xref: /linux/drivers/net/ethernet/intel/e1000/e1000_main.c (revision 95e9fd10f06cb5642028b6b851e32b8c8afb4571)
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 #include <linux/io.h>
32 #include <linux/prefetch.h>
33 #include <linux/bitops.h>
34 #include <linux/if_vlan.h>
35 
36 char e1000_driver_name[] = "e1000";
37 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
38 #define DRV_VERSION "7.3.21-k8-NAPI"
39 const char e1000_driver_version[] = DRV_VERSION;
40 static const char e1000_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
41 
42 /* e1000_pci_tbl - PCI Device ID Table
43  *
44  * Last entry must be all 0s
45  *
46  * Macro expands to...
47  *   {PCI_DEVICE(PCI_VENDOR_ID_INTEL, device_id)}
48  */
49 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
50 	INTEL_E1000_ETHERNET_DEVICE(0x1000),
51 	INTEL_E1000_ETHERNET_DEVICE(0x1001),
52 	INTEL_E1000_ETHERNET_DEVICE(0x1004),
53 	INTEL_E1000_ETHERNET_DEVICE(0x1008),
54 	INTEL_E1000_ETHERNET_DEVICE(0x1009),
55 	INTEL_E1000_ETHERNET_DEVICE(0x100C),
56 	INTEL_E1000_ETHERNET_DEVICE(0x100D),
57 	INTEL_E1000_ETHERNET_DEVICE(0x100E),
58 	INTEL_E1000_ETHERNET_DEVICE(0x100F),
59 	INTEL_E1000_ETHERNET_DEVICE(0x1010),
60 	INTEL_E1000_ETHERNET_DEVICE(0x1011),
61 	INTEL_E1000_ETHERNET_DEVICE(0x1012),
62 	INTEL_E1000_ETHERNET_DEVICE(0x1013),
63 	INTEL_E1000_ETHERNET_DEVICE(0x1014),
64 	INTEL_E1000_ETHERNET_DEVICE(0x1015),
65 	INTEL_E1000_ETHERNET_DEVICE(0x1016),
66 	INTEL_E1000_ETHERNET_DEVICE(0x1017),
67 	INTEL_E1000_ETHERNET_DEVICE(0x1018),
68 	INTEL_E1000_ETHERNET_DEVICE(0x1019),
69 	INTEL_E1000_ETHERNET_DEVICE(0x101A),
70 	INTEL_E1000_ETHERNET_DEVICE(0x101D),
71 	INTEL_E1000_ETHERNET_DEVICE(0x101E),
72 	INTEL_E1000_ETHERNET_DEVICE(0x1026),
73 	INTEL_E1000_ETHERNET_DEVICE(0x1027),
74 	INTEL_E1000_ETHERNET_DEVICE(0x1028),
75 	INTEL_E1000_ETHERNET_DEVICE(0x1075),
76 	INTEL_E1000_ETHERNET_DEVICE(0x1076),
77 	INTEL_E1000_ETHERNET_DEVICE(0x1077),
78 	INTEL_E1000_ETHERNET_DEVICE(0x1078),
79 	INTEL_E1000_ETHERNET_DEVICE(0x1079),
80 	INTEL_E1000_ETHERNET_DEVICE(0x107A),
81 	INTEL_E1000_ETHERNET_DEVICE(0x107B),
82 	INTEL_E1000_ETHERNET_DEVICE(0x107C),
83 	INTEL_E1000_ETHERNET_DEVICE(0x108A),
84 	INTEL_E1000_ETHERNET_DEVICE(0x1099),
85 	INTEL_E1000_ETHERNET_DEVICE(0x10B5),
86 	INTEL_E1000_ETHERNET_DEVICE(0x2E6E),
87 	/* required last entry */
88 	{0,}
89 };
90 
91 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
92 
93 int e1000_up(struct e1000_adapter *adapter);
94 void e1000_down(struct e1000_adapter *adapter);
95 void e1000_reinit_locked(struct e1000_adapter *adapter);
96 void e1000_reset(struct e1000_adapter *adapter);
97 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
98 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
99 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
100 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
101 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
102                              struct e1000_tx_ring *txdr);
103 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
104                              struct e1000_rx_ring *rxdr);
105 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
106                              struct e1000_tx_ring *tx_ring);
107 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
108                              struct e1000_rx_ring *rx_ring);
109 void e1000_update_stats(struct e1000_adapter *adapter);
110 
111 static int e1000_init_module(void);
112 static void e1000_exit_module(void);
113 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
114 static void __devexit e1000_remove(struct pci_dev *pdev);
115 static int e1000_alloc_queues(struct e1000_adapter *adapter);
116 static int e1000_sw_init(struct e1000_adapter *adapter);
117 static int e1000_open(struct net_device *netdev);
118 static int e1000_close(struct net_device *netdev);
119 static void e1000_configure_tx(struct e1000_adapter *adapter);
120 static void e1000_configure_rx(struct e1000_adapter *adapter);
121 static void e1000_setup_rctl(struct e1000_adapter *adapter);
122 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
123 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
124 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
125                                 struct e1000_tx_ring *tx_ring);
126 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
127                                 struct e1000_rx_ring *rx_ring);
128 static void e1000_set_rx_mode(struct net_device *netdev);
129 static void e1000_update_phy_info_task(struct work_struct *work);
130 static void e1000_watchdog(struct work_struct *work);
131 static void e1000_82547_tx_fifo_stall_task(struct work_struct *work);
132 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
133 				    struct net_device *netdev);
134 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
135 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
136 static int e1000_set_mac(struct net_device *netdev, void *p);
137 static irqreturn_t e1000_intr(int irq, void *data);
138 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
139 			       struct e1000_tx_ring *tx_ring);
140 static int e1000_clean(struct napi_struct *napi, int budget);
141 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
142 			       struct e1000_rx_ring *rx_ring,
143 			       int *work_done, int work_to_do);
144 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
145 				     struct e1000_rx_ring *rx_ring,
146 				     int *work_done, int work_to_do);
147 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
148 				   struct e1000_rx_ring *rx_ring,
149 				   int cleaned_count);
150 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
151 					 struct e1000_rx_ring *rx_ring,
152 					 int cleaned_count);
153 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
154 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
155 			   int cmd);
156 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
157 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
158 static void e1000_tx_timeout(struct net_device *dev);
159 static void e1000_reset_task(struct work_struct *work);
160 static void e1000_smartspeed(struct e1000_adapter *adapter);
161 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
162                                        struct sk_buff *skb);
163 
164 static bool e1000_vlan_used(struct e1000_adapter *adapter);
165 static void e1000_vlan_mode(struct net_device *netdev,
166 			    netdev_features_t features);
167 static void e1000_vlan_filter_on_off(struct e1000_adapter *adapter,
168 				     bool filter_on);
169 static int e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid);
170 static int e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid);
171 static void e1000_restore_vlan(struct e1000_adapter *adapter);
172 
173 #ifdef CONFIG_PM
174 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
175 static int e1000_resume(struct pci_dev *pdev);
176 #endif
177 static void e1000_shutdown(struct pci_dev *pdev);
178 
179 #ifdef CONFIG_NET_POLL_CONTROLLER
180 /* for netdump / net console */
181 static void e1000_netpoll (struct net_device *netdev);
182 #endif
183 
184 #define COPYBREAK_DEFAULT 256
185 static unsigned int copybreak __read_mostly = COPYBREAK_DEFAULT;
186 module_param(copybreak, uint, 0644);
187 MODULE_PARM_DESC(copybreak,
188 	"Maximum size of packet that is copied to a new buffer on receive");
189 
190 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
191                      pci_channel_state_t state);
192 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
193 static void e1000_io_resume(struct pci_dev *pdev);
194 
195 static struct pci_error_handlers e1000_err_handler = {
196 	.error_detected = e1000_io_error_detected,
197 	.slot_reset = e1000_io_slot_reset,
198 	.resume = e1000_io_resume,
199 };
200 
201 static struct pci_driver e1000_driver = {
202 	.name     = e1000_driver_name,
203 	.id_table = e1000_pci_tbl,
204 	.probe    = e1000_probe,
205 	.remove   = __devexit_p(e1000_remove),
206 #ifdef CONFIG_PM
207 	/* Power Management Hooks */
208 	.suspend  = e1000_suspend,
209 	.resume   = e1000_resume,
210 #endif
211 	.shutdown = e1000_shutdown,
212 	.err_handler = &e1000_err_handler
213 };
214 
215 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
216 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
217 MODULE_LICENSE("GPL");
218 MODULE_VERSION(DRV_VERSION);
219 
220 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
221 static int debug = -1;
222 module_param(debug, int, 0);
223 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
224 
225 /**
226  * e1000_get_hw_dev - return device
227  * used by hardware layer to print debugging information
228  *
229  **/
230 struct net_device *e1000_get_hw_dev(struct e1000_hw *hw)
231 {
232 	struct e1000_adapter *adapter = hw->back;
233 	return adapter->netdev;
234 }
235 
236 /**
237  * e1000_init_module - Driver Registration Routine
238  *
239  * e1000_init_module is the first routine called when the driver is
240  * loaded. All it does is register with the PCI subsystem.
241  **/
242 
243 static int __init e1000_init_module(void)
244 {
245 	int ret;
246 	pr_info("%s - version %s\n", e1000_driver_string, e1000_driver_version);
247 
248 	pr_info("%s\n", e1000_copyright);
249 
250 	ret = pci_register_driver(&e1000_driver);
251 	if (copybreak != COPYBREAK_DEFAULT) {
252 		if (copybreak == 0)
253 			pr_info("copybreak disabled\n");
254 		else
255 			pr_info("copybreak enabled for "
256 				   "packets <= %u bytes\n", copybreak);
257 	}
258 	return ret;
259 }
260 
261 module_init(e1000_init_module);
262 
263 /**
264  * e1000_exit_module - Driver Exit Cleanup Routine
265  *
266  * e1000_exit_module is called just before the driver is removed
267  * from memory.
268  **/
269 
270 static void __exit e1000_exit_module(void)
271 {
272 	pci_unregister_driver(&e1000_driver);
273 }
274 
275 module_exit(e1000_exit_module);
276 
277 static int e1000_request_irq(struct e1000_adapter *adapter)
278 {
279 	struct net_device *netdev = adapter->netdev;
280 	irq_handler_t handler = e1000_intr;
281 	int irq_flags = IRQF_SHARED;
282 	int err;
283 
284 	err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
285 	                  netdev);
286 	if (err) {
287 		e_err(probe, "Unable to allocate interrupt Error: %d\n", err);
288 	}
289 
290 	return err;
291 }
292 
293 static void e1000_free_irq(struct e1000_adapter *adapter)
294 {
295 	struct net_device *netdev = adapter->netdev;
296 
297 	free_irq(adapter->pdev->irq, netdev);
298 }
299 
300 /**
301  * e1000_irq_disable - Mask off interrupt generation on the NIC
302  * @adapter: board private structure
303  **/
304 
305 static void e1000_irq_disable(struct e1000_adapter *adapter)
306 {
307 	struct e1000_hw *hw = &adapter->hw;
308 
309 	ew32(IMC, ~0);
310 	E1000_WRITE_FLUSH();
311 	synchronize_irq(adapter->pdev->irq);
312 }
313 
314 /**
315  * e1000_irq_enable - Enable default interrupt generation settings
316  * @adapter: board private structure
317  **/
318 
319 static void e1000_irq_enable(struct e1000_adapter *adapter)
320 {
321 	struct e1000_hw *hw = &adapter->hw;
322 
323 	ew32(IMS, IMS_ENABLE_MASK);
324 	E1000_WRITE_FLUSH();
325 }
326 
327 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
328 {
329 	struct e1000_hw *hw = &adapter->hw;
330 	struct net_device *netdev = adapter->netdev;
331 	u16 vid = hw->mng_cookie.vlan_id;
332 	u16 old_vid = adapter->mng_vlan_id;
333 
334 	if (!e1000_vlan_used(adapter))
335 		return;
336 
337 	if (!test_bit(vid, adapter->active_vlans)) {
338 		if (hw->mng_cookie.status &
339 		    E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
340 			e1000_vlan_rx_add_vid(netdev, vid);
341 			adapter->mng_vlan_id = vid;
342 		} else {
343 			adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
344 		}
345 		if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
346 		    (vid != old_vid) &&
347 		    !test_bit(old_vid, adapter->active_vlans))
348 			e1000_vlan_rx_kill_vid(netdev, old_vid);
349 	} else {
350 		adapter->mng_vlan_id = vid;
351 	}
352 }
353 
354 static void e1000_init_manageability(struct e1000_adapter *adapter)
355 {
356 	struct e1000_hw *hw = &adapter->hw;
357 
358 	if (adapter->en_mng_pt) {
359 		u32 manc = er32(MANC);
360 
361 		/* disable hardware interception of ARP */
362 		manc &= ~(E1000_MANC_ARP_EN);
363 
364 		ew32(MANC, manc);
365 	}
366 }
367 
368 static void e1000_release_manageability(struct e1000_adapter *adapter)
369 {
370 	struct e1000_hw *hw = &adapter->hw;
371 
372 	if (adapter->en_mng_pt) {
373 		u32 manc = er32(MANC);
374 
375 		/* re-enable hardware interception of ARP */
376 		manc |= E1000_MANC_ARP_EN;
377 
378 		ew32(MANC, manc);
379 	}
380 }
381 
382 /**
383  * e1000_configure - configure the hardware for RX and TX
384  * @adapter = private board structure
385  **/
386 static void e1000_configure(struct e1000_adapter *adapter)
387 {
388 	struct net_device *netdev = adapter->netdev;
389 	int i;
390 
391 	e1000_set_rx_mode(netdev);
392 
393 	e1000_restore_vlan(adapter);
394 	e1000_init_manageability(adapter);
395 
396 	e1000_configure_tx(adapter);
397 	e1000_setup_rctl(adapter);
398 	e1000_configure_rx(adapter);
399 	/* call E1000_DESC_UNUSED which always leaves
400 	 * at least 1 descriptor unused to make sure
401 	 * next_to_use != next_to_clean */
402 	for (i = 0; i < adapter->num_rx_queues; i++) {
403 		struct e1000_rx_ring *ring = &adapter->rx_ring[i];
404 		adapter->alloc_rx_buf(adapter, ring,
405 		                      E1000_DESC_UNUSED(ring));
406 	}
407 }
408 
409 int e1000_up(struct e1000_adapter *adapter)
410 {
411 	struct e1000_hw *hw = &adapter->hw;
412 
413 	/* hardware has been reset, we need to reload some things */
414 	e1000_configure(adapter);
415 
416 	clear_bit(__E1000_DOWN, &adapter->flags);
417 
418 	napi_enable(&adapter->napi);
419 
420 	e1000_irq_enable(adapter);
421 
422 	netif_wake_queue(adapter->netdev);
423 
424 	/* fire a link change interrupt to start the watchdog */
425 	ew32(ICS, E1000_ICS_LSC);
426 	return 0;
427 }
428 
429 /**
430  * e1000_power_up_phy - restore link in case the phy was powered down
431  * @adapter: address of board private structure
432  *
433  * The phy may be powered down to save power and turn off link when the
434  * driver is unloaded and wake on lan is not enabled (among others)
435  * *** this routine MUST be followed by a call to e1000_reset ***
436  *
437  **/
438 
439 void e1000_power_up_phy(struct e1000_adapter *adapter)
440 {
441 	struct e1000_hw *hw = &adapter->hw;
442 	u16 mii_reg = 0;
443 
444 	/* Just clear the power down bit to wake the phy back up */
445 	if (hw->media_type == e1000_media_type_copper) {
446 		/* according to the manual, the phy will retain its
447 		 * settings across a power-down/up cycle */
448 		e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
449 		mii_reg &= ~MII_CR_POWER_DOWN;
450 		e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
451 	}
452 }
453 
454 static void e1000_power_down_phy(struct e1000_adapter *adapter)
455 {
456 	struct e1000_hw *hw = &adapter->hw;
457 
458 	/* Power down the PHY so no link is implied when interface is down *
459 	 * The PHY cannot be powered down if any of the following is true *
460 	 * (a) WoL is enabled
461 	 * (b) AMT is active
462 	 * (c) SoL/IDER session is active */
463 	if (!adapter->wol && hw->mac_type >= e1000_82540 &&
464 	   hw->media_type == e1000_media_type_copper) {
465 		u16 mii_reg = 0;
466 
467 		switch (hw->mac_type) {
468 		case e1000_82540:
469 		case e1000_82545:
470 		case e1000_82545_rev_3:
471 		case e1000_82546:
472 		case e1000_ce4100:
473 		case e1000_82546_rev_3:
474 		case e1000_82541:
475 		case e1000_82541_rev_2:
476 		case e1000_82547:
477 		case e1000_82547_rev_2:
478 			if (er32(MANC) & E1000_MANC_SMBUS_EN)
479 				goto out;
480 			break;
481 		default:
482 			goto out;
483 		}
484 		e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
485 		mii_reg |= MII_CR_POWER_DOWN;
486 		e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
487 		msleep(1);
488 	}
489 out:
490 	return;
491 }
492 
493 static void e1000_down_and_stop(struct e1000_adapter *adapter)
494 {
495 	set_bit(__E1000_DOWN, &adapter->flags);
496 
497 	/* Only kill reset task if adapter is not resetting */
498 	if (!test_bit(__E1000_RESETTING, &adapter->flags))
499 		cancel_work_sync(&adapter->reset_task);
500 
501 	cancel_delayed_work_sync(&adapter->watchdog_task);
502 	cancel_delayed_work_sync(&adapter->phy_info_task);
503 	cancel_delayed_work_sync(&adapter->fifo_stall_task);
504 }
505 
506 void e1000_down(struct e1000_adapter *adapter)
507 {
508 	struct e1000_hw *hw = &adapter->hw;
509 	struct net_device *netdev = adapter->netdev;
510 	u32 rctl, tctl;
511 
512 
513 	/* disable receives in the hardware */
514 	rctl = er32(RCTL);
515 	ew32(RCTL, rctl & ~E1000_RCTL_EN);
516 	/* flush and sleep below */
517 
518 	netif_tx_disable(netdev);
519 
520 	/* disable transmits in the hardware */
521 	tctl = er32(TCTL);
522 	tctl &= ~E1000_TCTL_EN;
523 	ew32(TCTL, tctl);
524 	/* flush both disables and wait for them to finish */
525 	E1000_WRITE_FLUSH();
526 	msleep(10);
527 
528 	napi_disable(&adapter->napi);
529 
530 	e1000_irq_disable(adapter);
531 
532 	/*
533 	 * Setting DOWN must be after irq_disable to prevent
534 	 * a screaming interrupt.  Setting DOWN also prevents
535 	 * tasks from rescheduling.
536 	 */
537 	e1000_down_and_stop(adapter);
538 
539 	adapter->link_speed = 0;
540 	adapter->link_duplex = 0;
541 	netif_carrier_off(netdev);
542 
543 	e1000_reset(adapter);
544 	e1000_clean_all_tx_rings(adapter);
545 	e1000_clean_all_rx_rings(adapter);
546 }
547 
548 static void e1000_reinit_safe(struct e1000_adapter *adapter)
549 {
550 	while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
551 		msleep(1);
552 	mutex_lock(&adapter->mutex);
553 	e1000_down(adapter);
554 	e1000_up(adapter);
555 	mutex_unlock(&adapter->mutex);
556 	clear_bit(__E1000_RESETTING, &adapter->flags);
557 }
558 
559 void e1000_reinit_locked(struct e1000_adapter *adapter)
560 {
561 	/* if rtnl_lock is not held the call path is bogus */
562 	ASSERT_RTNL();
563 	WARN_ON(in_interrupt());
564 	while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
565 		msleep(1);
566 	e1000_down(adapter);
567 	e1000_up(adapter);
568 	clear_bit(__E1000_RESETTING, &adapter->flags);
569 }
570 
571 void e1000_reset(struct e1000_adapter *adapter)
572 {
573 	struct e1000_hw *hw = &adapter->hw;
574 	u32 pba = 0, tx_space, min_tx_space, min_rx_space;
575 	bool legacy_pba_adjust = false;
576 	u16 hwm;
577 
578 	/* Repartition Pba for greater than 9k mtu
579 	 * To take effect CTRL.RST is required.
580 	 */
581 
582 	switch (hw->mac_type) {
583 	case e1000_82542_rev2_0:
584 	case e1000_82542_rev2_1:
585 	case e1000_82543:
586 	case e1000_82544:
587 	case e1000_82540:
588 	case e1000_82541:
589 	case e1000_82541_rev_2:
590 		legacy_pba_adjust = true;
591 		pba = E1000_PBA_48K;
592 		break;
593 	case e1000_82545:
594 	case e1000_82545_rev_3:
595 	case e1000_82546:
596 	case e1000_ce4100:
597 	case e1000_82546_rev_3:
598 		pba = E1000_PBA_48K;
599 		break;
600 	case e1000_82547:
601 	case e1000_82547_rev_2:
602 		legacy_pba_adjust = true;
603 		pba = E1000_PBA_30K;
604 		break;
605 	case e1000_undefined:
606 	case e1000_num_macs:
607 		break;
608 	}
609 
610 	if (legacy_pba_adjust) {
611 		if (hw->max_frame_size > E1000_RXBUFFER_8192)
612 			pba -= 8; /* allocate more FIFO for Tx */
613 
614 		if (hw->mac_type == e1000_82547) {
615 			adapter->tx_fifo_head = 0;
616 			adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
617 			adapter->tx_fifo_size =
618 				(E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
619 			atomic_set(&adapter->tx_fifo_stall, 0);
620 		}
621 	} else if (hw->max_frame_size >  ETH_FRAME_LEN + ETH_FCS_LEN) {
622 		/* adjust PBA for jumbo frames */
623 		ew32(PBA, pba);
624 
625 		/* To maintain wire speed transmits, the Tx FIFO should be
626 		 * large enough to accommodate two full transmit packets,
627 		 * rounded up to the next 1KB and expressed in KB.  Likewise,
628 		 * the Rx FIFO should be large enough to accommodate at least
629 		 * one full receive packet and is similarly rounded up and
630 		 * expressed in KB. */
631 		pba = er32(PBA);
632 		/* upper 16 bits has Tx packet buffer allocation size in KB */
633 		tx_space = pba >> 16;
634 		/* lower 16 bits has Rx packet buffer allocation size in KB */
635 		pba &= 0xffff;
636 		/*
637 		 * the tx fifo also stores 16 bytes of information about the tx
638 		 * but don't include ethernet FCS because hardware appends it
639 		 */
640 		min_tx_space = (hw->max_frame_size +
641 		                sizeof(struct e1000_tx_desc) -
642 		                ETH_FCS_LEN) * 2;
643 		min_tx_space = ALIGN(min_tx_space, 1024);
644 		min_tx_space >>= 10;
645 		/* software strips receive CRC, so leave room for it */
646 		min_rx_space = hw->max_frame_size;
647 		min_rx_space = ALIGN(min_rx_space, 1024);
648 		min_rx_space >>= 10;
649 
650 		/* If current Tx allocation is less than the min Tx FIFO size,
651 		 * and the min Tx FIFO size is less than the current Rx FIFO
652 		 * allocation, take space away from current Rx allocation */
653 		if (tx_space < min_tx_space &&
654 		    ((min_tx_space - tx_space) < pba)) {
655 			pba = pba - (min_tx_space - tx_space);
656 
657 			/* PCI/PCIx hardware has PBA alignment constraints */
658 			switch (hw->mac_type) {
659 			case e1000_82545 ... e1000_82546_rev_3:
660 				pba &= ~(E1000_PBA_8K - 1);
661 				break;
662 			default:
663 				break;
664 			}
665 
666 			/* if short on rx space, rx wins and must trump tx
667 			 * adjustment or use Early Receive if available */
668 			if (pba < min_rx_space)
669 				pba = min_rx_space;
670 		}
671 	}
672 
673 	ew32(PBA, pba);
674 
675 	/*
676 	 * flow control settings:
677 	 * The high water mark must be low enough to fit one full frame
678 	 * (or the size used for early receive) above it in the Rx FIFO.
679 	 * Set it to the lower of:
680 	 * - 90% of the Rx FIFO size, and
681 	 * - the full Rx FIFO size minus the early receive size (for parts
682 	 *   with ERT support assuming ERT set to E1000_ERT_2048), or
683 	 * - the full Rx FIFO size minus one full frame
684 	 */
685 	hwm = min(((pba << 10) * 9 / 10),
686 		  ((pba << 10) - hw->max_frame_size));
687 
688 	hw->fc_high_water = hwm & 0xFFF8;	/* 8-byte granularity */
689 	hw->fc_low_water = hw->fc_high_water - 8;
690 	hw->fc_pause_time = E1000_FC_PAUSE_TIME;
691 	hw->fc_send_xon = 1;
692 	hw->fc = hw->original_fc;
693 
694 	/* Allow time for pending master requests to run */
695 	e1000_reset_hw(hw);
696 	if (hw->mac_type >= e1000_82544)
697 		ew32(WUC, 0);
698 
699 	if (e1000_init_hw(hw))
700 		e_dev_err("Hardware Error\n");
701 	e1000_update_mng_vlan(adapter);
702 
703 	/* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
704 	if (hw->mac_type >= e1000_82544 &&
705 	    hw->autoneg == 1 &&
706 	    hw->autoneg_advertised == ADVERTISE_1000_FULL) {
707 		u32 ctrl = er32(CTRL);
708 		/* clear phy power management bit if we are in gig only mode,
709 		 * which if enabled will attempt negotiation to 100Mb, which
710 		 * can cause a loss of link at power off or driver unload */
711 		ctrl &= ~E1000_CTRL_SWDPIN3;
712 		ew32(CTRL, ctrl);
713 	}
714 
715 	/* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
716 	ew32(VET, ETHERNET_IEEE_VLAN_TYPE);
717 
718 	e1000_reset_adaptive(hw);
719 	e1000_phy_get_info(hw, &adapter->phy_info);
720 
721 	e1000_release_manageability(adapter);
722 }
723 
724 /* Dump the eeprom for users having checksum issues */
725 static void e1000_dump_eeprom(struct e1000_adapter *adapter)
726 {
727 	struct net_device *netdev = adapter->netdev;
728 	struct ethtool_eeprom eeprom;
729 	const struct ethtool_ops *ops = netdev->ethtool_ops;
730 	u8 *data;
731 	int i;
732 	u16 csum_old, csum_new = 0;
733 
734 	eeprom.len = ops->get_eeprom_len(netdev);
735 	eeprom.offset = 0;
736 
737 	data = kmalloc(eeprom.len, GFP_KERNEL);
738 	if (!data)
739 		return;
740 
741 	ops->get_eeprom(netdev, &eeprom, data);
742 
743 	csum_old = (data[EEPROM_CHECKSUM_REG * 2]) +
744 		   (data[EEPROM_CHECKSUM_REG * 2 + 1] << 8);
745 	for (i = 0; i < EEPROM_CHECKSUM_REG * 2; i += 2)
746 		csum_new += data[i] + (data[i + 1] << 8);
747 	csum_new = EEPROM_SUM - csum_new;
748 
749 	pr_err("/*********************/\n");
750 	pr_err("Current EEPROM Checksum : 0x%04x\n", csum_old);
751 	pr_err("Calculated              : 0x%04x\n", csum_new);
752 
753 	pr_err("Offset    Values\n");
754 	pr_err("========  ======\n");
755 	print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 16, 1, data, 128, 0);
756 
757 	pr_err("Include this output when contacting your support provider.\n");
758 	pr_err("This is not a software error! Something bad happened to\n");
759 	pr_err("your hardware or EEPROM image. Ignoring this problem could\n");
760 	pr_err("result in further problems, possibly loss of data,\n");
761 	pr_err("corruption or system hangs!\n");
762 	pr_err("The MAC Address will be reset to 00:00:00:00:00:00,\n");
763 	pr_err("which is invalid and requires you to set the proper MAC\n");
764 	pr_err("address manually before continuing to enable this network\n");
765 	pr_err("device. Please inspect the EEPROM dump and report the\n");
766 	pr_err("issue to your hardware vendor or Intel Customer Support.\n");
767 	pr_err("/*********************/\n");
768 
769 	kfree(data);
770 }
771 
772 /**
773  * e1000_is_need_ioport - determine if an adapter needs ioport resources or not
774  * @pdev: PCI device information struct
775  *
776  * Return true if an adapter needs ioport resources
777  **/
778 static int e1000_is_need_ioport(struct pci_dev *pdev)
779 {
780 	switch (pdev->device) {
781 	case E1000_DEV_ID_82540EM:
782 	case E1000_DEV_ID_82540EM_LOM:
783 	case E1000_DEV_ID_82540EP:
784 	case E1000_DEV_ID_82540EP_LOM:
785 	case E1000_DEV_ID_82540EP_LP:
786 	case E1000_DEV_ID_82541EI:
787 	case E1000_DEV_ID_82541EI_MOBILE:
788 	case E1000_DEV_ID_82541ER:
789 	case E1000_DEV_ID_82541ER_LOM:
790 	case E1000_DEV_ID_82541GI:
791 	case E1000_DEV_ID_82541GI_LF:
792 	case E1000_DEV_ID_82541GI_MOBILE:
793 	case E1000_DEV_ID_82544EI_COPPER:
794 	case E1000_DEV_ID_82544EI_FIBER:
795 	case E1000_DEV_ID_82544GC_COPPER:
796 	case E1000_DEV_ID_82544GC_LOM:
797 	case E1000_DEV_ID_82545EM_COPPER:
798 	case E1000_DEV_ID_82545EM_FIBER:
799 	case E1000_DEV_ID_82546EB_COPPER:
800 	case E1000_DEV_ID_82546EB_FIBER:
801 	case E1000_DEV_ID_82546EB_QUAD_COPPER:
802 		return true;
803 	default:
804 		return false;
805 	}
806 }
807 
808 static netdev_features_t e1000_fix_features(struct net_device *netdev,
809 	netdev_features_t features)
810 {
811 	/*
812 	 * Since there is no support for separate rx/tx vlan accel
813 	 * enable/disable make sure tx flag is always in same state as rx.
814 	 */
815 	if (features & NETIF_F_HW_VLAN_RX)
816 		features |= NETIF_F_HW_VLAN_TX;
817 	else
818 		features &= ~NETIF_F_HW_VLAN_TX;
819 
820 	return features;
821 }
822 
823 static int e1000_set_features(struct net_device *netdev,
824 	netdev_features_t features)
825 {
826 	struct e1000_adapter *adapter = netdev_priv(netdev);
827 	netdev_features_t changed = features ^ netdev->features;
828 
829 	if (changed & NETIF_F_HW_VLAN_RX)
830 		e1000_vlan_mode(netdev, features);
831 
832 	if (!(changed & (NETIF_F_RXCSUM | NETIF_F_RXALL)))
833 		return 0;
834 
835 	netdev->features = features;
836 	adapter->rx_csum = !!(features & NETIF_F_RXCSUM);
837 
838 	if (netif_running(netdev))
839 		e1000_reinit_locked(adapter);
840 	else
841 		e1000_reset(adapter);
842 
843 	return 0;
844 }
845 
846 static const struct net_device_ops e1000_netdev_ops = {
847 	.ndo_open		= e1000_open,
848 	.ndo_stop		= e1000_close,
849 	.ndo_start_xmit		= e1000_xmit_frame,
850 	.ndo_get_stats		= e1000_get_stats,
851 	.ndo_set_rx_mode	= e1000_set_rx_mode,
852 	.ndo_set_mac_address	= e1000_set_mac,
853 	.ndo_tx_timeout		= e1000_tx_timeout,
854 	.ndo_change_mtu		= e1000_change_mtu,
855 	.ndo_do_ioctl		= e1000_ioctl,
856 	.ndo_validate_addr	= eth_validate_addr,
857 	.ndo_vlan_rx_add_vid	= e1000_vlan_rx_add_vid,
858 	.ndo_vlan_rx_kill_vid	= e1000_vlan_rx_kill_vid,
859 #ifdef CONFIG_NET_POLL_CONTROLLER
860 	.ndo_poll_controller	= e1000_netpoll,
861 #endif
862 	.ndo_fix_features	= e1000_fix_features,
863 	.ndo_set_features	= e1000_set_features,
864 };
865 
866 /**
867  * e1000_init_hw_struct - initialize members of hw struct
868  * @adapter: board private struct
869  * @hw: structure used by e1000_hw.c
870  *
871  * Factors out initialization of the e1000_hw struct to its own function
872  * that can be called very early at init (just after struct allocation).
873  * Fields are initialized based on PCI device information and
874  * OS network device settings (MTU size).
875  * Returns negative error codes if MAC type setup fails.
876  */
877 static int e1000_init_hw_struct(struct e1000_adapter *adapter,
878 				struct e1000_hw *hw)
879 {
880 	struct pci_dev *pdev = adapter->pdev;
881 
882 	/* PCI config space info */
883 	hw->vendor_id = pdev->vendor;
884 	hw->device_id = pdev->device;
885 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
886 	hw->subsystem_id = pdev->subsystem_device;
887 	hw->revision_id = pdev->revision;
888 
889 	pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
890 
891 	hw->max_frame_size = adapter->netdev->mtu +
892 			     ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
893 	hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
894 
895 	/* identify the MAC */
896 	if (e1000_set_mac_type(hw)) {
897 		e_err(probe, "Unknown MAC Type\n");
898 		return -EIO;
899 	}
900 
901 	switch (hw->mac_type) {
902 	default:
903 		break;
904 	case e1000_82541:
905 	case e1000_82547:
906 	case e1000_82541_rev_2:
907 	case e1000_82547_rev_2:
908 		hw->phy_init_script = 1;
909 		break;
910 	}
911 
912 	e1000_set_media_type(hw);
913 	e1000_get_bus_info(hw);
914 
915 	hw->wait_autoneg_complete = false;
916 	hw->tbi_compatibility_en = true;
917 	hw->adaptive_ifs = true;
918 
919 	/* Copper options */
920 
921 	if (hw->media_type == e1000_media_type_copper) {
922 		hw->mdix = AUTO_ALL_MODES;
923 		hw->disable_polarity_correction = false;
924 		hw->master_slave = E1000_MASTER_SLAVE;
925 	}
926 
927 	return 0;
928 }
929 
930 /**
931  * e1000_probe - Device Initialization Routine
932  * @pdev: PCI device information struct
933  * @ent: entry in e1000_pci_tbl
934  *
935  * Returns 0 on success, negative on failure
936  *
937  * e1000_probe initializes an adapter identified by a pci_dev structure.
938  * The OS initialization, configuring of the adapter private structure,
939  * and a hardware reset occur.
940  **/
941 static int __devinit e1000_probe(struct pci_dev *pdev,
942 				 const struct pci_device_id *ent)
943 {
944 	struct net_device *netdev;
945 	struct e1000_adapter *adapter;
946 	struct e1000_hw *hw;
947 
948 	static int cards_found = 0;
949 	static int global_quad_port_a = 0; /* global ksp3 port a indication */
950 	int i, err, pci_using_dac;
951 	u16 eeprom_data = 0;
952 	u16 tmp = 0;
953 	u16 eeprom_apme_mask = E1000_EEPROM_APME;
954 	int bars, need_ioport;
955 
956 	/* do not allocate ioport bars when not needed */
957 	need_ioport = e1000_is_need_ioport(pdev);
958 	if (need_ioport) {
959 		bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
960 		err = pci_enable_device(pdev);
961 	} else {
962 		bars = pci_select_bars(pdev, IORESOURCE_MEM);
963 		err = pci_enable_device_mem(pdev);
964 	}
965 	if (err)
966 		return err;
967 
968 	err = pci_request_selected_regions(pdev, bars, e1000_driver_name);
969 	if (err)
970 		goto err_pci_reg;
971 
972 	pci_set_master(pdev);
973 	err = pci_save_state(pdev);
974 	if (err)
975 		goto err_alloc_etherdev;
976 
977 	err = -ENOMEM;
978 	netdev = alloc_etherdev(sizeof(struct e1000_adapter));
979 	if (!netdev)
980 		goto err_alloc_etherdev;
981 
982 	SET_NETDEV_DEV(netdev, &pdev->dev);
983 
984 	pci_set_drvdata(pdev, netdev);
985 	adapter = netdev_priv(netdev);
986 	adapter->netdev = netdev;
987 	adapter->pdev = pdev;
988 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
989 	adapter->bars = bars;
990 	adapter->need_ioport = need_ioport;
991 
992 	hw = &adapter->hw;
993 	hw->back = adapter;
994 
995 	err = -EIO;
996 	hw->hw_addr = pci_ioremap_bar(pdev, BAR_0);
997 	if (!hw->hw_addr)
998 		goto err_ioremap;
999 
1000 	if (adapter->need_ioport) {
1001 		for (i = BAR_1; i <= BAR_5; i++) {
1002 			if (pci_resource_len(pdev, i) == 0)
1003 				continue;
1004 			if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
1005 				hw->io_base = pci_resource_start(pdev, i);
1006 				break;
1007 			}
1008 		}
1009 	}
1010 
1011 	/* make ready for any if (hw->...) below */
1012 	err = e1000_init_hw_struct(adapter, hw);
1013 	if (err)
1014 		goto err_sw_init;
1015 
1016 	/*
1017 	 * there is a workaround being applied below that limits
1018 	 * 64-bit DMA addresses to 64-bit hardware.  There are some
1019 	 * 32-bit adapters that Tx hang when given 64-bit DMA addresses
1020 	 */
1021 	pci_using_dac = 0;
1022 	if ((hw->bus_type == e1000_bus_type_pcix) &&
1023 	    !dma_set_mask(&pdev->dev, DMA_BIT_MASK(64))) {
1024 		/*
1025 		 * according to DMA-API-HOWTO, coherent calls will always
1026 		 * succeed if the set call did
1027 		 */
1028 		dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
1029 		pci_using_dac = 1;
1030 	} else {
1031 		err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
1032 		if (err) {
1033 			pr_err("No usable DMA config, aborting\n");
1034 			goto err_dma;
1035 		}
1036 		dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32));
1037 	}
1038 
1039 	netdev->netdev_ops = &e1000_netdev_ops;
1040 	e1000_set_ethtool_ops(netdev);
1041 	netdev->watchdog_timeo = 5 * HZ;
1042 	netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
1043 
1044 	strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1045 
1046 	adapter->bd_number = cards_found;
1047 
1048 	/* setup the private structure */
1049 
1050 	err = e1000_sw_init(adapter);
1051 	if (err)
1052 		goto err_sw_init;
1053 
1054 	err = -EIO;
1055 	if (hw->mac_type == e1000_ce4100) {
1056 		hw->ce4100_gbe_mdio_base_virt =
1057 					ioremap(pci_resource_start(pdev, BAR_1),
1058 		                                pci_resource_len(pdev, BAR_1));
1059 
1060 		if (!hw->ce4100_gbe_mdio_base_virt)
1061 			goto err_mdio_ioremap;
1062 	}
1063 
1064 	if (hw->mac_type >= e1000_82543) {
1065 		netdev->hw_features = NETIF_F_SG |
1066 				   NETIF_F_HW_CSUM |
1067 				   NETIF_F_HW_VLAN_RX;
1068 		netdev->features = NETIF_F_HW_VLAN_TX |
1069 				   NETIF_F_HW_VLAN_FILTER;
1070 	}
1071 
1072 	if ((hw->mac_type >= e1000_82544) &&
1073 	   (hw->mac_type != e1000_82547))
1074 		netdev->hw_features |= NETIF_F_TSO;
1075 
1076 	netdev->priv_flags |= IFF_SUPP_NOFCS;
1077 
1078 	netdev->features |= netdev->hw_features;
1079 	netdev->hw_features |= (NETIF_F_RXCSUM |
1080 				NETIF_F_RXALL |
1081 				NETIF_F_RXFCS);
1082 
1083 	if (pci_using_dac) {
1084 		netdev->features |= NETIF_F_HIGHDMA;
1085 		netdev->vlan_features |= NETIF_F_HIGHDMA;
1086 	}
1087 
1088 	netdev->vlan_features |= (NETIF_F_TSO |
1089 				  NETIF_F_HW_CSUM |
1090 				  NETIF_F_SG);
1091 
1092 	netdev->priv_flags |= IFF_UNICAST_FLT;
1093 
1094 	adapter->en_mng_pt = e1000_enable_mng_pass_thru(hw);
1095 
1096 	/* initialize eeprom parameters */
1097 	if (e1000_init_eeprom_params(hw)) {
1098 		e_err(probe, "EEPROM initialization failed\n");
1099 		goto err_eeprom;
1100 	}
1101 
1102 	/* before reading the EEPROM, reset the controller to
1103 	 * put the device in a known good starting state */
1104 
1105 	e1000_reset_hw(hw);
1106 
1107 	/* make sure the EEPROM is good */
1108 	if (e1000_validate_eeprom_checksum(hw) < 0) {
1109 		e_err(probe, "The EEPROM Checksum Is Not Valid\n");
1110 		e1000_dump_eeprom(adapter);
1111 		/*
1112 		 * set MAC address to all zeroes to invalidate and temporary
1113 		 * disable this device for the user. This blocks regular
1114 		 * traffic while still permitting ethtool ioctls from reaching
1115 		 * the hardware as well as allowing the user to run the
1116 		 * interface after manually setting a hw addr using
1117 		 * `ip set address`
1118 		 */
1119 		memset(hw->mac_addr, 0, netdev->addr_len);
1120 	} else {
1121 		/* copy the MAC address out of the EEPROM */
1122 		if (e1000_read_mac_addr(hw))
1123 			e_err(probe, "EEPROM Read Error\n");
1124 	}
1125 	/* don't block initalization here due to bad MAC address */
1126 	memcpy(netdev->dev_addr, hw->mac_addr, netdev->addr_len);
1127 	memcpy(netdev->perm_addr, hw->mac_addr, netdev->addr_len);
1128 
1129 	if (!is_valid_ether_addr(netdev->perm_addr))
1130 		e_err(probe, "Invalid MAC Address\n");
1131 
1132 
1133 	INIT_DELAYED_WORK(&adapter->watchdog_task, e1000_watchdog);
1134 	INIT_DELAYED_WORK(&adapter->fifo_stall_task,
1135 			  e1000_82547_tx_fifo_stall_task);
1136 	INIT_DELAYED_WORK(&adapter->phy_info_task, e1000_update_phy_info_task);
1137 	INIT_WORK(&adapter->reset_task, e1000_reset_task);
1138 
1139 	e1000_check_options(adapter);
1140 
1141 	/* Initial Wake on LAN setting
1142 	 * If APM wake is enabled in the EEPROM,
1143 	 * enable the ACPI Magic Packet filter
1144 	 */
1145 
1146 	switch (hw->mac_type) {
1147 	case e1000_82542_rev2_0:
1148 	case e1000_82542_rev2_1:
1149 	case e1000_82543:
1150 		break;
1151 	case e1000_82544:
1152 		e1000_read_eeprom(hw,
1153 			EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
1154 		eeprom_apme_mask = E1000_EEPROM_82544_APM;
1155 		break;
1156 	case e1000_82546:
1157 	case e1000_82546_rev_3:
1158 		if (er32(STATUS) & E1000_STATUS_FUNC_1){
1159 			e1000_read_eeprom(hw,
1160 				EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1161 			break;
1162 		}
1163 		/* Fall Through */
1164 	default:
1165 		e1000_read_eeprom(hw,
1166 			EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1167 		break;
1168 	}
1169 	if (eeprom_data & eeprom_apme_mask)
1170 		adapter->eeprom_wol |= E1000_WUFC_MAG;
1171 
1172 	/* now that we have the eeprom settings, apply the special cases
1173 	 * where the eeprom may be wrong or the board simply won't support
1174 	 * wake on lan on a particular port */
1175 	switch (pdev->device) {
1176 	case E1000_DEV_ID_82546GB_PCIE:
1177 		adapter->eeprom_wol = 0;
1178 		break;
1179 	case E1000_DEV_ID_82546EB_FIBER:
1180 	case E1000_DEV_ID_82546GB_FIBER:
1181 		/* Wake events only supported on port A for dual fiber
1182 		 * regardless of eeprom setting */
1183 		if (er32(STATUS) & E1000_STATUS_FUNC_1)
1184 			adapter->eeprom_wol = 0;
1185 		break;
1186 	case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
1187 		/* if quad port adapter, disable WoL on all but port A */
1188 		if (global_quad_port_a != 0)
1189 			adapter->eeprom_wol = 0;
1190 		else
1191 			adapter->quad_port_a = true;
1192 		/* Reset for multiple quad port adapters */
1193 		if (++global_quad_port_a == 4)
1194 			global_quad_port_a = 0;
1195 		break;
1196 	}
1197 
1198 	/* initialize the wol settings based on the eeprom settings */
1199 	adapter->wol = adapter->eeprom_wol;
1200 	device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1201 
1202 	/* Auto detect PHY address */
1203 	if (hw->mac_type == e1000_ce4100) {
1204 		for (i = 0; i < 32; i++) {
1205 			hw->phy_addr = i;
1206 			e1000_read_phy_reg(hw, PHY_ID2, &tmp);
1207 			if (tmp == 0 || tmp == 0xFF) {
1208 				if (i == 31)
1209 					goto err_eeprom;
1210 				continue;
1211 			} else
1212 				break;
1213 		}
1214 	}
1215 
1216 	/* reset the hardware with the new settings */
1217 	e1000_reset(adapter);
1218 
1219 	strcpy(netdev->name, "eth%d");
1220 	err = register_netdev(netdev);
1221 	if (err)
1222 		goto err_register;
1223 
1224 	e1000_vlan_filter_on_off(adapter, false);
1225 
1226 	/* print bus type/speed/width info */
1227 	e_info(probe, "(PCI%s:%dMHz:%d-bit) %pM\n",
1228 	       ((hw->bus_type == e1000_bus_type_pcix) ? "-X" : ""),
1229 	       ((hw->bus_speed == e1000_bus_speed_133) ? 133 :
1230 		(hw->bus_speed == e1000_bus_speed_120) ? 120 :
1231 		(hw->bus_speed == e1000_bus_speed_100) ? 100 :
1232 		(hw->bus_speed == e1000_bus_speed_66) ? 66 : 33),
1233 	       ((hw->bus_width == e1000_bus_width_64) ? 64 : 32),
1234 	       netdev->dev_addr);
1235 
1236 	/* carrier off reporting is important to ethtool even BEFORE open */
1237 	netif_carrier_off(netdev);
1238 
1239 	e_info(probe, "Intel(R) PRO/1000 Network Connection\n");
1240 
1241 	cards_found++;
1242 	return 0;
1243 
1244 err_register:
1245 err_eeprom:
1246 	e1000_phy_hw_reset(hw);
1247 
1248 	if (hw->flash_address)
1249 		iounmap(hw->flash_address);
1250 	kfree(adapter->tx_ring);
1251 	kfree(adapter->rx_ring);
1252 err_dma:
1253 err_sw_init:
1254 err_mdio_ioremap:
1255 	iounmap(hw->ce4100_gbe_mdio_base_virt);
1256 	iounmap(hw->hw_addr);
1257 err_ioremap:
1258 	free_netdev(netdev);
1259 err_alloc_etherdev:
1260 	pci_release_selected_regions(pdev, bars);
1261 err_pci_reg:
1262 	pci_disable_device(pdev);
1263 	return err;
1264 }
1265 
1266 /**
1267  * e1000_remove - Device Removal Routine
1268  * @pdev: PCI device information struct
1269  *
1270  * e1000_remove is called by the PCI subsystem to alert the driver
1271  * that it should release a PCI device.  The could be caused by a
1272  * Hot-Plug event, or because the driver is going to be removed from
1273  * memory.
1274  **/
1275 
1276 static void __devexit e1000_remove(struct pci_dev *pdev)
1277 {
1278 	struct net_device *netdev = pci_get_drvdata(pdev);
1279 	struct e1000_adapter *adapter = netdev_priv(netdev);
1280 	struct e1000_hw *hw = &adapter->hw;
1281 
1282 	e1000_down_and_stop(adapter);
1283 	e1000_release_manageability(adapter);
1284 
1285 	unregister_netdev(netdev);
1286 
1287 	e1000_phy_hw_reset(hw);
1288 
1289 	kfree(adapter->tx_ring);
1290 	kfree(adapter->rx_ring);
1291 
1292 	if (hw->mac_type == e1000_ce4100)
1293 		iounmap(hw->ce4100_gbe_mdio_base_virt);
1294 	iounmap(hw->hw_addr);
1295 	if (hw->flash_address)
1296 		iounmap(hw->flash_address);
1297 	pci_release_selected_regions(pdev, adapter->bars);
1298 
1299 	free_netdev(netdev);
1300 
1301 	pci_disable_device(pdev);
1302 }
1303 
1304 /**
1305  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1306  * @adapter: board private structure to initialize
1307  *
1308  * e1000_sw_init initializes the Adapter private data structure.
1309  * e1000_init_hw_struct MUST be called before this function
1310  **/
1311 
1312 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
1313 {
1314 	adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1315 
1316 	adapter->num_tx_queues = 1;
1317 	adapter->num_rx_queues = 1;
1318 
1319 	if (e1000_alloc_queues(adapter)) {
1320 		e_err(probe, "Unable to allocate memory for queues\n");
1321 		return -ENOMEM;
1322 	}
1323 
1324 	/* Explicitly disable IRQ since the NIC can be in any state. */
1325 	e1000_irq_disable(adapter);
1326 
1327 	spin_lock_init(&adapter->stats_lock);
1328 	mutex_init(&adapter->mutex);
1329 
1330 	set_bit(__E1000_DOWN, &adapter->flags);
1331 
1332 	return 0;
1333 }
1334 
1335 /**
1336  * e1000_alloc_queues - Allocate memory for all rings
1337  * @adapter: board private structure to initialize
1338  *
1339  * We allocate one ring per queue at run-time since we don't know the
1340  * number of queues at compile-time.
1341  **/
1342 
1343 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1344 {
1345 	adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1346 	                           sizeof(struct e1000_tx_ring), GFP_KERNEL);
1347 	if (!adapter->tx_ring)
1348 		return -ENOMEM;
1349 
1350 	adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1351 	                           sizeof(struct e1000_rx_ring), GFP_KERNEL);
1352 	if (!adapter->rx_ring) {
1353 		kfree(adapter->tx_ring);
1354 		return -ENOMEM;
1355 	}
1356 
1357 	return E1000_SUCCESS;
1358 }
1359 
1360 /**
1361  * e1000_open - Called when a network interface is made active
1362  * @netdev: network interface device structure
1363  *
1364  * Returns 0 on success, negative value on failure
1365  *
1366  * The open entry point is called when a network interface is made
1367  * active by the system (IFF_UP).  At this point all resources needed
1368  * for transmit and receive operations are allocated, the interrupt
1369  * handler is registered with the OS, the watchdog task is started,
1370  * and the stack is notified that the interface is ready.
1371  **/
1372 
1373 static int e1000_open(struct net_device *netdev)
1374 {
1375 	struct e1000_adapter *adapter = netdev_priv(netdev);
1376 	struct e1000_hw *hw = &adapter->hw;
1377 	int err;
1378 
1379 	/* disallow open during test */
1380 	if (test_bit(__E1000_TESTING, &adapter->flags))
1381 		return -EBUSY;
1382 
1383 	netif_carrier_off(netdev);
1384 
1385 	/* allocate transmit descriptors */
1386 	err = e1000_setup_all_tx_resources(adapter);
1387 	if (err)
1388 		goto err_setup_tx;
1389 
1390 	/* allocate receive descriptors */
1391 	err = e1000_setup_all_rx_resources(adapter);
1392 	if (err)
1393 		goto err_setup_rx;
1394 
1395 	e1000_power_up_phy(adapter);
1396 
1397 	adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1398 	if ((hw->mng_cookie.status &
1399 			  E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1400 		e1000_update_mng_vlan(adapter);
1401 	}
1402 
1403 	/* before we allocate an interrupt, we must be ready to handle it.
1404 	 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1405 	 * as soon as we call pci_request_irq, so we have to setup our
1406 	 * clean_rx handler before we do so.  */
1407 	e1000_configure(adapter);
1408 
1409 	err = e1000_request_irq(adapter);
1410 	if (err)
1411 		goto err_req_irq;
1412 
1413 	/* From here on the code is the same as e1000_up() */
1414 	clear_bit(__E1000_DOWN, &adapter->flags);
1415 
1416 	napi_enable(&adapter->napi);
1417 
1418 	e1000_irq_enable(adapter);
1419 
1420 	netif_start_queue(netdev);
1421 
1422 	/* fire a link status change interrupt to start the watchdog */
1423 	ew32(ICS, E1000_ICS_LSC);
1424 
1425 	return E1000_SUCCESS;
1426 
1427 err_req_irq:
1428 	e1000_power_down_phy(adapter);
1429 	e1000_free_all_rx_resources(adapter);
1430 err_setup_rx:
1431 	e1000_free_all_tx_resources(adapter);
1432 err_setup_tx:
1433 	e1000_reset(adapter);
1434 
1435 	return err;
1436 }
1437 
1438 /**
1439  * e1000_close - Disables a network interface
1440  * @netdev: network interface device structure
1441  *
1442  * Returns 0, this is not allowed to fail
1443  *
1444  * The close entry point is called when an interface is de-activated
1445  * by the OS.  The hardware is still under the drivers control, but
1446  * needs to be disabled.  A global MAC reset is issued to stop the
1447  * hardware, and all transmit and receive resources are freed.
1448  **/
1449 
1450 static int e1000_close(struct net_device *netdev)
1451 {
1452 	struct e1000_adapter *adapter = netdev_priv(netdev);
1453 	struct e1000_hw *hw = &adapter->hw;
1454 
1455 	WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1456 	e1000_down(adapter);
1457 	e1000_power_down_phy(adapter);
1458 	e1000_free_irq(adapter);
1459 
1460 	e1000_free_all_tx_resources(adapter);
1461 	e1000_free_all_rx_resources(adapter);
1462 
1463 	/* kill manageability vlan ID if supported, but not if a vlan with
1464 	 * the same ID is registered on the host OS (let 8021q kill it) */
1465 	if ((hw->mng_cookie.status &
1466 			  E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1467 	     !test_bit(adapter->mng_vlan_id, adapter->active_vlans)) {
1468 		e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1469 	}
1470 
1471 	return 0;
1472 }
1473 
1474 /**
1475  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1476  * @adapter: address of board private structure
1477  * @start: address of beginning of memory
1478  * @len: length of memory
1479  **/
1480 static bool e1000_check_64k_bound(struct e1000_adapter *adapter, void *start,
1481 				  unsigned long len)
1482 {
1483 	struct e1000_hw *hw = &adapter->hw;
1484 	unsigned long begin = (unsigned long)start;
1485 	unsigned long end = begin + len;
1486 
1487 	/* First rev 82545 and 82546 need to not allow any memory
1488 	 * write location to cross 64k boundary due to errata 23 */
1489 	if (hw->mac_type == e1000_82545 ||
1490 	    hw->mac_type == e1000_ce4100 ||
1491 	    hw->mac_type == e1000_82546) {
1492 		return ((begin ^ (end - 1)) >> 16) != 0 ? false : true;
1493 	}
1494 
1495 	return true;
1496 }
1497 
1498 /**
1499  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1500  * @adapter: board private structure
1501  * @txdr:    tx descriptor ring (for a specific queue) to setup
1502  *
1503  * Return 0 on success, negative on failure
1504  **/
1505 
1506 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
1507 				    struct e1000_tx_ring *txdr)
1508 {
1509 	struct pci_dev *pdev = adapter->pdev;
1510 	int size;
1511 
1512 	size = sizeof(struct e1000_buffer) * txdr->count;
1513 	txdr->buffer_info = vzalloc(size);
1514 	if (!txdr->buffer_info) {
1515 		e_err(probe, "Unable to allocate memory for the Tx descriptor "
1516 		      "ring\n");
1517 		return -ENOMEM;
1518 	}
1519 
1520 	/* round up to nearest 4K */
1521 
1522 	txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1523 	txdr->size = ALIGN(txdr->size, 4096);
1524 
1525 	txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size, &txdr->dma,
1526 					GFP_KERNEL);
1527 	if (!txdr->desc) {
1528 setup_tx_desc_die:
1529 		vfree(txdr->buffer_info);
1530 		e_err(probe, "Unable to allocate memory for the Tx descriptor "
1531 		      "ring\n");
1532 		return -ENOMEM;
1533 	}
1534 
1535 	/* Fix for errata 23, can't cross 64kB boundary */
1536 	if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1537 		void *olddesc = txdr->desc;
1538 		dma_addr_t olddma = txdr->dma;
1539 		e_err(tx_err, "txdr align check failed: %u bytes at %p\n",
1540 		      txdr->size, txdr->desc);
1541 		/* Try again, without freeing the previous */
1542 		txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size,
1543 						&txdr->dma, GFP_KERNEL);
1544 		/* Failed allocation, critical failure */
1545 		if (!txdr->desc) {
1546 			dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1547 					  olddma);
1548 			goto setup_tx_desc_die;
1549 		}
1550 
1551 		if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1552 			/* give up */
1553 			dma_free_coherent(&pdev->dev, txdr->size, txdr->desc,
1554 					  txdr->dma);
1555 			dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1556 					  olddma);
1557 			e_err(probe, "Unable to allocate aligned memory "
1558 			      "for the transmit descriptor ring\n");
1559 			vfree(txdr->buffer_info);
1560 			return -ENOMEM;
1561 		} else {
1562 			/* Free old allocation, new allocation was successful */
1563 			dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1564 					  olddma);
1565 		}
1566 	}
1567 	memset(txdr->desc, 0, txdr->size);
1568 
1569 	txdr->next_to_use = 0;
1570 	txdr->next_to_clean = 0;
1571 
1572 	return 0;
1573 }
1574 
1575 /**
1576  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1577  * 				  (Descriptors) for all queues
1578  * @adapter: board private structure
1579  *
1580  * Return 0 on success, negative on failure
1581  **/
1582 
1583 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1584 {
1585 	int i, err = 0;
1586 
1587 	for (i = 0; i < adapter->num_tx_queues; i++) {
1588 		err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1589 		if (err) {
1590 			e_err(probe, "Allocation for Tx Queue %u failed\n", i);
1591 			for (i-- ; i >= 0; i--)
1592 				e1000_free_tx_resources(adapter,
1593 							&adapter->tx_ring[i]);
1594 			break;
1595 		}
1596 	}
1597 
1598 	return err;
1599 }
1600 
1601 /**
1602  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1603  * @adapter: board private structure
1604  *
1605  * Configure the Tx unit of the MAC after a reset.
1606  **/
1607 
1608 static void e1000_configure_tx(struct e1000_adapter *adapter)
1609 {
1610 	u64 tdba;
1611 	struct e1000_hw *hw = &adapter->hw;
1612 	u32 tdlen, tctl, tipg;
1613 	u32 ipgr1, ipgr2;
1614 
1615 	/* Setup the HW Tx Head and Tail descriptor pointers */
1616 
1617 	switch (adapter->num_tx_queues) {
1618 	case 1:
1619 	default:
1620 		tdba = adapter->tx_ring[0].dma;
1621 		tdlen = adapter->tx_ring[0].count *
1622 			sizeof(struct e1000_tx_desc);
1623 		ew32(TDLEN, tdlen);
1624 		ew32(TDBAH, (tdba >> 32));
1625 		ew32(TDBAL, (tdba & 0x00000000ffffffffULL));
1626 		ew32(TDT, 0);
1627 		ew32(TDH, 0);
1628 		adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ? E1000_TDH : E1000_82542_TDH);
1629 		adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ? E1000_TDT : E1000_82542_TDT);
1630 		break;
1631 	}
1632 
1633 	/* Set the default values for the Tx Inter Packet Gap timer */
1634 	if ((hw->media_type == e1000_media_type_fiber ||
1635 	     hw->media_type == e1000_media_type_internal_serdes))
1636 		tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1637 	else
1638 		tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1639 
1640 	switch (hw->mac_type) {
1641 	case e1000_82542_rev2_0:
1642 	case e1000_82542_rev2_1:
1643 		tipg = DEFAULT_82542_TIPG_IPGT;
1644 		ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1645 		ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1646 		break;
1647 	default:
1648 		ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1649 		ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1650 		break;
1651 	}
1652 	tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1653 	tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1654 	ew32(TIPG, tipg);
1655 
1656 	/* Set the Tx Interrupt Delay register */
1657 
1658 	ew32(TIDV, adapter->tx_int_delay);
1659 	if (hw->mac_type >= e1000_82540)
1660 		ew32(TADV, adapter->tx_abs_int_delay);
1661 
1662 	/* Program the Transmit Control Register */
1663 
1664 	tctl = er32(TCTL);
1665 	tctl &= ~E1000_TCTL_CT;
1666 	tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1667 		(E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1668 
1669 	e1000_config_collision_dist(hw);
1670 
1671 	/* Setup Transmit Descriptor Settings for eop descriptor */
1672 	adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1673 
1674 	/* only set IDE if we are delaying interrupts using the timers */
1675 	if (adapter->tx_int_delay)
1676 		adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1677 
1678 	if (hw->mac_type < e1000_82543)
1679 		adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1680 	else
1681 		adapter->txd_cmd |= E1000_TXD_CMD_RS;
1682 
1683 	/* Cache if we're 82544 running in PCI-X because we'll
1684 	 * need this to apply a workaround later in the send path. */
1685 	if (hw->mac_type == e1000_82544 &&
1686 	    hw->bus_type == e1000_bus_type_pcix)
1687 		adapter->pcix_82544 = true;
1688 
1689 	ew32(TCTL, tctl);
1690 
1691 }
1692 
1693 /**
1694  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1695  * @adapter: board private structure
1696  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1697  *
1698  * Returns 0 on success, negative on failure
1699  **/
1700 
1701 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
1702 				    struct e1000_rx_ring *rxdr)
1703 {
1704 	struct pci_dev *pdev = adapter->pdev;
1705 	int size, desc_len;
1706 
1707 	size = sizeof(struct e1000_buffer) * rxdr->count;
1708 	rxdr->buffer_info = vzalloc(size);
1709 	if (!rxdr->buffer_info) {
1710 		e_err(probe, "Unable to allocate memory for the Rx descriptor "
1711 		      "ring\n");
1712 		return -ENOMEM;
1713 	}
1714 
1715 	desc_len = sizeof(struct e1000_rx_desc);
1716 
1717 	/* Round up to nearest 4K */
1718 
1719 	rxdr->size = rxdr->count * desc_len;
1720 	rxdr->size = ALIGN(rxdr->size, 4096);
1721 
1722 	rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size, &rxdr->dma,
1723 					GFP_KERNEL);
1724 
1725 	if (!rxdr->desc) {
1726 		e_err(probe, "Unable to allocate memory for the Rx descriptor "
1727 		      "ring\n");
1728 setup_rx_desc_die:
1729 		vfree(rxdr->buffer_info);
1730 		return -ENOMEM;
1731 	}
1732 
1733 	/* Fix for errata 23, can't cross 64kB boundary */
1734 	if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1735 		void *olddesc = rxdr->desc;
1736 		dma_addr_t olddma = rxdr->dma;
1737 		e_err(rx_err, "rxdr align check failed: %u bytes at %p\n",
1738 		      rxdr->size, rxdr->desc);
1739 		/* Try again, without freeing the previous */
1740 		rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size,
1741 						&rxdr->dma, GFP_KERNEL);
1742 		/* Failed allocation, critical failure */
1743 		if (!rxdr->desc) {
1744 			dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1745 					  olddma);
1746 			e_err(probe, "Unable to allocate memory for the Rx "
1747 			      "descriptor ring\n");
1748 			goto setup_rx_desc_die;
1749 		}
1750 
1751 		if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1752 			/* give up */
1753 			dma_free_coherent(&pdev->dev, rxdr->size, rxdr->desc,
1754 					  rxdr->dma);
1755 			dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1756 					  olddma);
1757 			e_err(probe, "Unable to allocate aligned memory for "
1758 			      "the Rx descriptor ring\n");
1759 			goto setup_rx_desc_die;
1760 		} else {
1761 			/* Free old allocation, new allocation was successful */
1762 			dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1763 					  olddma);
1764 		}
1765 	}
1766 	memset(rxdr->desc, 0, rxdr->size);
1767 
1768 	rxdr->next_to_clean = 0;
1769 	rxdr->next_to_use = 0;
1770 	rxdr->rx_skb_top = NULL;
1771 
1772 	return 0;
1773 }
1774 
1775 /**
1776  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1777  * 				  (Descriptors) for all queues
1778  * @adapter: board private structure
1779  *
1780  * Return 0 on success, negative on failure
1781  **/
1782 
1783 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1784 {
1785 	int i, err = 0;
1786 
1787 	for (i = 0; i < adapter->num_rx_queues; i++) {
1788 		err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1789 		if (err) {
1790 			e_err(probe, "Allocation for Rx Queue %u failed\n", i);
1791 			for (i-- ; i >= 0; i--)
1792 				e1000_free_rx_resources(adapter,
1793 							&adapter->rx_ring[i]);
1794 			break;
1795 		}
1796 	}
1797 
1798 	return err;
1799 }
1800 
1801 /**
1802  * e1000_setup_rctl - configure the receive control registers
1803  * @adapter: Board private structure
1804  **/
1805 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1806 {
1807 	struct e1000_hw *hw = &adapter->hw;
1808 	u32 rctl;
1809 
1810 	rctl = er32(RCTL);
1811 
1812 	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1813 
1814 	rctl |= E1000_RCTL_BAM | E1000_RCTL_LBM_NO |
1815 		E1000_RCTL_RDMTS_HALF |
1816 		(hw->mc_filter_type << E1000_RCTL_MO_SHIFT);
1817 
1818 	if (hw->tbi_compatibility_on == 1)
1819 		rctl |= E1000_RCTL_SBP;
1820 	else
1821 		rctl &= ~E1000_RCTL_SBP;
1822 
1823 	if (adapter->netdev->mtu <= ETH_DATA_LEN)
1824 		rctl &= ~E1000_RCTL_LPE;
1825 	else
1826 		rctl |= E1000_RCTL_LPE;
1827 
1828 	/* Setup buffer sizes */
1829 	rctl &= ~E1000_RCTL_SZ_4096;
1830 	rctl |= E1000_RCTL_BSEX;
1831 	switch (adapter->rx_buffer_len) {
1832 		case E1000_RXBUFFER_2048:
1833 		default:
1834 			rctl |= E1000_RCTL_SZ_2048;
1835 			rctl &= ~E1000_RCTL_BSEX;
1836 			break;
1837 		case E1000_RXBUFFER_4096:
1838 			rctl |= E1000_RCTL_SZ_4096;
1839 			break;
1840 		case E1000_RXBUFFER_8192:
1841 			rctl |= E1000_RCTL_SZ_8192;
1842 			break;
1843 		case E1000_RXBUFFER_16384:
1844 			rctl |= E1000_RCTL_SZ_16384;
1845 			break;
1846 	}
1847 
1848 	/* This is useful for sniffing bad packets. */
1849 	if (adapter->netdev->features & NETIF_F_RXALL) {
1850 		/* UPE and MPE will be handled by normal PROMISC logic
1851 		 * in e1000e_set_rx_mode */
1852 		rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
1853 			 E1000_RCTL_BAM | /* RX All Bcast Pkts */
1854 			 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
1855 
1856 		rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
1857 			  E1000_RCTL_DPF | /* Allow filtered pause */
1858 			  E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
1859 		/* Do not mess with E1000_CTRL_VME, it affects transmit as well,
1860 		 * and that breaks VLANs.
1861 		 */
1862 	}
1863 
1864 	ew32(RCTL, rctl);
1865 }
1866 
1867 /**
1868  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1869  * @adapter: board private structure
1870  *
1871  * Configure the Rx unit of the MAC after a reset.
1872  **/
1873 
1874 static void e1000_configure_rx(struct e1000_adapter *adapter)
1875 {
1876 	u64 rdba;
1877 	struct e1000_hw *hw = &adapter->hw;
1878 	u32 rdlen, rctl, rxcsum;
1879 
1880 	if (adapter->netdev->mtu > ETH_DATA_LEN) {
1881 		rdlen = adapter->rx_ring[0].count *
1882 		        sizeof(struct e1000_rx_desc);
1883 		adapter->clean_rx = e1000_clean_jumbo_rx_irq;
1884 		adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
1885 	} else {
1886 		rdlen = adapter->rx_ring[0].count *
1887 		        sizeof(struct e1000_rx_desc);
1888 		adapter->clean_rx = e1000_clean_rx_irq;
1889 		adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1890 	}
1891 
1892 	/* disable receives while setting up the descriptors */
1893 	rctl = er32(RCTL);
1894 	ew32(RCTL, rctl & ~E1000_RCTL_EN);
1895 
1896 	/* set the Receive Delay Timer Register */
1897 	ew32(RDTR, adapter->rx_int_delay);
1898 
1899 	if (hw->mac_type >= e1000_82540) {
1900 		ew32(RADV, adapter->rx_abs_int_delay);
1901 		if (adapter->itr_setting != 0)
1902 			ew32(ITR, 1000000000 / (adapter->itr * 256));
1903 	}
1904 
1905 	/* Setup the HW Rx Head and Tail Descriptor Pointers and
1906 	 * the Base and Length of the Rx Descriptor Ring */
1907 	switch (adapter->num_rx_queues) {
1908 	case 1:
1909 	default:
1910 		rdba = adapter->rx_ring[0].dma;
1911 		ew32(RDLEN, rdlen);
1912 		ew32(RDBAH, (rdba >> 32));
1913 		ew32(RDBAL, (rdba & 0x00000000ffffffffULL));
1914 		ew32(RDT, 0);
1915 		ew32(RDH, 0);
1916 		adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ? E1000_RDH : E1000_82542_RDH);
1917 		adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ? E1000_RDT : E1000_82542_RDT);
1918 		break;
1919 	}
1920 
1921 	/* Enable 82543 Receive Checksum Offload for TCP and UDP */
1922 	if (hw->mac_type >= e1000_82543) {
1923 		rxcsum = er32(RXCSUM);
1924 		if (adapter->rx_csum)
1925 			rxcsum |= E1000_RXCSUM_TUOFL;
1926 		else
1927 			/* don't need to clear IPPCSE as it defaults to 0 */
1928 			rxcsum &= ~E1000_RXCSUM_TUOFL;
1929 		ew32(RXCSUM, rxcsum);
1930 	}
1931 
1932 	/* Enable Receives */
1933 	ew32(RCTL, rctl | E1000_RCTL_EN);
1934 }
1935 
1936 /**
1937  * e1000_free_tx_resources - Free Tx Resources per Queue
1938  * @adapter: board private structure
1939  * @tx_ring: Tx descriptor ring for a specific queue
1940  *
1941  * Free all transmit software resources
1942  **/
1943 
1944 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
1945 				    struct e1000_tx_ring *tx_ring)
1946 {
1947 	struct pci_dev *pdev = adapter->pdev;
1948 
1949 	e1000_clean_tx_ring(adapter, tx_ring);
1950 
1951 	vfree(tx_ring->buffer_info);
1952 	tx_ring->buffer_info = NULL;
1953 
1954 	dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1955 			  tx_ring->dma);
1956 
1957 	tx_ring->desc = NULL;
1958 }
1959 
1960 /**
1961  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1962  * @adapter: board private structure
1963  *
1964  * Free all transmit software resources
1965  **/
1966 
1967 void e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1968 {
1969 	int i;
1970 
1971 	for (i = 0; i < adapter->num_tx_queues; i++)
1972 		e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1973 }
1974 
1975 static void e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1976 					     struct e1000_buffer *buffer_info)
1977 {
1978 	if (buffer_info->dma) {
1979 		if (buffer_info->mapped_as_page)
1980 			dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1981 				       buffer_info->length, DMA_TO_DEVICE);
1982 		else
1983 			dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1984 					 buffer_info->length,
1985 					 DMA_TO_DEVICE);
1986 		buffer_info->dma = 0;
1987 	}
1988 	if (buffer_info->skb) {
1989 		dev_kfree_skb_any(buffer_info->skb);
1990 		buffer_info->skb = NULL;
1991 	}
1992 	buffer_info->time_stamp = 0;
1993 	/* buffer_info must be completely set up in the transmit path */
1994 }
1995 
1996 /**
1997  * e1000_clean_tx_ring - Free Tx Buffers
1998  * @adapter: board private structure
1999  * @tx_ring: ring to be cleaned
2000  **/
2001 
2002 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
2003 				struct e1000_tx_ring *tx_ring)
2004 {
2005 	struct e1000_hw *hw = &adapter->hw;
2006 	struct e1000_buffer *buffer_info;
2007 	unsigned long size;
2008 	unsigned int i;
2009 
2010 	/* Free all the Tx ring sk_buffs */
2011 
2012 	for (i = 0; i < tx_ring->count; i++) {
2013 		buffer_info = &tx_ring->buffer_info[i];
2014 		e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2015 	}
2016 
2017 	size = sizeof(struct e1000_buffer) * tx_ring->count;
2018 	memset(tx_ring->buffer_info, 0, size);
2019 
2020 	/* Zero out the descriptor ring */
2021 
2022 	memset(tx_ring->desc, 0, tx_ring->size);
2023 
2024 	tx_ring->next_to_use = 0;
2025 	tx_ring->next_to_clean = 0;
2026 	tx_ring->last_tx_tso = false;
2027 
2028 	writel(0, hw->hw_addr + tx_ring->tdh);
2029 	writel(0, hw->hw_addr + tx_ring->tdt);
2030 }
2031 
2032 /**
2033  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
2034  * @adapter: board private structure
2035  **/
2036 
2037 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
2038 {
2039 	int i;
2040 
2041 	for (i = 0; i < adapter->num_tx_queues; i++)
2042 		e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2043 }
2044 
2045 /**
2046  * e1000_free_rx_resources - Free Rx Resources
2047  * @adapter: board private structure
2048  * @rx_ring: ring to clean the resources from
2049  *
2050  * Free all receive software resources
2051  **/
2052 
2053 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
2054 				    struct e1000_rx_ring *rx_ring)
2055 {
2056 	struct pci_dev *pdev = adapter->pdev;
2057 
2058 	e1000_clean_rx_ring(adapter, rx_ring);
2059 
2060 	vfree(rx_ring->buffer_info);
2061 	rx_ring->buffer_info = NULL;
2062 
2063 	dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2064 			  rx_ring->dma);
2065 
2066 	rx_ring->desc = NULL;
2067 }
2068 
2069 /**
2070  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
2071  * @adapter: board private structure
2072  *
2073  * Free all receive software resources
2074  **/
2075 
2076 void e1000_free_all_rx_resources(struct e1000_adapter *adapter)
2077 {
2078 	int i;
2079 
2080 	for (i = 0; i < adapter->num_rx_queues; i++)
2081 		e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
2082 }
2083 
2084 /**
2085  * e1000_clean_rx_ring - Free Rx Buffers per Queue
2086  * @adapter: board private structure
2087  * @rx_ring: ring to free buffers from
2088  **/
2089 
2090 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
2091 				struct e1000_rx_ring *rx_ring)
2092 {
2093 	struct e1000_hw *hw = &adapter->hw;
2094 	struct e1000_buffer *buffer_info;
2095 	struct pci_dev *pdev = adapter->pdev;
2096 	unsigned long size;
2097 	unsigned int i;
2098 
2099 	/* Free all the Rx ring sk_buffs */
2100 	for (i = 0; i < rx_ring->count; i++) {
2101 		buffer_info = &rx_ring->buffer_info[i];
2102 		if (buffer_info->dma &&
2103 		    adapter->clean_rx == e1000_clean_rx_irq) {
2104 			dma_unmap_single(&pdev->dev, buffer_info->dma,
2105 			                 buffer_info->length,
2106 					 DMA_FROM_DEVICE);
2107 		} else if (buffer_info->dma &&
2108 		           adapter->clean_rx == e1000_clean_jumbo_rx_irq) {
2109 			dma_unmap_page(&pdev->dev, buffer_info->dma,
2110 				       buffer_info->length,
2111 				       DMA_FROM_DEVICE);
2112 		}
2113 
2114 		buffer_info->dma = 0;
2115 		if (buffer_info->page) {
2116 			put_page(buffer_info->page);
2117 			buffer_info->page = NULL;
2118 		}
2119 		if (buffer_info->skb) {
2120 			dev_kfree_skb(buffer_info->skb);
2121 			buffer_info->skb = NULL;
2122 		}
2123 	}
2124 
2125 	/* there also may be some cached data from a chained receive */
2126 	if (rx_ring->rx_skb_top) {
2127 		dev_kfree_skb(rx_ring->rx_skb_top);
2128 		rx_ring->rx_skb_top = NULL;
2129 	}
2130 
2131 	size = sizeof(struct e1000_buffer) * rx_ring->count;
2132 	memset(rx_ring->buffer_info, 0, size);
2133 
2134 	/* Zero out the descriptor ring */
2135 	memset(rx_ring->desc, 0, rx_ring->size);
2136 
2137 	rx_ring->next_to_clean = 0;
2138 	rx_ring->next_to_use = 0;
2139 
2140 	writel(0, hw->hw_addr + rx_ring->rdh);
2141 	writel(0, hw->hw_addr + rx_ring->rdt);
2142 }
2143 
2144 /**
2145  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2146  * @adapter: board private structure
2147  **/
2148 
2149 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2150 {
2151 	int i;
2152 
2153 	for (i = 0; i < adapter->num_rx_queues; i++)
2154 		e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2155 }
2156 
2157 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2158  * and memory write and invalidate disabled for certain operations
2159  */
2160 static void e1000_enter_82542_rst(struct e1000_adapter *adapter)
2161 {
2162 	struct e1000_hw *hw = &adapter->hw;
2163 	struct net_device *netdev = adapter->netdev;
2164 	u32 rctl;
2165 
2166 	e1000_pci_clear_mwi(hw);
2167 
2168 	rctl = er32(RCTL);
2169 	rctl |= E1000_RCTL_RST;
2170 	ew32(RCTL, rctl);
2171 	E1000_WRITE_FLUSH();
2172 	mdelay(5);
2173 
2174 	if (netif_running(netdev))
2175 		e1000_clean_all_rx_rings(adapter);
2176 }
2177 
2178 static void e1000_leave_82542_rst(struct e1000_adapter *adapter)
2179 {
2180 	struct e1000_hw *hw = &adapter->hw;
2181 	struct net_device *netdev = adapter->netdev;
2182 	u32 rctl;
2183 
2184 	rctl = er32(RCTL);
2185 	rctl &= ~E1000_RCTL_RST;
2186 	ew32(RCTL, rctl);
2187 	E1000_WRITE_FLUSH();
2188 	mdelay(5);
2189 
2190 	if (hw->pci_cmd_word & PCI_COMMAND_INVALIDATE)
2191 		e1000_pci_set_mwi(hw);
2192 
2193 	if (netif_running(netdev)) {
2194 		/* No need to loop, because 82542 supports only 1 queue */
2195 		struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2196 		e1000_configure_rx(adapter);
2197 		adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2198 	}
2199 }
2200 
2201 /**
2202  * e1000_set_mac - Change the Ethernet Address of the NIC
2203  * @netdev: network interface device structure
2204  * @p: pointer to an address structure
2205  *
2206  * Returns 0 on success, negative on failure
2207  **/
2208 
2209 static int e1000_set_mac(struct net_device *netdev, void *p)
2210 {
2211 	struct e1000_adapter *adapter = netdev_priv(netdev);
2212 	struct e1000_hw *hw = &adapter->hw;
2213 	struct sockaddr *addr = p;
2214 
2215 	if (!is_valid_ether_addr(addr->sa_data))
2216 		return -EADDRNOTAVAIL;
2217 
2218 	/* 82542 2.0 needs to be in reset to write receive address registers */
2219 
2220 	if (hw->mac_type == e1000_82542_rev2_0)
2221 		e1000_enter_82542_rst(adapter);
2222 
2223 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2224 	memcpy(hw->mac_addr, addr->sa_data, netdev->addr_len);
2225 
2226 	e1000_rar_set(hw, hw->mac_addr, 0);
2227 
2228 	if (hw->mac_type == e1000_82542_rev2_0)
2229 		e1000_leave_82542_rst(adapter);
2230 
2231 	return 0;
2232 }
2233 
2234 /**
2235  * e1000_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2236  * @netdev: network interface device structure
2237  *
2238  * The set_rx_mode entry point is called whenever the unicast or multicast
2239  * address lists or the network interface flags are updated. This routine is
2240  * responsible for configuring the hardware for proper unicast, multicast,
2241  * promiscuous mode, and all-multi behavior.
2242  **/
2243 
2244 static void e1000_set_rx_mode(struct net_device *netdev)
2245 {
2246 	struct e1000_adapter *adapter = netdev_priv(netdev);
2247 	struct e1000_hw *hw = &adapter->hw;
2248 	struct netdev_hw_addr *ha;
2249 	bool use_uc = false;
2250 	u32 rctl;
2251 	u32 hash_value;
2252 	int i, rar_entries = E1000_RAR_ENTRIES;
2253 	int mta_reg_count = E1000_NUM_MTA_REGISTERS;
2254 	u32 *mcarray = kcalloc(mta_reg_count, sizeof(u32), GFP_ATOMIC);
2255 
2256 	if (!mcarray) {
2257 		e_err(probe, "memory allocation failed\n");
2258 		return;
2259 	}
2260 
2261 	/* Check for Promiscuous and All Multicast modes */
2262 
2263 	rctl = er32(RCTL);
2264 
2265 	if (netdev->flags & IFF_PROMISC) {
2266 		rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2267 		rctl &= ~E1000_RCTL_VFE;
2268 	} else {
2269 		if (netdev->flags & IFF_ALLMULTI)
2270 			rctl |= E1000_RCTL_MPE;
2271 		else
2272 			rctl &= ~E1000_RCTL_MPE;
2273 		/* Enable VLAN filter if there is a VLAN */
2274 		if (e1000_vlan_used(adapter))
2275 			rctl |= E1000_RCTL_VFE;
2276 	}
2277 
2278 	if (netdev_uc_count(netdev) > rar_entries - 1) {
2279 		rctl |= E1000_RCTL_UPE;
2280 	} else if (!(netdev->flags & IFF_PROMISC)) {
2281 		rctl &= ~E1000_RCTL_UPE;
2282 		use_uc = true;
2283 	}
2284 
2285 	ew32(RCTL, rctl);
2286 
2287 	/* 82542 2.0 needs to be in reset to write receive address registers */
2288 
2289 	if (hw->mac_type == e1000_82542_rev2_0)
2290 		e1000_enter_82542_rst(adapter);
2291 
2292 	/* load the first 14 addresses into the exact filters 1-14. Unicast
2293 	 * addresses take precedence to avoid disabling unicast filtering
2294 	 * when possible.
2295 	 *
2296 	 * RAR 0 is used for the station MAC address
2297 	 * if there are not 14 addresses, go ahead and clear the filters
2298 	 */
2299 	i = 1;
2300 	if (use_uc)
2301 		netdev_for_each_uc_addr(ha, netdev) {
2302 			if (i == rar_entries)
2303 				break;
2304 			e1000_rar_set(hw, ha->addr, i++);
2305 		}
2306 
2307 	netdev_for_each_mc_addr(ha, netdev) {
2308 		if (i == rar_entries) {
2309 			/* load any remaining addresses into the hash table */
2310 			u32 hash_reg, hash_bit, mta;
2311 			hash_value = e1000_hash_mc_addr(hw, ha->addr);
2312 			hash_reg = (hash_value >> 5) & 0x7F;
2313 			hash_bit = hash_value & 0x1F;
2314 			mta = (1 << hash_bit);
2315 			mcarray[hash_reg] |= mta;
2316 		} else {
2317 			e1000_rar_set(hw, ha->addr, i++);
2318 		}
2319 	}
2320 
2321 	for (; i < rar_entries; i++) {
2322 		E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2323 		E1000_WRITE_FLUSH();
2324 		E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2325 		E1000_WRITE_FLUSH();
2326 	}
2327 
2328 	/* write the hash table completely, write from bottom to avoid
2329 	 * both stupid write combining chipsets, and flushing each write */
2330 	for (i = mta_reg_count - 1; i >= 0 ; i--) {
2331 		/*
2332 		 * If we are on an 82544 has an errata where writing odd
2333 		 * offsets overwrites the previous even offset, but writing
2334 		 * backwards over the range solves the issue by always
2335 		 * writing the odd offset first
2336 		 */
2337 		E1000_WRITE_REG_ARRAY(hw, MTA, i, mcarray[i]);
2338 	}
2339 	E1000_WRITE_FLUSH();
2340 
2341 	if (hw->mac_type == e1000_82542_rev2_0)
2342 		e1000_leave_82542_rst(adapter);
2343 
2344 	kfree(mcarray);
2345 }
2346 
2347 /**
2348  * e1000_update_phy_info_task - get phy info
2349  * @work: work struct contained inside adapter struct
2350  *
2351  * Need to wait a few seconds after link up to get diagnostic information from
2352  * the phy
2353  */
2354 static void e1000_update_phy_info_task(struct work_struct *work)
2355 {
2356 	struct e1000_adapter *adapter = container_of(work,
2357 						     struct e1000_adapter,
2358 						     phy_info_task.work);
2359 	if (test_bit(__E1000_DOWN, &adapter->flags))
2360 		return;
2361 	mutex_lock(&adapter->mutex);
2362 	e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
2363 	mutex_unlock(&adapter->mutex);
2364 }
2365 
2366 /**
2367  * e1000_82547_tx_fifo_stall_task - task to complete work
2368  * @work: work struct contained inside adapter struct
2369  **/
2370 static void e1000_82547_tx_fifo_stall_task(struct work_struct *work)
2371 {
2372 	struct e1000_adapter *adapter = container_of(work,
2373 						     struct e1000_adapter,
2374 						     fifo_stall_task.work);
2375 	struct e1000_hw *hw = &adapter->hw;
2376 	struct net_device *netdev = adapter->netdev;
2377 	u32 tctl;
2378 
2379 	if (test_bit(__E1000_DOWN, &adapter->flags))
2380 		return;
2381 	mutex_lock(&adapter->mutex);
2382 	if (atomic_read(&adapter->tx_fifo_stall)) {
2383 		if ((er32(TDT) == er32(TDH)) &&
2384 		   (er32(TDFT) == er32(TDFH)) &&
2385 		   (er32(TDFTS) == er32(TDFHS))) {
2386 			tctl = er32(TCTL);
2387 			ew32(TCTL, tctl & ~E1000_TCTL_EN);
2388 			ew32(TDFT, adapter->tx_head_addr);
2389 			ew32(TDFH, adapter->tx_head_addr);
2390 			ew32(TDFTS, adapter->tx_head_addr);
2391 			ew32(TDFHS, adapter->tx_head_addr);
2392 			ew32(TCTL, tctl);
2393 			E1000_WRITE_FLUSH();
2394 
2395 			adapter->tx_fifo_head = 0;
2396 			atomic_set(&adapter->tx_fifo_stall, 0);
2397 			netif_wake_queue(netdev);
2398 		} else if (!test_bit(__E1000_DOWN, &adapter->flags)) {
2399 			schedule_delayed_work(&adapter->fifo_stall_task, 1);
2400 		}
2401 	}
2402 	mutex_unlock(&adapter->mutex);
2403 }
2404 
2405 bool e1000_has_link(struct e1000_adapter *adapter)
2406 {
2407 	struct e1000_hw *hw = &adapter->hw;
2408 	bool link_active = false;
2409 
2410 	/* get_link_status is set on LSC (link status) interrupt or rx
2411 	 * sequence error interrupt (except on intel ce4100).
2412 	 * get_link_status will stay false until the
2413 	 * e1000_check_for_link establishes link for copper adapters
2414 	 * ONLY
2415 	 */
2416 	switch (hw->media_type) {
2417 	case e1000_media_type_copper:
2418 		if (hw->mac_type == e1000_ce4100)
2419 			hw->get_link_status = 1;
2420 		if (hw->get_link_status) {
2421 			e1000_check_for_link(hw);
2422 			link_active = !hw->get_link_status;
2423 		} else {
2424 			link_active = true;
2425 		}
2426 		break;
2427 	case e1000_media_type_fiber:
2428 		e1000_check_for_link(hw);
2429 		link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2430 		break;
2431 	case e1000_media_type_internal_serdes:
2432 		e1000_check_for_link(hw);
2433 		link_active = hw->serdes_has_link;
2434 		break;
2435 	default:
2436 		break;
2437 	}
2438 
2439 	return link_active;
2440 }
2441 
2442 /**
2443  * e1000_watchdog - work function
2444  * @work: work struct contained inside adapter struct
2445  **/
2446 static void e1000_watchdog(struct work_struct *work)
2447 {
2448 	struct e1000_adapter *adapter = container_of(work,
2449 						     struct e1000_adapter,
2450 						     watchdog_task.work);
2451 	struct e1000_hw *hw = &adapter->hw;
2452 	struct net_device *netdev = adapter->netdev;
2453 	struct e1000_tx_ring *txdr = adapter->tx_ring;
2454 	u32 link, tctl;
2455 
2456 	if (test_bit(__E1000_DOWN, &adapter->flags))
2457 		return;
2458 
2459 	mutex_lock(&adapter->mutex);
2460 	link = e1000_has_link(adapter);
2461 	if ((netif_carrier_ok(netdev)) && link)
2462 		goto link_up;
2463 
2464 	if (link) {
2465 		if (!netif_carrier_ok(netdev)) {
2466 			u32 ctrl;
2467 			bool txb2b = true;
2468 			/* update snapshot of PHY registers on LSC */
2469 			e1000_get_speed_and_duplex(hw,
2470 			                           &adapter->link_speed,
2471 			                           &adapter->link_duplex);
2472 
2473 			ctrl = er32(CTRL);
2474 			pr_info("%s NIC Link is Up %d Mbps %s, "
2475 				"Flow Control: %s\n",
2476 				netdev->name,
2477 				adapter->link_speed,
2478 				adapter->link_duplex == FULL_DUPLEX ?
2479 				"Full Duplex" : "Half Duplex",
2480 				((ctrl & E1000_CTRL_TFCE) && (ctrl &
2481 				E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2482 				E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2483 				E1000_CTRL_TFCE) ? "TX" : "None")));
2484 
2485 			/* adjust timeout factor according to speed/duplex */
2486 			adapter->tx_timeout_factor = 1;
2487 			switch (adapter->link_speed) {
2488 			case SPEED_10:
2489 				txb2b = false;
2490 				adapter->tx_timeout_factor = 16;
2491 				break;
2492 			case SPEED_100:
2493 				txb2b = false;
2494 				/* maybe add some timeout factor ? */
2495 				break;
2496 			}
2497 
2498 			/* enable transmits in the hardware */
2499 			tctl = er32(TCTL);
2500 			tctl |= E1000_TCTL_EN;
2501 			ew32(TCTL, tctl);
2502 
2503 			netif_carrier_on(netdev);
2504 			if (!test_bit(__E1000_DOWN, &adapter->flags))
2505 				schedule_delayed_work(&adapter->phy_info_task,
2506 						      2 * HZ);
2507 			adapter->smartspeed = 0;
2508 		}
2509 	} else {
2510 		if (netif_carrier_ok(netdev)) {
2511 			adapter->link_speed = 0;
2512 			adapter->link_duplex = 0;
2513 			pr_info("%s NIC Link is Down\n",
2514 				netdev->name);
2515 			netif_carrier_off(netdev);
2516 
2517 			if (!test_bit(__E1000_DOWN, &adapter->flags))
2518 				schedule_delayed_work(&adapter->phy_info_task,
2519 						      2 * HZ);
2520 		}
2521 
2522 		e1000_smartspeed(adapter);
2523 	}
2524 
2525 link_up:
2526 	e1000_update_stats(adapter);
2527 
2528 	hw->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2529 	adapter->tpt_old = adapter->stats.tpt;
2530 	hw->collision_delta = adapter->stats.colc - adapter->colc_old;
2531 	adapter->colc_old = adapter->stats.colc;
2532 
2533 	adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2534 	adapter->gorcl_old = adapter->stats.gorcl;
2535 	adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2536 	adapter->gotcl_old = adapter->stats.gotcl;
2537 
2538 	e1000_update_adaptive(hw);
2539 
2540 	if (!netif_carrier_ok(netdev)) {
2541 		if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2542 			/* We've lost link, so the controller stops DMA,
2543 			 * but we've got queued Tx work that's never going
2544 			 * to get done, so reset controller to flush Tx.
2545 			 * (Do the reset outside of interrupt context). */
2546 			adapter->tx_timeout_count++;
2547 			schedule_work(&adapter->reset_task);
2548 			/* exit immediately since reset is imminent */
2549 			goto unlock;
2550 		}
2551 	}
2552 
2553 	/* Simple mode for Interrupt Throttle Rate (ITR) */
2554 	if (hw->mac_type >= e1000_82540 && adapter->itr_setting == 4) {
2555 		/*
2556 		 * Symmetric Tx/Rx gets a reduced ITR=2000;
2557 		 * Total asymmetrical Tx or Rx gets ITR=8000;
2558 		 * everyone else is between 2000-8000.
2559 		 */
2560 		u32 goc = (adapter->gotcl + adapter->gorcl) / 10000;
2561 		u32 dif = (adapter->gotcl > adapter->gorcl ?
2562 			    adapter->gotcl - adapter->gorcl :
2563 			    adapter->gorcl - adapter->gotcl) / 10000;
2564 		u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
2565 
2566 		ew32(ITR, 1000000000 / (itr * 256));
2567 	}
2568 
2569 	/* Cause software interrupt to ensure rx ring is cleaned */
2570 	ew32(ICS, E1000_ICS_RXDMT0);
2571 
2572 	/* Force detection of hung controller every watchdog period */
2573 	adapter->detect_tx_hung = true;
2574 
2575 	/* Reschedule the task */
2576 	if (!test_bit(__E1000_DOWN, &adapter->flags))
2577 		schedule_delayed_work(&adapter->watchdog_task, 2 * HZ);
2578 
2579 unlock:
2580 	mutex_unlock(&adapter->mutex);
2581 }
2582 
2583 enum latency_range {
2584 	lowest_latency = 0,
2585 	low_latency = 1,
2586 	bulk_latency = 2,
2587 	latency_invalid = 255
2588 };
2589 
2590 /**
2591  * e1000_update_itr - update the dynamic ITR value based on statistics
2592  * @adapter: pointer to adapter
2593  * @itr_setting: current adapter->itr
2594  * @packets: the number of packets during this measurement interval
2595  * @bytes: the number of bytes during this measurement interval
2596  *
2597  *      Stores a new ITR value based on packets and byte
2598  *      counts during the last interrupt.  The advantage of per interrupt
2599  *      computation is faster updates and more accurate ITR for the current
2600  *      traffic pattern.  Constants in this function were computed
2601  *      based on theoretical maximum wire speed and thresholds were set based
2602  *      on testing data as well as attempting to minimize response time
2603  *      while increasing bulk throughput.
2604  *      this functionality is controlled by the InterruptThrottleRate module
2605  *      parameter (see e1000_param.c)
2606  **/
2607 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2608 				     u16 itr_setting, int packets, int bytes)
2609 {
2610 	unsigned int retval = itr_setting;
2611 	struct e1000_hw *hw = &adapter->hw;
2612 
2613 	if (unlikely(hw->mac_type < e1000_82540))
2614 		goto update_itr_done;
2615 
2616 	if (packets == 0)
2617 		goto update_itr_done;
2618 
2619 	switch (itr_setting) {
2620 	case lowest_latency:
2621 		/* jumbo frames get bulk treatment*/
2622 		if (bytes/packets > 8000)
2623 			retval = bulk_latency;
2624 		else if ((packets < 5) && (bytes > 512))
2625 			retval = low_latency;
2626 		break;
2627 	case low_latency:  /* 50 usec aka 20000 ints/s */
2628 		if (bytes > 10000) {
2629 			/* jumbo frames need bulk latency setting */
2630 			if (bytes/packets > 8000)
2631 				retval = bulk_latency;
2632 			else if ((packets < 10) || ((bytes/packets) > 1200))
2633 				retval = bulk_latency;
2634 			else if ((packets > 35))
2635 				retval = lowest_latency;
2636 		} else if (bytes/packets > 2000)
2637 			retval = bulk_latency;
2638 		else if (packets <= 2 && bytes < 512)
2639 			retval = lowest_latency;
2640 		break;
2641 	case bulk_latency: /* 250 usec aka 4000 ints/s */
2642 		if (bytes > 25000) {
2643 			if (packets > 35)
2644 				retval = low_latency;
2645 		} else if (bytes < 6000) {
2646 			retval = low_latency;
2647 		}
2648 		break;
2649 	}
2650 
2651 update_itr_done:
2652 	return retval;
2653 }
2654 
2655 static void e1000_set_itr(struct e1000_adapter *adapter)
2656 {
2657 	struct e1000_hw *hw = &adapter->hw;
2658 	u16 current_itr;
2659 	u32 new_itr = adapter->itr;
2660 
2661 	if (unlikely(hw->mac_type < e1000_82540))
2662 		return;
2663 
2664 	/* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2665 	if (unlikely(adapter->link_speed != SPEED_1000)) {
2666 		current_itr = 0;
2667 		new_itr = 4000;
2668 		goto set_itr_now;
2669 	}
2670 
2671 	adapter->tx_itr = e1000_update_itr(adapter,
2672 	                            adapter->tx_itr,
2673 	                            adapter->total_tx_packets,
2674 	                            adapter->total_tx_bytes);
2675 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
2676 	if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2677 		adapter->tx_itr = low_latency;
2678 
2679 	adapter->rx_itr = e1000_update_itr(adapter,
2680 	                            adapter->rx_itr,
2681 	                            adapter->total_rx_packets,
2682 	                            adapter->total_rx_bytes);
2683 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
2684 	if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2685 		adapter->rx_itr = low_latency;
2686 
2687 	current_itr = max(adapter->rx_itr, adapter->tx_itr);
2688 
2689 	switch (current_itr) {
2690 	/* counts and packets in update_itr are dependent on these numbers */
2691 	case lowest_latency:
2692 		new_itr = 70000;
2693 		break;
2694 	case low_latency:
2695 		new_itr = 20000; /* aka hwitr = ~200 */
2696 		break;
2697 	case bulk_latency:
2698 		new_itr = 4000;
2699 		break;
2700 	default:
2701 		break;
2702 	}
2703 
2704 set_itr_now:
2705 	if (new_itr != adapter->itr) {
2706 		/* this attempts to bias the interrupt rate towards Bulk
2707 		 * by adding intermediate steps when interrupt rate is
2708 		 * increasing */
2709 		new_itr = new_itr > adapter->itr ?
2710 		             min(adapter->itr + (new_itr >> 2), new_itr) :
2711 		             new_itr;
2712 		adapter->itr = new_itr;
2713 		ew32(ITR, 1000000000 / (new_itr * 256));
2714 	}
2715 }
2716 
2717 #define E1000_TX_FLAGS_CSUM		0x00000001
2718 #define E1000_TX_FLAGS_VLAN		0x00000002
2719 #define E1000_TX_FLAGS_TSO		0x00000004
2720 #define E1000_TX_FLAGS_IPV4		0x00000008
2721 #define E1000_TX_FLAGS_NO_FCS		0x00000010
2722 #define E1000_TX_FLAGS_VLAN_MASK	0xffff0000
2723 #define E1000_TX_FLAGS_VLAN_SHIFT	16
2724 
2725 static int e1000_tso(struct e1000_adapter *adapter,
2726 		     struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2727 {
2728 	struct e1000_context_desc *context_desc;
2729 	struct e1000_buffer *buffer_info;
2730 	unsigned int i;
2731 	u32 cmd_length = 0;
2732 	u16 ipcse = 0, tucse, mss;
2733 	u8 ipcss, ipcso, tucss, tucso, hdr_len;
2734 	int err;
2735 
2736 	if (skb_is_gso(skb)) {
2737 		if (skb_header_cloned(skb)) {
2738 			err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2739 			if (err)
2740 				return err;
2741 		}
2742 
2743 		hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2744 		mss = skb_shinfo(skb)->gso_size;
2745 		if (skb->protocol == htons(ETH_P_IP)) {
2746 			struct iphdr *iph = ip_hdr(skb);
2747 			iph->tot_len = 0;
2748 			iph->check = 0;
2749 			tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2750 								 iph->daddr, 0,
2751 								 IPPROTO_TCP,
2752 								 0);
2753 			cmd_length = E1000_TXD_CMD_IP;
2754 			ipcse = skb_transport_offset(skb) - 1;
2755 		} else if (skb->protocol == htons(ETH_P_IPV6)) {
2756 			ipv6_hdr(skb)->payload_len = 0;
2757 			tcp_hdr(skb)->check =
2758 				~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2759 						 &ipv6_hdr(skb)->daddr,
2760 						 0, IPPROTO_TCP, 0);
2761 			ipcse = 0;
2762 		}
2763 		ipcss = skb_network_offset(skb);
2764 		ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2765 		tucss = skb_transport_offset(skb);
2766 		tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2767 		tucse = 0;
2768 
2769 		cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2770 			       E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2771 
2772 		i = tx_ring->next_to_use;
2773 		context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2774 		buffer_info = &tx_ring->buffer_info[i];
2775 
2776 		context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2777 		context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2778 		context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2779 		context_desc->upper_setup.tcp_fields.tucss = tucss;
2780 		context_desc->upper_setup.tcp_fields.tucso = tucso;
2781 		context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2782 		context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2783 		context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2784 		context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2785 
2786 		buffer_info->time_stamp = jiffies;
2787 		buffer_info->next_to_watch = i;
2788 
2789 		if (++i == tx_ring->count) i = 0;
2790 		tx_ring->next_to_use = i;
2791 
2792 		return true;
2793 	}
2794 	return false;
2795 }
2796 
2797 static bool e1000_tx_csum(struct e1000_adapter *adapter,
2798 			  struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2799 {
2800 	struct e1000_context_desc *context_desc;
2801 	struct e1000_buffer *buffer_info;
2802 	unsigned int i;
2803 	u8 css;
2804 	u32 cmd_len = E1000_TXD_CMD_DEXT;
2805 
2806 	if (skb->ip_summed != CHECKSUM_PARTIAL)
2807 		return false;
2808 
2809 	switch (skb->protocol) {
2810 	case cpu_to_be16(ETH_P_IP):
2811 		if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2812 			cmd_len |= E1000_TXD_CMD_TCP;
2813 		break;
2814 	case cpu_to_be16(ETH_P_IPV6):
2815 		/* XXX not handling all IPV6 headers */
2816 		if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2817 			cmd_len |= E1000_TXD_CMD_TCP;
2818 		break;
2819 	default:
2820 		if (unlikely(net_ratelimit()))
2821 			e_warn(drv, "checksum_partial proto=%x!\n",
2822 			       skb->protocol);
2823 		break;
2824 	}
2825 
2826 	css = skb_checksum_start_offset(skb);
2827 
2828 	i = tx_ring->next_to_use;
2829 	buffer_info = &tx_ring->buffer_info[i];
2830 	context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2831 
2832 	context_desc->lower_setup.ip_config = 0;
2833 	context_desc->upper_setup.tcp_fields.tucss = css;
2834 	context_desc->upper_setup.tcp_fields.tucso =
2835 		css + skb->csum_offset;
2836 	context_desc->upper_setup.tcp_fields.tucse = 0;
2837 	context_desc->tcp_seg_setup.data = 0;
2838 	context_desc->cmd_and_length = cpu_to_le32(cmd_len);
2839 
2840 	buffer_info->time_stamp = jiffies;
2841 	buffer_info->next_to_watch = i;
2842 
2843 	if (unlikely(++i == tx_ring->count)) i = 0;
2844 	tx_ring->next_to_use = i;
2845 
2846 	return true;
2847 }
2848 
2849 #define E1000_MAX_TXD_PWR	12
2850 #define E1000_MAX_DATA_PER_TXD	(1<<E1000_MAX_TXD_PWR)
2851 
2852 static int e1000_tx_map(struct e1000_adapter *adapter,
2853 			struct e1000_tx_ring *tx_ring,
2854 			struct sk_buff *skb, unsigned int first,
2855 			unsigned int max_per_txd, unsigned int nr_frags,
2856 			unsigned int mss)
2857 {
2858 	struct e1000_hw *hw = &adapter->hw;
2859 	struct pci_dev *pdev = adapter->pdev;
2860 	struct e1000_buffer *buffer_info;
2861 	unsigned int len = skb_headlen(skb);
2862 	unsigned int offset = 0, size, count = 0, i;
2863 	unsigned int f, bytecount, segs;
2864 
2865 	i = tx_ring->next_to_use;
2866 
2867 	while (len) {
2868 		buffer_info = &tx_ring->buffer_info[i];
2869 		size = min(len, max_per_txd);
2870 		/* Workaround for Controller erratum --
2871 		 * descriptor for non-tso packet in a linear SKB that follows a
2872 		 * tso gets written back prematurely before the data is fully
2873 		 * DMA'd to the controller */
2874 		if (!skb->data_len && tx_ring->last_tx_tso &&
2875 		    !skb_is_gso(skb)) {
2876 			tx_ring->last_tx_tso = false;
2877 			size -= 4;
2878 		}
2879 
2880 		/* Workaround for premature desc write-backs
2881 		 * in TSO mode.  Append 4-byte sentinel desc */
2882 		if (unlikely(mss && !nr_frags && size == len && size > 8))
2883 			size -= 4;
2884 		/* work-around for errata 10 and it applies
2885 		 * to all controllers in PCI-X mode
2886 		 * The fix is to make sure that the first descriptor of a
2887 		 * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2888 		 */
2889 		if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
2890 		                (size > 2015) && count == 0))
2891 		        size = 2015;
2892 
2893 		/* Workaround for potential 82544 hang in PCI-X.  Avoid
2894 		 * terminating buffers within evenly-aligned dwords. */
2895 		if (unlikely(adapter->pcix_82544 &&
2896 		   !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2897 		   size > 4))
2898 			size -= 4;
2899 
2900 		buffer_info->length = size;
2901 		/* set time_stamp *before* dma to help avoid a possible race */
2902 		buffer_info->time_stamp = jiffies;
2903 		buffer_info->mapped_as_page = false;
2904 		buffer_info->dma = dma_map_single(&pdev->dev,
2905 						  skb->data + offset,
2906 						  size,	DMA_TO_DEVICE);
2907 		if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2908 			goto dma_error;
2909 		buffer_info->next_to_watch = i;
2910 
2911 		len -= size;
2912 		offset += size;
2913 		count++;
2914 		if (len) {
2915 			i++;
2916 			if (unlikely(i == tx_ring->count))
2917 				i = 0;
2918 		}
2919 	}
2920 
2921 	for (f = 0; f < nr_frags; f++) {
2922 		const struct skb_frag_struct *frag;
2923 
2924 		frag = &skb_shinfo(skb)->frags[f];
2925 		len = skb_frag_size(frag);
2926 		offset = 0;
2927 
2928 		while (len) {
2929 			unsigned long bufend;
2930 			i++;
2931 			if (unlikely(i == tx_ring->count))
2932 				i = 0;
2933 
2934 			buffer_info = &tx_ring->buffer_info[i];
2935 			size = min(len, max_per_txd);
2936 			/* Workaround for premature desc write-backs
2937 			 * in TSO mode.  Append 4-byte sentinel desc */
2938 			if (unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
2939 				size -= 4;
2940 			/* Workaround for potential 82544 hang in PCI-X.
2941 			 * Avoid terminating buffers within evenly-aligned
2942 			 * dwords. */
2943 			bufend = (unsigned long)
2944 				page_to_phys(skb_frag_page(frag));
2945 			bufend += offset + size - 1;
2946 			if (unlikely(adapter->pcix_82544 &&
2947 				     !(bufend & 4) &&
2948 				     size > 4))
2949 				size -= 4;
2950 
2951 			buffer_info->length = size;
2952 			buffer_info->time_stamp = jiffies;
2953 			buffer_info->mapped_as_page = true;
2954 			buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
2955 						offset, size, DMA_TO_DEVICE);
2956 			if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2957 				goto dma_error;
2958 			buffer_info->next_to_watch = i;
2959 
2960 			len -= size;
2961 			offset += size;
2962 			count++;
2963 		}
2964 	}
2965 
2966 	segs = skb_shinfo(skb)->gso_segs ?: 1;
2967 	/* multiply data chunks by size of headers */
2968 	bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
2969 
2970 	tx_ring->buffer_info[i].skb = skb;
2971 	tx_ring->buffer_info[i].segs = segs;
2972 	tx_ring->buffer_info[i].bytecount = bytecount;
2973 	tx_ring->buffer_info[first].next_to_watch = i;
2974 
2975 	return count;
2976 
2977 dma_error:
2978 	dev_err(&pdev->dev, "TX DMA map failed\n");
2979 	buffer_info->dma = 0;
2980 	if (count)
2981 		count--;
2982 
2983 	while (count--) {
2984 		if (i==0)
2985 			i += tx_ring->count;
2986 		i--;
2987 		buffer_info = &tx_ring->buffer_info[i];
2988 		e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2989 	}
2990 
2991 	return 0;
2992 }
2993 
2994 static void e1000_tx_queue(struct e1000_adapter *adapter,
2995 			   struct e1000_tx_ring *tx_ring, int tx_flags,
2996 			   int count)
2997 {
2998 	struct e1000_hw *hw = &adapter->hw;
2999 	struct e1000_tx_desc *tx_desc = NULL;
3000 	struct e1000_buffer *buffer_info;
3001 	u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3002 	unsigned int i;
3003 
3004 	if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
3005 		txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3006 		             E1000_TXD_CMD_TSE;
3007 		txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3008 
3009 		if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
3010 			txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3011 	}
3012 
3013 	if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
3014 		txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3015 		txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3016 	}
3017 
3018 	if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
3019 		txd_lower |= E1000_TXD_CMD_VLE;
3020 		txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3021 	}
3022 
3023 	if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
3024 		txd_lower &= ~(E1000_TXD_CMD_IFCS);
3025 
3026 	i = tx_ring->next_to_use;
3027 
3028 	while (count--) {
3029 		buffer_info = &tx_ring->buffer_info[i];
3030 		tx_desc = E1000_TX_DESC(*tx_ring, i);
3031 		tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3032 		tx_desc->lower.data =
3033 			cpu_to_le32(txd_lower | buffer_info->length);
3034 		tx_desc->upper.data = cpu_to_le32(txd_upper);
3035 		if (unlikely(++i == tx_ring->count)) i = 0;
3036 	}
3037 
3038 	tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3039 
3040 	/* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
3041 	if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
3042 		tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
3043 
3044 	/* Force memory writes to complete before letting h/w
3045 	 * know there are new descriptors to fetch.  (Only
3046 	 * applicable for weak-ordered memory model archs,
3047 	 * such as IA-64). */
3048 	wmb();
3049 
3050 	tx_ring->next_to_use = i;
3051 	writel(i, hw->hw_addr + tx_ring->tdt);
3052 	/* we need this if more than one processor can write to our tail
3053 	 * at a time, it syncronizes IO on IA64/Altix systems */
3054 	mmiowb();
3055 }
3056 
3057 /* 82547 workaround to avoid controller hang in half-duplex environment.
3058  * The workaround is to avoid queuing a large packet that would span
3059  * the internal Tx FIFO ring boundary by notifying the stack to resend
3060  * the packet at a later time.  This gives the Tx FIFO an opportunity to
3061  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
3062  * to the beginning of the Tx FIFO.
3063  */
3064 
3065 #define E1000_FIFO_HDR			0x10
3066 #define E1000_82547_PAD_LEN		0x3E0
3067 
3068 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
3069 				       struct sk_buff *skb)
3070 {
3071 	u32 fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
3072 	u32 skb_fifo_len = skb->len + E1000_FIFO_HDR;
3073 
3074 	skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
3075 
3076 	if (adapter->link_duplex != HALF_DUPLEX)
3077 		goto no_fifo_stall_required;
3078 
3079 	if (atomic_read(&adapter->tx_fifo_stall))
3080 		return 1;
3081 
3082 	if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
3083 		atomic_set(&adapter->tx_fifo_stall, 1);
3084 		return 1;
3085 	}
3086 
3087 no_fifo_stall_required:
3088 	adapter->tx_fifo_head += skb_fifo_len;
3089 	if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
3090 		adapter->tx_fifo_head -= adapter->tx_fifo_size;
3091 	return 0;
3092 }
3093 
3094 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3095 {
3096 	struct e1000_adapter *adapter = netdev_priv(netdev);
3097 	struct e1000_tx_ring *tx_ring = adapter->tx_ring;
3098 
3099 	netif_stop_queue(netdev);
3100 	/* Herbert's original patch had:
3101 	 *  smp_mb__after_netif_stop_queue();
3102 	 * but since that doesn't exist yet, just open code it. */
3103 	smp_mb();
3104 
3105 	/* We need to check again in a case another CPU has just
3106 	 * made room available. */
3107 	if (likely(E1000_DESC_UNUSED(tx_ring) < size))
3108 		return -EBUSY;
3109 
3110 	/* A reprieve! */
3111 	netif_start_queue(netdev);
3112 	++adapter->restart_queue;
3113 	return 0;
3114 }
3115 
3116 static int e1000_maybe_stop_tx(struct net_device *netdev,
3117                                struct e1000_tx_ring *tx_ring, int size)
3118 {
3119 	if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
3120 		return 0;
3121 	return __e1000_maybe_stop_tx(netdev, size);
3122 }
3123 
3124 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3125 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
3126 				    struct net_device *netdev)
3127 {
3128 	struct e1000_adapter *adapter = netdev_priv(netdev);
3129 	struct e1000_hw *hw = &adapter->hw;
3130 	struct e1000_tx_ring *tx_ring;
3131 	unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
3132 	unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3133 	unsigned int tx_flags = 0;
3134 	unsigned int len = skb_headlen(skb);
3135 	unsigned int nr_frags;
3136 	unsigned int mss;
3137 	int count = 0;
3138 	int tso;
3139 	unsigned int f;
3140 
3141 	/* This goes back to the question of how to logically map a tx queue
3142 	 * to a flow.  Right now, performance is impacted slightly negatively
3143 	 * if using multiple tx queues.  If the stack breaks away from a
3144 	 * single qdisc implementation, we can look at this again. */
3145 	tx_ring = adapter->tx_ring;
3146 
3147 	if (unlikely(skb->len <= 0)) {
3148 		dev_kfree_skb_any(skb);
3149 		return NETDEV_TX_OK;
3150 	}
3151 
3152 	mss = skb_shinfo(skb)->gso_size;
3153 	/* The controller does a simple calculation to
3154 	 * make sure there is enough room in the FIFO before
3155 	 * initiating the DMA for each buffer.  The calc is:
3156 	 * 4 = ceil(buffer len/mss).  To make sure we don't
3157 	 * overrun the FIFO, adjust the max buffer len if mss
3158 	 * drops. */
3159 	if (mss) {
3160 		u8 hdr_len;
3161 		max_per_txd = min(mss << 2, max_per_txd);
3162 		max_txd_pwr = fls(max_per_txd) - 1;
3163 
3164 		hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3165 		if (skb->data_len && hdr_len == len) {
3166 			switch (hw->mac_type) {
3167 				unsigned int pull_size;
3168 			case e1000_82544:
3169 				/* Make sure we have room to chop off 4 bytes,
3170 				 * and that the end alignment will work out to
3171 				 * this hardware's requirements
3172 				 * NOTE: this is a TSO only workaround
3173 				 * if end byte alignment not correct move us
3174 				 * into the next dword */
3175 				if ((unsigned long)(skb_tail_pointer(skb) - 1) & 4)
3176 					break;
3177 				/* fall through */
3178 				pull_size = min((unsigned int)4, skb->data_len);
3179 				if (!__pskb_pull_tail(skb, pull_size)) {
3180 					e_err(drv, "__pskb_pull_tail "
3181 					      "failed.\n");
3182 					dev_kfree_skb_any(skb);
3183 					return NETDEV_TX_OK;
3184 				}
3185 				len = skb_headlen(skb);
3186 				break;
3187 			default:
3188 				/* do nothing */
3189 				break;
3190 			}
3191 		}
3192 	}
3193 
3194 	/* reserve a descriptor for the offload context */
3195 	if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3196 		count++;
3197 	count++;
3198 
3199 	/* Controller Erratum workaround */
3200 	if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3201 		count++;
3202 
3203 	count += TXD_USE_COUNT(len, max_txd_pwr);
3204 
3205 	if (adapter->pcix_82544)
3206 		count++;
3207 
3208 	/* work-around for errata 10 and it applies to all controllers
3209 	 * in PCI-X mode, so add one more descriptor to the count
3210 	 */
3211 	if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
3212 			(len > 2015)))
3213 		count++;
3214 
3215 	nr_frags = skb_shinfo(skb)->nr_frags;
3216 	for (f = 0; f < nr_frags; f++)
3217 		count += TXD_USE_COUNT(skb_frag_size(&skb_shinfo(skb)->frags[f]),
3218 				       max_txd_pwr);
3219 	if (adapter->pcix_82544)
3220 		count += nr_frags;
3221 
3222 	/* need: count + 2 desc gap to keep tail from touching
3223 	 * head, otherwise try next time */
3224 	if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2)))
3225 		return NETDEV_TX_BUSY;
3226 
3227 	if (unlikely((hw->mac_type == e1000_82547) &&
3228 		     (e1000_82547_fifo_workaround(adapter, skb)))) {
3229 		netif_stop_queue(netdev);
3230 		if (!test_bit(__E1000_DOWN, &adapter->flags))
3231 			schedule_delayed_work(&adapter->fifo_stall_task, 1);
3232 		return NETDEV_TX_BUSY;
3233 	}
3234 
3235 	if (vlan_tx_tag_present(skb)) {
3236 		tx_flags |= E1000_TX_FLAGS_VLAN;
3237 		tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3238 	}
3239 
3240 	first = tx_ring->next_to_use;
3241 
3242 	tso = e1000_tso(adapter, tx_ring, skb);
3243 	if (tso < 0) {
3244 		dev_kfree_skb_any(skb);
3245 		return NETDEV_TX_OK;
3246 	}
3247 
3248 	if (likely(tso)) {
3249 		if (likely(hw->mac_type != e1000_82544))
3250 			tx_ring->last_tx_tso = true;
3251 		tx_flags |= E1000_TX_FLAGS_TSO;
3252 	} else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3253 		tx_flags |= E1000_TX_FLAGS_CSUM;
3254 
3255 	if (likely(skb->protocol == htons(ETH_P_IP)))
3256 		tx_flags |= E1000_TX_FLAGS_IPV4;
3257 
3258 	if (unlikely(skb->no_fcs))
3259 		tx_flags |= E1000_TX_FLAGS_NO_FCS;
3260 
3261 	count = e1000_tx_map(adapter, tx_ring, skb, first, max_per_txd,
3262 	                     nr_frags, mss);
3263 
3264 	if (count) {
3265 		skb_tx_timestamp(skb);
3266 
3267 		e1000_tx_queue(adapter, tx_ring, tx_flags, count);
3268 		/* Make sure there is space in the ring for the next send. */
3269 		e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3270 
3271 	} else {
3272 		dev_kfree_skb_any(skb);
3273 		tx_ring->buffer_info[first].time_stamp = 0;
3274 		tx_ring->next_to_use = first;
3275 	}
3276 
3277 	return NETDEV_TX_OK;
3278 }
3279 
3280 #define NUM_REGS 38 /* 1 based count */
3281 static void e1000_regdump(struct e1000_adapter *adapter)
3282 {
3283 	struct e1000_hw *hw = &adapter->hw;
3284 	u32 regs[NUM_REGS];
3285 	u32 *regs_buff = regs;
3286 	int i = 0;
3287 
3288 	static const char * const reg_name[] = {
3289 		"CTRL",  "STATUS",
3290 		"RCTL", "RDLEN", "RDH", "RDT", "RDTR",
3291 		"TCTL", "TDBAL", "TDBAH", "TDLEN", "TDH", "TDT",
3292 		"TIDV", "TXDCTL", "TADV", "TARC0",
3293 		"TDBAL1", "TDBAH1", "TDLEN1", "TDH1", "TDT1",
3294 		"TXDCTL1", "TARC1",
3295 		"CTRL_EXT", "ERT", "RDBAL", "RDBAH",
3296 		"TDFH", "TDFT", "TDFHS", "TDFTS", "TDFPC",
3297 		"RDFH", "RDFT", "RDFHS", "RDFTS", "RDFPC"
3298 	};
3299 
3300 	regs_buff[0]  = er32(CTRL);
3301 	regs_buff[1]  = er32(STATUS);
3302 
3303 	regs_buff[2]  = er32(RCTL);
3304 	regs_buff[3]  = er32(RDLEN);
3305 	regs_buff[4]  = er32(RDH);
3306 	regs_buff[5]  = er32(RDT);
3307 	regs_buff[6]  = er32(RDTR);
3308 
3309 	regs_buff[7]  = er32(TCTL);
3310 	regs_buff[8]  = er32(TDBAL);
3311 	regs_buff[9]  = er32(TDBAH);
3312 	regs_buff[10] = er32(TDLEN);
3313 	regs_buff[11] = er32(TDH);
3314 	regs_buff[12] = er32(TDT);
3315 	regs_buff[13] = er32(TIDV);
3316 	regs_buff[14] = er32(TXDCTL);
3317 	regs_buff[15] = er32(TADV);
3318 	regs_buff[16] = er32(TARC0);
3319 
3320 	regs_buff[17] = er32(TDBAL1);
3321 	regs_buff[18] = er32(TDBAH1);
3322 	regs_buff[19] = er32(TDLEN1);
3323 	regs_buff[20] = er32(TDH1);
3324 	regs_buff[21] = er32(TDT1);
3325 	regs_buff[22] = er32(TXDCTL1);
3326 	regs_buff[23] = er32(TARC1);
3327 	regs_buff[24] = er32(CTRL_EXT);
3328 	regs_buff[25] = er32(ERT);
3329 	regs_buff[26] = er32(RDBAL0);
3330 	regs_buff[27] = er32(RDBAH0);
3331 	regs_buff[28] = er32(TDFH);
3332 	regs_buff[29] = er32(TDFT);
3333 	regs_buff[30] = er32(TDFHS);
3334 	regs_buff[31] = er32(TDFTS);
3335 	regs_buff[32] = er32(TDFPC);
3336 	regs_buff[33] = er32(RDFH);
3337 	regs_buff[34] = er32(RDFT);
3338 	regs_buff[35] = er32(RDFHS);
3339 	regs_buff[36] = er32(RDFTS);
3340 	regs_buff[37] = er32(RDFPC);
3341 
3342 	pr_info("Register dump\n");
3343 	for (i = 0; i < NUM_REGS; i++)
3344 		pr_info("%-15s  %08x\n", reg_name[i], regs_buff[i]);
3345 }
3346 
3347 /*
3348  * e1000_dump: Print registers, tx ring and rx ring
3349  */
3350 static void e1000_dump(struct e1000_adapter *adapter)
3351 {
3352 	/* this code doesn't handle multiple rings */
3353 	struct e1000_tx_ring *tx_ring = adapter->tx_ring;
3354 	struct e1000_rx_ring *rx_ring = adapter->rx_ring;
3355 	int i;
3356 
3357 	if (!netif_msg_hw(adapter))
3358 		return;
3359 
3360 	/* Print Registers */
3361 	e1000_regdump(adapter);
3362 
3363 	/*
3364 	 * transmit dump
3365 	 */
3366 	pr_info("TX Desc ring0 dump\n");
3367 
3368 	/* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
3369 	 *
3370 	 * Legacy Transmit Descriptor
3371 	 *   +--------------------------------------------------------------+
3372 	 * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
3373 	 *   +--------------------------------------------------------------+
3374 	 * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
3375 	 *   +--------------------------------------------------------------+
3376 	 *   63       48 47        36 35    32 31     24 23    16 15        0
3377 	 *
3378 	 * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
3379 	 *   63      48 47    40 39       32 31             16 15    8 7      0
3380 	 *   +----------------------------------------------------------------+
3381 	 * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
3382 	 *   +----------------------------------------------------------------+
3383 	 * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
3384 	 *   +----------------------------------------------------------------+
3385 	 *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
3386 	 *
3387 	 * Extended Data Descriptor (DTYP=0x1)
3388 	 *   +----------------------------------------------------------------+
3389 	 * 0 |                     Buffer Address [63:0]                      |
3390 	 *   +----------------------------------------------------------------+
3391 	 * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
3392 	 *   +----------------------------------------------------------------+
3393 	 *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
3394 	 */
3395 	pr_info("Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma       ] leng  ntw timestmp         bi->skb\n");
3396 	pr_info("Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen] [bi->dma       ] leng  ntw timestmp         bi->skb\n");
3397 
3398 	if (!netif_msg_tx_done(adapter))
3399 		goto rx_ring_summary;
3400 
3401 	for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
3402 		struct e1000_tx_desc *tx_desc = E1000_TX_DESC(*tx_ring, i);
3403 		struct e1000_buffer *buffer_info = &tx_ring->buffer_info[i];
3404 		struct my_u { __le64 a; __le64 b; };
3405 		struct my_u *u = (struct my_u *)tx_desc;
3406 		const char *type;
3407 
3408 		if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
3409 			type = "NTC/U";
3410 		else if (i == tx_ring->next_to_use)
3411 			type = "NTU";
3412 		else if (i == tx_ring->next_to_clean)
3413 			type = "NTC";
3414 		else
3415 			type = "";
3416 
3417 		pr_info("T%c[0x%03X]    %016llX %016llX %016llX %04X  %3X %016llX %p %s\n",
3418 			((le64_to_cpu(u->b) & (1<<20)) ? 'd' : 'c'), i,
3419 			le64_to_cpu(u->a), le64_to_cpu(u->b),
3420 			(u64)buffer_info->dma, buffer_info->length,
3421 			buffer_info->next_to_watch,
3422 			(u64)buffer_info->time_stamp, buffer_info->skb, type);
3423 	}
3424 
3425 rx_ring_summary:
3426 	/*
3427 	 * receive dump
3428 	 */
3429 	pr_info("\nRX Desc ring dump\n");
3430 
3431 	/* Legacy Receive Descriptor Format
3432 	 *
3433 	 * +-----------------------------------------------------+
3434 	 * |                Buffer Address [63:0]                |
3435 	 * +-----------------------------------------------------+
3436 	 * | VLAN Tag | Errors | Status 0 | Packet csum | Length |
3437 	 * +-----------------------------------------------------+
3438 	 * 63       48 47    40 39      32 31         16 15      0
3439 	 */
3440 	pr_info("R[desc]      [address 63:0  ] [vl er S cks ln] [bi->dma       ] [bi->skb]\n");
3441 
3442 	if (!netif_msg_rx_status(adapter))
3443 		goto exit;
3444 
3445 	for (i = 0; rx_ring->desc && (i < rx_ring->count); i++) {
3446 		struct e1000_rx_desc *rx_desc = E1000_RX_DESC(*rx_ring, i);
3447 		struct e1000_buffer *buffer_info = &rx_ring->buffer_info[i];
3448 		struct my_u { __le64 a; __le64 b; };
3449 		struct my_u *u = (struct my_u *)rx_desc;
3450 		const char *type;
3451 
3452 		if (i == rx_ring->next_to_use)
3453 			type = "NTU";
3454 		else if (i == rx_ring->next_to_clean)
3455 			type = "NTC";
3456 		else
3457 			type = "";
3458 
3459 		pr_info("R[0x%03X]     %016llX %016llX %016llX %p %s\n",
3460 			i, le64_to_cpu(u->a), le64_to_cpu(u->b),
3461 			(u64)buffer_info->dma, buffer_info->skb, type);
3462 	} /* for */
3463 
3464 	/* dump the descriptor caches */
3465 	/* rx */
3466 	pr_info("Rx descriptor cache in 64bit format\n");
3467 	for (i = 0x6000; i <= 0x63FF ; i += 0x10) {
3468 		pr_info("R%04X: %08X|%08X %08X|%08X\n",
3469 			i,
3470 			readl(adapter->hw.hw_addr + i+4),
3471 			readl(adapter->hw.hw_addr + i),
3472 			readl(adapter->hw.hw_addr + i+12),
3473 			readl(adapter->hw.hw_addr + i+8));
3474 	}
3475 	/* tx */
3476 	pr_info("Tx descriptor cache in 64bit format\n");
3477 	for (i = 0x7000; i <= 0x73FF ; i += 0x10) {
3478 		pr_info("T%04X: %08X|%08X %08X|%08X\n",
3479 			i,
3480 			readl(adapter->hw.hw_addr + i+4),
3481 			readl(adapter->hw.hw_addr + i),
3482 			readl(adapter->hw.hw_addr + i+12),
3483 			readl(adapter->hw.hw_addr + i+8));
3484 	}
3485 exit:
3486 	return;
3487 }
3488 
3489 /**
3490  * e1000_tx_timeout - Respond to a Tx Hang
3491  * @netdev: network interface device structure
3492  **/
3493 
3494 static void e1000_tx_timeout(struct net_device *netdev)
3495 {
3496 	struct e1000_adapter *adapter = netdev_priv(netdev);
3497 
3498 	/* Do the reset outside of interrupt context */
3499 	adapter->tx_timeout_count++;
3500 	schedule_work(&adapter->reset_task);
3501 }
3502 
3503 static void e1000_reset_task(struct work_struct *work)
3504 {
3505 	struct e1000_adapter *adapter =
3506 		container_of(work, struct e1000_adapter, reset_task);
3507 
3508 	if (test_bit(__E1000_DOWN, &adapter->flags))
3509 		return;
3510 	e_err(drv, "Reset adapter\n");
3511 	e1000_reinit_safe(adapter);
3512 }
3513 
3514 /**
3515  * e1000_get_stats - Get System Network Statistics
3516  * @netdev: network interface device structure
3517  *
3518  * Returns the address of the device statistics structure.
3519  * The statistics are actually updated from the watchdog.
3520  **/
3521 
3522 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3523 {
3524 	/* only return the current stats */
3525 	return &netdev->stats;
3526 }
3527 
3528 /**
3529  * e1000_change_mtu - Change the Maximum Transfer Unit
3530  * @netdev: network interface device structure
3531  * @new_mtu: new value for maximum frame size
3532  *
3533  * Returns 0 on success, negative on failure
3534  **/
3535 
3536 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3537 {
3538 	struct e1000_adapter *adapter = netdev_priv(netdev);
3539 	struct e1000_hw *hw = &adapter->hw;
3540 	int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3541 
3542 	if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3543 	    (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3544 		e_err(probe, "Invalid MTU setting\n");
3545 		return -EINVAL;
3546 	}
3547 
3548 	/* Adapter-specific max frame size limits. */
3549 	switch (hw->mac_type) {
3550 	case e1000_undefined ... e1000_82542_rev2_1:
3551 		if (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)) {
3552 			e_err(probe, "Jumbo Frames not supported.\n");
3553 			return -EINVAL;
3554 		}
3555 		break;
3556 	default:
3557 		/* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3558 		break;
3559 	}
3560 
3561 	while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
3562 		msleep(1);
3563 	/* e1000_down has a dependency on max_frame_size */
3564 	hw->max_frame_size = max_frame;
3565 	if (netif_running(netdev))
3566 		e1000_down(adapter);
3567 
3568 	/* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3569 	 * means we reserve 2 more, this pushes us to allocate from the next
3570 	 * larger slab size.
3571 	 * i.e. RXBUFFER_2048 --> size-4096 slab
3572 	 *  however with the new *_jumbo_rx* routines, jumbo receives will use
3573 	 *  fragmented skbs */
3574 
3575 	if (max_frame <= E1000_RXBUFFER_2048)
3576 		adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3577 	else
3578 #if (PAGE_SIZE >= E1000_RXBUFFER_16384)
3579 		adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3580 #elif (PAGE_SIZE >= E1000_RXBUFFER_4096)
3581 		adapter->rx_buffer_len = PAGE_SIZE;
3582 #endif
3583 
3584 	/* adjust allocation if LPE protects us, and we aren't using SBP */
3585 	if (!hw->tbi_compatibility_on &&
3586 	    ((max_frame == (ETH_FRAME_LEN + ETH_FCS_LEN)) ||
3587 	     (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3588 		adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3589 
3590 	pr_info("%s changing MTU from %d to %d\n",
3591 		netdev->name, netdev->mtu, new_mtu);
3592 	netdev->mtu = new_mtu;
3593 
3594 	if (netif_running(netdev))
3595 		e1000_up(adapter);
3596 	else
3597 		e1000_reset(adapter);
3598 
3599 	clear_bit(__E1000_RESETTING, &adapter->flags);
3600 
3601 	return 0;
3602 }
3603 
3604 /**
3605  * e1000_update_stats - Update the board statistics counters
3606  * @adapter: board private structure
3607  **/
3608 
3609 void e1000_update_stats(struct e1000_adapter *adapter)
3610 {
3611 	struct net_device *netdev = adapter->netdev;
3612 	struct e1000_hw *hw = &adapter->hw;
3613 	struct pci_dev *pdev = adapter->pdev;
3614 	unsigned long flags;
3615 	u16 phy_tmp;
3616 
3617 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3618 
3619 	/*
3620 	 * Prevent stats update while adapter is being reset, or if the pci
3621 	 * connection is down.
3622 	 */
3623 	if (adapter->link_speed == 0)
3624 		return;
3625 	if (pci_channel_offline(pdev))
3626 		return;
3627 
3628 	spin_lock_irqsave(&adapter->stats_lock, flags);
3629 
3630 	/* these counters are modified from e1000_tbi_adjust_stats,
3631 	 * called from the interrupt context, so they must only
3632 	 * be written while holding adapter->stats_lock
3633 	 */
3634 
3635 	adapter->stats.crcerrs += er32(CRCERRS);
3636 	adapter->stats.gprc += er32(GPRC);
3637 	adapter->stats.gorcl += er32(GORCL);
3638 	adapter->stats.gorch += er32(GORCH);
3639 	adapter->stats.bprc += er32(BPRC);
3640 	adapter->stats.mprc += er32(MPRC);
3641 	adapter->stats.roc += er32(ROC);
3642 
3643 	adapter->stats.prc64 += er32(PRC64);
3644 	adapter->stats.prc127 += er32(PRC127);
3645 	adapter->stats.prc255 += er32(PRC255);
3646 	adapter->stats.prc511 += er32(PRC511);
3647 	adapter->stats.prc1023 += er32(PRC1023);
3648 	adapter->stats.prc1522 += er32(PRC1522);
3649 
3650 	adapter->stats.symerrs += er32(SYMERRS);
3651 	adapter->stats.mpc += er32(MPC);
3652 	adapter->stats.scc += er32(SCC);
3653 	adapter->stats.ecol += er32(ECOL);
3654 	adapter->stats.mcc += er32(MCC);
3655 	adapter->stats.latecol += er32(LATECOL);
3656 	adapter->stats.dc += er32(DC);
3657 	adapter->stats.sec += er32(SEC);
3658 	adapter->stats.rlec += er32(RLEC);
3659 	adapter->stats.xonrxc += er32(XONRXC);
3660 	adapter->stats.xontxc += er32(XONTXC);
3661 	adapter->stats.xoffrxc += er32(XOFFRXC);
3662 	adapter->stats.xofftxc += er32(XOFFTXC);
3663 	adapter->stats.fcruc += er32(FCRUC);
3664 	adapter->stats.gptc += er32(GPTC);
3665 	adapter->stats.gotcl += er32(GOTCL);
3666 	adapter->stats.gotch += er32(GOTCH);
3667 	adapter->stats.rnbc += er32(RNBC);
3668 	adapter->stats.ruc += er32(RUC);
3669 	adapter->stats.rfc += er32(RFC);
3670 	adapter->stats.rjc += er32(RJC);
3671 	adapter->stats.torl += er32(TORL);
3672 	adapter->stats.torh += er32(TORH);
3673 	adapter->stats.totl += er32(TOTL);
3674 	adapter->stats.toth += er32(TOTH);
3675 	adapter->stats.tpr += er32(TPR);
3676 
3677 	adapter->stats.ptc64 += er32(PTC64);
3678 	adapter->stats.ptc127 += er32(PTC127);
3679 	adapter->stats.ptc255 += er32(PTC255);
3680 	adapter->stats.ptc511 += er32(PTC511);
3681 	adapter->stats.ptc1023 += er32(PTC1023);
3682 	adapter->stats.ptc1522 += er32(PTC1522);
3683 
3684 	adapter->stats.mptc += er32(MPTC);
3685 	adapter->stats.bptc += er32(BPTC);
3686 
3687 	/* used for adaptive IFS */
3688 
3689 	hw->tx_packet_delta = er32(TPT);
3690 	adapter->stats.tpt += hw->tx_packet_delta;
3691 	hw->collision_delta = er32(COLC);
3692 	adapter->stats.colc += hw->collision_delta;
3693 
3694 	if (hw->mac_type >= e1000_82543) {
3695 		adapter->stats.algnerrc += er32(ALGNERRC);
3696 		adapter->stats.rxerrc += er32(RXERRC);
3697 		adapter->stats.tncrs += er32(TNCRS);
3698 		adapter->stats.cexterr += er32(CEXTERR);
3699 		adapter->stats.tsctc += er32(TSCTC);
3700 		adapter->stats.tsctfc += er32(TSCTFC);
3701 	}
3702 
3703 	/* Fill out the OS statistics structure */
3704 	netdev->stats.multicast = adapter->stats.mprc;
3705 	netdev->stats.collisions = adapter->stats.colc;
3706 
3707 	/* Rx Errors */
3708 
3709 	/* RLEC on some newer hardware can be incorrect so build
3710 	* our own version based on RUC and ROC */
3711 	netdev->stats.rx_errors = adapter->stats.rxerrc +
3712 		adapter->stats.crcerrs + adapter->stats.algnerrc +
3713 		adapter->stats.ruc + adapter->stats.roc +
3714 		adapter->stats.cexterr;
3715 	adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3716 	netdev->stats.rx_length_errors = adapter->stats.rlerrc;
3717 	netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3718 	netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3719 	netdev->stats.rx_missed_errors = adapter->stats.mpc;
3720 
3721 	/* Tx Errors */
3722 	adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3723 	netdev->stats.tx_errors = adapter->stats.txerrc;
3724 	netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3725 	netdev->stats.tx_window_errors = adapter->stats.latecol;
3726 	netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3727 	if (hw->bad_tx_carr_stats_fd &&
3728 	    adapter->link_duplex == FULL_DUPLEX) {
3729 		netdev->stats.tx_carrier_errors = 0;
3730 		adapter->stats.tncrs = 0;
3731 	}
3732 
3733 	/* Tx Dropped needs to be maintained elsewhere */
3734 
3735 	/* Phy Stats */
3736 	if (hw->media_type == e1000_media_type_copper) {
3737 		if ((adapter->link_speed == SPEED_1000) &&
3738 		   (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3739 			phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3740 			adapter->phy_stats.idle_errors += phy_tmp;
3741 		}
3742 
3743 		if ((hw->mac_type <= e1000_82546) &&
3744 		   (hw->phy_type == e1000_phy_m88) &&
3745 		   !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3746 			adapter->phy_stats.receive_errors += phy_tmp;
3747 	}
3748 
3749 	/* Management Stats */
3750 	if (hw->has_smbus) {
3751 		adapter->stats.mgptc += er32(MGTPTC);
3752 		adapter->stats.mgprc += er32(MGTPRC);
3753 		adapter->stats.mgpdc += er32(MGTPDC);
3754 	}
3755 
3756 	spin_unlock_irqrestore(&adapter->stats_lock, flags);
3757 }
3758 
3759 /**
3760  * e1000_intr - Interrupt Handler
3761  * @irq: interrupt number
3762  * @data: pointer to a network interface device structure
3763  **/
3764 
3765 static irqreturn_t e1000_intr(int irq, void *data)
3766 {
3767 	struct net_device *netdev = data;
3768 	struct e1000_adapter *adapter = netdev_priv(netdev);
3769 	struct e1000_hw *hw = &adapter->hw;
3770 	u32 icr = er32(ICR);
3771 
3772 	if (unlikely((!icr)))
3773 		return IRQ_NONE;  /* Not our interrupt */
3774 
3775 	/*
3776 	 * we might have caused the interrupt, but the above
3777 	 * read cleared it, and just in case the driver is
3778 	 * down there is nothing to do so return handled
3779 	 */
3780 	if (unlikely(test_bit(__E1000_DOWN, &adapter->flags)))
3781 		return IRQ_HANDLED;
3782 
3783 	if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3784 		hw->get_link_status = 1;
3785 		/* guard against interrupt when we're going down */
3786 		if (!test_bit(__E1000_DOWN, &adapter->flags))
3787 			schedule_delayed_work(&adapter->watchdog_task, 1);
3788 	}
3789 
3790 	/* disable interrupts, without the synchronize_irq bit */
3791 	ew32(IMC, ~0);
3792 	E1000_WRITE_FLUSH();
3793 
3794 	if (likely(napi_schedule_prep(&adapter->napi))) {
3795 		adapter->total_tx_bytes = 0;
3796 		adapter->total_tx_packets = 0;
3797 		adapter->total_rx_bytes = 0;
3798 		adapter->total_rx_packets = 0;
3799 		__napi_schedule(&adapter->napi);
3800 	} else {
3801 		/* this really should not happen! if it does it is basically a
3802 		 * bug, but not a hard error, so enable ints and continue */
3803 		if (!test_bit(__E1000_DOWN, &adapter->flags))
3804 			e1000_irq_enable(adapter);
3805 	}
3806 
3807 	return IRQ_HANDLED;
3808 }
3809 
3810 /**
3811  * e1000_clean - NAPI Rx polling callback
3812  * @adapter: board private structure
3813  **/
3814 static int e1000_clean(struct napi_struct *napi, int budget)
3815 {
3816 	struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
3817 	int tx_clean_complete = 0, work_done = 0;
3818 
3819 	tx_clean_complete = e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);
3820 
3821 	adapter->clean_rx(adapter, &adapter->rx_ring[0], &work_done, budget);
3822 
3823 	if (!tx_clean_complete)
3824 		work_done = budget;
3825 
3826 	/* If budget not fully consumed, exit the polling mode */
3827 	if (work_done < budget) {
3828 		if (likely(adapter->itr_setting & 3))
3829 			e1000_set_itr(adapter);
3830 		napi_complete(napi);
3831 		if (!test_bit(__E1000_DOWN, &adapter->flags))
3832 			e1000_irq_enable(adapter);
3833 	}
3834 
3835 	return work_done;
3836 }
3837 
3838 /**
3839  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3840  * @adapter: board private structure
3841  **/
3842 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
3843 			       struct e1000_tx_ring *tx_ring)
3844 {
3845 	struct e1000_hw *hw = &adapter->hw;
3846 	struct net_device *netdev = adapter->netdev;
3847 	struct e1000_tx_desc *tx_desc, *eop_desc;
3848 	struct e1000_buffer *buffer_info;
3849 	unsigned int i, eop;
3850 	unsigned int count = 0;
3851 	unsigned int total_tx_bytes=0, total_tx_packets=0;
3852 
3853 	i = tx_ring->next_to_clean;
3854 	eop = tx_ring->buffer_info[i].next_to_watch;
3855 	eop_desc = E1000_TX_DESC(*tx_ring, eop);
3856 
3857 	while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
3858 	       (count < tx_ring->count)) {
3859 		bool cleaned = false;
3860 		rmb();	/* read buffer_info after eop_desc */
3861 		for ( ; !cleaned; count++) {
3862 			tx_desc = E1000_TX_DESC(*tx_ring, i);
3863 			buffer_info = &tx_ring->buffer_info[i];
3864 			cleaned = (i == eop);
3865 
3866 			if (cleaned) {
3867 				total_tx_packets += buffer_info->segs;
3868 				total_tx_bytes += buffer_info->bytecount;
3869 			}
3870 			e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3871 			tx_desc->upper.data = 0;
3872 
3873 			if (unlikely(++i == tx_ring->count)) i = 0;
3874 		}
3875 
3876 		eop = tx_ring->buffer_info[i].next_to_watch;
3877 		eop_desc = E1000_TX_DESC(*tx_ring, eop);
3878 	}
3879 
3880 	tx_ring->next_to_clean = i;
3881 
3882 #define TX_WAKE_THRESHOLD 32
3883 	if (unlikely(count && netif_carrier_ok(netdev) &&
3884 		     E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
3885 		/* Make sure that anybody stopping the queue after this
3886 		 * sees the new next_to_clean.
3887 		 */
3888 		smp_mb();
3889 
3890 		if (netif_queue_stopped(netdev) &&
3891 		    !(test_bit(__E1000_DOWN, &adapter->flags))) {
3892 			netif_wake_queue(netdev);
3893 			++adapter->restart_queue;
3894 		}
3895 	}
3896 
3897 	if (adapter->detect_tx_hung) {
3898 		/* Detect a transmit hang in hardware, this serializes the
3899 		 * check with the clearing of time_stamp and movement of i */
3900 		adapter->detect_tx_hung = false;
3901 		if (tx_ring->buffer_info[eop].time_stamp &&
3902 		    time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
3903 		               (adapter->tx_timeout_factor * HZ)) &&
3904 		    !(er32(STATUS) & E1000_STATUS_TXOFF)) {
3905 
3906 			/* detected Tx unit hang */
3907 			e_err(drv, "Detected Tx Unit Hang\n"
3908 			      "  Tx Queue             <%lu>\n"
3909 			      "  TDH                  <%x>\n"
3910 			      "  TDT                  <%x>\n"
3911 			      "  next_to_use          <%x>\n"
3912 			      "  next_to_clean        <%x>\n"
3913 			      "buffer_info[next_to_clean]\n"
3914 			      "  time_stamp           <%lx>\n"
3915 			      "  next_to_watch        <%x>\n"
3916 			      "  jiffies              <%lx>\n"
3917 			      "  next_to_watch.status <%x>\n",
3918 				(unsigned long)((tx_ring - adapter->tx_ring) /
3919 					sizeof(struct e1000_tx_ring)),
3920 				readl(hw->hw_addr + tx_ring->tdh),
3921 				readl(hw->hw_addr + tx_ring->tdt),
3922 				tx_ring->next_to_use,
3923 				tx_ring->next_to_clean,
3924 				tx_ring->buffer_info[eop].time_stamp,
3925 				eop,
3926 				jiffies,
3927 				eop_desc->upper.fields.status);
3928 			e1000_dump(adapter);
3929 			netif_stop_queue(netdev);
3930 		}
3931 	}
3932 	adapter->total_tx_bytes += total_tx_bytes;
3933 	adapter->total_tx_packets += total_tx_packets;
3934 	netdev->stats.tx_bytes += total_tx_bytes;
3935 	netdev->stats.tx_packets += total_tx_packets;
3936 	return count < tx_ring->count;
3937 }
3938 
3939 /**
3940  * e1000_rx_checksum - Receive Checksum Offload for 82543
3941  * @adapter:     board private structure
3942  * @status_err:  receive descriptor status and error fields
3943  * @csum:        receive descriptor csum field
3944  * @sk_buff:     socket buffer with received data
3945  **/
3946 
3947 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
3948 			      u32 csum, struct sk_buff *skb)
3949 {
3950 	struct e1000_hw *hw = &adapter->hw;
3951 	u16 status = (u16)status_err;
3952 	u8 errors = (u8)(status_err >> 24);
3953 
3954 	skb_checksum_none_assert(skb);
3955 
3956 	/* 82543 or newer only */
3957 	if (unlikely(hw->mac_type < e1000_82543)) return;
3958 	/* Ignore Checksum bit is set */
3959 	if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
3960 	/* TCP/UDP checksum error bit is set */
3961 	if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
3962 		/* let the stack verify checksum errors */
3963 		adapter->hw_csum_err++;
3964 		return;
3965 	}
3966 	/* TCP/UDP Checksum has not been calculated */
3967 	if (!(status & E1000_RXD_STAT_TCPCS))
3968 		return;
3969 
3970 	/* It must be a TCP or UDP packet with a valid checksum */
3971 	if (likely(status & E1000_RXD_STAT_TCPCS)) {
3972 		/* TCP checksum is good */
3973 		skb->ip_summed = CHECKSUM_UNNECESSARY;
3974 	}
3975 	adapter->hw_csum_good++;
3976 }
3977 
3978 /**
3979  * e1000_consume_page - helper function
3980  **/
3981 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
3982                                u16 length)
3983 {
3984 	bi->page = NULL;
3985 	skb->len += length;
3986 	skb->data_len += length;
3987 	skb->truesize += PAGE_SIZE;
3988 }
3989 
3990 /**
3991  * e1000_receive_skb - helper function to handle rx indications
3992  * @adapter: board private structure
3993  * @status: descriptor status field as written by hardware
3994  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3995  * @skb: pointer to sk_buff to be indicated to stack
3996  */
3997 static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
3998 			      __le16 vlan, struct sk_buff *skb)
3999 {
4000 	skb->protocol = eth_type_trans(skb, adapter->netdev);
4001 
4002 	if (status & E1000_RXD_STAT_VP) {
4003 		u16 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
4004 
4005 		__vlan_hwaccel_put_tag(skb, vid);
4006 	}
4007 	napi_gro_receive(&adapter->napi, skb);
4008 }
4009 
4010 /**
4011  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
4012  * @adapter: board private structure
4013  * @rx_ring: ring to clean
4014  * @work_done: amount of napi work completed this call
4015  * @work_to_do: max amount of work allowed for this call to do
4016  *
4017  * the return value indicates whether actual cleaning was done, there
4018  * is no guarantee that everything was cleaned
4019  */
4020 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
4021 				     struct e1000_rx_ring *rx_ring,
4022 				     int *work_done, int work_to_do)
4023 {
4024 	struct e1000_hw *hw = &adapter->hw;
4025 	struct net_device *netdev = adapter->netdev;
4026 	struct pci_dev *pdev = adapter->pdev;
4027 	struct e1000_rx_desc *rx_desc, *next_rxd;
4028 	struct e1000_buffer *buffer_info, *next_buffer;
4029 	unsigned long irq_flags;
4030 	u32 length;
4031 	unsigned int i;
4032 	int cleaned_count = 0;
4033 	bool cleaned = false;
4034 	unsigned int total_rx_bytes=0, total_rx_packets=0;
4035 
4036 	i = rx_ring->next_to_clean;
4037 	rx_desc = E1000_RX_DESC(*rx_ring, i);
4038 	buffer_info = &rx_ring->buffer_info[i];
4039 
4040 	while (rx_desc->status & E1000_RXD_STAT_DD) {
4041 		struct sk_buff *skb;
4042 		u8 status;
4043 
4044 		if (*work_done >= work_to_do)
4045 			break;
4046 		(*work_done)++;
4047 		rmb(); /* read descriptor and rx_buffer_info after status DD */
4048 
4049 		status = rx_desc->status;
4050 		skb = buffer_info->skb;
4051 		buffer_info->skb = NULL;
4052 
4053 		if (++i == rx_ring->count) i = 0;
4054 		next_rxd = E1000_RX_DESC(*rx_ring, i);
4055 		prefetch(next_rxd);
4056 
4057 		next_buffer = &rx_ring->buffer_info[i];
4058 
4059 		cleaned = true;
4060 		cleaned_count++;
4061 		dma_unmap_page(&pdev->dev, buffer_info->dma,
4062 			       buffer_info->length, DMA_FROM_DEVICE);
4063 		buffer_info->dma = 0;
4064 
4065 		length = le16_to_cpu(rx_desc->length);
4066 
4067 		/* errors is only valid for DD + EOP descriptors */
4068 		if (unlikely((status & E1000_RXD_STAT_EOP) &&
4069 		    (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
4070 			u8 *mapped;
4071 			u8 last_byte;
4072 
4073 			mapped = page_address(buffer_info->page);
4074 			last_byte = *(mapped + length - 1);
4075 			if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
4076 				       last_byte)) {
4077 				spin_lock_irqsave(&adapter->stats_lock,
4078 				                  irq_flags);
4079 				e1000_tbi_adjust_stats(hw, &adapter->stats,
4080 						       length, mapped);
4081 				spin_unlock_irqrestore(&adapter->stats_lock,
4082 				                       irq_flags);
4083 				length--;
4084 			} else {
4085 				if (netdev->features & NETIF_F_RXALL)
4086 					goto process_skb;
4087 				/* recycle both page and skb */
4088 				buffer_info->skb = skb;
4089 				/* an error means any chain goes out the window
4090 				 * too */
4091 				if (rx_ring->rx_skb_top)
4092 					dev_kfree_skb(rx_ring->rx_skb_top);
4093 				rx_ring->rx_skb_top = NULL;
4094 				goto next_desc;
4095 			}
4096 		}
4097 
4098 #define rxtop rx_ring->rx_skb_top
4099 process_skb:
4100 		if (!(status & E1000_RXD_STAT_EOP)) {
4101 			/* this descriptor is only the beginning (or middle) */
4102 			if (!rxtop) {
4103 				/* this is the beginning of a chain */
4104 				rxtop = skb;
4105 				skb_fill_page_desc(rxtop, 0, buffer_info->page,
4106 				                   0, length);
4107 			} else {
4108 				/* this is the middle of a chain */
4109 				skb_fill_page_desc(rxtop,
4110 				    skb_shinfo(rxtop)->nr_frags,
4111 				    buffer_info->page, 0, length);
4112 				/* re-use the skb, only consumed the page */
4113 				buffer_info->skb = skb;
4114 			}
4115 			e1000_consume_page(buffer_info, rxtop, length);
4116 			goto next_desc;
4117 		} else {
4118 			if (rxtop) {
4119 				/* end of the chain */
4120 				skb_fill_page_desc(rxtop,
4121 				    skb_shinfo(rxtop)->nr_frags,
4122 				    buffer_info->page, 0, length);
4123 				/* re-use the current skb, we only consumed the
4124 				 * page */
4125 				buffer_info->skb = skb;
4126 				skb = rxtop;
4127 				rxtop = NULL;
4128 				e1000_consume_page(buffer_info, skb, length);
4129 			} else {
4130 				/* no chain, got EOP, this buf is the packet
4131 				 * copybreak to save the put_page/alloc_page */
4132 				if (length <= copybreak &&
4133 				    skb_tailroom(skb) >= length) {
4134 					u8 *vaddr;
4135 					vaddr = kmap_atomic(buffer_info->page);
4136 					memcpy(skb_tail_pointer(skb), vaddr, length);
4137 					kunmap_atomic(vaddr);
4138 					/* re-use the page, so don't erase
4139 					 * buffer_info->page */
4140 					skb_put(skb, length);
4141 				} else {
4142 					skb_fill_page_desc(skb, 0,
4143 					                   buffer_info->page, 0,
4144 				                           length);
4145 					e1000_consume_page(buffer_info, skb,
4146 					                   length);
4147 				}
4148 			}
4149 		}
4150 
4151 		/* Receive Checksum Offload XXX recompute due to CRC strip? */
4152 		e1000_rx_checksum(adapter,
4153 		                  (u32)(status) |
4154 		                  ((u32)(rx_desc->errors) << 24),
4155 		                  le16_to_cpu(rx_desc->csum), skb);
4156 
4157 		total_rx_bytes += (skb->len - 4); /* don't count FCS */
4158 		if (likely(!(netdev->features & NETIF_F_RXFCS)))
4159 			pskb_trim(skb, skb->len - 4);
4160 		total_rx_packets++;
4161 
4162 		/* eth type trans needs skb->data to point to something */
4163 		if (!pskb_may_pull(skb, ETH_HLEN)) {
4164 			e_err(drv, "pskb_may_pull failed.\n");
4165 			dev_kfree_skb(skb);
4166 			goto next_desc;
4167 		}
4168 
4169 		e1000_receive_skb(adapter, status, rx_desc->special, skb);
4170 
4171 next_desc:
4172 		rx_desc->status = 0;
4173 
4174 		/* return some buffers to hardware, one at a time is too slow */
4175 		if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4176 			adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4177 			cleaned_count = 0;
4178 		}
4179 
4180 		/* use prefetched values */
4181 		rx_desc = next_rxd;
4182 		buffer_info = next_buffer;
4183 	}
4184 	rx_ring->next_to_clean = i;
4185 
4186 	cleaned_count = E1000_DESC_UNUSED(rx_ring);
4187 	if (cleaned_count)
4188 		adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4189 
4190 	adapter->total_rx_packets += total_rx_packets;
4191 	adapter->total_rx_bytes += total_rx_bytes;
4192 	netdev->stats.rx_bytes += total_rx_bytes;
4193 	netdev->stats.rx_packets += total_rx_packets;
4194 	return cleaned;
4195 }
4196 
4197 /*
4198  * this should improve performance for small packets with large amounts
4199  * of reassembly being done in the stack
4200  */
4201 static void e1000_check_copybreak(struct net_device *netdev,
4202 				 struct e1000_buffer *buffer_info,
4203 				 u32 length, struct sk_buff **skb)
4204 {
4205 	struct sk_buff *new_skb;
4206 
4207 	if (length > copybreak)
4208 		return;
4209 
4210 	new_skb = netdev_alloc_skb_ip_align(netdev, length);
4211 	if (!new_skb)
4212 		return;
4213 
4214 	skb_copy_to_linear_data_offset(new_skb, -NET_IP_ALIGN,
4215 				       (*skb)->data - NET_IP_ALIGN,
4216 				       length + NET_IP_ALIGN);
4217 	/* save the skb in buffer_info as good */
4218 	buffer_info->skb = *skb;
4219 	*skb = new_skb;
4220 }
4221 
4222 /**
4223  * e1000_clean_rx_irq - Send received data up the network stack; legacy
4224  * @adapter: board private structure
4225  * @rx_ring: ring to clean
4226  * @work_done: amount of napi work completed this call
4227  * @work_to_do: max amount of work allowed for this call to do
4228  */
4229 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
4230 			       struct e1000_rx_ring *rx_ring,
4231 			       int *work_done, int work_to_do)
4232 {
4233 	struct e1000_hw *hw = &adapter->hw;
4234 	struct net_device *netdev = adapter->netdev;
4235 	struct pci_dev *pdev = adapter->pdev;
4236 	struct e1000_rx_desc *rx_desc, *next_rxd;
4237 	struct e1000_buffer *buffer_info, *next_buffer;
4238 	unsigned long flags;
4239 	u32 length;
4240 	unsigned int i;
4241 	int cleaned_count = 0;
4242 	bool cleaned = false;
4243 	unsigned int total_rx_bytes=0, total_rx_packets=0;
4244 
4245 	i = rx_ring->next_to_clean;
4246 	rx_desc = E1000_RX_DESC(*rx_ring, i);
4247 	buffer_info = &rx_ring->buffer_info[i];
4248 
4249 	while (rx_desc->status & E1000_RXD_STAT_DD) {
4250 		struct sk_buff *skb;
4251 		u8 status;
4252 
4253 		if (*work_done >= work_to_do)
4254 			break;
4255 		(*work_done)++;
4256 		rmb(); /* read descriptor and rx_buffer_info after status DD */
4257 
4258 		status = rx_desc->status;
4259 		skb = buffer_info->skb;
4260 		buffer_info->skb = NULL;
4261 
4262 		prefetch(skb->data - NET_IP_ALIGN);
4263 
4264 		if (++i == rx_ring->count) i = 0;
4265 		next_rxd = E1000_RX_DESC(*rx_ring, i);
4266 		prefetch(next_rxd);
4267 
4268 		next_buffer = &rx_ring->buffer_info[i];
4269 
4270 		cleaned = true;
4271 		cleaned_count++;
4272 		dma_unmap_single(&pdev->dev, buffer_info->dma,
4273 				 buffer_info->length, DMA_FROM_DEVICE);
4274 		buffer_info->dma = 0;
4275 
4276 		length = le16_to_cpu(rx_desc->length);
4277 		/* !EOP means multiple descriptors were used to store a single
4278 		 * packet, if thats the case we need to toss it.  In fact, we
4279 		 * to toss every packet with the EOP bit clear and the next
4280 		 * frame that _does_ have the EOP bit set, as it is by
4281 		 * definition only a frame fragment
4282 		 */
4283 		if (unlikely(!(status & E1000_RXD_STAT_EOP)))
4284 			adapter->discarding = true;
4285 
4286 		if (adapter->discarding) {
4287 			/* All receives must fit into a single buffer */
4288 			e_dbg("Receive packet consumed multiple buffers\n");
4289 			/* recycle */
4290 			buffer_info->skb = skb;
4291 			if (status & E1000_RXD_STAT_EOP)
4292 				adapter->discarding = false;
4293 			goto next_desc;
4294 		}
4295 
4296 		if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
4297 			u8 last_byte = *(skb->data + length - 1);
4298 			if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
4299 				       last_byte)) {
4300 				spin_lock_irqsave(&adapter->stats_lock, flags);
4301 				e1000_tbi_adjust_stats(hw, &adapter->stats,
4302 				                       length, skb->data);
4303 				spin_unlock_irqrestore(&adapter->stats_lock,
4304 				                       flags);
4305 				length--;
4306 			} else {
4307 				if (netdev->features & NETIF_F_RXALL)
4308 					goto process_skb;
4309 				/* recycle */
4310 				buffer_info->skb = skb;
4311 				goto next_desc;
4312 			}
4313 		}
4314 
4315 process_skb:
4316 		total_rx_bytes += (length - 4); /* don't count FCS */
4317 		total_rx_packets++;
4318 
4319 		if (likely(!(netdev->features & NETIF_F_RXFCS)))
4320 			/* adjust length to remove Ethernet CRC, this must be
4321 			 * done after the TBI_ACCEPT workaround above
4322 			 */
4323 			length -= 4;
4324 
4325 		e1000_check_copybreak(netdev, buffer_info, length, &skb);
4326 
4327 		skb_put(skb, length);
4328 
4329 		/* Receive Checksum Offload */
4330 		e1000_rx_checksum(adapter,
4331 				  (u32)(status) |
4332 				  ((u32)(rx_desc->errors) << 24),
4333 				  le16_to_cpu(rx_desc->csum), skb);
4334 
4335 		e1000_receive_skb(adapter, status, rx_desc->special, skb);
4336 
4337 next_desc:
4338 		rx_desc->status = 0;
4339 
4340 		/* return some buffers to hardware, one at a time is too slow */
4341 		if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4342 			adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4343 			cleaned_count = 0;
4344 		}
4345 
4346 		/* use prefetched values */
4347 		rx_desc = next_rxd;
4348 		buffer_info = next_buffer;
4349 	}
4350 	rx_ring->next_to_clean = i;
4351 
4352 	cleaned_count = E1000_DESC_UNUSED(rx_ring);
4353 	if (cleaned_count)
4354 		adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4355 
4356 	adapter->total_rx_packets += total_rx_packets;
4357 	adapter->total_rx_bytes += total_rx_bytes;
4358 	netdev->stats.rx_bytes += total_rx_bytes;
4359 	netdev->stats.rx_packets += total_rx_packets;
4360 	return cleaned;
4361 }
4362 
4363 /**
4364  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
4365  * @adapter: address of board private structure
4366  * @rx_ring: pointer to receive ring structure
4367  * @cleaned_count: number of buffers to allocate this pass
4368  **/
4369 
4370 static void
4371 e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
4372                              struct e1000_rx_ring *rx_ring, int cleaned_count)
4373 {
4374 	struct net_device *netdev = adapter->netdev;
4375 	struct pci_dev *pdev = adapter->pdev;
4376 	struct e1000_rx_desc *rx_desc;
4377 	struct e1000_buffer *buffer_info;
4378 	struct sk_buff *skb;
4379 	unsigned int i;
4380 	unsigned int bufsz = 256 - 16 /*for skb_reserve */ ;
4381 
4382 	i = rx_ring->next_to_use;
4383 	buffer_info = &rx_ring->buffer_info[i];
4384 
4385 	while (cleaned_count--) {
4386 		skb = buffer_info->skb;
4387 		if (skb) {
4388 			skb_trim(skb, 0);
4389 			goto check_page;
4390 		}
4391 
4392 		skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4393 		if (unlikely(!skb)) {
4394 			/* Better luck next round */
4395 			adapter->alloc_rx_buff_failed++;
4396 			break;
4397 		}
4398 
4399 		buffer_info->skb = skb;
4400 		buffer_info->length = adapter->rx_buffer_len;
4401 check_page:
4402 		/* allocate a new page if necessary */
4403 		if (!buffer_info->page) {
4404 			buffer_info->page = alloc_page(GFP_ATOMIC);
4405 			if (unlikely(!buffer_info->page)) {
4406 				adapter->alloc_rx_buff_failed++;
4407 				break;
4408 			}
4409 		}
4410 
4411 		if (!buffer_info->dma) {
4412 			buffer_info->dma = dma_map_page(&pdev->dev,
4413 			                                buffer_info->page, 0,
4414 							buffer_info->length,
4415 							DMA_FROM_DEVICE);
4416 			if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4417 				put_page(buffer_info->page);
4418 				dev_kfree_skb(skb);
4419 				buffer_info->page = NULL;
4420 				buffer_info->skb = NULL;
4421 				buffer_info->dma = 0;
4422 				adapter->alloc_rx_buff_failed++;
4423 				break; /* while !buffer_info->skb */
4424 			}
4425 		}
4426 
4427 		rx_desc = E1000_RX_DESC(*rx_ring, i);
4428 		rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4429 
4430 		if (unlikely(++i == rx_ring->count))
4431 			i = 0;
4432 		buffer_info = &rx_ring->buffer_info[i];
4433 	}
4434 
4435 	if (likely(rx_ring->next_to_use != i)) {
4436 		rx_ring->next_to_use = i;
4437 		if (unlikely(i-- == 0))
4438 			i = (rx_ring->count - 1);
4439 
4440 		/* Force memory writes to complete before letting h/w
4441 		 * know there are new descriptors to fetch.  (Only
4442 		 * applicable for weak-ordered memory model archs,
4443 		 * such as IA-64). */
4444 		wmb();
4445 		writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4446 	}
4447 }
4448 
4449 /**
4450  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4451  * @adapter: address of board private structure
4452  **/
4453 
4454 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4455 				   struct e1000_rx_ring *rx_ring,
4456 				   int cleaned_count)
4457 {
4458 	struct e1000_hw *hw = &adapter->hw;
4459 	struct net_device *netdev = adapter->netdev;
4460 	struct pci_dev *pdev = adapter->pdev;
4461 	struct e1000_rx_desc *rx_desc;
4462 	struct e1000_buffer *buffer_info;
4463 	struct sk_buff *skb;
4464 	unsigned int i;
4465 	unsigned int bufsz = adapter->rx_buffer_len;
4466 
4467 	i = rx_ring->next_to_use;
4468 	buffer_info = &rx_ring->buffer_info[i];
4469 
4470 	while (cleaned_count--) {
4471 		skb = buffer_info->skb;
4472 		if (skb) {
4473 			skb_trim(skb, 0);
4474 			goto map_skb;
4475 		}
4476 
4477 		skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4478 		if (unlikely(!skb)) {
4479 			/* Better luck next round */
4480 			adapter->alloc_rx_buff_failed++;
4481 			break;
4482 		}
4483 
4484 		/* Fix for errata 23, can't cross 64kB boundary */
4485 		if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4486 			struct sk_buff *oldskb = skb;
4487 			e_err(rx_err, "skb align check failed: %u bytes at "
4488 			      "%p\n", bufsz, skb->data);
4489 			/* Try again, without freeing the previous */
4490 			skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4491 			/* Failed allocation, critical failure */
4492 			if (!skb) {
4493 				dev_kfree_skb(oldskb);
4494 				adapter->alloc_rx_buff_failed++;
4495 				break;
4496 			}
4497 
4498 			if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4499 				/* give up */
4500 				dev_kfree_skb(skb);
4501 				dev_kfree_skb(oldskb);
4502 				adapter->alloc_rx_buff_failed++;
4503 				break; /* while !buffer_info->skb */
4504 			}
4505 
4506 			/* Use new allocation */
4507 			dev_kfree_skb(oldskb);
4508 		}
4509 		buffer_info->skb = skb;
4510 		buffer_info->length = adapter->rx_buffer_len;
4511 map_skb:
4512 		buffer_info->dma = dma_map_single(&pdev->dev,
4513 						  skb->data,
4514 						  buffer_info->length,
4515 						  DMA_FROM_DEVICE);
4516 		if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4517 			dev_kfree_skb(skb);
4518 			buffer_info->skb = NULL;
4519 			buffer_info->dma = 0;
4520 			adapter->alloc_rx_buff_failed++;
4521 			break; /* while !buffer_info->skb */
4522 		}
4523 
4524 		/*
4525 		 * XXX if it was allocated cleanly it will never map to a
4526 		 * boundary crossing
4527 		 */
4528 
4529 		/* Fix for errata 23, can't cross 64kB boundary */
4530 		if (!e1000_check_64k_bound(adapter,
4531 					(void *)(unsigned long)buffer_info->dma,
4532 					adapter->rx_buffer_len)) {
4533 			e_err(rx_err, "dma align check failed: %u bytes at "
4534 			      "%p\n", adapter->rx_buffer_len,
4535 			      (void *)(unsigned long)buffer_info->dma);
4536 			dev_kfree_skb(skb);
4537 			buffer_info->skb = NULL;
4538 
4539 			dma_unmap_single(&pdev->dev, buffer_info->dma,
4540 					 adapter->rx_buffer_len,
4541 					 DMA_FROM_DEVICE);
4542 			buffer_info->dma = 0;
4543 
4544 			adapter->alloc_rx_buff_failed++;
4545 			break; /* while !buffer_info->skb */
4546 		}
4547 		rx_desc = E1000_RX_DESC(*rx_ring, i);
4548 		rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4549 
4550 		if (unlikely(++i == rx_ring->count))
4551 			i = 0;
4552 		buffer_info = &rx_ring->buffer_info[i];
4553 	}
4554 
4555 	if (likely(rx_ring->next_to_use != i)) {
4556 		rx_ring->next_to_use = i;
4557 		if (unlikely(i-- == 0))
4558 			i = (rx_ring->count - 1);
4559 
4560 		/* Force memory writes to complete before letting h/w
4561 		 * know there are new descriptors to fetch.  (Only
4562 		 * applicable for weak-ordered memory model archs,
4563 		 * such as IA-64). */
4564 		wmb();
4565 		writel(i, hw->hw_addr + rx_ring->rdt);
4566 	}
4567 }
4568 
4569 /**
4570  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4571  * @adapter:
4572  **/
4573 
4574 static void e1000_smartspeed(struct e1000_adapter *adapter)
4575 {
4576 	struct e1000_hw *hw = &adapter->hw;
4577 	u16 phy_status;
4578 	u16 phy_ctrl;
4579 
4580 	if ((hw->phy_type != e1000_phy_igp) || !hw->autoneg ||
4581 	   !(hw->autoneg_advertised & ADVERTISE_1000_FULL))
4582 		return;
4583 
4584 	if (adapter->smartspeed == 0) {
4585 		/* If Master/Slave config fault is asserted twice,
4586 		 * we assume back-to-back */
4587 		e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4588 		if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4589 		e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4590 		if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4591 		e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4592 		if (phy_ctrl & CR_1000T_MS_ENABLE) {
4593 			phy_ctrl &= ~CR_1000T_MS_ENABLE;
4594 			e1000_write_phy_reg(hw, PHY_1000T_CTRL,
4595 					    phy_ctrl);
4596 			adapter->smartspeed++;
4597 			if (!e1000_phy_setup_autoneg(hw) &&
4598 			   !e1000_read_phy_reg(hw, PHY_CTRL,
4599 				   	       &phy_ctrl)) {
4600 				phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4601 					     MII_CR_RESTART_AUTO_NEG);
4602 				e1000_write_phy_reg(hw, PHY_CTRL,
4603 						    phy_ctrl);
4604 			}
4605 		}
4606 		return;
4607 	} else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4608 		/* If still no link, perhaps using 2/3 pair cable */
4609 		e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4610 		phy_ctrl |= CR_1000T_MS_ENABLE;
4611 		e1000_write_phy_reg(hw, PHY_1000T_CTRL, phy_ctrl);
4612 		if (!e1000_phy_setup_autoneg(hw) &&
4613 		   !e1000_read_phy_reg(hw, PHY_CTRL, &phy_ctrl)) {
4614 			phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4615 				     MII_CR_RESTART_AUTO_NEG);
4616 			e1000_write_phy_reg(hw, PHY_CTRL, phy_ctrl);
4617 		}
4618 	}
4619 	/* Restart process after E1000_SMARTSPEED_MAX iterations */
4620 	if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4621 		adapter->smartspeed = 0;
4622 }
4623 
4624 /**
4625  * e1000_ioctl -
4626  * @netdev:
4627  * @ifreq:
4628  * @cmd:
4629  **/
4630 
4631 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4632 {
4633 	switch (cmd) {
4634 	case SIOCGMIIPHY:
4635 	case SIOCGMIIREG:
4636 	case SIOCSMIIREG:
4637 		return e1000_mii_ioctl(netdev, ifr, cmd);
4638 	default:
4639 		return -EOPNOTSUPP;
4640 	}
4641 }
4642 
4643 /**
4644  * e1000_mii_ioctl -
4645  * @netdev:
4646  * @ifreq:
4647  * @cmd:
4648  **/
4649 
4650 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4651 			   int cmd)
4652 {
4653 	struct e1000_adapter *adapter = netdev_priv(netdev);
4654 	struct e1000_hw *hw = &adapter->hw;
4655 	struct mii_ioctl_data *data = if_mii(ifr);
4656 	int retval;
4657 	u16 mii_reg;
4658 	unsigned long flags;
4659 
4660 	if (hw->media_type != e1000_media_type_copper)
4661 		return -EOPNOTSUPP;
4662 
4663 	switch (cmd) {
4664 	case SIOCGMIIPHY:
4665 		data->phy_id = hw->phy_addr;
4666 		break;
4667 	case SIOCGMIIREG:
4668 		spin_lock_irqsave(&adapter->stats_lock, flags);
4669 		if (e1000_read_phy_reg(hw, data->reg_num & 0x1F,
4670 				   &data->val_out)) {
4671 			spin_unlock_irqrestore(&adapter->stats_lock, flags);
4672 			return -EIO;
4673 		}
4674 		spin_unlock_irqrestore(&adapter->stats_lock, flags);
4675 		break;
4676 	case SIOCSMIIREG:
4677 		if (data->reg_num & ~(0x1F))
4678 			return -EFAULT;
4679 		mii_reg = data->val_in;
4680 		spin_lock_irqsave(&adapter->stats_lock, flags);
4681 		if (e1000_write_phy_reg(hw, data->reg_num,
4682 					mii_reg)) {
4683 			spin_unlock_irqrestore(&adapter->stats_lock, flags);
4684 			return -EIO;
4685 		}
4686 		spin_unlock_irqrestore(&adapter->stats_lock, flags);
4687 		if (hw->media_type == e1000_media_type_copper) {
4688 			switch (data->reg_num) {
4689 			case PHY_CTRL:
4690 				if (mii_reg & MII_CR_POWER_DOWN)
4691 					break;
4692 				if (mii_reg & MII_CR_AUTO_NEG_EN) {
4693 					hw->autoneg = 1;
4694 					hw->autoneg_advertised = 0x2F;
4695 				} else {
4696 					u32 speed;
4697 					if (mii_reg & 0x40)
4698 						speed = SPEED_1000;
4699 					else if (mii_reg & 0x2000)
4700 						speed = SPEED_100;
4701 					else
4702 						speed = SPEED_10;
4703 					retval = e1000_set_spd_dplx(
4704 						adapter, speed,
4705 						((mii_reg & 0x100)
4706 						 ? DUPLEX_FULL :
4707 						 DUPLEX_HALF));
4708 					if (retval)
4709 						return retval;
4710 				}
4711 				if (netif_running(adapter->netdev))
4712 					e1000_reinit_locked(adapter);
4713 				else
4714 					e1000_reset(adapter);
4715 				break;
4716 			case M88E1000_PHY_SPEC_CTRL:
4717 			case M88E1000_EXT_PHY_SPEC_CTRL:
4718 				if (e1000_phy_reset(hw))
4719 					return -EIO;
4720 				break;
4721 			}
4722 		} else {
4723 			switch (data->reg_num) {
4724 			case PHY_CTRL:
4725 				if (mii_reg & MII_CR_POWER_DOWN)
4726 					break;
4727 				if (netif_running(adapter->netdev))
4728 					e1000_reinit_locked(adapter);
4729 				else
4730 					e1000_reset(adapter);
4731 				break;
4732 			}
4733 		}
4734 		break;
4735 	default:
4736 		return -EOPNOTSUPP;
4737 	}
4738 	return E1000_SUCCESS;
4739 }
4740 
4741 void e1000_pci_set_mwi(struct e1000_hw *hw)
4742 {
4743 	struct e1000_adapter *adapter = hw->back;
4744 	int ret_val = pci_set_mwi(adapter->pdev);
4745 
4746 	if (ret_val)
4747 		e_err(probe, "Error in setting MWI\n");
4748 }
4749 
4750 void e1000_pci_clear_mwi(struct e1000_hw *hw)
4751 {
4752 	struct e1000_adapter *adapter = hw->back;
4753 
4754 	pci_clear_mwi(adapter->pdev);
4755 }
4756 
4757 int e1000_pcix_get_mmrbc(struct e1000_hw *hw)
4758 {
4759 	struct e1000_adapter *adapter = hw->back;
4760 	return pcix_get_mmrbc(adapter->pdev);
4761 }
4762 
4763 void e1000_pcix_set_mmrbc(struct e1000_hw *hw, int mmrbc)
4764 {
4765 	struct e1000_adapter *adapter = hw->back;
4766 	pcix_set_mmrbc(adapter->pdev, mmrbc);
4767 }
4768 
4769 void e1000_io_write(struct e1000_hw *hw, unsigned long port, u32 value)
4770 {
4771 	outl(value, port);
4772 }
4773 
4774 static bool e1000_vlan_used(struct e1000_adapter *adapter)
4775 {
4776 	u16 vid;
4777 
4778 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
4779 		return true;
4780 	return false;
4781 }
4782 
4783 static void __e1000_vlan_mode(struct e1000_adapter *adapter,
4784 			      netdev_features_t features)
4785 {
4786 	struct e1000_hw *hw = &adapter->hw;
4787 	u32 ctrl;
4788 
4789 	ctrl = er32(CTRL);
4790 	if (features & NETIF_F_HW_VLAN_RX) {
4791 		/* enable VLAN tag insert/strip */
4792 		ctrl |= E1000_CTRL_VME;
4793 	} else {
4794 		/* disable VLAN tag insert/strip */
4795 		ctrl &= ~E1000_CTRL_VME;
4796 	}
4797 	ew32(CTRL, ctrl);
4798 }
4799 static void e1000_vlan_filter_on_off(struct e1000_adapter *adapter,
4800 				     bool filter_on)
4801 {
4802 	struct e1000_hw *hw = &adapter->hw;
4803 	u32 rctl;
4804 
4805 	if (!test_bit(__E1000_DOWN, &adapter->flags))
4806 		e1000_irq_disable(adapter);
4807 
4808 	__e1000_vlan_mode(adapter, adapter->netdev->features);
4809 	if (filter_on) {
4810 		/* enable VLAN receive filtering */
4811 		rctl = er32(RCTL);
4812 		rctl &= ~E1000_RCTL_CFIEN;
4813 		if (!(adapter->netdev->flags & IFF_PROMISC))
4814 			rctl |= E1000_RCTL_VFE;
4815 		ew32(RCTL, rctl);
4816 		e1000_update_mng_vlan(adapter);
4817 	} else {
4818 		/* disable VLAN receive filtering */
4819 		rctl = er32(RCTL);
4820 		rctl &= ~E1000_RCTL_VFE;
4821 		ew32(RCTL, rctl);
4822 	}
4823 
4824 	if (!test_bit(__E1000_DOWN, &adapter->flags))
4825 		e1000_irq_enable(adapter);
4826 }
4827 
4828 static void e1000_vlan_mode(struct net_device *netdev,
4829 			    netdev_features_t features)
4830 {
4831 	struct e1000_adapter *adapter = netdev_priv(netdev);
4832 
4833 	if (!test_bit(__E1000_DOWN, &adapter->flags))
4834 		e1000_irq_disable(adapter);
4835 
4836 	__e1000_vlan_mode(adapter, features);
4837 
4838 	if (!test_bit(__E1000_DOWN, &adapter->flags))
4839 		e1000_irq_enable(adapter);
4840 }
4841 
4842 static int e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4843 {
4844 	struct e1000_adapter *adapter = netdev_priv(netdev);
4845 	struct e1000_hw *hw = &adapter->hw;
4846 	u32 vfta, index;
4847 
4848 	if ((hw->mng_cookie.status &
4849 	     E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4850 	    (vid == adapter->mng_vlan_id))
4851 		return 0;
4852 
4853 	if (!e1000_vlan_used(adapter))
4854 		e1000_vlan_filter_on_off(adapter, true);
4855 
4856 	/* add VID to filter table */
4857 	index = (vid >> 5) & 0x7F;
4858 	vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4859 	vfta |= (1 << (vid & 0x1F));
4860 	e1000_write_vfta(hw, index, vfta);
4861 
4862 	set_bit(vid, adapter->active_vlans);
4863 
4864 	return 0;
4865 }
4866 
4867 static int e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
4868 {
4869 	struct e1000_adapter *adapter = netdev_priv(netdev);
4870 	struct e1000_hw *hw = &adapter->hw;
4871 	u32 vfta, index;
4872 
4873 	if (!test_bit(__E1000_DOWN, &adapter->flags))
4874 		e1000_irq_disable(adapter);
4875 	if (!test_bit(__E1000_DOWN, &adapter->flags))
4876 		e1000_irq_enable(adapter);
4877 
4878 	/* remove VID from filter table */
4879 	index = (vid >> 5) & 0x7F;
4880 	vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4881 	vfta &= ~(1 << (vid & 0x1F));
4882 	e1000_write_vfta(hw, index, vfta);
4883 
4884 	clear_bit(vid, adapter->active_vlans);
4885 
4886 	if (!e1000_vlan_used(adapter))
4887 		e1000_vlan_filter_on_off(adapter, false);
4888 
4889 	return 0;
4890 }
4891 
4892 static void e1000_restore_vlan(struct e1000_adapter *adapter)
4893 {
4894 	u16 vid;
4895 
4896 	if (!e1000_vlan_used(adapter))
4897 		return;
4898 
4899 	e1000_vlan_filter_on_off(adapter, true);
4900 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
4901 		e1000_vlan_rx_add_vid(adapter->netdev, vid);
4902 }
4903 
4904 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u32 spd, u8 dplx)
4905 {
4906 	struct e1000_hw *hw = &adapter->hw;
4907 
4908 	hw->autoneg = 0;
4909 
4910 	/* Make sure dplx is at most 1 bit and lsb of speed is not set
4911 	 * for the switch() below to work */
4912 	if ((spd & 1) || (dplx & ~1))
4913 		goto err_inval;
4914 
4915 	/* Fiber NICs only allow 1000 gbps Full duplex */
4916 	if ((hw->media_type == e1000_media_type_fiber) &&
4917 	    spd != SPEED_1000 &&
4918 	    dplx != DUPLEX_FULL)
4919 		goto err_inval;
4920 
4921 	switch (spd + dplx) {
4922 	case SPEED_10 + DUPLEX_HALF:
4923 		hw->forced_speed_duplex = e1000_10_half;
4924 		break;
4925 	case SPEED_10 + DUPLEX_FULL:
4926 		hw->forced_speed_duplex = e1000_10_full;
4927 		break;
4928 	case SPEED_100 + DUPLEX_HALF:
4929 		hw->forced_speed_duplex = e1000_100_half;
4930 		break;
4931 	case SPEED_100 + DUPLEX_FULL:
4932 		hw->forced_speed_duplex = e1000_100_full;
4933 		break;
4934 	case SPEED_1000 + DUPLEX_FULL:
4935 		hw->autoneg = 1;
4936 		hw->autoneg_advertised = ADVERTISE_1000_FULL;
4937 		break;
4938 	case SPEED_1000 + DUPLEX_HALF: /* not supported */
4939 	default:
4940 		goto err_inval;
4941 	}
4942 	return 0;
4943 
4944 err_inval:
4945 	e_err(probe, "Unsupported Speed/Duplex configuration\n");
4946 	return -EINVAL;
4947 }
4948 
4949 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4950 {
4951 	struct net_device *netdev = pci_get_drvdata(pdev);
4952 	struct e1000_adapter *adapter = netdev_priv(netdev);
4953 	struct e1000_hw *hw = &adapter->hw;
4954 	u32 ctrl, ctrl_ext, rctl, status;
4955 	u32 wufc = adapter->wol;
4956 #ifdef CONFIG_PM
4957 	int retval = 0;
4958 #endif
4959 
4960 	netif_device_detach(netdev);
4961 
4962 	if (netif_running(netdev)) {
4963 		WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
4964 		e1000_down(adapter);
4965 	}
4966 
4967 #ifdef CONFIG_PM
4968 	retval = pci_save_state(pdev);
4969 	if (retval)
4970 		return retval;
4971 #endif
4972 
4973 	status = er32(STATUS);
4974 	if (status & E1000_STATUS_LU)
4975 		wufc &= ~E1000_WUFC_LNKC;
4976 
4977 	if (wufc) {
4978 		e1000_setup_rctl(adapter);
4979 		e1000_set_rx_mode(netdev);
4980 
4981 		rctl = er32(RCTL);
4982 
4983 		/* turn on all-multi mode if wake on multicast is enabled */
4984 		if (wufc & E1000_WUFC_MC)
4985 			rctl |= E1000_RCTL_MPE;
4986 
4987 		/* enable receives in the hardware */
4988 		ew32(RCTL, rctl | E1000_RCTL_EN);
4989 
4990 		if (hw->mac_type >= e1000_82540) {
4991 			ctrl = er32(CTRL);
4992 			/* advertise wake from D3Cold */
4993 			#define E1000_CTRL_ADVD3WUC 0x00100000
4994 			/* phy power management enable */
4995 			#define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4996 			ctrl |= E1000_CTRL_ADVD3WUC |
4997 				E1000_CTRL_EN_PHY_PWR_MGMT;
4998 			ew32(CTRL, ctrl);
4999 		}
5000 
5001 		if (hw->media_type == e1000_media_type_fiber ||
5002 		    hw->media_type == e1000_media_type_internal_serdes) {
5003 			/* keep the laser running in D3 */
5004 			ctrl_ext = er32(CTRL_EXT);
5005 			ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
5006 			ew32(CTRL_EXT, ctrl_ext);
5007 		}
5008 
5009 		ew32(WUC, E1000_WUC_PME_EN);
5010 		ew32(WUFC, wufc);
5011 	} else {
5012 		ew32(WUC, 0);
5013 		ew32(WUFC, 0);
5014 	}
5015 
5016 	e1000_release_manageability(adapter);
5017 
5018 	*enable_wake = !!wufc;
5019 
5020 	/* make sure adapter isn't asleep if manageability is enabled */
5021 	if (adapter->en_mng_pt)
5022 		*enable_wake = true;
5023 
5024 	if (netif_running(netdev))
5025 		e1000_free_irq(adapter);
5026 
5027 	pci_disable_device(pdev);
5028 
5029 	return 0;
5030 }
5031 
5032 #ifdef CONFIG_PM
5033 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
5034 {
5035 	int retval;
5036 	bool wake;
5037 
5038 	retval = __e1000_shutdown(pdev, &wake);
5039 	if (retval)
5040 		return retval;
5041 
5042 	if (wake) {
5043 		pci_prepare_to_sleep(pdev);
5044 	} else {
5045 		pci_wake_from_d3(pdev, false);
5046 		pci_set_power_state(pdev, PCI_D3hot);
5047 	}
5048 
5049 	return 0;
5050 }
5051 
5052 static int e1000_resume(struct pci_dev *pdev)
5053 {
5054 	struct net_device *netdev = pci_get_drvdata(pdev);
5055 	struct e1000_adapter *adapter = netdev_priv(netdev);
5056 	struct e1000_hw *hw = &adapter->hw;
5057 	u32 err;
5058 
5059 	pci_set_power_state(pdev, PCI_D0);
5060 	pci_restore_state(pdev);
5061 	pci_save_state(pdev);
5062 
5063 	if (adapter->need_ioport)
5064 		err = pci_enable_device(pdev);
5065 	else
5066 		err = pci_enable_device_mem(pdev);
5067 	if (err) {
5068 		pr_err("Cannot enable PCI device from suspend\n");
5069 		return err;
5070 	}
5071 	pci_set_master(pdev);
5072 
5073 	pci_enable_wake(pdev, PCI_D3hot, 0);
5074 	pci_enable_wake(pdev, PCI_D3cold, 0);
5075 
5076 	if (netif_running(netdev)) {
5077 		err = e1000_request_irq(adapter);
5078 		if (err)
5079 			return err;
5080 	}
5081 
5082 	e1000_power_up_phy(adapter);
5083 	e1000_reset(adapter);
5084 	ew32(WUS, ~0);
5085 
5086 	e1000_init_manageability(adapter);
5087 
5088 	if (netif_running(netdev))
5089 		e1000_up(adapter);
5090 
5091 	netif_device_attach(netdev);
5092 
5093 	return 0;
5094 }
5095 #endif
5096 
5097 static void e1000_shutdown(struct pci_dev *pdev)
5098 {
5099 	bool wake;
5100 
5101 	__e1000_shutdown(pdev, &wake);
5102 
5103 	if (system_state == SYSTEM_POWER_OFF) {
5104 		pci_wake_from_d3(pdev, wake);
5105 		pci_set_power_state(pdev, PCI_D3hot);
5106 	}
5107 }
5108 
5109 #ifdef CONFIG_NET_POLL_CONTROLLER
5110 /*
5111  * Polling 'interrupt' - used by things like netconsole to send skbs
5112  * without having to re-enable interrupts. It's not called while
5113  * the interrupt routine is executing.
5114  */
5115 static void e1000_netpoll(struct net_device *netdev)
5116 {
5117 	struct e1000_adapter *adapter = netdev_priv(netdev);
5118 
5119 	disable_irq(adapter->pdev->irq);
5120 	e1000_intr(adapter->pdev->irq, netdev);
5121 	enable_irq(adapter->pdev->irq);
5122 }
5123 #endif
5124 
5125 /**
5126  * e1000_io_error_detected - called when PCI error is detected
5127  * @pdev: Pointer to PCI device
5128  * @state: The current pci connection state
5129  *
5130  * This function is called after a PCI bus error affecting
5131  * this device has been detected.
5132  */
5133 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5134 						pci_channel_state_t state)
5135 {
5136 	struct net_device *netdev = pci_get_drvdata(pdev);
5137 	struct e1000_adapter *adapter = netdev_priv(netdev);
5138 
5139 	netif_device_detach(netdev);
5140 
5141 	if (state == pci_channel_io_perm_failure)
5142 		return PCI_ERS_RESULT_DISCONNECT;
5143 
5144 	if (netif_running(netdev))
5145 		e1000_down(adapter);
5146 	pci_disable_device(pdev);
5147 
5148 	/* Request a slot slot reset. */
5149 	return PCI_ERS_RESULT_NEED_RESET;
5150 }
5151 
5152 /**
5153  * e1000_io_slot_reset - called after the pci bus has been reset.
5154  * @pdev: Pointer to PCI device
5155  *
5156  * Restart the card from scratch, as if from a cold-boot. Implementation
5157  * resembles the first-half of the e1000_resume routine.
5158  */
5159 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5160 {
5161 	struct net_device *netdev = pci_get_drvdata(pdev);
5162 	struct e1000_adapter *adapter = netdev_priv(netdev);
5163 	struct e1000_hw *hw = &adapter->hw;
5164 	int err;
5165 
5166 	if (adapter->need_ioport)
5167 		err = pci_enable_device(pdev);
5168 	else
5169 		err = pci_enable_device_mem(pdev);
5170 	if (err) {
5171 		pr_err("Cannot re-enable PCI device after reset.\n");
5172 		return PCI_ERS_RESULT_DISCONNECT;
5173 	}
5174 	pci_set_master(pdev);
5175 
5176 	pci_enable_wake(pdev, PCI_D3hot, 0);
5177 	pci_enable_wake(pdev, PCI_D3cold, 0);
5178 
5179 	e1000_reset(adapter);
5180 	ew32(WUS, ~0);
5181 
5182 	return PCI_ERS_RESULT_RECOVERED;
5183 }
5184 
5185 /**
5186  * e1000_io_resume - called when traffic can start flowing again.
5187  * @pdev: Pointer to PCI device
5188  *
5189  * This callback is called when the error recovery driver tells us that
5190  * its OK to resume normal operation. Implementation resembles the
5191  * second-half of the e1000_resume routine.
5192  */
5193 static void e1000_io_resume(struct pci_dev *pdev)
5194 {
5195 	struct net_device *netdev = pci_get_drvdata(pdev);
5196 	struct e1000_adapter *adapter = netdev_priv(netdev);
5197 
5198 	e1000_init_manageability(adapter);
5199 
5200 	if (netif_running(netdev)) {
5201 		if (e1000_up(adapter)) {
5202 			pr_info("can't bring device back up after reset\n");
5203 			return;
5204 		}
5205 	}
5206 
5207 	netif_device_attach(netdev);
5208 }
5209 
5210 /* e1000_main.c */
5211