mgprc += E1000_READ_REG(hw, MGTPRC);
3744                 adapter->stats.mgpdc += E1000_READ_REG(hw, MGTPDC);
3745         }
3746
3747         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3748 }
3749
3750 /**
3751  * e1000_intr_msi - Interrupt Handler
3752  * @irq: interrupt number
3753  * @data: pointer to a network interface device structure
3754  **/
3755
3756 static irqreturn_t
3757 e1000_intr_msi(int irq, void *data)
3758 {
3759         struct net_device *netdev = data;
3760         struct e1000_adapter *adapter = netdev_priv(netdev);
3761         struct e1000_hw *hw = &adapter->hw;
3762 #ifndef CONFIG_E1000_NAPI
3763         int i;
3764 #endif
3765         uint32_t icr = E1000_READ_REG(hw, ICR);
3766
3767 #ifdef CONFIG_E1000_NAPI
3768         /* read ICR disables interrupts using IAM, so keep up with our
3769          * enable/disable accounting */
3770         atomic_inc(&adapter->irq_sem);
3771 #endif
3772         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3773                 hw->get_link_status = 1;
3774                 /* 80003ES2LAN workaround-- For packet buffer work-around on
3775                  * link down event; disable receives here in the ISR and reset
3776                  * adapter in watchdog */
3777                 if (netif_carrier_ok(netdev) &&
3778                     (adapter->hw.mac_type == e1000_80003es2lan)) {
3779                         /* disable receives */
3780                         uint32_t rctl = E1000_READ_REG(hw, RCTL);
3781                         E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
3782                 }
3783                 /* guard against interrupt when we're going down */
3784                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3785                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3786         }
3787
3788 #ifdef CONFIG_E1000_NAPI
3789         if (likely(netif_rx_schedule_prep(netdev))) {
3790                 adapter->total_tx_bytes = 0;
3791                 adapter->total_tx_packets = 0;
3792                 adapter->total_rx_bytes = 0;
3793                 adapter->total_rx_packets = 0;
3794                 __netif_rx_schedule(netdev);
3795         } else
3796                 e1000_irq_enable(adapter);
3797 #else
3798         adapter->total_tx_bytes = 0;
3799         adapter->total_rx_bytes = 0;
3800         adapter->total_tx_packets = 0;
3801         adapter->total_rx_packets = 0;
3802
3803         for (i = 0; i < E1000_MAX_INTR; i++)
3804                 if (unlikely(!adapter->clean_rx(adapter, adapter->rx_ring) &
3805                    !e1000_clean_tx_irq(adapter, adapter->tx_ring)))
3806                         break;
3807
3808         if (likely(adapter->itr_setting & 3))
3809                 e1000_set_itr(adapter);
3810 #endif
3811
3812         return IRQ_HANDLED;
3813 }
3814
3815 /**
3816  * e1000_intr - Interrupt Handler
3817  * @irq: interrupt number
3818  * @data: pointer to a network interface device structure
3819  **/
3820
3821 static irqreturn_t
3822 e1000_intr(int irq, void *data)
3823 {
3824         struct net_device *netdev = data;
3825         struct e1000_adapter *adapter = netdev_priv(netdev);
3826         struct e1000_hw *hw = &adapter->hw;
3827         uint32_t rctl, icr = E1000_READ_REG(hw, ICR);
3828 #ifndef CONFIG_E1000_NAPI
3829         int i;
3830 #endif
3831         if (unlikely(!icr))
3832                 return IRQ_NONE;  /* Not our interrupt */
3833
3834 #ifdef CONFIG_E1000_NAPI
3835         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
3836          * not set, then the adapter didn't send an interrupt */
3837         if (unlikely(hw->mac_type >= e1000_82571 &&
3838                      !(icr & E1000_ICR_INT_ASSERTED)))
3839                 return IRQ_NONE;
3840
3841         /* Interrupt Auto-Mask...upon reading ICR,
3842          * interrupts are masked.  No need for the
3843          * IMC write, but it does mean we should
3844          * account for it ASAP. */
3845         if (likely(hw->mac_type >= e1000_82571))
3846                 atomic_inc(&adapter->irq_sem);
3847 #endif
3848
3849         if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3850                 hw->get_link_status = 1;
3851                 /* 80003ES2LAN workaround--
3852                  * For packet buffer work-around on link down event;
3853                  * disable receives here in the ISR and
3854                  * reset adapter in watchdog
3855                  */
3856                 if (netif_carrier_ok(netdev) &&
3857                     (adapter->hw.mac_type == e1000_80003es2lan)) {
3858                         /* disable receives */
3859                         rctl = E1000_READ_REG(hw, RCTL);
3860                         E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
3861                 }
3862                 /* guard against interrupt when we're going down */
3863                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3864                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3865         }
3866
3867 #ifdef CONFIG_E1000_NAPI
3868         if (unlikely(hw->mac_type < e1000_82571)) {
3869                 /* disable interrupts, without the synchronize_irq bit */
3870                 atomic_inc(&adapter->irq_sem);
3871                 E1000_WRITE_REG(hw, IMC, ~0);
3872                 E1000_WRITE_FLUSH(hw);
3873         }
3874         if (likely(netif_rx_schedule_prep(netdev))) {
3875                 adapter->total_tx_bytes = 0;
3876                 adapter->total_tx_packets = 0;
3877                 adapter->total_rx_bytes = 0;
3878                 adapter->total_rx_packets = 0;
3879                 __netif_rx_schedule(netdev);
3880         } else
3881                 /* this really should not happen! if it does it is basically a
3882                  * bug, but not a hard error, so enable ints and continue */
3883                 e1000_irq_enable(adapter);
3884 #else
3885         /* Writing IMC and IMS is needed for 82547.
3886          * Due to Hub Link bus being occupied, an interrupt
3887          * de-assertion message is not able to be sent.
3888          * When an interrupt assertion message is generated later,
3889          * two messages are re-ordered and sent out.
3890          * That causes APIC to think 82547 is in de-assertion
3891          * state, while 82547 is in assertion state, resulting
3892          * in dead lock. Writing IMC forces 82547 into
3893          * de-assertion state.
3894          */
3895         if (hw->mac_type == e1000_82547 || hw->mac_type == e1000_82547_rev_2) {
3896                 atomic_inc(&adapter->irq_sem);
3897                 E1000_WRITE_REG(hw, IMC, ~0);
3898         }
3899
3900         adapter->total_tx_bytes = 0;
3901         adapter->total_rx_bytes = 0;
3902         adapter->total_tx_packets = 0;
3903         adapter->total_rx_packets = 0;
3904
3905         for (i = 0; i < E1000_MAX_INTR; i++)
3906                 if (unlikely(!adapter->clean_rx(adapter, adapter->rx_ring) &
3907                    !e1000_clean_tx_irq(adapter, adapter->tx_ring)))
3908                         break;
3909
3910         if (likely(adapter->itr_setting & 3))
3911                 e1000_set_itr(adapter);
3912
3913         if (hw->mac_type == e1000_82547 || hw->mac_type == e1000_82547_rev_2)
3914                 e1000_irq_enable(adapter);
3915
3916 #endif
3917         return IRQ_HANDLED;
3918 }
3919
3920 #ifdef CONFIG_E1000_NAPI
3921 /**
3922  * e1000_clean - NAPI Rx polling callback
3923  * @adapter: board private structure
3924  **/
3925
3926 static int
3927 e1000_clean(struct net_device *poll_dev, int *budget)
3928 {
3929         struct e1000_adapter *adapter;
3930         int work_to_do = min(*budget, poll_dev->quota);
3931         int tx_cleaned = 0, work_done = 0;
3932
3933         /* Must NOT use netdev_priv macro here. */
3934         adapter = poll_dev->priv;
3935
3936         /* Keep link state information with original netdev */
3937         if (!netif_carrier_ok(poll_dev))
3938                 goto quit_polling;
3939
3940         /* e1000_clean is called per-cpu.  This lock protects
3941          * tx_ring[0] from being cleaned by multiple cpus
3942          * simultaneously.  A failure obtaining the lock means
3943          * tx_ring[0] is currently being cleaned anyway. */
3944         if (spin_trylock(&adapter->tx_queue_lock)) {
3945                 tx_cleaned = e1000_clean_tx_irq(adapter,
3946                                                 &adapter->tx_ring[0]);
3947                 spin_unlock(&adapter->tx_queue_lock);
3948         }
3949
3950         adapter->clean_rx(adapter, &adapter->rx_ring[0],
3951                           &work_done, work_to_do);
3952
3953         *budget -= work_done;
3954         poll_dev->quota -= work_done;
3955
3956         /* If no Tx and not enough Rx work done, exit the polling mode */
3957         if ((!tx_cleaned && (work_done == 0)) ||
3958            !netif_running(poll_dev)) {
3959 quit_polling:
3960                 if (likely(adapter->itr_setting & 3))
3961                         e1000_set_itr(adapter);
3962                 netif_rx_complete(poll_dev);
3963                 e1000_irq_enable(adapter);
3964                 return 0;
3965         }
3966
3967         return 1;
3968 }
3969
3970 #endif
3971 /**
3972  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3973  * @adapter: board private structure
3974  **/
3975
3976 static boolean_t
3977 e1000_clean_tx_irq(struct e1000_adapter *adapter,
3978                    struct e1000_tx_ring *tx_ring)
3979 {
3980         struct net_device *netdev = adapter->netdev;
3981         struct e1000_tx_desc *tx_desc, *eop_desc;
3982         struct e1000_buffer *buffer_info;
3983         unsigned int i, eop;
3984 #ifdef CONFIG_E1000_NAPI
3985         unsigned int count = 0;
3986 #endif
3987         boolean_t cleaned = FALSE;
3988         unsigned int total_tx_bytes=0, total_tx_packets=0;
3989
3990         i = tx_ring->next_to_clean;
3991         eop = tx_ring->buffer_info[i].next_to_watch;
3992         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3993
3994         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
3995                 for (cleaned = FALSE; !cleaned; ) {
3996                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3997                         buffer_info = &tx_ring->buffer_info[i];
3998                         cleaned = (i == eop);
3999
4000                         if (cleaned) {
4001                                 struct sk_buff *skb = buffer_info->skb;
4002                                 unsigned int segs, bytecount;
4003                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
4004                                 /* multiply data chunks by size of headers */
4005                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
4006                                             skb->len;
4007                                 total_tx_packets += segs;
4008                                 total_tx_bytes += bytecount;
4009                         }
4010                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
4011                         tx_desc->upper.data = 0;
4012
4013                         if (unlikely(++i == tx_ring->count)) i = 0;
4014                 }
4015
4016                 eop = tx_ring->buffer_info[i].next_to_watch;
4017                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
4018 #ifdef CONFIG_E1000_NAPI
4019 #define E1000_TX_WEIGHT 64
4020                 /* weight of a sort for tx, to avoid endless transmit cleanup */
4021                 if (count++ == E1000_TX_WEIGHT) break;
4022 #endif
4023         }
4024
4025         tx_ring->next_to_clean = i;
4026
4027 #define TX_WAKE_THRESHOLD 32
4028         if (unlikely(cleaned && netif_carrier_ok(netdev) &&
4029                      E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
4030                 /* Make sure that anybody stopping the queue after this
4031                  * sees the new next_to_clean.
4032                  */
4033                 smp_mb();
4034                 if (netif_queue_stopped(netdev)) {
4035                         netif_wake_queue(netdev);
4036                         ++adapter->restart_queue;
4037                 }
4038         }
4039
4040         if (adapter->detect_tx_hung) {
4041                 /* Detect a transmit hang in hardware, this serializes the
4042                  * check with the clearing of time_stamp and movement of i */
4043                 adapter->detect_tx_hung = FALSE;
4044                 if (tx_ring->buffer_info[eop].dma &&
4045                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
4046                                (adapter->tx_timeout_factor * HZ))
4047                     && !(E1000_READ_REG(&adapter->hw, STATUS) &
4048                          E1000_STATUS_TXOFF)) {
4049
4050                         /* detected Tx unit hang */
4051                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
4052                                         "  Tx Queue             <%lu>\n"
4053                                         "  TDH                  <%x>\n"
4054                                         "  TDT                  <%x>\n"
4055                                         "  next_to_use          <%x>\n"
4056                                         "  next_to_clean        <%x>\n"
4057                                         "buffer_info[next_to_clean]\n"
4058                                         "  time_stamp           <%lx>\n"
4059                                         "  next_to_watch        <%x>\n"
4060                                         "  jiffies              <%lx>\n"
4061                                         "  next_to_watch.status <%x>\n",
4062                                 (unsigned long)((tx_ring - adapter->tx_ring) /
4063                                         sizeof(struct e1000_tx_ring)),
4064                                 readl(adapter->hw.hw_addr + tx_ring->tdh),
4065                                 readl(adapter->hw.hw_addr + tx_ring->tdt),
4066                                 tx_ring->next_to_use,
4067                                 tx_ring->next_to_clean,
4068                                 tx_ring->buffer_info[eop].time_stamp,
4069                                 eop,
4070                                 jiffies,
4071                                 eop_desc->upper.fields.status);
4072                         netif_stop_queue(netdev);
4073                 }
4074         }
4075         adapter->total_tx_bytes += total_tx_bytes;
4076         adapter->total_tx_packets += total_tx_packets;
4077         return cleaned;
4078 }
4079
4080 /**
4081  * e1000_rx_checksum - Receive Checksum Offload for 82543
4082  * @adapter:     board private structure
4083  * @status_err:  receive descriptor status and error fields
4084  * @csum:        receive descriptor csum field
4085  * @sk_buff:     socket buffer with received data
4086  **/
4087
4088 static void
4089 e1000_rx_checksum(struct e1000_adapter *adapter,
4090                   uint32_t status_err, uint32_t csum,
4091                   struct sk_buff *skb)
4092 {
4093         uint16_t status = (uint16_t)status_err;
4094         uint8_t errors = (uint8_t)(status_err >> 24);
4095         skb->ip_summed = CHECKSUM_NONE;
4096
4097         /* 82543 or newer only */
4098         if (unlikely(adapter->hw.mac_type < e1000_82543)) return;
4099         /* Ignore Checksum bit is set */
4100         if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
4101         /* TCP/UDP checksum error bit is set */
4102         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
4103                 /* let the stack verify checksum errors */
4104                 adapter->hw_csum_err++;
4105                 return;
4106         }
4107         /* TCP/UDP Checksum has not been calculated */
4108         if (adapter->hw.mac_type <= e1000_82547_rev_2) {
4109                 if (!(status & E1000_RXD_STAT_TCPCS))
4110                         return;
4111         } else {
4112                 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
4113                         return;
4114         }
4115         /* It must be a TCP or UDP packet with a valid checksum */
4116         if (likely(status & E1000_RXD_STAT_TCPCS)) {
4117                 /* TCP checksum is good */
4118                 skb->ip_summed = CHECKSUM_UNNECESSARY;
4119         } else if (adapter->hw.mac_type > e1000_82547_rev_2) {
4120                 /* IP fragment with UDP payload */
4121                 /* Hardware complements the payload checksum, so we undo it
4122                  * and then put the value in host order for further stack use.
4123                  */
4124                 csum = ntohl(csum ^ 0xFFFF);
4125                 skb->csum = csum;
4126                 skb->ip_summed = CHECKSUM_COMPLETE;
4127         }
4128         adapter->hw_csum_good++;
4129 }
4130
4131 /**
4132  * e1000_clean_rx_irq - Send received data up the network stack; legacy
4133  * @adapter: board private structure
4134  **/
4135
4136 static boolean_t
4137 #ifdef CONFIG_E1000_NAPI
4138 e1000_clean_rx_irq(struct e1000_adapter *adapter,
4139                    struct e1000_rx_ring *rx_ring,
4140                    int *work_done, int work_to_do)
4141 #else
4142 e1000_clean_rx_irq(struct e1000_adapter *adapter,
4143                    struct e1000_rx_ring *rx_ring)
4144 #endif
4145 {
4146         struct net_device *netdev = adapter->netdev;
4147         struct pci_dev *pdev = adapter->pdev;
4148         struct e1000_rx_desc *rx_desc, *next_rxd;
4149         struct e1000_buffer *buffer_info, *next_buffer;
4150         unsigned long flags;
4151         uint32_t length;
4152         uint8_t last_byte;
4153         unsigned int i;
4154         int cleaned_count = 0;
4155         boolean_t cleaned = FALSE;
4156         unsigned int total_rx_bytes=0, total_rx_packets=0;
4157
4158         i = rx_ring->next_to_clean;
4159         rx_desc = E1000_RX_DESC(*rx_ring, i);
4160         buffer_info = &rx_ring->buffer_info[i];
4161
4162         while (rx_desc->status & E1000_RXD_STAT_DD) {
4163                 struct sk_buff *skb;
4164                 u8 status;
4165
4166 #ifdef CONFIG_E1000_NAPI
4167                 if (*work_done >= work_to_do)
4168                         break;
4169                 (*work_done)++;
4170 #endif
4171                 status = rx_desc->status;
4172                 skb = buffer_info->skb;
4173                 buffer_info->skb = NULL;
4174
4175                 prefetch(skb->data - NET_IP_ALIGN);
4176
4177                 if (++i == rx_ring->count) i = 0;
4178                 next_rxd = E1000_RX_DESC(*rx_ring, i);
4179                 prefetch(next_rxd);
4180
4181                 next_buffer = &rx_ring->buffer_info[i];
4182
4183                 cleaned = TRUE;
4184                 cleaned_count++;
4185                 pci_unmap_single(pdev,
4186                                  buffer_info->dma,
4187                                  buffer_info->length,
4188                                  PCI_DMA_FROMDEVICE);
4189
4190                 length = le16_to_cpu(rx_desc->length);
4191
4192                 if (unlikely(!(status & E1000_RXD_STAT_EOP))) {
4193                         /* All receives must fit into a single buffer */
4194                         E1000_DBG("%s: Receive packet consumed multiple"
4195                                   " buffers\n", netdev->name);
4196                         /* recycle */
4197                         buffer_info->skb = skb;
4198                         goto next_desc;
4199                 }
4200
4201                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
4202                         last_byte = *(skb->data + length - 1);
4203                         if (TBI_ACCEPT(&adapter->hw, status,
4204                                       rx_desc->errors, length, last_byte)) {
4205                                 spin_lock_irqsave(&adapter->stats_lock, flags);
4206                                 e1000_tbi_adjust_stats(&adapter->hw,
4207                                                        &adapter->stats,
4208                                                        length, skb->data);
4209                                 spin_unlock_irqrestore(&adapter->stats_lock,
4210                                                        flags);
4211                                 length--;
4212                         } else {
4213                                 /* recycle */
4214                                 buffer_info->skb = skb;
4215                                 goto next_desc;
4216                         }
4217                 }
4218
4219                 /* adjust length to remove Ethernet CRC, this must be
4220                  * done after the TBI_ACCEPT workaround above */
4221                 length -= 4;
4222
4223                 /* probably a little skewed due to removing CRC */
4224                 total_rx_bytes += length;
4225                 total_rx_packets++;
4226
4227                 /* code added for copybreak, this should improve
4228                  * performance for small packets with large amounts
4229                  * of reassembly being done in the stack */
4230                 if (length < copybreak) {
4231                         struct sk_buff *new_skb =
4232                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
4233                         if (new_skb) {
4234                                 skb_reserve(new_skb, NET_IP_ALIGN);
4235                                 skb_copy_to_linear_data_offset(new_skb,
4236                                                                -NET_IP_ALIGN,
4237                                                                (skb->data -
4238                                                                 NET_IP_ALIGN),
4239                                                                (length +
4240                                                                 NET_IP_ALIGN));
4241                                 /* save the skb in buffer_info as good */
4242                                 buffer_info->skb = skb;
4243                                 skb = new_skb;
4244                         }
4245                         /* else just continue with the old one */
4246                 }
4247                 /* end copybreak code */
4248                 skb_put(skb, length);
4249
4250                 /* Receive Checksum Offload */
4251                 e1000_rx_checksum(adapter,
4252                                   (uint32_t)(status) |
4253                                   ((uint32_t)(rx_desc->errors) << 24),
4254                                   le16_to_cpu(rx_desc->csum), skb);
4255
4256                 skb->protocol = eth_type_trans(skb, netdev);
4257 #ifdef CONFIG_E1000_NAPI
4258                 if (unlikely(adapter->vlgrp &&
4259                             (status & E1000_RXD_STAT_VP))) {
4260                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
4261                                                  le16_to_cpu(rx_desc->special) &
4262                                                  E1000_RXD_SPC_VLAN_MASK);
4263                 } else {
4264                         netif_receive_skb(skb);
4265                 }
4266 #else /* CONFIG_E1000_NAPI */
4267                 if (unlikely(adapter->vlgrp &&
4268                             (status & E1000_RXD_STAT_VP))) {
4269                         vlan_hwaccel_rx(skb, adapter->vlgrp,
4270                                         le16_to_cpu(rx_desc->special) &
4271                                         E1000_RXD_SPC_VLAN_MASK);
4272                 } else {
4273                         netif_rx(skb);
4274                 }
4275 #endif /* CONFIG_E1000_NAPI */
4276                 netdev->last_rx = jiffies;
4277
4278 next_desc:
4279                 rx_desc->status = 0;
4280
4281                 /* return some buffers to hardware, one at a time is too slow */
4282                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4283                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4284                         cleaned_count = 0;
4285                 }
4286
4287                 /* use prefetched values */
4288                 rx_desc = next_rxd;
4289                 buffer_info = next_buffer;
4290         }
4291         rx_ring->next_to_clean = i;
4292
4293         cleaned_count = E1000_DESC_UNUSED(rx_ring);
4294         if (cleaned_count)
4295                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4296
4297         adapter->total_rx_packets += total_rx_packets;
4298         adapter->total_rx_bytes += total_rx_bytes;
4299         return cleaned;
4300 }
4301
4302 /**
4303  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
4304  * @adapter: board private structure
4305  **/
4306
4307 static boolean_t
4308 #ifdef CONFIG_E1000_NAPI
4309 e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
4310                       struct e1000_rx_ring *rx_ring,
4311                       int *work_done, int work_to_do)
4312 #else
4313 e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
4314                       struct e1000_rx_ring *rx_ring)
4315 #endif
4316 {
4317         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
4318         struct net_device *netdev = adapter->netdev;
4319         struct pci_dev *pdev = adapter->pdev;
4320         struct e1000_buffer *buffer_info, *next_buffer;
4321         struct e1000_ps_page *ps_page;
4322         struct e1000_ps_page_dma *ps_page_dma;
4323         struct sk_buff *skb;
4324         unsigned int i, j;
4325         uint32_t length, staterr;
4326         int cleaned_count = 0;
4327         boolean_t cleaned = FALSE;
4328         unsigned int total_rx_bytes=0, total_rx_packets=0;
4329
4330         i = rx_ring->next_to_clean;
4331         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
4332         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
4333         buffer_info = &rx_ring->buffer_info[i];
4334
4335         while (staterr & E1000_RXD_STAT_DD) {
4336                 ps_page = &rx_ring->ps_page[i];
4337                 ps_page_dma = &rx_ring->ps_page_dma[i];
4338 #ifdef CONFIG_E1000_NAPI
4339                 if (unlikely(*work_done >= work_to_do))
4340                         break;
4341                 (*work_done)++;
4342 #endif
4343                 skb = buffer_info->skb;
4344
4345                 /* in the packet split case this is header only */
4346                 prefetch(skb->data - NET_IP_ALIGN);
4347
4348                 if (++i == rx_ring->count) i = 0;
4349                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
4350                 prefetch(next_rxd);
4351
4352                 next_buffer = &rx_ring->buffer_info[i];
4353
4354                 cleaned = TRUE;
4355                 cleaned_count++;
4356                 pci_unmap_single(pdev, buffer_info->dma,
4357                                  buffer_info->length,
4358                                  PCI_DMA_FROMDEVICE);
4359
4360                 if (unlikely(!(staterr & E1000_RXD_STAT_EOP))) {
4361                         E1000_DBG("%s: Packet Split buffers didn't pick up"
4362                                   " the full packet\n", netdev->name);
4363                         dev_kfree_skb_irq(skb);
4364                         goto next_desc;
4365                 }
4366
4367                 if (unlikely(staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK)) {
4368                         dev_kfree_skb_irq(skb);
4369                         goto next_desc;
4370                 }
4371
4372                 length = le16_to_cpu(rx_desc->wb.middle.length0);
4373
4374                 if (unlikely(!length)) {
4375                         E1000_DBG("%s: Last part of the packet spanning"
4376                                   " multiple descriptors\n", netdev->name);
4377                         dev_kfree_skb_irq(skb);
4378                         goto next_desc;
4379                 }
4380
4381                 /* Good Receive */
4382                 skb_put(skb, length);
4383
4384                 {
4385                 /* this looks ugly, but it seems compiler issues make it
4386                    more efficient than reusing j */
4387                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
4388
4389                 /* page alloc/put takes too long and effects small packet
4390                  * throughput, so unsplit small packets and save the alloc/put*/
4391                 if (l1 && (l1 <= copybreak) && ((length + l1) <= adapter->rx_ps_bsize0)) {
4392                         u8 *vaddr;
4393                         /* there is no documentation about how to call
4394                          * kmap_atomic, so we can't hold the mapping
4395                          * very long */
4396                         pci_dma_sync_single_for_cpu(pdev,
4397                                 ps_page_dma->ps_page_dma[0],
4398                                 PAGE_SIZE,
4399                                 PCI_DMA_FROMDEVICE);
4400                         vaddr = kmap_atomic(ps_page->ps_page[0],
4401                                             KM_SKB_DATA_SOFTIRQ);
4402                         memcpy(skb_tail_pointer(skb), vaddr, l1);
4403                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
4404                         pci_dma_sync_single_for_device(pdev,
4405                                 ps_page_dma->ps_page_dma[0],
4406                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
4407                         /* remove the CRC */
4408                         l1 -= 4;
4409                         skb_put(skb, l1);
4410                         goto copydone;
4411                 } /* if */
4412                 }
4413
4414                 for (j = 0; j < adapter->rx_ps_pages; j++) {
4415                         if (!(length= le16_to_cpu(rx_desc->wb.upper.length[j])))
4416                                 break;
4417                         pci_unmap_page(pdev, ps_page_dma->ps_page_dma[j],
4418                                         PAGE_SIZE, PCI_DMA_FROMDEVICE);
4419                         ps_page_dma->ps_page_dma[j] = 0;
4420                         skb_fill_page_desc(skb, j, ps_page->ps_page[j], 0,
4421                                            length);
4422                         ps_page->ps_page[j] = NULL;
4423                         skb->len += length;
4424                         skb->data_len += length;
4425                         skb->truesize += length;
4426                 }
4427
4428                 /* strip the ethernet crc, problem is we're using pages now so
4429                  * this whole operation can get a little cpu intensive */
4430                 pskb_trim(skb, skb->len - 4);
4431
4432 copydone:
4433                 total_rx_bytes += skb->len;
4434                 total_rx_packets++;
4435
4436                 e1000_rx_checksum(adapter, staterr,
4437                                   le16_to_cpu(rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
4438                 skb->protocol = eth_type_trans(skb, netdev);
4439
4440                 if (likely(rx_desc->wb.upper.header_status &
4441                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP)))
4442                         adapter->rx_hdr_split++;
4443 #ifdef CONFIG_E1000_NAPI
4444                 if (unlikely(adapter->vlgrp && (staterr & E1000_RXD_STAT_VP))) {
4445                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
4446                                 le16_to_cpu(rx_desc->wb.middle.vlan) &
4447                                 E1000_RXD_SPC_VLAN_MASK);
4448                 } else {
4449                         netif_receive_skb(skb);
4450                 }
4451 #else /* CONFIG_E1000_NAPI */
4452                 if (unlikely(adapter->vlgrp && (staterr & E1000_RXD_STAT_VP))) {
4453                         vlan_hwaccel_rx(skb, adapter->vlgrp,
4454                                 le16_to_cpu(rx_desc->wb.middle.vlan) &
4455                                 E1000_RXD_SPC_VLAN_MASK);
4456                 } else {
4457                         netif_rx(skb);
4458                 }
4459 #endif /* CONFIG_E1000_NAPI */
4460                 netdev->last_rx = jiffies;
4461
4462 next_desc:
4463                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
4464                 buffer_info->skb = NULL;
4465
4466                 /* return some buffers to hardware, one at a time is too slow */
4467                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4468                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4469                         cleaned_count = 0;
4470                 }
4471
4472                 /* use prefetched values */
4473                 rx_desc = next_rxd;
4474                 buffer_info = next_buffer;
4475
4476                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
4477         }
4478         rx_ring->next_to_clean = i;
4479
4480         cleaned_count = E1000_DESC_UNUSED(rx_ring);
4481         if (cleaned_count)
4482                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4483
4484         adapter->total_rx_packets += total_rx_packets;
4485         adapter->total_rx_bytes += total_rx_bytes;
4486         return cleaned;
4487 }
4488
4489 /**
4490  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4491  * @adapter: address of board private structure
4492  **/
4493
4494 static void
4495 e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4496                        struct e1000_rx_ring *rx_ring,
4497                        int cleaned_count)
4498 {
4499         struct net_device *netdev = adapter->netdev;
4500         struct pci_dev *pdev = adapter->pdev;
4501         struct e1000_rx_desc *rx_desc;
4502         struct e1000_buffer *buffer_info;
4503         struct sk_buff *skb;
4504         unsigned int i;
4505         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
4506
4507         i = rx_ring->next_to_use;
4508         buffer_info = &rx_ring->buffer_info[i];
4509
4510         while (cleaned_count--) {
4511                 skb = buffer_info->skb;
4512                 if (skb) {
4513                         skb_trim(skb, 0);
4514                         goto map_skb;
4515                 }
4516
4517                 skb = netdev_alloc_skb(netdev, bufsz);
4518                 if (unlikely(!skb)) {
4519                         /* Better luck next round */
4520                         adapter->alloc_rx_buff_failed++;
4521                         break;
4522                 }
4523
4524                 /* Fix for errata 23, can't cross 64kB boundary */
4525                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4526                         struct sk_buff *oldskb = skb;
4527                         DPRINTK(RX_ERR, ERR, "skb align check failed: %u bytes "
4528                                              "at %p\n", bufsz, skb->data);
4529                         /* Try again, without freeing the previous */
4530                         skb = netdev_alloc_skb(netdev, bufsz);
4531                         /* Failed allocation, critical failure */
4532                         if (!skb) {
4533                                 dev_kfree_skb(oldskb);
4534                                 break;
4535                         }
4536
4537                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4538                                 /* give up */
4539                                 dev_kfree_skb(skb);
4540                                 dev_kfree_skb(oldskb);
4541                                 break; /* while !buffer_info->skb */
4542                         }
4543
4544                         /* Use new allocation */
4545                         dev_kfree_skb(oldskb);
4546                 }
4547                 /* Make buffer alignment 2 beyond a 16 byte boundary
4548                  * this will result in a 16 byte aligned IP header after
4549                  * the 14 byte MAC header is removed
4550                  */
4551                 skb_reserve(skb, NET_IP_ALIGN);
4552
4553                 buffer_info->skb = skb;
4554                 buffer_info->length = adapter->rx_buffer_len;
4555 map_skb:
4556                 buffer_info->dma = pci_map_single(pdev,
4557                                                   skb->data,
4558                                                   adapter->rx_buffer_len,
4559                                                   PCI_DMA_FROMDEVICE);
4560
4561                 /* Fix for errata 23, can't cross 64kB boundary */
4562                 if (!e1000_check_64k_bound(adapter,
4563                                         (void *)(unsigned long)buffer_info->dma,
4564                                         adapter->rx_buffer_len)) {
4565                         DPRINTK(RX_ERR, ERR,
4566                                 "dma align check failed: %u bytes at %p\n",
4567                                 adapter->rx_buffer_len,
4568                                 (void *)(unsigned long)buffer_info->dma);
4569                         dev_kfree_skb(skb);
4570                         buffer_info->skb = NULL;
4571
4572                         pci_unmap_single(pdev, buffer_info->dma,
4573                                          adapter->rx_buffer_len,
4574                                          PCI_DMA_FROMDEVICE);
4575
4576                         break; /* while !buffer_info->skb */
4577                 }
4578                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4579                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4580
4581                 if (unlikely(++i == rx_ring->count))
4582                         i = 0;
4583                 buffer_info = &rx_ring->buffer_info[i];
4584         }
4585
4586         if (likely(rx_ring->next_to_use != i)) {
4587                 rx_ring->next_to_use = i;
4588                 if (unlikely(i-- == 0))
4589                         i = (rx_ring->count - 1);
4590
4591                 /* Force memory writes to complete before letting h/w
4592                  * know there are new descriptors to fetch.  (Only
4593                  * applicable for weak-ordered memory model archs,
4594                  * such as IA-64). */
4595                 wmb();
4596                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4597         }
4598 }
4599
4600 /**
4601  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
4602  * @adapter: address of board private structure
4603  **/
4604
4605 static void
4606 e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
4607                           struct e1000_rx_ring *rx_ring,
4608                           int cleaned_count)
4609 {
4610         struct net_device *netdev = adapter->netdev;
4611         struct pci_dev *pdev = adapter->pdev;
4612         union e1000_rx_desc_packet_split *rx_desc;
4613         struct e1000_buffer *buffer_info;
4614         struct e1000_ps_page *ps_page;
4615         struct e1000_ps_page_dma *ps_page_dma;
4616         struct sk_buff *skb;
4617         unsigned int i, j;
4618
4619         i = rx_ring->next_to_use;
4620         buffer_info = &rx_ring->buffer_info[i];
4621         ps_page = &rx_ring->ps_page[i];
4622         ps_page_dma = &rx_ring->ps_page_dma[i];
4623
4624         while (cleaned_count--) {
4625                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
4626
4627                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
4628                         if (j < adapter->rx_ps_pages) {
4629                                 if (likely(!ps_page->ps_page[j])) {
4630                                         ps_page->ps_page[j] =
4631                                                 alloc_page(GFP_ATOMIC);
4632                                         if (unlikely(!ps_page->ps_page[j])) {
4633                                                 adapter->alloc_rx_buff_failed++;
4634                                                 goto no_buffers;
4635                                         }
4636                                         ps_page_dma->ps_page_dma[j] =
4637                                                 pci_map_page(pdev,
4638                                                             ps_page->ps_page[j],
4639                                                             0, PAGE_SIZE,
4640                                                             PCI_DMA_FROMDEVICE);
4641                                 }
4642                                 /* Refresh the desc even if buffer_addrs didn't
4643                                  * change because each write-back erases
4644                                  * this info.
4645                                  */
4646                                 rx_desc->read.buffer_addr[j+1] =
4647                                      cpu_to_le64(ps_page_dma->ps_page_dma[j]);
4648                         } else
4649                                 rx_desc->read.buffer_addr[j+1] = ~0;
4650                 }
4651
4652                 skb = netdev_alloc_skb(netdev,
4653                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
4654
4655                 if (unlikely(!skb)) {
4656                         adapter->alloc_rx_buff_failed++;
4657                         break;
4658                 }
4659
4660                 /* Make buffer alignment 2 beyond a 16 byte boundary
4661                  * this will result in a 16 byte aligned IP header after
4662                  * the 14 byte MAC header is removed
4663                  */
4664                 skb_reserve(skb, NET_IP_ALIGN);
4665
4666                 buffer_info->skb = skb;
4667                 buffer_info->length = adapter->rx_ps_bsize0;
4668                 buffer_info->dma = pci_map_single(pdev, skb->data,
4669                                                   adapter->rx_ps_bsize0,
4670                                                   PCI_DMA_FROMDEVICE);
4671
4672                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
4673
4674                 if (unlikely(++i == rx_ring->count)) i = 0;
4675                 buffer_info = &rx_ring->buffer_info[i];
4676                 ps_page = &rx_ring->ps_page[i];
4677                 ps_page_dma = &rx_ring->ps_page_dma[i];
4678         }
4679
4680 no_buffers:
4681         if (likely(rx_ring->next_to_use != i)) {
4682                 rx_ring->next_to_use = i;
4683                 if (unlikely(i-- == 0)) i = (rx_ring->count - 1);
4684
4685                 /* Force memory writes to complete before letting h/w
4686                  * know there are new descriptors to fetch.  (Only
4687                  * applicable for weak-ordered memory model archs,
4688                  * such as IA-64). */
4689                 wmb();
4690                 /* Hardware increments by 16 bytes, but packet split
4691                  * descriptors are 32 bytes...so we increment tail
4692                  * twice as much.
4693                  */
4694                 writel(i<<1, adapter->hw.hw_addr + rx_ring->rdt);
4695         }
4696 }
4697
4698 /**
4699  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4700  * @adapter:
4701  **/
4702
4703 static void
4704 e1000_smartspeed(struct e1000_adapter *adapter)
4705 {
4706         uint16_t phy_status;
4707         uint16_t phy_ctrl;
4708
4709         if ((adapter->hw.phy_type != e1000_phy_igp) || !adapter->hw.autoneg ||
4710            !(adapter->hw.autoneg_advertised & ADVERTISE_1000_FULL))
4711                 return;
4712
4713         if (adapter->smartspeed == 0) {
4714                 /* If Master/Slave config fault is asserted twice,
4715                  * we assume back-to-back */
4716                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_status);
4717                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4718                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_status);
4719                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4720                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_ctrl);
4721                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4722                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4723                         e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL,
4724                                             phy_ctrl);
4725                         adapter->smartspeed++;
4726                         if (!e1000_phy_setup_autoneg(&adapter->hw) &&
4727                            !e1000_read_phy_reg(&adapter->hw, PHY_CTRL,
4728                                                &phy_ctrl)) {
4729                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4730                                              MII_CR_RESTART_AUTO_NEG);
4731                                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL,
4732                                                     phy_ctrl);
4733                         }
4734                 }
4735                 return;
4736         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4737                 /* If still no link, perhaps using 2/3 pair cable */
4738                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_ctrl);
4739                 phy_ctrl |= CR_1000T_MS_ENABLE;
4740                 e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL, phy_ctrl);
4741                 if (!e1000_phy_setup_autoneg(&adapter->hw) &&
4742                    !e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &phy_ctrl)) {
4743                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4744                                      MII_CR_RESTART_AUTO_NEG);
4745                         e1000_write_phy_reg(&adapter->hw, PHY_CTRL, phy_ctrl);
4746                 }
4747         }
4748         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4749         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4750                 adapter->smartspeed = 0;
4751 }
4752
4753 /**
4754  * e1000_ioctl -
4755  * @netdev:
4756  * @ifreq:
4757  * @cmd:
4758  **/
4759
4760 static int
4761 e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4762 {
4763         switch (cmd) {
4764         case SIOCGMIIPHY:
4765         case SIOCGMIIREG:
4766         case SIOCSMIIREG:
4767                 return e1000_mii_ioctl(netdev, ifr, cmd);
4768         default:
4769                 return -EOPNOTSUPP;
4770         }
4771 }
4772
4773 /**
4774  * e1000_mii_ioctl -
4775  * @netdev:
4776  * @ifreq:
4777  * @cmd:
4778  **/
4779
4780 static int
4781 e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4782 {
4783         struct e1000_adapter *adapter = netdev_priv(netdev);
4784         struct mii_ioctl_data *data = if_mii(ifr);
4785         int retval;
4786         uint16_t mii_reg;
4787         uint16_t spddplx;
4788         unsigned long flags;
4789
4790         if (adapter->hw.media_type != e1000_media_type_copper)
4791                 return -EOPNOTSUPP;
4792
4793         switch (cmd) {
4794         case SIOCGMIIPHY:
4795                 data->phy_id = adapter->hw.phy_addr;
4796                 break;
4797         case SIOCGMIIREG:
4798                 if (!capable(CAP_NET_ADMIN))
4799                         return -EPERM;
4800                 spin_lock_irqsave(&adapter->stats_lock, flags);
4801                 if (e1000_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
4802                                    &data->val_out)) {
4803                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4804                         return -EIO;
4805                 }
4806                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4807                 break;
4808         case SIOCSMIIREG:
4809                 if (!capable(CAP_NET_ADMIN))
4810                         return -EPERM;
4811                 if (data->reg_num & ~(0x1F))
4812                         return -EFAULT;
4813                 mii_reg = data->val_in;
4814                 spin_lock_irqsave(&adapter->stats_lock, flags);
4815                 if (e1000_write_phy_reg(&adapter->hw, data->reg_num,
4816                                         mii_reg)) {
4817                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4818                         return -EIO;
4819                 }
4820                 if (adapter->hw.media_type == e1000_media_type_copper) {
4821                         switch (data->reg_num) {
4822                         case PHY_CTRL:
4823                                 if (mii_reg & MII_CR_POWER_DOWN)
4824                                         break;
4825                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4826                                         adapter->hw.autoneg = 1;
4827                                         adapter->hw.autoneg_advertised = 0x2F;
4828                                 } else {
4829                                         if (mii_reg & 0x40)
4830                                                 spddplx = SPEED_1000;
4831                                         else if (mii_reg & 0x2000)
4832                                                 spddplx = SPEED_100;
4833                                         else
4834                                                 spddplx = SPEED_10;
4835                                         spddplx += (mii_reg & 0x100)
4836                                                    ? DUPLEX_FULL :
4837                                                    DUPLEX_HALF;
4838                                         retval = e1000_set_spd_dplx(adapter,
4839                                                                     spddplx);
4840                                         if (retval) {
4841                                                 spin_unlock_irqrestore(
4842                                                         &adapter->stats_lock,
4843                                                         flags);
4844                                                 return retval;
4845                                         }
4846                                 }
4847                                 if (netif_running(adapter->netdev))
4848                                         e1000_reinit_locked(adapter);
4849                                 else
4850                                         e1000_reset(adapter);
4851                                 break;
4852                         case M88E1000_PHY_SPEC_CTRL:
4853                         case M88E1000_EXT_PHY_SPEC_CTRL:
4854                                 if (e1000_phy_reset(&adapter->hw)) {
4855                                         spin_unlock_irqrestore(
4856                                                 &adapter->stats_lock, flags);
4857                                         return -EIO;
4858                                 }
4859                                 break;
4860                         }
4861                 } else {
4862                         switch (data->reg_num) {
4863                         case PHY_CTRL:
4864                                 if (mii_reg & MII_CR_POWER_DOWN)
4865                                         break;
4866                                 if (netif_running(adapter->netdev))
4867                                         e1000_reinit_locked(adapter);
4868                                 else
4869                                         e1000_reset(adapter);
4870                                 break;
4871                         }
4872                 }
4873                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4874                 break;
4875         default:
4876                 return -EOPNOTSUPP;
4877         }
4878         return E1000_SUCCESS;
4879 }
4880
4881 void
4882 e1000_pci_set_mwi(struct e1000_hw *hw)
4883 {
4884         struct e1000_adapter *adapter = hw->back;
4885         int ret_val = pci_set_mwi(adapter->pdev);
4886
4887         if (ret_val)
4888                 DPRINTK(PROBE, ERR, "Error in setting MWI\n");
4889 }
4890
4891 void
4892 e1000_pci_clear_mwi(struct e1000_hw *hw)
4893 {
4894         struct e1000_adapter *adapter = hw->back;
4895
4896         pci_clear_mwi(adapter->pdev);
4897 }
4898
4899 void
4900 e1000_read_pci_cfg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4901 {
4902         struct e1000_adapter *adapter = hw->back;
4903
4904         pci_read_config_word(adapter->pdev, reg, value);
4905 }
4906
4907 void
4908 e1000_write_pci_cfg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4909 {
4910         struct e1000_adapter *adapter = hw->back;
4911
4912         pci_write_config_word(adapter->pdev, reg, *value);
4913 }
4914
4915 int32_t
4916 e1000_read_pcie_cap_reg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4917 {
4918     struct e1000_adapter *adapter = hw->back;
4919     uint16_t cap_offset;
4920
4921     cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
4922     if (!cap_offset)
4923         return -E1000_ERR_CONFIG;
4924
4925     pci_read_config_word(adapter->pdev, cap_offset + reg, value);
4926
4927     return E1000_SUCCESS;
4928 }
4929
4930 void
4931 e1000_io_write(struct e1000_hw *hw, unsigned long port, uint32_t value)
4932 {
4933         outl(value, port);
4934 }
4935
4936 static void
4937 e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
4938 {
4939         struct e1000_adapter *adapter = netdev_priv(netdev);
4940         uint32_t ctrl, rctl;
4941
4942         e1000_irq_disable(adapter);
4943         adapter->vlgrp = grp;
4944
4945         if (grp) {
4946                 /* enable VLAN tag insert/strip */
4947                 ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4948                 ctrl |= E1000_CTRL_VME;
4949                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4950
4951                 if (adapter->hw.mac_type != e1000_ich8lan) {
4952                         /* enable VLAN receive filtering */
4953                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
4954                         rctl |= E1000_RCTL_VFE;
4955                         rctl &= ~E1000_RCTL_CFIEN;
4956                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4957                         e1000_update_mng_vlan(adapter);
4958                 }
4959         } else {
4960                 /* disable VLAN tag insert/strip */
4961                 ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4962                 ctrl &= ~E1000_CTRL_VME;
4963                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4964
4965                 if (adapter->hw.mac_type != e1000_ich8lan) {
4966                         /* disable VLAN filtering */
4967                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
4968                         rctl &= ~E1000_RCTL_VFE;
4969                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4970                         if (adapter->mng_vlan_id !=
4971                             (uint16_t)E1000_MNG_VLAN_NONE) {
4972                                 e1000_vlan_rx_kill_vid(netdev,
4973                                                        adapter->mng_vlan_id);
4974                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4975                         }
4976                 }
4977         }
4978
4979         e1000_irq_enable(adapter);
4980 }
4981
4982 static void
4983 e1000_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid)
4984 {
4985         struct e1000_adapter *adapter = netdev_priv(netdev);
4986         uint32_t vfta, index;
4987
4988         if ((adapter->hw.mng_cookie.status &
4989              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4990             (vid == adapter->mng_vlan_id))
4991                 return;
4992         /* add VID to filter table */
4993         index = (vid >> 5) & 0x7F;
4994         vfta = E1000_READ_REG_ARRAY(&adapter->hw, VFTA, index);
4995         vfta |= (1 << (vid & 0x1F));
4996         e1000_write_vfta(&adapter->hw, index, vfta);
4997 }
4998
4999 static void
5000 e1000_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid)
5001 {
5002         struct e1000_adapter *adapter = netdev_priv(netdev);
5003         uint32_t vfta, index;
5004
5005         e1000_irq_disable(adapter);
5006         vlan_group_set_device(adapter->vlgrp, vid, NULL);
5007         e1000_irq_enable(adapter);
5008
5009         if ((adapter->hw.mng_cookie.status &
5010              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
5011             (vid == adapter->mng_vlan_id)) {
5012                 /* release control to f/w */
5013                 e1000_release_hw_control(adapter);
5014                 return;
5015         }
5016
5017         /* remove VID from filter table */
5018         index = (vid >> 5) & 0x7F;
5019         vfta = E1000_READ_REG_ARRAY(&adapter->hw, VFTA, index);
5020         vfta &= ~(1 << (vid & 0x1F));
5021         e1000_write_vfta(&adapter->hw, index, vfta);
5022 }
5023
5024 static void
5025 e1000_restore_vlan(struct e1000_adapter *adapter)
5026 {
5027         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
5028
5029         if (adapter->vlgrp) {
5030                 uint16_t vid;
5031                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
5032                         if (!vlan_group_get_device(adapter->vlgrp, vid))
5033                                 continue;
5034                         e1000_vlan_rx_add_vid(adapter->netdev, vid);
5035                 }
5036         }
5037 }
5038
5039 int
5040 e1000_set_spd_dplx(struct e1000_adapter *adapter, uint16_t spddplx)
5041 {
5042         adapter->hw.autoneg = 0;
5043
5044         /* Fiber NICs only allow 1000 gbps Full duplex */
5045         if ((adapter->hw.media_type == e1000_media_type_fiber) &&
5046                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
5047                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
5048                 return -EINVAL;
5049         }
5050
5051         switch (spddplx) {
5052         case SPEED_10 + DUPLEX_HALF:
5053                 adapter->hw.forced_speed_duplex = e1000_10_half;
5054                 break;
5055         case SPEED_10 + DUPLEX_FULL:
5056                 adapter->hw.forced_speed_duplex = e1000_10_full;
5057                 break;
5058         case SPEED_100 + DUPLEX_HALF:
5059                 adapter->hw.forced_speed_duplex = e1000_100_half;
5060                 break;
5061         case SPEED_100 + DUPLEX_FULL:
5062                 adapter->hw.forced_speed_duplex = e1000_100_full;
5063                 break;
5064         case SPEED_1000 + DUPLEX_FULL:
5065                 adapter->hw.autoneg = 1;
5066                 adapter->hw.autoneg_advertised = ADVERTISE_1000_FULL;
5067                 break;
5068         case SPEED_1000 + DUPLEX_HALF: /* not supported */
5069         default:
5070                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
5071                 return -EINVAL;
5072         }
5073         return 0;
5074 }
5075
5076 static int
5077 e1000_suspend(struct pci_dev *pdev, pm_message_t state)
5078 {
5079         struct net_device *netdev = pci_get_drvdata(pdev);
5080         struct e1000_adapter *adapter = netdev_priv(netdev);
5081         uint32_t ctrl, ctrl_ext, rctl, status;
5082         uint32_t wufc = adapter->wol;
5083 #ifdef CONFIG_PM
5084         int retval = 0;
5085 #endif
5086
5087         netif_device_detach(netdev);
5088
5089         if (netif_running(netdev)) {
5090                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
5091                 e1000_down(adapter);
5092         }
5093
5094 #ifdef CONFIG_PM
5095         retval = pci_save_state(pdev);
5096         if (retval)
5097                 return retval;
5098 #endif
5099
5100         status = E1000_READ_REG(&adapter->hw, STATUS);
5101         if (status & E1000_STATUS_LU)
5102                 wufc &= ~E1000_WUFC_LNKC;
5103
5104         if (wufc) {
5105                 e1000_setup_rctl(adapter);
5106                 e1000_set_multi(netdev);
5107
5108                 /* turn on all-multi mode if wake on multicast is enabled */
5109                 if (wufc & E1000_WUFC_MC) {
5110                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
5111                         rctl |= E1000_RCTL_MPE;
5112                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
5113                 }
5114
5115                 if (adapter->hw.mac_type >= e1000_82540) {
5116                         ctrl = E1000_READ_REG(&adapter->hw, CTRL);
5117                         /* advertise wake from D3Cold */
5118                         #define E1000_CTRL_ADVD3WUC 0x00100000
5119                         /* phy power management enable */
5120                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5121                         ctrl |= E1000_CTRL_ADVD3WUC |
5122                                 E1000_CTRL_EN_PHY_PWR_MGMT;
5123                         E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
5124                 }
5125
5126                 if (adapter->hw.media_type == e1000_media_type_fiber ||
5127                    adapter->hw.media_type == e1000_media_type_internal_serdes) {
5128                         /* keep the laser running in D3 */
5129                         ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
5130                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
5131                         E1000_WRITE_REG(&adapter->hw, CTRL_EXT, ctrl_ext);
5132                 }
5133
5134                 /* Allow time for pending master requests to run */
5135                 e1000_disable_pciex_master(&adapter->hw);
5136
5137                 E1000_WRITE_REG(&adapter->hw, WUC, E1000_WUC_PME_EN);
5138                 E1000_WRITE_REG(&adapter->hw, WUFC, wufc);
5139                 pci_enable_wake(pdev, PCI_D3hot, 1);
5140                 pci_enable_wake(pdev, PCI_D3cold, 1);
5141         } else {
5142                 E1000_WRITE_REG(&adapter->hw, WUC, 0);
5143                 E1000_WRITE_REG(&adapter->hw, WUFC, 0);
5144                 pci_enable_wake(pdev, PCI_D3hot, 0);
5145                 pci_enable_wake(pdev, PCI_D3cold, 0);
5146         }
5147
5148         e1000_release_manageability(adapter);
5149
5150         /* make sure adapter isn't asleep if manageability is enabled */
5151         if (adapter->en_mng_pt) {
5152                 pci_enable_wake(pdev, PCI_D3hot, 1);
5153                 pci_enable_wake(pdev, PCI_D3cold, 1);
5154         }
5155
5156         if (adapter->hw.phy_type == e1000_phy_igp_3)
5157                 e1000_phy_powerdown_workaround(&adapter->hw);
5158
5159         if (netif_running(netdev))
5160                 e1000_free_irq(adapter);
5161
5162         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
5163          * would have already happened in close and is redundant. */
5164         e1000_release_hw_control(adapter);
5165
5166         pci_disable_device(pdev);
5167
5168         pci_set_power_state(pdev, pci_choose_state(pdev, state));
5169
5170         return 0;
5171 }
5172
5173 #ifdef CONFIG_PM
5174 static int
5175 e1000_resume(struct pci_dev *pdev)
5176 {
5177         struct net_device *netdev = pci_get_drvdata(pdev);
5178         struct e1000_adapter *adapter = netdev_priv(netdev);
5179         uint32_t err;
5180
5181         pci_set_power_state(pdev, PCI_D0);
5182         pci_restore_state(pdev);
5183         if ((err = pci_enable_device(pdev))) {
5184                 printk(KERN_ERR "e1000: Cannot enable PCI device from suspend\n");
5185                 return err;
5186         }
5187         pci_set_master(pdev);
5188
5189         pci_enable_wake(pdev, PCI_D3hot, 0);
5190         pci_enable_wake(pdev, PCI_D3cold, 0);
5191
5192         if (netif_running(netdev) && (err = e1000_request_irq(adapter)))
5193                 return err;
5194
5195         e1000_power_up_phy(adapter);
5196         e1000_reset(adapter);
5197         E1000_WRITE_REG(&adapter->hw, WUS, ~0);
5198
5199         e1000_init_manageability(adapter);
5200
5201         if (netif_running(netdev))
5202                 e1000_up(adapter);
5203
5204         netif_device_attach(netdev);
5205
5206         /* If the controller is 82573 and f/w is AMT, do not set
5207          * DRV_LOAD until the interface is up.  For all other cases,
5208          * let the f/w know that the h/w is now under the control
5209          * of the driver. */
5210         if (adapter->hw.mac_type != e1000_82573 ||
5211             !e1000_check_mng_mode(&adapter->hw))
5212                 e1000_get_hw_control(adapter);
5213
5214         return 0;
5215 }
5216 #endif
5217
5218 static void e1000_shutdown(struct pci_dev *pdev)
5219 {
5220         e1000_suspend(pdev, PMSG_SUSPEND);
5221 }
5222
5223 #ifdef CONFIG_NET_POLL_CONTROLLER
5224 /*
5225  * Polling 'interrupt' - used by things like netconsole to send skbs
5226  * without having to re-enable interrupts. It's not called while
5227  * the interrupt routine is executing.
5228  */
5229 static void
5230 e1000_netpoll(struct net_device *netdev)
5231 {
5232         struct e1000_adapter *adapter = netdev_priv(netdev);
5233
5234         disable_irq(adapter->pdev->irq);
5235         e1000_intr(adapter->pdev->irq, netdev);
5236         e1000_clean_tx_irq(adapter, adapter->tx_ring);
5237 #ifndef CONFIG_E1000_NAPI
5238         adapter->clean_rx(adapter, adapter->rx_ring);
5239 #endif
5240         enable_irq(adapter->pdev->irq);
5241 }
5242 #endif
5243
5244 /**
5245  * e1000_io_error_detected - called when PCI error is detected
5246  * @pdev: Pointer to PCI device
5247  * @state: The current pci conneection state
5248  *
5249  * This function is called after a PCI bus error affecting
5250  * this device has been detected.
5251  */
5252 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
5253 {
5254         struct net_device *netdev = pci_get_drvdata(pdev);
5255         struct e1000_adapter *adapter = netdev->priv;
5256
5257         netif_device_detach(netdev);
5258
5259         if (netif_running(netdev))
5260                 e1000_down(adapter);
5261         pci_disable_device(pdev);
5262
5263         /* Request a slot slot reset. */
5264         return PCI_ERS_RESULT_NEED_RESET;
5265 }
5266
5267 /**
5268  * e1000_io_slot_reset - called after the pci bus has been reset.
5269  * @pdev: Pointer to PCI device
5270  *
5271  * Restart the card from scratch, as if from a cold-boot. Implementation
5272  * resembles the first-half of the e1000_resume routine.
5273  */
5274 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5275 {
5276         struct net_device *netdev = pci_get_drvdata(pdev);
5277         struct e1000_adapter *adapter = netdev->priv;
5278
5279         if (pci_enable_device(pdev)) {
5280                 printk(KERN_ERR "e1000: Cannot re-enable PCI device after reset.\n");
5281                 return PCI_ERS_RESULT_DISCONNECT;
5282         }
5283         pci_set_master(pdev);
5284
5285         pci_enable_wake(pdev, PCI_D3hot, 0);
5286         pci_enable_wake(pdev, PCI_D3cold, 0);
5287
5288         e1000_reset(adapter);
5289         E1000_WRITE_REG(&adapter->hw, WUS, ~0);
5290
5291         return PCI_ERS_RESULT_RECOVERED;
5292 }
5293
5294 /**
5295  * e1000_io_resume - called when traffic can start flowing again.
5296  * @pdev: Pointer to PCI device
5297  *
5298  * This callback is called when the error recovery driver tells us that
5299  * its OK to resume normal operation. Implementation resembles the
5300  * second-half of the e1000_resume routine.
5301  */
5302 static void e1000_io_resume(struct pci_dev *pdev)
5303 {
5304         struct net_device *netdev = pci_get_drvdata(pdev);
5305         struct e1000_adapter *adapter = netdev->priv;
5306
5307         e1000_init_manageability(adapter);
5308
5309         if (netif_running(netdev)) {
5310                 if (e1000_up(adapter)) {
5311                         printk("e1000: can't bring device back up after reset\n");
5312                         return;
5313                 }
5314         }
5315
5316         netif_device_attach(netdev);
5317
5318         /* If the controller is 82573 and f/w is AMT, do not set
5319          * DRV_LOAD until the interface is up.  For all other cases,
5320          * let the f/w know that the h/w is now under the control
5321          * of the driver. */
5322         if (adapter->hw.mac_type != e1000_82573 ||
5323             !e1000_check_mng_mode(&adapter->hw))
5324                 e1000_get_hw_control(adapter);
5325
5326 }
5327
5328 /* e1000_main.c */