1 /*******************************************************************************
2 
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2012 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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30 
31 #include <linux/module.h>
32 #include <linux/types.h>
33 #include <linux/init.h>
34 #include <linux/pci.h>
35 #include <linux/vmalloc.h>
36 #include <linux/pagemap.h>
37 #include <linux/delay.h>
38 #include <linux/netdevice.h>
39 #include <linux/interrupt.h>
40 #include <linux/tcp.h>
41 #include <linux/ipv6.h>
42 #include <linux/slab.h>
43 #include <net/checksum.h>
44 #include <net/ip6_checksum.h>
45 #include <linux/mii.h>
46 #include <linux/ethtool.h>
47 #include <linux/if_vlan.h>
48 #include <linux/cpu.h>
49 #include <linux/smp.h>
50 #include <linux/pm_qos.h>
51 #include <linux/pm_runtime.h>
52 #include <linux/aer.h>
53 #include <linux/prefetch.h>
54 
55 #include "e1000.h"
56 
57 #define DRV_EXTRAVERSION "-k"
58 
59 #define DRV_VERSION "1.9.5" DRV_EXTRAVERSION
60 char e1000e_driver_name[] = "e1000e";
61 const char e1000e_driver_version[] = DRV_VERSION;
62 
63 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state);
64 
65 static const struct e1000_info *e1000_info_tbl[] = {
66 	[board_82571]		= &e1000_82571_info,
67 	[board_82572]		= &e1000_82572_info,
68 	[board_82573]		= &e1000_82573_info,
69 	[board_82574]		= &e1000_82574_info,
70 	[board_82583]		= &e1000_82583_info,
71 	[board_80003es2lan]	= &e1000_es2_info,
72 	[board_ich8lan]		= &e1000_ich8_info,
73 	[board_ich9lan]		= &e1000_ich9_info,
74 	[board_ich10lan]	= &e1000_ich10_info,
75 	[board_pchlan]		= &e1000_pch_info,
76 	[board_pch2lan]		= &e1000_pch2_info,
77 };
78 
79 struct e1000_reg_info {
80 	u32 ofs;
81 	char *name;
82 };
83 
84 #define E1000_RDFH	0x02410	/* Rx Data FIFO Head - RW */
85 #define E1000_RDFT	0x02418	/* Rx Data FIFO Tail - RW */
86 #define E1000_RDFHS	0x02420	/* Rx Data FIFO Head Saved - RW */
87 #define E1000_RDFTS	0x02428	/* Rx Data FIFO Tail Saved - RW */
88 #define E1000_RDFPC	0x02430	/* Rx Data FIFO Packet Count - RW */
89 
90 #define E1000_TDFH	0x03410	/* Tx Data FIFO Head - RW */
91 #define E1000_TDFT	0x03418	/* Tx Data FIFO Tail - RW */
92 #define E1000_TDFHS	0x03420	/* Tx Data FIFO Head Saved - RW */
93 #define E1000_TDFTS	0x03428	/* Tx Data FIFO Tail Saved - RW */
94 #define E1000_TDFPC	0x03430	/* Tx Data FIFO Packet Count - RW */
95 
96 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
97 
98 	/* General Registers */
99 	{E1000_CTRL, "CTRL"},
100 	{E1000_STATUS, "STATUS"},
101 	{E1000_CTRL_EXT, "CTRL_EXT"},
102 
103 	/* Interrupt Registers */
104 	{E1000_ICR, "ICR"},
105 
106 	/* Rx Registers */
107 	{E1000_RCTL, "RCTL"},
108 	{E1000_RDLEN, "RDLEN"},
109 	{E1000_RDH, "RDH"},
110 	{E1000_RDT, "RDT"},
111 	{E1000_RDTR, "RDTR"},
112 	{E1000_RXDCTL(0), "RXDCTL"},
113 	{E1000_ERT, "ERT"},
114 	{E1000_RDBAL, "RDBAL"},
115 	{E1000_RDBAH, "RDBAH"},
116 	{E1000_RDFH, "RDFH"},
117 	{E1000_RDFT, "RDFT"},
118 	{E1000_RDFHS, "RDFHS"},
119 	{E1000_RDFTS, "RDFTS"},
120 	{E1000_RDFPC, "RDFPC"},
121 
122 	/* Tx Registers */
123 	{E1000_TCTL, "TCTL"},
124 	{E1000_TDBAL, "TDBAL"},
125 	{E1000_TDBAH, "TDBAH"},
126 	{E1000_TDLEN, "TDLEN"},
127 	{E1000_TDH, "TDH"},
128 	{E1000_TDT, "TDT"},
129 	{E1000_TIDV, "TIDV"},
130 	{E1000_TXDCTL(0), "TXDCTL"},
131 	{E1000_TADV, "TADV"},
132 	{E1000_TARC(0), "TARC"},
133 	{E1000_TDFH, "TDFH"},
134 	{E1000_TDFT, "TDFT"},
135 	{E1000_TDFHS, "TDFHS"},
136 	{E1000_TDFTS, "TDFTS"},
137 	{E1000_TDFPC, "TDFPC"},
138 
139 	/* List Terminator */
140 	{0, NULL}
141 };
142 
143 /*
144  * e1000_regdump - register printout routine
145  */
146 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
147 {
148 	int n = 0;
149 	char rname[16];
150 	u32 regs[8];
151 
152 	switch (reginfo->ofs) {
153 	case E1000_RXDCTL(0):
154 		for (n = 0; n < 2; n++)
155 			regs[n] = __er32(hw, E1000_RXDCTL(n));
156 		break;
157 	case E1000_TXDCTL(0):
158 		for (n = 0; n < 2; n++)
159 			regs[n] = __er32(hw, E1000_TXDCTL(n));
160 		break;
161 	case E1000_TARC(0):
162 		for (n = 0; n < 2; n++)
163 			regs[n] = __er32(hw, E1000_TARC(n));
164 		break;
165 	default:
166 		pr_info("%-15s %08x\n",
167 			reginfo->name, __er32(hw, reginfo->ofs));
168 		return;
169 	}
170 
171 	snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
172 	pr_info("%-15s %08x %08x\n", rname, regs[0], regs[1]);
173 }
174 
175 /*
176  * e1000e_dump - Print registers, Tx-ring and Rx-ring
177  */
178 static void e1000e_dump(struct e1000_adapter *adapter)
179 {
180 	struct net_device *netdev = adapter->netdev;
181 	struct e1000_hw *hw = &adapter->hw;
182 	struct e1000_reg_info *reginfo;
183 	struct e1000_ring *tx_ring = adapter->tx_ring;
184 	struct e1000_tx_desc *tx_desc;
185 	struct my_u0 {
186 		__le64 a;
187 		__le64 b;
188 	} *u0;
189 	struct e1000_buffer *buffer_info;
190 	struct e1000_ring *rx_ring = adapter->rx_ring;
191 	union e1000_rx_desc_packet_split *rx_desc_ps;
192 	union e1000_rx_desc_extended *rx_desc;
193 	struct my_u1 {
194 		__le64 a;
195 		__le64 b;
196 		__le64 c;
197 		__le64 d;
198 	} *u1;
199 	u32 staterr;
200 	int i = 0;
201 
202 	if (!netif_msg_hw(adapter))
203 		return;
204 
205 	/* Print netdevice Info */
206 	if (netdev) {
207 		dev_info(&adapter->pdev->dev, "Net device Info\n");
208 		pr_info("Device Name     state            trans_start      last_rx\n");
209 		pr_info("%-15s %016lX %016lX %016lX\n",
210 			netdev->name, netdev->state, netdev->trans_start,
211 			netdev->last_rx);
212 	}
213 
214 	/* Print Registers */
215 	dev_info(&adapter->pdev->dev, "Register Dump\n");
216 	pr_info(" Register Name   Value\n");
217 	for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
218 	     reginfo->name; reginfo++) {
219 		e1000_regdump(hw, reginfo);
220 	}
221 
222 	/* Print Tx Ring Summary */
223 	if (!netdev || !netif_running(netdev))
224 		return;
225 
226 	dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
227 	pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
228 	buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
229 	pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
230 		0, tx_ring->next_to_use, tx_ring->next_to_clean,
231 		(unsigned long long)buffer_info->dma,
232 		buffer_info->length,
233 		buffer_info->next_to_watch,
234 		(unsigned long long)buffer_info->time_stamp);
235 
236 	/* Print Tx Ring */
237 	if (!netif_msg_tx_done(adapter))
238 		goto rx_ring_summary;
239 
240 	dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
241 
242 	/* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
243 	 *
244 	 * Legacy Transmit Descriptor
245 	 *   +--------------------------------------------------------------+
246 	 * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
247 	 *   +--------------------------------------------------------------+
248 	 * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
249 	 *   +--------------------------------------------------------------+
250 	 *   63       48 47        36 35    32 31     24 23    16 15        0
251 	 *
252 	 * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
253 	 *   63      48 47    40 39       32 31             16 15    8 7      0
254 	 *   +----------------------------------------------------------------+
255 	 * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
256 	 *   +----------------------------------------------------------------+
257 	 * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
258 	 *   +----------------------------------------------------------------+
259 	 *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
260 	 *
261 	 * Extended Data Descriptor (DTYP=0x1)
262 	 *   +----------------------------------------------------------------+
263 	 * 0 |                     Buffer Address [63:0]                      |
264 	 *   +----------------------------------------------------------------+
265 	 * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
266 	 *   +----------------------------------------------------------------+
267 	 *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
268 	 */
269 	pr_info("Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Legacy format\n");
270 	pr_info("Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Context format\n");
271 	pr_info("Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Data format\n");
272 	for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
273 		const char *next_desc;
274 		tx_desc = E1000_TX_DESC(*tx_ring, i);
275 		buffer_info = &tx_ring->buffer_info[i];
276 		u0 = (struct my_u0 *)tx_desc;
277 		if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
278 			next_desc = " NTC/U";
279 		else if (i == tx_ring->next_to_use)
280 			next_desc = " NTU";
281 		else if (i == tx_ring->next_to_clean)
282 			next_desc = " NTC";
283 		else
284 			next_desc = "";
285 		pr_info("T%c[0x%03X]    %016llX %016llX %016llX %04X  %3X %016llX %p%s\n",
286 			(!(le64_to_cpu(u0->b) & (1 << 29)) ? 'l' :
287 			 ((le64_to_cpu(u0->b) & (1 << 20)) ? 'd' : 'c')),
288 			i,
289 			(unsigned long long)le64_to_cpu(u0->a),
290 			(unsigned long long)le64_to_cpu(u0->b),
291 			(unsigned long long)buffer_info->dma,
292 			buffer_info->length, buffer_info->next_to_watch,
293 			(unsigned long long)buffer_info->time_stamp,
294 			buffer_info->skb, next_desc);
295 
296 		if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
297 			print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
298 				       16, 1, phys_to_virt(buffer_info->dma),
299 				       buffer_info->length, true);
300 	}
301 
302 	/* Print Rx Ring Summary */
303 rx_ring_summary:
304 	dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
305 	pr_info("Queue [NTU] [NTC]\n");
306 	pr_info(" %5d %5X %5X\n",
307 		0, rx_ring->next_to_use, rx_ring->next_to_clean);
308 
309 	/* Print Rx Ring */
310 	if (!netif_msg_rx_status(adapter))
311 		return;
312 
313 	dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
314 	switch (adapter->rx_ps_pages) {
315 	case 1:
316 	case 2:
317 	case 3:
318 		/* [Extended] Packet Split Receive Descriptor Format
319 		 *
320 		 *    +-----------------------------------------------------+
321 		 *  0 |                Buffer Address 0 [63:0]              |
322 		 *    +-----------------------------------------------------+
323 		 *  8 |                Buffer Address 1 [63:0]              |
324 		 *    +-----------------------------------------------------+
325 		 * 16 |                Buffer Address 2 [63:0]              |
326 		 *    +-----------------------------------------------------+
327 		 * 24 |                Buffer Address 3 [63:0]              |
328 		 *    +-----------------------------------------------------+
329 		 */
330 		pr_info("R  [desc]      [buffer 0 63:0 ] [buffer 1 63:0 ] [buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma       ] [bi->skb] <-- Ext Pkt Split format\n");
331 		/* [Extended] Receive Descriptor (Write-Back) Format
332 		 *
333 		 *   63       48 47    32 31     13 12    8 7    4 3        0
334 		 *   +------------------------------------------------------+
335 		 * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
336 		 *   | Checksum | Ident  |         | Queue |      |  Type   |
337 		 *   +------------------------------------------------------+
338 		 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
339 		 *   +------------------------------------------------------+
340 		 *   63       48 47    32 31            20 19               0
341 		 */
342 		pr_info("RWB[desc]      [ck ipid mrqhsh] [vl   l0 ee  es] [ l3  l2  l1 hs] [reserved      ] ---------------- [bi->skb] <-- Ext Rx Write-Back format\n");
343 		for (i = 0; i < rx_ring->count; i++) {
344 			const char *next_desc;
345 			buffer_info = &rx_ring->buffer_info[i];
346 			rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
347 			u1 = (struct my_u1 *)rx_desc_ps;
348 			staterr =
349 			    le32_to_cpu(rx_desc_ps->wb.middle.status_error);
350 
351 			if (i == rx_ring->next_to_use)
352 				next_desc = " NTU";
353 			else if (i == rx_ring->next_to_clean)
354 				next_desc = " NTC";
355 			else
356 				next_desc = "";
357 
358 			if (staterr & E1000_RXD_STAT_DD) {
359 				/* Descriptor Done */
360 				pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX ---------------- %p%s\n",
361 					"RWB", i,
362 					(unsigned long long)le64_to_cpu(u1->a),
363 					(unsigned long long)le64_to_cpu(u1->b),
364 					(unsigned long long)le64_to_cpu(u1->c),
365 					(unsigned long long)le64_to_cpu(u1->d),
366 					buffer_info->skb, next_desc);
367 			} else {
368 				pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX %016llX %p%s\n",
369 					"R  ", i,
370 					(unsigned long long)le64_to_cpu(u1->a),
371 					(unsigned long long)le64_to_cpu(u1->b),
372 					(unsigned long long)le64_to_cpu(u1->c),
373 					(unsigned long long)le64_to_cpu(u1->d),
374 					(unsigned long long)buffer_info->dma,
375 					buffer_info->skb, next_desc);
376 
377 				if (netif_msg_pktdata(adapter))
378 					print_hex_dump(KERN_INFO, "",
379 						DUMP_PREFIX_ADDRESS, 16, 1,
380 						phys_to_virt(buffer_info->dma),
381 						adapter->rx_ps_bsize0, true);
382 			}
383 		}
384 		break;
385 	default:
386 	case 0:
387 		/* Extended Receive Descriptor (Read) Format
388 		 *
389 		 *   +-----------------------------------------------------+
390 		 * 0 |                Buffer Address [63:0]                |
391 		 *   +-----------------------------------------------------+
392 		 * 8 |                      Reserved                       |
393 		 *   +-----------------------------------------------------+
394 		 */
395 		pr_info("R  [desc]      [buf addr 63:0 ] [reserved 63:0 ] [bi->dma       ] [bi->skb] <-- Ext (Read) format\n");
396 		/* Extended Receive Descriptor (Write-Back) Format
397 		 *
398 		 *   63       48 47    32 31    24 23            4 3        0
399 		 *   +------------------------------------------------------+
400 		 *   |     RSS Hash      |        |               |         |
401 		 * 0 +-------------------+  Rsvd  |   Reserved    | MRQ RSS |
402 		 *   | Packet   | IP     |        |               |  Type   |
403 		 *   | Checksum | Ident  |        |               |         |
404 		 *   +------------------------------------------------------+
405 		 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
406 		 *   +------------------------------------------------------+
407 		 *   63       48 47    32 31            20 19               0
408 		 */
409 		pr_info("RWB[desc]      [cs ipid    mrq] [vt   ln xe  xs] [bi->skb] <-- Ext (Write-Back) format\n");
410 
411 		for (i = 0; i < rx_ring->count; i++) {
412 			const char *next_desc;
413 
414 			buffer_info = &rx_ring->buffer_info[i];
415 			rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
416 			u1 = (struct my_u1 *)rx_desc;
417 			staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
418 
419 			if (i == rx_ring->next_to_use)
420 				next_desc = " NTU";
421 			else if (i == rx_ring->next_to_clean)
422 				next_desc = " NTC";
423 			else
424 				next_desc = "";
425 
426 			if (staterr & E1000_RXD_STAT_DD) {
427 				/* Descriptor Done */
428 				pr_info("%s[0x%03X]     %016llX %016llX ---------------- %p%s\n",
429 					"RWB", i,
430 					(unsigned long long)le64_to_cpu(u1->a),
431 					(unsigned long long)le64_to_cpu(u1->b),
432 					buffer_info->skb, next_desc);
433 			} else {
434 				pr_info("%s[0x%03X]     %016llX %016llX %016llX %p%s\n",
435 					"R  ", i,
436 					(unsigned long long)le64_to_cpu(u1->a),
437 					(unsigned long long)le64_to_cpu(u1->b),
438 					(unsigned long long)buffer_info->dma,
439 					buffer_info->skb, next_desc);
440 
441 				if (netif_msg_pktdata(adapter))
442 					print_hex_dump(KERN_INFO, "",
443 						       DUMP_PREFIX_ADDRESS, 16,
444 						       1,
445 						       phys_to_virt
446 						       (buffer_info->dma),
447 						       adapter->rx_buffer_len,
448 						       true);
449 			}
450 		}
451 	}
452 }
453 
454 /**
455  * e1000_desc_unused - calculate if we have unused descriptors
456  **/
457 static int e1000_desc_unused(struct e1000_ring *ring)
458 {
459 	if (ring->next_to_clean > ring->next_to_use)
460 		return ring->next_to_clean - ring->next_to_use - 1;
461 
462 	return ring->count + ring->next_to_clean - ring->next_to_use - 1;
463 }
464 
465 /**
466  * e1000_receive_skb - helper function to handle Rx indications
467  * @adapter: board private structure
468  * @status: descriptor status field as written by hardware
469  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
470  * @skb: pointer to sk_buff to be indicated to stack
471  **/
472 static void e1000_receive_skb(struct e1000_adapter *adapter,
473 			      struct net_device *netdev, struct sk_buff *skb,
474 			      u8 status, __le16 vlan)
475 {
476 	u16 tag = le16_to_cpu(vlan);
477 	skb->protocol = eth_type_trans(skb, netdev);
478 
479 	if (status & E1000_RXD_STAT_VP)
480 		__vlan_hwaccel_put_tag(skb, tag);
481 
482 	napi_gro_receive(&adapter->napi, skb);
483 }
484 
485 /**
486  * e1000_rx_checksum - Receive Checksum Offload
487  * @adapter: board private structure
488  * @status_err: receive descriptor status and error fields
489  * @csum: receive descriptor csum field
490  * @sk_buff: socket buffer with received data
491  **/
492 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
493 			      __le16 csum, struct sk_buff *skb)
494 {
495 	u16 status = (u16)status_err;
496 	u8 errors = (u8)(status_err >> 24);
497 
498 	skb_checksum_none_assert(skb);
499 
500 	/* Rx checksum disabled */
501 	if (!(adapter->netdev->features & NETIF_F_RXCSUM))
502 		return;
503 
504 	/* Ignore Checksum bit is set */
505 	if (status & E1000_RXD_STAT_IXSM)
506 		return;
507 
508 	/* TCP/UDP checksum error bit is set */
509 	if (errors & E1000_RXD_ERR_TCPE) {
510 		/* let the stack verify checksum errors */
511 		adapter->hw_csum_err++;
512 		return;
513 	}
514 
515 	/* TCP/UDP Checksum has not been calculated */
516 	if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
517 		return;
518 
519 	/* It must be a TCP or UDP packet with a valid checksum */
520 	if (status & E1000_RXD_STAT_TCPCS) {
521 		/* TCP checksum is good */
522 		skb->ip_summed = CHECKSUM_UNNECESSARY;
523 	} else {
524 		/*
525 		 * IP fragment with UDP payload
526 		 * Hardware complements the payload checksum, so we undo it
527 		 * and then put the value in host order for further stack use.
528 		 */
529 		__sum16 sum = (__force __sum16)swab16((__force u16)csum);
530 		skb->csum = csum_unfold(~sum);
531 		skb->ip_summed = CHECKSUM_COMPLETE;
532 	}
533 	adapter->hw_csum_good++;
534 }
535 
536 /**
537  * e1000e_update_tail_wa - helper function for e1000e_update_[rt]dt_wa()
538  * @hw: pointer to the HW structure
539  * @tail: address of tail descriptor register
540  * @i: value to write to tail descriptor register
541  *
542  * When updating the tail register, the ME could be accessing Host CSR
543  * registers at the same time.  Normally, this is handled in h/w by an
544  * arbiter but on some parts there is a bug that acknowledges Host accesses
545  * later than it should which could result in the descriptor register to
546  * have an incorrect value.  Workaround this by checking the FWSM register
547  * which has bit 24 set while ME is accessing Host CSR registers, wait
548  * if it is set and try again a number of times.
549  **/
550 static inline s32 e1000e_update_tail_wa(struct e1000_hw *hw, void __iomem *tail,
551 					unsigned int i)
552 {
553 	unsigned int j = 0;
554 
555 	while ((j++ < E1000_ICH_FWSM_PCIM2PCI_COUNT) &&
556 	       (er32(FWSM) & E1000_ICH_FWSM_PCIM2PCI))
557 		udelay(50);
558 
559 	writel(i, tail);
560 
561 	if ((j == E1000_ICH_FWSM_PCIM2PCI_COUNT) && (i != readl(tail)))
562 		return E1000_ERR_SWFW_SYNC;
563 
564 	return 0;
565 }
566 
567 static void e1000e_update_rdt_wa(struct e1000_ring *rx_ring, unsigned int i)
568 {
569 	struct e1000_adapter *adapter = rx_ring->adapter;
570 	struct e1000_hw *hw = &adapter->hw;
571 
572 	if (e1000e_update_tail_wa(hw, rx_ring->tail, i)) {
573 		u32 rctl = er32(RCTL);
574 		ew32(RCTL, rctl & ~E1000_RCTL_EN);
575 		e_err("ME firmware caused invalid RDT - resetting\n");
576 		schedule_work(&adapter->reset_task);
577 	}
578 }
579 
580 static void e1000e_update_tdt_wa(struct e1000_ring *tx_ring, unsigned int i)
581 {
582 	struct e1000_adapter *adapter = tx_ring->adapter;
583 	struct e1000_hw *hw = &adapter->hw;
584 
585 	if (e1000e_update_tail_wa(hw, tx_ring->tail, i)) {
586 		u32 tctl = er32(TCTL);
587 		ew32(TCTL, tctl & ~E1000_TCTL_EN);
588 		e_err("ME firmware caused invalid TDT - resetting\n");
589 		schedule_work(&adapter->reset_task);
590 	}
591 }
592 
593 /**
594  * e1000_alloc_rx_buffers - Replace used receive buffers
595  * @rx_ring: Rx descriptor ring
596  **/
597 static void e1000_alloc_rx_buffers(struct e1000_ring *rx_ring,
598 				   int cleaned_count, gfp_t gfp)
599 {
600 	struct e1000_adapter *adapter = rx_ring->adapter;
601 	struct net_device *netdev = adapter->netdev;
602 	struct pci_dev *pdev = adapter->pdev;
603 	union e1000_rx_desc_extended *rx_desc;
604 	struct e1000_buffer *buffer_info;
605 	struct sk_buff *skb;
606 	unsigned int i;
607 	unsigned int bufsz = adapter->rx_buffer_len;
608 
609 	i = rx_ring->next_to_use;
610 	buffer_info = &rx_ring->buffer_info[i];
611 
612 	while (cleaned_count--) {
613 		skb = buffer_info->skb;
614 		if (skb) {
615 			skb_trim(skb, 0);
616 			goto map_skb;
617 		}
618 
619 		skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
620 		if (!skb) {
621 			/* Better luck next round */
622 			adapter->alloc_rx_buff_failed++;
623 			break;
624 		}
625 
626 		buffer_info->skb = skb;
627 map_skb:
628 		buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
629 						  adapter->rx_buffer_len,
630 						  DMA_FROM_DEVICE);
631 		if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
632 			dev_err(&pdev->dev, "Rx DMA map failed\n");
633 			adapter->rx_dma_failed++;
634 			break;
635 		}
636 
637 		rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
638 		rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
639 
640 		if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
641 			/*
642 			 * Force memory writes to complete before letting h/w
643 			 * know there are new descriptors to fetch.  (Only
644 			 * applicable for weak-ordered memory model archs,
645 			 * such as IA-64).
646 			 */
647 			wmb();
648 			if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
649 				e1000e_update_rdt_wa(rx_ring, i);
650 			else
651 				writel(i, rx_ring->tail);
652 		}
653 		i++;
654 		if (i == rx_ring->count)
655 			i = 0;
656 		buffer_info = &rx_ring->buffer_info[i];
657 	}
658 
659 	rx_ring->next_to_use = i;
660 }
661 
662 /**
663  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
664  * @rx_ring: Rx descriptor ring
665  **/
666 static void e1000_alloc_rx_buffers_ps(struct e1000_ring *rx_ring,
667 				      int cleaned_count, gfp_t gfp)
668 {
669 	struct e1000_adapter *adapter = rx_ring->adapter;
670 	struct net_device *netdev = adapter->netdev;
671 	struct pci_dev *pdev = adapter->pdev;
672 	union e1000_rx_desc_packet_split *rx_desc;
673 	struct e1000_buffer *buffer_info;
674 	struct e1000_ps_page *ps_page;
675 	struct sk_buff *skb;
676 	unsigned int i, j;
677 
678 	i = rx_ring->next_to_use;
679 	buffer_info = &rx_ring->buffer_info[i];
680 
681 	while (cleaned_count--) {
682 		rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
683 
684 		for (j = 0; j < PS_PAGE_BUFFERS; j++) {
685 			ps_page = &buffer_info->ps_pages[j];
686 			if (j >= adapter->rx_ps_pages) {
687 				/* all unused desc entries get hw null ptr */
688 				rx_desc->read.buffer_addr[j + 1] =
689 				    ~cpu_to_le64(0);
690 				continue;
691 			}
692 			if (!ps_page->page) {
693 				ps_page->page = alloc_page(gfp);
694 				if (!ps_page->page) {
695 					adapter->alloc_rx_buff_failed++;
696 					goto no_buffers;
697 				}
698 				ps_page->dma = dma_map_page(&pdev->dev,
699 							    ps_page->page,
700 							    0, PAGE_SIZE,
701 							    DMA_FROM_DEVICE);
702 				if (dma_mapping_error(&pdev->dev,
703 						      ps_page->dma)) {
704 					dev_err(&adapter->pdev->dev,
705 						"Rx DMA page map failed\n");
706 					adapter->rx_dma_failed++;
707 					goto no_buffers;
708 				}
709 			}
710 			/*
711 			 * Refresh the desc even if buffer_addrs
712 			 * didn't change because each write-back
713 			 * erases this info.
714 			 */
715 			rx_desc->read.buffer_addr[j + 1] =
716 			    cpu_to_le64(ps_page->dma);
717 		}
718 
719 		skb = __netdev_alloc_skb_ip_align(netdev,
720 						  adapter->rx_ps_bsize0,
721 						  gfp);
722 
723 		if (!skb) {
724 			adapter->alloc_rx_buff_failed++;
725 			break;
726 		}
727 
728 		buffer_info->skb = skb;
729 		buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
730 						  adapter->rx_ps_bsize0,
731 						  DMA_FROM_DEVICE);
732 		if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
733 			dev_err(&pdev->dev, "Rx DMA map failed\n");
734 			adapter->rx_dma_failed++;
735 			/* cleanup skb */
736 			dev_kfree_skb_any(skb);
737 			buffer_info->skb = NULL;
738 			break;
739 		}
740 
741 		rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
742 
743 		if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
744 			/*
745 			 * Force memory writes to complete before letting h/w
746 			 * know there are new descriptors to fetch.  (Only
747 			 * applicable for weak-ordered memory model archs,
748 			 * such as IA-64).
749 			 */
750 			wmb();
751 			if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
752 				e1000e_update_rdt_wa(rx_ring, i << 1);
753 			else
754 				writel(i << 1, rx_ring->tail);
755 		}
756 
757 		i++;
758 		if (i == rx_ring->count)
759 			i = 0;
760 		buffer_info = &rx_ring->buffer_info[i];
761 	}
762 
763 no_buffers:
764 	rx_ring->next_to_use = i;
765 }
766 
767 /**
768  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
769  * @rx_ring: Rx descriptor ring
770  * @cleaned_count: number of buffers to allocate this pass
771  **/
772 
773 static void e1000_alloc_jumbo_rx_buffers(struct e1000_ring *rx_ring,
774 					 int cleaned_count, gfp_t gfp)
775 {
776 	struct e1000_adapter *adapter = rx_ring->adapter;
777 	struct net_device *netdev = adapter->netdev;
778 	struct pci_dev *pdev = adapter->pdev;
779 	union e1000_rx_desc_extended *rx_desc;
780 	struct e1000_buffer *buffer_info;
781 	struct sk_buff *skb;
782 	unsigned int i;
783 	unsigned int bufsz = 256 - 16 /* for skb_reserve */;
784 
785 	i = rx_ring->next_to_use;
786 	buffer_info = &rx_ring->buffer_info[i];
787 
788 	while (cleaned_count--) {
789 		skb = buffer_info->skb;
790 		if (skb) {
791 			skb_trim(skb, 0);
792 			goto check_page;
793 		}
794 
795 		skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
796 		if (unlikely(!skb)) {
797 			/* Better luck next round */
798 			adapter->alloc_rx_buff_failed++;
799 			break;
800 		}
801 
802 		buffer_info->skb = skb;
803 check_page:
804 		/* allocate a new page if necessary */
805 		if (!buffer_info->page) {
806 			buffer_info->page = alloc_page(gfp);
807 			if (unlikely(!buffer_info->page)) {
808 				adapter->alloc_rx_buff_failed++;
809 				break;
810 			}
811 		}
812 
813 		if (!buffer_info->dma)
814 			buffer_info->dma = dma_map_page(&pdev->dev,
815 			                                buffer_info->page, 0,
816 			                                PAGE_SIZE,
817 							DMA_FROM_DEVICE);
818 
819 		rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
820 		rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
821 
822 		if (unlikely(++i == rx_ring->count))
823 			i = 0;
824 		buffer_info = &rx_ring->buffer_info[i];
825 	}
826 
827 	if (likely(rx_ring->next_to_use != i)) {
828 		rx_ring->next_to_use = i;
829 		if (unlikely(i-- == 0))
830 			i = (rx_ring->count - 1);
831 
832 		/* Force memory writes to complete before letting h/w
833 		 * know there are new descriptors to fetch.  (Only
834 		 * applicable for weak-ordered memory model archs,
835 		 * such as IA-64). */
836 		wmb();
837 		if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
838 			e1000e_update_rdt_wa(rx_ring, i);
839 		else
840 			writel(i, rx_ring->tail);
841 	}
842 }
843 
844 static inline void e1000_rx_hash(struct net_device *netdev, __le32 rss,
845 				 struct sk_buff *skb)
846 {
847 	if (netdev->features & NETIF_F_RXHASH)
848 		skb->rxhash = le32_to_cpu(rss);
849 }
850 
851 /**
852  * e1000_clean_rx_irq - Send received data up the network stack
853  * @rx_ring: Rx descriptor ring
854  *
855  * the return value indicates whether actual cleaning was done, there
856  * is no guarantee that everything was cleaned
857  **/
858 static bool e1000_clean_rx_irq(struct e1000_ring *rx_ring, int *work_done,
859 			       int work_to_do)
860 {
861 	struct e1000_adapter *adapter = rx_ring->adapter;
862 	struct net_device *netdev = adapter->netdev;
863 	struct pci_dev *pdev = adapter->pdev;
864 	struct e1000_hw *hw = &adapter->hw;
865 	union e1000_rx_desc_extended *rx_desc, *next_rxd;
866 	struct e1000_buffer *buffer_info, *next_buffer;
867 	u32 length, staterr;
868 	unsigned int i;
869 	int cleaned_count = 0;
870 	bool cleaned = false;
871 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
872 
873 	i = rx_ring->next_to_clean;
874 	rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
875 	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
876 	buffer_info = &rx_ring->buffer_info[i];
877 
878 	while (staterr & E1000_RXD_STAT_DD) {
879 		struct sk_buff *skb;
880 
881 		if (*work_done >= work_to_do)
882 			break;
883 		(*work_done)++;
884 		rmb();	/* read descriptor and rx_buffer_info after status DD */
885 
886 		skb = buffer_info->skb;
887 		buffer_info->skb = NULL;
888 
889 		prefetch(skb->data - NET_IP_ALIGN);
890 
891 		i++;
892 		if (i == rx_ring->count)
893 			i = 0;
894 		next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
895 		prefetch(next_rxd);
896 
897 		next_buffer = &rx_ring->buffer_info[i];
898 
899 		cleaned = true;
900 		cleaned_count++;
901 		dma_unmap_single(&pdev->dev,
902 				 buffer_info->dma,
903 				 adapter->rx_buffer_len,
904 				 DMA_FROM_DEVICE);
905 		buffer_info->dma = 0;
906 
907 		length = le16_to_cpu(rx_desc->wb.upper.length);
908 
909 		/*
910 		 * !EOP means multiple descriptors were used to store a single
911 		 * packet, if that's the case we need to toss it.  In fact, we
912 		 * need to toss every packet with the EOP bit clear and the
913 		 * next frame that _does_ have the EOP bit set, as it is by
914 		 * definition only a frame fragment
915 		 */
916 		if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
917 			adapter->flags2 |= FLAG2_IS_DISCARDING;
918 
919 		if (adapter->flags2 & FLAG2_IS_DISCARDING) {
920 			/* All receives must fit into a single buffer */
921 			e_dbg("Receive packet consumed multiple buffers\n");
922 			/* recycle */
923 			buffer_info->skb = skb;
924 			if (staterr & E1000_RXD_STAT_EOP)
925 				adapter->flags2 &= ~FLAG2_IS_DISCARDING;
926 			goto next_desc;
927 		}
928 
929 		if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
930 			     !(netdev->features & NETIF_F_RXALL))) {
931 			/* recycle */
932 			buffer_info->skb = skb;
933 			goto next_desc;
934 		}
935 
936 		/* adjust length to remove Ethernet CRC */
937 		if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
938 			/* If configured to store CRC, don't subtract FCS,
939 			 * but keep the FCS bytes out of the total_rx_bytes
940 			 * counter
941 			 */
942 			if (netdev->features & NETIF_F_RXFCS)
943 				total_rx_bytes -= 4;
944 			else
945 				length -= 4;
946 		}
947 
948 		total_rx_bytes += length;
949 		total_rx_packets++;
950 
951 		/*
952 		 * code added for copybreak, this should improve
953 		 * performance for small packets with large amounts
954 		 * of reassembly being done in the stack
955 		 */
956 		if (length < copybreak) {
957 			struct sk_buff *new_skb =
958 			    netdev_alloc_skb_ip_align(netdev, length);
959 			if (new_skb) {
960 				skb_copy_to_linear_data_offset(new_skb,
961 							       -NET_IP_ALIGN,
962 							       (skb->data -
963 								NET_IP_ALIGN),
964 							       (length +
965 								NET_IP_ALIGN));
966 				/* save the skb in buffer_info as good */
967 				buffer_info->skb = skb;
968 				skb = new_skb;
969 			}
970 			/* else just continue with the old one */
971 		}
972 		/* end copybreak code */
973 		skb_put(skb, length);
974 
975 		/* Receive Checksum Offload */
976 		e1000_rx_checksum(adapter, staterr,
977 				  rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
978 
979 		e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
980 
981 		e1000_receive_skb(adapter, netdev, skb, staterr,
982 				  rx_desc->wb.upper.vlan);
983 
984 next_desc:
985 		rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
986 
987 		/* return some buffers to hardware, one at a time is too slow */
988 		if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
989 			adapter->alloc_rx_buf(rx_ring, cleaned_count,
990 					      GFP_ATOMIC);
991 			cleaned_count = 0;
992 		}
993 
994 		/* use prefetched values */
995 		rx_desc = next_rxd;
996 		buffer_info = next_buffer;
997 
998 		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
999 	}
1000 	rx_ring->next_to_clean = i;
1001 
1002 	cleaned_count = e1000_desc_unused(rx_ring);
1003 	if (cleaned_count)
1004 		adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1005 
1006 	adapter->total_rx_bytes += total_rx_bytes;
1007 	adapter->total_rx_packets += total_rx_packets;
1008 	return cleaned;
1009 }
1010 
1011 static void e1000_put_txbuf(struct e1000_ring *tx_ring,
1012 			    struct e1000_buffer *buffer_info)
1013 {
1014 	struct e1000_adapter *adapter = tx_ring->adapter;
1015 
1016 	if (buffer_info->dma) {
1017 		if (buffer_info->mapped_as_page)
1018 			dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1019 				       buffer_info->length, DMA_TO_DEVICE);
1020 		else
1021 			dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1022 					 buffer_info->length, DMA_TO_DEVICE);
1023 		buffer_info->dma = 0;
1024 	}
1025 	if (buffer_info->skb) {
1026 		dev_kfree_skb_any(buffer_info->skb);
1027 		buffer_info->skb = NULL;
1028 	}
1029 	buffer_info->time_stamp = 0;
1030 }
1031 
1032 static void e1000_print_hw_hang(struct work_struct *work)
1033 {
1034 	struct e1000_adapter *adapter = container_of(work,
1035 	                                             struct e1000_adapter,
1036 	                                             print_hang_task);
1037 	struct net_device *netdev = adapter->netdev;
1038 	struct e1000_ring *tx_ring = adapter->tx_ring;
1039 	unsigned int i = tx_ring->next_to_clean;
1040 	unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1041 	struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1042 	struct e1000_hw *hw = &adapter->hw;
1043 	u16 phy_status, phy_1000t_status, phy_ext_status;
1044 	u16 pci_status;
1045 
1046 	if (test_bit(__E1000_DOWN, &adapter->state))
1047 		return;
1048 
1049 	if (!adapter->tx_hang_recheck &&
1050 	    (adapter->flags2 & FLAG2_DMA_BURST)) {
1051 		/* May be block on write-back, flush and detect again
1052 		 * flush pending descriptor writebacks to memory
1053 		 */
1054 		ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1055 		/* execute the writes immediately */
1056 		e1e_flush();
1057 		adapter->tx_hang_recheck = true;
1058 		return;
1059 	}
1060 	/* Real hang detected */
1061 	adapter->tx_hang_recheck = false;
1062 	netif_stop_queue(netdev);
1063 
1064 	e1e_rphy(hw, PHY_STATUS, &phy_status);
1065 	e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
1066 	e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
1067 
1068 	pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1069 
1070 	/* detected Hardware unit hang */
1071 	e_err("Detected Hardware Unit Hang:\n"
1072 	      "  TDH                  <%x>\n"
1073 	      "  TDT                  <%x>\n"
1074 	      "  next_to_use          <%x>\n"
1075 	      "  next_to_clean        <%x>\n"
1076 	      "buffer_info[next_to_clean]:\n"
1077 	      "  time_stamp           <%lx>\n"
1078 	      "  next_to_watch        <%x>\n"
1079 	      "  jiffies              <%lx>\n"
1080 	      "  next_to_watch.status <%x>\n"
1081 	      "MAC Status             <%x>\n"
1082 	      "PHY Status             <%x>\n"
1083 	      "PHY 1000BASE-T Status  <%x>\n"
1084 	      "PHY Extended Status    <%x>\n"
1085 	      "PCI Status             <%x>\n",
1086 	      readl(tx_ring->head),
1087 	      readl(tx_ring->tail),
1088 	      tx_ring->next_to_use,
1089 	      tx_ring->next_to_clean,
1090 	      tx_ring->buffer_info[eop].time_stamp,
1091 	      eop,
1092 	      jiffies,
1093 	      eop_desc->upper.fields.status,
1094 	      er32(STATUS),
1095 	      phy_status,
1096 	      phy_1000t_status,
1097 	      phy_ext_status,
1098 	      pci_status);
1099 }
1100 
1101 /**
1102  * e1000_clean_tx_irq - Reclaim resources after transmit completes
1103  * @tx_ring: Tx descriptor ring
1104  *
1105  * the return value indicates whether actual cleaning was done, there
1106  * is no guarantee that everything was cleaned
1107  **/
1108 static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
1109 {
1110 	struct e1000_adapter *adapter = tx_ring->adapter;
1111 	struct net_device *netdev = adapter->netdev;
1112 	struct e1000_hw *hw = &adapter->hw;
1113 	struct e1000_tx_desc *tx_desc, *eop_desc;
1114 	struct e1000_buffer *buffer_info;
1115 	unsigned int i, eop;
1116 	unsigned int count = 0;
1117 	unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1118 	unsigned int bytes_compl = 0, pkts_compl = 0;
1119 
1120 	i = tx_ring->next_to_clean;
1121 	eop = tx_ring->buffer_info[i].next_to_watch;
1122 	eop_desc = E1000_TX_DESC(*tx_ring, eop);
1123 
1124 	while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1125 	       (count < tx_ring->count)) {
1126 		bool cleaned = false;
1127 		rmb(); /* read buffer_info after eop_desc */
1128 		for (; !cleaned; count++) {
1129 			tx_desc = E1000_TX_DESC(*tx_ring, i);
1130 			buffer_info = &tx_ring->buffer_info[i];
1131 			cleaned = (i == eop);
1132 
1133 			if (cleaned) {
1134 				total_tx_packets += buffer_info->segs;
1135 				total_tx_bytes += buffer_info->bytecount;
1136 				if (buffer_info->skb) {
1137 					bytes_compl += buffer_info->skb->len;
1138 					pkts_compl++;
1139 				}
1140 			}
1141 
1142 			e1000_put_txbuf(tx_ring, buffer_info);
1143 			tx_desc->upper.data = 0;
1144 
1145 			i++;
1146 			if (i == tx_ring->count)
1147 				i = 0;
1148 		}
1149 
1150 		if (i == tx_ring->next_to_use)
1151 			break;
1152 		eop = tx_ring->buffer_info[i].next_to_watch;
1153 		eop_desc = E1000_TX_DESC(*tx_ring, eop);
1154 	}
1155 
1156 	tx_ring->next_to_clean = i;
1157 
1158 	netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1159 
1160 #define TX_WAKE_THRESHOLD 32
1161 	if (count && netif_carrier_ok(netdev) &&
1162 	    e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1163 		/* Make sure that anybody stopping the queue after this
1164 		 * sees the new next_to_clean.
1165 		 */
1166 		smp_mb();
1167 
1168 		if (netif_queue_stopped(netdev) &&
1169 		    !(test_bit(__E1000_DOWN, &adapter->state))) {
1170 			netif_wake_queue(netdev);
1171 			++adapter->restart_queue;
1172 		}
1173 	}
1174 
1175 	if (adapter->detect_tx_hung) {
1176 		/*
1177 		 * Detect a transmit hang in hardware, this serializes the
1178 		 * check with the clearing of time_stamp and movement of i
1179 		 */
1180 		adapter->detect_tx_hung = false;
1181 		if (tx_ring->buffer_info[i].time_stamp &&
1182 		    time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1183 			       + (adapter->tx_timeout_factor * HZ)) &&
1184 		    !(er32(STATUS) & E1000_STATUS_TXOFF))
1185 			schedule_work(&adapter->print_hang_task);
1186 		else
1187 			adapter->tx_hang_recheck = false;
1188 	}
1189 	adapter->total_tx_bytes += total_tx_bytes;
1190 	adapter->total_tx_packets += total_tx_packets;
1191 	return count < tx_ring->count;
1192 }
1193 
1194 /**
1195  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1196  * @rx_ring: Rx descriptor ring
1197  *
1198  * the return value indicates whether actual cleaning was done, there
1199  * is no guarantee that everything was cleaned
1200  **/
1201 static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
1202 				  int work_to_do)
1203 {
1204 	struct e1000_adapter *adapter = rx_ring->adapter;
1205 	struct e1000_hw *hw = &adapter->hw;
1206 	union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1207 	struct net_device *netdev = adapter->netdev;
1208 	struct pci_dev *pdev = adapter->pdev;
1209 	struct e1000_buffer *buffer_info, *next_buffer;
1210 	struct e1000_ps_page *ps_page;
1211 	struct sk_buff *skb;
1212 	unsigned int i, j;
1213 	u32 length, staterr;
1214 	int cleaned_count = 0;
1215 	bool cleaned = false;
1216 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1217 
1218 	i = rx_ring->next_to_clean;
1219 	rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1220 	staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1221 	buffer_info = &rx_ring->buffer_info[i];
1222 
1223 	while (staterr & E1000_RXD_STAT_DD) {
1224 		if (*work_done >= work_to_do)
1225 			break;
1226 		(*work_done)++;
1227 		skb = buffer_info->skb;
1228 		rmb();	/* read descriptor and rx_buffer_info after status DD */
1229 
1230 		/* in the packet split case this is header only */
1231 		prefetch(skb->data - NET_IP_ALIGN);
1232 
1233 		i++;
1234 		if (i == rx_ring->count)
1235 			i = 0;
1236 		next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1237 		prefetch(next_rxd);
1238 
1239 		next_buffer = &rx_ring->buffer_info[i];
1240 
1241 		cleaned = true;
1242 		cleaned_count++;
1243 		dma_unmap_single(&pdev->dev, buffer_info->dma,
1244 				 adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1245 		buffer_info->dma = 0;
1246 
1247 		/* see !EOP comment in other Rx routine */
1248 		if (!(staterr & E1000_RXD_STAT_EOP))
1249 			adapter->flags2 |= FLAG2_IS_DISCARDING;
1250 
1251 		if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1252 			e_dbg("Packet Split buffers didn't pick up the full packet\n");
1253 			dev_kfree_skb_irq(skb);
1254 			if (staterr & E1000_RXD_STAT_EOP)
1255 				adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1256 			goto next_desc;
1257 		}
1258 
1259 		if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1260 			     !(netdev->features & NETIF_F_RXALL))) {
1261 			dev_kfree_skb_irq(skb);
1262 			goto next_desc;
1263 		}
1264 
1265 		length = le16_to_cpu(rx_desc->wb.middle.length0);
1266 
1267 		if (!length) {
1268 			e_dbg("Last part of the packet spanning multiple descriptors\n");
1269 			dev_kfree_skb_irq(skb);
1270 			goto next_desc;
1271 		}
1272 
1273 		/* Good Receive */
1274 		skb_put(skb, length);
1275 
1276 		{
1277 			/*
1278 			 * this looks ugly, but it seems compiler issues make
1279 			 * it more efficient than reusing j
1280 			 */
1281 			int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1282 
1283 			/*
1284 			 * page alloc/put takes too long and effects small
1285 			 * packet throughput, so unsplit small packets and
1286 			 * save the alloc/put only valid in softirq (napi)
1287 			 * context to call kmap_*
1288 			 */
1289 			if (l1 && (l1 <= copybreak) &&
1290 			    ((length + l1) <= adapter->rx_ps_bsize0)) {
1291 				u8 *vaddr;
1292 
1293 				ps_page = &buffer_info->ps_pages[0];
1294 
1295 				/*
1296 				 * there is no documentation about how to call
1297 				 * kmap_atomic, so we can't hold the mapping
1298 				 * very long
1299 				 */
1300 				dma_sync_single_for_cpu(&pdev->dev,
1301 							ps_page->dma,
1302 							PAGE_SIZE,
1303 							DMA_FROM_DEVICE);
1304 				vaddr = kmap_atomic(ps_page->page);
1305 				memcpy(skb_tail_pointer(skb), vaddr, l1);
1306 				kunmap_atomic(vaddr);
1307 				dma_sync_single_for_device(&pdev->dev,
1308 							   ps_page->dma,
1309 							   PAGE_SIZE,
1310 							   DMA_FROM_DEVICE);
1311 
1312 				/* remove the CRC */
1313 				if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1314 					if (!(netdev->features & NETIF_F_RXFCS))
1315 						l1 -= 4;
1316 				}
1317 
1318 				skb_put(skb, l1);
1319 				goto copydone;
1320 			} /* if */
1321 		}
1322 
1323 		for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1324 			length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1325 			if (!length)
1326 				break;
1327 
1328 			ps_page = &buffer_info->ps_pages[j];
1329 			dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1330 				       DMA_FROM_DEVICE);
1331 			ps_page->dma = 0;
1332 			skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1333 			ps_page->page = NULL;
1334 			skb->len += length;
1335 			skb->data_len += length;
1336 			skb->truesize += PAGE_SIZE;
1337 		}
1338 
1339 		/* strip the ethernet crc, problem is we're using pages now so
1340 		 * this whole operation can get a little cpu intensive
1341 		 */
1342 		if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1343 			if (!(netdev->features & NETIF_F_RXFCS))
1344 				pskb_trim(skb, skb->len - 4);
1345 		}
1346 
1347 copydone:
1348 		total_rx_bytes += skb->len;
1349 		total_rx_packets++;
1350 
1351 		e1000_rx_checksum(adapter, staterr,
1352 				  rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
1353 
1354 		e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1355 
1356 		if (rx_desc->wb.upper.header_status &
1357 			   cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1358 			adapter->rx_hdr_split++;
1359 
1360 		e1000_receive_skb(adapter, netdev, skb,
1361 				  staterr, rx_desc->wb.middle.vlan);
1362 
1363 next_desc:
1364 		rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1365 		buffer_info->skb = NULL;
1366 
1367 		/* return some buffers to hardware, one at a time is too slow */
1368 		if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1369 			adapter->alloc_rx_buf(rx_ring, cleaned_count,
1370 					      GFP_ATOMIC);
1371 			cleaned_count = 0;
1372 		}
1373 
1374 		/* use prefetched values */
1375 		rx_desc = next_rxd;
1376 		buffer_info = next_buffer;
1377 
1378 		staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1379 	}
1380 	rx_ring->next_to_clean = i;
1381 
1382 	cleaned_count = e1000_desc_unused(rx_ring);
1383 	if (cleaned_count)
1384 		adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1385 
1386 	adapter->total_rx_bytes += total_rx_bytes;
1387 	adapter->total_rx_packets += total_rx_packets;
1388 	return cleaned;
1389 }
1390 
1391 /**
1392  * e1000_consume_page - helper function
1393  **/
1394 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1395                                u16 length)
1396 {
1397 	bi->page = NULL;
1398 	skb->len += length;
1399 	skb->data_len += length;
1400 	skb->truesize += PAGE_SIZE;
1401 }
1402 
1403 /**
1404  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1405  * @adapter: board private structure
1406  *
1407  * the return value indicates whether actual cleaning was done, there
1408  * is no guarantee that everything was cleaned
1409  **/
1410 static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1411 				     int work_to_do)
1412 {
1413 	struct e1000_adapter *adapter = rx_ring->adapter;
1414 	struct net_device *netdev = adapter->netdev;
1415 	struct pci_dev *pdev = adapter->pdev;
1416 	union e1000_rx_desc_extended *rx_desc, *next_rxd;
1417 	struct e1000_buffer *buffer_info, *next_buffer;
1418 	u32 length, staterr;
1419 	unsigned int i;
1420 	int cleaned_count = 0;
1421 	bool cleaned = false;
1422 	unsigned int total_rx_bytes=0, total_rx_packets=0;
1423 
1424 	i = rx_ring->next_to_clean;
1425 	rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1426 	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1427 	buffer_info = &rx_ring->buffer_info[i];
1428 
1429 	while (staterr & E1000_RXD_STAT_DD) {
1430 		struct sk_buff *skb;
1431 
1432 		if (*work_done >= work_to_do)
1433 			break;
1434 		(*work_done)++;
1435 		rmb();	/* read descriptor and rx_buffer_info after status DD */
1436 
1437 		skb = buffer_info->skb;
1438 		buffer_info->skb = NULL;
1439 
1440 		++i;
1441 		if (i == rx_ring->count)
1442 			i = 0;
1443 		next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1444 		prefetch(next_rxd);
1445 
1446 		next_buffer = &rx_ring->buffer_info[i];
1447 
1448 		cleaned = true;
1449 		cleaned_count++;
1450 		dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1451 			       DMA_FROM_DEVICE);
1452 		buffer_info->dma = 0;
1453 
1454 		length = le16_to_cpu(rx_desc->wb.upper.length);
1455 
1456 		/* errors is only valid for DD + EOP descriptors */
1457 		if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1458 			     ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1459 			      !(netdev->features & NETIF_F_RXALL)))) {
1460 			/* recycle both page and skb */
1461 			buffer_info->skb = skb;
1462 			/* an error means any chain goes out the window too */
1463 			if (rx_ring->rx_skb_top)
1464 				dev_kfree_skb_irq(rx_ring->rx_skb_top);
1465 			rx_ring->rx_skb_top = NULL;
1466 			goto next_desc;
1467 		}
1468 
1469 #define rxtop (rx_ring->rx_skb_top)
1470 		if (!(staterr & E1000_RXD_STAT_EOP)) {
1471 			/* this descriptor is only the beginning (or middle) */
1472 			if (!rxtop) {
1473 				/* this is the beginning of a chain */
1474 				rxtop = skb;
1475 				skb_fill_page_desc(rxtop, 0, buffer_info->page,
1476 				                   0, length);
1477 			} else {
1478 				/* this is the middle of a chain */
1479 				skb_fill_page_desc(rxtop,
1480 				    skb_shinfo(rxtop)->nr_frags,
1481 				    buffer_info->page, 0, length);
1482 				/* re-use the skb, only consumed the page */
1483 				buffer_info->skb = skb;
1484 			}
1485 			e1000_consume_page(buffer_info, rxtop, length);
1486 			goto next_desc;
1487 		} else {
1488 			if (rxtop) {
1489 				/* end of the chain */
1490 				skb_fill_page_desc(rxtop,
1491 				    skb_shinfo(rxtop)->nr_frags,
1492 				    buffer_info->page, 0, length);
1493 				/* re-use the current skb, we only consumed the
1494 				 * page */
1495 				buffer_info->skb = skb;
1496 				skb = rxtop;
1497 				rxtop = NULL;
1498 				e1000_consume_page(buffer_info, skb, length);
1499 			} else {
1500 				/* no chain, got EOP, this buf is the packet
1501 				 * copybreak to save the put_page/alloc_page */
1502 				if (length <= copybreak &&
1503 				    skb_tailroom(skb) >= length) {
1504 					u8 *vaddr;
1505 					vaddr = kmap_atomic(buffer_info->page);
1506 					memcpy(skb_tail_pointer(skb), vaddr,
1507 					       length);
1508 					kunmap_atomic(vaddr);
1509 					/* re-use the page, so don't erase
1510 					 * buffer_info->page */
1511 					skb_put(skb, length);
1512 				} else {
1513 					skb_fill_page_desc(skb, 0,
1514 					                   buffer_info->page, 0,
1515 				                           length);
1516 					e1000_consume_page(buffer_info, skb,
1517 					                   length);
1518 				}
1519 			}
1520 		}
1521 
1522 		/* Receive Checksum Offload XXX recompute due to CRC strip? */
1523 		e1000_rx_checksum(adapter, staterr,
1524 				  rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
1525 
1526 		e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1527 
1528 		/* probably a little skewed due to removing CRC */
1529 		total_rx_bytes += skb->len;
1530 		total_rx_packets++;
1531 
1532 		/* eth type trans needs skb->data to point to something */
1533 		if (!pskb_may_pull(skb, ETH_HLEN)) {
1534 			e_err("pskb_may_pull failed.\n");
1535 			dev_kfree_skb_irq(skb);
1536 			goto next_desc;
1537 		}
1538 
1539 		e1000_receive_skb(adapter, netdev, skb, staterr,
1540 				  rx_desc->wb.upper.vlan);
1541 
1542 next_desc:
1543 		rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1544 
1545 		/* return some buffers to hardware, one at a time is too slow */
1546 		if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1547 			adapter->alloc_rx_buf(rx_ring, cleaned_count,
1548 					      GFP_ATOMIC);
1549 			cleaned_count = 0;
1550 		}
1551 
1552 		/* use prefetched values */
1553 		rx_desc = next_rxd;
1554 		buffer_info = next_buffer;
1555 
1556 		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1557 	}
1558 	rx_ring->next_to_clean = i;
1559 
1560 	cleaned_count = e1000_desc_unused(rx_ring);
1561 	if (cleaned_count)
1562 		adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1563 
1564 	adapter->total_rx_bytes += total_rx_bytes;
1565 	adapter->total_rx_packets += total_rx_packets;
1566 	return cleaned;
1567 }
1568 
1569 /**
1570  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1571  * @rx_ring: Rx descriptor ring
1572  **/
1573 static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1574 {
1575 	struct e1000_adapter *adapter = rx_ring->adapter;
1576 	struct e1000_buffer *buffer_info;
1577 	struct e1000_ps_page *ps_page;
1578 	struct pci_dev *pdev = adapter->pdev;
1579 	unsigned int i, j;
1580 
1581 	/* Free all the Rx ring sk_buffs */
1582 	for (i = 0; i < rx_ring->count; i++) {
1583 		buffer_info = &rx_ring->buffer_info[i];
1584 		if (buffer_info->dma) {
1585 			if (adapter->clean_rx == e1000_clean_rx_irq)
1586 				dma_unmap_single(&pdev->dev, buffer_info->dma,
1587 						 adapter->rx_buffer_len,
1588 						 DMA_FROM_DEVICE);
1589 			else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1590 				dma_unmap_page(&pdev->dev, buffer_info->dma,
1591 				               PAGE_SIZE,
1592 					       DMA_FROM_DEVICE);
1593 			else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1594 				dma_unmap_single(&pdev->dev, buffer_info->dma,
1595 						 adapter->rx_ps_bsize0,
1596 						 DMA_FROM_DEVICE);
1597 			buffer_info->dma = 0;
1598 		}
1599 
1600 		if (buffer_info->page) {
1601 			put_page(buffer_info->page);
1602 			buffer_info->page = NULL;
1603 		}
1604 
1605 		if (buffer_info->skb) {
1606 			dev_kfree_skb(buffer_info->skb);
1607 			buffer_info->skb = NULL;
1608 		}
1609 
1610 		for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1611 			ps_page = &buffer_info->ps_pages[j];
1612 			if (!ps_page->page)
1613 				break;
1614 			dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1615 				       DMA_FROM_DEVICE);
1616 			ps_page->dma = 0;
1617 			put_page(ps_page->page);
1618 			ps_page->page = NULL;
1619 		}
1620 	}
1621 
1622 	/* there also may be some cached data from a chained receive */
1623 	if (rx_ring->rx_skb_top) {
1624 		dev_kfree_skb(rx_ring->rx_skb_top);
1625 		rx_ring->rx_skb_top = NULL;
1626 	}
1627 
1628 	/* Zero out the descriptor ring */
1629 	memset(rx_ring->desc, 0, rx_ring->size);
1630 
1631 	rx_ring->next_to_clean = 0;
1632 	rx_ring->next_to_use = 0;
1633 	adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1634 
1635 	writel(0, rx_ring->head);
1636 	writel(0, rx_ring->tail);
1637 }
1638 
1639 static void e1000e_downshift_workaround(struct work_struct *work)
1640 {
1641 	struct e1000_adapter *adapter = container_of(work,
1642 					struct e1000_adapter, downshift_task);
1643 
1644 	if (test_bit(__E1000_DOWN, &adapter->state))
1645 		return;
1646 
1647 	e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1648 }
1649 
1650 /**
1651  * e1000_intr_msi - Interrupt Handler
1652  * @irq: interrupt number
1653  * @data: pointer to a network interface device structure
1654  **/
1655 static irqreturn_t e1000_intr_msi(int irq, void *data)
1656 {
1657 	struct net_device *netdev = data;
1658 	struct e1000_adapter *adapter = netdev_priv(netdev);
1659 	struct e1000_hw *hw = &adapter->hw;
1660 	u32 icr = er32(ICR);
1661 
1662 	/*
1663 	 * read ICR disables interrupts using IAM
1664 	 */
1665 
1666 	if (icr & E1000_ICR_LSC) {
1667 		hw->mac.get_link_status = true;
1668 		/*
1669 		 * ICH8 workaround-- Call gig speed drop workaround on cable
1670 		 * disconnect (LSC) before accessing any PHY registers
1671 		 */
1672 		if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1673 		    (!(er32(STATUS) & E1000_STATUS_LU)))
1674 			schedule_work(&adapter->downshift_task);
1675 
1676 		/*
1677 		 * 80003ES2LAN workaround-- For packet buffer work-around on
1678 		 * link down event; disable receives here in the ISR and reset
1679 		 * adapter in watchdog
1680 		 */
1681 		if (netif_carrier_ok(netdev) &&
1682 		    adapter->flags & FLAG_RX_NEEDS_RESTART) {
1683 			/* disable receives */
1684 			u32 rctl = er32(RCTL);
1685 			ew32(RCTL, rctl & ~E1000_RCTL_EN);
1686 			adapter->flags |= FLAG_RX_RESTART_NOW;
1687 		}
1688 		/* guard against interrupt when we're going down */
1689 		if (!test_bit(__E1000_DOWN, &adapter->state))
1690 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
1691 	}
1692 
1693 	if (napi_schedule_prep(&adapter->napi)) {
1694 		adapter->total_tx_bytes = 0;
1695 		adapter->total_tx_packets = 0;
1696 		adapter->total_rx_bytes = 0;
1697 		adapter->total_rx_packets = 0;
1698 		__napi_schedule(&adapter->napi);
1699 	}
1700 
1701 	return IRQ_HANDLED;
1702 }
1703 
1704 /**
1705  * e1000_intr - Interrupt Handler
1706  * @irq: interrupt number
1707  * @data: pointer to a network interface device structure
1708  **/
1709 static irqreturn_t e1000_intr(int irq, void *data)
1710 {
1711 	struct net_device *netdev = data;
1712 	struct e1000_adapter *adapter = netdev_priv(netdev);
1713 	struct e1000_hw *hw = &adapter->hw;
1714 	u32 rctl, icr = er32(ICR);
1715 
1716 	if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1717 		return IRQ_NONE;  /* Not our interrupt */
1718 
1719 	/*
1720 	 * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1721 	 * not set, then the adapter didn't send an interrupt
1722 	 */
1723 	if (!(icr & E1000_ICR_INT_ASSERTED))
1724 		return IRQ_NONE;
1725 
1726 	/*
1727 	 * Interrupt Auto-Mask...upon reading ICR,
1728 	 * interrupts are masked.  No need for the
1729 	 * IMC write
1730 	 */
1731 
1732 	if (icr & E1000_ICR_LSC) {
1733 		hw->mac.get_link_status = true;
1734 		/*
1735 		 * ICH8 workaround-- Call gig speed drop workaround on cable
1736 		 * disconnect (LSC) before accessing any PHY registers
1737 		 */
1738 		if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1739 		    (!(er32(STATUS) & E1000_STATUS_LU)))
1740 			schedule_work(&adapter->downshift_task);
1741 
1742 		/*
1743 		 * 80003ES2LAN workaround--
1744 		 * For packet buffer work-around on link down event;
1745 		 * disable receives here in the ISR and
1746 		 * reset adapter in watchdog
1747 		 */
1748 		if (netif_carrier_ok(netdev) &&
1749 		    (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1750 			/* disable receives */
1751 			rctl = er32(RCTL);
1752 			ew32(RCTL, rctl & ~E1000_RCTL_EN);
1753 			adapter->flags |= FLAG_RX_RESTART_NOW;
1754 		}
1755 		/* guard against interrupt when we're going down */
1756 		if (!test_bit(__E1000_DOWN, &adapter->state))
1757 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
1758 	}
1759 
1760 	if (napi_schedule_prep(&adapter->napi)) {
1761 		adapter->total_tx_bytes = 0;
1762 		adapter->total_tx_packets = 0;
1763 		adapter->total_rx_bytes = 0;
1764 		adapter->total_rx_packets = 0;
1765 		__napi_schedule(&adapter->napi);
1766 	}
1767 
1768 	return IRQ_HANDLED;
1769 }
1770 
1771 static irqreturn_t e1000_msix_other(int irq, void *data)
1772 {
1773 	struct net_device *netdev = data;
1774 	struct e1000_adapter *adapter = netdev_priv(netdev);
1775 	struct e1000_hw *hw = &adapter->hw;
1776 	u32 icr = er32(ICR);
1777 
1778 	if (!(icr & E1000_ICR_INT_ASSERTED)) {
1779 		if (!test_bit(__E1000_DOWN, &adapter->state))
1780 			ew32(IMS, E1000_IMS_OTHER);
1781 		return IRQ_NONE;
1782 	}
1783 
1784 	if (icr & adapter->eiac_mask)
1785 		ew32(ICS, (icr & adapter->eiac_mask));
1786 
1787 	if (icr & E1000_ICR_OTHER) {
1788 		if (!(icr & E1000_ICR_LSC))
1789 			goto no_link_interrupt;
1790 		hw->mac.get_link_status = true;
1791 		/* guard against interrupt when we're going down */
1792 		if (!test_bit(__E1000_DOWN, &adapter->state))
1793 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
1794 	}
1795 
1796 no_link_interrupt:
1797 	if (!test_bit(__E1000_DOWN, &adapter->state))
1798 		ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1799 
1800 	return IRQ_HANDLED;
1801 }
1802 
1803 
1804 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1805 {
1806 	struct net_device *netdev = data;
1807 	struct e1000_adapter *adapter = netdev_priv(netdev);
1808 	struct e1000_hw *hw = &adapter->hw;
1809 	struct e1000_ring *tx_ring = adapter->tx_ring;
1810 
1811 
1812 	adapter->total_tx_bytes = 0;
1813 	adapter->total_tx_packets = 0;
1814 
1815 	if (!e1000_clean_tx_irq(tx_ring))
1816 		/* Ring was not completely cleaned, so fire another interrupt */
1817 		ew32(ICS, tx_ring->ims_val);
1818 
1819 	return IRQ_HANDLED;
1820 }
1821 
1822 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1823 {
1824 	struct net_device *netdev = data;
1825 	struct e1000_adapter *adapter = netdev_priv(netdev);
1826 	struct e1000_ring *rx_ring = adapter->rx_ring;
1827 
1828 	/* Write the ITR value calculated at the end of the
1829 	 * previous interrupt.
1830 	 */
1831 	if (rx_ring->set_itr) {
1832 		writel(1000000000 / (rx_ring->itr_val * 256),
1833 		       rx_ring->itr_register);
1834 		rx_ring->set_itr = 0;
1835 	}
1836 
1837 	if (napi_schedule_prep(&adapter->napi)) {
1838 		adapter->total_rx_bytes = 0;
1839 		adapter->total_rx_packets = 0;
1840 		__napi_schedule(&adapter->napi);
1841 	}
1842 	return IRQ_HANDLED;
1843 }
1844 
1845 /**
1846  * e1000_configure_msix - Configure MSI-X hardware
1847  *
1848  * e1000_configure_msix sets up the hardware to properly
1849  * generate MSI-X interrupts.
1850  **/
1851 static void e1000_configure_msix(struct e1000_adapter *adapter)
1852 {
1853 	struct e1000_hw *hw = &adapter->hw;
1854 	struct e1000_ring *rx_ring = adapter->rx_ring;
1855 	struct e1000_ring *tx_ring = adapter->tx_ring;
1856 	int vector = 0;
1857 	u32 ctrl_ext, ivar = 0;
1858 
1859 	adapter->eiac_mask = 0;
1860 
1861 	/* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1862 	if (hw->mac.type == e1000_82574) {
1863 		u32 rfctl = er32(RFCTL);
1864 		rfctl |= E1000_RFCTL_ACK_DIS;
1865 		ew32(RFCTL, rfctl);
1866 	}
1867 
1868 #define E1000_IVAR_INT_ALLOC_VALID	0x8
1869 	/* Configure Rx vector */
1870 	rx_ring->ims_val = E1000_IMS_RXQ0;
1871 	adapter->eiac_mask |= rx_ring->ims_val;
1872 	if (rx_ring->itr_val)
1873 		writel(1000000000 / (rx_ring->itr_val * 256),
1874 		       rx_ring->itr_register);
1875 	else
1876 		writel(1, rx_ring->itr_register);
1877 	ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1878 
1879 	/* Configure Tx vector */
1880 	tx_ring->ims_val = E1000_IMS_TXQ0;
1881 	vector++;
1882 	if (tx_ring->itr_val)
1883 		writel(1000000000 / (tx_ring->itr_val * 256),
1884 		       tx_ring->itr_register);
1885 	else
1886 		writel(1, tx_ring->itr_register);
1887 	adapter->eiac_mask |= tx_ring->ims_val;
1888 	ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1889 
1890 	/* set vector for Other Causes, e.g. link changes */
1891 	vector++;
1892 	ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1893 	if (rx_ring->itr_val)
1894 		writel(1000000000 / (rx_ring->itr_val * 256),
1895 		       hw->hw_addr + E1000_EITR_82574(vector));
1896 	else
1897 		writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1898 
1899 	/* Cause Tx interrupts on every write back */
1900 	ivar |= (1 << 31);
1901 
1902 	ew32(IVAR, ivar);
1903 
1904 	/* enable MSI-X PBA support */
1905 	ctrl_ext = er32(CTRL_EXT);
1906 	ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1907 
1908 	/* Auto-Mask Other interrupts upon ICR read */
1909 #define E1000_EIAC_MASK_82574   0x01F00000
1910 	ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1911 	ctrl_ext |= E1000_CTRL_EXT_EIAME;
1912 	ew32(CTRL_EXT, ctrl_ext);
1913 	e1e_flush();
1914 }
1915 
1916 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1917 {
1918 	if (adapter->msix_entries) {
1919 		pci_disable_msix(adapter->pdev);
1920 		kfree(adapter->msix_entries);
1921 		adapter->msix_entries = NULL;
1922 	} else if (adapter->flags & FLAG_MSI_ENABLED) {
1923 		pci_disable_msi(adapter->pdev);
1924 		adapter->flags &= ~FLAG_MSI_ENABLED;
1925 	}
1926 }
1927 
1928 /**
1929  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1930  *
1931  * Attempt to configure interrupts using the best available
1932  * capabilities of the hardware and kernel.
1933  **/
1934 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1935 {
1936 	int err;
1937 	int i;
1938 
1939 	switch (adapter->int_mode) {
1940 	case E1000E_INT_MODE_MSIX:
1941 		if (adapter->flags & FLAG_HAS_MSIX) {
1942 			adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1943 			adapter->msix_entries = kcalloc(adapter->num_vectors,
1944 						      sizeof(struct msix_entry),
1945 						      GFP_KERNEL);
1946 			if (adapter->msix_entries) {
1947 				for (i = 0; i < adapter->num_vectors; i++)
1948 					adapter->msix_entries[i].entry = i;
1949 
1950 				err = pci_enable_msix(adapter->pdev,
1951 						      adapter->msix_entries,
1952 						      adapter->num_vectors);
1953 				if (err == 0)
1954 					return;
1955 			}
1956 			/* MSI-X failed, so fall through and try MSI */
1957 			e_err("Failed to initialize MSI-X interrupts.  Falling back to MSI interrupts.\n");
1958 			e1000e_reset_interrupt_capability(adapter);
1959 		}
1960 		adapter->int_mode = E1000E_INT_MODE_MSI;
1961 		/* Fall through */
1962 	case E1000E_INT_MODE_MSI:
1963 		if (!pci_enable_msi(adapter->pdev)) {
1964 			adapter->flags |= FLAG_MSI_ENABLED;
1965 		} else {
1966 			adapter->int_mode = E1000E_INT_MODE_LEGACY;
1967 			e_err("Failed to initialize MSI interrupts.  Falling back to legacy interrupts.\n");
1968 		}
1969 		/* Fall through */
1970 	case E1000E_INT_MODE_LEGACY:
1971 		/* Don't do anything; this is the system default */
1972 		break;
1973 	}
1974 
1975 	/* store the number of vectors being used */
1976 	adapter->num_vectors = 1;
1977 }
1978 
1979 /**
1980  * e1000_request_msix - Initialize MSI-X interrupts
1981  *
1982  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1983  * kernel.
1984  **/
1985 static int e1000_request_msix(struct e1000_adapter *adapter)
1986 {
1987 	struct net_device *netdev = adapter->netdev;
1988 	int err = 0, vector = 0;
1989 
1990 	if (strlen(netdev->name) < (IFNAMSIZ - 5))
1991 		snprintf(adapter->rx_ring->name,
1992 			 sizeof(adapter->rx_ring->name) - 1,
1993 			 "%s-rx-0", netdev->name);
1994 	else
1995 		memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1996 	err = request_irq(adapter->msix_entries[vector].vector,
1997 			  e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1998 			  netdev);
1999 	if (err)
2000 		return err;
2001 	adapter->rx_ring->itr_register = adapter->hw.hw_addr +
2002 	    E1000_EITR_82574(vector);
2003 	adapter->rx_ring->itr_val = adapter->itr;
2004 	vector++;
2005 
2006 	if (strlen(netdev->name) < (IFNAMSIZ - 5))
2007 		snprintf(adapter->tx_ring->name,
2008 			 sizeof(adapter->tx_ring->name) - 1,
2009 			 "%s-tx-0", netdev->name);
2010 	else
2011 		memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
2012 	err = request_irq(adapter->msix_entries[vector].vector,
2013 			  e1000_intr_msix_tx, 0, adapter->tx_ring->name,
2014 			  netdev);
2015 	if (err)
2016 		return err;
2017 	adapter->tx_ring->itr_register = adapter->hw.hw_addr +
2018 	    E1000_EITR_82574(vector);
2019 	adapter->tx_ring->itr_val = adapter->itr;
2020 	vector++;
2021 
2022 	err = request_irq(adapter->msix_entries[vector].vector,
2023 			  e1000_msix_other, 0, netdev->name, netdev);
2024 	if (err)
2025 		return err;
2026 
2027 	e1000_configure_msix(adapter);
2028 
2029 	return 0;
2030 }
2031 
2032 /**
2033  * e1000_request_irq - initialize interrupts
2034  *
2035  * Attempts to configure interrupts using the best available
2036  * capabilities of the hardware and kernel.
2037  **/
2038 static int e1000_request_irq(struct e1000_adapter *adapter)
2039 {
2040 	struct net_device *netdev = adapter->netdev;
2041 	int err;
2042 
2043 	if (adapter->msix_entries) {
2044 		err = e1000_request_msix(adapter);
2045 		if (!err)
2046 			return err;
2047 		/* fall back to MSI */
2048 		e1000e_reset_interrupt_capability(adapter);
2049 		adapter->int_mode = E1000E_INT_MODE_MSI;
2050 		e1000e_set_interrupt_capability(adapter);
2051 	}
2052 	if (adapter->flags & FLAG_MSI_ENABLED) {
2053 		err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2054 				  netdev->name, netdev);
2055 		if (!err)
2056 			return err;
2057 
2058 		/* fall back to legacy interrupt */
2059 		e1000e_reset_interrupt_capability(adapter);
2060 		adapter->int_mode = E1000E_INT_MODE_LEGACY;
2061 	}
2062 
2063 	err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2064 			  netdev->name, netdev);
2065 	if (err)
2066 		e_err("Unable to allocate interrupt, Error: %d\n", err);
2067 
2068 	return err;
2069 }
2070 
2071 static void e1000_free_irq(struct e1000_adapter *adapter)
2072 {
2073 	struct net_device *netdev = adapter->netdev;
2074 
2075 	if (adapter->msix_entries) {
2076 		int vector = 0;
2077 
2078 		free_irq(adapter->msix_entries[vector].vector, netdev);
2079 		vector++;
2080 
2081 		free_irq(adapter->msix_entries[vector].vector, netdev);
2082 		vector++;
2083 
2084 		/* Other Causes interrupt vector */
2085 		free_irq(adapter->msix_entries[vector].vector, netdev);
2086 		return;
2087 	}
2088 
2089 	free_irq(adapter->pdev->irq, netdev);
2090 }
2091 
2092 /**
2093  * e1000_irq_disable - Mask off interrupt generation on the NIC
2094  **/
2095 static void e1000_irq_disable(struct e1000_adapter *adapter)
2096 {
2097 	struct e1000_hw *hw = &adapter->hw;
2098 
2099 	ew32(IMC, ~0);
2100 	if (adapter->msix_entries)
2101 		ew32(EIAC_82574, 0);
2102 	e1e_flush();
2103 
2104 	if (adapter->msix_entries) {
2105 		int i;
2106 		for (i = 0; i < adapter->num_vectors; i++)
2107 			synchronize_irq(adapter->msix_entries[i].vector);
2108 	} else {
2109 		synchronize_irq(adapter->pdev->irq);
2110 	}
2111 }
2112 
2113 /**
2114  * e1000_irq_enable - Enable default interrupt generation settings
2115  **/
2116 static void e1000_irq_enable(struct e1000_adapter *adapter)
2117 {
2118 	struct e1000_hw *hw = &adapter->hw;
2119 
2120 	if (adapter->msix_entries) {
2121 		ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2122 		ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
2123 	} else {
2124 		ew32(IMS, IMS_ENABLE_MASK);
2125 	}
2126 	e1e_flush();
2127 }
2128 
2129 /**
2130  * e1000e_get_hw_control - get control of the h/w from f/w
2131  * @adapter: address of board private structure
2132  *
2133  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2134  * For ASF and Pass Through versions of f/w this means that
2135  * the driver is loaded. For AMT version (only with 82573)
2136  * of the f/w this means that the network i/f is open.
2137  **/
2138 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2139 {
2140 	struct e1000_hw *hw = &adapter->hw;
2141 	u32 ctrl_ext;
2142 	u32 swsm;
2143 
2144 	/* Let firmware know the driver has taken over */
2145 	if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2146 		swsm = er32(SWSM);
2147 		ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2148 	} else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2149 		ctrl_ext = er32(CTRL_EXT);
2150 		ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2151 	}
2152 }
2153 
2154 /**
2155  * e1000e_release_hw_control - release control of the h/w to f/w
2156  * @adapter: address of board private structure
2157  *
2158  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2159  * For ASF and Pass Through versions of f/w this means that the
2160  * driver is no longer loaded. For AMT version (only with 82573) i
2161  * of the f/w this means that the network i/f is closed.
2162  *
2163  **/
2164 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2165 {
2166 	struct e1000_hw *hw = &adapter->hw;
2167 	u32 ctrl_ext;
2168 	u32 swsm;
2169 
2170 	/* Let firmware taken over control of h/w */
2171 	if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2172 		swsm = er32(SWSM);
2173 		ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2174 	} else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2175 		ctrl_ext = er32(CTRL_EXT);
2176 		ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2177 	}
2178 }
2179 
2180 /**
2181  * @e1000_alloc_ring - allocate memory for a ring structure
2182  **/
2183 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2184 				struct e1000_ring *ring)
2185 {
2186 	struct pci_dev *pdev = adapter->pdev;
2187 
2188 	ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2189 					GFP_KERNEL);
2190 	if (!ring->desc)
2191 		return -ENOMEM;
2192 
2193 	return 0;
2194 }
2195 
2196 /**
2197  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2198  * @tx_ring: Tx descriptor ring
2199  *
2200  * Return 0 on success, negative on failure
2201  **/
2202 int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2203 {
2204 	struct e1000_adapter *adapter = tx_ring->adapter;
2205 	int err = -ENOMEM, size;
2206 
2207 	size = sizeof(struct e1000_buffer) * tx_ring->count;
2208 	tx_ring->buffer_info = vzalloc(size);
2209 	if (!tx_ring->buffer_info)
2210 		goto err;
2211 
2212 	/* round up to nearest 4K */
2213 	tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2214 	tx_ring->size = ALIGN(tx_ring->size, 4096);
2215 
2216 	err = e1000_alloc_ring_dma(adapter, tx_ring);
2217 	if (err)
2218 		goto err;
2219 
2220 	tx_ring->next_to_use = 0;
2221 	tx_ring->next_to_clean = 0;
2222 
2223 	return 0;
2224 err:
2225 	vfree(tx_ring->buffer_info);
2226 	e_err("Unable to allocate memory for the transmit descriptor ring\n");
2227 	return err;
2228 }
2229 
2230 /**
2231  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2232  * @rx_ring: Rx descriptor ring
2233  *
2234  * Returns 0 on success, negative on failure
2235  **/
2236 int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2237 {
2238 	struct e1000_adapter *adapter = rx_ring->adapter;
2239 	struct e1000_buffer *buffer_info;
2240 	int i, size, desc_len, err = -ENOMEM;
2241 
2242 	size = sizeof(struct e1000_buffer) * rx_ring->count;
2243 	rx_ring->buffer_info = vzalloc(size);
2244 	if (!rx_ring->buffer_info)
2245 		goto err;
2246 
2247 	for (i = 0; i < rx_ring->count; i++) {
2248 		buffer_info = &rx_ring->buffer_info[i];
2249 		buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2250 						sizeof(struct e1000_ps_page),
2251 						GFP_KERNEL);
2252 		if (!buffer_info->ps_pages)
2253 			goto err_pages;
2254 	}
2255 
2256 	desc_len = sizeof(union e1000_rx_desc_packet_split);
2257 
2258 	/* Round up to nearest 4K */
2259 	rx_ring->size = rx_ring->count * desc_len;
2260 	rx_ring->size = ALIGN(rx_ring->size, 4096);
2261 
2262 	err = e1000_alloc_ring_dma(adapter, rx_ring);
2263 	if (err)
2264 		goto err_pages;
2265 
2266 	rx_ring->next_to_clean = 0;
2267 	rx_ring->next_to_use = 0;
2268 	rx_ring->rx_skb_top = NULL;
2269 
2270 	return 0;
2271 
2272 err_pages:
2273 	for (i = 0; i < rx_ring->count; i++) {
2274 		buffer_info = &rx_ring->buffer_info[i];
2275 		kfree(buffer_info->ps_pages);
2276 	}
2277 err:
2278 	vfree(rx_ring->buffer_info);
2279 	e_err("Unable to allocate memory for the receive descriptor ring\n");
2280 	return err;
2281 }
2282 
2283 /**
2284  * e1000_clean_tx_ring - Free Tx Buffers
2285  * @tx_ring: Tx descriptor ring
2286  **/
2287 static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2288 {
2289 	struct e1000_adapter *adapter = tx_ring->adapter;
2290 	struct e1000_buffer *buffer_info;
2291 	unsigned long size;
2292 	unsigned int i;
2293 
2294 	for (i = 0; i < tx_ring->count; i++) {
2295 		buffer_info = &tx_ring->buffer_info[i];
2296 		e1000_put_txbuf(tx_ring, buffer_info);
2297 	}
2298 
2299 	netdev_reset_queue(adapter->netdev);
2300 	size = sizeof(struct e1000_buffer) * tx_ring->count;
2301 	memset(tx_ring->buffer_info, 0, size);
2302 
2303 	memset(tx_ring->desc, 0, tx_ring->size);
2304 
2305 	tx_ring->next_to_use = 0;
2306 	tx_ring->next_to_clean = 0;
2307 
2308 	writel(0, tx_ring->head);
2309 	writel(0, tx_ring->tail);
2310 }
2311 
2312 /**
2313  * e1000e_free_tx_resources - Free Tx Resources per Queue
2314  * @tx_ring: Tx descriptor ring
2315  *
2316  * Free all transmit software resources
2317  **/
2318 void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2319 {
2320 	struct e1000_adapter *adapter = tx_ring->adapter;
2321 	struct pci_dev *pdev = adapter->pdev;
2322 
2323 	e1000_clean_tx_ring(tx_ring);
2324 
2325 	vfree(tx_ring->buffer_info);
2326 	tx_ring->buffer_info = NULL;
2327 
2328 	dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2329 			  tx_ring->dma);
2330 	tx_ring->desc = NULL;
2331 }
2332 
2333 /**
2334  * e1000e_free_rx_resources - Free Rx Resources
2335  * @rx_ring: Rx descriptor ring
2336  *
2337  * Free all receive software resources
2338  **/
2339 void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2340 {
2341 	struct e1000_adapter *adapter = rx_ring->adapter;
2342 	struct pci_dev *pdev = adapter->pdev;
2343 	int i;
2344 
2345 	e1000_clean_rx_ring(rx_ring);
2346 
2347 	for (i = 0; i < rx_ring->count; i++)
2348 		kfree(rx_ring->buffer_info[i].ps_pages);
2349 
2350 	vfree(rx_ring->buffer_info);
2351 	rx_ring->buffer_info = NULL;
2352 
2353 	dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2354 			  rx_ring->dma);
2355 	rx_ring->desc = NULL;
2356 }
2357 
2358 /**
2359  * e1000_update_itr - update the dynamic ITR value based on statistics
2360  * @adapter: pointer to adapter
2361  * @itr_setting: current adapter->itr
2362  * @packets: the number of packets during this measurement interval
2363  * @bytes: the number of bytes during this measurement interval
2364  *
2365  *      Stores a new ITR value based on packets and byte
2366  *      counts during the last interrupt.  The advantage of per interrupt
2367  *      computation is faster updates and more accurate ITR for the current
2368  *      traffic pattern.  Constants in this function were computed
2369  *      based on theoretical maximum wire speed and thresholds were set based
2370  *      on testing data as well as attempting to minimize response time
2371  *      while increasing bulk throughput.  This functionality is controlled
2372  *      by the InterruptThrottleRate module parameter.
2373  **/
2374 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2375 				     u16 itr_setting, int packets,
2376 				     int bytes)
2377 {
2378 	unsigned int retval = itr_setting;
2379 
2380 	if (packets == 0)
2381 		return itr_setting;
2382 
2383 	switch (itr_setting) {
2384 	case lowest_latency:
2385 		/* handle TSO and jumbo frames */
2386 		if (bytes/packets > 8000)
2387 			retval = bulk_latency;
2388 		else if ((packets < 5) && (bytes > 512))
2389 			retval = low_latency;
2390 		break;
2391 	case low_latency:  /* 50 usec aka 20000 ints/s */
2392 		if (bytes > 10000) {
2393 			/* this if handles the TSO accounting */
2394 			if (bytes/packets > 8000)
2395 				retval = bulk_latency;
2396 			else if ((packets < 10) || ((bytes/packets) > 1200))
2397 				retval = bulk_latency;
2398 			else if ((packets > 35))
2399 				retval = lowest_latency;
2400 		} else if (bytes/packets > 2000) {
2401 			retval = bulk_latency;
2402 		} else if (packets <= 2 && bytes < 512) {
2403 			retval = lowest_latency;
2404 		}
2405 		break;
2406 	case bulk_latency: /* 250 usec aka 4000 ints/s */
2407 		if (bytes > 25000) {
2408 			if (packets > 35)
2409 				retval = low_latency;
2410 		} else if (bytes < 6000) {
2411 			retval = low_latency;
2412 		}
2413 		break;
2414 	}
2415 
2416 	return retval;
2417 }
2418 
2419 static void e1000_set_itr(struct e1000_adapter *adapter)
2420 {
2421 	struct e1000_hw *hw = &adapter->hw;
2422 	u16 current_itr;
2423 	u32 new_itr = adapter->itr;
2424 
2425 	/* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2426 	if (adapter->link_speed != SPEED_1000) {
2427 		current_itr = 0;
2428 		new_itr = 4000;
2429 		goto set_itr_now;
2430 	}
2431 
2432 	if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2433 		new_itr = 0;
2434 		goto set_itr_now;
2435 	}
2436 
2437 	adapter->tx_itr = e1000_update_itr(adapter,
2438 				    adapter->tx_itr,
2439 				    adapter->total_tx_packets,
2440 				    adapter->total_tx_bytes);
2441 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
2442 	if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2443 		adapter->tx_itr = low_latency;
2444 
2445 	adapter->rx_itr = e1000_update_itr(adapter,
2446 				    adapter->rx_itr,
2447 				    adapter->total_rx_packets,
2448 				    adapter->total_rx_bytes);
2449 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
2450 	if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2451 		adapter->rx_itr = low_latency;
2452 
2453 	current_itr = max(adapter->rx_itr, adapter->tx_itr);
2454 
2455 	switch (current_itr) {
2456 	/* counts and packets in update_itr are dependent on these numbers */
2457 	case lowest_latency:
2458 		new_itr = 70000;
2459 		break;
2460 	case low_latency:
2461 		new_itr = 20000; /* aka hwitr = ~200 */
2462 		break;
2463 	case bulk_latency:
2464 		new_itr = 4000;
2465 		break;
2466 	default:
2467 		break;
2468 	}
2469 
2470 set_itr_now:
2471 	if (new_itr != adapter->itr) {
2472 		/*
2473 		 * this attempts to bias the interrupt rate towards Bulk
2474 		 * by adding intermediate steps when interrupt rate is
2475 		 * increasing
2476 		 */
2477 		new_itr = new_itr > adapter->itr ?
2478 			     min(adapter->itr + (new_itr >> 2), new_itr) :
2479 			     new_itr;
2480 		adapter->itr = new_itr;
2481 		adapter->rx_ring->itr_val = new_itr;
2482 		if (adapter->msix_entries)
2483 			adapter->rx_ring->set_itr = 1;
2484 		else
2485 			if (new_itr)
2486 				ew32(ITR, 1000000000 / (new_itr * 256));
2487 			else
2488 				ew32(ITR, 0);
2489 	}
2490 }
2491 
2492 /**
2493  * e1000_alloc_queues - Allocate memory for all rings
2494  * @adapter: board private structure to initialize
2495  **/
2496 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2497 {
2498 	int size = sizeof(struct e1000_ring);
2499 
2500 	adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2501 	if (!adapter->tx_ring)
2502 		goto err;
2503 	adapter->tx_ring->count = adapter->tx_ring_count;
2504 	adapter->tx_ring->adapter = adapter;
2505 
2506 	adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2507 	if (!adapter->rx_ring)
2508 		goto err;
2509 	adapter->rx_ring->count = adapter->rx_ring_count;
2510 	adapter->rx_ring->adapter = adapter;
2511 
2512 	return 0;
2513 err:
2514 	e_err("Unable to allocate memory for queues\n");
2515 	kfree(adapter->rx_ring);
2516 	kfree(adapter->tx_ring);
2517 	return -ENOMEM;
2518 }
2519 
2520 /**
2521  * e1000_clean - NAPI Rx polling callback
2522  * @napi: struct associated with this polling callback
2523  * @budget: amount of packets driver is allowed to process this poll
2524  **/
2525 static int e1000_clean(struct napi_struct *napi, int budget)
2526 {
2527 	struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2528 	struct e1000_hw *hw = &adapter->hw;
2529 	struct net_device *poll_dev = adapter->netdev;
2530 	int tx_cleaned = 1, work_done = 0;
2531 
2532 	adapter = netdev_priv(poll_dev);
2533 
2534 	if (adapter->msix_entries &&
2535 	    !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2536 		goto clean_rx;
2537 
2538 	tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2539 
2540 clean_rx:
2541 	adapter->clean_rx(adapter->rx_ring, &work_done, budget);
2542 
2543 	if (!tx_cleaned)
2544 		work_done = budget;
2545 
2546 	/* If budget not fully consumed, exit the polling mode */
2547 	if (work_done < budget) {
2548 		if (adapter->itr_setting & 3)
2549 			e1000_set_itr(adapter);
2550 		napi_complete(napi);
2551 		if (!test_bit(__E1000_DOWN, &adapter->state)) {
2552 			if (adapter->msix_entries)
2553 				ew32(IMS, adapter->rx_ring->ims_val);
2554 			else
2555 				e1000_irq_enable(adapter);
2556 		}
2557 	}
2558 
2559 	return work_done;
2560 }
2561 
2562 static int e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2563 {
2564 	struct e1000_adapter *adapter = netdev_priv(netdev);
2565 	struct e1000_hw *hw = &adapter->hw;
2566 	u32 vfta, index;
2567 
2568 	/* don't update vlan cookie if already programmed */
2569 	if ((adapter->hw.mng_cookie.status &
2570 	     E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2571 	    (vid == adapter->mng_vlan_id))
2572 		return 0;
2573 
2574 	/* add VID to filter table */
2575 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2576 		index = (vid >> 5) & 0x7F;
2577 		vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2578 		vfta |= (1 << (vid & 0x1F));
2579 		hw->mac.ops.write_vfta(hw, index, vfta);
2580 	}
2581 
2582 	set_bit(vid, adapter->active_vlans);
2583 
2584 	return 0;
2585 }
2586 
2587 static int e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2588 {
2589 	struct e1000_adapter *adapter = netdev_priv(netdev);
2590 	struct e1000_hw *hw = &adapter->hw;
2591 	u32 vfta, index;
2592 
2593 	if ((adapter->hw.mng_cookie.status &
2594 	     E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2595 	    (vid == adapter->mng_vlan_id)) {
2596 		/* release control to f/w */
2597 		e1000e_release_hw_control(adapter);
2598 		return 0;
2599 	}
2600 
2601 	/* remove VID from filter table */
2602 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2603 		index = (vid >> 5) & 0x7F;
2604 		vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2605 		vfta &= ~(1 << (vid & 0x1F));
2606 		hw->mac.ops.write_vfta(hw, index, vfta);
2607 	}
2608 
2609 	clear_bit(vid, adapter->active_vlans);
2610 
2611 	return 0;
2612 }
2613 
2614 /**
2615  * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2616  * @adapter: board private structure to initialize
2617  **/
2618 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2619 {
2620 	struct net_device *netdev = adapter->netdev;
2621 	struct e1000_hw *hw = &adapter->hw;
2622 	u32 rctl;
2623 
2624 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2625 		/* disable VLAN receive filtering */
2626 		rctl = er32(RCTL);
2627 		rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2628 		ew32(RCTL, rctl);
2629 
2630 		if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2631 			e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2632 			adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2633 		}
2634 	}
2635 }
2636 
2637 /**
2638  * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2639  * @adapter: board private structure to initialize
2640  **/
2641 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2642 {
2643 	struct e1000_hw *hw = &adapter->hw;
2644 	u32 rctl;
2645 
2646 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2647 		/* enable VLAN receive filtering */
2648 		rctl = er32(RCTL);
2649 		rctl |= E1000_RCTL_VFE;
2650 		rctl &= ~E1000_RCTL_CFIEN;
2651 		ew32(RCTL, rctl);
2652 	}
2653 }
2654 
2655 /**
2656  * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2657  * @adapter: board private structure to initialize
2658  **/
2659 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2660 {
2661 	struct e1000_hw *hw = &adapter->hw;
2662 	u32 ctrl;
2663 
2664 	/* disable VLAN tag insert/strip */
2665 	ctrl = er32(CTRL);
2666 	ctrl &= ~E1000_CTRL_VME;
2667 	ew32(CTRL, ctrl);
2668 }
2669 
2670 /**
2671  * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2672  * @adapter: board private structure to initialize
2673  **/
2674 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2675 {
2676 	struct e1000_hw *hw = &adapter->hw;
2677 	u32 ctrl;
2678 
2679 	/* enable VLAN tag insert/strip */
2680 	ctrl = er32(CTRL);
2681 	ctrl |= E1000_CTRL_VME;
2682 	ew32(CTRL, ctrl);
2683 }
2684 
2685 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2686 {
2687 	struct net_device *netdev = adapter->netdev;
2688 	u16 vid = adapter->hw.mng_cookie.vlan_id;
2689 	u16 old_vid = adapter->mng_vlan_id;
2690 
2691 	if (adapter->hw.mng_cookie.status &
2692 	    E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2693 		e1000_vlan_rx_add_vid(netdev, vid);
2694 		adapter->mng_vlan_id = vid;
2695 	}
2696 
2697 	if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2698 		e1000_vlan_rx_kill_vid(netdev, old_vid);
2699 }
2700 
2701 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2702 {
2703 	u16 vid;
2704 
2705 	e1000_vlan_rx_add_vid(adapter->netdev, 0);
2706 
2707 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2708 		e1000_vlan_rx_add_vid(adapter->netdev, vid);
2709 }
2710 
2711 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2712 {
2713 	struct e1000_hw *hw = &adapter->hw;
2714 	u32 manc, manc2h, mdef, i, j;
2715 
2716 	if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2717 		return;
2718 
2719 	manc = er32(MANC);
2720 
2721 	/*
2722 	 * enable receiving management packets to the host. this will probably
2723 	 * generate destination unreachable messages from the host OS, but
2724 	 * the packets will be handled on SMBUS
2725 	 */
2726 	manc |= E1000_MANC_EN_MNG2HOST;
2727 	manc2h = er32(MANC2H);
2728 
2729 	switch (hw->mac.type) {
2730 	default:
2731 		manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2732 		break;
2733 	case e1000_82574:
2734 	case e1000_82583:
2735 		/*
2736 		 * Check if IPMI pass-through decision filter already exists;
2737 		 * if so, enable it.
2738 		 */
2739 		for (i = 0, j = 0; i < 8; i++) {
2740 			mdef = er32(MDEF(i));
2741 
2742 			/* Ignore filters with anything other than IPMI ports */
2743 			if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2744 				continue;
2745 
2746 			/* Enable this decision filter in MANC2H */
2747 			if (mdef)
2748 				manc2h |= (1 << i);
2749 
2750 			j |= mdef;
2751 		}
2752 
2753 		if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2754 			break;
2755 
2756 		/* Create new decision filter in an empty filter */
2757 		for (i = 0, j = 0; i < 8; i++)
2758 			if (er32(MDEF(i)) == 0) {
2759 				ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2760 					       E1000_MDEF_PORT_664));
2761 				manc2h |= (1 << 1);
2762 				j++;
2763 				break;
2764 			}
2765 
2766 		if (!j)
2767 			e_warn("Unable to create IPMI pass-through filter\n");
2768 		break;
2769 	}
2770 
2771 	ew32(MANC2H, manc2h);
2772 	ew32(MANC, manc);
2773 }
2774 
2775 /**
2776  * e1000_configure_tx - Configure Transmit Unit after Reset
2777  * @adapter: board private structure
2778  *
2779  * Configure the Tx unit of the MAC after a reset.
2780  **/
2781 static void e1000_configure_tx(struct e1000_adapter *adapter)
2782 {
2783 	struct e1000_hw *hw = &adapter->hw;
2784 	struct e1000_ring *tx_ring = adapter->tx_ring;
2785 	u64 tdba;
2786 	u32 tdlen, tarc;
2787 
2788 	/* Setup the HW Tx Head and Tail descriptor pointers */
2789 	tdba = tx_ring->dma;
2790 	tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2791 	ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2792 	ew32(TDBAH, (tdba >> 32));
2793 	ew32(TDLEN, tdlen);
2794 	ew32(TDH, 0);
2795 	ew32(TDT, 0);
2796 	tx_ring->head = adapter->hw.hw_addr + E1000_TDH;
2797 	tx_ring->tail = adapter->hw.hw_addr + E1000_TDT;
2798 
2799 	/* Set the Tx Interrupt Delay register */
2800 	ew32(TIDV, adapter->tx_int_delay);
2801 	/* Tx irq moderation */
2802 	ew32(TADV, adapter->tx_abs_int_delay);
2803 
2804 	if (adapter->flags2 & FLAG2_DMA_BURST) {
2805 		u32 txdctl = er32(TXDCTL(0));
2806 		txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2807 			    E1000_TXDCTL_WTHRESH);
2808 		/*
2809 		 * set up some performance related parameters to encourage the
2810 		 * hardware to use the bus more efficiently in bursts, depends
2811 		 * on the tx_int_delay to be enabled,
2812 		 * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2813 		 * hthresh = 1 ==> prefetch when one or more available
2814 		 * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2815 		 * BEWARE: this seems to work but should be considered first if
2816 		 * there are Tx hangs or other Tx related bugs
2817 		 */
2818 		txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2819 		ew32(TXDCTL(0), txdctl);
2820 	}
2821 	/* erratum work around: set txdctl the same for both queues */
2822 	ew32(TXDCTL(1), er32(TXDCTL(0)));
2823 
2824 	if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2825 		tarc = er32(TARC(0));
2826 		/*
2827 		 * set the speed mode bit, we'll clear it if we're not at
2828 		 * gigabit link later
2829 		 */
2830 #define SPEED_MODE_BIT (1 << 21)
2831 		tarc |= SPEED_MODE_BIT;
2832 		ew32(TARC(0), tarc);
2833 	}
2834 
2835 	/* errata: program both queues to unweighted RR */
2836 	if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2837 		tarc = er32(TARC(0));
2838 		tarc |= 1;
2839 		ew32(TARC(0), tarc);
2840 		tarc = er32(TARC(1));
2841 		tarc |= 1;
2842 		ew32(TARC(1), tarc);
2843 	}
2844 
2845 	/* Setup Transmit Descriptor Settings for eop descriptor */
2846 	adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2847 
2848 	/* only set IDE if we are delaying interrupts using the timers */
2849 	if (adapter->tx_int_delay)
2850 		adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2851 
2852 	/* enable Report Status bit */
2853 	adapter->txd_cmd |= E1000_TXD_CMD_RS;
2854 
2855 	hw->mac.ops.config_collision_dist(hw);
2856 }
2857 
2858 /**
2859  * e1000_setup_rctl - configure the receive control registers
2860  * @adapter: Board private structure
2861  **/
2862 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2863 			   (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2864 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2865 {
2866 	struct e1000_hw *hw = &adapter->hw;
2867 	u32 rctl, rfctl;
2868 	u32 pages = 0;
2869 
2870 	/* Workaround Si errata on 82579 - configure jumbo frame flow */
2871 	if (hw->mac.type == e1000_pch2lan) {
2872 		s32 ret_val;
2873 
2874 		if (adapter->netdev->mtu > ETH_DATA_LEN)
2875 			ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2876 		else
2877 			ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2878 
2879 		if (ret_val)
2880 			e_dbg("failed to enable jumbo frame workaround mode\n");
2881 	}
2882 
2883 	/* Program MC offset vector base */
2884 	rctl = er32(RCTL);
2885 	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2886 	rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2887 		E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2888 		(adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2889 
2890 	/* Do not Store bad packets */
2891 	rctl &= ~E1000_RCTL_SBP;
2892 
2893 	/* Enable Long Packet receive */
2894 	if (adapter->netdev->mtu <= ETH_DATA_LEN)
2895 		rctl &= ~E1000_RCTL_LPE;
2896 	else
2897 		rctl |= E1000_RCTL_LPE;
2898 
2899 	/* Some systems expect that the CRC is included in SMBUS traffic. The
2900 	 * hardware strips the CRC before sending to both SMBUS (BMC) and to
2901 	 * host memory when this is enabled
2902 	 */
2903 	if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2904 		rctl |= E1000_RCTL_SECRC;
2905 
2906 	/* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2907 	if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2908 		u16 phy_data;
2909 
2910 		e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2911 		phy_data &= 0xfff8;
2912 		phy_data |= (1 << 2);
2913 		e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2914 
2915 		e1e_rphy(hw, 22, &phy_data);
2916 		phy_data &= 0x0fff;
2917 		phy_data |= (1 << 14);
2918 		e1e_wphy(hw, 0x10, 0x2823);
2919 		e1e_wphy(hw, 0x11, 0x0003);
2920 		e1e_wphy(hw, 22, phy_data);
2921 	}
2922 
2923 	/* Setup buffer sizes */
2924 	rctl &= ~E1000_RCTL_SZ_4096;
2925 	rctl |= E1000_RCTL_BSEX;
2926 	switch (adapter->rx_buffer_len) {
2927 	case 2048:
2928 	default:
2929 		rctl |= E1000_RCTL_SZ_2048;
2930 		rctl &= ~E1000_RCTL_BSEX;
2931 		break;
2932 	case 4096:
2933 		rctl |= E1000_RCTL_SZ_4096;
2934 		break;
2935 	case 8192:
2936 		rctl |= E1000_RCTL_SZ_8192;
2937 		break;
2938 	case 16384:
2939 		rctl |= E1000_RCTL_SZ_16384;
2940 		break;
2941 	}
2942 
2943 	/* Enable Extended Status in all Receive Descriptors */
2944 	rfctl = er32(RFCTL);
2945 	rfctl |= E1000_RFCTL_EXTEN;
2946 
2947 	/*
2948 	 * 82571 and greater support packet-split where the protocol
2949 	 * header is placed in skb->data and the packet data is
2950 	 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2951 	 * In the case of a non-split, skb->data is linearly filled,
2952 	 * followed by the page buffers.  Therefore, skb->data is
2953 	 * sized to hold the largest protocol header.
2954 	 *
2955 	 * allocations using alloc_page take too long for regular MTU
2956 	 * so only enable packet split for jumbo frames
2957 	 *
2958 	 * Using pages when the page size is greater than 16k wastes
2959 	 * a lot of memory, since we allocate 3 pages at all times
2960 	 * per packet.
2961 	 */
2962 	pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2963 	if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2964 		adapter->rx_ps_pages = pages;
2965 	else
2966 		adapter->rx_ps_pages = 0;
2967 
2968 	if (adapter->rx_ps_pages) {
2969 		u32 psrctl = 0;
2970 
2971 		/*
2972 		 * disable packet split support for IPv6 extension headers,
2973 		 * because some malformed IPv6 headers can hang the Rx
2974 		 */
2975 		rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2976 			  E1000_RFCTL_NEW_IPV6_EXT_DIS);
2977 
2978 		/* Enable Packet split descriptors */
2979 		rctl |= E1000_RCTL_DTYP_PS;
2980 
2981 		psrctl |= adapter->rx_ps_bsize0 >>
2982 			E1000_PSRCTL_BSIZE0_SHIFT;
2983 
2984 		switch (adapter->rx_ps_pages) {
2985 		case 3:
2986 			psrctl |= PAGE_SIZE <<
2987 				E1000_PSRCTL_BSIZE3_SHIFT;
2988 		case 2:
2989 			psrctl |= PAGE_SIZE <<
2990 				E1000_PSRCTL_BSIZE2_SHIFT;
2991 		case 1:
2992 			psrctl |= PAGE_SIZE >>
2993 				E1000_PSRCTL_BSIZE1_SHIFT;
2994 			break;
2995 		}
2996 
2997 		ew32(PSRCTL, psrctl);
2998 	}
2999 
3000 	/* This is useful for sniffing bad packets. */
3001 	if (adapter->netdev->features & NETIF_F_RXALL) {
3002 		/* UPE and MPE will be handled by normal PROMISC logic
3003 		 * in e1000e_set_rx_mode */
3004 		rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3005 			 E1000_RCTL_BAM | /* RX All Bcast Pkts */
3006 			 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3007 
3008 		rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3009 			  E1000_RCTL_DPF | /* Allow filtered pause */
3010 			  E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3011 		/* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3012 		 * and that breaks VLANs.
3013 		 */
3014 	}
3015 
3016 	ew32(RFCTL, rfctl);
3017 	ew32(RCTL, rctl);
3018 	/* just started the receive unit, no need to restart */
3019 	adapter->flags &= ~FLAG_RX_RESTART_NOW;
3020 }
3021 
3022 /**
3023  * e1000_configure_rx - Configure Receive Unit after Reset
3024  * @adapter: board private structure
3025  *
3026  * Configure the Rx unit of the MAC after a reset.
3027  **/
3028 static void e1000_configure_rx(struct e1000_adapter *adapter)
3029 {
3030 	struct e1000_hw *hw = &adapter->hw;
3031 	struct e1000_ring *rx_ring = adapter->rx_ring;
3032 	u64 rdba;
3033 	u32 rdlen, rctl, rxcsum, ctrl_ext;
3034 
3035 	if (adapter->rx_ps_pages) {
3036 		/* this is a 32 byte descriptor */
3037 		rdlen = rx_ring->count *
3038 		    sizeof(union e1000_rx_desc_packet_split);
3039 		adapter->clean_rx = e1000_clean_rx_irq_ps;
3040 		adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3041 	} else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3042 		rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3043 		adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3044 		adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3045 	} else {
3046 		rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3047 		adapter->clean_rx = e1000_clean_rx_irq;
3048 		adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3049 	}
3050 
3051 	/* disable receives while setting up the descriptors */
3052 	rctl = er32(RCTL);
3053 	if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3054 		ew32(RCTL, rctl & ~E1000_RCTL_EN);
3055 	e1e_flush();
3056 	usleep_range(10000, 20000);
3057 
3058 	if (adapter->flags2 & FLAG2_DMA_BURST) {
3059 		/*
3060 		 * set the writeback threshold (only takes effect if the RDTR
3061 		 * is set). set GRAN=1 and write back up to 0x4 worth, and
3062 		 * enable prefetching of 0x20 Rx descriptors
3063 		 * granularity = 01
3064 		 * wthresh = 04,
3065 		 * hthresh = 04,
3066 		 * pthresh = 0x20
3067 		 */
3068 		ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3069 		ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3070 
3071 		/*
3072 		 * override the delay timers for enabling bursting, only if
3073 		 * the value was not set by the user via module options
3074 		 */
3075 		if (adapter->rx_int_delay == DEFAULT_RDTR)
3076 			adapter->rx_int_delay = BURST_RDTR;
3077 		if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3078 			adapter->rx_abs_int_delay = BURST_RADV;
3079 	}
3080 
3081 	/* set the Receive Delay Timer Register */
3082 	ew32(RDTR, adapter->rx_int_delay);
3083 
3084 	/* irq moderation */
3085 	ew32(RADV, adapter->rx_abs_int_delay);
3086 	if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3087 		ew32(ITR, 1000000000 / (adapter->itr * 256));
3088 
3089 	ctrl_ext = er32(CTRL_EXT);
3090 	/* Auto-Mask interrupts upon ICR access */
3091 	ctrl_ext |= E1000_CTRL_EXT_IAME;
3092 	ew32(IAM, 0xffffffff);
3093 	ew32(CTRL_EXT, ctrl_ext);
3094 	e1e_flush();
3095 
3096 	/*
3097 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
3098 	 * the Base and Length of the Rx Descriptor Ring
3099 	 */
3100 	rdba = rx_ring->dma;
3101 	ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
3102 	ew32(RDBAH, (rdba >> 32));
3103 	ew32(RDLEN, rdlen);
3104 	ew32(RDH, 0);
3105 	ew32(RDT, 0);
3106 	rx_ring->head = adapter->hw.hw_addr + E1000_RDH;
3107 	rx_ring->tail = adapter->hw.hw_addr + E1000_RDT;
3108 
3109 	/* Enable Receive Checksum Offload for TCP and UDP */
3110 	rxcsum = er32(RXCSUM);
3111 	if (adapter->netdev->features & NETIF_F_RXCSUM) {
3112 		rxcsum |= E1000_RXCSUM_TUOFL;
3113 
3114 		/*
3115 		 * IPv4 payload checksum for UDP fragments must be
3116 		 * used in conjunction with packet-split.
3117 		 */
3118 		if (adapter->rx_ps_pages)
3119 			rxcsum |= E1000_RXCSUM_IPPCSE;
3120 	} else {
3121 		rxcsum &= ~E1000_RXCSUM_TUOFL;
3122 		/* no need to clear IPPCSE as it defaults to 0 */
3123 	}
3124 	ew32(RXCSUM, rxcsum);
3125 
3126 	if (adapter->hw.mac.type == e1000_pch2lan) {
3127 		/*
3128 		 * With jumbo frames, excessive C-state transition
3129 		 * latencies result in dropped transactions.
3130 		 */
3131 		if (adapter->netdev->mtu > ETH_DATA_LEN) {
3132 			u32 rxdctl = er32(RXDCTL(0));
3133 			ew32(RXDCTL(0), rxdctl | 0x3);
3134 			pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
3135 		} else {
3136 			pm_qos_update_request(&adapter->netdev->pm_qos_req,
3137 					      PM_QOS_DEFAULT_VALUE);
3138 		}
3139 	}
3140 
3141 	/* Enable Receives */
3142 	ew32(RCTL, rctl);
3143 }
3144 
3145 /**
3146  * e1000e_write_mc_addr_list - write multicast addresses to MTA
3147  * @netdev: network interface device structure
3148  *
3149  * Writes multicast address list to the MTA hash table.
3150  * Returns: -ENOMEM on failure
3151  *                0 on no addresses written
3152  *                X on writing X addresses to MTA
3153  */
3154 static int e1000e_write_mc_addr_list(struct net_device *netdev)
3155 {
3156 	struct e1000_adapter *adapter = netdev_priv(netdev);
3157 	struct e1000_hw *hw = &adapter->hw;
3158 	struct netdev_hw_addr *ha;
3159 	u8 *mta_list;
3160 	int i;
3161 
3162 	if (netdev_mc_empty(netdev)) {
3163 		/* nothing to program, so clear mc list */
3164 		hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3165 		return 0;
3166 	}
3167 
3168 	mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3169 	if (!mta_list)
3170 		return -ENOMEM;
3171 
3172 	/* update_mc_addr_list expects a packed array of only addresses. */
3173 	i = 0;
3174 	netdev_for_each_mc_addr(ha, netdev)
3175 		memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3176 
3177 	hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3178 	kfree(mta_list);
3179 
3180 	return netdev_mc_count(netdev);
3181 }
3182 
3183 /**
3184  * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3185  * @netdev: network interface device structure
3186  *
3187  * Writes unicast address list to the RAR table.
3188  * Returns: -ENOMEM on failure/insufficient address space
3189  *                0 on no addresses written
3190  *                X on writing X addresses to the RAR table
3191  **/
3192 static int e1000e_write_uc_addr_list(struct net_device *netdev)
3193 {
3194 	struct e1000_adapter *adapter = netdev_priv(netdev);
3195 	struct e1000_hw *hw = &adapter->hw;
3196 	unsigned int rar_entries = hw->mac.rar_entry_count;
3197 	int count = 0;
3198 
3199 	/* save a rar entry for our hardware address */
3200 	rar_entries--;
3201 
3202 	/* save a rar entry for the LAA workaround */
3203 	if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3204 		rar_entries--;
3205 
3206 	/* return ENOMEM indicating insufficient memory for addresses */
3207 	if (netdev_uc_count(netdev) > rar_entries)
3208 		return -ENOMEM;
3209 
3210 	if (!netdev_uc_empty(netdev) && rar_entries) {
3211 		struct netdev_hw_addr *ha;
3212 
3213 		/*
3214 		 * write the addresses in reverse order to avoid write
3215 		 * combining
3216 		 */
3217 		netdev_for_each_uc_addr(ha, netdev) {
3218 			if (!rar_entries)
3219 				break;
3220 			e1000e_rar_set(hw, ha->addr, rar_entries--);
3221 			count++;
3222 		}
3223 	}
3224 
3225 	/* zero out the remaining RAR entries not used above */
3226 	for (; rar_entries > 0; rar_entries--) {
3227 		ew32(RAH(rar_entries), 0);
3228 		ew32(RAL(rar_entries), 0);
3229 	}
3230 	e1e_flush();
3231 
3232 	return count;
3233 }
3234 
3235 /**
3236  * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3237  * @netdev: network interface device structure
3238  *
3239  * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3240  * address list or the network interface flags are updated.  This routine is
3241  * responsible for configuring the hardware for proper unicast, multicast,
3242  * promiscuous mode, and all-multi behavior.
3243  **/
3244 static void e1000e_set_rx_mode(struct net_device *netdev)
3245 {
3246 	struct e1000_adapter *adapter = netdev_priv(netdev);
3247 	struct e1000_hw *hw = &adapter->hw;
3248 	u32 rctl;
3249 
3250 	/* Check for Promiscuous and All Multicast modes */
3251 	rctl = er32(RCTL);
3252 
3253 	/* clear the affected bits */
3254 	rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3255 
3256 	if (netdev->flags & IFF_PROMISC) {
3257 		rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3258 		/* Do not hardware filter VLANs in promisc mode */
3259 		e1000e_vlan_filter_disable(adapter);
3260 	} else {
3261 		int count;
3262 
3263 		if (netdev->flags & IFF_ALLMULTI) {
3264 			rctl |= E1000_RCTL_MPE;
3265 		} else {
3266 			/*
3267 			 * Write addresses to the MTA, if the attempt fails
3268 			 * then we should just turn on promiscuous mode so
3269 			 * that we can at least receive multicast traffic
3270 			 */
3271 			count = e1000e_write_mc_addr_list(netdev);
3272 			if (count < 0)
3273 				rctl |= E1000_RCTL_MPE;
3274 		}
3275 		e1000e_vlan_filter_enable(adapter);
3276 		/*
3277 		 * Write addresses to available RAR registers, if there is not
3278 		 * sufficient space to store all the addresses then enable
3279 		 * unicast promiscuous mode
3280 		 */
3281 		count = e1000e_write_uc_addr_list(netdev);
3282 		if (count < 0)
3283 			rctl |= E1000_RCTL_UPE;
3284 	}
3285 
3286 	ew32(RCTL, rctl);
3287 
3288 	if (netdev->features & NETIF_F_HW_VLAN_RX)
3289 		e1000e_vlan_strip_enable(adapter);
3290 	else
3291 		e1000e_vlan_strip_disable(adapter);
3292 }
3293 
3294 static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3295 {
3296 	struct e1000_hw *hw = &adapter->hw;
3297 	u32 mrqc, rxcsum;
3298 	int i;
3299 	static const u32 rsskey[10] = {
3300 		0xda565a6d, 0xc20e5b25, 0x3d256741, 0xb08fa343, 0xcb2bcad0,
3301 		0xb4307bae, 0xa32dcb77, 0x0cf23080, 0x3bb7426a, 0xfa01acbe
3302 	};
3303 
3304 	/* Fill out hash function seed */
3305 	for (i = 0; i < 10; i++)
3306 		ew32(RSSRK(i), rsskey[i]);
3307 
3308 	/* Direct all traffic to queue 0 */
3309 	for (i = 0; i < 32; i++)
3310 		ew32(RETA(i), 0);
3311 
3312 	/*
3313 	 * Disable raw packet checksumming so that RSS hash is placed in
3314 	 * descriptor on writeback.
3315 	 */
3316 	rxcsum = er32(RXCSUM);
3317 	rxcsum |= E1000_RXCSUM_PCSD;
3318 
3319 	ew32(RXCSUM, rxcsum);
3320 
3321 	mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3322 		E1000_MRQC_RSS_FIELD_IPV4_TCP |
3323 		E1000_MRQC_RSS_FIELD_IPV6 |
3324 		E1000_MRQC_RSS_FIELD_IPV6_TCP |
3325 		E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3326 
3327 	ew32(MRQC, mrqc);
3328 }
3329 
3330 /**
3331  * e1000_configure - configure the hardware for Rx and Tx
3332  * @adapter: private board structure
3333  **/
3334 static void e1000_configure(struct e1000_adapter *adapter)
3335 {
3336 	struct e1000_ring *rx_ring = adapter->rx_ring;
3337 
3338 	e1000e_set_rx_mode(adapter->netdev);
3339 
3340 	e1000_restore_vlan(adapter);
3341 	e1000_init_manageability_pt(adapter);
3342 
3343 	e1000_configure_tx(adapter);
3344 
3345 	if (adapter->netdev->features & NETIF_F_RXHASH)
3346 		e1000e_setup_rss_hash(adapter);
3347 	e1000_setup_rctl(adapter);
3348 	e1000_configure_rx(adapter);
3349 	adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3350 }
3351 
3352 /**
3353  * e1000e_power_up_phy - restore link in case the phy was powered down
3354  * @adapter: address of board private structure
3355  *
3356  * The phy may be powered down to save power and turn off link when the
3357  * driver is unloaded and wake on lan is not enabled (among others)
3358  * *** this routine MUST be followed by a call to e1000e_reset ***
3359  **/
3360 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3361 {
3362 	if (adapter->hw.phy.ops.power_up)
3363 		adapter->hw.phy.ops.power_up(&adapter->hw);
3364 
3365 	adapter->hw.mac.ops.setup_link(&adapter->hw);
3366 }
3367 
3368 /**
3369  * e1000_power_down_phy - Power down the PHY
3370  *
3371  * Power down the PHY so no link is implied when interface is down.
3372  * The PHY cannot be powered down if management or WoL is active.
3373  */
3374 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3375 {
3376 	/* WoL is enabled */
3377 	if (adapter->wol)
3378 		return;
3379 
3380 	if (adapter->hw.phy.ops.power_down)
3381 		adapter->hw.phy.ops.power_down(&adapter->hw);
3382 }
3383 
3384 /**
3385  * e1000e_reset - bring the hardware into a known good state
3386  *
3387  * This function boots the hardware and enables some settings that
3388  * require a configuration cycle of the hardware - those cannot be
3389  * set/changed during runtime. After reset the device needs to be
3390  * properly configured for Rx, Tx etc.
3391  */
3392 void e1000e_reset(struct e1000_adapter *adapter)
3393 {
3394 	struct e1000_mac_info *mac = &adapter->hw.mac;
3395 	struct e1000_fc_info *fc = &adapter->hw.fc;
3396 	struct e1000_hw *hw = &adapter->hw;
3397 	u32 tx_space, min_tx_space, min_rx_space;
3398 	u32 pba = adapter->pba;
3399 	u16 hwm;
3400 
3401 	/* reset Packet Buffer Allocation to default */
3402 	ew32(PBA, pba);
3403 
3404 	if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3405 		/*
3406 		 * To maintain wire speed transmits, the Tx FIFO should be
3407 		 * large enough to accommodate two full transmit packets,
3408 		 * rounded up to the next 1KB and expressed in KB.  Likewise,
3409 		 * the Rx FIFO should be large enough to accommodate at least
3410 		 * one full receive packet and is similarly rounded up and
3411 		 * expressed in KB.
3412 		 */
3413 		pba = er32(PBA);
3414 		/* upper 16 bits has Tx packet buffer allocation size in KB */
3415 		tx_space = pba >> 16;
3416 		/* lower 16 bits has Rx packet buffer allocation size in KB */
3417 		pba &= 0xffff;
3418 		/*
3419 		 * the Tx fifo also stores 16 bytes of information about the Tx
3420 		 * but don't include ethernet FCS because hardware appends it
3421 		 */
3422 		min_tx_space = (adapter->max_frame_size +
3423 				sizeof(struct e1000_tx_desc) -
3424 				ETH_FCS_LEN) * 2;
3425 		min_tx_space = ALIGN(min_tx_space, 1024);
3426 		min_tx_space >>= 10;
3427 		/* software strips receive CRC, so leave room for it */
3428 		min_rx_space = adapter->max_frame_size;
3429 		min_rx_space = ALIGN(min_rx_space, 1024);
3430 		min_rx_space >>= 10;
3431 
3432 		/*
3433 		 * If current Tx allocation is less than the min Tx FIFO size,
3434 		 * and the min Tx FIFO size is less than the current Rx FIFO
3435 		 * allocation, take space away from current Rx allocation
3436 		 */
3437 		if ((tx_space < min_tx_space) &&
3438 		    ((min_tx_space - tx_space) < pba)) {
3439 			pba -= min_tx_space - tx_space;
3440 
3441 			/*
3442 			 * if short on Rx space, Rx wins and must trump Tx
3443 			 * adjustment or use Early Receive if available
3444 			 */
3445 			if (pba < min_rx_space)
3446 				pba = min_rx_space;
3447 		}
3448 
3449 		ew32(PBA, pba);
3450 	}
3451 
3452 	/*
3453 	 * flow control settings
3454 	 *
3455 	 * The high water mark must be low enough to fit one full frame
3456 	 * (or the size used for early receive) above it in the Rx FIFO.
3457 	 * Set it to the lower of:
3458 	 * - 90% of the Rx FIFO size, and
3459 	 * - the full Rx FIFO size minus one full frame
3460 	 */
3461 	if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3462 		fc->pause_time = 0xFFFF;
3463 	else
3464 		fc->pause_time = E1000_FC_PAUSE_TIME;
3465 	fc->send_xon = true;
3466 	fc->current_mode = fc->requested_mode;
3467 
3468 	switch (hw->mac.type) {
3469 	case e1000_ich9lan:
3470 	case e1000_ich10lan:
3471 		if (adapter->netdev->mtu > ETH_DATA_LEN) {
3472 			pba = 14;
3473 			ew32(PBA, pba);
3474 			fc->high_water = 0x2800;
3475 			fc->low_water = fc->high_water - 8;
3476 			break;
3477 		}
3478 		/* fall-through */
3479 	default:
3480 		hwm = min(((pba << 10) * 9 / 10),
3481 			  ((pba << 10) - adapter->max_frame_size));
3482 
3483 		fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3484 		fc->low_water = fc->high_water - 8;
3485 		break;
3486 	case e1000_pchlan:
3487 		/*
3488 		 * Workaround PCH LOM adapter hangs with certain network
3489 		 * loads.  If hangs persist, try disabling Tx flow control.
3490 		 */
3491 		if (adapter->netdev->mtu > ETH_DATA_LEN) {
3492 			fc->high_water = 0x3500;
3493 			fc->low_water  = 0x1500;
3494 		} else {
3495 			fc->high_water = 0x5000;
3496 			fc->low_water  = 0x3000;
3497 		}
3498 		fc->refresh_time = 0x1000;
3499 		break;
3500 	case e1000_pch2lan:
3501 		fc->high_water = 0x05C20;
3502 		fc->low_water = 0x05048;
3503 		fc->pause_time = 0x0650;
3504 		fc->refresh_time = 0x0400;
3505 		if (adapter->netdev->mtu > ETH_DATA_LEN) {
3506 			pba = 14;
3507 			ew32(PBA, pba);
3508 		}
3509 		break;
3510 	}
3511 
3512 	/*
3513 	 * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3514 	 * fit in receive buffer.
3515 	 */
3516 	if (adapter->itr_setting & 0x3) {
3517 		if ((adapter->max_frame_size * 2) > (pba << 10)) {
3518 			if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3519 				dev_info(&adapter->pdev->dev,
3520 					"Interrupt Throttle Rate turned off\n");
3521 				adapter->flags2 |= FLAG2_DISABLE_AIM;
3522 				ew32(ITR, 0);
3523 			}
3524 		} else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3525 			dev_info(&adapter->pdev->dev,
3526 				 "Interrupt Throttle Rate turned on\n");
3527 			adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3528 			adapter->itr = 20000;
3529 			ew32(ITR, 1000000000 / (adapter->itr * 256));
3530 		}
3531 	}
3532 
3533 	/* Allow time for pending master requests to run */
3534 	mac->ops.reset_hw(hw);
3535 
3536 	/*
3537 	 * For parts with AMT enabled, let the firmware know
3538 	 * that the network interface is in control
3539 	 */
3540 	if (adapter->flags & FLAG_HAS_AMT)
3541 		e1000e_get_hw_control(adapter);
3542 
3543 	ew32(WUC, 0);
3544 
3545 	if (mac->ops.init_hw(hw))
3546 		e_err("Hardware Error\n");
3547 
3548 	e1000_update_mng_vlan(adapter);
3549 
3550 	/* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3551 	ew32(VET, ETH_P_8021Q);
3552 
3553 	e1000e_reset_adaptive(hw);
3554 
3555 	if (!netif_running(adapter->netdev) &&
3556 	    !test_bit(__E1000_TESTING, &adapter->state)) {
3557 		e1000_power_down_phy(adapter);
3558 		return;
3559 	}
3560 
3561 	e1000_get_phy_info(hw);
3562 
3563 	if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3564 	    !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3565 		u16 phy_data = 0;
3566 		/*
3567 		 * speed up time to link by disabling smart power down, ignore
3568 		 * the return value of this function because there is nothing
3569 		 * different we would do if it failed
3570 		 */
3571 		e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3572 		phy_data &= ~IGP02E1000_PM_SPD;
3573 		e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3574 	}
3575 }
3576 
3577 int e1000e_up(struct e1000_adapter *adapter)
3578 {
3579 	struct e1000_hw *hw = &adapter->hw;
3580 
3581 	/* hardware has been reset, we need to reload some things */
3582 	e1000_configure(adapter);
3583 
3584 	clear_bit(__E1000_DOWN, &adapter->state);
3585 
3586 	if (adapter->msix_entries)
3587 		e1000_configure_msix(adapter);
3588 	e1000_irq_enable(adapter);
3589 
3590 	netif_start_queue(adapter->netdev);
3591 
3592 	/* fire a link change interrupt to start the watchdog */
3593 	if (adapter->msix_entries)
3594 		ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3595 	else
3596 		ew32(ICS, E1000_ICS_LSC);
3597 
3598 	return 0;
3599 }
3600 
3601 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3602 {
3603 	struct e1000_hw *hw = &adapter->hw;
3604 
3605 	if (!(adapter->flags2 & FLAG2_DMA_BURST))
3606 		return;
3607 
3608 	/* flush pending descriptor writebacks to memory */
3609 	ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3610 	ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3611 
3612 	/* execute the writes immediately */
3613 	e1e_flush();
3614 }
3615 
3616 static void e1000e_update_stats(struct e1000_adapter *adapter);
3617 
3618 void e1000e_down(struct e1000_adapter *adapter)
3619 {
3620 	struct net_device *netdev = adapter->netdev;
3621 	struct e1000_hw *hw = &adapter->hw;
3622 	u32 tctl, rctl;
3623 
3624 	/*
3625 	 * signal that we're down so the interrupt handler does not
3626 	 * reschedule our watchdog timer
3627 	 */
3628 	set_bit(__E1000_DOWN, &adapter->state);
3629 
3630 	/* disable receives in the hardware */
3631 	rctl = er32(RCTL);
3632 	if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3633 		ew32(RCTL, rctl & ~E1000_RCTL_EN);
3634 	/* flush and sleep below */
3635 
3636 	netif_stop_queue(netdev);
3637 
3638 	/* disable transmits in the hardware */
3639 	tctl = er32(TCTL);
3640 	tctl &= ~E1000_TCTL_EN;
3641 	ew32(TCTL, tctl);
3642 
3643 	/* flush both disables and wait for them to finish */
3644 	e1e_flush();
3645 	usleep_range(10000, 20000);
3646 
3647 	e1000_irq_disable(adapter);
3648 
3649 	del_timer_sync(&adapter->watchdog_timer);
3650 	del_timer_sync(&adapter->phy_info_timer);
3651 
3652 	netif_carrier_off(netdev);
3653 
3654 	spin_lock(&adapter->stats64_lock);
3655 	e1000e_update_stats(adapter);
3656 	spin_unlock(&adapter->stats64_lock);
3657 
3658 	e1000e_flush_descriptors(adapter);
3659 	e1000_clean_tx_ring(adapter->tx_ring);
3660 	e1000_clean_rx_ring(adapter->rx_ring);
3661 
3662 	adapter->link_speed = 0;
3663 	adapter->link_duplex = 0;
3664 
3665 	if (!pci_channel_offline(adapter->pdev))
3666 		e1000e_reset(adapter);
3667 
3668 	/*
3669 	 * TODO: for power management, we could drop the link and
3670 	 * pci_disable_device here.
3671 	 */
3672 }
3673 
3674 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3675 {
3676 	might_sleep();
3677 	while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3678 		usleep_range(1000, 2000);
3679 	e1000e_down(adapter);
3680 	e1000e_up(adapter);
3681 	clear_bit(__E1000_RESETTING, &adapter->state);
3682 }
3683 
3684 /**
3685  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3686  * @adapter: board private structure to initialize
3687  *
3688  * e1000_sw_init initializes the Adapter private data structure.
3689  * Fields are initialized based on PCI device information and
3690  * OS network device settings (MTU size).
3691  **/
3692 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3693 {
3694 	struct net_device *netdev = adapter->netdev;
3695 
3696 	adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3697 	adapter->rx_ps_bsize0 = 128;
3698 	adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3699 	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3700 	adapter->tx_ring_count = E1000_DEFAULT_TXD;
3701 	adapter->rx_ring_count = E1000_DEFAULT_RXD;
3702 
3703 	spin_lock_init(&adapter->stats64_lock);
3704 
3705 	e1000e_set_interrupt_capability(adapter);
3706 
3707 	if (e1000_alloc_queues(adapter))
3708 		return -ENOMEM;
3709 
3710 	/* Explicitly disable IRQ since the NIC can be in any state. */
3711 	e1000_irq_disable(adapter);
3712 
3713 	set_bit(__E1000_DOWN, &adapter->state);
3714 	return 0;
3715 }
3716 
3717 /**
3718  * e1000_intr_msi_test - Interrupt Handler
3719  * @irq: interrupt number
3720  * @data: pointer to a network interface device structure
3721  **/
3722 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3723 {
3724 	struct net_device *netdev = data;
3725 	struct e1000_adapter *adapter = netdev_priv(netdev);
3726 	struct e1000_hw *hw = &adapter->hw;
3727 	u32 icr = er32(ICR);
3728 
3729 	e_dbg("icr is %08X\n", icr);
3730 	if (icr & E1000_ICR_RXSEQ) {
3731 		adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3732 		wmb();
3733 	}
3734 
3735 	return IRQ_HANDLED;
3736 }
3737 
3738 /**
3739  * e1000_test_msi_interrupt - Returns 0 for successful test
3740  * @adapter: board private struct
3741  *
3742  * code flow taken from tg3.c
3743  **/
3744 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3745 {
3746 	struct net_device *netdev = adapter->netdev;
3747 	struct e1000_hw *hw = &adapter->hw;
3748 	int err;
3749 
3750 	/* poll_enable hasn't been called yet, so don't need disable */
3751 	/* clear any pending events */
3752 	er32(ICR);
3753 
3754 	/* free the real vector and request a test handler */
3755 	e1000_free_irq(adapter);
3756 	e1000e_reset_interrupt_capability(adapter);
3757 
3758 	/* Assume that the test fails, if it succeeds then the test
3759 	 * MSI irq handler will unset this flag */
3760 	adapter->flags |= FLAG_MSI_TEST_FAILED;
3761 
3762 	err = pci_enable_msi(adapter->pdev);
3763 	if (err)
3764 		goto msi_test_failed;
3765 
3766 	err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3767 			  netdev->name, netdev);
3768 	if (err) {
3769 		pci_disable_msi(adapter->pdev);
3770 		goto msi_test_failed;
3771 	}
3772 
3773 	wmb();
3774 
3775 	e1000_irq_enable(adapter);
3776 
3777 	/* fire an unusual interrupt on the test handler */
3778 	ew32(ICS, E1000_ICS_RXSEQ);
3779 	e1e_flush();
3780 	msleep(50);
3781 
3782 	e1000_irq_disable(adapter);
3783 
3784 	rmb();
3785 
3786 	if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3787 		adapter->int_mode = E1000E_INT_MODE_LEGACY;
3788 		e_info("MSI interrupt test failed, using legacy interrupt.\n");
3789 	} else {
3790 		e_dbg("MSI interrupt test succeeded!\n");
3791 	}
3792 
3793 	free_irq(adapter->pdev->irq, netdev);
3794 	pci_disable_msi(adapter->pdev);
3795 
3796 msi_test_failed:
3797 	e1000e_set_interrupt_capability(adapter);
3798 	return e1000_request_irq(adapter);
3799 }
3800 
3801 /**
3802  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3803  * @adapter: board private struct
3804  *
3805  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3806  **/
3807 static int e1000_test_msi(struct e1000_adapter *adapter)
3808 {
3809 	int err;
3810 	u16 pci_cmd;
3811 
3812 	if (!(adapter->flags & FLAG_MSI_ENABLED))
3813 		return 0;
3814 
3815 	/* disable SERR in case the MSI write causes a master abort */
3816 	pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3817 	if (pci_cmd & PCI_COMMAND_SERR)
3818 		pci_write_config_word(adapter->pdev, PCI_COMMAND,
3819 				      pci_cmd & ~PCI_COMMAND_SERR);
3820 
3821 	err = e1000_test_msi_interrupt(adapter);
3822 
3823 	/* re-enable SERR */
3824 	if (pci_cmd & PCI_COMMAND_SERR) {
3825 		pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3826 		pci_cmd |= PCI_COMMAND_SERR;
3827 		pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3828 	}
3829 
3830 	return err;
3831 }
3832 
3833 /**
3834  * e1000_open - Called when a network interface is made active
3835  * @netdev: network interface device structure
3836  *
3837  * Returns 0 on success, negative value on failure
3838  *
3839  * The open entry point is called when a network interface is made
3840  * active by the system (IFF_UP).  At this point all resources needed
3841  * for transmit and receive operations are allocated, the interrupt
3842  * handler is registered with the OS, the watchdog timer is started,
3843  * and the stack is notified that the interface is ready.
3844  **/
3845 static int e1000_open(struct net_device *netdev)
3846 {
3847 	struct e1000_adapter *adapter = netdev_priv(netdev);
3848 	struct e1000_hw *hw = &adapter->hw;
3849 	struct pci_dev *pdev = adapter->pdev;
3850 	int err;
3851 
3852 	/* disallow open during test */
3853 	if (test_bit(__E1000_TESTING, &adapter->state))
3854 		return -EBUSY;
3855 
3856 	pm_runtime_get_sync(&pdev->dev);
3857 
3858 	netif_carrier_off(netdev);
3859 
3860 	/* allocate transmit descriptors */
3861 	err = e1000e_setup_tx_resources(adapter->tx_ring);
3862 	if (err)
3863 		goto err_setup_tx;
3864 
3865 	/* allocate receive descriptors */
3866 	err = e1000e_setup_rx_resources(adapter->rx_ring);
3867 	if (err)
3868 		goto err_setup_rx;
3869 
3870 	/*
3871 	 * If AMT is enabled, let the firmware know that the network
3872 	 * interface is now open and reset the part to a known state.
3873 	 */
3874 	if (adapter->flags & FLAG_HAS_AMT) {
3875 		e1000e_get_hw_control(adapter);
3876 		e1000e_reset(adapter);
3877 	}
3878 
3879 	e1000e_power_up_phy(adapter);
3880 
3881 	adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3882 	if ((adapter->hw.mng_cookie.status &
3883 	     E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3884 		e1000_update_mng_vlan(adapter);
3885 
3886 	/* DMA latency requirement to workaround jumbo issue */
3887 	if (adapter->hw.mac.type == e1000_pch2lan)
3888 		pm_qos_add_request(&adapter->netdev->pm_qos_req,
3889 				   PM_QOS_CPU_DMA_LATENCY,
3890 				   PM_QOS_DEFAULT_VALUE);
3891 
3892 	/*
3893 	 * before we allocate an interrupt, we must be ready to handle it.
3894 	 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3895 	 * as soon as we call pci_request_irq, so we have to setup our
3896 	 * clean_rx handler before we do so.
3897 	 */
3898 	e1000_configure(adapter);
3899 
3900 	err = e1000_request_irq(adapter);
3901 	if (err)
3902 		goto err_req_irq;
3903 
3904 	/*
3905 	 * Work around PCIe errata with MSI interrupts causing some chipsets to
3906 	 * ignore e1000e MSI messages, which means we need to test our MSI
3907 	 * interrupt now
3908 	 */
3909 	if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3910 		err = e1000_test_msi(adapter);
3911 		if (err) {
3912 			e_err("Interrupt allocation failed\n");
3913 			goto err_req_irq;
3914 		}
3915 	}
3916 
3917 	/* From here on the code is the same as e1000e_up() */
3918 	clear_bit(__E1000_DOWN, &adapter->state);
3919 
3920 	napi_enable(&adapter->napi);
3921 
3922 	e1000_irq_enable(adapter);
3923 
3924 	adapter->tx_hang_recheck = false;
3925 	netif_start_queue(netdev);
3926 
3927 	adapter->idle_check = true;
3928 	pm_runtime_put(&pdev->dev);
3929 
3930 	/* fire a link status change interrupt to start the watchdog */
3931 	if (adapter->msix_entries)
3932 		ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3933 	else
3934 		ew32(ICS, E1000_ICS_LSC);
3935 
3936 	return 0;
3937 
3938 err_req_irq:
3939 	e1000e_release_hw_control(adapter);
3940 	e1000_power_down_phy(adapter);
3941 	e1000e_free_rx_resources(adapter->rx_ring);
3942 err_setup_rx:
3943 	e1000e_free_tx_resources(adapter->tx_ring);
3944 err_setup_tx:
3945 	e1000e_reset(adapter);
3946 	pm_runtime_put_sync(&pdev->dev);
3947 
3948 	return err;
3949 }
3950 
3951 /**
3952  * e1000_close - Disables a network interface
3953  * @netdev: network interface device structure
3954  *
3955  * Returns 0, this is not allowed to fail
3956  *
3957  * The close entry point is called when an interface is de-activated
3958  * by the OS.  The hardware is still under the drivers control, but
3959  * needs to be disabled.  A global MAC reset is issued to stop the
3960  * hardware, and all transmit and receive resources are freed.
3961  **/
3962 static int e1000_close(struct net_device *netdev)
3963 {
3964 	struct e1000_adapter *adapter = netdev_priv(netdev);
3965 	struct pci_dev *pdev = adapter->pdev;
3966 
3967 	WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3968 
3969 	pm_runtime_get_sync(&pdev->dev);
3970 
3971 	napi_disable(&adapter->napi);
3972 
3973 	if (!test_bit(__E1000_DOWN, &adapter->state)) {
3974 		e1000e_down(adapter);
3975 		e1000_free_irq(adapter);
3976 	}
3977 	e1000_power_down_phy(adapter);
3978 
3979 	e1000e_free_tx_resources(adapter->tx_ring);
3980 	e1000e_free_rx_resources(adapter->rx_ring);
3981 
3982 	/*
3983 	 * kill manageability vlan ID if supported, but not if a vlan with
3984 	 * the same ID is registered on the host OS (let 8021q kill it)
3985 	 */
3986 	if (adapter->hw.mng_cookie.status &
3987 	    E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
3988 		e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3989 
3990 	/*
3991 	 * If AMT is enabled, let the firmware know that the network
3992 	 * interface is now closed
3993 	 */
3994 	if ((adapter->flags & FLAG_HAS_AMT) &&
3995 	    !test_bit(__E1000_TESTING, &adapter->state))
3996 		e1000e_release_hw_control(adapter);
3997 
3998 	if (adapter->hw.mac.type == e1000_pch2lan)
3999 		pm_qos_remove_request(&adapter->netdev->pm_qos_req);
4000 
4001 	pm_runtime_put_sync(&pdev->dev);
4002 
4003 	return 0;
4004 }
4005 /**
4006  * e1000_set_mac - Change the Ethernet Address of the NIC
4007  * @netdev: network interface device structure
4008  * @p: pointer to an address structure
4009  *
4010  * Returns 0 on success, negative on failure
4011  **/
4012 static int e1000_set_mac(struct net_device *netdev, void *p)
4013 {
4014 	struct e1000_adapter *adapter = netdev_priv(netdev);
4015 	struct sockaddr *addr = p;
4016 
4017 	if (!is_valid_ether_addr(addr->sa_data))
4018 		return -EADDRNOTAVAIL;
4019 
4020 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4021 	memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
4022 
4023 	e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
4024 
4025 	if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
4026 		/* activate the work around */
4027 		e1000e_set_laa_state_82571(&adapter->hw, 1);
4028 
4029 		/*
4030 		 * Hold a copy of the LAA in RAR[14] This is done so that
4031 		 * between the time RAR[0] gets clobbered  and the time it
4032 		 * gets fixed (in e1000_watchdog), the actual LAA is in one
4033 		 * of the RARs and no incoming packets directed to this port
4034 		 * are dropped. Eventually the LAA will be in RAR[0] and
4035 		 * RAR[14]
4036 		 */
4037 		e1000e_rar_set(&adapter->hw,
4038 			      adapter->hw.mac.addr,
4039 			      adapter->hw.mac.rar_entry_count - 1);
4040 	}
4041 
4042 	return 0;
4043 }
4044 
4045 /**
4046  * e1000e_update_phy_task - work thread to update phy
4047  * @work: pointer to our work struct
4048  *
4049  * this worker thread exists because we must acquire a
4050  * semaphore to read the phy, which we could msleep while
4051  * waiting for it, and we can't msleep in a timer.
4052  **/
4053 static void e1000e_update_phy_task(struct work_struct *work)
4054 {
4055 	struct e1000_adapter *adapter = container_of(work,
4056 					struct e1000_adapter, update_phy_task);
4057 
4058 	if (test_bit(__E1000_DOWN, &adapter->state))
4059 		return;
4060 
4061 	e1000_get_phy_info(&adapter->hw);
4062 }
4063 
4064 /*
4065  * Need to wait a few seconds after link up to get diagnostic information from
4066  * the phy
4067  */
4068 static void e1000_update_phy_info(unsigned long data)
4069 {
4070 	struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4071 
4072 	if (test_bit(__E1000_DOWN, &adapter->state))
4073 		return;
4074 
4075 	schedule_work(&adapter->update_phy_task);
4076 }
4077 
4078 /**
4079  * e1000e_update_phy_stats - Update the PHY statistics counters
4080  * @adapter: board private structure
4081  *
4082  * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4083  **/
4084 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4085 {
4086 	struct e1000_hw *hw = &adapter->hw;
4087 	s32 ret_val;
4088 	u16 phy_data;
4089 
4090 	ret_val = hw->phy.ops.acquire(hw);
4091 	if (ret_val)
4092 		return;
4093 
4094 	/*
4095 	 * A page set is expensive so check if already on desired page.
4096 	 * If not, set to the page with the PHY status registers.
4097 	 */
4098 	hw->phy.addr = 1;
4099 	ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4100 					   &phy_data);
4101 	if (ret_val)
4102 		goto release;
4103 	if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4104 		ret_val = hw->phy.ops.set_page(hw,
4105 					       HV_STATS_PAGE << IGP_PAGE_SHIFT);
4106 		if (ret_val)
4107 			goto release;
4108 	}
4109 
4110 	/* Single Collision Count */
4111 	hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4112 	ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4113 	if (!ret_val)
4114 		adapter->stats.scc += phy_data;
4115 
4116 	/* Excessive Collision Count */
4117 	hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4118 	ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4119 	if (!ret_val)
4120 		adapter->stats.ecol += phy_data;
4121 
4122 	/* Multiple Collision Count */
4123 	hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4124 	ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4125 	if (!ret_val)
4126 		adapter->stats.mcc += phy_data;
4127 
4128 	/* Late Collision Count */
4129 	hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4130 	ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4131 	if (!ret_val)
4132 		adapter->stats.latecol += phy_data;
4133 
4134 	/* Collision Count - also used for adaptive IFS */
4135 	hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4136 	ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4137 	if (!ret_val)
4138 		hw->mac.collision_delta = phy_data;
4139 
4140 	/* Defer Count */
4141 	hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4142 	ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4143 	if (!ret_val)
4144 		adapter->stats.dc += phy_data;
4145 
4146 	/* Transmit with no CRS */
4147 	hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4148 	ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4149 	if (!ret_val)
4150 		adapter->stats.tncrs += phy_data;
4151 
4152 release:
4153 	hw->phy.ops.release(hw);
4154 }
4155 
4156 /**
4157  * e1000e_update_stats - Update the board statistics counters
4158  * @adapter: board private structure
4159  **/
4160 static void e1000e_update_stats(struct e1000_adapter *adapter)
4161 {
4162 	struct net_device *netdev = adapter->netdev;
4163 	struct e1000_hw *hw = &adapter->hw;
4164 	struct pci_dev *pdev = adapter->pdev;
4165 
4166 	/*
4167 	 * Prevent stats update while adapter is being reset, or if the pci
4168 	 * connection is down.
4169 	 */
4170 	if (adapter->link_speed == 0)
4171 		return;
4172 	if (pci_channel_offline(pdev))
4173 		return;
4174 
4175 	adapter->stats.crcerrs += er32(CRCERRS);
4176 	adapter->stats.gprc += er32(GPRC);
4177 	adapter->stats.gorc += er32(GORCL);
4178 	er32(GORCH); /* Clear gorc */
4179 	adapter->stats.bprc += er32(BPRC);
4180 	adapter->stats.mprc += er32(MPRC);
4181 	adapter->stats.roc += er32(ROC);
4182 
4183 	adapter->stats.mpc += er32(MPC);
4184 
4185 	/* Half-duplex statistics */
4186 	if (adapter->link_duplex == HALF_DUPLEX) {
4187 		if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4188 			e1000e_update_phy_stats(adapter);
4189 		} else {
4190 			adapter->stats.scc += er32(SCC);
4191 			adapter->stats.ecol += er32(ECOL);
4192 			adapter->stats.mcc += er32(MCC);
4193 			adapter->stats.latecol += er32(LATECOL);
4194 			adapter->stats.dc += er32(DC);
4195 
4196 			hw->mac.collision_delta = er32(COLC);
4197 
4198 			if ((hw->mac.type != e1000_82574) &&
4199 			    (hw->mac.type != e1000_82583))
4200 				adapter->stats.tncrs += er32(TNCRS);
4201 		}
4202 		adapter->stats.colc += hw->mac.collision_delta;
4203 	}
4204 
4205 	adapter->stats.xonrxc += er32(XONRXC);
4206 	adapter->stats.xontxc += er32(XONTXC);
4207 	adapter->stats.xoffrxc += er32(XOFFRXC);
4208 	adapter->stats.xofftxc += er32(XOFFTXC);
4209 	adapter->stats.gptc += er32(GPTC);
4210 	adapter->stats.gotc += er32(GOTCL);
4211 	er32(GOTCH); /* Clear gotc */
4212 	adapter->stats.rnbc += er32(RNBC);
4213 	adapter->stats.ruc += er32(RUC);
4214 
4215 	adapter->stats.mptc += er32(MPTC);
4216 	adapter->stats.bptc += er32(BPTC);
4217 
4218 	/* used for adaptive IFS */
4219 
4220 	hw->mac.tx_packet_delta = er32(TPT);
4221 	adapter->stats.tpt += hw->mac.tx_packet_delta;
4222 
4223 	adapter->stats.algnerrc += er32(ALGNERRC);
4224 	adapter->stats.rxerrc += er32(RXERRC);
4225 	adapter->stats.cexterr += er32(CEXTERR);
4226 	adapter->stats.tsctc += er32(TSCTC);
4227 	adapter->stats.tsctfc += er32(TSCTFC);
4228 
4229 	/* Fill out the OS statistics structure */
4230 	netdev->stats.multicast = adapter->stats.mprc;
4231 	netdev->stats.collisions = adapter->stats.colc;
4232 
4233 	/* Rx Errors */
4234 
4235 	/*
4236 	 * RLEC on some newer hardware can be incorrect so build
4237 	 * our own version based on RUC and ROC
4238 	 */
4239 	netdev->stats.rx_errors = adapter->stats.rxerrc +
4240 		adapter->stats.crcerrs + adapter->stats.algnerrc +
4241 		adapter->stats.ruc + adapter->stats.roc +
4242 		adapter->stats.cexterr;
4243 	netdev->stats.rx_length_errors = adapter->stats.ruc +
4244 					      adapter->stats.roc;
4245 	netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4246 	netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4247 	netdev->stats.rx_missed_errors = adapter->stats.mpc;
4248 
4249 	/* Tx Errors */
4250 	netdev->stats.tx_errors = adapter->stats.ecol +
4251 				       adapter->stats.latecol;
4252 	netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4253 	netdev->stats.tx_window_errors = adapter->stats.latecol;
4254 	netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4255 
4256 	/* Tx Dropped needs to be maintained elsewhere */
4257 
4258 	/* Management Stats */
4259 	adapter->stats.mgptc += er32(MGTPTC);
4260 	adapter->stats.mgprc += er32(MGTPRC);
4261 	adapter->stats.mgpdc += er32(MGTPDC);
4262 }
4263 
4264 /**
4265  * e1000_phy_read_status - Update the PHY register status snapshot
4266  * @adapter: board private structure
4267  **/
4268 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4269 {
4270 	struct e1000_hw *hw = &adapter->hw;
4271 	struct e1000_phy_regs *phy = &adapter->phy_regs;
4272 
4273 	if ((er32(STATUS) & E1000_STATUS_LU) &&
4274 	    (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4275 		int ret_val;
4276 
4277 		ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4278 		ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4279 		ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4280 		ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4281 		ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4282 		ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4283 		ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4284 		ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4285 		if (ret_val)
4286 			e_warn("Error reading PHY register\n");
4287 	} else {
4288 		/*
4289 		 * Do not read PHY registers if link is not up
4290 		 * Set values to typical power-on defaults
4291 		 */
4292 		phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4293 		phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4294 			     BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4295 			     BMSR_ERCAP);
4296 		phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4297 				  ADVERTISE_ALL | ADVERTISE_CSMA);
4298 		phy->lpa = 0;
4299 		phy->expansion = EXPANSION_ENABLENPAGE;
4300 		phy->ctrl1000 = ADVERTISE_1000FULL;
4301 		phy->stat1000 = 0;
4302 		phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4303 	}
4304 }
4305 
4306 static void e1000_print_link_info(struct e1000_adapter *adapter)
4307 {
4308 	struct e1000_hw *hw = &adapter->hw;
4309 	u32 ctrl = er32(CTRL);
4310 
4311 	/* Link status message must follow this format for user tools */
4312 	printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4313 		adapter->netdev->name,
4314 		adapter->link_speed,
4315 		adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
4316 		(ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
4317 		(ctrl & E1000_CTRL_RFCE) ? "Rx" :
4318 		(ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
4319 }
4320 
4321 static bool e1000e_has_link(struct e1000_adapter *adapter)
4322 {
4323 	struct e1000_hw *hw = &adapter->hw;
4324 	bool link_active = false;
4325 	s32 ret_val = 0;
4326 
4327 	/*
4328 	 * get_link_status is set on LSC (link status) interrupt or
4329 	 * Rx sequence error interrupt.  get_link_status will stay
4330 	 * false until the check_for_link establishes link
4331 	 * for copper adapters ONLY
4332 	 */
4333 	switch (hw->phy.media_type) {
4334 	case e1000_media_type_copper:
4335 		if (hw->mac.get_link_status) {
4336 			ret_val = hw->mac.ops.check_for_link(hw);
4337 			link_active = !hw->mac.get_link_status;
4338 		} else {
4339 			link_active = true;
4340 		}
4341 		break;
4342 	case e1000_media_type_fiber:
4343 		ret_val = hw->mac.ops.check_for_link(hw);
4344 		link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4345 		break;
4346 	case e1000_media_type_internal_serdes:
4347 		ret_val = hw->mac.ops.check_for_link(hw);
4348 		link_active = adapter->hw.mac.serdes_has_link;
4349 		break;
4350 	default:
4351 	case e1000_media_type_unknown:
4352 		break;
4353 	}
4354 
4355 	if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4356 	    (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4357 		/* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4358 		e_info("Gigabit has been disabled, downgrading speed\n");
4359 	}
4360 
4361 	return link_active;
4362 }
4363 
4364 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4365 {
4366 	/* make sure the receive unit is started */
4367 	if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4368 	    (adapter->flags & FLAG_RX_RESTART_NOW)) {
4369 		struct e1000_hw *hw = &adapter->hw;
4370 		u32 rctl = er32(RCTL);
4371 		ew32(RCTL, rctl | E1000_RCTL_EN);
4372 		adapter->flags &= ~FLAG_RX_RESTART_NOW;
4373 	}
4374 }
4375 
4376 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4377 {
4378 	struct e1000_hw *hw = &adapter->hw;
4379 
4380 	/*
4381 	 * With 82574 controllers, PHY needs to be checked periodically
4382 	 * for hung state and reset, if two calls return true
4383 	 */
4384 	if (e1000_check_phy_82574(hw))
4385 		adapter->phy_hang_count++;
4386 	else
4387 		adapter->phy_hang_count = 0;
4388 
4389 	if (adapter->phy_hang_count > 1) {
4390 		adapter->phy_hang_count = 0;
4391 		schedule_work(&adapter->reset_task);
4392 	}
4393 }
4394 
4395 /**
4396  * e1000_watchdog - Timer Call-back
4397  * @data: pointer to adapter cast into an unsigned long
4398  **/
4399 static void e1000_watchdog(unsigned long data)
4400 {
4401 	struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4402 
4403 	/* Do the rest outside of interrupt context */
4404 	schedule_work(&adapter->watchdog_task);
4405 
4406 	/* TODO: make this use queue_delayed_work() */
4407 }
4408 
4409 static void e1000_watchdog_task(struct work_struct *work)
4410 {
4411 	struct e1000_adapter *adapter = container_of(work,
4412 					struct e1000_adapter, watchdog_task);
4413 	struct net_device *netdev = adapter->netdev;
4414 	struct e1000_mac_info *mac = &adapter->hw.mac;
4415 	struct e1000_phy_info *phy = &adapter->hw.phy;
4416 	struct e1000_ring *tx_ring = adapter->tx_ring;
4417 	struct e1000_hw *hw = &adapter->hw;
4418 	u32 link, tctl;
4419 
4420 	if (test_bit(__E1000_DOWN, &adapter->state))
4421 		return;
4422 
4423 	link = e1000e_has_link(adapter);
4424 	if ((netif_carrier_ok(netdev)) && link) {
4425 		/* Cancel scheduled suspend requests. */
4426 		pm_runtime_resume(netdev->dev.parent);
4427 
4428 		e1000e_enable_receives(adapter);
4429 		goto link_up;
4430 	}
4431 
4432 	if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4433 	    (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4434 		e1000_update_mng_vlan(adapter);
4435 
4436 	if (link) {
4437 		if (!netif_carrier_ok(netdev)) {
4438 			bool txb2b = true;
4439 
4440 			/* Cancel scheduled suspend requests. */
4441 			pm_runtime_resume(netdev->dev.parent);
4442 
4443 			/* update snapshot of PHY registers on LSC */
4444 			e1000_phy_read_status(adapter);
4445 			mac->ops.get_link_up_info(&adapter->hw,
4446 						   &adapter->link_speed,
4447 						   &adapter->link_duplex);
4448 			e1000_print_link_info(adapter);
4449 			/*
4450 			 * On supported PHYs, check for duplex mismatch only
4451 			 * if link has autonegotiated at 10/100 half
4452 			 */
4453 			if ((hw->phy.type == e1000_phy_igp_3 ||
4454 			     hw->phy.type == e1000_phy_bm) &&
4455 			    (hw->mac.autoneg == true) &&
4456 			    (adapter->link_speed == SPEED_10 ||
4457 			     adapter->link_speed == SPEED_100) &&
4458 			    (adapter->link_duplex == HALF_DUPLEX)) {
4459 				u16 autoneg_exp;
4460 
4461 				e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4462 
4463 				if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4464 					e_info("Autonegotiated half duplex but link partner cannot autoneg.  Try forcing full duplex if link gets many collisions.\n");
4465 			}
4466 
4467 			/* adjust timeout factor according to speed/duplex */
4468 			adapter->tx_timeout_factor = 1;
4469 			switch (adapter->link_speed) {
4470 			case SPEED_10:
4471 				txb2b = false;
4472 				adapter->tx_timeout_factor = 16;
4473 				break;
4474 			case SPEED_100:
4475 				txb2b = false;
4476 				adapter->tx_timeout_factor = 10;
4477 				break;
4478 			}
4479 
4480 			/*
4481 			 * workaround: re-program speed mode bit after
4482 			 * link-up event
4483 			 */
4484 			if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4485 			    !txb2b) {
4486 				u32 tarc0;
4487 				tarc0 = er32(TARC(0));
4488 				tarc0 &= ~SPEED_MODE_BIT;
4489 				ew32(TARC(0), tarc0);
4490 			}
4491 
4492 			/*
4493 			 * disable TSO for pcie and 10/100 speeds, to avoid
4494 			 * some hardware issues
4495 			 */
4496 			if (!(adapter->flags & FLAG_TSO_FORCE)) {
4497 				switch (adapter->link_speed) {
4498 				case SPEED_10:
4499 				case SPEED_100:
4500 					e_info("10/100 speed: disabling TSO\n");
4501 					netdev->features &= ~NETIF_F_TSO;
4502 					netdev->features &= ~NETIF_F_TSO6;
4503 					break;
4504 				case SPEED_1000:
4505 					netdev->features |= NETIF_F_TSO;
4506 					netdev->features |= NETIF_F_TSO6;
4507 					break;
4508 				default:
4509 					/* oops */
4510 					break;
4511 				}
4512 			}
4513 
4514 			/*
4515 			 * enable transmits in the hardware, need to do this
4516 			 * after setting TARC(0)
4517 			 */
4518 			tctl = er32(TCTL);
4519 			tctl |= E1000_TCTL_EN;
4520 			ew32(TCTL, tctl);
4521 
4522                         /*
4523 			 * Perform any post-link-up configuration before
4524 			 * reporting link up.
4525 			 */
4526 			if (phy->ops.cfg_on_link_up)
4527 				phy->ops.cfg_on_link_up(hw);
4528 
4529 			netif_carrier_on(netdev);
4530 
4531 			if (!test_bit(__E1000_DOWN, &adapter->state))
4532 				mod_timer(&adapter->phy_info_timer,
4533 					  round_jiffies(jiffies + 2 * HZ));
4534 		}
4535 	} else {
4536 		if (netif_carrier_ok(netdev)) {
4537 			adapter->link_speed = 0;
4538 			adapter->link_duplex = 0;
4539 			/* Link status message must follow this format */
4540 			printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4541 			       adapter->netdev->name);
4542 			netif_carrier_off(netdev);
4543 			if (!test_bit(__E1000_DOWN, &adapter->state))
4544 				mod_timer(&adapter->phy_info_timer,
4545 					  round_jiffies(jiffies + 2 * HZ));
4546 
4547 			if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4548 				schedule_work(&adapter->reset_task);
4549 			else
4550 				pm_schedule_suspend(netdev->dev.parent,
4551 							LINK_TIMEOUT);
4552 		}
4553 	}
4554 
4555 link_up:
4556 	spin_lock(&adapter->stats64_lock);
4557 	e1000e_update_stats(adapter);
4558 
4559 	mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4560 	adapter->tpt_old = adapter->stats.tpt;
4561 	mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4562 	adapter->colc_old = adapter->stats.colc;
4563 
4564 	adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4565 	adapter->gorc_old = adapter->stats.gorc;
4566 	adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4567 	adapter->gotc_old = adapter->stats.gotc;
4568 	spin_unlock(&adapter->stats64_lock);
4569 
4570 	e1000e_update_adaptive(&adapter->hw);
4571 
4572 	if (!netif_carrier_ok(netdev) &&
4573 	    (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4574 		/*
4575 		 * We've lost link, so the controller stops DMA,
4576 		 * but we've got queued Tx work that's never going
4577 		 * to get done, so reset controller to flush Tx.
4578 		 * (Do the reset outside of interrupt context).
4579 		 */
4580 		schedule_work(&adapter->reset_task);
4581 		/* return immediately since reset is imminent */
4582 		return;
4583 	}
4584 
4585 	/* Simple mode for Interrupt Throttle Rate (ITR) */
4586 	if (adapter->itr_setting == 4) {
4587 		/*
4588 		 * Symmetric Tx/Rx gets a reduced ITR=2000;
4589 		 * Total asymmetrical Tx or Rx gets ITR=8000;
4590 		 * everyone else is between 2000-8000.
4591 		 */
4592 		u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4593 		u32 dif = (adapter->gotc > adapter->gorc ?
4594 			    adapter->gotc - adapter->gorc :
4595 			    adapter->gorc - adapter->gotc) / 10000;
4596 		u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4597 
4598 		ew32(ITR, 1000000000 / (itr * 256));
4599 	}
4600 
4601 	/* Cause software interrupt to ensure Rx ring is cleaned */
4602 	if (adapter->msix_entries)
4603 		ew32(ICS, adapter->rx_ring->ims_val);
4604 	else
4605 		ew32(ICS, E1000_ICS_RXDMT0);
4606 
4607 	/* flush pending descriptors to memory before detecting Tx hang */
4608 	e1000e_flush_descriptors(adapter);
4609 
4610 	/* Force detection of hung controller every watchdog period */
4611 	adapter->detect_tx_hung = true;
4612 
4613 	/*
4614 	 * With 82571 controllers, LAA may be overwritten due to controller
4615 	 * reset from the other port. Set the appropriate LAA in RAR[0]
4616 	 */
4617 	if (e1000e_get_laa_state_82571(hw))
4618 		e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4619 
4620 	if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4621 		e1000e_check_82574_phy_workaround(adapter);
4622 
4623 	/* Reset the timer */
4624 	if (!test_bit(__E1000_DOWN, &adapter->state))
4625 		mod_timer(&adapter->watchdog_timer,
4626 			  round_jiffies(jiffies + 2 * HZ));
4627 }
4628 
4629 #define E1000_TX_FLAGS_CSUM		0x00000001
4630 #define E1000_TX_FLAGS_VLAN		0x00000002
4631 #define E1000_TX_FLAGS_TSO		0x00000004
4632 #define E1000_TX_FLAGS_IPV4		0x00000008
4633 #define E1000_TX_FLAGS_NO_FCS		0x00000010
4634 #define E1000_TX_FLAGS_VLAN_MASK	0xffff0000
4635 #define E1000_TX_FLAGS_VLAN_SHIFT	16
4636 
4637 static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb)
4638 {
4639 	struct e1000_context_desc *context_desc;
4640 	struct e1000_buffer *buffer_info;
4641 	unsigned int i;
4642 	u32 cmd_length = 0;
4643 	u16 ipcse = 0, tucse, mss;
4644 	u8 ipcss, ipcso, tucss, tucso, hdr_len;
4645 
4646 	if (!skb_is_gso(skb))
4647 		return 0;
4648 
4649 	if (skb_header_cloned(skb)) {
4650 		int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4651 
4652 		if (err)
4653 			return err;
4654 	}
4655 
4656 	hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4657 	mss = skb_shinfo(skb)->gso_size;
4658 	if (skb->protocol == htons(ETH_P_IP)) {
4659 		struct iphdr *iph = ip_hdr(skb);
4660 		iph->tot_len = 0;
4661 		iph->check = 0;
4662 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4663 		                                         0, IPPROTO_TCP, 0);
4664 		cmd_length = E1000_TXD_CMD_IP;
4665 		ipcse = skb_transport_offset(skb) - 1;
4666 	} else if (skb_is_gso_v6(skb)) {
4667 		ipv6_hdr(skb)->payload_len = 0;
4668 		tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4669 		                                       &ipv6_hdr(skb)->daddr,
4670 		                                       0, IPPROTO_TCP, 0);
4671 		ipcse = 0;
4672 	}
4673 	ipcss = skb_network_offset(skb);
4674 	ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4675 	tucss = skb_transport_offset(skb);
4676 	tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4677 	tucse = 0;
4678 
4679 	cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4680 	               E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4681 
4682 	i = tx_ring->next_to_use;
4683 	context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4684 	buffer_info = &tx_ring->buffer_info[i];
4685 
4686 	context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4687 	context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4688 	context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4689 	context_desc->upper_setup.tcp_fields.tucss = tucss;
4690 	context_desc->upper_setup.tcp_fields.tucso = tucso;
4691 	context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4692 	context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4693 	context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4694 	context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4695 
4696 	buffer_info->time_stamp = jiffies;
4697 	buffer_info->next_to_watch = i;
4698 
4699 	i++;
4700 	if (i == tx_ring->count)
4701 		i = 0;
4702 	tx_ring->next_to_use = i;
4703 
4704 	return 1;
4705 }
4706 
4707 static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb)
4708 {
4709 	struct e1000_adapter *adapter = tx_ring->adapter;
4710 	struct e1000_context_desc *context_desc;
4711 	struct e1000_buffer *buffer_info;
4712 	unsigned int i;
4713 	u8 css;
4714 	u32 cmd_len = E1000_TXD_CMD_DEXT;
4715 	__be16 protocol;
4716 
4717 	if (skb->ip_summed != CHECKSUM_PARTIAL)
4718 		return 0;
4719 
4720 	if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4721 		protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4722 	else
4723 		protocol = skb->protocol;
4724 
4725 	switch (protocol) {
4726 	case cpu_to_be16(ETH_P_IP):
4727 		if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4728 			cmd_len |= E1000_TXD_CMD_TCP;
4729 		break;
4730 	case cpu_to_be16(ETH_P_IPV6):
4731 		/* XXX not handling all IPV6 headers */
4732 		if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4733 			cmd_len |= E1000_TXD_CMD_TCP;
4734 		break;
4735 	default:
4736 		if (unlikely(net_ratelimit()))
4737 			e_warn("checksum_partial proto=%x!\n",
4738 			       be16_to_cpu(protocol));
4739 		break;
4740 	}
4741 
4742 	css = skb_checksum_start_offset(skb);
4743 
4744 	i = tx_ring->next_to_use;
4745 	buffer_info = &tx_ring->buffer_info[i];
4746 	context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4747 
4748 	context_desc->lower_setup.ip_config = 0;
4749 	context_desc->upper_setup.tcp_fields.tucss = css;
4750 	context_desc->upper_setup.tcp_fields.tucso =
4751 				css + skb->csum_offset;
4752 	context_desc->upper_setup.tcp_fields.tucse = 0;
4753 	context_desc->tcp_seg_setup.data = 0;
4754 	context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4755 
4756 	buffer_info->time_stamp = jiffies;
4757 	buffer_info->next_to_watch = i;
4758 
4759 	i++;
4760 	if (i == tx_ring->count)
4761 		i = 0;
4762 	tx_ring->next_to_use = i;
4763 
4764 	return 1;
4765 }
4766 
4767 #define E1000_MAX_PER_TXD	8192
4768 #define E1000_MAX_TXD_PWR	12
4769 
4770 static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
4771 			unsigned int first, unsigned int max_per_txd,
4772 			unsigned int nr_frags, unsigned int mss)
4773 {
4774 	struct e1000_adapter *adapter = tx_ring->adapter;
4775 	struct pci_dev *pdev = adapter->pdev;
4776 	struct e1000_buffer *buffer_info;
4777 	unsigned int len = skb_headlen(skb);
4778 	unsigned int offset = 0, size, count = 0, i;
4779 	unsigned int f, bytecount, segs;
4780 
4781 	i = tx_ring->next_to_use;
4782 
4783 	while (len) {
4784 		buffer_info = &tx_ring->buffer_info[i];
4785 		size = min(len, max_per_txd);
4786 
4787 		buffer_info->length = size;
4788 		buffer_info->time_stamp = jiffies;
4789 		buffer_info->next_to_watch = i;
4790 		buffer_info->dma = dma_map_single(&pdev->dev,
4791 						  skb->data + offset,
4792 						  size, DMA_TO_DEVICE);
4793 		buffer_info->mapped_as_page = false;
4794 		if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4795 			goto dma_error;
4796 
4797 		len -= size;
4798 		offset += size;
4799 		count++;
4800 
4801 		if (len) {
4802 			i++;
4803 			if (i == tx_ring->count)
4804 				i = 0;
4805 		}
4806 	}
4807 
4808 	for (f = 0; f < nr_frags; f++) {
4809 		const struct skb_frag_struct *frag;
4810 
4811 		frag = &skb_shinfo(skb)->frags[f];
4812 		len = skb_frag_size(frag);
4813 		offset = 0;
4814 
4815 		while (len) {
4816 			i++;
4817 			if (i == tx_ring->count)
4818 				i = 0;
4819 
4820 			buffer_info = &tx_ring->buffer_info[i];
4821 			size = min(len, max_per_txd);
4822 
4823 			buffer_info->length = size;
4824 			buffer_info->time_stamp = jiffies;
4825 			buffer_info->next_to_watch = i;
4826 			buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
4827 						offset, size, DMA_TO_DEVICE);
4828 			buffer_info->mapped_as_page = true;
4829 			if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4830 				goto dma_error;
4831 
4832 			len -= size;
4833 			offset += size;
4834 			count++;
4835 		}
4836 	}
4837 
4838 	segs = skb_shinfo(skb)->gso_segs ? : 1;
4839 	/* multiply data chunks by size of headers */
4840 	bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4841 
4842 	tx_ring->buffer_info[i].skb = skb;
4843 	tx_ring->buffer_info[i].segs = segs;
4844 	tx_ring->buffer_info[i].bytecount = bytecount;
4845 	tx_ring->buffer_info[first].next_to_watch = i;
4846 
4847 	return count;
4848 
4849 dma_error:
4850 	dev_err(&pdev->dev, "Tx DMA map failed\n");
4851 	buffer_info->dma = 0;
4852 	if (count)
4853 		count--;
4854 
4855 	while (count--) {
4856 		if (i == 0)
4857 			i += tx_ring->count;
4858 		i--;
4859 		buffer_info = &tx_ring->buffer_info[i];
4860 		e1000_put_txbuf(tx_ring, buffer_info);
4861 	}
4862 
4863 	return 0;
4864 }
4865 
4866 static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
4867 {
4868 	struct e1000_adapter *adapter = tx_ring->adapter;
4869 	struct e1000_tx_desc *tx_desc = NULL;
4870 	struct e1000_buffer *buffer_info;
4871 	u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4872 	unsigned int i;
4873 
4874 	if (tx_flags & E1000_TX_FLAGS_TSO) {
4875 		txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4876 			     E1000_TXD_CMD_TSE;
4877 		txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4878 
4879 		if (tx_flags & E1000_TX_FLAGS_IPV4)
4880 			txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4881 	}
4882 
4883 	if (tx_flags & E1000_TX_FLAGS_CSUM) {
4884 		txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4885 		txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4886 	}
4887 
4888 	if (tx_flags & E1000_TX_FLAGS_VLAN) {
4889 		txd_lower |= E1000_TXD_CMD_VLE;
4890 		txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4891 	}
4892 
4893 	if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
4894 		txd_lower &= ~(E1000_TXD_CMD_IFCS);
4895 
4896 	i = tx_ring->next_to_use;
4897 
4898 	do {
4899 		buffer_info = &tx_ring->buffer_info[i];
4900 		tx_desc = E1000_TX_DESC(*tx_ring, i);
4901 		tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4902 		tx_desc->lower.data =
4903 			cpu_to_le32(txd_lower | buffer_info->length);
4904 		tx_desc->upper.data = cpu_to_le32(txd_upper);
4905 
4906 		i++;
4907 		if (i == tx_ring->count)
4908 			i = 0;
4909 	} while (--count > 0);
4910 
4911 	tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4912 
4913 	/* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
4914 	if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
4915 		tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
4916 
4917 	/*
4918 	 * Force memory writes to complete before letting h/w
4919 	 * know there are new descriptors to fetch.  (Only
4920 	 * applicable for weak-ordered memory model archs,
4921 	 * such as IA-64).
4922 	 */
4923 	wmb();
4924 
4925 	tx_ring->next_to_use = i;
4926 
4927 	if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
4928 		e1000e_update_tdt_wa(tx_ring, i);
4929 	else
4930 		writel(i, tx_ring->tail);
4931 
4932 	/*
4933 	 * we need this if more than one processor can write to our tail
4934 	 * at a time, it synchronizes IO on IA64/Altix systems
4935 	 */
4936 	mmiowb();
4937 }
4938 
4939 #define MINIMUM_DHCP_PACKET_SIZE 282
4940 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4941 				    struct sk_buff *skb)
4942 {
4943 	struct e1000_hw *hw =  &adapter->hw;
4944 	u16 length, offset;
4945 
4946 	if (vlan_tx_tag_present(skb)) {
4947 		if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4948 		    (adapter->hw.mng_cookie.status &
4949 			E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4950 			return 0;
4951 	}
4952 
4953 	if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4954 		return 0;
4955 
4956 	if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4957 		return 0;
4958 
4959 	{
4960 		const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4961 		struct udphdr *udp;
4962 
4963 		if (ip->protocol != IPPROTO_UDP)
4964 			return 0;
4965 
4966 		udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4967 		if (ntohs(udp->dest) != 67)
4968 			return 0;
4969 
4970 		offset = (u8 *)udp + 8 - skb->data;
4971 		length = skb->len - offset;
4972 		return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4973 	}
4974 
4975 	return 0;
4976 }
4977 
4978 static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
4979 {
4980 	struct e1000_adapter *adapter = tx_ring->adapter;
4981 
4982 	netif_stop_queue(adapter->netdev);
4983 	/*
4984 	 * Herbert's original patch had:
4985 	 *  smp_mb__after_netif_stop_queue();
4986 	 * but since that doesn't exist yet, just open code it.
4987 	 */
4988 	smp_mb();
4989 
4990 	/*
4991 	 * We need to check again in a case another CPU has just
4992 	 * made room available.
4993 	 */
4994 	if (e1000_desc_unused(tx_ring) < size)
4995 		return -EBUSY;
4996 
4997 	/* A reprieve! */
4998 	netif_start_queue(adapter->netdev);
4999 	++adapter->restart_queue;
5000 	return 0;
5001 }
5002 
5003 static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5004 {
5005 	if (e1000_desc_unused(tx_ring) >= size)
5006 		return 0;
5007 	return __e1000_maybe_stop_tx(tx_ring, size);
5008 }
5009 
5010 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1)
5011 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
5012 				    struct net_device *netdev)
5013 {
5014 	struct e1000_adapter *adapter = netdev_priv(netdev);
5015 	struct e1000_ring *tx_ring = adapter->tx_ring;
5016 	unsigned int first;
5017 	unsigned int max_per_txd = E1000_MAX_PER_TXD;
5018 	unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
5019 	unsigned int tx_flags = 0;
5020 	unsigned int len = skb_headlen(skb);
5021 	unsigned int nr_frags;
5022 	unsigned int mss;
5023 	int count = 0;
5024 	int tso;
5025 	unsigned int f;
5026 
5027 	if (test_bit(__E1000_DOWN, &adapter->state)) {
5028 		dev_kfree_skb_any(skb);
5029 		return NETDEV_TX_OK;
5030 	}
5031 
5032 	if (skb->len <= 0) {
5033 		dev_kfree_skb_any(skb);
5034 		return NETDEV_TX_OK;
5035 	}
5036 
5037 	mss = skb_shinfo(skb)->gso_size;
5038 	/*
5039 	 * The controller does a simple calculation to
5040 	 * make sure there is enough room in the FIFO before
5041 	 * initiating the DMA for each buffer.  The calc is:
5042 	 * 4 = ceil(buffer len/mss).  To make sure we don't
5043 	 * overrun the FIFO, adjust the max buffer len if mss
5044 	 * drops.
5045 	 */
5046 	if (mss) {
5047 		u8 hdr_len;
5048 		max_per_txd = min(mss << 2, max_per_txd);
5049 		max_txd_pwr = fls(max_per_txd) - 1;
5050 
5051 		/*
5052 		 * TSO Workaround for 82571/2/3 Controllers -- if skb->data
5053 		 * points to just header, pull a few bytes of payload from
5054 		 * frags into skb->data
5055 		 */
5056 		hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5057 		/*
5058 		 * we do this workaround for ES2LAN, but it is un-necessary,
5059 		 * avoiding it could save a lot of cycles
5060 		 */
5061 		if (skb->data_len && (hdr_len == len)) {
5062 			unsigned int pull_size;
5063 
5064 			pull_size = min_t(unsigned int, 4, skb->data_len);
5065 			if (!__pskb_pull_tail(skb, pull_size)) {
5066 				e_err("__pskb_pull_tail failed.\n");
5067 				dev_kfree_skb_any(skb);
5068 				return NETDEV_TX_OK;
5069 			}
5070 			len = skb_headlen(skb);
5071 		}
5072 	}
5073 
5074 	/* reserve a descriptor for the offload context */
5075 	if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5076 		count++;
5077 	count++;
5078 
5079 	count += TXD_USE_COUNT(len, max_txd_pwr);
5080 
5081 	nr_frags = skb_shinfo(skb)->nr_frags;
5082 	for (f = 0; f < nr_frags; f++)
5083 		count += TXD_USE_COUNT(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5084 				       max_txd_pwr);
5085 
5086 	if (adapter->hw.mac.tx_pkt_filtering)
5087 		e1000_transfer_dhcp_info(adapter, skb);
5088 
5089 	/*
5090 	 * need: count + 2 desc gap to keep tail from touching
5091 	 * head, otherwise try next time
5092 	 */
5093 	if (e1000_maybe_stop_tx(tx_ring, count + 2))
5094 		return NETDEV_TX_BUSY;
5095 
5096 	if (vlan_tx_tag_present(skb)) {
5097 		tx_flags |= E1000_TX_FLAGS_VLAN;
5098 		tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
5099 	}
5100 
5101 	first = tx_ring->next_to_use;
5102 
5103 	tso = e1000_tso(tx_ring, skb);
5104 	if (tso < 0) {
5105 		dev_kfree_skb_any(skb);
5106 		return NETDEV_TX_OK;
5107 	}
5108 
5109 	if (tso)
5110 		tx_flags |= E1000_TX_FLAGS_TSO;
5111 	else if (e1000_tx_csum(tx_ring, skb))
5112 		tx_flags |= E1000_TX_FLAGS_CSUM;
5113 
5114 	/*
5115 	 * Old method was to assume IPv4 packet by default if TSO was enabled.
5116 	 * 82571 hardware supports TSO capabilities for IPv6 as well...
5117 	 * no longer assume, we must.
5118 	 */
5119 	if (skb->protocol == htons(ETH_P_IP))
5120 		tx_flags |= E1000_TX_FLAGS_IPV4;
5121 
5122 	if (unlikely(skb->no_fcs))
5123 		tx_flags |= E1000_TX_FLAGS_NO_FCS;
5124 
5125 	/* if count is 0 then mapping error has occurred */
5126 	count = e1000_tx_map(tx_ring, skb, first, max_per_txd, nr_frags, mss);
5127 	if (count) {
5128 		netdev_sent_queue(netdev, skb->len);
5129 		e1000_tx_queue(tx_ring, tx_flags, count);
5130 		/* Make sure there is space in the ring for the next send. */
5131 		e1000_maybe_stop_tx(tx_ring, MAX_SKB_FRAGS + 2);
5132 
5133 	} else {
5134 		dev_kfree_skb_any(skb);
5135 		tx_ring->buffer_info[first].time_stamp = 0;
5136 		tx_ring->next_to_use = first;
5137 	}
5138 
5139 	return NETDEV_TX_OK;
5140 }
5141 
5142 /**
5143  * e1000_tx_timeout - Respond to a Tx Hang
5144  * @netdev: network interface device structure
5145  **/
5146 static void e1000_tx_timeout(struct net_device *netdev)
5147 {
5148 	struct e1000_adapter *adapter = netdev_priv(netdev);
5149 
5150 	/* Do the reset outside of interrupt context */
5151 	adapter->tx_timeout_count++;
5152 	schedule_work(&adapter->reset_task);
5153 }
5154 
5155 static void e1000_reset_task(struct work_struct *work)
5156 {
5157 	struct e1000_adapter *adapter;
5158 	adapter = container_of(work, struct e1000_adapter, reset_task);
5159 
5160 	/* don't run the task if already down */
5161 	if (test_bit(__E1000_DOWN, &adapter->state))
5162 		return;
5163 
5164 	if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5165 	      (adapter->flags & FLAG_RX_RESTART_NOW))) {
5166 		e1000e_dump(adapter);
5167 		e_err("Reset adapter\n");
5168 	}
5169 	e1000e_reinit_locked(adapter);
5170 }
5171 
5172 /**
5173  * e1000_get_stats64 - Get System Network Statistics
5174  * @netdev: network interface device structure
5175  * @stats: rtnl_link_stats64 pointer
5176  *
5177  * Returns the address of the device statistics structure.
5178  **/
5179 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
5180                                              struct rtnl_link_stats64 *stats)
5181 {
5182 	struct e1000_adapter *adapter = netdev_priv(netdev);
5183 
5184 	memset(stats, 0, sizeof(struct rtnl_link_stats64));
5185 	spin_lock(&adapter->stats64_lock);
5186 	e1000e_update_stats(adapter);
5187 	/* Fill out the OS statistics structure */
5188 	stats->rx_bytes = adapter->stats.gorc;
5189 	stats->rx_packets = adapter->stats.gprc;
5190 	stats->tx_bytes = adapter->stats.gotc;
5191 	stats->tx_packets = adapter->stats.gptc;
5192 	stats->multicast = adapter->stats.mprc;
5193 	stats->collisions = adapter->stats.colc;
5194 
5195 	/* Rx Errors */
5196 
5197 	/*
5198 	 * RLEC on some newer hardware can be incorrect so build
5199 	 * our own version based on RUC and ROC
5200 	 */
5201 	stats->rx_errors = adapter->stats.rxerrc +
5202 		adapter->stats.crcerrs + adapter->stats.algnerrc +
5203 		adapter->stats.ruc + adapter->stats.roc +
5204 		adapter->stats.cexterr;
5205 	stats->rx_length_errors = adapter->stats.ruc +
5206 					      adapter->stats.roc;
5207 	stats->rx_crc_errors = adapter->stats.crcerrs;
5208 	stats->rx_frame_errors = adapter->stats.algnerrc;
5209 	stats->rx_missed_errors = adapter->stats.mpc;
5210 
5211 	/* Tx Errors */
5212 	stats->tx_errors = adapter->stats.ecol +
5213 				       adapter->stats.latecol;
5214 	stats->tx_aborted_errors = adapter->stats.ecol;
5215 	stats->tx_window_errors = adapter->stats.latecol;
5216 	stats->tx_carrier_errors = adapter->stats.tncrs;
5217 
5218 	/* Tx Dropped needs to be maintained elsewhere */
5219 
5220 	spin_unlock(&adapter->stats64_lock);
5221 	return stats;
5222 }
5223 
5224 /**
5225  * e1000_change_mtu - Change the Maximum Transfer Unit
5226  * @netdev: network interface device structure
5227  * @new_mtu: new value for maximum frame size
5228  *
5229  * Returns 0 on success, negative on failure
5230  **/
5231 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5232 {
5233 	struct e1000_adapter *adapter = netdev_priv(netdev);
5234 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5235 
5236 	/* Jumbo frame support */
5237 	if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
5238 		if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5239 			e_err("Jumbo Frames not supported.\n");
5240 			return -EINVAL;
5241 		}
5242 
5243 		/*
5244 		 * IP payload checksum (enabled with jumbos/packet-split when
5245 		 * Rx checksum is enabled) and generation of RSS hash is
5246 		 * mutually exclusive in the hardware.
5247 		 */
5248 		if ((netdev->features & NETIF_F_RXCSUM) &&
5249 		    (netdev->features & NETIF_F_RXHASH)) {
5250 			e_err("Jumbo frames cannot be enabled when both receive checksum offload and receive hashing are enabled.  Disable one of the receive offload features before enabling jumbos.\n");
5251 			return -EINVAL;
5252 		}
5253 	}
5254 
5255 	/* Supported frame sizes */
5256 	if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5257 	    (max_frame > adapter->max_hw_frame_size)) {
5258 		e_err("Unsupported MTU setting\n");
5259 		return -EINVAL;
5260 	}
5261 
5262 	/* Jumbo frame workaround on 82579 requires CRC be stripped */
5263 	if ((adapter->hw.mac.type == e1000_pch2lan) &&
5264 	    !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5265 	    (new_mtu > ETH_DATA_LEN)) {
5266 		e_err("Jumbo Frames not supported on 82579 when CRC stripping is disabled.\n");
5267 		return -EINVAL;
5268 	}
5269 
5270 	/* 82573 Errata 17 */
5271 	if (((adapter->hw.mac.type == e1000_82573) ||
5272 	     (adapter->hw.mac.type == e1000_82574)) &&
5273 	    (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
5274 		adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
5275 		e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
5276 	}
5277 
5278 	while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5279 		usleep_range(1000, 2000);
5280 	/* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5281 	adapter->max_frame_size = max_frame;
5282 	e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5283 	netdev->mtu = new_mtu;
5284 	if (netif_running(netdev))
5285 		e1000e_down(adapter);
5286 
5287 	/*
5288 	 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5289 	 * means we reserve 2 more, this pushes us to allocate from the next
5290 	 * larger slab size.
5291 	 * i.e. RXBUFFER_2048 --> size-4096 slab
5292 	 * However with the new *_jumbo_rx* routines, jumbo receives will use
5293 	 * fragmented skbs
5294 	 */
5295 
5296 	if (max_frame <= 2048)
5297 		adapter->rx_buffer_len = 2048;
5298 	else
5299 		adapter->rx_buffer_len = 4096;
5300 
5301 	/* adjust allocation if LPE protects us, and we aren't using SBP */
5302 	if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5303 	     (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5304 		adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5305 					 + ETH_FCS_LEN;
5306 
5307 	if (netif_running(netdev))
5308 		e1000e_up(adapter);
5309 	else
5310 		e1000e_reset(adapter);
5311 
5312 	clear_bit(__E1000_RESETTING, &adapter->state);
5313 
5314 	return 0;
5315 }
5316 
5317 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5318 			   int cmd)
5319 {
5320 	struct e1000_adapter *adapter = netdev_priv(netdev);
5321 	struct mii_ioctl_data *data = if_mii(ifr);
5322 
5323 	if (adapter->hw.phy.media_type != e1000_media_type_copper)
5324 		return -EOPNOTSUPP;
5325 
5326 	switch (cmd) {
5327 	case SIOCGMIIPHY:
5328 		data->phy_id = adapter->hw.phy.addr;
5329 		break;
5330 	case SIOCGMIIREG:
5331 		e1000_phy_read_status(adapter);
5332 
5333 		switch (data->reg_num & 0x1F) {
5334 		case MII_BMCR:
5335 			data->val_out = adapter->phy_regs.bmcr;
5336 			break;
5337 		case MII_BMSR:
5338 			data->val_out = adapter->phy_regs.bmsr;
5339 			break;
5340 		case MII_PHYSID1:
5341 			data->val_out = (adapter->hw.phy.id >> 16);
5342 			break;
5343 		case MII_PHYSID2:
5344 			data->val_out = (adapter->hw.phy.id & 0xFFFF);
5345 			break;
5346 		case MII_ADVERTISE:
5347 			data->val_out = adapter->phy_regs.advertise;
5348 			break;
5349 		case MII_LPA:
5350 			data->val_out = adapter->phy_regs.lpa;
5351 			break;
5352 		case MII_EXPANSION:
5353 			data->val_out = adapter->phy_regs.expansion;
5354 			break;
5355 		case MII_CTRL1000:
5356 			data->val_out = adapter->phy_regs.ctrl1000;
5357 			break;
5358 		case MII_STAT1000:
5359 			data->val_out = adapter->phy_regs.stat1000;
5360 			break;
5361 		case MII_ESTATUS:
5362 			data->val_out = adapter->phy_regs.estatus;
5363 			break;
5364 		default:
5365 			return -EIO;
5366 		}
5367 		break;
5368 	case SIOCSMIIREG:
5369 	default:
5370 		return -EOPNOTSUPP;
5371 	}
5372 	return 0;
5373 }
5374 
5375 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5376 {
5377 	switch (cmd) {
5378 	case SIOCGMIIPHY:
5379 	case SIOCGMIIREG:
5380 	case SIOCSMIIREG:
5381 		return e1000_mii_ioctl(netdev, ifr, cmd);
5382 	default:
5383 		return -EOPNOTSUPP;
5384 	}
5385 }
5386 
5387 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5388 {
5389 	struct e1000_hw *hw = &adapter->hw;
5390 	u32 i, mac_reg;
5391 	u16 phy_reg, wuc_enable;
5392 	int retval = 0;
5393 
5394 	/* copy MAC RARs to PHY RARs */
5395 	e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5396 
5397 	retval = hw->phy.ops.acquire(hw);
5398 	if (retval) {
5399 		e_err("Could not acquire PHY\n");
5400 		return retval;
5401 	}
5402 
5403 	/* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
5404 	retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5405 	if (retval)
5406 		goto release;
5407 
5408 	/* copy MAC MTA to PHY MTA - only needed for pchlan */
5409 	for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5410 		mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5411 		hw->phy.ops.write_reg_page(hw, BM_MTA(i),
5412 					   (u16)(mac_reg & 0xFFFF));
5413 		hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
5414 					   (u16)((mac_reg >> 16) & 0xFFFF));
5415 	}
5416 
5417 	/* configure PHY Rx Control register */
5418 	hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
5419 	mac_reg = er32(RCTL);
5420 	if (mac_reg & E1000_RCTL_UPE)
5421 		phy_reg |= BM_RCTL_UPE;
5422 	if (mac_reg & E1000_RCTL_MPE)
5423 		phy_reg |= BM_RCTL_MPE;
5424 	phy_reg &= ~(BM_RCTL_MO_MASK);
5425 	if (mac_reg & E1000_RCTL_MO_3)
5426 		phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5427 				<< BM_RCTL_MO_SHIFT);
5428 	if (mac_reg & E1000_RCTL_BAM)
5429 		phy_reg |= BM_RCTL_BAM;
5430 	if (mac_reg & E1000_RCTL_PMCF)
5431 		phy_reg |= BM_RCTL_PMCF;
5432 	mac_reg = er32(CTRL);
5433 	if (mac_reg & E1000_CTRL_RFCE)
5434 		phy_reg |= BM_RCTL_RFCE;
5435 	hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
5436 
5437 	/* enable PHY wakeup in MAC register */
5438 	ew32(WUFC, wufc);
5439 	ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5440 
5441 	/* configure and enable PHY wakeup in PHY registers */
5442 	hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
5443 	hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5444 
5445 	/* activate PHY wakeup */
5446 	wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5447 	retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5448 	if (retval)
5449 		e_err("Could not set PHY Host Wakeup bit\n");
5450 release:
5451 	hw->phy.ops.release(hw);
5452 
5453 	return retval;
5454 }
5455 
5456 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5457 			    bool runtime)
5458 {
5459 	struct net_device *netdev = pci_get_drvdata(pdev);
5460 	struct e1000_adapter *adapter = netdev_priv(netdev);
5461 	struct e1000_hw *hw = &adapter->hw;
5462 	u32 ctrl, ctrl_ext, rctl, status;
5463 	/* Runtime suspend should only enable wakeup for link changes */
5464 	u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5465 	int retval = 0;
5466 
5467 	netif_device_detach(netdev);
5468 
5469 	if (netif_running(netdev)) {
5470 		WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5471 		e1000e_down(adapter);
5472 		e1000_free_irq(adapter);
5473 	}
5474 	e1000e_reset_interrupt_capability(adapter);
5475 
5476 	retval = pci_save_state(pdev);
5477 	if (retval)
5478 		return retval;
5479 
5480 	status = er32(STATUS);
5481 	if (status & E1000_STATUS_LU)
5482 		wufc &= ~E1000_WUFC_LNKC;
5483 
5484 	if (wufc) {
5485 		e1000_setup_rctl(adapter);
5486 		e1000e_set_rx_mode(netdev);
5487 
5488 		/* turn on all-multi mode if wake on multicast is enabled */
5489 		if (wufc & E1000_WUFC_MC) {
5490 			rctl = er32(RCTL);
5491 			rctl |= E1000_RCTL_MPE;
5492 			ew32(RCTL, rctl);
5493 		}
5494 
5495 		ctrl = er32(CTRL);
5496 		/* advertise wake from D3Cold */
5497 		#define E1000_CTRL_ADVD3WUC 0x00100000
5498 		/* phy power management enable */
5499 		#define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5500 		ctrl |= E1000_CTRL_ADVD3WUC;
5501 		if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5502 			ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5503 		ew32(CTRL, ctrl);
5504 
5505 		if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5506 		    adapter->hw.phy.media_type ==
5507 		    e1000_media_type_internal_serdes) {
5508 			/* keep the laser running in D3 */
5509 			ctrl_ext = er32(CTRL_EXT);
5510 			ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5511 			ew32(CTRL_EXT, ctrl_ext);
5512 		}
5513 
5514 		if (adapter->flags & FLAG_IS_ICH)
5515 			e1000_suspend_workarounds_ich8lan(&adapter->hw);
5516 
5517 		/* Allow time for pending master requests to run */
5518 		e1000e_disable_pcie_master(&adapter->hw);
5519 
5520 		if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5521 			/* enable wakeup by the PHY */
5522 			retval = e1000_init_phy_wakeup(adapter, wufc);
5523 			if (retval)
5524 				return retval;
5525 		} else {
5526 			/* enable wakeup by the MAC */
5527 			ew32(WUFC, wufc);
5528 			ew32(WUC, E1000_WUC_PME_EN);
5529 		}
5530 	} else {
5531 		ew32(WUC, 0);
5532 		ew32(WUFC, 0);
5533 	}
5534 
5535 	*enable_wake = !!wufc;
5536 
5537 	/* make sure adapter isn't asleep if manageability is enabled */
5538 	if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5539 	    (hw->mac.ops.check_mng_mode(hw)))
5540 		*enable_wake = true;
5541 
5542 	if (adapter->hw.phy.type == e1000_phy_igp_3)
5543 		e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5544 
5545 	/*
5546 	 * Release control of h/w to f/w.  If f/w is AMT enabled, this
5547 	 * would have already happened in close and is redundant.
5548 	 */
5549 	e1000e_release_hw_control(adapter);
5550 
5551 	pci_disable_device(pdev);
5552 
5553 	return 0;
5554 }
5555 
5556 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5557 {
5558 	if (sleep && wake) {
5559 		pci_prepare_to_sleep(pdev);
5560 		return;
5561 	}
5562 
5563 	pci_wake_from_d3(pdev, wake);
5564 	pci_set_power_state(pdev, PCI_D3hot);
5565 }
5566 
5567 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5568                                     bool wake)
5569 {
5570 	struct net_device *netdev = pci_get_drvdata(pdev);
5571 	struct e1000_adapter *adapter = netdev_priv(netdev);
5572 
5573 	/*
5574 	 * The pci-e switch on some quad port adapters will report a
5575 	 * correctable error when the MAC transitions from D0 to D3.  To
5576 	 * prevent this we need to mask off the correctable errors on the
5577 	 * downstream port of the pci-e switch.
5578 	 */
5579 	if (adapter->flags & FLAG_IS_QUAD_PORT) {
5580 		struct pci_dev *us_dev = pdev->bus->self;
5581 		int pos = pci_pcie_cap(us_dev);
5582 		u16 devctl;
5583 
5584 		pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5585 		pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5586 		                      (devctl & ~PCI_EXP_DEVCTL_CERE));
5587 
5588 		e1000_power_off(pdev, sleep, wake);
5589 
5590 		pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5591 	} else {
5592 		e1000_power_off(pdev, sleep, wake);
5593 	}
5594 }
5595 
5596 #ifdef CONFIG_PCIEASPM
5597 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5598 {
5599 	pci_disable_link_state_locked(pdev, state);
5600 }
5601 #else
5602 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5603 {
5604 	int pos;
5605 	u16 reg16;
5606 
5607 	/*
5608 	 * Both device and parent should have the same ASPM setting.
5609 	 * Disable ASPM in downstream component first and then upstream.
5610 	 */
5611 	pos = pci_pcie_cap(pdev);
5612 	pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5613 	reg16 &= ~state;
5614 	pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5615 
5616 	if (!pdev->bus->self)
5617 		return;
5618 
5619 	pos = pci_pcie_cap(pdev->bus->self);
5620 	pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5621 	reg16 &= ~state;
5622 	pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5623 }
5624 #endif
5625 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5626 {
5627 	dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5628 		 (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5629 		 (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5630 
5631 	__e1000e_disable_aspm(pdev, state);
5632 }
5633 
5634 #ifdef CONFIG_PM
5635 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5636 {
5637 	return !!adapter->tx_ring->buffer_info;
5638 }
5639 
5640 static int __e1000_resume(struct pci_dev *pdev)
5641 {
5642 	struct net_device *netdev = pci_get_drvdata(pdev);
5643 	struct e1000_adapter *adapter = netdev_priv(netdev);
5644 	struct e1000_hw *hw = &adapter->hw;
5645 	u16 aspm_disable_flag = 0;
5646 	u32 err;
5647 
5648 	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5649 		aspm_disable_flag = PCIE_LINK_STATE_L0S;
5650 	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5651 		aspm_disable_flag |= PCIE_LINK_STATE_L1;
5652 	if (aspm_disable_flag)
5653 		e1000e_disable_aspm(pdev, aspm_disable_flag);
5654 
5655 	pci_set_power_state(pdev, PCI_D0);
5656 	pci_restore_state(pdev);
5657 	pci_save_state(pdev);
5658 
5659 	e1000e_set_interrupt_capability(adapter);
5660 	if (netif_running(netdev)) {
5661 		err = e1000_request_irq(adapter);
5662 		if (err)
5663 			return err;
5664 	}
5665 
5666 	if (hw->mac.type == e1000_pch2lan)
5667 		e1000_resume_workarounds_pchlan(&adapter->hw);
5668 
5669 	e1000e_power_up_phy(adapter);
5670 
5671 	/* report the system wakeup cause from S3/S4 */
5672 	if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5673 		u16 phy_data;
5674 
5675 		e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5676 		if (phy_data) {
5677 			e_info("PHY Wakeup cause - %s\n",
5678 				phy_data & E1000_WUS_EX ? "Unicast Packet" :
5679 				phy_data & E1000_WUS_MC ? "Multicast Packet" :
5680 				phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5681 				phy_data & E1000_WUS_MAG ? "Magic Packet" :
5682 				phy_data & E1000_WUS_LNKC ?
5683 				"Link Status Change" : "other");
5684 		}
5685 		e1e_wphy(&adapter->hw, BM_WUS, ~0);
5686 	} else {
5687 		u32 wus = er32(WUS);
5688 		if (wus) {
5689 			e_info("MAC Wakeup cause - %s\n",
5690 				wus & E1000_WUS_EX ? "Unicast Packet" :
5691 				wus & E1000_WUS_MC ? "Multicast Packet" :
5692 				wus & E1000_WUS_BC ? "Broadcast Packet" :
5693 				wus & E1000_WUS_MAG ? "Magic Packet" :
5694 				wus & E1000_WUS_LNKC ? "Link Status Change" :
5695 				"other");
5696 		}
5697 		ew32(WUS, ~0);
5698 	}
5699 
5700 	e1000e_reset(adapter);
5701 
5702 	e1000_init_manageability_pt(adapter);
5703 
5704 	if (netif_running(netdev))
5705 		e1000e_up(adapter);
5706 
5707 	netif_device_attach(netdev);
5708 
5709 	/*
5710 	 * If the controller has AMT, do not set DRV_LOAD until the interface
5711 	 * is up.  For all other cases, let the f/w know that the h/w is now
5712 	 * under the control of the driver.
5713 	 */
5714 	if (!(adapter->flags & FLAG_HAS_AMT))
5715 		e1000e_get_hw_control(adapter);
5716 
5717 	return 0;
5718 }
5719 
5720 #ifdef CONFIG_PM_SLEEP
5721 static int e1000_suspend(struct device *dev)
5722 {
5723 	struct pci_dev *pdev = to_pci_dev(dev);
5724 	int retval;
5725 	bool wake;
5726 
5727 	retval = __e1000_shutdown(pdev, &wake, false);
5728 	if (!retval)
5729 		e1000_complete_shutdown(pdev, true, wake);
5730 
5731 	return retval;
5732 }
5733 
5734 static int e1000_resume(struct device *dev)
5735 {
5736 	struct pci_dev *pdev = to_pci_dev(dev);
5737 	struct net_device *netdev = pci_get_drvdata(pdev);
5738 	struct e1000_adapter *adapter = netdev_priv(netdev);
5739 
5740 	if (e1000e_pm_ready(adapter))
5741 		adapter->idle_check = true;
5742 
5743 	return __e1000_resume(pdev);
5744 }
5745 #endif /* CONFIG_PM_SLEEP */
5746 
5747 #ifdef CONFIG_PM_RUNTIME
5748 static int e1000_runtime_suspend(struct device *dev)
5749 {
5750 	struct pci_dev *pdev = to_pci_dev(dev);
5751 	struct net_device *netdev = pci_get_drvdata(pdev);
5752 	struct e1000_adapter *adapter = netdev_priv(netdev);
5753 
5754 	if (e1000e_pm_ready(adapter)) {
5755 		bool wake;
5756 
5757 		__e1000_shutdown(pdev, &wake, true);
5758 	}
5759 
5760 	return 0;
5761 }
5762 
5763 static int e1000_idle(struct device *dev)
5764 {
5765 	struct pci_dev *pdev = to_pci_dev(dev);
5766 	struct net_device *netdev = pci_get_drvdata(pdev);
5767 	struct e1000_adapter *adapter = netdev_priv(netdev);
5768 
5769 	if (!e1000e_pm_ready(adapter))
5770 		return 0;
5771 
5772 	if (adapter->idle_check) {
5773 		adapter->idle_check = false;
5774 		if (!e1000e_has_link(adapter))
5775 			pm_schedule_suspend(dev, MSEC_PER_SEC);
5776 	}
5777 
5778 	return -EBUSY;
5779 }
5780 
5781 static int e1000_runtime_resume(struct device *dev)
5782 {
5783 	struct pci_dev *pdev = to_pci_dev(dev);
5784 	struct net_device *netdev = pci_get_drvdata(pdev);
5785 	struct e1000_adapter *adapter = netdev_priv(netdev);
5786 
5787 	if (!e1000e_pm_ready(adapter))
5788 		return 0;
5789 
5790 	adapter->idle_check = !dev->power.runtime_auto;
5791 	return __e1000_resume(pdev);
5792 }
5793 #endif /* CONFIG_PM_RUNTIME */
5794 #endif /* CONFIG_PM */
5795 
5796 static void e1000_shutdown(struct pci_dev *pdev)
5797 {
5798 	bool wake = false;
5799 
5800 	__e1000_shutdown(pdev, &wake, false);
5801 
5802 	if (system_state == SYSTEM_POWER_OFF)
5803 		e1000_complete_shutdown(pdev, false, wake);
5804 }
5805 
5806 #ifdef CONFIG_NET_POLL_CONTROLLER
5807 
5808 static irqreturn_t e1000_intr_msix(int irq, void *data)
5809 {
5810 	struct net_device *netdev = data;
5811 	struct e1000_adapter *adapter = netdev_priv(netdev);
5812 
5813 	if (adapter->msix_entries) {
5814 		int vector, msix_irq;
5815 
5816 		vector = 0;
5817 		msix_irq = adapter->msix_entries[vector].vector;
5818 		disable_irq(msix_irq);
5819 		e1000_intr_msix_rx(msix_irq, netdev);
5820 		enable_irq(msix_irq);
5821 
5822 		vector++;
5823 		msix_irq = adapter->msix_entries[vector].vector;
5824 		disable_irq(msix_irq);
5825 		e1000_intr_msix_tx(msix_irq, netdev);
5826 		enable_irq(msix_irq);
5827 
5828 		vector++;
5829 		msix_irq = adapter->msix_entries[vector].vector;
5830 		disable_irq(msix_irq);
5831 		e1000_msix_other(msix_irq, netdev);
5832 		enable_irq(msix_irq);
5833 	}
5834 
5835 	return IRQ_HANDLED;
5836 }
5837 
5838 /*
5839  * Polling 'interrupt' - used by things like netconsole to send skbs
5840  * without having to re-enable interrupts. It's not called while
5841  * the interrupt routine is executing.
5842  */
5843 static void e1000_netpoll(struct net_device *netdev)
5844 {
5845 	struct e1000_adapter *adapter = netdev_priv(netdev);
5846 
5847 	switch (adapter->int_mode) {
5848 	case E1000E_INT_MODE_MSIX:
5849 		e1000_intr_msix(adapter->pdev->irq, netdev);
5850 		break;
5851 	case E1000E_INT_MODE_MSI:
5852 		disable_irq(adapter->pdev->irq);
5853 		e1000_intr_msi(adapter->pdev->irq, netdev);
5854 		enable_irq(adapter->pdev->irq);
5855 		break;
5856 	default: /* E1000E_INT_MODE_LEGACY */
5857 		disable_irq(adapter->pdev->irq);
5858 		e1000_intr(adapter->pdev->irq, netdev);
5859 		enable_irq(adapter->pdev->irq);
5860 		break;
5861 	}
5862 }
5863 #endif
5864 
5865 /**
5866  * e1000_io_error_detected - called when PCI error is detected
5867  * @pdev: Pointer to PCI device
5868  * @state: The current pci connection state
5869  *
5870  * This function is called after a PCI bus error affecting
5871  * this device has been detected.
5872  */
5873 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5874 						pci_channel_state_t state)
5875 {
5876 	struct net_device *netdev = pci_get_drvdata(pdev);
5877 	struct e1000_adapter *adapter = netdev_priv(netdev);
5878 
5879 	netif_device_detach(netdev);
5880 
5881 	if (state == pci_channel_io_perm_failure)
5882 		return PCI_ERS_RESULT_DISCONNECT;
5883 
5884 	if (netif_running(netdev))
5885 		e1000e_down(adapter);
5886 	pci_disable_device(pdev);
5887 
5888 	/* Request a slot slot reset. */
5889 	return PCI_ERS_RESULT_NEED_RESET;
5890 }
5891 
5892 /**
5893  * e1000_io_slot_reset - called after the pci bus has been reset.
5894  * @pdev: Pointer to PCI device
5895  *
5896  * Restart the card from scratch, as if from a cold-boot. Implementation
5897  * resembles the first-half of the e1000_resume routine.
5898  */
5899 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5900 {
5901 	struct net_device *netdev = pci_get_drvdata(pdev);
5902 	struct e1000_adapter *adapter = netdev_priv(netdev);
5903 	struct e1000_hw *hw = &adapter->hw;
5904 	u16 aspm_disable_flag = 0;
5905 	int err;
5906 	pci_ers_result_t result;
5907 
5908 	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5909 		aspm_disable_flag = PCIE_LINK_STATE_L0S;
5910 	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5911 		aspm_disable_flag |= PCIE_LINK_STATE_L1;
5912 	if (aspm_disable_flag)
5913 		e1000e_disable_aspm(pdev, aspm_disable_flag);
5914 
5915 	err = pci_enable_device_mem(pdev);
5916 	if (err) {
5917 		dev_err(&pdev->dev,
5918 			"Cannot re-enable PCI device after reset.\n");
5919 		result = PCI_ERS_RESULT_DISCONNECT;
5920 	} else {
5921 		pci_set_master(pdev);
5922 		pdev->state_saved = true;
5923 		pci_restore_state(pdev);
5924 
5925 		pci_enable_wake(pdev, PCI_D3hot, 0);
5926 		pci_enable_wake(pdev, PCI_D3cold, 0);
5927 
5928 		e1000e_reset(adapter);
5929 		ew32(WUS, ~0);
5930 		result = PCI_ERS_RESULT_RECOVERED;
5931 	}
5932 
5933 	pci_cleanup_aer_uncorrect_error_status(pdev);
5934 
5935 	return result;
5936 }
5937 
5938 /**
5939  * e1000_io_resume - called when traffic can start flowing again.
5940  * @pdev: Pointer to PCI device
5941  *
5942  * This callback is called when the error recovery driver tells us that
5943  * its OK to resume normal operation. Implementation resembles the
5944  * second-half of the e1000_resume routine.
5945  */
5946 static void e1000_io_resume(struct pci_dev *pdev)
5947 {
5948 	struct net_device *netdev = pci_get_drvdata(pdev);
5949 	struct e1000_adapter *adapter = netdev_priv(netdev);
5950 
5951 	e1000_init_manageability_pt(adapter);
5952 
5953 	if (netif_running(netdev)) {
5954 		if (e1000e_up(adapter)) {
5955 			dev_err(&pdev->dev,
5956 				"can't bring device back up after reset\n");
5957 			return;
5958 		}
5959 	}
5960 
5961 	netif_device_attach(netdev);
5962 
5963 	/*
5964 	 * If the controller has AMT, do not set DRV_LOAD until the interface
5965 	 * is up.  For all other cases, let the f/w know that the h/w is now
5966 	 * under the control of the driver.
5967 	 */
5968 	if (!(adapter->flags & FLAG_HAS_AMT))
5969 		e1000e_get_hw_control(adapter);
5970 
5971 }
5972 
5973 static void e1000_print_device_info(struct e1000_adapter *adapter)
5974 {
5975 	struct e1000_hw *hw = &adapter->hw;
5976 	struct net_device *netdev = adapter->netdev;
5977 	u32 ret_val;
5978 	u8 pba_str[E1000_PBANUM_LENGTH];
5979 
5980 	/* print bus type/speed/width info */
5981 	e_info("(PCI Express:2.5GT/s:%s) %pM\n",
5982 	       /* bus width */
5983 	       ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5984 	        "Width x1"),
5985 	       /* MAC address */
5986 	       netdev->dev_addr);
5987 	e_info("Intel(R) PRO/%s Network Connection\n",
5988 	       (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5989 	ret_val = e1000_read_pba_string_generic(hw, pba_str,
5990 						E1000_PBANUM_LENGTH);
5991 	if (ret_val)
5992 		strlcpy((char *)pba_str, "Unknown", sizeof(pba_str));
5993 	e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5994 	       hw->mac.type, hw->phy.type, pba_str);
5995 }
5996 
5997 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5998 {
5999 	struct e1000_hw *hw = &adapter->hw;
6000 	int ret_val;
6001 	u16 buf = 0;
6002 
6003 	if (hw->mac.type != e1000_82573)
6004 		return;
6005 
6006 	ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
6007 	le16_to_cpus(&buf);
6008 	if (!ret_val && (!(buf & (1 << 0)))) {
6009 		/* Deep Smart Power Down (DSPD) */
6010 		dev_warn(&adapter->pdev->dev,
6011 			 "Warning: detected DSPD enabled in EEPROM\n");
6012 	}
6013 }
6014 
6015 static int e1000_set_features(struct net_device *netdev,
6016 			      netdev_features_t features)
6017 {
6018 	struct e1000_adapter *adapter = netdev_priv(netdev);
6019 	netdev_features_t changed = features ^ netdev->features;
6020 
6021 	if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
6022 		adapter->flags |= FLAG_TSO_FORCE;
6023 
6024 	if (!(changed & (NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX |
6025 			 NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_RXFCS |
6026 			 NETIF_F_RXALL)))
6027 		return 0;
6028 
6029 	/*
6030 	 * IP payload checksum (enabled with jumbos/packet-split when Rx
6031 	 * checksum is enabled) and generation of RSS hash is mutually
6032 	 * exclusive in the hardware.
6033 	 */
6034 	if (adapter->rx_ps_pages &&
6035 	    (features & NETIF_F_RXCSUM) && (features & NETIF_F_RXHASH)) {
6036 		e_err("Enabling both receive checksum offload and receive hashing is not possible with jumbo frames.  Disable jumbos or enable only one of the receive offload features.\n");
6037 		return -EINVAL;
6038 	}
6039 
6040 	if (changed & NETIF_F_RXFCS) {
6041 		if (features & NETIF_F_RXFCS) {
6042 			adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6043 		} else {
6044 			/* We need to take it back to defaults, which might mean
6045 			 * stripping is still disabled at the adapter level.
6046 			 */
6047 			if (adapter->flags2 & FLAG2_DFLT_CRC_STRIPPING)
6048 				adapter->flags2 |= FLAG2_CRC_STRIPPING;
6049 			else
6050 				adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6051 		}
6052 	}
6053 
6054 	netdev->features = features;
6055 
6056 	if (netif_running(netdev))
6057 		e1000e_reinit_locked(adapter);
6058 	else
6059 		e1000e_reset(adapter);
6060 
6061 	return 0;
6062 }
6063 
6064 static const struct net_device_ops e1000e_netdev_ops = {
6065 	.ndo_open		= e1000_open,
6066 	.ndo_stop		= e1000_close,
6067 	.ndo_start_xmit		= e1000_xmit_frame,
6068 	.ndo_get_stats64	= e1000e_get_stats64,
6069 	.ndo_set_rx_mode	= e1000e_set_rx_mode,
6070 	.ndo_set_mac_address	= e1000_set_mac,
6071 	.ndo_change_mtu		= e1000_change_mtu,
6072 	.ndo_do_ioctl		= e1000_ioctl,
6073 	.ndo_tx_timeout		= e1000_tx_timeout,
6074 	.ndo_validate_addr	= eth_validate_addr,
6075 
6076 	.ndo_vlan_rx_add_vid	= e1000_vlan_rx_add_vid,
6077 	.ndo_vlan_rx_kill_vid	= e1000_vlan_rx_kill_vid,
6078 #ifdef CONFIG_NET_POLL_CONTROLLER
6079 	.ndo_poll_controller	= e1000_netpoll,
6080 #endif
6081 	.ndo_set_features = e1000_set_features,
6082 };
6083 
6084 /**
6085  * e1000_probe - Device Initialization Routine
6086  * @pdev: PCI device information struct
6087  * @ent: entry in e1000_pci_tbl
6088  *
6089  * Returns 0 on success, negative on failure
6090  *
6091  * e1000_probe initializes an adapter identified by a pci_dev structure.
6092  * The OS initialization, configuring of the adapter private structure,
6093  * and a hardware reset occur.
6094  **/
6095 static int __devinit e1000_probe(struct pci_dev *pdev,
6096 				 const struct pci_device_id *ent)
6097 {
6098 	struct net_device *netdev;
6099 	struct e1000_adapter *adapter;
6100 	struct e1000_hw *hw;
6101 	const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
6102 	resource_size_t mmio_start, mmio_len;
6103 	resource_size_t flash_start, flash_len;
6104 	static int cards_found;
6105 	u16 aspm_disable_flag = 0;
6106 	int i, err, pci_using_dac;
6107 	u16 eeprom_data = 0;
6108 	u16 eeprom_apme_mask = E1000_EEPROM_APME;
6109 
6110 	if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
6111 		aspm_disable_flag = PCIE_LINK_STATE_L0S;
6112 	if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
6113 		aspm_disable_flag |= PCIE_LINK_STATE_L1;
6114 	if (aspm_disable_flag)
6115 		e1000e_disable_aspm(pdev, aspm_disable_flag);
6116 
6117 	err = pci_enable_device_mem(pdev);
6118 	if (err)
6119 		return err;
6120 
6121 	pci_using_dac = 0;
6122 	err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
6123 	if (!err) {
6124 		err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
6125 		if (!err)
6126 			pci_using_dac = 1;
6127 	} else {
6128 		err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6129 		if (err) {
6130 			err = dma_set_coherent_mask(&pdev->dev,
6131 						    DMA_BIT_MASK(32));
6132 			if (err) {
6133 				dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
6134 				goto err_dma;
6135 			}
6136 		}
6137 	}
6138 
6139 	err = pci_request_selected_regions_exclusive(pdev,
6140 	                                  pci_select_bars(pdev, IORESOURCE_MEM),
6141 	                                  e1000e_driver_name);
6142 	if (err)
6143 		goto err_pci_reg;
6144 
6145 	/* AER (Advanced Error Reporting) hooks */
6146 	pci_enable_pcie_error_reporting(pdev);
6147 
6148 	pci_set_master(pdev);
6149 	/* PCI config space info */
6150 	err = pci_save_state(pdev);
6151 	if (err)
6152 		goto err_alloc_etherdev;
6153 
6154 	err = -ENOMEM;
6155 	netdev = alloc_etherdev(sizeof(struct e1000_adapter));
6156 	if (!netdev)
6157 		goto err_alloc_etherdev;
6158 
6159 	SET_NETDEV_DEV(netdev, &pdev->dev);
6160 
6161 	netdev->irq = pdev->irq;
6162 
6163 	pci_set_drvdata(pdev, netdev);
6164 	adapter = netdev_priv(netdev);
6165 	hw = &adapter->hw;
6166 	adapter->netdev = netdev;
6167 	adapter->pdev = pdev;
6168 	adapter->ei = ei;
6169 	adapter->pba = ei->pba;
6170 	adapter->flags = ei->flags;
6171 	adapter->flags2 = ei->flags2;
6172 	adapter->hw.adapter = adapter;
6173 	adapter->hw.mac.type = ei->mac;
6174 	adapter->max_hw_frame_size = ei->max_hw_frame_size;
6175 	adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
6176 
6177 	mmio_start = pci_resource_start(pdev, 0);
6178 	mmio_len = pci_resource_len(pdev, 0);
6179 
6180 	err = -EIO;
6181 	adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
6182 	if (!adapter->hw.hw_addr)
6183 		goto err_ioremap;
6184 
6185 	if ((adapter->flags & FLAG_HAS_FLASH) &&
6186 	    (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
6187 		flash_start = pci_resource_start(pdev, 1);
6188 		flash_len = pci_resource_len(pdev, 1);
6189 		adapter->hw.flash_address = ioremap(flash_start, flash_len);
6190 		if (!adapter->hw.flash_address)
6191 			goto err_flashmap;
6192 	}
6193 
6194 	/* construct the net_device struct */
6195 	netdev->netdev_ops		= &e1000e_netdev_ops;
6196 	e1000e_set_ethtool_ops(netdev);
6197 	netdev->watchdog_timeo		= 5 * HZ;
6198 	netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
6199 	strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
6200 
6201 	netdev->mem_start = mmio_start;
6202 	netdev->mem_end = mmio_start + mmio_len;
6203 
6204 	adapter->bd_number = cards_found++;
6205 
6206 	e1000e_check_options(adapter);
6207 
6208 	/* setup adapter struct */
6209 	err = e1000_sw_init(adapter);
6210 	if (err)
6211 		goto err_sw_init;
6212 
6213 	memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
6214 	memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
6215 	memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
6216 
6217 	err = ei->get_variants(adapter);
6218 	if (err)
6219 		goto err_hw_init;
6220 
6221 	if ((adapter->flags & FLAG_IS_ICH) &&
6222 	    (adapter->flags & FLAG_READ_ONLY_NVM))
6223 		e1000e_write_protect_nvm_ich8lan(&adapter->hw);
6224 
6225 	hw->mac.ops.get_bus_info(&adapter->hw);
6226 
6227 	adapter->hw.phy.autoneg_wait_to_complete = 0;
6228 
6229 	/* Copper options */
6230 	if (adapter->hw.phy.media_type == e1000_media_type_copper) {
6231 		adapter->hw.phy.mdix = AUTO_ALL_MODES;
6232 		adapter->hw.phy.disable_polarity_correction = 0;
6233 		adapter->hw.phy.ms_type = e1000_ms_hw_default;
6234 	}
6235 
6236 	if (hw->phy.ops.check_reset_block(hw))
6237 		e_info("PHY reset is blocked due to SOL/IDER session.\n");
6238 
6239 	/* Set initial default active device features */
6240 	netdev->features = (NETIF_F_SG |
6241 			    NETIF_F_HW_VLAN_RX |
6242 			    NETIF_F_HW_VLAN_TX |
6243 			    NETIF_F_TSO |
6244 			    NETIF_F_TSO6 |
6245 			    NETIF_F_RXHASH |
6246 			    NETIF_F_RXCSUM |
6247 			    NETIF_F_HW_CSUM);
6248 
6249 	/* Set user-changeable features (subset of all device features) */
6250 	netdev->hw_features = netdev->features;
6251 	netdev->hw_features |= NETIF_F_RXFCS;
6252 	netdev->priv_flags |= IFF_SUPP_NOFCS;
6253 	netdev->hw_features |= NETIF_F_RXALL;
6254 
6255 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
6256 		netdev->features |= NETIF_F_HW_VLAN_FILTER;
6257 
6258 	netdev->vlan_features |= (NETIF_F_SG |
6259 				  NETIF_F_TSO |
6260 				  NETIF_F_TSO6 |
6261 				  NETIF_F_HW_CSUM);
6262 
6263 	netdev->priv_flags |= IFF_UNICAST_FLT;
6264 
6265 	if (pci_using_dac) {
6266 		netdev->features |= NETIF_F_HIGHDMA;
6267 		netdev->vlan_features |= NETIF_F_HIGHDMA;
6268 	}
6269 
6270 	if (e1000e_enable_mng_pass_thru(&adapter->hw))
6271 		adapter->flags |= FLAG_MNG_PT_ENABLED;
6272 
6273 	/*
6274 	 * before reading the NVM, reset the controller to
6275 	 * put the device in a known good starting state
6276 	 */
6277 	adapter->hw.mac.ops.reset_hw(&adapter->hw);
6278 
6279 	/*
6280 	 * systems with ASPM and others may see the checksum fail on the first
6281 	 * attempt. Let's give it a few tries
6282 	 */
6283 	for (i = 0;; i++) {
6284 		if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
6285 			break;
6286 		if (i == 2) {
6287 			e_err("The NVM Checksum Is Not Valid\n");
6288 			err = -EIO;
6289 			goto err_eeprom;
6290 		}
6291 	}
6292 
6293 	e1000_eeprom_checks(adapter);
6294 
6295 	/* copy the MAC address */
6296 	if (e1000e_read_mac_addr(&adapter->hw))
6297 		e_err("NVM Read Error while reading MAC address\n");
6298 
6299 	memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
6300 	memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
6301 
6302 	if (!is_valid_ether_addr(netdev->perm_addr)) {
6303 		e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
6304 		err = -EIO;
6305 		goto err_eeprom;
6306 	}
6307 
6308 	init_timer(&adapter->watchdog_timer);
6309 	adapter->watchdog_timer.function = e1000_watchdog;
6310 	adapter->watchdog_timer.data = (unsigned long) adapter;
6311 
6312 	init_timer(&adapter->phy_info_timer);
6313 	adapter->phy_info_timer.function = e1000_update_phy_info;
6314 	adapter->phy_info_timer.data = (unsigned long) adapter;
6315 
6316 	INIT_WORK(&adapter->reset_task, e1000_reset_task);
6317 	INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
6318 	INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
6319 	INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
6320 	INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
6321 
6322 	/* Initialize link parameters. User can change them with ethtool */
6323 	adapter->hw.mac.autoneg = 1;
6324 	adapter->fc_autoneg = true;
6325 	adapter->hw.fc.requested_mode = e1000_fc_default;
6326 	adapter->hw.fc.current_mode = e1000_fc_default;
6327 	adapter->hw.phy.autoneg_advertised = 0x2f;
6328 
6329 	/* ring size defaults */
6330 	adapter->rx_ring->count = 256;
6331 	adapter->tx_ring->count = 256;
6332 
6333 	/*
6334 	 * Initial Wake on LAN setting - If APM wake is enabled in
6335 	 * the EEPROM, enable the ACPI Magic Packet filter
6336 	 */
6337 	if (adapter->flags & FLAG_APME_IN_WUC) {
6338 		/* APME bit in EEPROM is mapped to WUC.APME */
6339 		eeprom_data = er32(WUC);
6340 		eeprom_apme_mask = E1000_WUC_APME;
6341 		if ((hw->mac.type > e1000_ich10lan) &&
6342 		    (eeprom_data & E1000_WUC_PHY_WAKE))
6343 			adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
6344 	} else if (adapter->flags & FLAG_APME_IN_CTRL3) {
6345 		if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
6346 		    (adapter->hw.bus.func == 1))
6347 			e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_B,
6348 				       1, &eeprom_data);
6349 		else
6350 			e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_A,
6351 				       1, &eeprom_data);
6352 	}
6353 
6354 	/* fetch WoL from EEPROM */
6355 	if (eeprom_data & eeprom_apme_mask)
6356 		adapter->eeprom_wol |= E1000_WUFC_MAG;
6357 
6358 	/*
6359 	 * now that we have the eeprom settings, apply the special cases
6360 	 * where the eeprom may be wrong or the board simply won't support
6361 	 * wake on lan on a particular port
6362 	 */
6363 	if (!(adapter->flags & FLAG_HAS_WOL))
6364 		adapter->eeprom_wol = 0;
6365 
6366 	/* initialize the wol settings based on the eeprom settings */
6367 	adapter->wol = adapter->eeprom_wol;
6368 	device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6369 
6370 	/* save off EEPROM version number */
6371 	e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6372 
6373 	/* reset the hardware with the new settings */
6374 	e1000e_reset(adapter);
6375 
6376 	/*
6377 	 * If the controller has AMT, do not set DRV_LOAD until the interface
6378 	 * is up.  For all other cases, let the f/w know that the h/w is now
6379 	 * under the control of the driver.
6380 	 */
6381 	if (!(adapter->flags & FLAG_HAS_AMT))
6382 		e1000e_get_hw_control(adapter);
6383 
6384 	strlcpy(netdev->name, "eth%d", sizeof(netdev->name));
6385 	err = register_netdev(netdev);
6386 	if (err)
6387 		goto err_register;
6388 
6389 	/* carrier off reporting is important to ethtool even BEFORE open */
6390 	netif_carrier_off(netdev);
6391 
6392 	e1000_print_device_info(adapter);
6393 
6394 	if (pci_dev_run_wake(pdev))
6395 		pm_runtime_put_noidle(&pdev->dev);
6396 
6397 	return 0;
6398 
6399 err_register:
6400 	if (!(adapter->flags & FLAG_HAS_AMT))
6401 		e1000e_release_hw_control(adapter);
6402 err_eeprom:
6403 	if (!hw->phy.ops.check_reset_block(hw))
6404 		e1000_phy_hw_reset(&adapter->hw);
6405 err_hw_init:
6406 	kfree(adapter->tx_ring);
6407 	kfree(adapter->rx_ring);
6408 err_sw_init:
6409 	if (adapter->hw.flash_address)
6410 		iounmap(adapter->hw.flash_address);
6411 	e1000e_reset_interrupt_capability(adapter);
6412 err_flashmap:
6413 	iounmap(adapter->hw.hw_addr);
6414 err_ioremap:
6415 	free_netdev(netdev);
6416 err_alloc_etherdev:
6417 	pci_release_selected_regions(pdev,
6418 	                             pci_select_bars(pdev, IORESOURCE_MEM));
6419 err_pci_reg:
6420 err_dma:
6421 	pci_disable_device(pdev);
6422 	return err;
6423 }
6424 
6425 /**
6426  * e1000_remove - Device Removal Routine
6427  * @pdev: PCI device information struct
6428  *
6429  * e1000_remove is called by the PCI subsystem to alert the driver
6430  * that it should release a PCI device.  The could be caused by a
6431  * Hot-Plug event, or because the driver is going to be removed from
6432  * memory.
6433  **/
6434 static void __devexit e1000_remove(struct pci_dev *pdev)
6435 {
6436 	struct net_device *netdev = pci_get_drvdata(pdev);
6437 	struct e1000_adapter *adapter = netdev_priv(netdev);
6438 	bool down = test_bit(__E1000_DOWN, &adapter->state);
6439 
6440 	/*
6441 	 * The timers may be rescheduled, so explicitly disable them
6442 	 * from being rescheduled.
6443 	 */
6444 	if (!down)
6445 		set_bit(__E1000_DOWN, &adapter->state);
6446 	del_timer_sync(&adapter->watchdog_timer);
6447 	del_timer_sync(&adapter->phy_info_timer);
6448 
6449 	cancel_work_sync(&adapter->reset_task);
6450 	cancel_work_sync(&adapter->watchdog_task);
6451 	cancel_work_sync(&adapter->downshift_task);
6452 	cancel_work_sync(&adapter->update_phy_task);
6453 	cancel_work_sync(&adapter->print_hang_task);
6454 
6455 	if (!(netdev->flags & IFF_UP))
6456 		e1000_power_down_phy(adapter);
6457 
6458 	/* Don't lie to e1000_close() down the road. */
6459 	if (!down)
6460 		clear_bit(__E1000_DOWN, &adapter->state);
6461 	unregister_netdev(netdev);
6462 
6463 	if (pci_dev_run_wake(pdev))
6464 		pm_runtime_get_noresume(&pdev->dev);
6465 
6466 	/*
6467 	 * Release control of h/w to f/w.  If f/w is AMT enabled, this
6468 	 * would have already happened in close and is redundant.
6469 	 */
6470 	e1000e_release_hw_control(adapter);
6471 
6472 	e1000e_reset_interrupt_capability(adapter);
6473 	kfree(adapter->tx_ring);
6474 	kfree(adapter->rx_ring);
6475 
6476 	iounmap(adapter->hw.hw_addr);
6477 	if (adapter->hw.flash_address)
6478 		iounmap(adapter->hw.flash_address);
6479 	pci_release_selected_regions(pdev,
6480 	                             pci_select_bars(pdev, IORESOURCE_MEM));
6481 
6482 	free_netdev(netdev);
6483 
6484 	/* AER disable */
6485 	pci_disable_pcie_error_reporting(pdev);
6486 
6487 	pci_disable_device(pdev);
6488 }
6489 
6490 /* PCI Error Recovery (ERS) */
6491 static struct pci_error_handlers e1000_err_handler = {
6492 	.error_detected = e1000_io_error_detected,
6493 	.slot_reset = e1000_io_slot_reset,
6494 	.resume = e1000_io_resume,
6495 };
6496 
6497 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6498 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6499 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6500 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6501 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6502 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6503 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6504 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6505 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6506 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6507 
6508 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6509 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6510 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6511 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6512 
6513 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6514 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6515 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6516 
6517 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6518 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6519 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6520 
6521 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6522 	  board_80003es2lan },
6523 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6524 	  board_80003es2lan },
6525 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6526 	  board_80003es2lan },
6527 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6528 	  board_80003es2lan },
6529 
6530 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6531 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6532 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6533 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6534 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6535 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6536 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6537 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6538 
6539 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6540 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6541 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6542 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6543 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6544 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6545 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6546 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6547 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6548 
6549 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6550 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6551 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6552 
6553 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6554 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6555 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6556 
6557 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6558 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6559 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6560 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6561 
6562 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6563 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6564 
6565 	{ 0, 0, 0, 0, 0, 0, 0 }	/* terminate list */
6566 };
6567 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6568 
6569 #ifdef CONFIG_PM
6570 static const struct dev_pm_ops e1000_pm_ops = {
6571 	SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6572 	SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6573 				e1000_runtime_resume, e1000_idle)
6574 };
6575 #endif
6576 
6577 /* PCI Device API Driver */
6578 static struct pci_driver e1000_driver = {
6579 	.name     = e1000e_driver_name,
6580 	.id_table = e1000_pci_tbl,
6581 	.probe    = e1000_probe,
6582 	.remove   = __devexit_p(e1000_remove),
6583 #ifdef CONFIG_PM
6584 	.driver   = {
6585 		.pm = &e1000_pm_ops,
6586 	},
6587 #endif
6588 	.shutdown = e1000_shutdown,
6589 	.err_handler = &e1000_err_handler
6590 };
6591 
6592 /**
6593  * e1000_init_module - Driver Registration Routine
6594  *
6595  * e1000_init_module is the first routine called when the driver is
6596  * loaded. All it does is register with the PCI subsystem.
6597  **/
6598 static int __init e1000_init_module(void)
6599 {
6600 	int ret;
6601 	pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6602 		e1000e_driver_version);
6603 	pr_info("Copyright(c) 1999 - 2012 Intel Corporation.\n");
6604 	ret = pci_register_driver(&e1000_driver);
6605 
6606 	return ret;
6607 }
6608 module_init(e1000_init_module);
6609 
6610 /**
6611  * e1000_exit_module - Driver Exit Cleanup Routine
6612  *
6613  * e1000_exit_module is called just before the driver is removed
6614  * from memory.
6615  **/
6616 static void __exit e1000_exit_module(void)
6617 {
6618 	pci_unregister_driver(&e1000_driver);
6619 }
6620 module_exit(e1000_exit_module);
6621 
6622 
6623 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6624 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6625 MODULE_LICENSE("GPL");
6626 MODULE_VERSION(DRV_VERSION);
6627 
6628 /* netdev.c */
6629