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 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,
170 				 __be16 proto, u16 vid);
171 static int e1000_vlan_rx_kill_vid(struct net_device *netdev,
172 				  __be16 proto, u16 vid);
173 static void e1000_restore_vlan(struct e1000_adapter *adapter);
174 
175 #ifdef CONFIG_PM
176 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
177 static int e1000_resume(struct pci_dev *pdev);
178 #endif
179 static void e1000_shutdown(struct pci_dev *pdev);
180 
181 #ifdef CONFIG_NET_POLL_CONTROLLER
182 /* for netdump / net console */
183 static void e1000_netpoll (struct net_device *netdev);
184 #endif
185 
186 #define COPYBREAK_DEFAULT 256
187 static unsigned int copybreak __read_mostly = COPYBREAK_DEFAULT;
188 module_param(copybreak, uint, 0644);
189 MODULE_PARM_DESC(copybreak,
190 	"Maximum size of packet that is copied to a new buffer on receive");
191 
192 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
193                      pci_channel_state_t state);
194 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
195 static void e1000_io_resume(struct pci_dev *pdev);
196 
197 static const struct pci_error_handlers e1000_err_handler = {
198 	.error_detected = e1000_io_error_detected,
199 	.slot_reset = e1000_io_slot_reset,
200 	.resume = e1000_io_resume,
201 };
202 
203 static struct pci_driver e1000_driver = {
204 	.name     = e1000_driver_name,
205 	.id_table = e1000_pci_tbl,
206 	.probe    = e1000_probe,
207 	.remove   = e1000_remove,
208 #ifdef CONFIG_PM
209 	/* Power Management Hooks */
210 	.suspend  = e1000_suspend,
211 	.resume   = e1000_resume,
212 #endif
213 	.shutdown = e1000_shutdown,
214 	.err_handler = &e1000_err_handler
215 };
216 
217 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
218 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
219 MODULE_LICENSE("GPL");
220 MODULE_VERSION(DRV_VERSION);
221 
222 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
223 static int debug = -1;
224 module_param(debug, int, 0);
225 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
226 
227 /**
228  * e1000_get_hw_dev - return device
229  * used by hardware layer to print debugging information
230  *
231  **/
232 struct net_device *e1000_get_hw_dev(struct e1000_hw *hw)
233 {
234 	struct e1000_adapter *adapter = hw->back;
235 	return adapter->netdev;
236 }
237 
238 /**
239  * e1000_init_module - Driver Registration Routine
240  *
241  * e1000_init_module is the first routine called when the driver is
242  * loaded. All it does is register with the PCI subsystem.
243  **/
244 static int __init e1000_init_module(void)
245 {
246 	int ret;
247 	pr_info("%s - version %s\n", e1000_driver_string, e1000_driver_version);
248 
249 	pr_info("%s\n", e1000_copyright);
250 
251 	ret = pci_register_driver(&e1000_driver);
252 	if (copybreak != COPYBREAK_DEFAULT) {
253 		if (copybreak == 0)
254 			pr_info("copybreak disabled\n");
255 		else
256 			pr_info("copybreak enabled for "
257 				   "packets <= %u bytes\n", copybreak);
258 	}
259 	return ret;
260 }
261 
262 module_init(e1000_init_module);
263 
264 /**
265  * e1000_exit_module - Driver Exit Cleanup Routine
266  *
267  * e1000_exit_module is called just before the driver is removed
268  * from memory.
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 static void e1000_irq_disable(struct e1000_adapter *adapter)
305 {
306 	struct e1000_hw *hw = &adapter->hw;
307 
308 	ew32(IMC, ~0);
309 	E1000_WRITE_FLUSH();
310 	synchronize_irq(adapter->pdev->irq);
311 }
312 
313 /**
314  * e1000_irq_enable - Enable default interrupt generation settings
315  * @adapter: board private structure
316  **/
317 static void e1000_irq_enable(struct e1000_adapter *adapter)
318 {
319 	struct e1000_hw *hw = &adapter->hw;
320 
321 	ew32(IMS, IMS_ENABLE_MASK);
322 	E1000_WRITE_FLUSH();
323 }
324 
325 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
326 {
327 	struct e1000_hw *hw = &adapter->hw;
328 	struct net_device *netdev = adapter->netdev;
329 	u16 vid = hw->mng_cookie.vlan_id;
330 	u16 old_vid = adapter->mng_vlan_id;
331 
332 	if (!e1000_vlan_used(adapter))
333 		return;
334 
335 	if (!test_bit(vid, adapter->active_vlans)) {
336 		if (hw->mng_cookie.status &
337 		    E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
338 			e1000_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
339 			adapter->mng_vlan_id = vid;
340 		} else {
341 			adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
342 		}
343 		if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
344 		    (vid != old_vid) &&
345 		    !test_bit(old_vid, adapter->active_vlans))
346 			e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
347 					       old_vid);
348 	} else {
349 		adapter->mng_vlan_id = vid;
350 	}
351 }
352 
353 static void e1000_init_manageability(struct e1000_adapter *adapter)
354 {
355 	struct e1000_hw *hw = &adapter->hw;
356 
357 	if (adapter->en_mng_pt) {
358 		u32 manc = er32(MANC);
359 
360 		/* disable hardware interception of ARP */
361 		manc &= ~(E1000_MANC_ARP_EN);
362 
363 		ew32(MANC, manc);
364 	}
365 }
366 
367 static void e1000_release_manageability(struct e1000_adapter *adapter)
368 {
369 	struct e1000_hw *hw = &adapter->hw;
370 
371 	if (adapter->en_mng_pt) {
372 		u32 manc = er32(MANC);
373 
374 		/* re-enable hardware interception of ARP */
375 		manc |= E1000_MANC_ARP_EN;
376 
377 		ew32(MANC, manc);
378 	}
379 }
380 
381 /**
382  * e1000_configure - configure the hardware for RX and TX
383  * @adapter = private board structure
384  **/
385 static void e1000_configure(struct e1000_adapter *adapter)
386 {
387 	struct net_device *netdev = adapter->netdev;
388 	int i;
389 
390 	e1000_set_rx_mode(netdev);
391 
392 	e1000_restore_vlan(adapter);
393 	e1000_init_manageability(adapter);
394 
395 	e1000_configure_tx(adapter);
396 	e1000_setup_rctl(adapter);
397 	e1000_configure_rx(adapter);
398 	/* call E1000_DESC_UNUSED which always leaves
399 	 * at least 1 descriptor unused to make sure
400 	 * next_to_use != next_to_clean
401 	 */
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 void e1000_power_up_phy(struct e1000_adapter *adapter)
438 {
439 	struct e1000_hw *hw = &adapter->hw;
440 	u16 mii_reg = 0;
441 
442 	/* Just clear the power down bit to wake the phy back up */
443 	if (hw->media_type == e1000_media_type_copper) {
444 		/* according to the manual, the phy will retain its
445 		 * settings across a power-down/up cycle
446 		 */
447 		e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
448 		mii_reg &= ~MII_CR_POWER_DOWN;
449 		e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
450 	}
451 }
452 
453 static void e1000_power_down_phy(struct e1000_adapter *adapter)
454 {
455 	struct e1000_hw *hw = &adapter->hw;
456 
457 	/* Power down the PHY so no link is implied when interface is down *
458 	 * The PHY cannot be powered down if any of the following is true *
459 	 * (a) WoL is enabled
460 	 * (b) AMT is active
461 	 * (c) SoL/IDER session is active
462 	 */
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 	/* Setting DOWN must be after irq_disable to prevent
533 	 * a screaming interrupt.  Setting DOWN also prevents
534 	 * tasks from rescheduling.
535 	 */
536 	e1000_down_and_stop(adapter);
537 
538 	adapter->link_speed = 0;
539 	adapter->link_duplex = 0;
540 	netif_carrier_off(netdev);
541 
542 	e1000_reset(adapter);
543 	e1000_clean_all_tx_rings(adapter);
544 	e1000_clean_all_rx_rings(adapter);
545 }
546 
547 static void e1000_reinit_safe(struct e1000_adapter *adapter)
548 {
549 	while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
550 		msleep(1);
551 	mutex_lock(&adapter->mutex);
552 	e1000_down(adapter);
553 	e1000_up(adapter);
554 	mutex_unlock(&adapter->mutex);
555 	clear_bit(__E1000_RESETTING, &adapter->flags);
556 }
557 
558 void e1000_reinit_locked(struct e1000_adapter *adapter)
559 {
560 	/* if rtnl_lock is not held the call path is bogus */
561 	ASSERT_RTNL();
562 	WARN_ON(in_interrupt());
563 	while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
564 		msleep(1);
565 	e1000_down(adapter);
566 	e1000_up(adapter);
567 	clear_bit(__E1000_RESETTING, &adapter->flags);
568 }
569 
570 void e1000_reset(struct e1000_adapter *adapter)
571 {
572 	struct e1000_hw *hw = &adapter->hw;
573 	u32 pba = 0, tx_space, min_tx_space, min_rx_space;
574 	bool legacy_pba_adjust = false;
575 	u16 hwm;
576 
577 	/* Repartition Pba for greater than 9k mtu
578 	 * To take effect CTRL.RST is required.
579 	 */
580 
581 	switch (hw->mac_type) {
582 	case e1000_82542_rev2_0:
583 	case e1000_82542_rev2_1:
584 	case e1000_82543:
585 	case e1000_82544:
586 	case e1000_82540:
587 	case e1000_82541:
588 	case e1000_82541_rev_2:
589 		legacy_pba_adjust = true;
590 		pba = E1000_PBA_48K;
591 		break;
592 	case e1000_82545:
593 	case e1000_82545_rev_3:
594 	case e1000_82546:
595 	case e1000_ce4100:
596 	case e1000_82546_rev_3:
597 		pba = E1000_PBA_48K;
598 		break;
599 	case e1000_82547:
600 	case e1000_82547_rev_2:
601 		legacy_pba_adjust = true;
602 		pba = E1000_PBA_30K;
603 		break;
604 	case e1000_undefined:
605 	case e1000_num_macs:
606 		break;
607 	}
608 
609 	if (legacy_pba_adjust) {
610 		if (hw->max_frame_size > E1000_RXBUFFER_8192)
611 			pba -= 8; /* allocate more FIFO for Tx */
612 
613 		if (hw->mac_type == e1000_82547) {
614 			adapter->tx_fifo_head = 0;
615 			adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
616 			adapter->tx_fifo_size =
617 				(E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
618 			atomic_set(&adapter->tx_fifo_stall, 0);
619 		}
620 	} else if (hw->max_frame_size >  ETH_FRAME_LEN + ETH_FCS_LEN) {
621 		/* adjust PBA for jumbo frames */
622 		ew32(PBA, pba);
623 
624 		/* To maintain wire speed transmits, the Tx FIFO should be
625 		 * large enough to accommodate two full transmit packets,
626 		 * rounded up to the next 1KB and expressed in KB.  Likewise,
627 		 * the Rx FIFO should be large enough to accommodate at least
628 		 * one full receive packet and is similarly rounded up and
629 		 * expressed in KB.
630 		 */
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 		/* the Tx fifo also stores 16 bytes of information about the Tx
637 		 * but don't include ethernet FCS because hardware appends it
638 		 */
639 		min_tx_space = (hw->max_frame_size +
640 		                sizeof(struct e1000_tx_desc) -
641 		                ETH_FCS_LEN) * 2;
642 		min_tx_space = ALIGN(min_tx_space, 1024);
643 		min_tx_space >>= 10;
644 		/* software strips receive CRC, so leave room for it */
645 		min_rx_space = hw->max_frame_size;
646 		min_rx_space = ALIGN(min_rx_space, 1024);
647 		min_rx_space >>= 10;
648 
649 		/* If current Tx allocation is less than the min Tx FIFO size,
650 		 * and the min Tx FIFO size is less than the current Rx FIFO
651 		 * allocation, take space away from current Rx allocation
652 		 */
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 			 */
669 			if (pba < min_rx_space)
670 				pba = min_rx_space;
671 		}
672 	}
673 
674 	ew32(PBA, pba);
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 		 */
712 		ctrl &= ~E1000_CTRL_SWDPIN3;
713 		ew32(CTRL, ctrl);
714 	}
715 
716 	/* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
717 	ew32(VET, ETHERNET_IEEE_VLAN_TYPE);
718 
719 	e1000_reset_adaptive(hw);
720 	e1000_phy_get_info(hw, &adapter->phy_info);
721 
722 	e1000_release_manageability(adapter);
723 }
724 
725 /* Dump the eeprom for users having checksum issues */
726 static void e1000_dump_eeprom(struct e1000_adapter *adapter)
727 {
728 	struct net_device *netdev = adapter->netdev;
729 	struct ethtool_eeprom eeprom;
730 	const struct ethtool_ops *ops = netdev->ethtool_ops;
731 	u8 *data;
732 	int i;
733 	u16 csum_old, csum_new = 0;
734 
735 	eeprom.len = ops->get_eeprom_len(netdev);
736 	eeprom.offset = 0;
737 
738 	data = kmalloc(eeprom.len, GFP_KERNEL);
739 	if (!data)
740 		return;
741 
742 	ops->get_eeprom(netdev, &eeprom, data);
743 
744 	csum_old = (data[EEPROM_CHECKSUM_REG * 2]) +
745 		   (data[EEPROM_CHECKSUM_REG * 2 + 1] << 8);
746 	for (i = 0; i < EEPROM_CHECKSUM_REG * 2; i += 2)
747 		csum_new += data[i] + (data[i + 1] << 8);
748 	csum_new = EEPROM_SUM - csum_new;
749 
750 	pr_err("/*********************/\n");
751 	pr_err("Current EEPROM Checksum : 0x%04x\n", csum_old);
752 	pr_err("Calculated              : 0x%04x\n", csum_new);
753 
754 	pr_err("Offset    Values\n");
755 	pr_err("========  ======\n");
756 	print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 16, 1, data, 128, 0);
757 
758 	pr_err("Include this output when contacting your support provider.\n");
759 	pr_err("This is not a software error! Something bad happened to\n");
760 	pr_err("your hardware or EEPROM image. Ignoring this problem could\n");
761 	pr_err("result in further problems, possibly loss of data,\n");
762 	pr_err("corruption or system hangs!\n");
763 	pr_err("The MAC Address will be reset to 00:00:00:00:00:00,\n");
764 	pr_err("which is invalid and requires you to set the proper MAC\n");
765 	pr_err("address manually before continuing to enable this network\n");
766 	pr_err("device. Please inspect the EEPROM dump and report the\n");
767 	pr_err("issue to your hardware vendor or Intel Customer Support.\n");
768 	pr_err("/*********************/\n");
769 
770 	kfree(data);
771 }
772 
773 /**
774  * e1000_is_need_ioport - determine if an adapter needs ioport resources or not
775  * @pdev: PCI device information struct
776  *
777  * Return true if an adapter needs ioport resources
778  **/
779 static int e1000_is_need_ioport(struct pci_dev *pdev)
780 {
781 	switch (pdev->device) {
782 	case E1000_DEV_ID_82540EM:
783 	case E1000_DEV_ID_82540EM_LOM:
784 	case E1000_DEV_ID_82540EP:
785 	case E1000_DEV_ID_82540EP_LOM:
786 	case E1000_DEV_ID_82540EP_LP:
787 	case E1000_DEV_ID_82541EI:
788 	case E1000_DEV_ID_82541EI_MOBILE:
789 	case E1000_DEV_ID_82541ER:
790 	case E1000_DEV_ID_82541ER_LOM:
791 	case E1000_DEV_ID_82541GI:
792 	case E1000_DEV_ID_82541GI_LF:
793 	case E1000_DEV_ID_82541GI_MOBILE:
794 	case E1000_DEV_ID_82544EI_COPPER:
795 	case E1000_DEV_ID_82544EI_FIBER:
796 	case E1000_DEV_ID_82544GC_COPPER:
797 	case E1000_DEV_ID_82544GC_LOM:
798 	case E1000_DEV_ID_82545EM_COPPER:
799 	case E1000_DEV_ID_82545EM_FIBER:
800 	case E1000_DEV_ID_82546EB_COPPER:
801 	case E1000_DEV_ID_82546EB_FIBER:
802 	case E1000_DEV_ID_82546EB_QUAD_COPPER:
803 		return true;
804 	default:
805 		return false;
806 	}
807 }
808 
809 static netdev_features_t e1000_fix_features(struct net_device *netdev,
810 	netdev_features_t features)
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_CTAG_RX)
816 		features |= NETIF_F_HW_VLAN_CTAG_TX;
817 	else
818 		features &= ~NETIF_F_HW_VLAN_CTAG_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_CTAG_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 e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
942 {
943 	struct net_device *netdev;
944 	struct e1000_adapter *adapter;
945 	struct e1000_hw *hw;
946 
947 	static int cards_found = 0;
948 	static int global_quad_port_a = 0; /* global ksp3 port a indication */
949 	int i, err, pci_using_dac;
950 	u16 eeprom_data = 0;
951 	u16 tmp = 0;
952 	u16 eeprom_apme_mask = E1000_EEPROM_APME;
953 	int bars, need_ioport;
954 
955 	/* do not allocate ioport bars when not needed */
956 	need_ioport = e1000_is_need_ioport(pdev);
957 	if (need_ioport) {
958 		bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
959 		err = pci_enable_device(pdev);
960 	} else {
961 		bars = pci_select_bars(pdev, IORESOURCE_MEM);
962 		err = pci_enable_device_mem(pdev);
963 	}
964 	if (err)
965 		return err;
966 
967 	err = pci_request_selected_regions(pdev, bars, e1000_driver_name);
968 	if (err)
969 		goto err_pci_reg;
970 
971 	pci_set_master(pdev);
972 	err = pci_save_state(pdev);
973 	if (err)
974 		goto err_alloc_etherdev;
975 
976 	err = -ENOMEM;
977 	netdev = alloc_etherdev(sizeof(struct e1000_adapter));
978 	if (!netdev)
979 		goto err_alloc_etherdev;
980 
981 	SET_NETDEV_DEV(netdev, &pdev->dev);
982 
983 	pci_set_drvdata(pdev, netdev);
984 	adapter = netdev_priv(netdev);
985 	adapter->netdev = netdev;
986 	adapter->pdev = pdev;
987 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
988 	adapter->bars = bars;
989 	adapter->need_ioport = need_ioport;
990 
991 	hw = &adapter->hw;
992 	hw->back = adapter;
993 
994 	err = -EIO;
995 	hw->hw_addr = pci_ioremap_bar(pdev, BAR_0);
996 	if (!hw->hw_addr)
997 		goto err_ioremap;
998 
999 	if (adapter->need_ioport) {
1000 		for (i = BAR_1; i <= BAR_5; i++) {
1001 			if (pci_resource_len(pdev, i) == 0)
1002 				continue;
1003 			if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
1004 				hw->io_base = pci_resource_start(pdev, i);
1005 				break;
1006 			}
1007 		}
1008 	}
1009 
1010 	/* make ready for any if (hw->...) below */
1011 	err = e1000_init_hw_struct(adapter, hw);
1012 	if (err)
1013 		goto err_sw_init;
1014 
1015 	/* there is a workaround being applied below that limits
1016 	 * 64-bit DMA addresses to 64-bit hardware.  There are some
1017 	 * 32-bit adapters that Tx hang when given 64-bit DMA addresses
1018 	 */
1019 	pci_using_dac = 0;
1020 	if ((hw->bus_type == e1000_bus_type_pcix) &&
1021 	    !dma_set_mask(&pdev->dev, DMA_BIT_MASK(64))) {
1022 		/* according to DMA-API-HOWTO, coherent calls will always
1023 		 * succeed if the set call did
1024 		 */
1025 		dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
1026 		pci_using_dac = 1;
1027 	} else {
1028 		err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
1029 		if (err) {
1030 			pr_err("No usable DMA config, aborting\n");
1031 			goto err_dma;
1032 		}
1033 		dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32));
1034 	}
1035 
1036 	netdev->netdev_ops = &e1000_netdev_ops;
1037 	e1000_set_ethtool_ops(netdev);
1038 	netdev->watchdog_timeo = 5 * HZ;
1039 	netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
1040 
1041 	strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1042 
1043 	adapter->bd_number = cards_found;
1044 
1045 	/* setup the private structure */
1046 
1047 	err = e1000_sw_init(adapter);
1048 	if (err)
1049 		goto err_sw_init;
1050 
1051 	err = -EIO;
1052 	if (hw->mac_type == e1000_ce4100) {
1053 		hw->ce4100_gbe_mdio_base_virt =
1054 					ioremap(pci_resource_start(pdev, BAR_1),
1055 		                                pci_resource_len(pdev, BAR_1));
1056 
1057 		if (!hw->ce4100_gbe_mdio_base_virt)
1058 			goto err_mdio_ioremap;
1059 	}
1060 
1061 	if (hw->mac_type >= e1000_82543) {
1062 		netdev->hw_features = NETIF_F_SG |
1063 				   NETIF_F_HW_CSUM |
1064 				   NETIF_F_HW_VLAN_CTAG_RX;
1065 		netdev->features = NETIF_F_HW_VLAN_CTAG_TX |
1066 				   NETIF_F_HW_VLAN_CTAG_FILTER;
1067 	}
1068 
1069 	if ((hw->mac_type >= e1000_82544) &&
1070 	   (hw->mac_type != e1000_82547))
1071 		netdev->hw_features |= NETIF_F_TSO;
1072 
1073 	netdev->priv_flags |= IFF_SUPP_NOFCS;
1074 
1075 	netdev->features |= netdev->hw_features;
1076 	netdev->hw_features |= (NETIF_F_RXCSUM |
1077 				NETIF_F_RXALL |
1078 				NETIF_F_RXFCS);
1079 
1080 	if (pci_using_dac) {
1081 		netdev->features |= NETIF_F_HIGHDMA;
1082 		netdev->vlan_features |= NETIF_F_HIGHDMA;
1083 	}
1084 
1085 	netdev->vlan_features |= (NETIF_F_TSO |
1086 				  NETIF_F_HW_CSUM |
1087 				  NETIF_F_SG);
1088 
1089 	netdev->priv_flags |= IFF_UNICAST_FLT;
1090 
1091 	adapter->en_mng_pt = e1000_enable_mng_pass_thru(hw);
1092 
1093 	/* initialize eeprom parameters */
1094 	if (e1000_init_eeprom_params(hw)) {
1095 		e_err(probe, "EEPROM initialization failed\n");
1096 		goto err_eeprom;
1097 	}
1098 
1099 	/* before reading the EEPROM, reset the controller to
1100 	 * put the device in a known good starting state
1101 	 */
1102 
1103 	e1000_reset_hw(hw);
1104 
1105 	/* make sure the EEPROM is good */
1106 	if (e1000_validate_eeprom_checksum(hw) < 0) {
1107 		e_err(probe, "The EEPROM Checksum Is Not Valid\n");
1108 		e1000_dump_eeprom(adapter);
1109 		/* set MAC address to all zeroes to invalidate and temporary
1110 		 * disable this device for the user. This blocks regular
1111 		 * traffic while still permitting ethtool ioctls from reaching
1112 		 * the hardware as well as allowing the user to run the
1113 		 * interface after manually setting a hw addr using
1114 		 * `ip set address`
1115 		 */
1116 		memset(hw->mac_addr, 0, netdev->addr_len);
1117 	} else {
1118 		/* copy the MAC address out of the EEPROM */
1119 		if (e1000_read_mac_addr(hw))
1120 			e_err(probe, "EEPROM Read Error\n");
1121 	}
1122 	/* don't block initalization here due to bad MAC address */
1123 	memcpy(netdev->dev_addr, hw->mac_addr, netdev->addr_len);
1124 
1125 	if (!is_valid_ether_addr(netdev->dev_addr))
1126 		e_err(probe, "Invalid MAC Address\n");
1127 
1128 
1129 	INIT_DELAYED_WORK(&adapter->watchdog_task, e1000_watchdog);
1130 	INIT_DELAYED_WORK(&adapter->fifo_stall_task,
1131 			  e1000_82547_tx_fifo_stall_task);
1132 	INIT_DELAYED_WORK(&adapter->phy_info_task, e1000_update_phy_info_task);
1133 	INIT_WORK(&adapter->reset_task, e1000_reset_task);
1134 
1135 	e1000_check_options(adapter);
1136 
1137 	/* Initial Wake on LAN setting
1138 	 * If APM wake is enabled in the EEPROM,
1139 	 * enable the ACPI Magic Packet filter
1140 	 */
1141 
1142 	switch (hw->mac_type) {
1143 	case e1000_82542_rev2_0:
1144 	case e1000_82542_rev2_1:
1145 	case e1000_82543:
1146 		break;
1147 	case e1000_82544:
1148 		e1000_read_eeprom(hw,
1149 			EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
1150 		eeprom_apme_mask = E1000_EEPROM_82544_APM;
1151 		break;
1152 	case e1000_82546:
1153 	case e1000_82546_rev_3:
1154 		if (er32(STATUS) & E1000_STATUS_FUNC_1){
1155 			e1000_read_eeprom(hw,
1156 				EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1157 			break;
1158 		}
1159 		/* Fall Through */
1160 	default:
1161 		e1000_read_eeprom(hw,
1162 			EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1163 		break;
1164 	}
1165 	if (eeprom_data & eeprom_apme_mask)
1166 		adapter->eeprom_wol |= E1000_WUFC_MAG;
1167 
1168 	/* now that we have the eeprom settings, apply the special cases
1169 	 * where the eeprom may be wrong or the board simply won't support
1170 	 * wake on lan on a particular port
1171 	 */
1172 	switch (pdev->device) {
1173 	case E1000_DEV_ID_82546GB_PCIE:
1174 		adapter->eeprom_wol = 0;
1175 		break;
1176 	case E1000_DEV_ID_82546EB_FIBER:
1177 	case E1000_DEV_ID_82546GB_FIBER:
1178 		/* Wake events only supported on port A for dual fiber
1179 		 * regardless of eeprom setting
1180 		 */
1181 		if (er32(STATUS) & E1000_STATUS_FUNC_1)
1182 			adapter->eeprom_wol = 0;
1183 		break;
1184 	case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
1185 		/* if quad port adapter, disable WoL on all but port A */
1186 		if (global_quad_port_a != 0)
1187 			adapter->eeprom_wol = 0;
1188 		else
1189 			adapter->quad_port_a = true;
1190 		/* Reset for multiple quad port adapters */
1191 		if (++global_quad_port_a == 4)
1192 			global_quad_port_a = 0;
1193 		break;
1194 	}
1195 
1196 	/* initialize the wol settings based on the eeprom settings */
1197 	adapter->wol = adapter->eeprom_wol;
1198 	device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1199 
1200 	/* Auto detect PHY address */
1201 	if (hw->mac_type == e1000_ce4100) {
1202 		for (i = 0; i < 32; i++) {
1203 			hw->phy_addr = i;
1204 			e1000_read_phy_reg(hw, PHY_ID2, &tmp);
1205 			if (tmp == 0 || tmp == 0xFF) {
1206 				if (i == 31)
1207 					goto err_eeprom;
1208 				continue;
1209 			} else
1210 				break;
1211 		}
1212 	}
1213 
1214 	/* reset the hardware with the new settings */
1215 	e1000_reset(adapter);
1216 
1217 	strcpy(netdev->name, "eth%d");
1218 	err = register_netdev(netdev);
1219 	if (err)
1220 		goto err_register;
1221 
1222 	e1000_vlan_filter_on_off(adapter, false);
1223 
1224 	/* print bus type/speed/width info */
1225 	e_info(probe, "(PCI%s:%dMHz:%d-bit) %pM\n",
1226 	       ((hw->bus_type == e1000_bus_type_pcix) ? "-X" : ""),
1227 	       ((hw->bus_speed == e1000_bus_speed_133) ? 133 :
1228 		(hw->bus_speed == e1000_bus_speed_120) ? 120 :
1229 		(hw->bus_speed == e1000_bus_speed_100) ? 100 :
1230 		(hw->bus_speed == e1000_bus_speed_66) ? 66 : 33),
1231 	       ((hw->bus_width == e1000_bus_width_64) ? 64 : 32),
1232 	       netdev->dev_addr);
1233 
1234 	/* carrier off reporting is important to ethtool even BEFORE open */
1235 	netif_carrier_off(netdev);
1236 
1237 	e_info(probe, "Intel(R) PRO/1000 Network Connection\n");
1238 
1239 	cards_found++;
1240 	return 0;
1241 
1242 err_register:
1243 err_eeprom:
1244 	e1000_phy_hw_reset(hw);
1245 
1246 	if (hw->flash_address)
1247 		iounmap(hw->flash_address);
1248 	kfree(adapter->tx_ring);
1249 	kfree(adapter->rx_ring);
1250 err_dma:
1251 err_sw_init:
1252 err_mdio_ioremap:
1253 	iounmap(hw->ce4100_gbe_mdio_base_virt);
1254 	iounmap(hw->hw_addr);
1255 err_ioremap:
1256 	free_netdev(netdev);
1257 err_alloc_etherdev:
1258 	pci_release_selected_regions(pdev, bars);
1259 err_pci_reg:
1260 	pci_disable_device(pdev);
1261 	return err;
1262 }
1263 
1264 /**
1265  * e1000_remove - Device Removal Routine
1266  * @pdev: PCI device information struct
1267  *
1268  * e1000_remove is called by the PCI subsystem to alert the driver
1269  * that it should release a PCI device.  The could be caused by a
1270  * Hot-Plug event, or because the driver is going to be removed from
1271  * memory.
1272  **/
1273 static void e1000_remove(struct pci_dev *pdev)
1274 {
1275 	struct net_device *netdev = pci_get_drvdata(pdev);
1276 	struct e1000_adapter *adapter = netdev_priv(netdev);
1277 	struct e1000_hw *hw = &adapter->hw;
1278 
1279 	e1000_down_and_stop(adapter);
1280 	e1000_release_manageability(adapter);
1281 
1282 	unregister_netdev(netdev);
1283 
1284 	e1000_phy_hw_reset(hw);
1285 
1286 	kfree(adapter->tx_ring);
1287 	kfree(adapter->rx_ring);
1288 
1289 	if (hw->mac_type == e1000_ce4100)
1290 		iounmap(hw->ce4100_gbe_mdio_base_virt);
1291 	iounmap(hw->hw_addr);
1292 	if (hw->flash_address)
1293 		iounmap(hw->flash_address);
1294 	pci_release_selected_regions(pdev, adapter->bars);
1295 
1296 	free_netdev(netdev);
1297 
1298 	pci_disable_device(pdev);
1299 }
1300 
1301 /**
1302  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1303  * @adapter: board private structure to initialize
1304  *
1305  * e1000_sw_init initializes the Adapter private data structure.
1306  * e1000_init_hw_struct MUST be called before this function
1307  **/
1308 static int e1000_sw_init(struct e1000_adapter *adapter)
1309 {
1310 	adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1311 
1312 	adapter->num_tx_queues = 1;
1313 	adapter->num_rx_queues = 1;
1314 
1315 	if (e1000_alloc_queues(adapter)) {
1316 		e_err(probe, "Unable to allocate memory for queues\n");
1317 		return -ENOMEM;
1318 	}
1319 
1320 	/* Explicitly disable IRQ since the NIC can be in any state. */
1321 	e1000_irq_disable(adapter);
1322 
1323 	spin_lock_init(&adapter->stats_lock);
1324 	mutex_init(&adapter->mutex);
1325 
1326 	set_bit(__E1000_DOWN, &adapter->flags);
1327 
1328 	return 0;
1329 }
1330 
1331 /**
1332  * e1000_alloc_queues - Allocate memory for all rings
1333  * @adapter: board private structure to initialize
1334  *
1335  * We allocate one ring per queue at run-time since we don't know the
1336  * number of queues at compile-time.
1337  **/
1338 static int e1000_alloc_queues(struct e1000_adapter *adapter)
1339 {
1340 	adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1341 	                           sizeof(struct e1000_tx_ring), GFP_KERNEL);
1342 	if (!adapter->tx_ring)
1343 		return -ENOMEM;
1344 
1345 	adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1346 	                           sizeof(struct e1000_rx_ring), GFP_KERNEL);
1347 	if (!adapter->rx_ring) {
1348 		kfree(adapter->tx_ring);
1349 		return -ENOMEM;
1350 	}
1351 
1352 	return E1000_SUCCESS;
1353 }
1354 
1355 /**
1356  * e1000_open - Called when a network interface is made active
1357  * @netdev: network interface device structure
1358  *
1359  * Returns 0 on success, negative value on failure
1360  *
1361  * The open entry point is called when a network interface is made
1362  * active by the system (IFF_UP).  At this point all resources needed
1363  * for transmit and receive operations are allocated, the interrupt
1364  * handler is registered with the OS, the watchdog task is started,
1365  * and the stack is notified that the interface is ready.
1366  **/
1367 static int e1000_open(struct net_device *netdev)
1368 {
1369 	struct e1000_adapter *adapter = netdev_priv(netdev);
1370 	struct e1000_hw *hw = &adapter->hw;
1371 	int err;
1372 
1373 	/* disallow open during test */
1374 	if (test_bit(__E1000_TESTING, &adapter->flags))
1375 		return -EBUSY;
1376 
1377 	netif_carrier_off(netdev);
1378 
1379 	/* allocate transmit descriptors */
1380 	err = e1000_setup_all_tx_resources(adapter);
1381 	if (err)
1382 		goto err_setup_tx;
1383 
1384 	/* allocate receive descriptors */
1385 	err = e1000_setup_all_rx_resources(adapter);
1386 	if (err)
1387 		goto err_setup_rx;
1388 
1389 	e1000_power_up_phy(adapter);
1390 
1391 	adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1392 	if ((hw->mng_cookie.status &
1393 			  E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1394 		e1000_update_mng_vlan(adapter);
1395 	}
1396 
1397 	/* before we allocate an interrupt, we must be ready to handle it.
1398 	 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1399 	 * as soon as we call pci_request_irq, so we have to setup our
1400 	 * clean_rx handler before we do so.
1401 	 */
1402 	e1000_configure(adapter);
1403 
1404 	err = e1000_request_irq(adapter);
1405 	if (err)
1406 		goto err_req_irq;
1407 
1408 	/* From here on the code is the same as e1000_up() */
1409 	clear_bit(__E1000_DOWN, &adapter->flags);
1410 
1411 	napi_enable(&adapter->napi);
1412 
1413 	e1000_irq_enable(adapter);
1414 
1415 	netif_start_queue(netdev);
1416 
1417 	/* fire a link status change interrupt to start the watchdog */
1418 	ew32(ICS, E1000_ICS_LSC);
1419 
1420 	return E1000_SUCCESS;
1421 
1422 err_req_irq:
1423 	e1000_power_down_phy(adapter);
1424 	e1000_free_all_rx_resources(adapter);
1425 err_setup_rx:
1426 	e1000_free_all_tx_resources(adapter);
1427 err_setup_tx:
1428 	e1000_reset(adapter);
1429 
1430 	return err;
1431 }
1432 
1433 /**
1434  * e1000_close - Disables a network interface
1435  * @netdev: network interface device structure
1436  *
1437  * Returns 0, this is not allowed to fail
1438  *
1439  * The close entry point is called when an interface is de-activated
1440  * by the OS.  The hardware is still under the drivers control, but
1441  * needs to be disabled.  A global MAC reset is issued to stop the
1442  * hardware, and all transmit and receive resources are freed.
1443  **/
1444 static int e1000_close(struct net_device *netdev)
1445 {
1446 	struct e1000_adapter *adapter = netdev_priv(netdev);
1447 	struct e1000_hw *hw = &adapter->hw;
1448 
1449 	WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1450 	e1000_down(adapter);
1451 	e1000_power_down_phy(adapter);
1452 	e1000_free_irq(adapter);
1453 
1454 	e1000_free_all_tx_resources(adapter);
1455 	e1000_free_all_rx_resources(adapter);
1456 
1457 	/* kill manageability vlan ID if supported, but not if a vlan with
1458 	 * the same ID is registered on the host OS (let 8021q kill it)
1459 	 */
1460 	if ((hw->mng_cookie.status &
1461 	     E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1462 	    !test_bit(adapter->mng_vlan_id, adapter->active_vlans)) {
1463 		e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
1464 				       adapter->mng_vlan_id);
1465 	}
1466 
1467 	return 0;
1468 }
1469 
1470 /**
1471  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1472  * @adapter: address of board private structure
1473  * @start: address of beginning of memory
1474  * @len: length of memory
1475  **/
1476 static bool e1000_check_64k_bound(struct e1000_adapter *adapter, void *start,
1477 				  unsigned long len)
1478 {
1479 	struct e1000_hw *hw = &adapter->hw;
1480 	unsigned long begin = (unsigned long)start;
1481 	unsigned long end = begin + len;
1482 
1483 	/* First rev 82545 and 82546 need to not allow any memory
1484 	 * write location to cross 64k boundary due to errata 23
1485 	 */
1486 	if (hw->mac_type == e1000_82545 ||
1487 	    hw->mac_type == e1000_ce4100 ||
1488 	    hw->mac_type == e1000_82546) {
1489 		return ((begin ^ (end - 1)) >> 16) != 0 ? false : true;
1490 	}
1491 
1492 	return true;
1493 }
1494 
1495 /**
1496  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1497  * @adapter: board private structure
1498  * @txdr:    tx descriptor ring (for a specific queue) to setup
1499  *
1500  * Return 0 on success, negative on failure
1501  **/
1502 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
1503 				    struct e1000_tx_ring *txdr)
1504 {
1505 	struct pci_dev *pdev = adapter->pdev;
1506 	int size;
1507 
1508 	size = sizeof(struct e1000_buffer) * txdr->count;
1509 	txdr->buffer_info = vzalloc(size);
1510 	if (!txdr->buffer_info)
1511 		return -ENOMEM;
1512 
1513 	/* round up to nearest 4K */
1514 
1515 	txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1516 	txdr->size = ALIGN(txdr->size, 4096);
1517 
1518 	txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size, &txdr->dma,
1519 					GFP_KERNEL);
1520 	if (!txdr->desc) {
1521 setup_tx_desc_die:
1522 		vfree(txdr->buffer_info);
1523 		return -ENOMEM;
1524 	}
1525 
1526 	/* Fix for errata 23, can't cross 64kB boundary */
1527 	if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1528 		void *olddesc = txdr->desc;
1529 		dma_addr_t olddma = txdr->dma;
1530 		e_err(tx_err, "txdr align check failed: %u bytes at %p\n",
1531 		      txdr->size, txdr->desc);
1532 		/* Try again, without freeing the previous */
1533 		txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size,
1534 						&txdr->dma, GFP_KERNEL);
1535 		/* Failed allocation, critical failure */
1536 		if (!txdr->desc) {
1537 			dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1538 					  olddma);
1539 			goto setup_tx_desc_die;
1540 		}
1541 
1542 		if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1543 			/* give up */
1544 			dma_free_coherent(&pdev->dev, txdr->size, txdr->desc,
1545 					  txdr->dma);
1546 			dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1547 					  olddma);
1548 			e_err(probe, "Unable to allocate aligned memory "
1549 			      "for the transmit descriptor ring\n");
1550 			vfree(txdr->buffer_info);
1551 			return -ENOMEM;
1552 		} else {
1553 			/* Free old allocation, new allocation was successful */
1554 			dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1555 					  olddma);
1556 		}
1557 	}
1558 	memset(txdr->desc, 0, txdr->size);
1559 
1560 	txdr->next_to_use = 0;
1561 	txdr->next_to_clean = 0;
1562 
1563 	return 0;
1564 }
1565 
1566 /**
1567  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1568  * 				  (Descriptors) for all queues
1569  * @adapter: board private structure
1570  *
1571  * Return 0 on success, negative on failure
1572  **/
1573 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1574 {
1575 	int i, err = 0;
1576 
1577 	for (i = 0; i < adapter->num_tx_queues; i++) {
1578 		err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1579 		if (err) {
1580 			e_err(probe, "Allocation for Tx Queue %u failed\n", i);
1581 			for (i-- ; i >= 0; i--)
1582 				e1000_free_tx_resources(adapter,
1583 							&adapter->tx_ring[i]);
1584 			break;
1585 		}
1586 	}
1587 
1588 	return err;
1589 }
1590 
1591 /**
1592  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1593  * @adapter: board private structure
1594  *
1595  * Configure the Tx unit of the MAC after a reset.
1596  **/
1597 static void e1000_configure_tx(struct e1000_adapter *adapter)
1598 {
1599 	u64 tdba;
1600 	struct e1000_hw *hw = &adapter->hw;
1601 	u32 tdlen, tctl, tipg;
1602 	u32 ipgr1, ipgr2;
1603 
1604 	/* Setup the HW Tx Head and Tail descriptor pointers */
1605 
1606 	switch (adapter->num_tx_queues) {
1607 	case 1:
1608 	default:
1609 		tdba = adapter->tx_ring[0].dma;
1610 		tdlen = adapter->tx_ring[0].count *
1611 			sizeof(struct e1000_tx_desc);
1612 		ew32(TDLEN, tdlen);
1613 		ew32(TDBAH, (tdba >> 32));
1614 		ew32(TDBAL, (tdba & 0x00000000ffffffffULL));
1615 		ew32(TDT, 0);
1616 		ew32(TDH, 0);
1617 		adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ?
1618 					   E1000_TDH : E1000_82542_TDH);
1619 		adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ?
1620 					   E1000_TDT : E1000_82542_TDT);
1621 		break;
1622 	}
1623 
1624 	/* Set the default values for the Tx Inter Packet Gap timer */
1625 	if ((hw->media_type == e1000_media_type_fiber ||
1626 	     hw->media_type == e1000_media_type_internal_serdes))
1627 		tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1628 	else
1629 		tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1630 
1631 	switch (hw->mac_type) {
1632 	case e1000_82542_rev2_0:
1633 	case e1000_82542_rev2_1:
1634 		tipg = DEFAULT_82542_TIPG_IPGT;
1635 		ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1636 		ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1637 		break;
1638 	default:
1639 		ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1640 		ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1641 		break;
1642 	}
1643 	tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1644 	tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1645 	ew32(TIPG, tipg);
1646 
1647 	/* Set the Tx Interrupt Delay register */
1648 
1649 	ew32(TIDV, adapter->tx_int_delay);
1650 	if (hw->mac_type >= e1000_82540)
1651 		ew32(TADV, adapter->tx_abs_int_delay);
1652 
1653 	/* Program the Transmit Control Register */
1654 
1655 	tctl = er32(TCTL);
1656 	tctl &= ~E1000_TCTL_CT;
1657 	tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1658 		(E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1659 
1660 	e1000_config_collision_dist(hw);
1661 
1662 	/* Setup Transmit Descriptor Settings for eop descriptor */
1663 	adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1664 
1665 	/* only set IDE if we are delaying interrupts using the timers */
1666 	if (adapter->tx_int_delay)
1667 		adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1668 
1669 	if (hw->mac_type < e1000_82543)
1670 		adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1671 	else
1672 		adapter->txd_cmd |= E1000_TXD_CMD_RS;
1673 
1674 	/* Cache if we're 82544 running in PCI-X because we'll
1675 	 * need this to apply a workaround later in the send path.
1676 	 */
1677 	if (hw->mac_type == e1000_82544 &&
1678 	    hw->bus_type == e1000_bus_type_pcix)
1679 		adapter->pcix_82544 = true;
1680 
1681 	ew32(TCTL, tctl);
1682 
1683 }
1684 
1685 /**
1686  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1687  * @adapter: board private structure
1688  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1689  *
1690  * Returns 0 on success, negative on failure
1691  **/
1692 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
1693 				    struct e1000_rx_ring *rxdr)
1694 {
1695 	struct pci_dev *pdev = adapter->pdev;
1696 	int size, desc_len;
1697 
1698 	size = sizeof(struct e1000_buffer) * rxdr->count;
1699 	rxdr->buffer_info = vzalloc(size);
1700 	if (!rxdr->buffer_info)
1701 		return -ENOMEM;
1702 
1703 	desc_len = sizeof(struct e1000_rx_desc);
1704 
1705 	/* Round up to nearest 4K */
1706 
1707 	rxdr->size = rxdr->count * desc_len;
1708 	rxdr->size = ALIGN(rxdr->size, 4096);
1709 
1710 	rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size, &rxdr->dma,
1711 					GFP_KERNEL);
1712 	if (!rxdr->desc) {
1713 setup_rx_desc_die:
1714 		vfree(rxdr->buffer_info);
1715 		return -ENOMEM;
1716 	}
1717 
1718 	/* Fix for errata 23, can't cross 64kB boundary */
1719 	if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1720 		void *olddesc = rxdr->desc;
1721 		dma_addr_t olddma = rxdr->dma;
1722 		e_err(rx_err, "rxdr align check failed: %u bytes at %p\n",
1723 		      rxdr->size, rxdr->desc);
1724 		/* Try again, without freeing the previous */
1725 		rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size,
1726 						&rxdr->dma, GFP_KERNEL);
1727 		/* Failed allocation, critical failure */
1728 		if (!rxdr->desc) {
1729 			dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1730 					  olddma);
1731 			goto setup_rx_desc_die;
1732 		}
1733 
1734 		if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1735 			/* give up */
1736 			dma_free_coherent(&pdev->dev, rxdr->size, rxdr->desc,
1737 					  rxdr->dma);
1738 			dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1739 					  olddma);
1740 			e_err(probe, "Unable to allocate aligned memory for "
1741 			      "the Rx descriptor ring\n");
1742 			goto setup_rx_desc_die;
1743 		} else {
1744 			/* Free old allocation, new allocation was successful */
1745 			dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1746 					  olddma);
1747 		}
1748 	}
1749 	memset(rxdr->desc, 0, rxdr->size);
1750 
1751 	rxdr->next_to_clean = 0;
1752 	rxdr->next_to_use = 0;
1753 	rxdr->rx_skb_top = NULL;
1754 
1755 	return 0;
1756 }
1757 
1758 /**
1759  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1760  * 				  (Descriptors) for all queues
1761  * @adapter: board private structure
1762  *
1763  * Return 0 on success, negative on failure
1764  **/
1765 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1766 {
1767 	int i, err = 0;
1768 
1769 	for (i = 0; i < adapter->num_rx_queues; i++) {
1770 		err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1771 		if (err) {
1772 			e_err(probe, "Allocation for Rx Queue %u failed\n", i);
1773 			for (i-- ; i >= 0; i--)
1774 				e1000_free_rx_resources(adapter,
1775 							&adapter->rx_ring[i]);
1776 			break;
1777 		}
1778 	}
1779 
1780 	return err;
1781 }
1782 
1783 /**
1784  * e1000_setup_rctl - configure the receive control registers
1785  * @adapter: Board private structure
1786  **/
1787 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1788 {
1789 	struct e1000_hw *hw = &adapter->hw;
1790 	u32 rctl;
1791 
1792 	rctl = er32(RCTL);
1793 
1794 	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1795 
1796 	rctl |= E1000_RCTL_BAM | E1000_RCTL_LBM_NO |
1797 		E1000_RCTL_RDMTS_HALF |
1798 		(hw->mc_filter_type << E1000_RCTL_MO_SHIFT);
1799 
1800 	if (hw->tbi_compatibility_on == 1)
1801 		rctl |= E1000_RCTL_SBP;
1802 	else
1803 		rctl &= ~E1000_RCTL_SBP;
1804 
1805 	if (adapter->netdev->mtu <= ETH_DATA_LEN)
1806 		rctl &= ~E1000_RCTL_LPE;
1807 	else
1808 		rctl |= E1000_RCTL_LPE;
1809 
1810 	/* Setup buffer sizes */
1811 	rctl &= ~E1000_RCTL_SZ_4096;
1812 	rctl |= E1000_RCTL_BSEX;
1813 	switch (adapter->rx_buffer_len) {
1814 		case E1000_RXBUFFER_2048:
1815 		default:
1816 			rctl |= E1000_RCTL_SZ_2048;
1817 			rctl &= ~E1000_RCTL_BSEX;
1818 			break;
1819 		case E1000_RXBUFFER_4096:
1820 			rctl |= E1000_RCTL_SZ_4096;
1821 			break;
1822 		case E1000_RXBUFFER_8192:
1823 			rctl |= E1000_RCTL_SZ_8192;
1824 			break;
1825 		case E1000_RXBUFFER_16384:
1826 			rctl |= E1000_RCTL_SZ_16384;
1827 			break;
1828 	}
1829 
1830 	/* This is useful for sniffing bad packets. */
1831 	if (adapter->netdev->features & NETIF_F_RXALL) {
1832 		/* UPE and MPE will be handled by normal PROMISC logic
1833 		 * in e1000e_set_rx_mode
1834 		 */
1835 		rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
1836 			 E1000_RCTL_BAM | /* RX All Bcast Pkts */
1837 			 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
1838 
1839 		rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
1840 			  E1000_RCTL_DPF | /* Allow filtered pause */
1841 			  E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
1842 		/* Do not mess with E1000_CTRL_VME, it affects transmit as well,
1843 		 * and that breaks VLANs.
1844 		 */
1845 	}
1846 
1847 	ew32(RCTL, rctl);
1848 }
1849 
1850 /**
1851  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1852  * @adapter: board private structure
1853  *
1854  * Configure the Rx unit of the MAC after a reset.
1855  **/
1856 static void e1000_configure_rx(struct e1000_adapter *adapter)
1857 {
1858 	u64 rdba;
1859 	struct e1000_hw *hw = &adapter->hw;
1860 	u32 rdlen, rctl, rxcsum;
1861 
1862 	if (adapter->netdev->mtu > ETH_DATA_LEN) {
1863 		rdlen = adapter->rx_ring[0].count *
1864 		        sizeof(struct e1000_rx_desc);
1865 		adapter->clean_rx = e1000_clean_jumbo_rx_irq;
1866 		adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
1867 	} else {
1868 		rdlen = adapter->rx_ring[0].count *
1869 		        sizeof(struct e1000_rx_desc);
1870 		adapter->clean_rx = e1000_clean_rx_irq;
1871 		adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1872 	}
1873 
1874 	/* disable receives while setting up the descriptors */
1875 	rctl = er32(RCTL);
1876 	ew32(RCTL, rctl & ~E1000_RCTL_EN);
1877 
1878 	/* set the Receive Delay Timer Register */
1879 	ew32(RDTR, adapter->rx_int_delay);
1880 
1881 	if (hw->mac_type >= e1000_82540) {
1882 		ew32(RADV, adapter->rx_abs_int_delay);
1883 		if (adapter->itr_setting != 0)
1884 			ew32(ITR, 1000000000 / (adapter->itr * 256));
1885 	}
1886 
1887 	/* Setup the HW Rx Head and Tail Descriptor Pointers and
1888 	 * the Base and Length of the Rx Descriptor Ring
1889 	 */
1890 	switch (adapter->num_rx_queues) {
1891 	case 1:
1892 	default:
1893 		rdba = adapter->rx_ring[0].dma;
1894 		ew32(RDLEN, rdlen);
1895 		ew32(RDBAH, (rdba >> 32));
1896 		ew32(RDBAL, (rdba & 0x00000000ffffffffULL));
1897 		ew32(RDT, 0);
1898 		ew32(RDH, 0);
1899 		adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ?
1900 					   E1000_RDH : E1000_82542_RDH);
1901 		adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ?
1902 					   E1000_RDT : E1000_82542_RDT);
1903 		break;
1904 	}
1905 
1906 	/* Enable 82543 Receive Checksum Offload for TCP and UDP */
1907 	if (hw->mac_type >= e1000_82543) {
1908 		rxcsum = er32(RXCSUM);
1909 		if (adapter->rx_csum)
1910 			rxcsum |= E1000_RXCSUM_TUOFL;
1911 		else
1912 			/* don't need to clear IPPCSE as it defaults to 0 */
1913 			rxcsum &= ~E1000_RXCSUM_TUOFL;
1914 		ew32(RXCSUM, rxcsum);
1915 	}
1916 
1917 	/* Enable Receives */
1918 	ew32(RCTL, rctl | E1000_RCTL_EN);
1919 }
1920 
1921 /**
1922  * e1000_free_tx_resources - Free Tx Resources per Queue
1923  * @adapter: board private structure
1924  * @tx_ring: Tx descriptor ring for a specific queue
1925  *
1926  * Free all transmit software resources
1927  **/
1928 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
1929 				    struct e1000_tx_ring *tx_ring)
1930 {
1931 	struct pci_dev *pdev = adapter->pdev;
1932 
1933 	e1000_clean_tx_ring(adapter, tx_ring);
1934 
1935 	vfree(tx_ring->buffer_info);
1936 	tx_ring->buffer_info = NULL;
1937 
1938 	dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1939 			  tx_ring->dma);
1940 
1941 	tx_ring->desc = NULL;
1942 }
1943 
1944 /**
1945  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1946  * @adapter: board private structure
1947  *
1948  * Free all transmit software resources
1949  **/
1950 void e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1951 {
1952 	int i;
1953 
1954 	for (i = 0; i < adapter->num_tx_queues; i++)
1955 		e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1956 }
1957 
1958 static void e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1959 					     struct e1000_buffer *buffer_info)
1960 {
1961 	if (buffer_info->dma) {
1962 		if (buffer_info->mapped_as_page)
1963 			dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1964 				       buffer_info->length, DMA_TO_DEVICE);
1965 		else
1966 			dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1967 					 buffer_info->length,
1968 					 DMA_TO_DEVICE);
1969 		buffer_info->dma = 0;
1970 	}
1971 	if (buffer_info->skb) {
1972 		dev_kfree_skb_any(buffer_info->skb);
1973 		buffer_info->skb = NULL;
1974 	}
1975 	buffer_info->time_stamp = 0;
1976 	/* buffer_info must be completely set up in the transmit path */
1977 }
1978 
1979 /**
1980  * e1000_clean_tx_ring - Free Tx Buffers
1981  * @adapter: board private structure
1982  * @tx_ring: ring to be cleaned
1983  **/
1984 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
1985 				struct e1000_tx_ring *tx_ring)
1986 {
1987 	struct e1000_hw *hw = &adapter->hw;
1988 	struct e1000_buffer *buffer_info;
1989 	unsigned long size;
1990 	unsigned int i;
1991 
1992 	/* Free all the Tx ring sk_buffs */
1993 
1994 	for (i = 0; i < tx_ring->count; i++) {
1995 		buffer_info = &tx_ring->buffer_info[i];
1996 		e1000_unmap_and_free_tx_resource(adapter, buffer_info);
1997 	}
1998 
1999 	netdev_reset_queue(adapter->netdev);
2000 	size = sizeof(struct e1000_buffer) * tx_ring->count;
2001 	memset(tx_ring->buffer_info, 0, size);
2002 
2003 	/* Zero out the descriptor ring */
2004 
2005 	memset(tx_ring->desc, 0, tx_ring->size);
2006 
2007 	tx_ring->next_to_use = 0;
2008 	tx_ring->next_to_clean = 0;
2009 	tx_ring->last_tx_tso = false;
2010 
2011 	writel(0, hw->hw_addr + tx_ring->tdh);
2012 	writel(0, hw->hw_addr + tx_ring->tdt);
2013 }
2014 
2015 /**
2016  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
2017  * @adapter: board private structure
2018  **/
2019 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
2020 {
2021 	int i;
2022 
2023 	for (i = 0; i < adapter->num_tx_queues; i++)
2024 		e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2025 }
2026 
2027 /**
2028  * e1000_free_rx_resources - Free Rx Resources
2029  * @adapter: board private structure
2030  * @rx_ring: ring to clean the resources from
2031  *
2032  * Free all receive software resources
2033  **/
2034 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
2035 				    struct e1000_rx_ring *rx_ring)
2036 {
2037 	struct pci_dev *pdev = adapter->pdev;
2038 
2039 	e1000_clean_rx_ring(adapter, rx_ring);
2040 
2041 	vfree(rx_ring->buffer_info);
2042 	rx_ring->buffer_info = NULL;
2043 
2044 	dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2045 			  rx_ring->dma);
2046 
2047 	rx_ring->desc = NULL;
2048 }
2049 
2050 /**
2051  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
2052  * @adapter: board private structure
2053  *
2054  * Free all receive software resources
2055  **/
2056 void e1000_free_all_rx_resources(struct e1000_adapter *adapter)
2057 {
2058 	int i;
2059 
2060 	for (i = 0; i < adapter->num_rx_queues; i++)
2061 		e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
2062 }
2063 
2064 /**
2065  * e1000_clean_rx_ring - Free Rx Buffers per Queue
2066  * @adapter: board private structure
2067  * @rx_ring: ring to free buffers from
2068  **/
2069 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
2070 				struct e1000_rx_ring *rx_ring)
2071 {
2072 	struct e1000_hw *hw = &adapter->hw;
2073 	struct e1000_buffer *buffer_info;
2074 	struct pci_dev *pdev = adapter->pdev;
2075 	unsigned long size;
2076 	unsigned int i;
2077 
2078 	/* Free all the Rx ring sk_buffs */
2079 	for (i = 0; i < rx_ring->count; i++) {
2080 		buffer_info = &rx_ring->buffer_info[i];
2081 		if (buffer_info->dma &&
2082 		    adapter->clean_rx == e1000_clean_rx_irq) {
2083 			dma_unmap_single(&pdev->dev, buffer_info->dma,
2084 			                 buffer_info->length,
2085 					 DMA_FROM_DEVICE);
2086 		} else if (buffer_info->dma &&
2087 		           adapter->clean_rx == e1000_clean_jumbo_rx_irq) {
2088 			dma_unmap_page(&pdev->dev, buffer_info->dma,
2089 				       buffer_info->length,
2090 				       DMA_FROM_DEVICE);
2091 		}
2092 
2093 		buffer_info->dma = 0;
2094 		if (buffer_info->page) {
2095 			put_page(buffer_info->page);
2096 			buffer_info->page = NULL;
2097 		}
2098 		if (buffer_info->skb) {
2099 			dev_kfree_skb(buffer_info->skb);
2100 			buffer_info->skb = NULL;
2101 		}
2102 	}
2103 
2104 	/* there also may be some cached data from a chained receive */
2105 	if (rx_ring->rx_skb_top) {
2106 		dev_kfree_skb(rx_ring->rx_skb_top);
2107 		rx_ring->rx_skb_top = NULL;
2108 	}
2109 
2110 	size = sizeof(struct e1000_buffer) * rx_ring->count;
2111 	memset(rx_ring->buffer_info, 0, size);
2112 
2113 	/* Zero out the descriptor ring */
2114 	memset(rx_ring->desc, 0, rx_ring->size);
2115 
2116 	rx_ring->next_to_clean = 0;
2117 	rx_ring->next_to_use = 0;
2118 
2119 	writel(0, hw->hw_addr + rx_ring->rdh);
2120 	writel(0, hw->hw_addr + rx_ring->rdt);
2121 }
2122 
2123 /**
2124  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2125  * @adapter: board private structure
2126  **/
2127 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2128 {
2129 	int i;
2130 
2131 	for (i = 0; i < adapter->num_rx_queues; i++)
2132 		e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2133 }
2134 
2135 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2136  * and memory write and invalidate disabled for certain operations
2137  */
2138 static void e1000_enter_82542_rst(struct e1000_adapter *adapter)
2139 {
2140 	struct e1000_hw *hw = &adapter->hw;
2141 	struct net_device *netdev = adapter->netdev;
2142 	u32 rctl;
2143 
2144 	e1000_pci_clear_mwi(hw);
2145 
2146 	rctl = er32(RCTL);
2147 	rctl |= E1000_RCTL_RST;
2148 	ew32(RCTL, rctl);
2149 	E1000_WRITE_FLUSH();
2150 	mdelay(5);
2151 
2152 	if (netif_running(netdev))
2153 		e1000_clean_all_rx_rings(adapter);
2154 }
2155 
2156 static void e1000_leave_82542_rst(struct e1000_adapter *adapter)
2157 {
2158 	struct e1000_hw *hw = &adapter->hw;
2159 	struct net_device *netdev = adapter->netdev;
2160 	u32 rctl;
2161 
2162 	rctl = er32(RCTL);
2163 	rctl &= ~E1000_RCTL_RST;
2164 	ew32(RCTL, rctl);
2165 	E1000_WRITE_FLUSH();
2166 	mdelay(5);
2167 
2168 	if (hw->pci_cmd_word & PCI_COMMAND_INVALIDATE)
2169 		e1000_pci_set_mwi(hw);
2170 
2171 	if (netif_running(netdev)) {
2172 		/* No need to loop, because 82542 supports only 1 queue */
2173 		struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2174 		e1000_configure_rx(adapter);
2175 		adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2176 	}
2177 }
2178 
2179 /**
2180  * e1000_set_mac - Change the Ethernet Address of the NIC
2181  * @netdev: network interface device structure
2182  * @p: pointer to an address structure
2183  *
2184  * Returns 0 on success, negative on failure
2185  **/
2186 static int e1000_set_mac(struct net_device *netdev, void *p)
2187 {
2188 	struct e1000_adapter *adapter = netdev_priv(netdev);
2189 	struct e1000_hw *hw = &adapter->hw;
2190 	struct sockaddr *addr = p;
2191 
2192 	if (!is_valid_ether_addr(addr->sa_data))
2193 		return -EADDRNOTAVAIL;
2194 
2195 	/* 82542 2.0 needs to be in reset to write receive address registers */
2196 
2197 	if (hw->mac_type == e1000_82542_rev2_0)
2198 		e1000_enter_82542_rst(adapter);
2199 
2200 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2201 	memcpy(hw->mac_addr, addr->sa_data, netdev->addr_len);
2202 
2203 	e1000_rar_set(hw, hw->mac_addr, 0);
2204 
2205 	if (hw->mac_type == e1000_82542_rev2_0)
2206 		e1000_leave_82542_rst(adapter);
2207 
2208 	return 0;
2209 }
2210 
2211 /**
2212  * e1000_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2213  * @netdev: network interface device structure
2214  *
2215  * The set_rx_mode entry point is called whenever the unicast or multicast
2216  * address lists or the network interface flags are updated. This routine is
2217  * responsible for configuring the hardware for proper unicast, multicast,
2218  * promiscuous mode, and all-multi behavior.
2219  **/
2220 static void e1000_set_rx_mode(struct net_device *netdev)
2221 {
2222 	struct e1000_adapter *adapter = netdev_priv(netdev);
2223 	struct e1000_hw *hw = &adapter->hw;
2224 	struct netdev_hw_addr *ha;
2225 	bool use_uc = false;
2226 	u32 rctl;
2227 	u32 hash_value;
2228 	int i, rar_entries = E1000_RAR_ENTRIES;
2229 	int mta_reg_count = E1000_NUM_MTA_REGISTERS;
2230 	u32 *mcarray = kcalloc(mta_reg_count, sizeof(u32), GFP_ATOMIC);
2231 
2232 	if (!mcarray)
2233 		return;
2234 
2235 	/* Check for Promiscuous and All Multicast modes */
2236 
2237 	rctl = er32(RCTL);
2238 
2239 	if (netdev->flags & IFF_PROMISC) {
2240 		rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2241 		rctl &= ~E1000_RCTL_VFE;
2242 	} else {
2243 		if (netdev->flags & IFF_ALLMULTI)
2244 			rctl |= E1000_RCTL_MPE;
2245 		else
2246 			rctl &= ~E1000_RCTL_MPE;
2247 		/* Enable VLAN filter if there is a VLAN */
2248 		if (e1000_vlan_used(adapter))
2249 			rctl |= E1000_RCTL_VFE;
2250 	}
2251 
2252 	if (netdev_uc_count(netdev) > rar_entries - 1) {
2253 		rctl |= E1000_RCTL_UPE;
2254 	} else if (!(netdev->flags & IFF_PROMISC)) {
2255 		rctl &= ~E1000_RCTL_UPE;
2256 		use_uc = true;
2257 	}
2258 
2259 	ew32(RCTL, rctl);
2260 
2261 	/* 82542 2.0 needs to be in reset to write receive address registers */
2262 
2263 	if (hw->mac_type == e1000_82542_rev2_0)
2264 		e1000_enter_82542_rst(adapter);
2265 
2266 	/* load the first 14 addresses into the exact filters 1-14. Unicast
2267 	 * addresses take precedence to avoid disabling unicast filtering
2268 	 * when possible.
2269 	 *
2270 	 * RAR 0 is used for the station MAC address
2271 	 * if there are not 14 addresses, go ahead and clear the filters
2272 	 */
2273 	i = 1;
2274 	if (use_uc)
2275 		netdev_for_each_uc_addr(ha, netdev) {
2276 			if (i == rar_entries)
2277 				break;
2278 			e1000_rar_set(hw, ha->addr, i++);
2279 		}
2280 
2281 	netdev_for_each_mc_addr(ha, netdev) {
2282 		if (i == rar_entries) {
2283 			/* load any remaining addresses into the hash table */
2284 			u32 hash_reg, hash_bit, mta;
2285 			hash_value = e1000_hash_mc_addr(hw, ha->addr);
2286 			hash_reg = (hash_value >> 5) & 0x7F;
2287 			hash_bit = hash_value & 0x1F;
2288 			mta = (1 << hash_bit);
2289 			mcarray[hash_reg] |= mta;
2290 		} else {
2291 			e1000_rar_set(hw, ha->addr, i++);
2292 		}
2293 	}
2294 
2295 	for (; i < rar_entries; i++) {
2296 		E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2297 		E1000_WRITE_FLUSH();
2298 		E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2299 		E1000_WRITE_FLUSH();
2300 	}
2301 
2302 	/* write the hash table completely, write from bottom to avoid
2303 	 * both stupid write combining chipsets, and flushing each write
2304 	 */
2305 	for (i = mta_reg_count - 1; i >= 0 ; i--) {
2306 		/* If we are on an 82544 has an errata where writing odd
2307 		 * offsets overwrites the previous even offset, but writing
2308 		 * backwards over the range solves the issue by always
2309 		 * writing the odd offset first
2310 		 */
2311 		E1000_WRITE_REG_ARRAY(hw, MTA, i, mcarray[i]);
2312 	}
2313 	E1000_WRITE_FLUSH();
2314 
2315 	if (hw->mac_type == e1000_82542_rev2_0)
2316 		e1000_leave_82542_rst(adapter);
2317 
2318 	kfree(mcarray);
2319 }
2320 
2321 /**
2322  * e1000_update_phy_info_task - get phy info
2323  * @work: work struct contained inside adapter struct
2324  *
2325  * Need to wait a few seconds after link up to get diagnostic information from
2326  * the phy
2327  */
2328 static void e1000_update_phy_info_task(struct work_struct *work)
2329 {
2330 	struct e1000_adapter *adapter = container_of(work,
2331 						     struct e1000_adapter,
2332 						     phy_info_task.work);
2333 	if (test_bit(__E1000_DOWN, &adapter->flags))
2334 		return;
2335 	mutex_lock(&adapter->mutex);
2336 	e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
2337 	mutex_unlock(&adapter->mutex);
2338 }
2339 
2340 /**
2341  * e1000_82547_tx_fifo_stall_task - task to complete work
2342  * @work: work struct contained inside adapter struct
2343  **/
2344 static void e1000_82547_tx_fifo_stall_task(struct work_struct *work)
2345 {
2346 	struct e1000_adapter *adapter = container_of(work,
2347 						     struct e1000_adapter,
2348 						     fifo_stall_task.work);
2349 	struct e1000_hw *hw = &adapter->hw;
2350 	struct net_device *netdev = adapter->netdev;
2351 	u32 tctl;
2352 
2353 	if (test_bit(__E1000_DOWN, &adapter->flags))
2354 		return;
2355 	mutex_lock(&adapter->mutex);
2356 	if (atomic_read(&adapter->tx_fifo_stall)) {
2357 		if ((er32(TDT) == er32(TDH)) &&
2358 		   (er32(TDFT) == er32(TDFH)) &&
2359 		   (er32(TDFTS) == er32(TDFHS))) {
2360 			tctl = er32(TCTL);
2361 			ew32(TCTL, tctl & ~E1000_TCTL_EN);
2362 			ew32(TDFT, adapter->tx_head_addr);
2363 			ew32(TDFH, adapter->tx_head_addr);
2364 			ew32(TDFTS, adapter->tx_head_addr);
2365 			ew32(TDFHS, adapter->tx_head_addr);
2366 			ew32(TCTL, tctl);
2367 			E1000_WRITE_FLUSH();
2368 
2369 			adapter->tx_fifo_head = 0;
2370 			atomic_set(&adapter->tx_fifo_stall, 0);
2371 			netif_wake_queue(netdev);
2372 		} else if (!test_bit(__E1000_DOWN, &adapter->flags)) {
2373 			schedule_delayed_work(&adapter->fifo_stall_task, 1);
2374 		}
2375 	}
2376 	mutex_unlock(&adapter->mutex);
2377 }
2378 
2379 bool e1000_has_link(struct e1000_adapter *adapter)
2380 {
2381 	struct e1000_hw *hw = &adapter->hw;
2382 	bool link_active = false;
2383 
2384 	/* get_link_status is set on LSC (link status) interrupt or rx
2385 	 * sequence error interrupt (except on intel ce4100).
2386 	 * get_link_status will stay false until the
2387 	 * e1000_check_for_link establishes link for copper adapters
2388 	 * ONLY
2389 	 */
2390 	switch (hw->media_type) {
2391 	case e1000_media_type_copper:
2392 		if (hw->mac_type == e1000_ce4100)
2393 			hw->get_link_status = 1;
2394 		if (hw->get_link_status) {
2395 			e1000_check_for_link(hw);
2396 			link_active = !hw->get_link_status;
2397 		} else {
2398 			link_active = true;
2399 		}
2400 		break;
2401 	case e1000_media_type_fiber:
2402 		e1000_check_for_link(hw);
2403 		link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2404 		break;
2405 	case e1000_media_type_internal_serdes:
2406 		e1000_check_for_link(hw);
2407 		link_active = hw->serdes_has_link;
2408 		break;
2409 	default:
2410 		break;
2411 	}
2412 
2413 	return link_active;
2414 }
2415 
2416 /**
2417  * e1000_watchdog - work function
2418  * @work: work struct contained inside adapter struct
2419  **/
2420 static void e1000_watchdog(struct work_struct *work)
2421 {
2422 	struct e1000_adapter *adapter = container_of(work,
2423 						     struct e1000_adapter,
2424 						     watchdog_task.work);
2425 	struct e1000_hw *hw = &adapter->hw;
2426 	struct net_device *netdev = adapter->netdev;
2427 	struct e1000_tx_ring *txdr = adapter->tx_ring;
2428 	u32 link, tctl;
2429 
2430 	if (test_bit(__E1000_DOWN, &adapter->flags))
2431 		return;
2432 
2433 	mutex_lock(&adapter->mutex);
2434 	link = e1000_has_link(adapter);
2435 	if ((netif_carrier_ok(netdev)) && link)
2436 		goto link_up;
2437 
2438 	if (link) {
2439 		if (!netif_carrier_ok(netdev)) {
2440 			u32 ctrl;
2441 			bool txb2b = true;
2442 			/* update snapshot of PHY registers on LSC */
2443 			e1000_get_speed_and_duplex(hw,
2444 						   &adapter->link_speed,
2445 						   &adapter->link_duplex);
2446 
2447 			ctrl = er32(CTRL);
2448 			pr_info("%s NIC Link is Up %d Mbps %s, "
2449 				"Flow Control: %s\n",
2450 				netdev->name,
2451 				adapter->link_speed,
2452 				adapter->link_duplex == FULL_DUPLEX ?
2453 				"Full Duplex" : "Half Duplex",
2454 				((ctrl & E1000_CTRL_TFCE) && (ctrl &
2455 				E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2456 				E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2457 				E1000_CTRL_TFCE) ? "TX" : "None")));
2458 
2459 			/* adjust timeout factor according to speed/duplex */
2460 			adapter->tx_timeout_factor = 1;
2461 			switch (adapter->link_speed) {
2462 			case SPEED_10:
2463 				txb2b = false;
2464 				adapter->tx_timeout_factor = 16;
2465 				break;
2466 			case SPEED_100:
2467 				txb2b = false;
2468 				/* maybe add some timeout factor ? */
2469 				break;
2470 			}
2471 
2472 			/* enable transmits in the hardware */
2473 			tctl = er32(TCTL);
2474 			tctl |= E1000_TCTL_EN;
2475 			ew32(TCTL, tctl);
2476 
2477 			netif_carrier_on(netdev);
2478 			if (!test_bit(__E1000_DOWN, &adapter->flags))
2479 				schedule_delayed_work(&adapter->phy_info_task,
2480 						      2 * HZ);
2481 			adapter->smartspeed = 0;
2482 		}
2483 	} else {
2484 		if (netif_carrier_ok(netdev)) {
2485 			adapter->link_speed = 0;
2486 			adapter->link_duplex = 0;
2487 			pr_info("%s NIC Link is Down\n",
2488 				netdev->name);
2489 			netif_carrier_off(netdev);
2490 
2491 			if (!test_bit(__E1000_DOWN, &adapter->flags))
2492 				schedule_delayed_work(&adapter->phy_info_task,
2493 						      2 * HZ);
2494 		}
2495 
2496 		e1000_smartspeed(adapter);
2497 	}
2498 
2499 link_up:
2500 	e1000_update_stats(adapter);
2501 
2502 	hw->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2503 	adapter->tpt_old = adapter->stats.tpt;
2504 	hw->collision_delta = adapter->stats.colc - adapter->colc_old;
2505 	adapter->colc_old = adapter->stats.colc;
2506 
2507 	adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2508 	adapter->gorcl_old = adapter->stats.gorcl;
2509 	adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2510 	adapter->gotcl_old = adapter->stats.gotcl;
2511 
2512 	e1000_update_adaptive(hw);
2513 
2514 	if (!netif_carrier_ok(netdev)) {
2515 		if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2516 			/* We've lost link, so the controller stops DMA,
2517 			 * but we've got queued Tx work that's never going
2518 			 * to get done, so reset controller to flush Tx.
2519 			 * (Do the reset outside of interrupt context).
2520 			 */
2521 			adapter->tx_timeout_count++;
2522 			schedule_work(&adapter->reset_task);
2523 			/* exit immediately since reset is imminent */
2524 			goto unlock;
2525 		}
2526 	}
2527 
2528 	/* Simple mode for Interrupt Throttle Rate (ITR) */
2529 	if (hw->mac_type >= e1000_82540 && adapter->itr_setting == 4) {
2530 		/* Symmetric Tx/Rx gets a reduced ITR=2000;
2531 		 * Total asymmetrical Tx or Rx gets ITR=8000;
2532 		 * everyone else is between 2000-8000.
2533 		 */
2534 		u32 goc = (adapter->gotcl + adapter->gorcl) / 10000;
2535 		u32 dif = (adapter->gotcl > adapter->gorcl ?
2536 			    adapter->gotcl - adapter->gorcl :
2537 			    adapter->gorcl - adapter->gotcl) / 10000;
2538 		u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
2539 
2540 		ew32(ITR, 1000000000 / (itr * 256));
2541 	}
2542 
2543 	/* Cause software interrupt to ensure rx ring is cleaned */
2544 	ew32(ICS, E1000_ICS_RXDMT0);
2545 
2546 	/* Force detection of hung controller every watchdog period */
2547 	adapter->detect_tx_hung = true;
2548 
2549 	/* Reschedule the task */
2550 	if (!test_bit(__E1000_DOWN, &adapter->flags))
2551 		schedule_delayed_work(&adapter->watchdog_task, 2 * HZ);
2552 
2553 unlock:
2554 	mutex_unlock(&adapter->mutex);
2555 }
2556 
2557 enum latency_range {
2558 	lowest_latency = 0,
2559 	low_latency = 1,
2560 	bulk_latency = 2,
2561 	latency_invalid = 255
2562 };
2563 
2564 /**
2565  * e1000_update_itr - update the dynamic ITR value based on statistics
2566  * @adapter: pointer to adapter
2567  * @itr_setting: current adapter->itr
2568  * @packets: the number of packets during this measurement interval
2569  * @bytes: the number of bytes during this measurement interval
2570  *
2571  *      Stores a new ITR value based on packets and byte
2572  *      counts during the last interrupt.  The advantage of per interrupt
2573  *      computation is faster updates and more accurate ITR for the current
2574  *      traffic pattern.  Constants in this function were computed
2575  *      based on theoretical maximum wire speed and thresholds were set based
2576  *      on testing data as well as attempting to minimize response time
2577  *      while increasing bulk throughput.
2578  *      this functionality is controlled by the InterruptThrottleRate module
2579  *      parameter (see e1000_param.c)
2580  **/
2581 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2582 				     u16 itr_setting, int packets, int bytes)
2583 {
2584 	unsigned int retval = itr_setting;
2585 	struct e1000_hw *hw = &adapter->hw;
2586 
2587 	if (unlikely(hw->mac_type < e1000_82540))
2588 		goto update_itr_done;
2589 
2590 	if (packets == 0)
2591 		goto update_itr_done;
2592 
2593 	switch (itr_setting) {
2594 	case lowest_latency:
2595 		/* jumbo frames get bulk treatment*/
2596 		if (bytes/packets > 8000)
2597 			retval = bulk_latency;
2598 		else if ((packets < 5) && (bytes > 512))
2599 			retval = low_latency;
2600 		break;
2601 	case low_latency:  /* 50 usec aka 20000 ints/s */
2602 		if (bytes > 10000) {
2603 			/* jumbo frames need bulk latency setting */
2604 			if (bytes/packets > 8000)
2605 				retval = bulk_latency;
2606 			else if ((packets < 10) || ((bytes/packets) > 1200))
2607 				retval = bulk_latency;
2608 			else if ((packets > 35))
2609 				retval = lowest_latency;
2610 		} else if (bytes/packets > 2000)
2611 			retval = bulk_latency;
2612 		else if (packets <= 2 && bytes < 512)
2613 			retval = lowest_latency;
2614 		break;
2615 	case bulk_latency: /* 250 usec aka 4000 ints/s */
2616 		if (bytes > 25000) {
2617 			if (packets > 35)
2618 				retval = low_latency;
2619 		} else if (bytes < 6000) {
2620 			retval = low_latency;
2621 		}
2622 		break;
2623 	}
2624 
2625 update_itr_done:
2626 	return retval;
2627 }
2628 
2629 static void e1000_set_itr(struct e1000_adapter *adapter)
2630 {
2631 	struct e1000_hw *hw = &adapter->hw;
2632 	u16 current_itr;
2633 	u32 new_itr = adapter->itr;
2634 
2635 	if (unlikely(hw->mac_type < e1000_82540))
2636 		return;
2637 
2638 	/* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2639 	if (unlikely(adapter->link_speed != SPEED_1000)) {
2640 		current_itr = 0;
2641 		new_itr = 4000;
2642 		goto set_itr_now;
2643 	}
2644 
2645 	adapter->tx_itr = e1000_update_itr(adapter, adapter->tx_itr,
2646 					   adapter->total_tx_packets,
2647 					   adapter->total_tx_bytes);
2648 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
2649 	if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2650 		adapter->tx_itr = low_latency;
2651 
2652 	adapter->rx_itr = e1000_update_itr(adapter, adapter->rx_itr,
2653 					   adapter->total_rx_packets,
2654 					   adapter->total_rx_bytes);
2655 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
2656 	if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2657 		adapter->rx_itr = low_latency;
2658 
2659 	current_itr = max(adapter->rx_itr, adapter->tx_itr);
2660 
2661 	switch (current_itr) {
2662 	/* counts and packets in update_itr are dependent on these numbers */
2663 	case lowest_latency:
2664 		new_itr = 70000;
2665 		break;
2666 	case low_latency:
2667 		new_itr = 20000; /* aka hwitr = ~200 */
2668 		break;
2669 	case bulk_latency:
2670 		new_itr = 4000;
2671 		break;
2672 	default:
2673 		break;
2674 	}
2675 
2676 set_itr_now:
2677 	if (new_itr != adapter->itr) {
2678 		/* this attempts to bias the interrupt rate towards Bulk
2679 		 * by adding intermediate steps when interrupt rate is
2680 		 * increasing
2681 		 */
2682 		new_itr = new_itr > adapter->itr ?
2683 			  min(adapter->itr + (new_itr >> 2), new_itr) :
2684 			  new_itr;
2685 		adapter->itr = new_itr;
2686 		ew32(ITR, 1000000000 / (new_itr * 256));
2687 	}
2688 }
2689 
2690 #define E1000_TX_FLAGS_CSUM		0x00000001
2691 #define E1000_TX_FLAGS_VLAN		0x00000002
2692 #define E1000_TX_FLAGS_TSO		0x00000004
2693 #define E1000_TX_FLAGS_IPV4		0x00000008
2694 #define E1000_TX_FLAGS_NO_FCS		0x00000010
2695 #define E1000_TX_FLAGS_VLAN_MASK	0xffff0000
2696 #define E1000_TX_FLAGS_VLAN_SHIFT	16
2697 
2698 static int e1000_tso(struct e1000_adapter *adapter,
2699 		     struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2700 {
2701 	struct e1000_context_desc *context_desc;
2702 	struct e1000_buffer *buffer_info;
2703 	unsigned int i;
2704 	u32 cmd_length = 0;
2705 	u16 ipcse = 0, tucse, mss;
2706 	u8 ipcss, ipcso, tucss, tucso, hdr_len;
2707 	int err;
2708 
2709 	if (skb_is_gso(skb)) {
2710 		if (skb_header_cloned(skb)) {
2711 			err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2712 			if (err)
2713 				return err;
2714 		}
2715 
2716 		hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2717 		mss = skb_shinfo(skb)->gso_size;
2718 		if (skb->protocol == htons(ETH_P_IP)) {
2719 			struct iphdr *iph = ip_hdr(skb);
2720 			iph->tot_len = 0;
2721 			iph->check = 0;
2722 			tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2723 								 iph->daddr, 0,
2724 								 IPPROTO_TCP,
2725 								 0);
2726 			cmd_length = E1000_TXD_CMD_IP;
2727 			ipcse = skb_transport_offset(skb) - 1;
2728 		} else if (skb->protocol == htons(ETH_P_IPV6)) {
2729 			ipv6_hdr(skb)->payload_len = 0;
2730 			tcp_hdr(skb)->check =
2731 				~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2732 						 &ipv6_hdr(skb)->daddr,
2733 						 0, IPPROTO_TCP, 0);
2734 			ipcse = 0;
2735 		}
2736 		ipcss = skb_network_offset(skb);
2737 		ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2738 		tucss = skb_transport_offset(skb);
2739 		tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2740 		tucse = 0;
2741 
2742 		cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2743 			       E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2744 
2745 		i = tx_ring->next_to_use;
2746 		context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2747 		buffer_info = &tx_ring->buffer_info[i];
2748 
2749 		context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2750 		context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2751 		context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2752 		context_desc->upper_setup.tcp_fields.tucss = tucss;
2753 		context_desc->upper_setup.tcp_fields.tucso = tucso;
2754 		context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2755 		context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2756 		context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2757 		context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2758 
2759 		buffer_info->time_stamp = jiffies;
2760 		buffer_info->next_to_watch = i;
2761 
2762 		if (++i == tx_ring->count) i = 0;
2763 		tx_ring->next_to_use = i;
2764 
2765 		return true;
2766 	}
2767 	return false;
2768 }
2769 
2770 static bool e1000_tx_csum(struct e1000_adapter *adapter,
2771 			  struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2772 {
2773 	struct e1000_context_desc *context_desc;
2774 	struct e1000_buffer *buffer_info;
2775 	unsigned int i;
2776 	u8 css;
2777 	u32 cmd_len = E1000_TXD_CMD_DEXT;
2778 
2779 	if (skb->ip_summed != CHECKSUM_PARTIAL)
2780 		return false;
2781 
2782 	switch (skb->protocol) {
2783 	case cpu_to_be16(ETH_P_IP):
2784 		if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2785 			cmd_len |= E1000_TXD_CMD_TCP;
2786 		break;
2787 	case cpu_to_be16(ETH_P_IPV6):
2788 		/* XXX not handling all IPV6 headers */
2789 		if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2790 			cmd_len |= E1000_TXD_CMD_TCP;
2791 		break;
2792 	default:
2793 		if (unlikely(net_ratelimit()))
2794 			e_warn(drv, "checksum_partial proto=%x!\n",
2795 			       skb->protocol);
2796 		break;
2797 	}
2798 
2799 	css = skb_checksum_start_offset(skb);
2800 
2801 	i = tx_ring->next_to_use;
2802 	buffer_info = &tx_ring->buffer_info[i];
2803 	context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2804 
2805 	context_desc->lower_setup.ip_config = 0;
2806 	context_desc->upper_setup.tcp_fields.tucss = css;
2807 	context_desc->upper_setup.tcp_fields.tucso =
2808 		css + skb->csum_offset;
2809 	context_desc->upper_setup.tcp_fields.tucse = 0;
2810 	context_desc->tcp_seg_setup.data = 0;
2811 	context_desc->cmd_and_length = cpu_to_le32(cmd_len);
2812 
2813 	buffer_info->time_stamp = jiffies;
2814 	buffer_info->next_to_watch = i;
2815 
2816 	if (unlikely(++i == tx_ring->count)) i = 0;
2817 	tx_ring->next_to_use = i;
2818 
2819 	return true;
2820 }
2821 
2822 #define E1000_MAX_TXD_PWR	12
2823 #define E1000_MAX_DATA_PER_TXD	(1<<E1000_MAX_TXD_PWR)
2824 
2825 static int e1000_tx_map(struct e1000_adapter *adapter,
2826 			struct e1000_tx_ring *tx_ring,
2827 			struct sk_buff *skb, unsigned int first,
2828 			unsigned int max_per_txd, unsigned int nr_frags,
2829 			unsigned int mss)
2830 {
2831 	struct e1000_hw *hw = &adapter->hw;
2832 	struct pci_dev *pdev = adapter->pdev;
2833 	struct e1000_buffer *buffer_info;
2834 	unsigned int len = skb_headlen(skb);
2835 	unsigned int offset = 0, size, count = 0, i;
2836 	unsigned int f, bytecount, segs;
2837 
2838 	i = tx_ring->next_to_use;
2839 
2840 	while (len) {
2841 		buffer_info = &tx_ring->buffer_info[i];
2842 		size = min(len, max_per_txd);
2843 		/* Workaround for Controller erratum --
2844 		 * descriptor for non-tso packet in a linear SKB that follows a
2845 		 * tso gets written back prematurely before the data is fully
2846 		 * DMA'd to the controller
2847 		 */
2848 		if (!skb->data_len && tx_ring->last_tx_tso &&
2849 		    !skb_is_gso(skb)) {
2850 			tx_ring->last_tx_tso = false;
2851 			size -= 4;
2852 		}
2853 
2854 		/* Workaround for premature desc write-backs
2855 		 * in TSO mode.  Append 4-byte sentinel desc
2856 		 */
2857 		if (unlikely(mss && !nr_frags && size == len && size > 8))
2858 			size -= 4;
2859 		/* work-around for errata 10 and it applies
2860 		 * to all controllers in PCI-X mode
2861 		 * The fix is to make sure that the first descriptor of a
2862 		 * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2863 		 */
2864 		if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
2865 		                (size > 2015) && count == 0))
2866 		        size = 2015;
2867 
2868 		/* Workaround for potential 82544 hang in PCI-X.  Avoid
2869 		 * terminating buffers within evenly-aligned dwords.
2870 		 */
2871 		if (unlikely(adapter->pcix_82544 &&
2872 		   !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2873 		   size > 4))
2874 			size -= 4;
2875 
2876 		buffer_info->length = size;
2877 		/* set time_stamp *before* dma to help avoid a possible race */
2878 		buffer_info->time_stamp = jiffies;
2879 		buffer_info->mapped_as_page = false;
2880 		buffer_info->dma = dma_map_single(&pdev->dev,
2881 						  skb->data + offset,
2882 						  size, DMA_TO_DEVICE);
2883 		if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2884 			goto dma_error;
2885 		buffer_info->next_to_watch = i;
2886 
2887 		len -= size;
2888 		offset += size;
2889 		count++;
2890 		if (len) {
2891 			i++;
2892 			if (unlikely(i == tx_ring->count))
2893 				i = 0;
2894 		}
2895 	}
2896 
2897 	for (f = 0; f < nr_frags; f++) {
2898 		const struct skb_frag_struct *frag;
2899 
2900 		frag = &skb_shinfo(skb)->frags[f];
2901 		len = skb_frag_size(frag);
2902 		offset = 0;
2903 
2904 		while (len) {
2905 			unsigned long bufend;
2906 			i++;
2907 			if (unlikely(i == tx_ring->count))
2908 				i = 0;
2909 
2910 			buffer_info = &tx_ring->buffer_info[i];
2911 			size = min(len, max_per_txd);
2912 			/* Workaround for premature desc write-backs
2913 			 * in TSO mode.  Append 4-byte sentinel desc
2914 			 */
2915 			if (unlikely(mss && f == (nr_frags-1) &&
2916 			    size == len && size > 8))
2917 				size -= 4;
2918 			/* Workaround for potential 82544 hang in PCI-X.
2919 			 * Avoid terminating buffers within evenly-aligned
2920 			 * dwords.
2921 			 */
2922 			bufend = (unsigned long)
2923 				page_to_phys(skb_frag_page(frag));
2924 			bufend += offset + size - 1;
2925 			if (unlikely(adapter->pcix_82544 &&
2926 				     !(bufend & 4) &&
2927 				     size > 4))
2928 				size -= 4;
2929 
2930 			buffer_info->length = size;
2931 			buffer_info->time_stamp = jiffies;
2932 			buffer_info->mapped_as_page = true;
2933 			buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
2934 						offset, size, DMA_TO_DEVICE);
2935 			if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2936 				goto dma_error;
2937 			buffer_info->next_to_watch = i;
2938 
2939 			len -= size;
2940 			offset += size;
2941 			count++;
2942 		}
2943 	}
2944 
2945 	segs = skb_shinfo(skb)->gso_segs ?: 1;
2946 	/* multiply data chunks by size of headers */
2947 	bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
2948 
2949 	tx_ring->buffer_info[i].skb = skb;
2950 	tx_ring->buffer_info[i].segs = segs;
2951 	tx_ring->buffer_info[i].bytecount = bytecount;
2952 	tx_ring->buffer_info[first].next_to_watch = i;
2953 
2954 	return count;
2955 
2956 dma_error:
2957 	dev_err(&pdev->dev, "TX DMA map failed\n");
2958 	buffer_info->dma = 0;
2959 	if (count)
2960 		count--;
2961 
2962 	while (count--) {
2963 		if (i==0)
2964 			i += tx_ring->count;
2965 		i--;
2966 		buffer_info = &tx_ring->buffer_info[i];
2967 		e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2968 	}
2969 
2970 	return 0;
2971 }
2972 
2973 static void e1000_tx_queue(struct e1000_adapter *adapter,
2974 			   struct e1000_tx_ring *tx_ring, int tx_flags,
2975 			   int count)
2976 {
2977 	struct e1000_hw *hw = &adapter->hw;
2978 	struct e1000_tx_desc *tx_desc = NULL;
2979 	struct e1000_buffer *buffer_info;
2980 	u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2981 	unsigned int i;
2982 
2983 	if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
2984 		txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2985 			     E1000_TXD_CMD_TSE;
2986 		txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2987 
2988 		if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
2989 			txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2990 	}
2991 
2992 	if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
2993 		txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
2994 		txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2995 	}
2996 
2997 	if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
2998 		txd_lower |= E1000_TXD_CMD_VLE;
2999 		txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3000 	}
3001 
3002 	if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
3003 		txd_lower &= ~(E1000_TXD_CMD_IFCS);
3004 
3005 	i = tx_ring->next_to_use;
3006 
3007 	while (count--) {
3008 		buffer_info = &tx_ring->buffer_info[i];
3009 		tx_desc = E1000_TX_DESC(*tx_ring, i);
3010 		tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3011 		tx_desc->lower.data =
3012 			cpu_to_le32(txd_lower | buffer_info->length);
3013 		tx_desc->upper.data = cpu_to_le32(txd_upper);
3014 		if (unlikely(++i == tx_ring->count)) i = 0;
3015 	}
3016 
3017 	tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3018 
3019 	/* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
3020 	if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
3021 		tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
3022 
3023 	/* Force memory writes to complete before letting h/w
3024 	 * know there are new descriptors to fetch.  (Only
3025 	 * applicable for weak-ordered memory model archs,
3026 	 * such as IA-64).
3027 	 */
3028 	wmb();
3029 
3030 	tx_ring->next_to_use = i;
3031 	writel(i, hw->hw_addr + tx_ring->tdt);
3032 	/* we need this if more than one processor can write to our tail
3033 	 * at a time, it synchronizes IO on IA64/Altix systems
3034 	 */
3035 	mmiowb();
3036 }
3037 
3038 /* 82547 workaround to avoid controller hang in half-duplex environment.
3039  * The workaround is to avoid queuing a large packet that would span
3040  * the internal Tx FIFO ring boundary by notifying the stack to resend
3041  * the packet at a later time.  This gives the Tx FIFO an opportunity to
3042  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
3043  * to the beginning of the Tx FIFO.
3044  */
3045 
3046 #define E1000_FIFO_HDR			0x10
3047 #define E1000_82547_PAD_LEN		0x3E0
3048 
3049 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
3050 				       struct sk_buff *skb)
3051 {
3052 	u32 fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
3053 	u32 skb_fifo_len = skb->len + E1000_FIFO_HDR;
3054 
3055 	skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
3056 
3057 	if (adapter->link_duplex != HALF_DUPLEX)
3058 		goto no_fifo_stall_required;
3059 
3060 	if (atomic_read(&adapter->tx_fifo_stall))
3061 		return 1;
3062 
3063 	if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
3064 		atomic_set(&adapter->tx_fifo_stall, 1);
3065 		return 1;
3066 	}
3067 
3068 no_fifo_stall_required:
3069 	adapter->tx_fifo_head += skb_fifo_len;
3070 	if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
3071 		adapter->tx_fifo_head -= adapter->tx_fifo_size;
3072 	return 0;
3073 }
3074 
3075 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3076 {
3077 	struct e1000_adapter *adapter = netdev_priv(netdev);
3078 	struct e1000_tx_ring *tx_ring = adapter->tx_ring;
3079 
3080 	netif_stop_queue(netdev);
3081 	/* Herbert's original patch had:
3082 	 *  smp_mb__after_netif_stop_queue();
3083 	 * but since that doesn't exist yet, just open code it.
3084 	 */
3085 	smp_mb();
3086 
3087 	/* We need to check again in a case another CPU has just
3088 	 * made room available.
3089 	 */
3090 	if (likely(E1000_DESC_UNUSED(tx_ring) < size))
3091 		return -EBUSY;
3092 
3093 	/* A reprieve! */
3094 	netif_start_queue(netdev);
3095 	++adapter->restart_queue;
3096 	return 0;
3097 }
3098 
3099 static int e1000_maybe_stop_tx(struct net_device *netdev,
3100 			       struct e1000_tx_ring *tx_ring, int size)
3101 {
3102 	if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
3103 		return 0;
3104 	return __e1000_maybe_stop_tx(netdev, size);
3105 }
3106 
3107 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3108 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
3109 				    struct net_device *netdev)
3110 {
3111 	struct e1000_adapter *adapter = netdev_priv(netdev);
3112 	struct e1000_hw *hw = &adapter->hw;
3113 	struct e1000_tx_ring *tx_ring;
3114 	unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
3115 	unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3116 	unsigned int tx_flags = 0;
3117 	unsigned int len = skb_headlen(skb);
3118 	unsigned int nr_frags;
3119 	unsigned int mss;
3120 	int count = 0;
3121 	int tso;
3122 	unsigned int f;
3123 
3124 	/* This goes back to the question of how to logically map a Tx queue
3125 	 * to a flow.  Right now, performance is impacted slightly negatively
3126 	 * if using multiple Tx queues.  If the stack breaks away from a
3127 	 * single qdisc implementation, we can look at this again.
3128 	 */
3129 	tx_ring = adapter->tx_ring;
3130 
3131 	if (unlikely(skb->len <= 0)) {
3132 		dev_kfree_skb_any(skb);
3133 		return NETDEV_TX_OK;
3134 	}
3135 
3136 	/* On PCI/PCI-X HW, if packet size is less than ETH_ZLEN,
3137 	 * packets may get corrupted during padding by HW.
3138 	 * To WA this issue, pad all small packets manually.
3139 	 */
3140 	if (skb->len < ETH_ZLEN) {
3141 		if (skb_pad(skb, ETH_ZLEN - skb->len))
3142 			return NETDEV_TX_OK;
3143 		skb->len = ETH_ZLEN;
3144 		skb_set_tail_pointer(skb, ETH_ZLEN);
3145 	}
3146 
3147 	mss = skb_shinfo(skb)->gso_size;
3148 	/* The controller does a simple calculation to
3149 	 * make sure there is enough room in the FIFO before
3150 	 * initiating the DMA for each buffer.  The calc is:
3151 	 * 4 = ceil(buffer len/mss).  To make sure we don't
3152 	 * overrun the FIFO, adjust the max buffer len if mss
3153 	 * drops.
3154 	 */
3155 	if (mss) {
3156 		u8 hdr_len;
3157 		max_per_txd = min(mss << 2, max_per_txd);
3158 		max_txd_pwr = fls(max_per_txd) - 1;
3159 
3160 		hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3161 		if (skb->data_len && hdr_len == len) {
3162 			switch (hw->mac_type) {
3163 				unsigned int pull_size;
3164 			case e1000_82544:
3165 				/* Make sure we have room to chop off 4 bytes,
3166 				 * and that the end alignment will work out to
3167 				 * this hardware's requirements
3168 				 * NOTE: this is a TSO only workaround
3169 				 * if end byte alignment not correct move us
3170 				 * into the next dword
3171 				 */
3172 				if ((unsigned long)(skb_tail_pointer(skb) - 1)
3173 				    & 4)
3174 					break;
3175 				/* fall through */
3176 				pull_size = min((unsigned int)4, skb->data_len);
3177 				if (!__pskb_pull_tail(skb, pull_size)) {
3178 					e_err(drv, "__pskb_pull_tail "
3179 					      "failed.\n");
3180 					dev_kfree_skb_any(skb);
3181 					return NETDEV_TX_OK;
3182 				}
3183 				len = skb_headlen(skb);
3184 				break;
3185 			default:
3186 				/* do nothing */
3187 				break;
3188 			}
3189 		}
3190 	}
3191 
3192 	/* reserve a descriptor for the offload context */
3193 	if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3194 		count++;
3195 	count++;
3196 
3197 	/* Controller Erratum workaround */
3198 	if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3199 		count++;
3200 
3201 	count += TXD_USE_COUNT(len, max_txd_pwr);
3202 
3203 	if (adapter->pcix_82544)
3204 		count++;
3205 
3206 	/* work-around for errata 10 and it applies to all controllers
3207 	 * in PCI-X mode, so add one more descriptor to the count
3208 	 */
3209 	if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
3210 			(len > 2015)))
3211 		count++;
3212 
3213 	nr_frags = skb_shinfo(skb)->nr_frags;
3214 	for (f = 0; f < nr_frags; f++)
3215 		count += TXD_USE_COUNT(skb_frag_size(&skb_shinfo(skb)->frags[f]),
3216 				       max_txd_pwr);
3217 	if (adapter->pcix_82544)
3218 		count += nr_frags;
3219 
3220 	/* need: count + 2 desc gap to keep tail from touching
3221 	 * head, otherwise try next time
3222 	 */
3223 	if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2)))
3224 		return NETDEV_TX_BUSY;
3225 
3226 	if (unlikely((hw->mac_type == e1000_82547) &&
3227 		     (e1000_82547_fifo_workaround(adapter, skb)))) {
3228 		netif_stop_queue(netdev);
3229 		if (!test_bit(__E1000_DOWN, &adapter->flags))
3230 			schedule_delayed_work(&adapter->fifo_stall_task, 1);
3231 		return NETDEV_TX_BUSY;
3232 	}
3233 
3234 	if (vlan_tx_tag_present(skb)) {
3235 		tx_flags |= E1000_TX_FLAGS_VLAN;
3236 		tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3237 	}
3238 
3239 	first = tx_ring->next_to_use;
3240 
3241 	tso = e1000_tso(adapter, tx_ring, skb);
3242 	if (tso < 0) {
3243 		dev_kfree_skb_any(skb);
3244 		return NETDEV_TX_OK;
3245 	}
3246 
3247 	if (likely(tso)) {
3248 		if (likely(hw->mac_type != e1000_82544))
3249 			tx_ring->last_tx_tso = true;
3250 		tx_flags |= E1000_TX_FLAGS_TSO;
3251 	} else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3252 		tx_flags |= E1000_TX_FLAGS_CSUM;
3253 
3254 	if (likely(skb->protocol == htons(ETH_P_IP)))
3255 		tx_flags |= E1000_TX_FLAGS_IPV4;
3256 
3257 	if (unlikely(skb->no_fcs))
3258 		tx_flags |= E1000_TX_FLAGS_NO_FCS;
3259 
3260 	count = e1000_tx_map(adapter, tx_ring, skb, first, max_per_txd,
3261 			     nr_frags, mss);
3262 
3263 	if (count) {
3264 		netdev_sent_queue(netdev, skb->len);
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 	/* transmit dump */
3364 	pr_info("TX Desc ring0 dump\n");
3365 
3366 	/* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
3367 	 *
3368 	 * Legacy Transmit Descriptor
3369 	 *   +--------------------------------------------------------------+
3370 	 * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
3371 	 *   +--------------------------------------------------------------+
3372 	 * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
3373 	 *   +--------------------------------------------------------------+
3374 	 *   63       48 47        36 35    32 31     24 23    16 15        0
3375 	 *
3376 	 * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
3377 	 *   63      48 47    40 39       32 31             16 15    8 7      0
3378 	 *   +----------------------------------------------------------------+
3379 	 * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
3380 	 *   +----------------------------------------------------------------+
3381 	 * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
3382 	 *   +----------------------------------------------------------------+
3383 	 *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
3384 	 *
3385 	 * Extended Data Descriptor (DTYP=0x1)
3386 	 *   +----------------------------------------------------------------+
3387 	 * 0 |                     Buffer Address [63:0]                      |
3388 	 *   +----------------------------------------------------------------+
3389 	 * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
3390 	 *   +----------------------------------------------------------------+
3391 	 *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
3392 	 */
3393 	pr_info("Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma       ] leng  ntw timestmp         bi->skb\n");
3394 	pr_info("Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen] [bi->dma       ] leng  ntw timestmp         bi->skb\n");
3395 
3396 	if (!netif_msg_tx_done(adapter))
3397 		goto rx_ring_summary;
3398 
3399 	for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
3400 		struct e1000_tx_desc *tx_desc = E1000_TX_DESC(*tx_ring, i);
3401 		struct e1000_buffer *buffer_info = &tx_ring->buffer_info[i];
3402 		struct my_u { __le64 a; __le64 b; };
3403 		struct my_u *u = (struct my_u *)tx_desc;
3404 		const char *type;
3405 
3406 		if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
3407 			type = "NTC/U";
3408 		else if (i == tx_ring->next_to_use)
3409 			type = "NTU";
3410 		else if (i == tx_ring->next_to_clean)
3411 			type = "NTC";
3412 		else
3413 			type = "";
3414 
3415 		pr_info("T%c[0x%03X]    %016llX %016llX %016llX %04X  %3X %016llX %p %s\n",
3416 			((le64_to_cpu(u->b) & (1<<20)) ? 'd' : 'c'), i,
3417 			le64_to_cpu(u->a), le64_to_cpu(u->b),
3418 			(u64)buffer_info->dma, buffer_info->length,
3419 			buffer_info->next_to_watch,
3420 			(u64)buffer_info->time_stamp, buffer_info->skb, type);
3421 	}
3422 
3423 rx_ring_summary:
3424 	/* receive dump */
3425 	pr_info("\nRX Desc ring dump\n");
3426 
3427 	/* Legacy Receive Descriptor Format
3428 	 *
3429 	 * +-----------------------------------------------------+
3430 	 * |                Buffer Address [63:0]                |
3431 	 * +-----------------------------------------------------+
3432 	 * | VLAN Tag | Errors | Status 0 | Packet csum | Length |
3433 	 * +-----------------------------------------------------+
3434 	 * 63       48 47    40 39      32 31         16 15      0
3435 	 */
3436 	pr_info("R[desc]      [address 63:0  ] [vl er S cks ln] [bi->dma       ] [bi->skb]\n");
3437 
3438 	if (!netif_msg_rx_status(adapter))
3439 		goto exit;
3440 
3441 	for (i = 0; rx_ring->desc && (i < rx_ring->count); i++) {
3442 		struct e1000_rx_desc *rx_desc = E1000_RX_DESC(*rx_ring, i);
3443 		struct e1000_buffer *buffer_info = &rx_ring->buffer_info[i];
3444 		struct my_u { __le64 a; __le64 b; };
3445 		struct my_u *u = (struct my_u *)rx_desc;
3446 		const char *type;
3447 
3448 		if (i == rx_ring->next_to_use)
3449 			type = "NTU";
3450 		else if (i == rx_ring->next_to_clean)
3451 			type = "NTC";
3452 		else
3453 			type = "";
3454 
3455 		pr_info("R[0x%03X]     %016llX %016llX %016llX %p %s\n",
3456 			i, le64_to_cpu(u->a), le64_to_cpu(u->b),
3457 			(u64)buffer_info->dma, buffer_info->skb, type);
3458 	} /* for */
3459 
3460 	/* dump the descriptor caches */
3461 	/* rx */
3462 	pr_info("Rx descriptor cache in 64bit format\n");
3463 	for (i = 0x6000; i <= 0x63FF ; i += 0x10) {
3464 		pr_info("R%04X: %08X|%08X %08X|%08X\n",
3465 			i,
3466 			readl(adapter->hw.hw_addr + i+4),
3467 			readl(adapter->hw.hw_addr + i),
3468 			readl(adapter->hw.hw_addr + i+12),
3469 			readl(adapter->hw.hw_addr + i+8));
3470 	}
3471 	/* tx */
3472 	pr_info("Tx descriptor cache in 64bit format\n");
3473 	for (i = 0x7000; i <= 0x73FF ; i += 0x10) {
3474 		pr_info("T%04X: %08X|%08X %08X|%08X\n",
3475 			i,
3476 			readl(adapter->hw.hw_addr + i+4),
3477 			readl(adapter->hw.hw_addr + i),
3478 			readl(adapter->hw.hw_addr + i+12),
3479 			readl(adapter->hw.hw_addr + i+8));
3480 	}
3481 exit:
3482 	return;
3483 }
3484 
3485 /**
3486  * e1000_tx_timeout - Respond to a Tx Hang
3487  * @netdev: network interface device structure
3488  **/
3489 static void e1000_tx_timeout(struct net_device *netdev)
3490 {
3491 	struct e1000_adapter *adapter = netdev_priv(netdev);
3492 
3493 	/* Do the reset outside of interrupt context */
3494 	adapter->tx_timeout_count++;
3495 	schedule_work(&adapter->reset_task);
3496 }
3497 
3498 static void e1000_reset_task(struct work_struct *work)
3499 {
3500 	struct e1000_adapter *adapter =
3501 		container_of(work, struct e1000_adapter, reset_task);
3502 
3503 	if (test_bit(__E1000_DOWN, &adapter->flags))
3504 		return;
3505 	e_err(drv, "Reset adapter\n");
3506 	e1000_reinit_safe(adapter);
3507 }
3508 
3509 /**
3510  * e1000_get_stats - Get System Network Statistics
3511  * @netdev: network interface device structure
3512  *
3513  * Returns the address of the device statistics structure.
3514  * The statistics are actually updated from the watchdog.
3515  **/
3516 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3517 {
3518 	/* only return the current stats */
3519 	return &netdev->stats;
3520 }
3521 
3522 /**
3523  * e1000_change_mtu - Change the Maximum Transfer Unit
3524  * @netdev: network interface device structure
3525  * @new_mtu: new value for maximum frame size
3526  *
3527  * Returns 0 on success, negative on failure
3528  **/
3529 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3530 {
3531 	struct e1000_adapter *adapter = netdev_priv(netdev);
3532 	struct e1000_hw *hw = &adapter->hw;
3533 	int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3534 
3535 	if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3536 	    (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3537 		e_err(probe, "Invalid MTU setting\n");
3538 		return -EINVAL;
3539 	}
3540 
3541 	/* Adapter-specific max frame size limits. */
3542 	switch (hw->mac_type) {
3543 	case e1000_undefined ... e1000_82542_rev2_1:
3544 		if (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)) {
3545 			e_err(probe, "Jumbo Frames not supported.\n");
3546 			return -EINVAL;
3547 		}
3548 		break;
3549 	default:
3550 		/* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3551 		break;
3552 	}
3553 
3554 	while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
3555 		msleep(1);
3556 	/* e1000_down has a dependency on max_frame_size */
3557 	hw->max_frame_size = max_frame;
3558 	if (netif_running(netdev))
3559 		e1000_down(adapter);
3560 
3561 	/* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3562 	 * means we reserve 2 more, this pushes us to allocate from the next
3563 	 * larger slab size.
3564 	 * i.e. RXBUFFER_2048 --> size-4096 slab
3565 	 * however with the new *_jumbo_rx* routines, jumbo receives will use
3566 	 * fragmented skbs
3567 	 */
3568 
3569 	if (max_frame <= E1000_RXBUFFER_2048)
3570 		adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3571 	else
3572 #if (PAGE_SIZE >= E1000_RXBUFFER_16384)
3573 		adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3574 #elif (PAGE_SIZE >= E1000_RXBUFFER_4096)
3575 		adapter->rx_buffer_len = PAGE_SIZE;
3576 #endif
3577 
3578 	/* adjust allocation if LPE protects us, and we aren't using SBP */
3579 	if (!hw->tbi_compatibility_on &&
3580 	    ((max_frame == (ETH_FRAME_LEN + ETH_FCS_LEN)) ||
3581 	     (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3582 		adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3583 
3584 	pr_info("%s changing MTU from %d to %d\n",
3585 		netdev->name, netdev->mtu, new_mtu);
3586 	netdev->mtu = new_mtu;
3587 
3588 	if (netif_running(netdev))
3589 		e1000_up(adapter);
3590 	else
3591 		e1000_reset(adapter);
3592 
3593 	clear_bit(__E1000_RESETTING, &adapter->flags);
3594 
3595 	return 0;
3596 }
3597 
3598 /**
3599  * e1000_update_stats - Update the board statistics counters
3600  * @adapter: board private structure
3601  **/
3602 void e1000_update_stats(struct e1000_adapter *adapter)
3603 {
3604 	struct net_device *netdev = adapter->netdev;
3605 	struct e1000_hw *hw = &adapter->hw;
3606 	struct pci_dev *pdev = adapter->pdev;
3607 	unsigned long flags;
3608 	u16 phy_tmp;
3609 
3610 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3611 
3612 	/* Prevent stats update while adapter is being reset, or if the pci
3613 	 * connection is down.
3614 	 */
3615 	if (adapter->link_speed == 0)
3616 		return;
3617 	if (pci_channel_offline(pdev))
3618 		return;
3619 
3620 	spin_lock_irqsave(&adapter->stats_lock, flags);
3621 
3622 	/* these counters are modified from e1000_tbi_adjust_stats,
3623 	 * called from the interrupt context, so they must only
3624 	 * be written while holding adapter->stats_lock
3625 	 */
3626 
3627 	adapter->stats.crcerrs += er32(CRCERRS);
3628 	adapter->stats.gprc += er32(GPRC);
3629 	adapter->stats.gorcl += er32(GORCL);
3630 	adapter->stats.gorch += er32(GORCH);
3631 	adapter->stats.bprc += er32(BPRC);
3632 	adapter->stats.mprc += er32(MPRC);
3633 	adapter->stats.roc += er32(ROC);
3634 
3635 	adapter->stats.prc64 += er32(PRC64);
3636 	adapter->stats.prc127 += er32(PRC127);
3637 	adapter->stats.prc255 += er32(PRC255);
3638 	adapter->stats.prc511 += er32(PRC511);
3639 	adapter->stats.prc1023 += er32(PRC1023);
3640 	adapter->stats.prc1522 += er32(PRC1522);
3641 
3642 	adapter->stats.symerrs += er32(SYMERRS);
3643 	adapter->stats.mpc += er32(MPC);
3644 	adapter->stats.scc += er32(SCC);
3645 	adapter->stats.ecol += er32(ECOL);
3646 	adapter->stats.mcc += er32(MCC);
3647 	adapter->stats.latecol += er32(LATECOL);
3648 	adapter->stats.dc += er32(DC);
3649 	adapter->stats.sec += er32(SEC);
3650 	adapter->stats.rlec += er32(RLEC);
3651 	adapter->stats.xonrxc += er32(XONRXC);
3652 	adapter->stats.xontxc += er32(XONTXC);
3653 	adapter->stats.xoffrxc += er32(XOFFRXC);
3654 	adapter->stats.xofftxc += er32(XOFFTXC);
3655 	adapter->stats.fcruc += er32(FCRUC);
3656 	adapter->stats.gptc += er32(GPTC);
3657 	adapter->stats.gotcl += er32(GOTCL);
3658 	adapter->stats.gotch += er32(GOTCH);
3659 	adapter->stats.rnbc += er32(RNBC);
3660 	adapter->stats.ruc += er32(RUC);
3661 	adapter->stats.rfc += er32(RFC);
3662 	adapter->stats.rjc += er32(RJC);
3663 	adapter->stats.torl += er32(TORL);
3664 	adapter->stats.torh += er32(TORH);
3665 	adapter->stats.totl += er32(TOTL);
3666 	adapter->stats.toth += er32(TOTH);
3667 	adapter->stats.tpr += er32(TPR);
3668 
3669 	adapter->stats.ptc64 += er32(PTC64);
3670 	adapter->stats.ptc127 += er32(PTC127);
3671 	adapter->stats.ptc255 += er32(PTC255);
3672 	adapter->stats.ptc511 += er32(PTC511);
3673 	adapter->stats.ptc1023 += er32(PTC1023);
3674 	adapter->stats.ptc1522 += er32(PTC1522);
3675 
3676 	adapter->stats.mptc += er32(MPTC);
3677 	adapter->stats.bptc += er32(BPTC);
3678 
3679 	/* used for adaptive IFS */
3680 
3681 	hw->tx_packet_delta = er32(TPT);
3682 	adapter->stats.tpt += hw->tx_packet_delta;
3683 	hw->collision_delta = er32(COLC);
3684 	adapter->stats.colc += hw->collision_delta;
3685 
3686 	if (hw->mac_type >= e1000_82543) {
3687 		adapter->stats.algnerrc += er32(ALGNERRC);
3688 		adapter->stats.rxerrc += er32(RXERRC);
3689 		adapter->stats.tncrs += er32(TNCRS);
3690 		adapter->stats.cexterr += er32(CEXTERR);
3691 		adapter->stats.tsctc += er32(TSCTC);
3692 		adapter->stats.tsctfc += er32(TSCTFC);
3693 	}
3694 
3695 	/* Fill out the OS statistics structure */
3696 	netdev->stats.multicast = adapter->stats.mprc;
3697 	netdev->stats.collisions = adapter->stats.colc;
3698 
3699 	/* Rx Errors */
3700 
3701 	/* RLEC on some newer hardware can be incorrect so build
3702 	 * our own version based on RUC and ROC
3703 	 */
3704 	netdev->stats.rx_errors = adapter->stats.rxerrc +
3705 		adapter->stats.crcerrs + adapter->stats.algnerrc +
3706 		adapter->stats.ruc + adapter->stats.roc +
3707 		adapter->stats.cexterr;
3708 	adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3709 	netdev->stats.rx_length_errors = adapter->stats.rlerrc;
3710 	netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3711 	netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3712 	netdev->stats.rx_missed_errors = adapter->stats.mpc;
3713 
3714 	/* Tx Errors */
3715 	adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3716 	netdev->stats.tx_errors = adapter->stats.txerrc;
3717 	netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3718 	netdev->stats.tx_window_errors = adapter->stats.latecol;
3719 	netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3720 	if (hw->bad_tx_carr_stats_fd &&
3721 	    adapter->link_duplex == FULL_DUPLEX) {
3722 		netdev->stats.tx_carrier_errors = 0;
3723 		adapter->stats.tncrs = 0;
3724 	}
3725 
3726 	/* Tx Dropped needs to be maintained elsewhere */
3727 
3728 	/* Phy Stats */
3729 	if (hw->media_type == e1000_media_type_copper) {
3730 		if ((adapter->link_speed == SPEED_1000) &&
3731 		   (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3732 			phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3733 			adapter->phy_stats.idle_errors += phy_tmp;
3734 		}
3735 
3736 		if ((hw->mac_type <= e1000_82546) &&
3737 		   (hw->phy_type == e1000_phy_m88) &&
3738 		   !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3739 			adapter->phy_stats.receive_errors += phy_tmp;
3740 	}
3741 
3742 	/* Management Stats */
3743 	if (hw->has_smbus) {
3744 		adapter->stats.mgptc += er32(MGTPTC);
3745 		adapter->stats.mgprc += er32(MGTPRC);
3746 		adapter->stats.mgpdc += er32(MGTPDC);
3747 	}
3748 
3749 	spin_unlock_irqrestore(&adapter->stats_lock, flags);
3750 }
3751 
3752 /**
3753  * e1000_intr - Interrupt Handler
3754  * @irq: interrupt number
3755  * @data: pointer to a network interface device structure
3756  **/
3757 static irqreturn_t e1000_intr(int irq, void *data)
3758 {
3759 	struct net_device *netdev = data;
3760 	struct e1000_adapter *adapter = netdev_priv(netdev);
3761 	struct e1000_hw *hw = &adapter->hw;
3762 	u32 icr = er32(ICR);
3763 
3764 	if (unlikely((!icr)))
3765 		return IRQ_NONE;  /* Not our interrupt */
3766 
3767 	/* we might have caused the interrupt, but the above
3768 	 * read cleared it, and just in case the driver is
3769 	 * down there is nothing to do so return handled
3770 	 */
3771 	if (unlikely(test_bit(__E1000_DOWN, &adapter->flags)))
3772 		return IRQ_HANDLED;
3773 
3774 	if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3775 		hw->get_link_status = 1;
3776 		/* guard against interrupt when we're going down */
3777 		if (!test_bit(__E1000_DOWN, &adapter->flags))
3778 			schedule_delayed_work(&adapter->watchdog_task, 1);
3779 	}
3780 
3781 	/* disable interrupts, without the synchronize_irq bit */
3782 	ew32(IMC, ~0);
3783 	E1000_WRITE_FLUSH();
3784 
3785 	if (likely(napi_schedule_prep(&adapter->napi))) {
3786 		adapter->total_tx_bytes = 0;
3787 		adapter->total_tx_packets = 0;
3788 		adapter->total_rx_bytes = 0;
3789 		adapter->total_rx_packets = 0;
3790 		__napi_schedule(&adapter->napi);
3791 	} else {
3792 		/* this really should not happen! if it does it is basically a
3793 		 * bug, but not a hard error, so enable ints and continue
3794 		 */
3795 		if (!test_bit(__E1000_DOWN, &adapter->flags))
3796 			e1000_irq_enable(adapter);
3797 	}
3798 
3799 	return IRQ_HANDLED;
3800 }
3801 
3802 /**
3803  * e1000_clean - NAPI Rx polling callback
3804  * @adapter: board private structure
3805  **/
3806 static int e1000_clean(struct napi_struct *napi, int budget)
3807 {
3808 	struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
3809 						     napi);
3810 	int tx_clean_complete = 0, work_done = 0;
3811 
3812 	tx_clean_complete = e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);
3813 
3814 	adapter->clean_rx(adapter, &adapter->rx_ring[0], &work_done, budget);
3815 
3816 	if (!tx_clean_complete)
3817 		work_done = budget;
3818 
3819 	/* If budget not fully consumed, exit the polling mode */
3820 	if (work_done < budget) {
3821 		if (likely(adapter->itr_setting & 3))
3822 			e1000_set_itr(adapter);
3823 		napi_complete(napi);
3824 		if (!test_bit(__E1000_DOWN, &adapter->flags))
3825 			e1000_irq_enable(adapter);
3826 	}
3827 
3828 	return work_done;
3829 }
3830 
3831 /**
3832  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3833  * @adapter: board private structure
3834  **/
3835 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
3836 			       struct e1000_tx_ring *tx_ring)
3837 {
3838 	struct e1000_hw *hw = &adapter->hw;
3839 	struct net_device *netdev = adapter->netdev;
3840 	struct e1000_tx_desc *tx_desc, *eop_desc;
3841 	struct e1000_buffer *buffer_info;
3842 	unsigned int i, eop;
3843 	unsigned int count = 0;
3844 	unsigned int total_tx_bytes=0, total_tx_packets=0;
3845 	unsigned int bytes_compl = 0, pkts_compl = 0;
3846 
3847 	i = tx_ring->next_to_clean;
3848 	eop = tx_ring->buffer_info[i].next_to_watch;
3849 	eop_desc = E1000_TX_DESC(*tx_ring, eop);
3850 
3851 	while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
3852 	       (count < tx_ring->count)) {
3853 		bool cleaned = false;
3854 		rmb();	/* read buffer_info after eop_desc */
3855 		for ( ; !cleaned; count++) {
3856 			tx_desc = E1000_TX_DESC(*tx_ring, i);
3857 			buffer_info = &tx_ring->buffer_info[i];
3858 			cleaned = (i == eop);
3859 
3860 			if (cleaned) {
3861 				total_tx_packets += buffer_info->segs;
3862 				total_tx_bytes += buffer_info->bytecount;
3863 				if (buffer_info->skb) {
3864 					bytes_compl += buffer_info->skb->len;
3865 					pkts_compl++;
3866 				}
3867 
3868 			}
3869 			e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3870 			tx_desc->upper.data = 0;
3871 
3872 			if (unlikely(++i == tx_ring->count)) i = 0;
3873 		}
3874 
3875 		eop = tx_ring->buffer_info[i].next_to_watch;
3876 		eop_desc = E1000_TX_DESC(*tx_ring, eop);
3877 	}
3878 
3879 	tx_ring->next_to_clean = i;
3880 
3881 	netdev_completed_queue(netdev, pkts_compl, bytes_compl);
3882 
3883 #define TX_WAKE_THRESHOLD 32
3884 	if (unlikely(count && netif_carrier_ok(netdev) &&
3885 		     E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
3886 		/* Make sure that anybody stopping the queue after this
3887 		 * sees the new next_to_clean.
3888 		 */
3889 		smp_mb();
3890 
3891 		if (netif_queue_stopped(netdev) &&
3892 		    !(test_bit(__E1000_DOWN, &adapter->flags))) {
3893 			netif_wake_queue(netdev);
3894 			++adapter->restart_queue;
3895 		}
3896 	}
3897 
3898 	if (adapter->detect_tx_hung) {
3899 		/* Detect a transmit hang in hardware, this serializes the
3900 		 * check with the clearing of time_stamp and movement of i
3901 		 */
3902 		adapter->detect_tx_hung = false;
3903 		if (tx_ring->buffer_info[eop].time_stamp &&
3904 		    time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
3905 			       (adapter->tx_timeout_factor * HZ)) &&
3906 		    !(er32(STATUS) & E1000_STATUS_TXOFF)) {
3907 
3908 			/* detected Tx unit hang */
3909 			e_err(drv, "Detected Tx Unit Hang\n"
3910 			      "  Tx Queue             <%lu>\n"
3911 			      "  TDH                  <%x>\n"
3912 			      "  TDT                  <%x>\n"
3913 			      "  next_to_use          <%x>\n"
3914 			      "  next_to_clean        <%x>\n"
3915 			      "buffer_info[next_to_clean]\n"
3916 			      "  time_stamp           <%lx>\n"
3917 			      "  next_to_watch        <%x>\n"
3918 			      "  jiffies              <%lx>\n"
3919 			      "  next_to_watch.status <%x>\n",
3920 				(unsigned long)((tx_ring - adapter->tx_ring) /
3921 					sizeof(struct e1000_tx_ring)),
3922 				readl(hw->hw_addr + tx_ring->tdh),
3923 				readl(hw->hw_addr + tx_ring->tdt),
3924 				tx_ring->next_to_use,
3925 				tx_ring->next_to_clean,
3926 				tx_ring->buffer_info[eop].time_stamp,
3927 				eop,
3928 				jiffies,
3929 				eop_desc->upper.fields.status);
3930 			e1000_dump(adapter);
3931 			netif_stop_queue(netdev);
3932 		}
3933 	}
3934 	adapter->total_tx_bytes += total_tx_bytes;
3935 	adapter->total_tx_packets += total_tx_packets;
3936 	netdev->stats.tx_bytes += total_tx_bytes;
3937 	netdev->stats.tx_packets += total_tx_packets;
3938 	return count < tx_ring->count;
3939 }
3940 
3941 /**
3942  * e1000_rx_checksum - Receive Checksum Offload for 82543
3943  * @adapter:     board private structure
3944  * @status_err:  receive descriptor status and error fields
3945  * @csum:        receive descriptor csum field
3946  * @sk_buff:     socket buffer with received data
3947  **/
3948 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
3949 			      u32 csum, struct sk_buff *skb)
3950 {
3951 	struct e1000_hw *hw = &adapter->hw;
3952 	u16 status = (u16)status_err;
3953 	u8 errors = (u8)(status_err >> 24);
3954 
3955 	skb_checksum_none_assert(skb);
3956 
3957 	/* 82543 or newer only */
3958 	if (unlikely(hw->mac_type < e1000_82543)) return;
3959 	/* Ignore Checksum bit is set */
3960 	if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
3961 	/* TCP/UDP checksum error bit is set */
3962 	if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
3963 		/* let the stack verify checksum errors */
3964 		adapter->hw_csum_err++;
3965 		return;
3966 	}
3967 	/* TCP/UDP Checksum has not been calculated */
3968 	if (!(status & E1000_RXD_STAT_TCPCS))
3969 		return;
3970 
3971 	/* It must be a TCP or UDP packet with a valid checksum */
3972 	if (likely(status & E1000_RXD_STAT_TCPCS)) {
3973 		/* TCP checksum is good */
3974 		skb->ip_summed = CHECKSUM_UNNECESSARY;
3975 	}
3976 	adapter->hw_csum_good++;
3977 }
3978 
3979 /**
3980  * e1000_consume_page - helper function
3981  **/
3982 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
3983 			       u16 length)
3984 {
3985 	bi->page = NULL;
3986 	skb->len += length;
3987 	skb->data_len += length;
3988 	skb->truesize += PAGE_SIZE;
3989 }
3990 
3991 /**
3992  * e1000_receive_skb - helper function to handle rx indications
3993  * @adapter: board private structure
3994  * @status: descriptor status field as written by hardware
3995  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3996  * @skb: pointer to sk_buff to be indicated to stack
3997  */
3998 static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
3999 			      __le16 vlan, struct sk_buff *skb)
4000 {
4001 	skb->protocol = eth_type_trans(skb, adapter->netdev);
4002 
4003 	if (status & E1000_RXD_STAT_VP) {
4004 		u16 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
4005 
4006 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
4007 	}
4008 	napi_gro_receive(&adapter->napi, skb);
4009 }
4010 
4011 /**
4012  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
4013  * @adapter: board private structure
4014  * @rx_ring: ring to clean
4015  * @work_done: amount of napi work completed this call
4016  * @work_to_do: max amount of work allowed for this call to do
4017  *
4018  * the return value indicates whether actual cleaning was done, there
4019  * is no guarantee that everything was cleaned
4020  */
4021 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
4022 				     struct e1000_rx_ring *rx_ring,
4023 				     int *work_done, int work_to_do)
4024 {
4025 	struct e1000_hw *hw = &adapter->hw;
4026 	struct net_device *netdev = adapter->netdev;
4027 	struct pci_dev *pdev = adapter->pdev;
4028 	struct e1000_rx_desc *rx_desc, *next_rxd;
4029 	struct e1000_buffer *buffer_info, *next_buffer;
4030 	unsigned long irq_flags;
4031 	u32 length;
4032 	unsigned int i;
4033 	int cleaned_count = 0;
4034 	bool cleaned = false;
4035 	unsigned int total_rx_bytes=0, total_rx_packets=0;
4036 
4037 	i = rx_ring->next_to_clean;
4038 	rx_desc = E1000_RX_DESC(*rx_ring, i);
4039 	buffer_info = &rx_ring->buffer_info[i];
4040 
4041 	while (rx_desc->status & E1000_RXD_STAT_DD) {
4042 		struct sk_buff *skb;
4043 		u8 status;
4044 
4045 		if (*work_done >= work_to_do)
4046 			break;
4047 		(*work_done)++;
4048 		rmb(); /* read descriptor and rx_buffer_info after status DD */
4049 
4050 		status = rx_desc->status;
4051 		skb = buffer_info->skb;
4052 		buffer_info->skb = NULL;
4053 
4054 		if (++i == rx_ring->count) i = 0;
4055 		next_rxd = E1000_RX_DESC(*rx_ring, i);
4056 		prefetch(next_rxd);
4057 
4058 		next_buffer = &rx_ring->buffer_info[i];
4059 
4060 		cleaned = true;
4061 		cleaned_count++;
4062 		dma_unmap_page(&pdev->dev, buffer_info->dma,
4063 			       buffer_info->length, DMA_FROM_DEVICE);
4064 		buffer_info->dma = 0;
4065 
4066 		length = le16_to_cpu(rx_desc->length);
4067 
4068 		/* errors is only valid for DD + EOP descriptors */
4069 		if (unlikely((status & E1000_RXD_STAT_EOP) &&
4070 		    (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
4071 			u8 *mapped;
4072 			u8 last_byte;
4073 
4074 			mapped = page_address(buffer_info->page);
4075 			last_byte = *(mapped + length - 1);
4076 			if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
4077 				       last_byte)) {
4078 				spin_lock_irqsave(&adapter->stats_lock,
4079 						  irq_flags);
4080 				e1000_tbi_adjust_stats(hw, &adapter->stats,
4081 						       length, mapped);
4082 				spin_unlock_irqrestore(&adapter->stats_lock,
4083 						       irq_flags);
4084 				length--;
4085 			} else {
4086 				if (netdev->features & NETIF_F_RXALL)
4087 					goto process_skb;
4088 				/* recycle both page and skb */
4089 				buffer_info->skb = skb;
4090 				/* an error means any chain goes out the window
4091 				 * too
4092 				 */
4093 				if (rx_ring->rx_skb_top)
4094 					dev_kfree_skb(rx_ring->rx_skb_top);
4095 				rx_ring->rx_skb_top = NULL;
4096 				goto next_desc;
4097 			}
4098 		}
4099 
4100 #define rxtop rx_ring->rx_skb_top
4101 process_skb:
4102 		if (!(status & E1000_RXD_STAT_EOP)) {
4103 			/* this descriptor is only the beginning (or middle) */
4104 			if (!rxtop) {
4105 				/* this is the beginning of a chain */
4106 				rxtop = skb;
4107 				skb_fill_page_desc(rxtop, 0, buffer_info->page,
4108 						   0, length);
4109 			} else {
4110 				/* this is the middle of a chain */
4111 				skb_fill_page_desc(rxtop,
4112 				    skb_shinfo(rxtop)->nr_frags,
4113 				    buffer_info->page, 0, length);
4114 				/* re-use the skb, only consumed the page */
4115 				buffer_info->skb = skb;
4116 			}
4117 			e1000_consume_page(buffer_info, rxtop, length);
4118 			goto next_desc;
4119 		} else {
4120 			if (rxtop) {
4121 				/* end of the chain */
4122 				skb_fill_page_desc(rxtop,
4123 				    skb_shinfo(rxtop)->nr_frags,
4124 				    buffer_info->page, 0, length);
4125 				/* re-use the current skb, we only consumed the
4126 				 * page
4127 				 */
4128 				buffer_info->skb = skb;
4129 				skb = rxtop;
4130 				rxtop = NULL;
4131 				e1000_consume_page(buffer_info, skb, length);
4132 			} else {
4133 				/* no chain, got EOP, this buf is the packet
4134 				 * copybreak to save the put_page/alloc_page
4135 				 */
4136 				if (length <= copybreak &&
4137 				    skb_tailroom(skb) >= length) {
4138 					u8 *vaddr;
4139 					vaddr = kmap_atomic(buffer_info->page);
4140 					memcpy(skb_tail_pointer(skb), vaddr,
4141 					       length);
4142 					kunmap_atomic(vaddr);
4143 					/* re-use the page, so don't erase
4144 					 * buffer_info->page
4145 					 */
4146 					skb_put(skb, length);
4147 				} else {
4148 					skb_fill_page_desc(skb, 0,
4149 							   buffer_info->page, 0,
4150 							   length);
4151 					e1000_consume_page(buffer_info, skb,
4152 							   length);
4153 				}
4154 			}
4155 		}
4156 
4157 		/* Receive Checksum Offload XXX recompute due to CRC strip? */
4158 		e1000_rx_checksum(adapter,
4159 				  (u32)(status) |
4160 				  ((u32)(rx_desc->errors) << 24),
4161 				  le16_to_cpu(rx_desc->csum), skb);
4162 
4163 		total_rx_bytes += (skb->len - 4); /* don't count FCS */
4164 		if (likely(!(netdev->features & NETIF_F_RXFCS)))
4165 			pskb_trim(skb, skb->len - 4);
4166 		total_rx_packets++;
4167 
4168 		/* eth type trans needs skb->data to point to something */
4169 		if (!pskb_may_pull(skb, ETH_HLEN)) {
4170 			e_err(drv, "pskb_may_pull failed.\n");
4171 			dev_kfree_skb(skb);
4172 			goto next_desc;
4173 		}
4174 
4175 		e1000_receive_skb(adapter, status, rx_desc->special, skb);
4176 
4177 next_desc:
4178 		rx_desc->status = 0;
4179 
4180 		/* return some buffers to hardware, one at a time is too slow */
4181 		if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4182 			adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4183 			cleaned_count = 0;
4184 		}
4185 
4186 		/* use prefetched values */
4187 		rx_desc = next_rxd;
4188 		buffer_info = next_buffer;
4189 	}
4190 	rx_ring->next_to_clean = i;
4191 
4192 	cleaned_count = E1000_DESC_UNUSED(rx_ring);
4193 	if (cleaned_count)
4194 		adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4195 
4196 	adapter->total_rx_packets += total_rx_packets;
4197 	adapter->total_rx_bytes += total_rx_bytes;
4198 	netdev->stats.rx_bytes += total_rx_bytes;
4199 	netdev->stats.rx_packets += total_rx_packets;
4200 	return cleaned;
4201 }
4202 
4203 /* this should improve performance for small packets with large amounts
4204  * of reassembly being done in the stack
4205  */
4206 static void e1000_check_copybreak(struct net_device *netdev,
4207 				 struct e1000_buffer *buffer_info,
4208 				 u32 length, struct sk_buff **skb)
4209 {
4210 	struct sk_buff *new_skb;
4211 
4212 	if (length > copybreak)
4213 		return;
4214 
4215 	new_skb = netdev_alloc_skb_ip_align(netdev, length);
4216 	if (!new_skb)
4217 		return;
4218 
4219 	skb_copy_to_linear_data_offset(new_skb, -NET_IP_ALIGN,
4220 				       (*skb)->data - NET_IP_ALIGN,
4221 				       length + NET_IP_ALIGN);
4222 	/* save the skb in buffer_info as good */
4223 	buffer_info->skb = *skb;
4224 	*skb = new_skb;
4225 }
4226 
4227 /**
4228  * e1000_clean_rx_irq - Send received data up the network stack; legacy
4229  * @adapter: board private structure
4230  * @rx_ring: ring to clean
4231  * @work_done: amount of napi work completed this call
4232  * @work_to_do: max amount of work allowed for this call to do
4233  */
4234 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
4235 			       struct e1000_rx_ring *rx_ring,
4236 			       int *work_done, int work_to_do)
4237 {
4238 	struct e1000_hw *hw = &adapter->hw;
4239 	struct net_device *netdev = adapter->netdev;
4240 	struct pci_dev *pdev = adapter->pdev;
4241 	struct e1000_rx_desc *rx_desc, *next_rxd;
4242 	struct e1000_buffer *buffer_info, *next_buffer;
4243 	unsigned long flags;
4244 	u32 length;
4245 	unsigned int i;
4246 	int cleaned_count = 0;
4247 	bool cleaned = false;
4248 	unsigned int total_rx_bytes=0, total_rx_packets=0;
4249 
4250 	i = rx_ring->next_to_clean;
4251 	rx_desc = E1000_RX_DESC(*rx_ring, i);
4252 	buffer_info = &rx_ring->buffer_info[i];
4253 
4254 	while (rx_desc->status & E1000_RXD_STAT_DD) {
4255 		struct sk_buff *skb;
4256 		u8 status;
4257 
4258 		if (*work_done >= work_to_do)
4259 			break;
4260 		(*work_done)++;
4261 		rmb(); /* read descriptor and rx_buffer_info after status DD */
4262 
4263 		status = rx_desc->status;
4264 		skb = buffer_info->skb;
4265 		buffer_info->skb = NULL;
4266 
4267 		prefetch(skb->data - NET_IP_ALIGN);
4268 
4269 		if (++i == rx_ring->count) i = 0;
4270 		next_rxd = E1000_RX_DESC(*rx_ring, i);
4271 		prefetch(next_rxd);
4272 
4273 		next_buffer = &rx_ring->buffer_info[i];
4274 
4275 		cleaned = true;
4276 		cleaned_count++;
4277 		dma_unmap_single(&pdev->dev, buffer_info->dma,
4278 				 buffer_info->length, DMA_FROM_DEVICE);
4279 		buffer_info->dma = 0;
4280 
4281 		length = le16_to_cpu(rx_desc->length);
4282 		/* !EOP means multiple descriptors were used to store a single
4283 		 * packet, if thats the case we need to toss it.  In fact, we
4284 		 * to toss every packet with the EOP bit clear and the next
4285 		 * frame that _does_ have the EOP bit set, as it is by
4286 		 * definition only a frame fragment
4287 		 */
4288 		if (unlikely(!(status & E1000_RXD_STAT_EOP)))
4289 			adapter->discarding = true;
4290 
4291 		if (adapter->discarding) {
4292 			/* All receives must fit into a single buffer */
4293 			e_dbg("Receive packet consumed multiple buffers\n");
4294 			/* recycle */
4295 			buffer_info->skb = skb;
4296 			if (status & E1000_RXD_STAT_EOP)
4297 				adapter->discarding = false;
4298 			goto next_desc;
4299 		}
4300 
4301 		if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
4302 			u8 last_byte = *(skb->data + length - 1);
4303 			if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
4304 				       last_byte)) {
4305 				spin_lock_irqsave(&adapter->stats_lock, flags);
4306 				e1000_tbi_adjust_stats(hw, &adapter->stats,
4307 						       length, skb->data);
4308 				spin_unlock_irqrestore(&adapter->stats_lock,
4309 						       flags);
4310 				length--;
4311 			} else {
4312 				if (netdev->features & NETIF_F_RXALL)
4313 					goto process_skb;
4314 				/* recycle */
4315 				buffer_info->skb = skb;
4316 				goto next_desc;
4317 			}
4318 		}
4319 
4320 process_skb:
4321 		total_rx_bytes += (length - 4); /* don't count FCS */
4322 		total_rx_packets++;
4323 
4324 		if (likely(!(netdev->features & NETIF_F_RXFCS)))
4325 			/* adjust length to remove Ethernet CRC, this must be
4326 			 * done after the TBI_ACCEPT workaround above
4327 			 */
4328 			length -= 4;
4329 
4330 		e1000_check_copybreak(netdev, buffer_info, length, &skb);
4331 
4332 		skb_put(skb, length);
4333 
4334 		/* Receive Checksum Offload */
4335 		e1000_rx_checksum(adapter,
4336 				  (u32)(status) |
4337 				  ((u32)(rx_desc->errors) << 24),
4338 				  le16_to_cpu(rx_desc->csum), skb);
4339 
4340 		e1000_receive_skb(adapter, status, rx_desc->special, skb);
4341 
4342 next_desc:
4343 		rx_desc->status = 0;
4344 
4345 		/* return some buffers to hardware, one at a time is too slow */
4346 		if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4347 			adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4348 			cleaned_count = 0;
4349 		}
4350 
4351 		/* use prefetched values */
4352 		rx_desc = next_rxd;
4353 		buffer_info = next_buffer;
4354 	}
4355 	rx_ring->next_to_clean = i;
4356 
4357 	cleaned_count = E1000_DESC_UNUSED(rx_ring);
4358 	if (cleaned_count)
4359 		adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4360 
4361 	adapter->total_rx_packets += total_rx_packets;
4362 	adapter->total_rx_bytes += total_rx_bytes;
4363 	netdev->stats.rx_bytes += total_rx_bytes;
4364 	netdev->stats.rx_packets += total_rx_packets;
4365 	return cleaned;
4366 }
4367 
4368 /**
4369  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
4370  * @adapter: address of board private structure
4371  * @rx_ring: pointer to receive ring structure
4372  * @cleaned_count: number of buffers to allocate this pass
4373  **/
4374 static void
4375 e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
4376 			     struct e1000_rx_ring *rx_ring, int cleaned_count)
4377 {
4378 	struct net_device *netdev = adapter->netdev;
4379 	struct pci_dev *pdev = adapter->pdev;
4380 	struct e1000_rx_desc *rx_desc;
4381 	struct e1000_buffer *buffer_info;
4382 	struct sk_buff *skb;
4383 	unsigned int i;
4384 	unsigned int bufsz = 256 - 16 /*for skb_reserve */ ;
4385 
4386 	i = rx_ring->next_to_use;
4387 	buffer_info = &rx_ring->buffer_info[i];
4388 
4389 	while (cleaned_count--) {
4390 		skb = buffer_info->skb;
4391 		if (skb) {
4392 			skb_trim(skb, 0);
4393 			goto check_page;
4394 		}
4395 
4396 		skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4397 		if (unlikely(!skb)) {
4398 			/* Better luck next round */
4399 			adapter->alloc_rx_buff_failed++;
4400 			break;
4401 		}
4402 
4403 		buffer_info->skb = skb;
4404 		buffer_info->length = adapter->rx_buffer_len;
4405 check_page:
4406 		/* allocate a new page if necessary */
4407 		if (!buffer_info->page) {
4408 			buffer_info->page = alloc_page(GFP_ATOMIC);
4409 			if (unlikely(!buffer_info->page)) {
4410 				adapter->alloc_rx_buff_failed++;
4411 				break;
4412 			}
4413 		}
4414 
4415 		if (!buffer_info->dma) {
4416 			buffer_info->dma = dma_map_page(&pdev->dev,
4417 							buffer_info->page, 0,
4418 							buffer_info->length,
4419 							DMA_FROM_DEVICE);
4420 			if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4421 				put_page(buffer_info->page);
4422 				dev_kfree_skb(skb);
4423 				buffer_info->page = NULL;
4424 				buffer_info->skb = NULL;
4425 				buffer_info->dma = 0;
4426 				adapter->alloc_rx_buff_failed++;
4427 				break; /* while !buffer_info->skb */
4428 			}
4429 		}
4430 
4431 		rx_desc = E1000_RX_DESC(*rx_ring, i);
4432 		rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4433 
4434 		if (unlikely(++i == rx_ring->count))
4435 			i = 0;
4436 		buffer_info = &rx_ring->buffer_info[i];
4437 	}
4438 
4439 	if (likely(rx_ring->next_to_use != i)) {
4440 		rx_ring->next_to_use = i;
4441 		if (unlikely(i-- == 0))
4442 			i = (rx_ring->count - 1);
4443 
4444 		/* Force memory writes to complete before letting h/w
4445 		 * know there are new descriptors to fetch.  (Only
4446 		 * applicable for weak-ordered memory model archs,
4447 		 * such as IA-64).
4448 		 */
4449 		wmb();
4450 		writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4451 	}
4452 }
4453 
4454 /**
4455  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4456  * @adapter: address of board private structure
4457  **/
4458 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4459 				   struct e1000_rx_ring *rx_ring,
4460 				   int cleaned_count)
4461 {
4462 	struct e1000_hw *hw = &adapter->hw;
4463 	struct net_device *netdev = adapter->netdev;
4464 	struct pci_dev *pdev = adapter->pdev;
4465 	struct e1000_rx_desc *rx_desc;
4466 	struct e1000_buffer *buffer_info;
4467 	struct sk_buff *skb;
4468 	unsigned int i;
4469 	unsigned int bufsz = adapter->rx_buffer_len;
4470 
4471 	i = rx_ring->next_to_use;
4472 	buffer_info = &rx_ring->buffer_info[i];
4473 
4474 	while (cleaned_count--) {
4475 		skb = buffer_info->skb;
4476 		if (skb) {
4477 			skb_trim(skb, 0);
4478 			goto map_skb;
4479 		}
4480 
4481 		skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4482 		if (unlikely(!skb)) {
4483 			/* Better luck next round */
4484 			adapter->alloc_rx_buff_failed++;
4485 			break;
4486 		}
4487 
4488 		/* Fix for errata 23, can't cross 64kB boundary */
4489 		if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4490 			struct sk_buff *oldskb = skb;
4491 			e_err(rx_err, "skb align check failed: %u bytes at "
4492 			      "%p\n", bufsz, skb->data);
4493 			/* Try again, without freeing the previous */
4494 			skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4495 			/* Failed allocation, critical failure */
4496 			if (!skb) {
4497 				dev_kfree_skb(oldskb);
4498 				adapter->alloc_rx_buff_failed++;
4499 				break;
4500 			}
4501 
4502 			if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4503 				/* give up */
4504 				dev_kfree_skb(skb);
4505 				dev_kfree_skb(oldskb);
4506 				adapter->alloc_rx_buff_failed++;
4507 				break; /* while !buffer_info->skb */
4508 			}
4509 
4510 			/* Use new allocation */
4511 			dev_kfree_skb(oldskb);
4512 		}
4513 		buffer_info->skb = skb;
4514 		buffer_info->length = adapter->rx_buffer_len;
4515 map_skb:
4516 		buffer_info->dma = dma_map_single(&pdev->dev,
4517 						  skb->data,
4518 						  buffer_info->length,
4519 						  DMA_FROM_DEVICE);
4520 		if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4521 			dev_kfree_skb(skb);
4522 			buffer_info->skb = NULL;
4523 			buffer_info->dma = 0;
4524 			adapter->alloc_rx_buff_failed++;
4525 			break; /* while !buffer_info->skb */
4526 		}
4527 
4528 		/* XXX if it was allocated cleanly it will never map to a
4529 		 * boundary crossing
4530 		 */
4531 
4532 		/* Fix for errata 23, can't cross 64kB boundary */
4533 		if (!e1000_check_64k_bound(adapter,
4534 					(void *)(unsigned long)buffer_info->dma,
4535 					adapter->rx_buffer_len)) {
4536 			e_err(rx_err, "dma align check failed: %u bytes at "
4537 			      "%p\n", adapter->rx_buffer_len,
4538 			      (void *)(unsigned long)buffer_info->dma);
4539 			dev_kfree_skb(skb);
4540 			buffer_info->skb = NULL;
4541 
4542 			dma_unmap_single(&pdev->dev, buffer_info->dma,
4543 					 adapter->rx_buffer_len,
4544 					 DMA_FROM_DEVICE);
4545 			buffer_info->dma = 0;
4546 
4547 			adapter->alloc_rx_buff_failed++;
4548 			break; /* while !buffer_info->skb */
4549 		}
4550 		rx_desc = E1000_RX_DESC(*rx_ring, i);
4551 		rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4552 
4553 		if (unlikely(++i == rx_ring->count))
4554 			i = 0;
4555 		buffer_info = &rx_ring->buffer_info[i];
4556 	}
4557 
4558 	if (likely(rx_ring->next_to_use != i)) {
4559 		rx_ring->next_to_use = i;
4560 		if (unlikely(i-- == 0))
4561 			i = (rx_ring->count - 1);
4562 
4563 		/* Force memory writes to complete before letting h/w
4564 		 * know there are new descriptors to fetch.  (Only
4565 		 * applicable for weak-ordered memory model archs,
4566 		 * such as IA-64).
4567 		 */
4568 		wmb();
4569 		writel(i, hw->hw_addr + rx_ring->rdt);
4570 	}
4571 }
4572 
4573 /**
4574  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4575  * @adapter:
4576  **/
4577 static void e1000_smartspeed(struct e1000_adapter *adapter)
4578 {
4579 	struct e1000_hw *hw = &adapter->hw;
4580 	u16 phy_status;
4581 	u16 phy_ctrl;
4582 
4583 	if ((hw->phy_type != e1000_phy_igp) || !hw->autoneg ||
4584 	   !(hw->autoneg_advertised & ADVERTISE_1000_FULL))
4585 		return;
4586 
4587 	if (adapter->smartspeed == 0) {
4588 		/* If Master/Slave config fault is asserted twice,
4589 		 * we assume back-to-back
4590 		 */
4591 		e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4592 		if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4593 		e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4594 		if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4595 		e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4596 		if (phy_ctrl & CR_1000T_MS_ENABLE) {
4597 			phy_ctrl &= ~CR_1000T_MS_ENABLE;
4598 			e1000_write_phy_reg(hw, PHY_1000T_CTRL,
4599 					    phy_ctrl);
4600 			adapter->smartspeed++;
4601 			if (!e1000_phy_setup_autoneg(hw) &&
4602 			   !e1000_read_phy_reg(hw, PHY_CTRL,
4603 					       &phy_ctrl)) {
4604 				phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4605 					     MII_CR_RESTART_AUTO_NEG);
4606 				e1000_write_phy_reg(hw, PHY_CTRL,
4607 						    phy_ctrl);
4608 			}
4609 		}
4610 		return;
4611 	} else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4612 		/* If still no link, perhaps using 2/3 pair cable */
4613 		e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4614 		phy_ctrl |= CR_1000T_MS_ENABLE;
4615 		e1000_write_phy_reg(hw, PHY_1000T_CTRL, phy_ctrl);
4616 		if (!e1000_phy_setup_autoneg(hw) &&
4617 		   !e1000_read_phy_reg(hw, PHY_CTRL, &phy_ctrl)) {
4618 			phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4619 				     MII_CR_RESTART_AUTO_NEG);
4620 			e1000_write_phy_reg(hw, PHY_CTRL, phy_ctrl);
4621 		}
4622 	}
4623 	/* Restart process after E1000_SMARTSPEED_MAX iterations */
4624 	if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4625 		adapter->smartspeed = 0;
4626 }
4627 
4628 /**
4629  * e1000_ioctl -
4630  * @netdev:
4631  * @ifreq:
4632  * @cmd:
4633  **/
4634 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4635 {
4636 	switch (cmd) {
4637 	case SIOCGMIIPHY:
4638 	case SIOCGMIIREG:
4639 	case SIOCSMIIREG:
4640 		return e1000_mii_ioctl(netdev, ifr, cmd);
4641 	default:
4642 		return -EOPNOTSUPP;
4643 	}
4644 }
4645 
4646 /**
4647  * e1000_mii_ioctl -
4648  * @netdev:
4649  * @ifreq:
4650  * @cmd:
4651  **/
4652 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4653 			   int cmd)
4654 {
4655 	struct e1000_adapter *adapter = netdev_priv(netdev);
4656 	struct e1000_hw *hw = &adapter->hw;
4657 	struct mii_ioctl_data *data = if_mii(ifr);
4658 	int retval;
4659 	u16 mii_reg;
4660 	unsigned long flags;
4661 
4662 	if (hw->media_type != e1000_media_type_copper)
4663 		return -EOPNOTSUPP;
4664 
4665 	switch (cmd) {
4666 	case SIOCGMIIPHY:
4667 		data->phy_id = hw->phy_addr;
4668 		break;
4669 	case SIOCGMIIREG:
4670 		spin_lock_irqsave(&adapter->stats_lock, flags);
4671 		if (e1000_read_phy_reg(hw, data->reg_num & 0x1F,
4672 				   &data->val_out)) {
4673 			spin_unlock_irqrestore(&adapter->stats_lock, flags);
4674 			return -EIO;
4675 		}
4676 		spin_unlock_irqrestore(&adapter->stats_lock, flags);
4677 		break;
4678 	case SIOCSMIIREG:
4679 		if (data->reg_num & ~(0x1F))
4680 			return -EFAULT;
4681 		mii_reg = data->val_in;
4682 		spin_lock_irqsave(&adapter->stats_lock, flags);
4683 		if (e1000_write_phy_reg(hw, data->reg_num,
4684 					mii_reg)) {
4685 			spin_unlock_irqrestore(&adapter->stats_lock, flags);
4686 			return -EIO;
4687 		}
4688 		spin_unlock_irqrestore(&adapter->stats_lock, flags);
4689 		if (hw->media_type == e1000_media_type_copper) {
4690 			switch (data->reg_num) {
4691 			case PHY_CTRL:
4692 				if (mii_reg & MII_CR_POWER_DOWN)
4693 					break;
4694 				if (mii_reg & MII_CR_AUTO_NEG_EN) {
4695 					hw->autoneg = 1;
4696 					hw->autoneg_advertised = 0x2F;
4697 				} else {
4698 					u32 speed;
4699 					if (mii_reg & 0x40)
4700 						speed = SPEED_1000;
4701 					else if (mii_reg & 0x2000)
4702 						speed = SPEED_100;
4703 					else
4704 						speed = SPEED_10;
4705 					retval = e1000_set_spd_dplx(
4706 						adapter, speed,
4707 						((mii_reg & 0x100)
4708 						 ? DUPLEX_FULL :
4709 						 DUPLEX_HALF));
4710 					if (retval)
4711 						return retval;
4712 				}
4713 				if (netif_running(adapter->netdev))
4714 					e1000_reinit_locked(adapter);
4715 				else
4716 					e1000_reset(adapter);
4717 				break;
4718 			case M88E1000_PHY_SPEC_CTRL:
4719 			case M88E1000_EXT_PHY_SPEC_CTRL:
4720 				if (e1000_phy_reset(hw))
4721 					return -EIO;
4722 				break;
4723 			}
4724 		} else {
4725 			switch (data->reg_num) {
4726 			case PHY_CTRL:
4727 				if (mii_reg & MII_CR_POWER_DOWN)
4728 					break;
4729 				if (netif_running(adapter->netdev))
4730 					e1000_reinit_locked(adapter);
4731 				else
4732 					e1000_reset(adapter);
4733 				break;
4734 			}
4735 		}
4736 		break;
4737 	default:
4738 		return -EOPNOTSUPP;
4739 	}
4740 	return E1000_SUCCESS;
4741 }
4742 
4743 void e1000_pci_set_mwi(struct e1000_hw *hw)
4744 {
4745 	struct e1000_adapter *adapter = hw->back;
4746 	int ret_val = pci_set_mwi(adapter->pdev);
4747 
4748 	if (ret_val)
4749 		e_err(probe, "Error in setting MWI\n");
4750 }
4751 
4752 void e1000_pci_clear_mwi(struct e1000_hw *hw)
4753 {
4754 	struct e1000_adapter *adapter = hw->back;
4755 
4756 	pci_clear_mwi(adapter->pdev);
4757 }
4758 
4759 int e1000_pcix_get_mmrbc(struct e1000_hw *hw)
4760 {
4761 	struct e1000_adapter *adapter = hw->back;
4762 	return pcix_get_mmrbc(adapter->pdev);
4763 }
4764 
4765 void e1000_pcix_set_mmrbc(struct e1000_hw *hw, int mmrbc)
4766 {
4767 	struct e1000_adapter *adapter = hw->back;
4768 	pcix_set_mmrbc(adapter->pdev, mmrbc);
4769 }
4770 
4771 void e1000_io_write(struct e1000_hw *hw, unsigned long port, u32 value)
4772 {
4773 	outl(value, port);
4774 }
4775 
4776 static bool e1000_vlan_used(struct e1000_adapter *adapter)
4777 {
4778 	u16 vid;
4779 
4780 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
4781 		return true;
4782 	return false;
4783 }
4784 
4785 static void __e1000_vlan_mode(struct e1000_adapter *adapter,
4786 			      netdev_features_t features)
4787 {
4788 	struct e1000_hw *hw = &adapter->hw;
4789 	u32 ctrl;
4790 
4791 	ctrl = er32(CTRL);
4792 	if (features & NETIF_F_HW_VLAN_CTAG_RX) {
4793 		/* enable VLAN tag insert/strip */
4794 		ctrl |= E1000_CTRL_VME;
4795 	} else {
4796 		/* disable VLAN tag insert/strip */
4797 		ctrl &= ~E1000_CTRL_VME;
4798 	}
4799 	ew32(CTRL, ctrl);
4800 }
4801 static void e1000_vlan_filter_on_off(struct e1000_adapter *adapter,
4802 				     bool filter_on)
4803 {
4804 	struct e1000_hw *hw = &adapter->hw;
4805 	u32 rctl;
4806 
4807 	if (!test_bit(__E1000_DOWN, &adapter->flags))
4808 		e1000_irq_disable(adapter);
4809 
4810 	__e1000_vlan_mode(adapter, adapter->netdev->features);
4811 	if (filter_on) {
4812 		/* enable VLAN receive filtering */
4813 		rctl = er32(RCTL);
4814 		rctl &= ~E1000_RCTL_CFIEN;
4815 		if (!(adapter->netdev->flags & IFF_PROMISC))
4816 			rctl |= E1000_RCTL_VFE;
4817 		ew32(RCTL, rctl);
4818 		e1000_update_mng_vlan(adapter);
4819 	} else {
4820 		/* disable VLAN receive filtering */
4821 		rctl = er32(RCTL);
4822 		rctl &= ~E1000_RCTL_VFE;
4823 		ew32(RCTL, rctl);
4824 	}
4825 
4826 	if (!test_bit(__E1000_DOWN, &adapter->flags))
4827 		e1000_irq_enable(adapter);
4828 }
4829 
4830 static void e1000_vlan_mode(struct net_device *netdev,
4831 			    netdev_features_t features)
4832 {
4833 	struct e1000_adapter *adapter = netdev_priv(netdev);
4834 
4835 	if (!test_bit(__E1000_DOWN, &adapter->flags))
4836 		e1000_irq_disable(adapter);
4837 
4838 	__e1000_vlan_mode(adapter, features);
4839 
4840 	if (!test_bit(__E1000_DOWN, &adapter->flags))
4841 		e1000_irq_enable(adapter);
4842 }
4843 
4844 static int e1000_vlan_rx_add_vid(struct net_device *netdev,
4845 				 __be16 proto, u16 vid)
4846 {
4847 	struct e1000_adapter *adapter = netdev_priv(netdev);
4848 	struct e1000_hw *hw = &adapter->hw;
4849 	u32 vfta, index;
4850 
4851 	if ((hw->mng_cookie.status &
4852 	     E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4853 	    (vid == adapter->mng_vlan_id))
4854 		return 0;
4855 
4856 	if (!e1000_vlan_used(adapter))
4857 		e1000_vlan_filter_on_off(adapter, true);
4858 
4859 	/* add VID to filter table */
4860 	index = (vid >> 5) & 0x7F;
4861 	vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4862 	vfta |= (1 << (vid & 0x1F));
4863 	e1000_write_vfta(hw, index, vfta);
4864 
4865 	set_bit(vid, adapter->active_vlans);
4866 
4867 	return 0;
4868 }
4869 
4870 static int e1000_vlan_rx_kill_vid(struct net_device *netdev,
4871 				  __be16 proto, u16 vid)
4872 {
4873 	struct e1000_adapter *adapter = netdev_priv(netdev);
4874 	struct e1000_hw *hw = &adapter->hw;
4875 	u32 vfta, index;
4876 
4877 	if (!test_bit(__E1000_DOWN, &adapter->flags))
4878 		e1000_irq_disable(adapter);
4879 	if (!test_bit(__E1000_DOWN, &adapter->flags))
4880 		e1000_irq_enable(adapter);
4881 
4882 	/* remove VID from filter table */
4883 	index = (vid >> 5) & 0x7F;
4884 	vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4885 	vfta &= ~(1 << (vid & 0x1F));
4886 	e1000_write_vfta(hw, index, vfta);
4887 
4888 	clear_bit(vid, adapter->active_vlans);
4889 
4890 	if (!e1000_vlan_used(adapter))
4891 		e1000_vlan_filter_on_off(adapter, false);
4892 
4893 	return 0;
4894 }
4895 
4896 static void e1000_restore_vlan(struct e1000_adapter *adapter)
4897 {
4898 	u16 vid;
4899 
4900 	if (!e1000_vlan_used(adapter))
4901 		return;
4902 
4903 	e1000_vlan_filter_on_off(adapter, true);
4904 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
4905 		e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
4906 }
4907 
4908 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u32 spd, u8 dplx)
4909 {
4910 	struct e1000_hw *hw = &adapter->hw;
4911 
4912 	hw->autoneg = 0;
4913 
4914 	/* Make sure dplx is at most 1 bit and lsb of speed is not set
4915 	 * for the switch() below to work
4916 	 */
4917 	if ((spd & 1) || (dplx & ~1))
4918 		goto err_inval;
4919 
4920 	/* Fiber NICs only allow 1000 gbps Full duplex */
4921 	if ((hw->media_type == e1000_media_type_fiber) &&
4922 	    spd != SPEED_1000 &&
4923 	    dplx != DUPLEX_FULL)
4924 		goto err_inval;
4925 
4926 	switch (spd + dplx) {
4927 	case SPEED_10 + DUPLEX_HALF:
4928 		hw->forced_speed_duplex = e1000_10_half;
4929 		break;
4930 	case SPEED_10 + DUPLEX_FULL:
4931 		hw->forced_speed_duplex = e1000_10_full;
4932 		break;
4933 	case SPEED_100 + DUPLEX_HALF:
4934 		hw->forced_speed_duplex = e1000_100_half;
4935 		break;
4936 	case SPEED_100 + DUPLEX_FULL:
4937 		hw->forced_speed_duplex = e1000_100_full;
4938 		break;
4939 	case SPEED_1000 + DUPLEX_FULL:
4940 		hw->autoneg = 1;
4941 		hw->autoneg_advertised = ADVERTISE_1000_FULL;
4942 		break;
4943 	case SPEED_1000 + DUPLEX_HALF: /* not supported */
4944 	default:
4945 		goto err_inval;
4946 	}
4947 
4948 	/* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
4949 	hw->mdix = AUTO_ALL_MODES;
4950 
4951 	return 0;
4952 
4953 err_inval:
4954 	e_err(probe, "Unsupported Speed/Duplex configuration\n");
4955 	return -EINVAL;
4956 }
4957 
4958 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4959 {
4960 	struct net_device *netdev = pci_get_drvdata(pdev);
4961 	struct e1000_adapter *adapter = netdev_priv(netdev);
4962 	struct e1000_hw *hw = &adapter->hw;
4963 	u32 ctrl, ctrl_ext, rctl, status;
4964 	u32 wufc = adapter->wol;
4965 #ifdef CONFIG_PM
4966 	int retval = 0;
4967 #endif
4968 
4969 	netif_device_detach(netdev);
4970 
4971 	if (netif_running(netdev)) {
4972 		WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
4973 		e1000_down(adapter);
4974 	}
4975 
4976 #ifdef CONFIG_PM
4977 	retval = pci_save_state(pdev);
4978 	if (retval)
4979 		return retval;
4980 #endif
4981 
4982 	status = er32(STATUS);
4983 	if (status & E1000_STATUS_LU)
4984 		wufc &= ~E1000_WUFC_LNKC;
4985 
4986 	if (wufc) {
4987 		e1000_setup_rctl(adapter);
4988 		e1000_set_rx_mode(netdev);
4989 
4990 		rctl = er32(RCTL);
4991 
4992 		/* turn on all-multi mode if wake on multicast is enabled */
4993 		if (wufc & E1000_WUFC_MC)
4994 			rctl |= E1000_RCTL_MPE;
4995 
4996 		/* enable receives in the hardware */
4997 		ew32(RCTL, rctl | E1000_RCTL_EN);
4998 
4999 		if (hw->mac_type >= e1000_82540) {
5000 			ctrl = er32(CTRL);
5001 			/* advertise wake from D3Cold */
5002 			#define E1000_CTRL_ADVD3WUC 0x00100000
5003 			/* phy power management enable */
5004 			#define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5005 			ctrl |= E1000_CTRL_ADVD3WUC |
5006 				E1000_CTRL_EN_PHY_PWR_MGMT;
5007 			ew32(CTRL, ctrl);
5008 		}
5009 
5010 		if (hw->media_type == e1000_media_type_fiber ||
5011 		    hw->media_type == e1000_media_type_internal_serdes) {
5012 			/* keep the laser running in D3 */
5013 			ctrl_ext = er32(CTRL_EXT);
5014 			ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
5015 			ew32(CTRL_EXT, ctrl_ext);
5016 		}
5017 
5018 		ew32(WUC, E1000_WUC_PME_EN);
5019 		ew32(WUFC, wufc);
5020 	} else {
5021 		ew32(WUC, 0);
5022 		ew32(WUFC, 0);
5023 	}
5024 
5025 	e1000_release_manageability(adapter);
5026 
5027 	*enable_wake = !!wufc;
5028 
5029 	/* make sure adapter isn't asleep if manageability is enabled */
5030 	if (adapter->en_mng_pt)
5031 		*enable_wake = true;
5032 
5033 	if (netif_running(netdev))
5034 		e1000_free_irq(adapter);
5035 
5036 	pci_disable_device(pdev);
5037 
5038 	return 0;
5039 }
5040 
5041 #ifdef CONFIG_PM
5042 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
5043 {
5044 	int retval;
5045 	bool wake;
5046 
5047 	retval = __e1000_shutdown(pdev, &wake);
5048 	if (retval)
5049 		return retval;
5050 
5051 	if (wake) {
5052 		pci_prepare_to_sleep(pdev);
5053 	} else {
5054 		pci_wake_from_d3(pdev, false);
5055 		pci_set_power_state(pdev, PCI_D3hot);
5056 	}
5057 
5058 	return 0;
5059 }
5060 
5061 static int e1000_resume(struct pci_dev *pdev)
5062 {
5063 	struct net_device *netdev = pci_get_drvdata(pdev);
5064 	struct e1000_adapter *adapter = netdev_priv(netdev);
5065 	struct e1000_hw *hw = &adapter->hw;
5066 	u32 err;
5067 
5068 	pci_set_power_state(pdev, PCI_D0);
5069 	pci_restore_state(pdev);
5070 	pci_save_state(pdev);
5071 
5072 	if (adapter->need_ioport)
5073 		err = pci_enable_device(pdev);
5074 	else
5075 		err = pci_enable_device_mem(pdev);
5076 	if (err) {
5077 		pr_err("Cannot enable PCI device from suspend\n");
5078 		return err;
5079 	}
5080 	pci_set_master(pdev);
5081 
5082 	pci_enable_wake(pdev, PCI_D3hot, 0);
5083 	pci_enable_wake(pdev, PCI_D3cold, 0);
5084 
5085 	if (netif_running(netdev)) {
5086 		err = e1000_request_irq(adapter);
5087 		if (err)
5088 			return err;
5089 	}
5090 
5091 	e1000_power_up_phy(adapter);
5092 	e1000_reset(adapter);
5093 	ew32(WUS, ~0);
5094 
5095 	e1000_init_manageability(adapter);
5096 
5097 	if (netif_running(netdev))
5098 		e1000_up(adapter);
5099 
5100 	netif_device_attach(netdev);
5101 
5102 	return 0;
5103 }
5104 #endif
5105 
5106 static void e1000_shutdown(struct pci_dev *pdev)
5107 {
5108 	bool wake;
5109 
5110 	__e1000_shutdown(pdev, &wake);
5111 
5112 	if (system_state == SYSTEM_POWER_OFF) {
5113 		pci_wake_from_d3(pdev, wake);
5114 		pci_set_power_state(pdev, PCI_D3hot);
5115 	}
5116 }
5117 
5118 #ifdef CONFIG_NET_POLL_CONTROLLER
5119 /* Polling 'interrupt' - used by things like netconsole to send skbs
5120  * without having to re-enable interrupts. It's not called while
5121  * the interrupt routine is executing.
5122  */
5123 static void e1000_netpoll(struct net_device *netdev)
5124 {
5125 	struct e1000_adapter *adapter = netdev_priv(netdev);
5126 
5127 	disable_irq(adapter->pdev->irq);
5128 	e1000_intr(adapter->pdev->irq, netdev);
5129 	enable_irq(adapter->pdev->irq);
5130 }
5131 #endif
5132 
5133 /**
5134  * e1000_io_error_detected - called when PCI error is detected
5135  * @pdev: Pointer to PCI device
5136  * @state: The current pci connection state
5137  *
5138  * This function is called after a PCI bus error affecting
5139  * this device has been detected.
5140  */
5141 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5142 						pci_channel_state_t state)
5143 {
5144 	struct net_device *netdev = pci_get_drvdata(pdev);
5145 	struct e1000_adapter *adapter = netdev_priv(netdev);
5146 
5147 	netif_device_detach(netdev);
5148 
5149 	if (state == pci_channel_io_perm_failure)
5150 		return PCI_ERS_RESULT_DISCONNECT;
5151 
5152 	if (netif_running(netdev))
5153 		e1000_down(adapter);
5154 	pci_disable_device(pdev);
5155 
5156 	/* Request a slot slot reset. */
5157 	return PCI_ERS_RESULT_NEED_RESET;
5158 }
5159 
5160 /**
5161  * e1000_io_slot_reset - called after the pci bus has been reset.
5162  * @pdev: Pointer to PCI device
5163  *
5164  * Restart the card from scratch, as if from a cold-boot. Implementation
5165  * resembles the first-half of the e1000_resume routine.
5166  */
5167 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5168 {
5169 	struct net_device *netdev = pci_get_drvdata(pdev);
5170 	struct e1000_adapter *adapter = netdev_priv(netdev);
5171 	struct e1000_hw *hw = &adapter->hw;
5172 	int err;
5173 
5174 	if (adapter->need_ioport)
5175 		err = pci_enable_device(pdev);
5176 	else
5177 		err = pci_enable_device_mem(pdev);
5178 	if (err) {
5179 		pr_err("Cannot re-enable PCI device after reset.\n");
5180 		return PCI_ERS_RESULT_DISCONNECT;
5181 	}
5182 	pci_set_master(pdev);
5183 
5184 	pci_enable_wake(pdev, PCI_D3hot, 0);
5185 	pci_enable_wake(pdev, PCI_D3cold, 0);
5186 
5187 	e1000_reset(adapter);
5188 	ew32(WUS, ~0);
5189 
5190 	return PCI_ERS_RESULT_RECOVERED;
5191 }
5192 
5193 /**
5194  * e1000_io_resume - called when traffic can start flowing again.
5195  * @pdev: Pointer to PCI device
5196  *
5197  * This callback is called when the error recovery driver tells us that
5198  * its OK to resume normal operation. Implementation resembles the
5199  * second-half of the e1000_resume routine.
5200  */
5201 static void e1000_io_resume(struct pci_dev *pdev)
5202 {
5203 	struct net_device *netdev = pci_get_drvdata(pdev);
5204 	struct e1000_adapter *adapter = netdev_priv(netdev);
5205 
5206 	e1000_init_manageability(adapter);
5207 
5208 	if (netif_running(netdev)) {
5209 		if (e1000_up(adapter)) {
5210 			pr_info("can't bring device back up after reset\n");
5211 			return;
5212 		}
5213 	}
5214 
5215 	netif_device_attach(netdev);
5216 }
5217 
5218 /* e1000_main.c */
5219