xref: /linux/drivers/net/ethernet/intel/e1000e/netdev.c (revision 800c5eb7b5eba6cb2a32738d763fd59f0fbcdde4)
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 		goto exit;
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 		goto exit;
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 exit:
454 	return;
455 }
456 
457 /**
458  * e1000_desc_unused - calculate if we have unused descriptors
459  **/
460 static int e1000_desc_unused(struct e1000_ring *ring)
461 {
462 	if (ring->next_to_clean > ring->next_to_use)
463 		return ring->next_to_clean - ring->next_to_use - 1;
464 
465 	return ring->count + ring->next_to_clean - ring->next_to_use - 1;
466 }
467 
468 /**
469  * e1000_receive_skb - helper function to handle Rx indications
470  * @adapter: board private structure
471  * @status: descriptor status field as written by hardware
472  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
473  * @skb: pointer to sk_buff to be indicated to stack
474  **/
475 static void e1000_receive_skb(struct e1000_adapter *adapter,
476 			      struct net_device *netdev, struct sk_buff *skb,
477 			      u8 status, __le16 vlan)
478 {
479 	u16 tag = le16_to_cpu(vlan);
480 	skb->protocol = eth_type_trans(skb, netdev);
481 
482 	if (status & E1000_RXD_STAT_VP)
483 		__vlan_hwaccel_put_tag(skb, tag);
484 
485 	napi_gro_receive(&adapter->napi, skb);
486 }
487 
488 /**
489  * e1000_rx_checksum - Receive Checksum Offload
490  * @adapter: board private structure
491  * @status_err: receive descriptor status and error fields
492  * @csum: receive descriptor csum field
493  * @sk_buff: socket buffer with received data
494  **/
495 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
496 			      __le16 csum, struct sk_buff *skb)
497 {
498 	u16 status = (u16)status_err;
499 	u8 errors = (u8)(status_err >> 24);
500 
501 	skb_checksum_none_assert(skb);
502 
503 	/* Rx checksum disabled */
504 	if (!(adapter->netdev->features & NETIF_F_RXCSUM))
505 		return;
506 
507 	/* Ignore Checksum bit is set */
508 	if (status & E1000_RXD_STAT_IXSM)
509 		return;
510 
511 	/* TCP/UDP checksum error bit is set */
512 	if (errors & E1000_RXD_ERR_TCPE) {
513 		/* let the stack verify checksum errors */
514 		adapter->hw_csum_err++;
515 		return;
516 	}
517 
518 	/* TCP/UDP Checksum has not been calculated */
519 	if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
520 		return;
521 
522 	/* It must be a TCP or UDP packet with a valid checksum */
523 	if (status & E1000_RXD_STAT_TCPCS) {
524 		/* TCP checksum is good */
525 		skb->ip_summed = CHECKSUM_UNNECESSARY;
526 	} else {
527 		/*
528 		 * IP fragment with UDP payload
529 		 * Hardware complements the payload checksum, so we undo it
530 		 * and then put the value in host order for further stack use.
531 		 */
532 		__sum16 sum = (__force __sum16)swab16((__force u16)csum);
533 		skb->csum = csum_unfold(~sum);
534 		skb->ip_summed = CHECKSUM_COMPLETE;
535 	}
536 	adapter->hw_csum_good++;
537 }
538 
539 /**
540  * e1000e_update_tail_wa - helper function for e1000e_update_[rt]dt_wa()
541  * @hw: pointer to the HW structure
542  * @tail: address of tail descriptor register
543  * @i: value to write to tail descriptor register
544  *
545  * When updating the tail register, the ME could be accessing Host CSR
546  * registers at the same time.  Normally, this is handled in h/w by an
547  * arbiter but on some parts there is a bug that acknowledges Host accesses
548  * later than it should which could result in the descriptor register to
549  * have an incorrect value.  Workaround this by checking the FWSM register
550  * which has bit 24 set while ME is accessing Host CSR registers, wait
551  * if it is set and try again a number of times.
552  **/
553 static inline s32 e1000e_update_tail_wa(struct e1000_hw *hw, void __iomem *tail,
554 					unsigned int i)
555 {
556 	unsigned int j = 0;
557 
558 	while ((j++ < E1000_ICH_FWSM_PCIM2PCI_COUNT) &&
559 	       (er32(FWSM) & E1000_ICH_FWSM_PCIM2PCI))
560 		udelay(50);
561 
562 	writel(i, tail);
563 
564 	if ((j == E1000_ICH_FWSM_PCIM2PCI_COUNT) && (i != readl(tail)))
565 		return E1000_ERR_SWFW_SYNC;
566 
567 	return 0;
568 }
569 
570 static void e1000e_update_rdt_wa(struct e1000_ring *rx_ring, unsigned int i)
571 {
572 	struct e1000_adapter *adapter = rx_ring->adapter;
573 	struct e1000_hw *hw = &adapter->hw;
574 
575 	if (e1000e_update_tail_wa(hw, rx_ring->tail, i)) {
576 		u32 rctl = er32(RCTL);
577 		ew32(RCTL, rctl & ~E1000_RCTL_EN);
578 		e_err("ME firmware caused invalid RDT - resetting\n");
579 		schedule_work(&adapter->reset_task);
580 	}
581 }
582 
583 static void e1000e_update_tdt_wa(struct e1000_ring *tx_ring, unsigned int i)
584 {
585 	struct e1000_adapter *adapter = tx_ring->adapter;
586 	struct e1000_hw *hw = &adapter->hw;
587 
588 	if (e1000e_update_tail_wa(hw, tx_ring->tail, i)) {
589 		u32 tctl = er32(TCTL);
590 		ew32(TCTL, tctl & ~E1000_TCTL_EN);
591 		e_err("ME firmware caused invalid TDT - resetting\n");
592 		schedule_work(&adapter->reset_task);
593 	}
594 }
595 
596 /**
597  * e1000_alloc_rx_buffers - Replace used receive buffers
598  * @rx_ring: Rx descriptor ring
599  **/
600 static void e1000_alloc_rx_buffers(struct e1000_ring *rx_ring,
601 				   int cleaned_count, gfp_t gfp)
602 {
603 	struct e1000_adapter *adapter = rx_ring->adapter;
604 	struct net_device *netdev = adapter->netdev;
605 	struct pci_dev *pdev = adapter->pdev;
606 	union e1000_rx_desc_extended *rx_desc;
607 	struct e1000_buffer *buffer_info;
608 	struct sk_buff *skb;
609 	unsigned int i;
610 	unsigned int bufsz = adapter->rx_buffer_len;
611 
612 	i = rx_ring->next_to_use;
613 	buffer_info = &rx_ring->buffer_info[i];
614 
615 	while (cleaned_count--) {
616 		skb = buffer_info->skb;
617 		if (skb) {
618 			skb_trim(skb, 0);
619 			goto map_skb;
620 		}
621 
622 		skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
623 		if (!skb) {
624 			/* Better luck next round */
625 			adapter->alloc_rx_buff_failed++;
626 			break;
627 		}
628 
629 		buffer_info->skb = skb;
630 map_skb:
631 		buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
632 						  adapter->rx_buffer_len,
633 						  DMA_FROM_DEVICE);
634 		if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
635 			dev_err(&pdev->dev, "Rx DMA map failed\n");
636 			adapter->rx_dma_failed++;
637 			break;
638 		}
639 
640 		rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
641 		rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
642 
643 		if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
644 			/*
645 			 * Force memory writes to complete before letting h/w
646 			 * know there are new descriptors to fetch.  (Only
647 			 * applicable for weak-ordered memory model archs,
648 			 * such as IA-64).
649 			 */
650 			wmb();
651 			if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
652 				e1000e_update_rdt_wa(rx_ring, i);
653 			else
654 				writel(i, rx_ring->tail);
655 		}
656 		i++;
657 		if (i == rx_ring->count)
658 			i = 0;
659 		buffer_info = &rx_ring->buffer_info[i];
660 	}
661 
662 	rx_ring->next_to_use = i;
663 }
664 
665 /**
666  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
667  * @rx_ring: Rx descriptor ring
668  **/
669 static void e1000_alloc_rx_buffers_ps(struct e1000_ring *rx_ring,
670 				      int cleaned_count, gfp_t gfp)
671 {
672 	struct e1000_adapter *adapter = rx_ring->adapter;
673 	struct net_device *netdev = adapter->netdev;
674 	struct pci_dev *pdev = adapter->pdev;
675 	union e1000_rx_desc_packet_split *rx_desc;
676 	struct e1000_buffer *buffer_info;
677 	struct e1000_ps_page *ps_page;
678 	struct sk_buff *skb;
679 	unsigned int i, j;
680 
681 	i = rx_ring->next_to_use;
682 	buffer_info = &rx_ring->buffer_info[i];
683 
684 	while (cleaned_count--) {
685 		rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
686 
687 		for (j = 0; j < PS_PAGE_BUFFERS; j++) {
688 			ps_page = &buffer_info->ps_pages[j];
689 			if (j >= adapter->rx_ps_pages) {
690 				/* all unused desc entries get hw null ptr */
691 				rx_desc->read.buffer_addr[j + 1] =
692 				    ~cpu_to_le64(0);
693 				continue;
694 			}
695 			if (!ps_page->page) {
696 				ps_page->page = alloc_page(gfp);
697 				if (!ps_page->page) {
698 					adapter->alloc_rx_buff_failed++;
699 					goto no_buffers;
700 				}
701 				ps_page->dma = dma_map_page(&pdev->dev,
702 							    ps_page->page,
703 							    0, PAGE_SIZE,
704 							    DMA_FROM_DEVICE);
705 				if (dma_mapping_error(&pdev->dev,
706 						      ps_page->dma)) {
707 					dev_err(&adapter->pdev->dev,
708 						"Rx DMA page map failed\n");
709 					adapter->rx_dma_failed++;
710 					goto no_buffers;
711 				}
712 			}
713 			/*
714 			 * Refresh the desc even if buffer_addrs
715 			 * didn't change because each write-back
716 			 * erases this info.
717 			 */
718 			rx_desc->read.buffer_addr[j + 1] =
719 			    cpu_to_le64(ps_page->dma);
720 		}
721 
722 		skb = __netdev_alloc_skb_ip_align(netdev,
723 						  adapter->rx_ps_bsize0,
724 						  gfp);
725 
726 		if (!skb) {
727 			adapter->alloc_rx_buff_failed++;
728 			break;
729 		}
730 
731 		buffer_info->skb = skb;
732 		buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
733 						  adapter->rx_ps_bsize0,
734 						  DMA_FROM_DEVICE);
735 		if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
736 			dev_err(&pdev->dev, "Rx DMA map failed\n");
737 			adapter->rx_dma_failed++;
738 			/* cleanup skb */
739 			dev_kfree_skb_any(skb);
740 			buffer_info->skb = NULL;
741 			break;
742 		}
743 
744 		rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
745 
746 		if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
747 			/*
748 			 * Force memory writes to complete before letting h/w
749 			 * know there are new descriptors to fetch.  (Only
750 			 * applicable for weak-ordered memory model archs,
751 			 * such as IA-64).
752 			 */
753 			wmb();
754 			if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
755 				e1000e_update_rdt_wa(rx_ring, i << 1);
756 			else
757 				writel(i << 1, rx_ring->tail);
758 		}
759 
760 		i++;
761 		if (i == rx_ring->count)
762 			i = 0;
763 		buffer_info = &rx_ring->buffer_info[i];
764 	}
765 
766 no_buffers:
767 	rx_ring->next_to_use = i;
768 }
769 
770 /**
771  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
772  * @rx_ring: Rx descriptor ring
773  * @cleaned_count: number of buffers to allocate this pass
774  **/
775 
776 static void e1000_alloc_jumbo_rx_buffers(struct e1000_ring *rx_ring,
777 					 int cleaned_count, gfp_t gfp)
778 {
779 	struct e1000_adapter *adapter = rx_ring->adapter;
780 	struct net_device *netdev = adapter->netdev;
781 	struct pci_dev *pdev = adapter->pdev;
782 	union e1000_rx_desc_extended *rx_desc;
783 	struct e1000_buffer *buffer_info;
784 	struct sk_buff *skb;
785 	unsigned int i;
786 	unsigned int bufsz = 256 - 16 /* for skb_reserve */;
787 
788 	i = rx_ring->next_to_use;
789 	buffer_info = &rx_ring->buffer_info[i];
790 
791 	while (cleaned_count--) {
792 		skb = buffer_info->skb;
793 		if (skb) {
794 			skb_trim(skb, 0);
795 			goto check_page;
796 		}
797 
798 		skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
799 		if (unlikely(!skb)) {
800 			/* Better luck next round */
801 			adapter->alloc_rx_buff_failed++;
802 			break;
803 		}
804 
805 		buffer_info->skb = skb;
806 check_page:
807 		/* allocate a new page if necessary */
808 		if (!buffer_info->page) {
809 			buffer_info->page = alloc_page(gfp);
810 			if (unlikely(!buffer_info->page)) {
811 				adapter->alloc_rx_buff_failed++;
812 				break;
813 			}
814 		}
815 
816 		if (!buffer_info->dma)
817 			buffer_info->dma = dma_map_page(&pdev->dev,
818 			                                buffer_info->page, 0,
819 			                                PAGE_SIZE,
820 							DMA_FROM_DEVICE);
821 
822 		rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
823 		rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
824 
825 		if (unlikely(++i == rx_ring->count))
826 			i = 0;
827 		buffer_info = &rx_ring->buffer_info[i];
828 	}
829 
830 	if (likely(rx_ring->next_to_use != i)) {
831 		rx_ring->next_to_use = i;
832 		if (unlikely(i-- == 0))
833 			i = (rx_ring->count - 1);
834 
835 		/* Force memory writes to complete before letting h/w
836 		 * know there are new descriptors to fetch.  (Only
837 		 * applicable for weak-ordered memory model archs,
838 		 * such as IA-64). */
839 		wmb();
840 		if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
841 			e1000e_update_rdt_wa(rx_ring, i);
842 		else
843 			writel(i, rx_ring->tail);
844 	}
845 }
846 
847 static inline void e1000_rx_hash(struct net_device *netdev, __le32 rss,
848 				 struct sk_buff *skb)
849 {
850 	if (netdev->features & NETIF_F_RXHASH)
851 		skb->rxhash = le32_to_cpu(rss);
852 }
853 
854 /**
855  * e1000_clean_rx_irq - Send received data up the network stack
856  * @rx_ring: Rx descriptor ring
857  *
858  * the return value indicates whether actual cleaning was done, there
859  * is no guarantee that everything was cleaned
860  **/
861 static bool e1000_clean_rx_irq(struct e1000_ring *rx_ring, int *work_done,
862 			       int work_to_do)
863 {
864 	struct e1000_adapter *adapter = rx_ring->adapter;
865 	struct net_device *netdev = adapter->netdev;
866 	struct pci_dev *pdev = adapter->pdev;
867 	struct e1000_hw *hw = &adapter->hw;
868 	union e1000_rx_desc_extended *rx_desc, *next_rxd;
869 	struct e1000_buffer *buffer_info, *next_buffer;
870 	u32 length, staterr;
871 	unsigned int i;
872 	int cleaned_count = 0;
873 	bool cleaned = false;
874 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
875 
876 	i = rx_ring->next_to_clean;
877 	rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
878 	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
879 	buffer_info = &rx_ring->buffer_info[i];
880 
881 	while (staterr & E1000_RXD_STAT_DD) {
882 		struct sk_buff *skb;
883 
884 		if (*work_done >= work_to_do)
885 			break;
886 		(*work_done)++;
887 		rmb();	/* read descriptor and rx_buffer_info after status DD */
888 
889 		skb = buffer_info->skb;
890 		buffer_info->skb = NULL;
891 
892 		prefetch(skb->data - NET_IP_ALIGN);
893 
894 		i++;
895 		if (i == rx_ring->count)
896 			i = 0;
897 		next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
898 		prefetch(next_rxd);
899 
900 		next_buffer = &rx_ring->buffer_info[i];
901 
902 		cleaned = true;
903 		cleaned_count++;
904 		dma_unmap_single(&pdev->dev,
905 				 buffer_info->dma,
906 				 adapter->rx_buffer_len,
907 				 DMA_FROM_DEVICE);
908 		buffer_info->dma = 0;
909 
910 		length = le16_to_cpu(rx_desc->wb.upper.length);
911 
912 		/*
913 		 * !EOP means multiple descriptors were used to store a single
914 		 * packet, if that's the case we need to toss it.  In fact, we
915 		 * need to toss every packet with the EOP bit clear and the
916 		 * next frame that _does_ have the EOP bit set, as it is by
917 		 * definition only a frame fragment
918 		 */
919 		if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
920 			adapter->flags2 |= FLAG2_IS_DISCARDING;
921 
922 		if (adapter->flags2 & FLAG2_IS_DISCARDING) {
923 			/* All receives must fit into a single buffer */
924 			e_dbg("Receive packet consumed multiple buffers\n");
925 			/* recycle */
926 			buffer_info->skb = skb;
927 			if (staterr & E1000_RXD_STAT_EOP)
928 				adapter->flags2 &= ~FLAG2_IS_DISCARDING;
929 			goto next_desc;
930 		}
931 
932 		if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
933 			/* recycle */
934 			buffer_info->skb = skb;
935 			goto next_desc;
936 		}
937 
938 		/* adjust length to remove Ethernet CRC */
939 		if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
940 			length -= 4;
941 
942 		total_rx_bytes += length;
943 		total_rx_packets++;
944 
945 		/*
946 		 * code added for copybreak, this should improve
947 		 * performance for small packets with large amounts
948 		 * of reassembly being done in the stack
949 		 */
950 		if (length < copybreak) {
951 			struct sk_buff *new_skb =
952 			    netdev_alloc_skb_ip_align(netdev, length);
953 			if (new_skb) {
954 				skb_copy_to_linear_data_offset(new_skb,
955 							       -NET_IP_ALIGN,
956 							       (skb->data -
957 								NET_IP_ALIGN),
958 							       (length +
959 								NET_IP_ALIGN));
960 				/* save the skb in buffer_info as good */
961 				buffer_info->skb = skb;
962 				skb = new_skb;
963 			}
964 			/* else just continue with the old one */
965 		}
966 		/* end copybreak code */
967 		skb_put(skb, length);
968 
969 		/* Receive Checksum Offload */
970 		e1000_rx_checksum(adapter, staterr,
971 				  rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
972 
973 		e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
974 
975 		e1000_receive_skb(adapter, netdev, skb, staterr,
976 				  rx_desc->wb.upper.vlan);
977 
978 next_desc:
979 		rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
980 
981 		/* return some buffers to hardware, one at a time is too slow */
982 		if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
983 			adapter->alloc_rx_buf(rx_ring, cleaned_count,
984 					      GFP_ATOMIC);
985 			cleaned_count = 0;
986 		}
987 
988 		/* use prefetched values */
989 		rx_desc = next_rxd;
990 		buffer_info = next_buffer;
991 
992 		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
993 	}
994 	rx_ring->next_to_clean = i;
995 
996 	cleaned_count = e1000_desc_unused(rx_ring);
997 	if (cleaned_count)
998 		adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
999 
1000 	adapter->total_rx_bytes += total_rx_bytes;
1001 	adapter->total_rx_packets += total_rx_packets;
1002 	return cleaned;
1003 }
1004 
1005 static void e1000_put_txbuf(struct e1000_ring *tx_ring,
1006 			    struct e1000_buffer *buffer_info)
1007 {
1008 	struct e1000_adapter *adapter = tx_ring->adapter;
1009 
1010 	if (buffer_info->dma) {
1011 		if (buffer_info->mapped_as_page)
1012 			dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1013 				       buffer_info->length, DMA_TO_DEVICE);
1014 		else
1015 			dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1016 					 buffer_info->length, DMA_TO_DEVICE);
1017 		buffer_info->dma = 0;
1018 	}
1019 	if (buffer_info->skb) {
1020 		dev_kfree_skb_any(buffer_info->skb);
1021 		buffer_info->skb = NULL;
1022 	}
1023 	buffer_info->time_stamp = 0;
1024 }
1025 
1026 static void e1000_print_hw_hang(struct work_struct *work)
1027 {
1028 	struct e1000_adapter *adapter = container_of(work,
1029 	                                             struct e1000_adapter,
1030 	                                             print_hang_task);
1031 	struct net_device *netdev = adapter->netdev;
1032 	struct e1000_ring *tx_ring = adapter->tx_ring;
1033 	unsigned int i = tx_ring->next_to_clean;
1034 	unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1035 	struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1036 	struct e1000_hw *hw = &adapter->hw;
1037 	u16 phy_status, phy_1000t_status, phy_ext_status;
1038 	u16 pci_status;
1039 
1040 	if (test_bit(__E1000_DOWN, &adapter->state))
1041 		return;
1042 
1043 	if (!adapter->tx_hang_recheck &&
1044 	    (adapter->flags2 & FLAG2_DMA_BURST)) {
1045 		/* May be block on write-back, flush and detect again
1046 		 * flush pending descriptor writebacks to memory
1047 		 */
1048 		ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1049 		/* execute the writes immediately */
1050 		e1e_flush();
1051 		adapter->tx_hang_recheck = true;
1052 		return;
1053 	}
1054 	/* Real hang detected */
1055 	adapter->tx_hang_recheck = false;
1056 	netif_stop_queue(netdev);
1057 
1058 	e1e_rphy(hw, PHY_STATUS, &phy_status);
1059 	e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
1060 	e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
1061 
1062 	pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1063 
1064 	/* detected Hardware unit hang */
1065 	e_err("Detected Hardware Unit Hang:\n"
1066 	      "  TDH                  <%x>\n"
1067 	      "  TDT                  <%x>\n"
1068 	      "  next_to_use          <%x>\n"
1069 	      "  next_to_clean        <%x>\n"
1070 	      "buffer_info[next_to_clean]:\n"
1071 	      "  time_stamp           <%lx>\n"
1072 	      "  next_to_watch        <%x>\n"
1073 	      "  jiffies              <%lx>\n"
1074 	      "  next_to_watch.status <%x>\n"
1075 	      "MAC Status             <%x>\n"
1076 	      "PHY Status             <%x>\n"
1077 	      "PHY 1000BASE-T Status  <%x>\n"
1078 	      "PHY Extended Status    <%x>\n"
1079 	      "PCI Status             <%x>\n",
1080 	      readl(tx_ring->head),
1081 	      readl(tx_ring->tail),
1082 	      tx_ring->next_to_use,
1083 	      tx_ring->next_to_clean,
1084 	      tx_ring->buffer_info[eop].time_stamp,
1085 	      eop,
1086 	      jiffies,
1087 	      eop_desc->upper.fields.status,
1088 	      er32(STATUS),
1089 	      phy_status,
1090 	      phy_1000t_status,
1091 	      phy_ext_status,
1092 	      pci_status);
1093 }
1094 
1095 /**
1096  * e1000_clean_tx_irq - Reclaim resources after transmit completes
1097  * @tx_ring: Tx descriptor ring
1098  *
1099  * the return value indicates whether actual cleaning was done, there
1100  * is no guarantee that everything was cleaned
1101  **/
1102 static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
1103 {
1104 	struct e1000_adapter *adapter = tx_ring->adapter;
1105 	struct net_device *netdev = adapter->netdev;
1106 	struct e1000_hw *hw = &adapter->hw;
1107 	struct e1000_tx_desc *tx_desc, *eop_desc;
1108 	struct e1000_buffer *buffer_info;
1109 	unsigned int i, eop;
1110 	unsigned int count = 0;
1111 	unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1112 	unsigned int bytes_compl = 0, pkts_compl = 0;
1113 
1114 	i = tx_ring->next_to_clean;
1115 	eop = tx_ring->buffer_info[i].next_to_watch;
1116 	eop_desc = E1000_TX_DESC(*tx_ring, eop);
1117 
1118 	while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1119 	       (count < tx_ring->count)) {
1120 		bool cleaned = false;
1121 		rmb(); /* read buffer_info after eop_desc */
1122 		for (; !cleaned; count++) {
1123 			tx_desc = E1000_TX_DESC(*tx_ring, i);
1124 			buffer_info = &tx_ring->buffer_info[i];
1125 			cleaned = (i == eop);
1126 
1127 			if (cleaned) {
1128 				total_tx_packets += buffer_info->segs;
1129 				total_tx_bytes += buffer_info->bytecount;
1130 				if (buffer_info->skb) {
1131 					bytes_compl += buffer_info->skb->len;
1132 					pkts_compl++;
1133 				}
1134 			}
1135 
1136 			e1000_put_txbuf(tx_ring, buffer_info);
1137 			tx_desc->upper.data = 0;
1138 
1139 			i++;
1140 			if (i == tx_ring->count)
1141 				i = 0;
1142 		}
1143 
1144 		if (i == tx_ring->next_to_use)
1145 			break;
1146 		eop = tx_ring->buffer_info[i].next_to_watch;
1147 		eop_desc = E1000_TX_DESC(*tx_ring, eop);
1148 	}
1149 
1150 	tx_ring->next_to_clean = i;
1151 
1152 	netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1153 
1154 #define TX_WAKE_THRESHOLD 32
1155 	if (count && netif_carrier_ok(netdev) &&
1156 	    e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1157 		/* Make sure that anybody stopping the queue after this
1158 		 * sees the new next_to_clean.
1159 		 */
1160 		smp_mb();
1161 
1162 		if (netif_queue_stopped(netdev) &&
1163 		    !(test_bit(__E1000_DOWN, &adapter->state))) {
1164 			netif_wake_queue(netdev);
1165 			++adapter->restart_queue;
1166 		}
1167 	}
1168 
1169 	if (adapter->detect_tx_hung) {
1170 		/*
1171 		 * Detect a transmit hang in hardware, this serializes the
1172 		 * check with the clearing of time_stamp and movement of i
1173 		 */
1174 		adapter->detect_tx_hung = false;
1175 		if (tx_ring->buffer_info[i].time_stamp &&
1176 		    time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1177 			       + (adapter->tx_timeout_factor * HZ)) &&
1178 		    !(er32(STATUS) & E1000_STATUS_TXOFF))
1179 			schedule_work(&adapter->print_hang_task);
1180 		else
1181 			adapter->tx_hang_recheck = false;
1182 	}
1183 	adapter->total_tx_bytes += total_tx_bytes;
1184 	adapter->total_tx_packets += total_tx_packets;
1185 	return count < tx_ring->count;
1186 }
1187 
1188 /**
1189  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1190  * @rx_ring: Rx descriptor ring
1191  *
1192  * the return value indicates whether actual cleaning was done, there
1193  * is no guarantee that everything was cleaned
1194  **/
1195 static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
1196 				  int work_to_do)
1197 {
1198 	struct e1000_adapter *adapter = rx_ring->adapter;
1199 	struct e1000_hw *hw = &adapter->hw;
1200 	union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1201 	struct net_device *netdev = adapter->netdev;
1202 	struct pci_dev *pdev = adapter->pdev;
1203 	struct e1000_buffer *buffer_info, *next_buffer;
1204 	struct e1000_ps_page *ps_page;
1205 	struct sk_buff *skb;
1206 	unsigned int i, j;
1207 	u32 length, staterr;
1208 	int cleaned_count = 0;
1209 	bool cleaned = false;
1210 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1211 
1212 	i = rx_ring->next_to_clean;
1213 	rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1214 	staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1215 	buffer_info = &rx_ring->buffer_info[i];
1216 
1217 	while (staterr & E1000_RXD_STAT_DD) {
1218 		if (*work_done >= work_to_do)
1219 			break;
1220 		(*work_done)++;
1221 		skb = buffer_info->skb;
1222 		rmb();	/* read descriptor and rx_buffer_info after status DD */
1223 
1224 		/* in the packet split case this is header only */
1225 		prefetch(skb->data - NET_IP_ALIGN);
1226 
1227 		i++;
1228 		if (i == rx_ring->count)
1229 			i = 0;
1230 		next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1231 		prefetch(next_rxd);
1232 
1233 		next_buffer = &rx_ring->buffer_info[i];
1234 
1235 		cleaned = true;
1236 		cleaned_count++;
1237 		dma_unmap_single(&pdev->dev, buffer_info->dma,
1238 				 adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1239 		buffer_info->dma = 0;
1240 
1241 		/* see !EOP comment in other Rx routine */
1242 		if (!(staterr & E1000_RXD_STAT_EOP))
1243 			adapter->flags2 |= FLAG2_IS_DISCARDING;
1244 
1245 		if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1246 			e_dbg("Packet Split buffers didn't pick up the full packet\n");
1247 			dev_kfree_skb_irq(skb);
1248 			if (staterr & E1000_RXD_STAT_EOP)
1249 				adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1250 			goto next_desc;
1251 		}
1252 
1253 		if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1254 			dev_kfree_skb_irq(skb);
1255 			goto next_desc;
1256 		}
1257 
1258 		length = le16_to_cpu(rx_desc->wb.middle.length0);
1259 
1260 		if (!length) {
1261 			e_dbg("Last part of the packet spanning multiple descriptors\n");
1262 			dev_kfree_skb_irq(skb);
1263 			goto next_desc;
1264 		}
1265 
1266 		/* Good Receive */
1267 		skb_put(skb, length);
1268 
1269 		{
1270 			/*
1271 			 * this looks ugly, but it seems compiler issues make
1272 			 * it more efficient than reusing j
1273 			 */
1274 			int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1275 
1276 			/*
1277 			 * page alloc/put takes too long and effects small
1278 			 * packet throughput, so unsplit small packets and
1279 			 * save the alloc/put only valid in softirq (napi)
1280 			 * context to call kmap_*
1281 			 */
1282 			if (l1 && (l1 <= copybreak) &&
1283 			    ((length + l1) <= adapter->rx_ps_bsize0)) {
1284 				u8 *vaddr;
1285 
1286 				ps_page = &buffer_info->ps_pages[0];
1287 
1288 				/*
1289 				 * there is no documentation about how to call
1290 				 * kmap_atomic, so we can't hold the mapping
1291 				 * very long
1292 				 */
1293 				dma_sync_single_for_cpu(&pdev->dev,
1294 							ps_page->dma,
1295 							PAGE_SIZE,
1296 							DMA_FROM_DEVICE);
1297 				vaddr = kmap_atomic(ps_page->page,
1298 						    KM_SKB_DATA_SOFTIRQ);
1299 				memcpy(skb_tail_pointer(skb), vaddr, l1);
1300 				kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1301 				dma_sync_single_for_device(&pdev->dev,
1302 							   ps_page->dma,
1303 							   PAGE_SIZE,
1304 							   DMA_FROM_DEVICE);
1305 
1306 				/* remove the CRC */
1307 				if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1308 					l1 -= 4;
1309 
1310 				skb_put(skb, l1);
1311 				goto copydone;
1312 			} /* if */
1313 		}
1314 
1315 		for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1316 			length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1317 			if (!length)
1318 				break;
1319 
1320 			ps_page = &buffer_info->ps_pages[j];
1321 			dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1322 				       DMA_FROM_DEVICE);
1323 			ps_page->dma = 0;
1324 			skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1325 			ps_page->page = NULL;
1326 			skb->len += length;
1327 			skb->data_len += length;
1328 			skb->truesize += PAGE_SIZE;
1329 		}
1330 
1331 		/* strip the ethernet crc, problem is we're using pages now so
1332 		 * this whole operation can get a little cpu intensive
1333 		 */
1334 		if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1335 			pskb_trim(skb, skb->len - 4);
1336 
1337 copydone:
1338 		total_rx_bytes += skb->len;
1339 		total_rx_packets++;
1340 
1341 		e1000_rx_checksum(adapter, staterr,
1342 				  rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
1343 
1344 		e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1345 
1346 		if (rx_desc->wb.upper.header_status &
1347 			   cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1348 			adapter->rx_hdr_split++;
1349 
1350 		e1000_receive_skb(adapter, netdev, skb,
1351 				  staterr, rx_desc->wb.middle.vlan);
1352 
1353 next_desc:
1354 		rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1355 		buffer_info->skb = NULL;
1356 
1357 		/* return some buffers to hardware, one at a time is too slow */
1358 		if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1359 			adapter->alloc_rx_buf(rx_ring, cleaned_count,
1360 					      GFP_ATOMIC);
1361 			cleaned_count = 0;
1362 		}
1363 
1364 		/* use prefetched values */
1365 		rx_desc = next_rxd;
1366 		buffer_info = next_buffer;
1367 
1368 		staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1369 	}
1370 	rx_ring->next_to_clean = i;
1371 
1372 	cleaned_count = e1000_desc_unused(rx_ring);
1373 	if (cleaned_count)
1374 		adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1375 
1376 	adapter->total_rx_bytes += total_rx_bytes;
1377 	adapter->total_rx_packets += total_rx_packets;
1378 	return cleaned;
1379 }
1380 
1381 /**
1382  * e1000_consume_page - helper function
1383  **/
1384 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1385                                u16 length)
1386 {
1387 	bi->page = NULL;
1388 	skb->len += length;
1389 	skb->data_len += length;
1390 	skb->truesize += PAGE_SIZE;
1391 }
1392 
1393 /**
1394  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1395  * @adapter: board private structure
1396  *
1397  * the return value indicates whether actual cleaning was done, there
1398  * is no guarantee that everything was cleaned
1399  **/
1400 static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1401 				     int work_to_do)
1402 {
1403 	struct e1000_adapter *adapter = rx_ring->adapter;
1404 	struct net_device *netdev = adapter->netdev;
1405 	struct pci_dev *pdev = adapter->pdev;
1406 	union e1000_rx_desc_extended *rx_desc, *next_rxd;
1407 	struct e1000_buffer *buffer_info, *next_buffer;
1408 	u32 length, staterr;
1409 	unsigned int i;
1410 	int cleaned_count = 0;
1411 	bool cleaned = false;
1412 	unsigned int total_rx_bytes=0, total_rx_packets=0;
1413 
1414 	i = rx_ring->next_to_clean;
1415 	rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1416 	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1417 	buffer_info = &rx_ring->buffer_info[i];
1418 
1419 	while (staterr & E1000_RXD_STAT_DD) {
1420 		struct sk_buff *skb;
1421 
1422 		if (*work_done >= work_to_do)
1423 			break;
1424 		(*work_done)++;
1425 		rmb();	/* read descriptor and rx_buffer_info after status DD */
1426 
1427 		skb = buffer_info->skb;
1428 		buffer_info->skb = NULL;
1429 
1430 		++i;
1431 		if (i == rx_ring->count)
1432 			i = 0;
1433 		next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1434 		prefetch(next_rxd);
1435 
1436 		next_buffer = &rx_ring->buffer_info[i];
1437 
1438 		cleaned = true;
1439 		cleaned_count++;
1440 		dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1441 			       DMA_FROM_DEVICE);
1442 		buffer_info->dma = 0;
1443 
1444 		length = le16_to_cpu(rx_desc->wb.upper.length);
1445 
1446 		/* errors is only valid for DD + EOP descriptors */
1447 		if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1448 			     (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK))) {
1449 			/* recycle both page and skb */
1450 			buffer_info->skb = skb;
1451 			/* an error means any chain goes out the window too */
1452 			if (rx_ring->rx_skb_top)
1453 				dev_kfree_skb_irq(rx_ring->rx_skb_top);
1454 			rx_ring->rx_skb_top = NULL;
1455 			goto next_desc;
1456 		}
1457 
1458 #define rxtop (rx_ring->rx_skb_top)
1459 		if (!(staterr & E1000_RXD_STAT_EOP)) {
1460 			/* this descriptor is only the beginning (or middle) */
1461 			if (!rxtop) {
1462 				/* this is the beginning of a chain */
1463 				rxtop = skb;
1464 				skb_fill_page_desc(rxtop, 0, buffer_info->page,
1465 				                   0, length);
1466 			} else {
1467 				/* this is the middle of a chain */
1468 				skb_fill_page_desc(rxtop,
1469 				    skb_shinfo(rxtop)->nr_frags,
1470 				    buffer_info->page, 0, length);
1471 				/* re-use the skb, only consumed the page */
1472 				buffer_info->skb = skb;
1473 			}
1474 			e1000_consume_page(buffer_info, rxtop, length);
1475 			goto next_desc;
1476 		} else {
1477 			if (rxtop) {
1478 				/* end of the chain */
1479 				skb_fill_page_desc(rxtop,
1480 				    skb_shinfo(rxtop)->nr_frags,
1481 				    buffer_info->page, 0, length);
1482 				/* re-use the current skb, we only consumed the
1483 				 * page */
1484 				buffer_info->skb = skb;
1485 				skb = rxtop;
1486 				rxtop = NULL;
1487 				e1000_consume_page(buffer_info, skb, length);
1488 			} else {
1489 				/* no chain, got EOP, this buf is the packet
1490 				 * copybreak to save the put_page/alloc_page */
1491 				if (length <= copybreak &&
1492 				    skb_tailroom(skb) >= length) {
1493 					u8 *vaddr;
1494 					vaddr = kmap_atomic(buffer_info->page,
1495 					                   KM_SKB_DATA_SOFTIRQ);
1496 					memcpy(skb_tail_pointer(skb), vaddr,
1497 					       length);
1498 					kunmap_atomic(vaddr,
1499 					              KM_SKB_DATA_SOFTIRQ);
1500 					/* re-use the page, so don't erase
1501 					 * buffer_info->page */
1502 					skb_put(skb, length);
1503 				} else {
1504 					skb_fill_page_desc(skb, 0,
1505 					                   buffer_info->page, 0,
1506 				                           length);
1507 					e1000_consume_page(buffer_info, skb,
1508 					                   length);
1509 				}
1510 			}
1511 		}
1512 
1513 		/* Receive Checksum Offload XXX recompute due to CRC strip? */
1514 		e1000_rx_checksum(adapter, staterr,
1515 				  rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
1516 
1517 		e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1518 
1519 		/* probably a little skewed due to removing CRC */
1520 		total_rx_bytes += skb->len;
1521 		total_rx_packets++;
1522 
1523 		/* eth type trans needs skb->data to point to something */
1524 		if (!pskb_may_pull(skb, ETH_HLEN)) {
1525 			e_err("pskb_may_pull failed.\n");
1526 			dev_kfree_skb_irq(skb);
1527 			goto next_desc;
1528 		}
1529 
1530 		e1000_receive_skb(adapter, netdev, skb, staterr,
1531 				  rx_desc->wb.upper.vlan);
1532 
1533 next_desc:
1534 		rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1535 
1536 		/* return some buffers to hardware, one at a time is too slow */
1537 		if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1538 			adapter->alloc_rx_buf(rx_ring, cleaned_count,
1539 					      GFP_ATOMIC);
1540 			cleaned_count = 0;
1541 		}
1542 
1543 		/* use prefetched values */
1544 		rx_desc = next_rxd;
1545 		buffer_info = next_buffer;
1546 
1547 		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1548 	}
1549 	rx_ring->next_to_clean = i;
1550 
1551 	cleaned_count = e1000_desc_unused(rx_ring);
1552 	if (cleaned_count)
1553 		adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1554 
1555 	adapter->total_rx_bytes += total_rx_bytes;
1556 	adapter->total_rx_packets += total_rx_packets;
1557 	return cleaned;
1558 }
1559 
1560 /**
1561  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1562  * @rx_ring: Rx descriptor ring
1563  **/
1564 static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1565 {
1566 	struct e1000_adapter *adapter = rx_ring->adapter;
1567 	struct e1000_buffer *buffer_info;
1568 	struct e1000_ps_page *ps_page;
1569 	struct pci_dev *pdev = adapter->pdev;
1570 	unsigned int i, j;
1571 
1572 	/* Free all the Rx ring sk_buffs */
1573 	for (i = 0; i < rx_ring->count; i++) {
1574 		buffer_info = &rx_ring->buffer_info[i];
1575 		if (buffer_info->dma) {
1576 			if (adapter->clean_rx == e1000_clean_rx_irq)
1577 				dma_unmap_single(&pdev->dev, buffer_info->dma,
1578 						 adapter->rx_buffer_len,
1579 						 DMA_FROM_DEVICE);
1580 			else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1581 				dma_unmap_page(&pdev->dev, buffer_info->dma,
1582 				               PAGE_SIZE,
1583 					       DMA_FROM_DEVICE);
1584 			else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1585 				dma_unmap_single(&pdev->dev, buffer_info->dma,
1586 						 adapter->rx_ps_bsize0,
1587 						 DMA_FROM_DEVICE);
1588 			buffer_info->dma = 0;
1589 		}
1590 
1591 		if (buffer_info->page) {
1592 			put_page(buffer_info->page);
1593 			buffer_info->page = NULL;
1594 		}
1595 
1596 		if (buffer_info->skb) {
1597 			dev_kfree_skb(buffer_info->skb);
1598 			buffer_info->skb = NULL;
1599 		}
1600 
1601 		for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1602 			ps_page = &buffer_info->ps_pages[j];
1603 			if (!ps_page->page)
1604 				break;
1605 			dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1606 				       DMA_FROM_DEVICE);
1607 			ps_page->dma = 0;
1608 			put_page(ps_page->page);
1609 			ps_page->page = NULL;
1610 		}
1611 	}
1612 
1613 	/* there also may be some cached data from a chained receive */
1614 	if (rx_ring->rx_skb_top) {
1615 		dev_kfree_skb(rx_ring->rx_skb_top);
1616 		rx_ring->rx_skb_top = NULL;
1617 	}
1618 
1619 	/* Zero out the descriptor ring */
1620 	memset(rx_ring->desc, 0, rx_ring->size);
1621 
1622 	rx_ring->next_to_clean = 0;
1623 	rx_ring->next_to_use = 0;
1624 	adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1625 
1626 	writel(0, rx_ring->head);
1627 	writel(0, rx_ring->tail);
1628 }
1629 
1630 static void e1000e_downshift_workaround(struct work_struct *work)
1631 {
1632 	struct e1000_adapter *adapter = container_of(work,
1633 					struct e1000_adapter, downshift_task);
1634 
1635 	if (test_bit(__E1000_DOWN, &adapter->state))
1636 		return;
1637 
1638 	e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1639 }
1640 
1641 /**
1642  * e1000_intr_msi - Interrupt Handler
1643  * @irq: interrupt number
1644  * @data: pointer to a network interface device structure
1645  **/
1646 static irqreturn_t e1000_intr_msi(int irq, void *data)
1647 {
1648 	struct net_device *netdev = data;
1649 	struct e1000_adapter *adapter = netdev_priv(netdev);
1650 	struct e1000_hw *hw = &adapter->hw;
1651 	u32 icr = er32(ICR);
1652 
1653 	/*
1654 	 * read ICR disables interrupts using IAM
1655 	 */
1656 
1657 	if (icr & E1000_ICR_LSC) {
1658 		hw->mac.get_link_status = 1;
1659 		/*
1660 		 * ICH8 workaround-- Call gig speed drop workaround on cable
1661 		 * disconnect (LSC) before accessing any PHY registers
1662 		 */
1663 		if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1664 		    (!(er32(STATUS) & E1000_STATUS_LU)))
1665 			schedule_work(&adapter->downshift_task);
1666 
1667 		/*
1668 		 * 80003ES2LAN workaround-- For packet buffer work-around on
1669 		 * link down event; disable receives here in the ISR and reset
1670 		 * adapter in watchdog
1671 		 */
1672 		if (netif_carrier_ok(netdev) &&
1673 		    adapter->flags & FLAG_RX_NEEDS_RESTART) {
1674 			/* disable receives */
1675 			u32 rctl = er32(RCTL);
1676 			ew32(RCTL, rctl & ~E1000_RCTL_EN);
1677 			adapter->flags |= FLAG_RX_RESTART_NOW;
1678 		}
1679 		/* guard against interrupt when we're going down */
1680 		if (!test_bit(__E1000_DOWN, &adapter->state))
1681 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
1682 	}
1683 
1684 	if (napi_schedule_prep(&adapter->napi)) {
1685 		adapter->total_tx_bytes = 0;
1686 		adapter->total_tx_packets = 0;
1687 		adapter->total_rx_bytes = 0;
1688 		adapter->total_rx_packets = 0;
1689 		__napi_schedule(&adapter->napi);
1690 	}
1691 
1692 	return IRQ_HANDLED;
1693 }
1694 
1695 /**
1696  * e1000_intr - Interrupt Handler
1697  * @irq: interrupt number
1698  * @data: pointer to a network interface device structure
1699  **/
1700 static irqreturn_t e1000_intr(int irq, void *data)
1701 {
1702 	struct net_device *netdev = data;
1703 	struct e1000_adapter *adapter = netdev_priv(netdev);
1704 	struct e1000_hw *hw = &adapter->hw;
1705 	u32 rctl, icr = er32(ICR);
1706 
1707 	if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1708 		return IRQ_NONE;  /* Not our interrupt */
1709 
1710 	/*
1711 	 * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1712 	 * not set, then the adapter didn't send an interrupt
1713 	 */
1714 	if (!(icr & E1000_ICR_INT_ASSERTED))
1715 		return IRQ_NONE;
1716 
1717 	/*
1718 	 * Interrupt Auto-Mask...upon reading ICR,
1719 	 * interrupts are masked.  No need for the
1720 	 * IMC write
1721 	 */
1722 
1723 	if (icr & E1000_ICR_LSC) {
1724 		hw->mac.get_link_status = 1;
1725 		/*
1726 		 * ICH8 workaround-- Call gig speed drop workaround on cable
1727 		 * disconnect (LSC) before accessing any PHY registers
1728 		 */
1729 		if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1730 		    (!(er32(STATUS) & E1000_STATUS_LU)))
1731 			schedule_work(&adapter->downshift_task);
1732 
1733 		/*
1734 		 * 80003ES2LAN workaround--
1735 		 * For packet buffer work-around on link down event;
1736 		 * disable receives here in the ISR and
1737 		 * reset adapter in watchdog
1738 		 */
1739 		if (netif_carrier_ok(netdev) &&
1740 		    (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1741 			/* disable receives */
1742 			rctl = er32(RCTL);
1743 			ew32(RCTL, rctl & ~E1000_RCTL_EN);
1744 			adapter->flags |= FLAG_RX_RESTART_NOW;
1745 		}
1746 		/* guard against interrupt when we're going down */
1747 		if (!test_bit(__E1000_DOWN, &adapter->state))
1748 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
1749 	}
1750 
1751 	if (napi_schedule_prep(&adapter->napi)) {
1752 		adapter->total_tx_bytes = 0;
1753 		adapter->total_tx_packets = 0;
1754 		adapter->total_rx_bytes = 0;
1755 		adapter->total_rx_packets = 0;
1756 		__napi_schedule(&adapter->napi);
1757 	}
1758 
1759 	return IRQ_HANDLED;
1760 }
1761 
1762 static irqreturn_t e1000_msix_other(int irq, void *data)
1763 {
1764 	struct net_device *netdev = data;
1765 	struct e1000_adapter *adapter = netdev_priv(netdev);
1766 	struct e1000_hw *hw = &adapter->hw;
1767 	u32 icr = er32(ICR);
1768 
1769 	if (!(icr & E1000_ICR_INT_ASSERTED)) {
1770 		if (!test_bit(__E1000_DOWN, &adapter->state))
1771 			ew32(IMS, E1000_IMS_OTHER);
1772 		return IRQ_NONE;
1773 	}
1774 
1775 	if (icr & adapter->eiac_mask)
1776 		ew32(ICS, (icr & adapter->eiac_mask));
1777 
1778 	if (icr & E1000_ICR_OTHER) {
1779 		if (!(icr & E1000_ICR_LSC))
1780 			goto no_link_interrupt;
1781 		hw->mac.get_link_status = 1;
1782 		/* guard against interrupt when we're going down */
1783 		if (!test_bit(__E1000_DOWN, &adapter->state))
1784 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
1785 	}
1786 
1787 no_link_interrupt:
1788 	if (!test_bit(__E1000_DOWN, &adapter->state))
1789 		ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1790 
1791 	return IRQ_HANDLED;
1792 }
1793 
1794 
1795 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1796 {
1797 	struct net_device *netdev = data;
1798 	struct e1000_adapter *adapter = netdev_priv(netdev);
1799 	struct e1000_hw *hw = &adapter->hw;
1800 	struct e1000_ring *tx_ring = adapter->tx_ring;
1801 
1802 
1803 	adapter->total_tx_bytes = 0;
1804 	adapter->total_tx_packets = 0;
1805 
1806 	if (!e1000_clean_tx_irq(tx_ring))
1807 		/* Ring was not completely cleaned, so fire another interrupt */
1808 		ew32(ICS, tx_ring->ims_val);
1809 
1810 	return IRQ_HANDLED;
1811 }
1812 
1813 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1814 {
1815 	struct net_device *netdev = data;
1816 	struct e1000_adapter *adapter = netdev_priv(netdev);
1817 	struct e1000_ring *rx_ring = adapter->rx_ring;
1818 
1819 	/* Write the ITR value calculated at the end of the
1820 	 * previous interrupt.
1821 	 */
1822 	if (rx_ring->set_itr) {
1823 		writel(1000000000 / (rx_ring->itr_val * 256),
1824 		       rx_ring->itr_register);
1825 		rx_ring->set_itr = 0;
1826 	}
1827 
1828 	if (napi_schedule_prep(&adapter->napi)) {
1829 		adapter->total_rx_bytes = 0;
1830 		adapter->total_rx_packets = 0;
1831 		__napi_schedule(&adapter->napi);
1832 	}
1833 	return IRQ_HANDLED;
1834 }
1835 
1836 /**
1837  * e1000_configure_msix - Configure MSI-X hardware
1838  *
1839  * e1000_configure_msix sets up the hardware to properly
1840  * generate MSI-X interrupts.
1841  **/
1842 static void e1000_configure_msix(struct e1000_adapter *adapter)
1843 {
1844 	struct e1000_hw *hw = &adapter->hw;
1845 	struct e1000_ring *rx_ring = adapter->rx_ring;
1846 	struct e1000_ring *tx_ring = adapter->tx_ring;
1847 	int vector = 0;
1848 	u32 ctrl_ext, ivar = 0;
1849 
1850 	adapter->eiac_mask = 0;
1851 
1852 	/* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1853 	if (hw->mac.type == e1000_82574) {
1854 		u32 rfctl = er32(RFCTL);
1855 		rfctl |= E1000_RFCTL_ACK_DIS;
1856 		ew32(RFCTL, rfctl);
1857 	}
1858 
1859 #define E1000_IVAR_INT_ALLOC_VALID	0x8
1860 	/* Configure Rx vector */
1861 	rx_ring->ims_val = E1000_IMS_RXQ0;
1862 	adapter->eiac_mask |= rx_ring->ims_val;
1863 	if (rx_ring->itr_val)
1864 		writel(1000000000 / (rx_ring->itr_val * 256),
1865 		       rx_ring->itr_register);
1866 	else
1867 		writel(1, rx_ring->itr_register);
1868 	ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1869 
1870 	/* Configure Tx vector */
1871 	tx_ring->ims_val = E1000_IMS_TXQ0;
1872 	vector++;
1873 	if (tx_ring->itr_val)
1874 		writel(1000000000 / (tx_ring->itr_val * 256),
1875 		       tx_ring->itr_register);
1876 	else
1877 		writel(1, tx_ring->itr_register);
1878 	adapter->eiac_mask |= tx_ring->ims_val;
1879 	ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1880 
1881 	/* set vector for Other Causes, e.g. link changes */
1882 	vector++;
1883 	ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1884 	if (rx_ring->itr_val)
1885 		writel(1000000000 / (rx_ring->itr_val * 256),
1886 		       hw->hw_addr + E1000_EITR_82574(vector));
1887 	else
1888 		writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1889 
1890 	/* Cause Tx interrupts on every write back */
1891 	ivar |= (1 << 31);
1892 
1893 	ew32(IVAR, ivar);
1894 
1895 	/* enable MSI-X PBA support */
1896 	ctrl_ext = er32(CTRL_EXT);
1897 	ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1898 
1899 	/* Auto-Mask Other interrupts upon ICR read */
1900 #define E1000_EIAC_MASK_82574   0x01F00000
1901 	ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1902 	ctrl_ext |= E1000_CTRL_EXT_EIAME;
1903 	ew32(CTRL_EXT, ctrl_ext);
1904 	e1e_flush();
1905 }
1906 
1907 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1908 {
1909 	if (adapter->msix_entries) {
1910 		pci_disable_msix(adapter->pdev);
1911 		kfree(adapter->msix_entries);
1912 		adapter->msix_entries = NULL;
1913 	} else if (adapter->flags & FLAG_MSI_ENABLED) {
1914 		pci_disable_msi(adapter->pdev);
1915 		adapter->flags &= ~FLAG_MSI_ENABLED;
1916 	}
1917 }
1918 
1919 /**
1920  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1921  *
1922  * Attempt to configure interrupts using the best available
1923  * capabilities of the hardware and kernel.
1924  **/
1925 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1926 {
1927 	int err;
1928 	int i;
1929 
1930 	switch (adapter->int_mode) {
1931 	case E1000E_INT_MODE_MSIX:
1932 		if (adapter->flags & FLAG_HAS_MSIX) {
1933 			adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1934 			adapter->msix_entries = kcalloc(adapter->num_vectors,
1935 						      sizeof(struct msix_entry),
1936 						      GFP_KERNEL);
1937 			if (adapter->msix_entries) {
1938 				for (i = 0; i < adapter->num_vectors; i++)
1939 					adapter->msix_entries[i].entry = i;
1940 
1941 				err = pci_enable_msix(adapter->pdev,
1942 						      adapter->msix_entries,
1943 						      adapter->num_vectors);
1944 				if (err == 0)
1945 					return;
1946 			}
1947 			/* MSI-X failed, so fall through and try MSI */
1948 			e_err("Failed to initialize MSI-X interrupts.  Falling back to MSI interrupts.\n");
1949 			e1000e_reset_interrupt_capability(adapter);
1950 		}
1951 		adapter->int_mode = E1000E_INT_MODE_MSI;
1952 		/* Fall through */
1953 	case E1000E_INT_MODE_MSI:
1954 		if (!pci_enable_msi(adapter->pdev)) {
1955 			adapter->flags |= FLAG_MSI_ENABLED;
1956 		} else {
1957 			adapter->int_mode = E1000E_INT_MODE_LEGACY;
1958 			e_err("Failed to initialize MSI interrupts.  Falling back to legacy interrupts.\n");
1959 		}
1960 		/* Fall through */
1961 	case E1000E_INT_MODE_LEGACY:
1962 		/* Don't do anything; this is the system default */
1963 		break;
1964 	}
1965 
1966 	/* store the number of vectors being used */
1967 	adapter->num_vectors = 1;
1968 }
1969 
1970 /**
1971  * e1000_request_msix - Initialize MSI-X interrupts
1972  *
1973  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1974  * kernel.
1975  **/
1976 static int e1000_request_msix(struct e1000_adapter *adapter)
1977 {
1978 	struct net_device *netdev = adapter->netdev;
1979 	int err = 0, vector = 0;
1980 
1981 	if (strlen(netdev->name) < (IFNAMSIZ - 5))
1982 		snprintf(adapter->rx_ring->name,
1983 			 sizeof(adapter->rx_ring->name) - 1,
1984 			 "%s-rx-0", netdev->name);
1985 	else
1986 		memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1987 	err = request_irq(adapter->msix_entries[vector].vector,
1988 			  e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1989 			  netdev);
1990 	if (err)
1991 		goto out;
1992 	adapter->rx_ring->itr_register = adapter->hw.hw_addr +
1993 	    E1000_EITR_82574(vector);
1994 	adapter->rx_ring->itr_val = adapter->itr;
1995 	vector++;
1996 
1997 	if (strlen(netdev->name) < (IFNAMSIZ - 5))
1998 		snprintf(adapter->tx_ring->name,
1999 			 sizeof(adapter->tx_ring->name) - 1,
2000 			 "%s-tx-0", netdev->name);
2001 	else
2002 		memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
2003 	err = request_irq(adapter->msix_entries[vector].vector,
2004 			  e1000_intr_msix_tx, 0, adapter->tx_ring->name,
2005 			  netdev);
2006 	if (err)
2007 		goto out;
2008 	adapter->tx_ring->itr_register = adapter->hw.hw_addr +
2009 	    E1000_EITR_82574(vector);
2010 	adapter->tx_ring->itr_val = adapter->itr;
2011 	vector++;
2012 
2013 	err = request_irq(adapter->msix_entries[vector].vector,
2014 			  e1000_msix_other, 0, netdev->name, netdev);
2015 	if (err)
2016 		goto out;
2017 
2018 	e1000_configure_msix(adapter);
2019 	return 0;
2020 out:
2021 	return err;
2022 }
2023 
2024 /**
2025  * e1000_request_irq - initialize interrupts
2026  *
2027  * Attempts to configure interrupts using the best available
2028  * capabilities of the hardware and kernel.
2029  **/
2030 static int e1000_request_irq(struct e1000_adapter *adapter)
2031 {
2032 	struct net_device *netdev = adapter->netdev;
2033 	int err;
2034 
2035 	if (adapter->msix_entries) {
2036 		err = e1000_request_msix(adapter);
2037 		if (!err)
2038 			return err;
2039 		/* fall back to MSI */
2040 		e1000e_reset_interrupt_capability(adapter);
2041 		adapter->int_mode = E1000E_INT_MODE_MSI;
2042 		e1000e_set_interrupt_capability(adapter);
2043 	}
2044 	if (adapter->flags & FLAG_MSI_ENABLED) {
2045 		err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2046 				  netdev->name, netdev);
2047 		if (!err)
2048 			return err;
2049 
2050 		/* fall back to legacy interrupt */
2051 		e1000e_reset_interrupt_capability(adapter);
2052 		adapter->int_mode = E1000E_INT_MODE_LEGACY;
2053 	}
2054 
2055 	err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2056 			  netdev->name, netdev);
2057 	if (err)
2058 		e_err("Unable to allocate interrupt, Error: %d\n", err);
2059 
2060 	return err;
2061 }
2062 
2063 static void e1000_free_irq(struct e1000_adapter *adapter)
2064 {
2065 	struct net_device *netdev = adapter->netdev;
2066 
2067 	if (adapter->msix_entries) {
2068 		int vector = 0;
2069 
2070 		free_irq(adapter->msix_entries[vector].vector, netdev);
2071 		vector++;
2072 
2073 		free_irq(adapter->msix_entries[vector].vector, netdev);
2074 		vector++;
2075 
2076 		/* Other Causes interrupt vector */
2077 		free_irq(adapter->msix_entries[vector].vector, netdev);
2078 		return;
2079 	}
2080 
2081 	free_irq(adapter->pdev->irq, netdev);
2082 }
2083 
2084 /**
2085  * e1000_irq_disable - Mask off interrupt generation on the NIC
2086  **/
2087 static void e1000_irq_disable(struct e1000_adapter *adapter)
2088 {
2089 	struct e1000_hw *hw = &adapter->hw;
2090 
2091 	ew32(IMC, ~0);
2092 	if (adapter->msix_entries)
2093 		ew32(EIAC_82574, 0);
2094 	e1e_flush();
2095 
2096 	if (adapter->msix_entries) {
2097 		int i;
2098 		for (i = 0; i < adapter->num_vectors; i++)
2099 			synchronize_irq(adapter->msix_entries[i].vector);
2100 	} else {
2101 		synchronize_irq(adapter->pdev->irq);
2102 	}
2103 }
2104 
2105 /**
2106  * e1000_irq_enable - Enable default interrupt generation settings
2107  **/
2108 static void e1000_irq_enable(struct e1000_adapter *adapter)
2109 {
2110 	struct e1000_hw *hw = &adapter->hw;
2111 
2112 	if (adapter->msix_entries) {
2113 		ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2114 		ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
2115 	} else {
2116 		ew32(IMS, IMS_ENABLE_MASK);
2117 	}
2118 	e1e_flush();
2119 }
2120 
2121 /**
2122  * e1000e_get_hw_control - get control of the h/w from f/w
2123  * @adapter: address of board private structure
2124  *
2125  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2126  * For ASF and Pass Through versions of f/w this means that
2127  * the driver is loaded. For AMT version (only with 82573)
2128  * of the f/w this means that the network i/f is open.
2129  **/
2130 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2131 {
2132 	struct e1000_hw *hw = &adapter->hw;
2133 	u32 ctrl_ext;
2134 	u32 swsm;
2135 
2136 	/* Let firmware know the driver has taken over */
2137 	if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2138 		swsm = er32(SWSM);
2139 		ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2140 	} else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2141 		ctrl_ext = er32(CTRL_EXT);
2142 		ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2143 	}
2144 }
2145 
2146 /**
2147  * e1000e_release_hw_control - release control of the h/w to f/w
2148  * @adapter: address of board private structure
2149  *
2150  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2151  * For ASF and Pass Through versions of f/w this means that the
2152  * driver is no longer loaded. For AMT version (only with 82573) i
2153  * of the f/w this means that the network i/f is closed.
2154  *
2155  **/
2156 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2157 {
2158 	struct e1000_hw *hw = &adapter->hw;
2159 	u32 ctrl_ext;
2160 	u32 swsm;
2161 
2162 	/* Let firmware taken over control of h/w */
2163 	if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2164 		swsm = er32(SWSM);
2165 		ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2166 	} else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2167 		ctrl_ext = er32(CTRL_EXT);
2168 		ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2169 	}
2170 }
2171 
2172 /**
2173  * @e1000_alloc_ring - allocate memory for a ring structure
2174  **/
2175 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2176 				struct e1000_ring *ring)
2177 {
2178 	struct pci_dev *pdev = adapter->pdev;
2179 
2180 	ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2181 					GFP_KERNEL);
2182 	if (!ring->desc)
2183 		return -ENOMEM;
2184 
2185 	return 0;
2186 }
2187 
2188 /**
2189  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2190  * @tx_ring: Tx descriptor ring
2191  *
2192  * Return 0 on success, negative on failure
2193  **/
2194 int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2195 {
2196 	struct e1000_adapter *adapter = tx_ring->adapter;
2197 	int err = -ENOMEM, size;
2198 
2199 	size = sizeof(struct e1000_buffer) * tx_ring->count;
2200 	tx_ring->buffer_info = vzalloc(size);
2201 	if (!tx_ring->buffer_info)
2202 		goto err;
2203 
2204 	/* round up to nearest 4K */
2205 	tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2206 	tx_ring->size = ALIGN(tx_ring->size, 4096);
2207 
2208 	err = e1000_alloc_ring_dma(adapter, tx_ring);
2209 	if (err)
2210 		goto err;
2211 
2212 	tx_ring->next_to_use = 0;
2213 	tx_ring->next_to_clean = 0;
2214 
2215 	return 0;
2216 err:
2217 	vfree(tx_ring->buffer_info);
2218 	e_err("Unable to allocate memory for the transmit descriptor ring\n");
2219 	return err;
2220 }
2221 
2222 /**
2223  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2224  * @rx_ring: Rx descriptor ring
2225  *
2226  * Returns 0 on success, negative on failure
2227  **/
2228 int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2229 {
2230 	struct e1000_adapter *adapter = rx_ring->adapter;
2231 	struct e1000_buffer *buffer_info;
2232 	int i, size, desc_len, err = -ENOMEM;
2233 
2234 	size = sizeof(struct e1000_buffer) * rx_ring->count;
2235 	rx_ring->buffer_info = vzalloc(size);
2236 	if (!rx_ring->buffer_info)
2237 		goto err;
2238 
2239 	for (i = 0; i < rx_ring->count; i++) {
2240 		buffer_info = &rx_ring->buffer_info[i];
2241 		buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2242 						sizeof(struct e1000_ps_page),
2243 						GFP_KERNEL);
2244 		if (!buffer_info->ps_pages)
2245 			goto err_pages;
2246 	}
2247 
2248 	desc_len = sizeof(union e1000_rx_desc_packet_split);
2249 
2250 	/* Round up to nearest 4K */
2251 	rx_ring->size = rx_ring->count * desc_len;
2252 	rx_ring->size = ALIGN(rx_ring->size, 4096);
2253 
2254 	err = e1000_alloc_ring_dma(adapter, rx_ring);
2255 	if (err)
2256 		goto err_pages;
2257 
2258 	rx_ring->next_to_clean = 0;
2259 	rx_ring->next_to_use = 0;
2260 	rx_ring->rx_skb_top = NULL;
2261 
2262 	return 0;
2263 
2264 err_pages:
2265 	for (i = 0; i < rx_ring->count; i++) {
2266 		buffer_info = &rx_ring->buffer_info[i];
2267 		kfree(buffer_info->ps_pages);
2268 	}
2269 err:
2270 	vfree(rx_ring->buffer_info);
2271 	e_err("Unable to allocate memory for the receive descriptor ring\n");
2272 	return err;
2273 }
2274 
2275 /**
2276  * e1000_clean_tx_ring - Free Tx Buffers
2277  * @tx_ring: Tx descriptor ring
2278  **/
2279 static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2280 {
2281 	struct e1000_adapter *adapter = tx_ring->adapter;
2282 	struct e1000_buffer *buffer_info;
2283 	unsigned long size;
2284 	unsigned int i;
2285 
2286 	for (i = 0; i < tx_ring->count; i++) {
2287 		buffer_info = &tx_ring->buffer_info[i];
2288 		e1000_put_txbuf(tx_ring, buffer_info);
2289 	}
2290 
2291 	netdev_reset_queue(adapter->netdev);
2292 	size = sizeof(struct e1000_buffer) * tx_ring->count;
2293 	memset(tx_ring->buffer_info, 0, size);
2294 
2295 	memset(tx_ring->desc, 0, tx_ring->size);
2296 
2297 	tx_ring->next_to_use = 0;
2298 	tx_ring->next_to_clean = 0;
2299 
2300 	writel(0, tx_ring->head);
2301 	writel(0, tx_ring->tail);
2302 }
2303 
2304 /**
2305  * e1000e_free_tx_resources - Free Tx Resources per Queue
2306  * @tx_ring: Tx descriptor ring
2307  *
2308  * Free all transmit software resources
2309  **/
2310 void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2311 {
2312 	struct e1000_adapter *adapter = tx_ring->adapter;
2313 	struct pci_dev *pdev = adapter->pdev;
2314 
2315 	e1000_clean_tx_ring(tx_ring);
2316 
2317 	vfree(tx_ring->buffer_info);
2318 	tx_ring->buffer_info = NULL;
2319 
2320 	dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2321 			  tx_ring->dma);
2322 	tx_ring->desc = NULL;
2323 }
2324 
2325 /**
2326  * e1000e_free_rx_resources - Free Rx Resources
2327  * @rx_ring: Rx descriptor ring
2328  *
2329  * Free all receive software resources
2330  **/
2331 void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2332 {
2333 	struct e1000_adapter *adapter = rx_ring->adapter;
2334 	struct pci_dev *pdev = adapter->pdev;
2335 	int i;
2336 
2337 	e1000_clean_rx_ring(rx_ring);
2338 
2339 	for (i = 0; i < rx_ring->count; i++)
2340 		kfree(rx_ring->buffer_info[i].ps_pages);
2341 
2342 	vfree(rx_ring->buffer_info);
2343 	rx_ring->buffer_info = NULL;
2344 
2345 	dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2346 			  rx_ring->dma);
2347 	rx_ring->desc = NULL;
2348 }
2349 
2350 /**
2351  * e1000_update_itr - update the dynamic ITR value based on statistics
2352  * @adapter: pointer to adapter
2353  * @itr_setting: current adapter->itr
2354  * @packets: the number of packets during this measurement interval
2355  * @bytes: the number of bytes during this measurement interval
2356  *
2357  *      Stores a new ITR value based on packets and byte
2358  *      counts during the last interrupt.  The advantage of per interrupt
2359  *      computation is faster updates and more accurate ITR for the current
2360  *      traffic pattern.  Constants in this function were computed
2361  *      based on theoretical maximum wire speed and thresholds were set based
2362  *      on testing data as well as attempting to minimize response time
2363  *      while increasing bulk throughput.  This functionality is controlled
2364  *      by the InterruptThrottleRate module parameter.
2365  **/
2366 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2367 				     u16 itr_setting, int packets,
2368 				     int bytes)
2369 {
2370 	unsigned int retval = itr_setting;
2371 
2372 	if (packets == 0)
2373 		goto update_itr_done;
2374 
2375 	switch (itr_setting) {
2376 	case lowest_latency:
2377 		/* handle TSO and jumbo frames */
2378 		if (bytes/packets > 8000)
2379 			retval = bulk_latency;
2380 		else if ((packets < 5) && (bytes > 512))
2381 			retval = low_latency;
2382 		break;
2383 	case low_latency:  /* 50 usec aka 20000 ints/s */
2384 		if (bytes > 10000) {
2385 			/* this if handles the TSO accounting */
2386 			if (bytes/packets > 8000)
2387 				retval = bulk_latency;
2388 			else if ((packets < 10) || ((bytes/packets) > 1200))
2389 				retval = bulk_latency;
2390 			else if ((packets > 35))
2391 				retval = lowest_latency;
2392 		} else if (bytes/packets > 2000) {
2393 			retval = bulk_latency;
2394 		} else if (packets <= 2 && bytes < 512) {
2395 			retval = lowest_latency;
2396 		}
2397 		break;
2398 	case bulk_latency: /* 250 usec aka 4000 ints/s */
2399 		if (bytes > 25000) {
2400 			if (packets > 35)
2401 				retval = low_latency;
2402 		} else if (bytes < 6000) {
2403 			retval = low_latency;
2404 		}
2405 		break;
2406 	}
2407 
2408 update_itr_done:
2409 	return retval;
2410 }
2411 
2412 static void e1000_set_itr(struct e1000_adapter *adapter)
2413 {
2414 	struct e1000_hw *hw = &adapter->hw;
2415 	u16 current_itr;
2416 	u32 new_itr = adapter->itr;
2417 
2418 	/* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2419 	if (adapter->link_speed != SPEED_1000) {
2420 		current_itr = 0;
2421 		new_itr = 4000;
2422 		goto set_itr_now;
2423 	}
2424 
2425 	if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2426 		new_itr = 0;
2427 		goto set_itr_now;
2428 	}
2429 
2430 	adapter->tx_itr = e1000_update_itr(adapter,
2431 				    adapter->tx_itr,
2432 				    adapter->total_tx_packets,
2433 				    adapter->total_tx_bytes);
2434 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
2435 	if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2436 		adapter->tx_itr = low_latency;
2437 
2438 	adapter->rx_itr = e1000_update_itr(adapter,
2439 				    adapter->rx_itr,
2440 				    adapter->total_rx_packets,
2441 				    adapter->total_rx_bytes);
2442 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
2443 	if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2444 		adapter->rx_itr = low_latency;
2445 
2446 	current_itr = max(adapter->rx_itr, adapter->tx_itr);
2447 
2448 	switch (current_itr) {
2449 	/* counts and packets in update_itr are dependent on these numbers */
2450 	case lowest_latency:
2451 		new_itr = 70000;
2452 		break;
2453 	case low_latency:
2454 		new_itr = 20000; /* aka hwitr = ~200 */
2455 		break;
2456 	case bulk_latency:
2457 		new_itr = 4000;
2458 		break;
2459 	default:
2460 		break;
2461 	}
2462 
2463 set_itr_now:
2464 	if (new_itr != adapter->itr) {
2465 		/*
2466 		 * this attempts to bias the interrupt rate towards Bulk
2467 		 * by adding intermediate steps when interrupt rate is
2468 		 * increasing
2469 		 */
2470 		new_itr = new_itr > adapter->itr ?
2471 			     min(adapter->itr + (new_itr >> 2), new_itr) :
2472 			     new_itr;
2473 		adapter->itr = new_itr;
2474 		adapter->rx_ring->itr_val = new_itr;
2475 		if (adapter->msix_entries)
2476 			adapter->rx_ring->set_itr = 1;
2477 		else
2478 			if (new_itr)
2479 				ew32(ITR, 1000000000 / (new_itr * 256));
2480 			else
2481 				ew32(ITR, 0);
2482 	}
2483 }
2484 
2485 /**
2486  * e1000_alloc_queues - Allocate memory for all rings
2487  * @adapter: board private structure to initialize
2488  **/
2489 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2490 {
2491 	int size = sizeof(struct e1000_ring);
2492 
2493 	adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2494 	if (!adapter->tx_ring)
2495 		goto err;
2496 	adapter->tx_ring->count = adapter->tx_ring_count;
2497 	adapter->tx_ring->adapter = adapter;
2498 
2499 	adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2500 	if (!adapter->rx_ring)
2501 		goto err;
2502 	adapter->rx_ring->count = adapter->rx_ring_count;
2503 	adapter->rx_ring->adapter = adapter;
2504 
2505 	return 0;
2506 err:
2507 	e_err("Unable to allocate memory for queues\n");
2508 	kfree(adapter->rx_ring);
2509 	kfree(adapter->tx_ring);
2510 	return -ENOMEM;
2511 }
2512 
2513 /**
2514  * e1000_clean - NAPI Rx polling callback
2515  * @napi: struct associated with this polling callback
2516  * @budget: amount of packets driver is allowed to process this poll
2517  **/
2518 static int e1000_clean(struct napi_struct *napi, int budget)
2519 {
2520 	struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2521 	struct e1000_hw *hw = &adapter->hw;
2522 	struct net_device *poll_dev = adapter->netdev;
2523 	int tx_cleaned = 1, work_done = 0;
2524 
2525 	adapter = netdev_priv(poll_dev);
2526 
2527 	if (adapter->msix_entries &&
2528 	    !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2529 		goto clean_rx;
2530 
2531 	tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2532 
2533 clean_rx:
2534 	adapter->clean_rx(adapter->rx_ring, &work_done, budget);
2535 
2536 	if (!tx_cleaned)
2537 		work_done = budget;
2538 
2539 	/* If budget not fully consumed, exit the polling mode */
2540 	if (work_done < budget) {
2541 		if (adapter->itr_setting & 3)
2542 			e1000_set_itr(adapter);
2543 		napi_complete(napi);
2544 		if (!test_bit(__E1000_DOWN, &adapter->state)) {
2545 			if (adapter->msix_entries)
2546 				ew32(IMS, adapter->rx_ring->ims_val);
2547 			else
2548 				e1000_irq_enable(adapter);
2549 		}
2550 	}
2551 
2552 	return work_done;
2553 }
2554 
2555 static int e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2556 {
2557 	struct e1000_adapter *adapter = netdev_priv(netdev);
2558 	struct e1000_hw *hw = &adapter->hw;
2559 	u32 vfta, index;
2560 
2561 	/* don't update vlan cookie if already programmed */
2562 	if ((adapter->hw.mng_cookie.status &
2563 	     E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2564 	    (vid == adapter->mng_vlan_id))
2565 		return 0;
2566 
2567 	/* add VID to filter table */
2568 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2569 		index = (vid >> 5) & 0x7F;
2570 		vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2571 		vfta |= (1 << (vid & 0x1F));
2572 		hw->mac.ops.write_vfta(hw, index, vfta);
2573 	}
2574 
2575 	set_bit(vid, adapter->active_vlans);
2576 
2577 	return 0;
2578 }
2579 
2580 static int e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2581 {
2582 	struct e1000_adapter *adapter = netdev_priv(netdev);
2583 	struct e1000_hw *hw = &adapter->hw;
2584 	u32 vfta, index;
2585 
2586 	if ((adapter->hw.mng_cookie.status &
2587 	     E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2588 	    (vid == adapter->mng_vlan_id)) {
2589 		/* release control to f/w */
2590 		e1000e_release_hw_control(adapter);
2591 		return 0;
2592 	}
2593 
2594 	/* remove VID from filter table */
2595 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2596 		index = (vid >> 5) & 0x7F;
2597 		vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2598 		vfta &= ~(1 << (vid & 0x1F));
2599 		hw->mac.ops.write_vfta(hw, index, vfta);
2600 	}
2601 
2602 	clear_bit(vid, adapter->active_vlans);
2603 
2604 	return 0;
2605 }
2606 
2607 /**
2608  * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2609  * @adapter: board private structure to initialize
2610  **/
2611 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2612 {
2613 	struct net_device *netdev = adapter->netdev;
2614 	struct e1000_hw *hw = &adapter->hw;
2615 	u32 rctl;
2616 
2617 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2618 		/* disable VLAN receive filtering */
2619 		rctl = er32(RCTL);
2620 		rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2621 		ew32(RCTL, rctl);
2622 
2623 		if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2624 			e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2625 			adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2626 		}
2627 	}
2628 }
2629 
2630 /**
2631  * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2632  * @adapter: board private structure to initialize
2633  **/
2634 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2635 {
2636 	struct e1000_hw *hw = &adapter->hw;
2637 	u32 rctl;
2638 
2639 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2640 		/* enable VLAN receive filtering */
2641 		rctl = er32(RCTL);
2642 		rctl |= E1000_RCTL_VFE;
2643 		rctl &= ~E1000_RCTL_CFIEN;
2644 		ew32(RCTL, rctl);
2645 	}
2646 }
2647 
2648 /**
2649  * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2650  * @adapter: board private structure to initialize
2651  **/
2652 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2653 {
2654 	struct e1000_hw *hw = &adapter->hw;
2655 	u32 ctrl;
2656 
2657 	/* disable VLAN tag insert/strip */
2658 	ctrl = er32(CTRL);
2659 	ctrl &= ~E1000_CTRL_VME;
2660 	ew32(CTRL, ctrl);
2661 }
2662 
2663 /**
2664  * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2665  * @adapter: board private structure to initialize
2666  **/
2667 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2668 {
2669 	struct e1000_hw *hw = &adapter->hw;
2670 	u32 ctrl;
2671 
2672 	/* enable VLAN tag insert/strip */
2673 	ctrl = er32(CTRL);
2674 	ctrl |= E1000_CTRL_VME;
2675 	ew32(CTRL, ctrl);
2676 }
2677 
2678 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2679 {
2680 	struct net_device *netdev = adapter->netdev;
2681 	u16 vid = adapter->hw.mng_cookie.vlan_id;
2682 	u16 old_vid = adapter->mng_vlan_id;
2683 
2684 	if (adapter->hw.mng_cookie.status &
2685 	    E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2686 		e1000_vlan_rx_add_vid(netdev, vid);
2687 		adapter->mng_vlan_id = vid;
2688 	}
2689 
2690 	if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2691 		e1000_vlan_rx_kill_vid(netdev, old_vid);
2692 }
2693 
2694 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2695 {
2696 	u16 vid;
2697 
2698 	e1000_vlan_rx_add_vid(adapter->netdev, 0);
2699 
2700 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2701 		e1000_vlan_rx_add_vid(adapter->netdev, vid);
2702 }
2703 
2704 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2705 {
2706 	struct e1000_hw *hw = &adapter->hw;
2707 	u32 manc, manc2h, mdef, i, j;
2708 
2709 	if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2710 		return;
2711 
2712 	manc = er32(MANC);
2713 
2714 	/*
2715 	 * enable receiving management packets to the host. this will probably
2716 	 * generate destination unreachable messages from the host OS, but
2717 	 * the packets will be handled on SMBUS
2718 	 */
2719 	manc |= E1000_MANC_EN_MNG2HOST;
2720 	manc2h = er32(MANC2H);
2721 
2722 	switch (hw->mac.type) {
2723 	default:
2724 		manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2725 		break;
2726 	case e1000_82574:
2727 	case e1000_82583:
2728 		/*
2729 		 * Check if IPMI pass-through decision filter already exists;
2730 		 * if so, enable it.
2731 		 */
2732 		for (i = 0, j = 0; i < 8; i++) {
2733 			mdef = er32(MDEF(i));
2734 
2735 			/* Ignore filters with anything other than IPMI ports */
2736 			if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2737 				continue;
2738 
2739 			/* Enable this decision filter in MANC2H */
2740 			if (mdef)
2741 				manc2h |= (1 << i);
2742 
2743 			j |= mdef;
2744 		}
2745 
2746 		if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2747 			break;
2748 
2749 		/* Create new decision filter in an empty filter */
2750 		for (i = 0, j = 0; i < 8; i++)
2751 			if (er32(MDEF(i)) == 0) {
2752 				ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2753 					       E1000_MDEF_PORT_664));
2754 				manc2h |= (1 << 1);
2755 				j++;
2756 				break;
2757 			}
2758 
2759 		if (!j)
2760 			e_warn("Unable to create IPMI pass-through filter\n");
2761 		break;
2762 	}
2763 
2764 	ew32(MANC2H, manc2h);
2765 	ew32(MANC, manc);
2766 }
2767 
2768 /**
2769  * e1000_configure_tx - Configure Transmit Unit after Reset
2770  * @adapter: board private structure
2771  *
2772  * Configure the Tx unit of the MAC after a reset.
2773  **/
2774 static void e1000_configure_tx(struct e1000_adapter *adapter)
2775 {
2776 	struct e1000_hw *hw = &adapter->hw;
2777 	struct e1000_ring *tx_ring = adapter->tx_ring;
2778 	u64 tdba;
2779 	u32 tdlen, tarc;
2780 
2781 	/* Setup the HW Tx Head and Tail descriptor pointers */
2782 	tdba = tx_ring->dma;
2783 	tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2784 	ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2785 	ew32(TDBAH, (tdba >> 32));
2786 	ew32(TDLEN, tdlen);
2787 	ew32(TDH, 0);
2788 	ew32(TDT, 0);
2789 	tx_ring->head = adapter->hw.hw_addr + E1000_TDH;
2790 	tx_ring->tail = adapter->hw.hw_addr + E1000_TDT;
2791 
2792 	/* Set the Tx Interrupt Delay register */
2793 	ew32(TIDV, adapter->tx_int_delay);
2794 	/* Tx irq moderation */
2795 	ew32(TADV, adapter->tx_abs_int_delay);
2796 
2797 	if (adapter->flags2 & FLAG2_DMA_BURST) {
2798 		u32 txdctl = er32(TXDCTL(0));
2799 		txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2800 			    E1000_TXDCTL_WTHRESH);
2801 		/*
2802 		 * set up some performance related parameters to encourage the
2803 		 * hardware to use the bus more efficiently in bursts, depends
2804 		 * on the tx_int_delay to be enabled,
2805 		 * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2806 		 * hthresh = 1 ==> prefetch when one or more available
2807 		 * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2808 		 * BEWARE: this seems to work but should be considered first if
2809 		 * there are Tx hangs or other Tx related bugs
2810 		 */
2811 		txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2812 		ew32(TXDCTL(0), txdctl);
2813 	}
2814 	/* erratum work around: set txdctl the same for both queues */
2815 	ew32(TXDCTL(1), er32(TXDCTL(0)));
2816 
2817 	if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2818 		tarc = er32(TARC(0));
2819 		/*
2820 		 * set the speed mode bit, we'll clear it if we're not at
2821 		 * gigabit link later
2822 		 */
2823 #define SPEED_MODE_BIT (1 << 21)
2824 		tarc |= SPEED_MODE_BIT;
2825 		ew32(TARC(0), tarc);
2826 	}
2827 
2828 	/* errata: program both queues to unweighted RR */
2829 	if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2830 		tarc = er32(TARC(0));
2831 		tarc |= 1;
2832 		ew32(TARC(0), tarc);
2833 		tarc = er32(TARC(1));
2834 		tarc |= 1;
2835 		ew32(TARC(1), tarc);
2836 	}
2837 
2838 	/* Setup Transmit Descriptor Settings for eop descriptor */
2839 	adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2840 
2841 	/* only set IDE if we are delaying interrupts using the timers */
2842 	if (adapter->tx_int_delay)
2843 		adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2844 
2845 	/* enable Report Status bit */
2846 	adapter->txd_cmd |= E1000_TXD_CMD_RS;
2847 
2848 	e1000e_config_collision_dist(hw);
2849 }
2850 
2851 /**
2852  * e1000_setup_rctl - configure the receive control registers
2853  * @adapter: Board private structure
2854  **/
2855 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2856 			   (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2857 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2858 {
2859 	struct e1000_hw *hw = &adapter->hw;
2860 	u32 rctl, rfctl;
2861 	u32 pages = 0;
2862 
2863 	/* Workaround Si errata on 82579 - configure jumbo frame flow */
2864 	if (hw->mac.type == e1000_pch2lan) {
2865 		s32 ret_val;
2866 
2867 		if (adapter->netdev->mtu > ETH_DATA_LEN)
2868 			ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2869 		else
2870 			ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2871 
2872 		if (ret_val)
2873 			e_dbg("failed to enable jumbo frame workaround mode\n");
2874 	}
2875 
2876 	/* Program MC offset vector base */
2877 	rctl = er32(RCTL);
2878 	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2879 	rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2880 		E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2881 		(adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2882 
2883 	/* Do not Store bad packets */
2884 	rctl &= ~E1000_RCTL_SBP;
2885 
2886 	/* Enable Long Packet receive */
2887 	if (adapter->netdev->mtu <= ETH_DATA_LEN)
2888 		rctl &= ~E1000_RCTL_LPE;
2889 	else
2890 		rctl |= E1000_RCTL_LPE;
2891 
2892 	/* Some systems expect that the CRC is included in SMBUS traffic. The
2893 	 * hardware strips the CRC before sending to both SMBUS (BMC) and to
2894 	 * host memory when this is enabled
2895 	 */
2896 	if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2897 		rctl |= E1000_RCTL_SECRC;
2898 
2899 	/* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2900 	if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2901 		u16 phy_data;
2902 
2903 		e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2904 		phy_data &= 0xfff8;
2905 		phy_data |= (1 << 2);
2906 		e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2907 
2908 		e1e_rphy(hw, 22, &phy_data);
2909 		phy_data &= 0x0fff;
2910 		phy_data |= (1 << 14);
2911 		e1e_wphy(hw, 0x10, 0x2823);
2912 		e1e_wphy(hw, 0x11, 0x0003);
2913 		e1e_wphy(hw, 22, phy_data);
2914 	}
2915 
2916 	/* Setup buffer sizes */
2917 	rctl &= ~E1000_RCTL_SZ_4096;
2918 	rctl |= E1000_RCTL_BSEX;
2919 	switch (adapter->rx_buffer_len) {
2920 	case 2048:
2921 	default:
2922 		rctl |= E1000_RCTL_SZ_2048;
2923 		rctl &= ~E1000_RCTL_BSEX;
2924 		break;
2925 	case 4096:
2926 		rctl |= E1000_RCTL_SZ_4096;
2927 		break;
2928 	case 8192:
2929 		rctl |= E1000_RCTL_SZ_8192;
2930 		break;
2931 	case 16384:
2932 		rctl |= E1000_RCTL_SZ_16384;
2933 		break;
2934 	}
2935 
2936 	/* Enable Extended Status in all Receive Descriptors */
2937 	rfctl = er32(RFCTL);
2938 	rfctl |= E1000_RFCTL_EXTEN;
2939 
2940 	/*
2941 	 * 82571 and greater support packet-split where the protocol
2942 	 * header is placed in skb->data and the packet data is
2943 	 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2944 	 * In the case of a non-split, skb->data is linearly filled,
2945 	 * followed by the page buffers.  Therefore, skb->data is
2946 	 * sized to hold the largest protocol header.
2947 	 *
2948 	 * allocations using alloc_page take too long for regular MTU
2949 	 * so only enable packet split for jumbo frames
2950 	 *
2951 	 * Using pages when the page size is greater than 16k wastes
2952 	 * a lot of memory, since we allocate 3 pages at all times
2953 	 * per packet.
2954 	 */
2955 	pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2956 	if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2957 		adapter->rx_ps_pages = pages;
2958 	else
2959 		adapter->rx_ps_pages = 0;
2960 
2961 	if (adapter->rx_ps_pages) {
2962 		u32 psrctl = 0;
2963 
2964 		/*
2965 		 * disable packet split support for IPv6 extension headers,
2966 		 * because some malformed IPv6 headers can hang the Rx
2967 		 */
2968 		rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2969 			  E1000_RFCTL_NEW_IPV6_EXT_DIS);
2970 
2971 		/* Enable Packet split descriptors */
2972 		rctl |= E1000_RCTL_DTYP_PS;
2973 
2974 		psrctl |= adapter->rx_ps_bsize0 >>
2975 			E1000_PSRCTL_BSIZE0_SHIFT;
2976 
2977 		switch (adapter->rx_ps_pages) {
2978 		case 3:
2979 			psrctl |= PAGE_SIZE <<
2980 				E1000_PSRCTL_BSIZE3_SHIFT;
2981 		case 2:
2982 			psrctl |= PAGE_SIZE <<
2983 				E1000_PSRCTL_BSIZE2_SHIFT;
2984 		case 1:
2985 			psrctl |= PAGE_SIZE >>
2986 				E1000_PSRCTL_BSIZE1_SHIFT;
2987 			break;
2988 		}
2989 
2990 		ew32(PSRCTL, psrctl);
2991 	}
2992 
2993 	ew32(RFCTL, rfctl);
2994 	ew32(RCTL, rctl);
2995 	/* just started the receive unit, no need to restart */
2996 	adapter->flags &= ~FLAG_RX_RESTART_NOW;
2997 }
2998 
2999 /**
3000  * e1000_configure_rx - Configure Receive Unit after Reset
3001  * @adapter: board private structure
3002  *
3003  * Configure the Rx unit of the MAC after a reset.
3004  **/
3005 static void e1000_configure_rx(struct e1000_adapter *adapter)
3006 {
3007 	struct e1000_hw *hw = &adapter->hw;
3008 	struct e1000_ring *rx_ring = adapter->rx_ring;
3009 	u64 rdba;
3010 	u32 rdlen, rctl, rxcsum, ctrl_ext;
3011 
3012 	if (adapter->rx_ps_pages) {
3013 		/* this is a 32 byte descriptor */
3014 		rdlen = rx_ring->count *
3015 		    sizeof(union e1000_rx_desc_packet_split);
3016 		adapter->clean_rx = e1000_clean_rx_irq_ps;
3017 		adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3018 	} else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3019 		rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3020 		adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3021 		adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3022 	} else {
3023 		rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3024 		adapter->clean_rx = e1000_clean_rx_irq;
3025 		adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3026 	}
3027 
3028 	/* disable receives while setting up the descriptors */
3029 	rctl = er32(RCTL);
3030 	if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3031 		ew32(RCTL, rctl & ~E1000_RCTL_EN);
3032 	e1e_flush();
3033 	usleep_range(10000, 20000);
3034 
3035 	if (adapter->flags2 & FLAG2_DMA_BURST) {
3036 		/*
3037 		 * set the writeback threshold (only takes effect if the RDTR
3038 		 * is set). set GRAN=1 and write back up to 0x4 worth, and
3039 		 * enable prefetching of 0x20 Rx descriptors
3040 		 * granularity = 01
3041 		 * wthresh = 04,
3042 		 * hthresh = 04,
3043 		 * pthresh = 0x20
3044 		 */
3045 		ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3046 		ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3047 
3048 		/*
3049 		 * override the delay timers for enabling bursting, only if
3050 		 * the value was not set by the user via module options
3051 		 */
3052 		if (adapter->rx_int_delay == DEFAULT_RDTR)
3053 			adapter->rx_int_delay = BURST_RDTR;
3054 		if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3055 			adapter->rx_abs_int_delay = BURST_RADV;
3056 	}
3057 
3058 	/* set the Receive Delay Timer Register */
3059 	ew32(RDTR, adapter->rx_int_delay);
3060 
3061 	/* irq moderation */
3062 	ew32(RADV, adapter->rx_abs_int_delay);
3063 	if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3064 		ew32(ITR, 1000000000 / (adapter->itr * 256));
3065 
3066 	ctrl_ext = er32(CTRL_EXT);
3067 	/* Auto-Mask interrupts upon ICR access */
3068 	ctrl_ext |= E1000_CTRL_EXT_IAME;
3069 	ew32(IAM, 0xffffffff);
3070 	ew32(CTRL_EXT, ctrl_ext);
3071 	e1e_flush();
3072 
3073 	/*
3074 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
3075 	 * the Base and Length of the Rx Descriptor Ring
3076 	 */
3077 	rdba = rx_ring->dma;
3078 	ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
3079 	ew32(RDBAH, (rdba >> 32));
3080 	ew32(RDLEN, rdlen);
3081 	ew32(RDH, 0);
3082 	ew32(RDT, 0);
3083 	rx_ring->head = adapter->hw.hw_addr + E1000_RDH;
3084 	rx_ring->tail = adapter->hw.hw_addr + E1000_RDT;
3085 
3086 	/* Enable Receive Checksum Offload for TCP and UDP */
3087 	rxcsum = er32(RXCSUM);
3088 	if (adapter->netdev->features & NETIF_F_RXCSUM) {
3089 		rxcsum |= E1000_RXCSUM_TUOFL;
3090 
3091 		/*
3092 		 * IPv4 payload checksum for UDP fragments must be
3093 		 * used in conjunction with packet-split.
3094 		 */
3095 		if (adapter->rx_ps_pages)
3096 			rxcsum |= E1000_RXCSUM_IPPCSE;
3097 	} else {
3098 		rxcsum &= ~E1000_RXCSUM_TUOFL;
3099 		/* no need to clear IPPCSE as it defaults to 0 */
3100 	}
3101 	ew32(RXCSUM, rxcsum);
3102 
3103 	if (adapter->hw.mac.type == e1000_pch2lan) {
3104 		/*
3105 		 * With jumbo frames, excessive C-state transition
3106 		 * latencies result in dropped transactions.
3107 		 */
3108 		if (adapter->netdev->mtu > ETH_DATA_LEN) {
3109 			u32 rxdctl = er32(RXDCTL(0));
3110 			ew32(RXDCTL(0), rxdctl | 0x3);
3111 			pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
3112 		} else {
3113 			pm_qos_update_request(&adapter->netdev->pm_qos_req,
3114 					      PM_QOS_DEFAULT_VALUE);
3115 		}
3116 	}
3117 
3118 	/* Enable Receives */
3119 	ew32(RCTL, rctl);
3120 }
3121 
3122 /**
3123  * e1000e_write_mc_addr_list - write multicast addresses to MTA
3124  * @netdev: network interface device structure
3125  *
3126  * Writes multicast address list to the MTA hash table.
3127  * Returns: -ENOMEM on failure
3128  *                0 on no addresses written
3129  *                X on writing X addresses to MTA
3130  */
3131 static int e1000e_write_mc_addr_list(struct net_device *netdev)
3132 {
3133 	struct e1000_adapter *adapter = netdev_priv(netdev);
3134 	struct e1000_hw *hw = &adapter->hw;
3135 	struct netdev_hw_addr *ha;
3136 	u8 *mta_list;
3137 	int i;
3138 
3139 	if (netdev_mc_empty(netdev)) {
3140 		/* nothing to program, so clear mc list */
3141 		hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3142 		return 0;
3143 	}
3144 
3145 	mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3146 	if (!mta_list)
3147 		return -ENOMEM;
3148 
3149 	/* update_mc_addr_list expects a packed array of only addresses. */
3150 	i = 0;
3151 	netdev_for_each_mc_addr(ha, netdev)
3152 		memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3153 
3154 	hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3155 	kfree(mta_list);
3156 
3157 	return netdev_mc_count(netdev);
3158 }
3159 
3160 /**
3161  * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3162  * @netdev: network interface device structure
3163  *
3164  * Writes unicast address list to the RAR table.
3165  * Returns: -ENOMEM on failure/insufficient address space
3166  *                0 on no addresses written
3167  *                X on writing X addresses to the RAR table
3168  **/
3169 static int e1000e_write_uc_addr_list(struct net_device *netdev)
3170 {
3171 	struct e1000_adapter *adapter = netdev_priv(netdev);
3172 	struct e1000_hw *hw = &adapter->hw;
3173 	unsigned int rar_entries = hw->mac.rar_entry_count;
3174 	int count = 0;
3175 
3176 	/* save a rar entry for our hardware address */
3177 	rar_entries--;
3178 
3179 	/* save a rar entry for the LAA workaround */
3180 	if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3181 		rar_entries--;
3182 
3183 	/* return ENOMEM indicating insufficient memory for addresses */
3184 	if (netdev_uc_count(netdev) > rar_entries)
3185 		return -ENOMEM;
3186 
3187 	if (!netdev_uc_empty(netdev) && rar_entries) {
3188 		struct netdev_hw_addr *ha;
3189 
3190 		/*
3191 		 * write the addresses in reverse order to avoid write
3192 		 * combining
3193 		 */
3194 		netdev_for_each_uc_addr(ha, netdev) {
3195 			if (!rar_entries)
3196 				break;
3197 			e1000e_rar_set(hw, ha->addr, rar_entries--);
3198 			count++;
3199 		}
3200 	}
3201 
3202 	/* zero out the remaining RAR entries not used above */
3203 	for (; rar_entries > 0; rar_entries--) {
3204 		ew32(RAH(rar_entries), 0);
3205 		ew32(RAL(rar_entries), 0);
3206 	}
3207 	e1e_flush();
3208 
3209 	return count;
3210 }
3211 
3212 /**
3213  * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3214  * @netdev: network interface device structure
3215  *
3216  * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3217  * address list or the network interface flags are updated.  This routine is
3218  * responsible for configuring the hardware for proper unicast, multicast,
3219  * promiscuous mode, and all-multi behavior.
3220  **/
3221 static void e1000e_set_rx_mode(struct net_device *netdev)
3222 {
3223 	struct e1000_adapter *adapter = netdev_priv(netdev);
3224 	struct e1000_hw *hw = &adapter->hw;
3225 	u32 rctl;
3226 
3227 	/* Check for Promiscuous and All Multicast modes */
3228 	rctl = er32(RCTL);
3229 
3230 	/* clear the affected bits */
3231 	rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3232 
3233 	if (netdev->flags & IFF_PROMISC) {
3234 		rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3235 		/* Do not hardware filter VLANs in promisc mode */
3236 		e1000e_vlan_filter_disable(adapter);
3237 	} else {
3238 		int count;
3239 		if (netdev->flags & IFF_ALLMULTI) {
3240 			rctl |= E1000_RCTL_MPE;
3241 		} else {
3242 			/*
3243 			 * Write addresses to the MTA, if the attempt fails
3244 			 * then we should just turn on promiscuous mode so
3245 			 * that we can at least receive multicast traffic
3246 			 */
3247 			count = e1000e_write_mc_addr_list(netdev);
3248 			if (count < 0)
3249 				rctl |= E1000_RCTL_MPE;
3250 		}
3251 		e1000e_vlan_filter_enable(adapter);
3252 		/*
3253 		 * Write addresses to available RAR registers, if there is not
3254 		 * sufficient space to store all the addresses then enable
3255 		 * unicast promiscuous mode
3256 		 */
3257 		count = e1000e_write_uc_addr_list(netdev);
3258 		if (count < 0)
3259 			rctl |= E1000_RCTL_UPE;
3260 	}
3261 
3262 	ew32(RCTL, rctl);
3263 
3264 	if (netdev->features & NETIF_F_HW_VLAN_RX)
3265 		e1000e_vlan_strip_enable(adapter);
3266 	else
3267 		e1000e_vlan_strip_disable(adapter);
3268 }
3269 
3270 static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3271 {
3272 	struct e1000_hw *hw = &adapter->hw;
3273 	u32 mrqc, rxcsum;
3274 	int i;
3275 	static const u32 rsskey[10] = {
3276 		0xda565a6d, 0xc20e5b25, 0x3d256741, 0xb08fa343, 0xcb2bcad0,
3277 		0xb4307bae, 0xa32dcb77, 0x0cf23080, 0x3bb7426a, 0xfa01acbe
3278 	};
3279 
3280 	/* Fill out hash function seed */
3281 	for (i = 0; i < 10; i++)
3282 		ew32(RSSRK(i), rsskey[i]);
3283 
3284 	/* Direct all traffic to queue 0 */
3285 	for (i = 0; i < 32; i++)
3286 		ew32(RETA(i), 0);
3287 
3288 	/*
3289 	 * Disable raw packet checksumming so that RSS hash is placed in
3290 	 * descriptor on writeback.
3291 	 */
3292 	rxcsum = er32(RXCSUM);
3293 	rxcsum |= E1000_RXCSUM_PCSD;
3294 
3295 	ew32(RXCSUM, rxcsum);
3296 
3297 	mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3298 		E1000_MRQC_RSS_FIELD_IPV4_TCP |
3299 		E1000_MRQC_RSS_FIELD_IPV6 |
3300 		E1000_MRQC_RSS_FIELD_IPV6_TCP |
3301 		E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3302 
3303 	ew32(MRQC, mrqc);
3304 }
3305 
3306 /**
3307  * e1000_configure - configure the hardware for Rx and Tx
3308  * @adapter: private board structure
3309  **/
3310 static void e1000_configure(struct e1000_adapter *adapter)
3311 {
3312 	struct e1000_ring *rx_ring = adapter->rx_ring;
3313 
3314 	e1000e_set_rx_mode(adapter->netdev);
3315 
3316 	e1000_restore_vlan(adapter);
3317 	e1000_init_manageability_pt(adapter);
3318 
3319 	e1000_configure_tx(adapter);
3320 
3321 	if (adapter->netdev->features & NETIF_F_RXHASH)
3322 		e1000e_setup_rss_hash(adapter);
3323 	e1000_setup_rctl(adapter);
3324 	e1000_configure_rx(adapter);
3325 	adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3326 }
3327 
3328 /**
3329  * e1000e_power_up_phy - restore link in case the phy was powered down
3330  * @adapter: address of board private structure
3331  *
3332  * The phy may be powered down to save power and turn off link when the
3333  * driver is unloaded and wake on lan is not enabled (among others)
3334  * *** this routine MUST be followed by a call to e1000e_reset ***
3335  **/
3336 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3337 {
3338 	if (adapter->hw.phy.ops.power_up)
3339 		adapter->hw.phy.ops.power_up(&adapter->hw);
3340 
3341 	adapter->hw.mac.ops.setup_link(&adapter->hw);
3342 }
3343 
3344 /**
3345  * e1000_power_down_phy - Power down the PHY
3346  *
3347  * Power down the PHY so no link is implied when interface is down.
3348  * The PHY cannot be powered down if management or WoL is active.
3349  */
3350 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3351 {
3352 	/* WoL is enabled */
3353 	if (adapter->wol)
3354 		return;
3355 
3356 	if (adapter->hw.phy.ops.power_down)
3357 		adapter->hw.phy.ops.power_down(&adapter->hw);
3358 }
3359 
3360 /**
3361  * e1000e_reset - bring the hardware into a known good state
3362  *
3363  * This function boots the hardware and enables some settings that
3364  * require a configuration cycle of the hardware - those cannot be
3365  * set/changed during runtime. After reset the device needs to be
3366  * properly configured for Rx, Tx etc.
3367  */
3368 void e1000e_reset(struct e1000_adapter *adapter)
3369 {
3370 	struct e1000_mac_info *mac = &adapter->hw.mac;
3371 	struct e1000_fc_info *fc = &adapter->hw.fc;
3372 	struct e1000_hw *hw = &adapter->hw;
3373 	u32 tx_space, min_tx_space, min_rx_space;
3374 	u32 pba = adapter->pba;
3375 	u16 hwm;
3376 
3377 	/* reset Packet Buffer Allocation to default */
3378 	ew32(PBA, pba);
3379 
3380 	if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3381 		/*
3382 		 * To maintain wire speed transmits, the Tx FIFO should be
3383 		 * large enough to accommodate two full transmit packets,
3384 		 * rounded up to the next 1KB and expressed in KB.  Likewise,
3385 		 * the Rx FIFO should be large enough to accommodate at least
3386 		 * one full receive packet and is similarly rounded up and
3387 		 * expressed in KB.
3388 		 */
3389 		pba = er32(PBA);
3390 		/* upper 16 bits has Tx packet buffer allocation size in KB */
3391 		tx_space = pba >> 16;
3392 		/* lower 16 bits has Rx packet buffer allocation size in KB */
3393 		pba &= 0xffff;
3394 		/*
3395 		 * the Tx fifo also stores 16 bytes of information about the Tx
3396 		 * but don't include ethernet FCS because hardware appends it
3397 		 */
3398 		min_tx_space = (adapter->max_frame_size +
3399 				sizeof(struct e1000_tx_desc) -
3400 				ETH_FCS_LEN) * 2;
3401 		min_tx_space = ALIGN(min_tx_space, 1024);
3402 		min_tx_space >>= 10;
3403 		/* software strips receive CRC, so leave room for it */
3404 		min_rx_space = adapter->max_frame_size;
3405 		min_rx_space = ALIGN(min_rx_space, 1024);
3406 		min_rx_space >>= 10;
3407 
3408 		/*
3409 		 * If current Tx allocation is less than the min Tx FIFO size,
3410 		 * and the min Tx FIFO size is less than the current Rx FIFO
3411 		 * allocation, take space away from current Rx allocation
3412 		 */
3413 		if ((tx_space < min_tx_space) &&
3414 		    ((min_tx_space - tx_space) < pba)) {
3415 			pba -= min_tx_space - tx_space;
3416 
3417 			/*
3418 			 * if short on Rx space, Rx wins and must trump Tx
3419 			 * adjustment or use Early Receive if available
3420 			 */
3421 			if (pba < min_rx_space)
3422 				pba = min_rx_space;
3423 		}
3424 
3425 		ew32(PBA, pba);
3426 	}
3427 
3428 	/*
3429 	 * flow control settings
3430 	 *
3431 	 * The high water mark must be low enough to fit one full frame
3432 	 * (or the size used for early receive) above it in the Rx FIFO.
3433 	 * Set it to the lower of:
3434 	 * - 90% of the Rx FIFO size, and
3435 	 * - the full Rx FIFO size minus one full frame
3436 	 */
3437 	if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3438 		fc->pause_time = 0xFFFF;
3439 	else
3440 		fc->pause_time = E1000_FC_PAUSE_TIME;
3441 	fc->send_xon = 1;
3442 	fc->current_mode = fc->requested_mode;
3443 
3444 	switch (hw->mac.type) {
3445 	case e1000_ich9lan:
3446 	case e1000_ich10lan:
3447 		if (adapter->netdev->mtu > ETH_DATA_LEN) {
3448 			pba = 14;
3449 			ew32(PBA, pba);
3450 			fc->high_water = 0x2800;
3451 			fc->low_water = fc->high_water - 8;
3452 			break;
3453 		}
3454 		/* fall-through */
3455 	default:
3456 		hwm = min(((pba << 10) * 9 / 10),
3457 			  ((pba << 10) - adapter->max_frame_size));
3458 
3459 		fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3460 		fc->low_water = fc->high_water - 8;
3461 		break;
3462 	case e1000_pchlan:
3463 		/*
3464 		 * Workaround PCH LOM adapter hangs with certain network
3465 		 * loads.  If hangs persist, try disabling Tx flow control.
3466 		 */
3467 		if (adapter->netdev->mtu > ETH_DATA_LEN) {
3468 			fc->high_water = 0x3500;
3469 			fc->low_water  = 0x1500;
3470 		} else {
3471 			fc->high_water = 0x5000;
3472 			fc->low_water  = 0x3000;
3473 		}
3474 		fc->refresh_time = 0x1000;
3475 		break;
3476 	case e1000_pch2lan:
3477 		fc->high_water = 0x05C20;
3478 		fc->low_water = 0x05048;
3479 		fc->pause_time = 0x0650;
3480 		fc->refresh_time = 0x0400;
3481 		if (adapter->netdev->mtu > ETH_DATA_LEN) {
3482 			pba = 14;
3483 			ew32(PBA, pba);
3484 		}
3485 		break;
3486 	}
3487 
3488 	/*
3489 	 * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3490 	 * fit in receive buffer.
3491 	 */
3492 	if (adapter->itr_setting & 0x3) {
3493 		if ((adapter->max_frame_size * 2) > (pba << 10)) {
3494 			if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3495 				dev_info(&adapter->pdev->dev,
3496 					"Interrupt Throttle Rate turned off\n");
3497 				adapter->flags2 |= FLAG2_DISABLE_AIM;
3498 				ew32(ITR, 0);
3499 			}
3500 		} else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3501 			dev_info(&adapter->pdev->dev,
3502 				 "Interrupt Throttle Rate turned on\n");
3503 			adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3504 			adapter->itr = 20000;
3505 			ew32(ITR, 1000000000 / (adapter->itr * 256));
3506 		}
3507 	}
3508 
3509 	/* Allow time for pending master requests to run */
3510 	mac->ops.reset_hw(hw);
3511 
3512 	/*
3513 	 * For parts with AMT enabled, let the firmware know
3514 	 * that the network interface is in control
3515 	 */
3516 	if (adapter->flags & FLAG_HAS_AMT)
3517 		e1000e_get_hw_control(adapter);
3518 
3519 	ew32(WUC, 0);
3520 
3521 	if (mac->ops.init_hw(hw))
3522 		e_err("Hardware Error\n");
3523 
3524 	e1000_update_mng_vlan(adapter);
3525 
3526 	/* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3527 	ew32(VET, ETH_P_8021Q);
3528 
3529 	e1000e_reset_adaptive(hw);
3530 
3531 	if (!netif_running(adapter->netdev) &&
3532 	    !test_bit(__E1000_TESTING, &adapter->state)) {
3533 		e1000_power_down_phy(adapter);
3534 		return;
3535 	}
3536 
3537 	e1000_get_phy_info(hw);
3538 
3539 	if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3540 	    !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3541 		u16 phy_data = 0;
3542 		/*
3543 		 * speed up time to link by disabling smart power down, ignore
3544 		 * the return value of this function because there is nothing
3545 		 * different we would do if it failed
3546 		 */
3547 		e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3548 		phy_data &= ~IGP02E1000_PM_SPD;
3549 		e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3550 	}
3551 }
3552 
3553 int e1000e_up(struct e1000_adapter *adapter)
3554 {
3555 	struct e1000_hw *hw = &adapter->hw;
3556 
3557 	/* hardware has been reset, we need to reload some things */
3558 	e1000_configure(adapter);
3559 
3560 	clear_bit(__E1000_DOWN, &adapter->state);
3561 
3562 	if (adapter->msix_entries)
3563 		e1000_configure_msix(adapter);
3564 	e1000_irq_enable(adapter);
3565 
3566 	netif_start_queue(adapter->netdev);
3567 
3568 	/* fire a link change interrupt to start the watchdog */
3569 	if (adapter->msix_entries)
3570 		ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3571 	else
3572 		ew32(ICS, E1000_ICS_LSC);
3573 
3574 	return 0;
3575 }
3576 
3577 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3578 {
3579 	struct e1000_hw *hw = &adapter->hw;
3580 
3581 	if (!(adapter->flags2 & FLAG2_DMA_BURST))
3582 		return;
3583 
3584 	/* flush pending descriptor writebacks to memory */
3585 	ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3586 	ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3587 
3588 	/* execute the writes immediately */
3589 	e1e_flush();
3590 }
3591 
3592 static void e1000e_update_stats(struct e1000_adapter *adapter);
3593 
3594 void e1000e_down(struct e1000_adapter *adapter)
3595 {
3596 	struct net_device *netdev = adapter->netdev;
3597 	struct e1000_hw *hw = &adapter->hw;
3598 	u32 tctl, rctl;
3599 
3600 	/*
3601 	 * signal that we're down so the interrupt handler does not
3602 	 * reschedule our watchdog timer
3603 	 */
3604 	set_bit(__E1000_DOWN, &adapter->state);
3605 
3606 	/* disable receives in the hardware */
3607 	rctl = er32(RCTL);
3608 	if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3609 		ew32(RCTL, rctl & ~E1000_RCTL_EN);
3610 	/* flush and sleep below */
3611 
3612 	netif_stop_queue(netdev);
3613 
3614 	/* disable transmits in the hardware */
3615 	tctl = er32(TCTL);
3616 	tctl &= ~E1000_TCTL_EN;
3617 	ew32(TCTL, tctl);
3618 
3619 	/* flush both disables and wait for them to finish */
3620 	e1e_flush();
3621 	usleep_range(10000, 20000);
3622 
3623 	e1000_irq_disable(adapter);
3624 
3625 	del_timer_sync(&adapter->watchdog_timer);
3626 	del_timer_sync(&adapter->phy_info_timer);
3627 
3628 	netif_carrier_off(netdev);
3629 
3630 	spin_lock(&adapter->stats64_lock);
3631 	e1000e_update_stats(adapter);
3632 	spin_unlock(&adapter->stats64_lock);
3633 
3634 	e1000e_flush_descriptors(adapter);
3635 	e1000_clean_tx_ring(adapter->tx_ring);
3636 	e1000_clean_rx_ring(adapter->rx_ring);
3637 
3638 	adapter->link_speed = 0;
3639 	adapter->link_duplex = 0;
3640 
3641 	if (!pci_channel_offline(adapter->pdev))
3642 		e1000e_reset(adapter);
3643 
3644 	/*
3645 	 * TODO: for power management, we could drop the link and
3646 	 * pci_disable_device here.
3647 	 */
3648 }
3649 
3650 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3651 {
3652 	might_sleep();
3653 	while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3654 		usleep_range(1000, 2000);
3655 	e1000e_down(adapter);
3656 	e1000e_up(adapter);
3657 	clear_bit(__E1000_RESETTING, &adapter->state);
3658 }
3659 
3660 /**
3661  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3662  * @adapter: board private structure to initialize
3663  *
3664  * e1000_sw_init initializes the Adapter private data structure.
3665  * Fields are initialized based on PCI device information and
3666  * OS network device settings (MTU size).
3667  **/
3668 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3669 {
3670 	struct net_device *netdev = adapter->netdev;
3671 
3672 	adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3673 	adapter->rx_ps_bsize0 = 128;
3674 	adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3675 	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3676 	adapter->tx_ring_count = E1000_DEFAULT_TXD;
3677 	adapter->rx_ring_count = E1000_DEFAULT_RXD;
3678 
3679 	spin_lock_init(&adapter->stats64_lock);
3680 
3681 	e1000e_set_interrupt_capability(adapter);
3682 
3683 	if (e1000_alloc_queues(adapter))
3684 		return -ENOMEM;
3685 
3686 	/* Explicitly disable IRQ since the NIC can be in any state. */
3687 	e1000_irq_disable(adapter);
3688 
3689 	set_bit(__E1000_DOWN, &adapter->state);
3690 	return 0;
3691 }
3692 
3693 /**
3694  * e1000_intr_msi_test - Interrupt Handler
3695  * @irq: interrupt number
3696  * @data: pointer to a network interface device structure
3697  **/
3698 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3699 {
3700 	struct net_device *netdev = data;
3701 	struct e1000_adapter *adapter = netdev_priv(netdev);
3702 	struct e1000_hw *hw = &adapter->hw;
3703 	u32 icr = er32(ICR);
3704 
3705 	e_dbg("icr is %08X\n", icr);
3706 	if (icr & E1000_ICR_RXSEQ) {
3707 		adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3708 		wmb();
3709 	}
3710 
3711 	return IRQ_HANDLED;
3712 }
3713 
3714 /**
3715  * e1000_test_msi_interrupt - Returns 0 for successful test
3716  * @adapter: board private struct
3717  *
3718  * code flow taken from tg3.c
3719  **/
3720 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3721 {
3722 	struct net_device *netdev = adapter->netdev;
3723 	struct e1000_hw *hw = &adapter->hw;
3724 	int err;
3725 
3726 	/* poll_enable hasn't been called yet, so don't need disable */
3727 	/* clear any pending events */
3728 	er32(ICR);
3729 
3730 	/* free the real vector and request a test handler */
3731 	e1000_free_irq(adapter);
3732 	e1000e_reset_interrupt_capability(adapter);
3733 
3734 	/* Assume that the test fails, if it succeeds then the test
3735 	 * MSI irq handler will unset this flag */
3736 	adapter->flags |= FLAG_MSI_TEST_FAILED;
3737 
3738 	err = pci_enable_msi(adapter->pdev);
3739 	if (err)
3740 		goto msi_test_failed;
3741 
3742 	err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3743 			  netdev->name, netdev);
3744 	if (err) {
3745 		pci_disable_msi(adapter->pdev);
3746 		goto msi_test_failed;
3747 	}
3748 
3749 	wmb();
3750 
3751 	e1000_irq_enable(adapter);
3752 
3753 	/* fire an unusual interrupt on the test handler */
3754 	ew32(ICS, E1000_ICS_RXSEQ);
3755 	e1e_flush();
3756 	msleep(50);
3757 
3758 	e1000_irq_disable(adapter);
3759 
3760 	rmb();
3761 
3762 	if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3763 		adapter->int_mode = E1000E_INT_MODE_LEGACY;
3764 		e_info("MSI interrupt test failed, using legacy interrupt.\n");
3765 	} else {
3766 		e_dbg("MSI interrupt test succeeded!\n");
3767 	}
3768 
3769 	free_irq(adapter->pdev->irq, netdev);
3770 	pci_disable_msi(adapter->pdev);
3771 
3772 msi_test_failed:
3773 	e1000e_set_interrupt_capability(adapter);
3774 	return e1000_request_irq(adapter);
3775 }
3776 
3777 /**
3778  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3779  * @adapter: board private struct
3780  *
3781  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3782  **/
3783 static int e1000_test_msi(struct e1000_adapter *adapter)
3784 {
3785 	int err;
3786 	u16 pci_cmd;
3787 
3788 	if (!(adapter->flags & FLAG_MSI_ENABLED))
3789 		return 0;
3790 
3791 	/* disable SERR in case the MSI write causes a master abort */
3792 	pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3793 	if (pci_cmd & PCI_COMMAND_SERR)
3794 		pci_write_config_word(adapter->pdev, PCI_COMMAND,
3795 				      pci_cmd & ~PCI_COMMAND_SERR);
3796 
3797 	err = e1000_test_msi_interrupt(adapter);
3798 
3799 	/* re-enable SERR */
3800 	if (pci_cmd & PCI_COMMAND_SERR) {
3801 		pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3802 		pci_cmd |= PCI_COMMAND_SERR;
3803 		pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3804 	}
3805 
3806 	return err;
3807 }
3808 
3809 /**
3810  * e1000_open - Called when a network interface is made active
3811  * @netdev: network interface device structure
3812  *
3813  * Returns 0 on success, negative value on failure
3814  *
3815  * The open entry point is called when a network interface is made
3816  * active by the system (IFF_UP).  At this point all resources needed
3817  * for transmit and receive operations are allocated, the interrupt
3818  * handler is registered with the OS, the watchdog timer is started,
3819  * and the stack is notified that the interface is ready.
3820  **/
3821 static int e1000_open(struct net_device *netdev)
3822 {
3823 	struct e1000_adapter *adapter = netdev_priv(netdev);
3824 	struct e1000_hw *hw = &adapter->hw;
3825 	struct pci_dev *pdev = adapter->pdev;
3826 	int err;
3827 
3828 	/* disallow open during test */
3829 	if (test_bit(__E1000_TESTING, &adapter->state))
3830 		return -EBUSY;
3831 
3832 	pm_runtime_get_sync(&pdev->dev);
3833 
3834 	netif_carrier_off(netdev);
3835 
3836 	/* allocate transmit descriptors */
3837 	err = e1000e_setup_tx_resources(adapter->tx_ring);
3838 	if (err)
3839 		goto err_setup_tx;
3840 
3841 	/* allocate receive descriptors */
3842 	err = e1000e_setup_rx_resources(adapter->rx_ring);
3843 	if (err)
3844 		goto err_setup_rx;
3845 
3846 	/*
3847 	 * If AMT is enabled, let the firmware know that the network
3848 	 * interface is now open and reset the part to a known state.
3849 	 */
3850 	if (adapter->flags & FLAG_HAS_AMT) {
3851 		e1000e_get_hw_control(adapter);
3852 		e1000e_reset(adapter);
3853 	}
3854 
3855 	e1000e_power_up_phy(adapter);
3856 
3857 	adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3858 	if ((adapter->hw.mng_cookie.status &
3859 	     E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3860 		e1000_update_mng_vlan(adapter);
3861 
3862 	/* DMA latency requirement to workaround jumbo issue */
3863 	if (adapter->hw.mac.type == e1000_pch2lan)
3864 		pm_qos_add_request(&adapter->netdev->pm_qos_req,
3865 				   PM_QOS_CPU_DMA_LATENCY,
3866 				   PM_QOS_DEFAULT_VALUE);
3867 
3868 	/*
3869 	 * before we allocate an interrupt, we must be ready to handle it.
3870 	 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3871 	 * as soon as we call pci_request_irq, so we have to setup our
3872 	 * clean_rx handler before we do so.
3873 	 */
3874 	e1000_configure(adapter);
3875 
3876 	err = e1000_request_irq(adapter);
3877 	if (err)
3878 		goto err_req_irq;
3879 
3880 	/*
3881 	 * Work around PCIe errata with MSI interrupts causing some chipsets to
3882 	 * ignore e1000e MSI messages, which means we need to test our MSI
3883 	 * interrupt now
3884 	 */
3885 	if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3886 		err = e1000_test_msi(adapter);
3887 		if (err) {
3888 			e_err("Interrupt allocation failed\n");
3889 			goto err_req_irq;
3890 		}
3891 	}
3892 
3893 	/* From here on the code is the same as e1000e_up() */
3894 	clear_bit(__E1000_DOWN, &adapter->state);
3895 
3896 	napi_enable(&adapter->napi);
3897 
3898 	e1000_irq_enable(adapter);
3899 
3900 	adapter->tx_hang_recheck = false;
3901 	netif_start_queue(netdev);
3902 
3903 	adapter->idle_check = true;
3904 	pm_runtime_put(&pdev->dev);
3905 
3906 	/* fire a link status change interrupt to start the watchdog */
3907 	if (adapter->msix_entries)
3908 		ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3909 	else
3910 		ew32(ICS, E1000_ICS_LSC);
3911 
3912 	return 0;
3913 
3914 err_req_irq:
3915 	e1000e_release_hw_control(adapter);
3916 	e1000_power_down_phy(adapter);
3917 	e1000e_free_rx_resources(adapter->rx_ring);
3918 err_setup_rx:
3919 	e1000e_free_tx_resources(adapter->tx_ring);
3920 err_setup_tx:
3921 	e1000e_reset(adapter);
3922 	pm_runtime_put_sync(&pdev->dev);
3923 
3924 	return err;
3925 }
3926 
3927 /**
3928  * e1000_close - Disables a network interface
3929  * @netdev: network interface device structure
3930  *
3931  * Returns 0, this is not allowed to fail
3932  *
3933  * The close entry point is called when an interface is de-activated
3934  * by the OS.  The hardware is still under the drivers control, but
3935  * needs to be disabled.  A global MAC reset is issued to stop the
3936  * hardware, and all transmit and receive resources are freed.
3937  **/
3938 static int e1000_close(struct net_device *netdev)
3939 {
3940 	struct e1000_adapter *adapter = netdev_priv(netdev);
3941 	struct pci_dev *pdev = adapter->pdev;
3942 
3943 	WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3944 
3945 	pm_runtime_get_sync(&pdev->dev);
3946 
3947 	napi_disable(&adapter->napi);
3948 
3949 	if (!test_bit(__E1000_DOWN, &adapter->state)) {
3950 		e1000e_down(adapter);
3951 		e1000_free_irq(adapter);
3952 	}
3953 	e1000_power_down_phy(adapter);
3954 
3955 	e1000e_free_tx_resources(adapter->tx_ring);
3956 	e1000e_free_rx_resources(adapter->rx_ring);
3957 
3958 	/*
3959 	 * kill manageability vlan ID if supported, but not if a vlan with
3960 	 * the same ID is registered on the host OS (let 8021q kill it)
3961 	 */
3962 	if (adapter->hw.mng_cookie.status &
3963 	    E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
3964 		e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3965 
3966 	/*
3967 	 * If AMT is enabled, let the firmware know that the network
3968 	 * interface is now closed
3969 	 */
3970 	if ((adapter->flags & FLAG_HAS_AMT) &&
3971 	    !test_bit(__E1000_TESTING, &adapter->state))
3972 		e1000e_release_hw_control(adapter);
3973 
3974 	if (adapter->hw.mac.type == e1000_pch2lan)
3975 		pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3976 
3977 	pm_runtime_put_sync(&pdev->dev);
3978 
3979 	return 0;
3980 }
3981 /**
3982  * e1000_set_mac - Change the Ethernet Address of the NIC
3983  * @netdev: network interface device structure
3984  * @p: pointer to an address structure
3985  *
3986  * Returns 0 on success, negative on failure
3987  **/
3988 static int e1000_set_mac(struct net_device *netdev, void *p)
3989 {
3990 	struct e1000_adapter *adapter = netdev_priv(netdev);
3991 	struct sockaddr *addr = p;
3992 
3993 	if (!is_valid_ether_addr(addr->sa_data))
3994 		return -EADDRNOTAVAIL;
3995 
3996 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3997 	memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3998 
3999 	e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
4000 
4001 	if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
4002 		/* activate the work around */
4003 		e1000e_set_laa_state_82571(&adapter->hw, 1);
4004 
4005 		/*
4006 		 * Hold a copy of the LAA in RAR[14] This is done so that
4007 		 * between the time RAR[0] gets clobbered  and the time it
4008 		 * gets fixed (in e1000_watchdog), the actual LAA is in one
4009 		 * of the RARs and no incoming packets directed to this port
4010 		 * are dropped. Eventually the LAA will be in RAR[0] and
4011 		 * RAR[14]
4012 		 */
4013 		e1000e_rar_set(&adapter->hw,
4014 			      adapter->hw.mac.addr,
4015 			      adapter->hw.mac.rar_entry_count - 1);
4016 	}
4017 
4018 	return 0;
4019 }
4020 
4021 /**
4022  * e1000e_update_phy_task - work thread to update phy
4023  * @work: pointer to our work struct
4024  *
4025  * this worker thread exists because we must acquire a
4026  * semaphore to read the phy, which we could msleep while
4027  * waiting for it, and we can't msleep in a timer.
4028  **/
4029 static void e1000e_update_phy_task(struct work_struct *work)
4030 {
4031 	struct e1000_adapter *adapter = container_of(work,
4032 					struct e1000_adapter, update_phy_task);
4033 
4034 	if (test_bit(__E1000_DOWN, &adapter->state))
4035 		return;
4036 
4037 	e1000_get_phy_info(&adapter->hw);
4038 }
4039 
4040 /*
4041  * Need to wait a few seconds after link up to get diagnostic information from
4042  * the phy
4043  */
4044 static void e1000_update_phy_info(unsigned long data)
4045 {
4046 	struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4047 
4048 	if (test_bit(__E1000_DOWN, &adapter->state))
4049 		return;
4050 
4051 	schedule_work(&adapter->update_phy_task);
4052 }
4053 
4054 /**
4055  * e1000e_update_phy_stats - Update the PHY statistics counters
4056  * @adapter: board private structure
4057  *
4058  * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4059  **/
4060 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4061 {
4062 	struct e1000_hw *hw = &adapter->hw;
4063 	s32 ret_val;
4064 	u16 phy_data;
4065 
4066 	ret_val = hw->phy.ops.acquire(hw);
4067 	if (ret_val)
4068 		return;
4069 
4070 	/*
4071 	 * A page set is expensive so check if already on desired page.
4072 	 * If not, set to the page with the PHY status registers.
4073 	 */
4074 	hw->phy.addr = 1;
4075 	ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4076 					   &phy_data);
4077 	if (ret_val)
4078 		goto release;
4079 	if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4080 		ret_val = hw->phy.ops.set_page(hw,
4081 					       HV_STATS_PAGE << IGP_PAGE_SHIFT);
4082 		if (ret_val)
4083 			goto release;
4084 	}
4085 
4086 	/* Single Collision Count */
4087 	hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4088 	ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4089 	if (!ret_val)
4090 		adapter->stats.scc += phy_data;
4091 
4092 	/* Excessive Collision Count */
4093 	hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4094 	ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4095 	if (!ret_val)
4096 		adapter->stats.ecol += phy_data;
4097 
4098 	/* Multiple Collision Count */
4099 	hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4100 	ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4101 	if (!ret_val)
4102 		adapter->stats.mcc += phy_data;
4103 
4104 	/* Late Collision Count */
4105 	hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4106 	ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4107 	if (!ret_val)
4108 		adapter->stats.latecol += phy_data;
4109 
4110 	/* Collision Count - also used for adaptive IFS */
4111 	hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4112 	ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4113 	if (!ret_val)
4114 		hw->mac.collision_delta = phy_data;
4115 
4116 	/* Defer Count */
4117 	hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4118 	ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4119 	if (!ret_val)
4120 		adapter->stats.dc += phy_data;
4121 
4122 	/* Transmit with no CRS */
4123 	hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4124 	ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4125 	if (!ret_val)
4126 		adapter->stats.tncrs += phy_data;
4127 
4128 release:
4129 	hw->phy.ops.release(hw);
4130 }
4131 
4132 /**
4133  * e1000e_update_stats - Update the board statistics counters
4134  * @adapter: board private structure
4135  **/
4136 static void e1000e_update_stats(struct e1000_adapter *adapter)
4137 {
4138 	struct net_device *netdev = adapter->netdev;
4139 	struct e1000_hw *hw = &adapter->hw;
4140 	struct pci_dev *pdev = adapter->pdev;
4141 
4142 	/*
4143 	 * Prevent stats update while adapter is being reset, or if the pci
4144 	 * connection is down.
4145 	 */
4146 	if (adapter->link_speed == 0)
4147 		return;
4148 	if (pci_channel_offline(pdev))
4149 		return;
4150 
4151 	adapter->stats.crcerrs += er32(CRCERRS);
4152 	adapter->stats.gprc += er32(GPRC);
4153 	adapter->stats.gorc += er32(GORCL);
4154 	er32(GORCH); /* Clear gorc */
4155 	adapter->stats.bprc += er32(BPRC);
4156 	adapter->stats.mprc += er32(MPRC);
4157 	adapter->stats.roc += er32(ROC);
4158 
4159 	adapter->stats.mpc += er32(MPC);
4160 
4161 	/* Half-duplex statistics */
4162 	if (adapter->link_duplex == HALF_DUPLEX) {
4163 		if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4164 			e1000e_update_phy_stats(adapter);
4165 		} else {
4166 			adapter->stats.scc += er32(SCC);
4167 			adapter->stats.ecol += er32(ECOL);
4168 			adapter->stats.mcc += er32(MCC);
4169 			adapter->stats.latecol += er32(LATECOL);
4170 			adapter->stats.dc += er32(DC);
4171 
4172 			hw->mac.collision_delta = er32(COLC);
4173 
4174 			if ((hw->mac.type != e1000_82574) &&
4175 			    (hw->mac.type != e1000_82583))
4176 				adapter->stats.tncrs += er32(TNCRS);
4177 		}
4178 		adapter->stats.colc += hw->mac.collision_delta;
4179 	}
4180 
4181 	adapter->stats.xonrxc += er32(XONRXC);
4182 	adapter->stats.xontxc += er32(XONTXC);
4183 	adapter->stats.xoffrxc += er32(XOFFRXC);
4184 	adapter->stats.xofftxc += er32(XOFFTXC);
4185 	adapter->stats.gptc += er32(GPTC);
4186 	adapter->stats.gotc += er32(GOTCL);
4187 	er32(GOTCH); /* Clear gotc */
4188 	adapter->stats.rnbc += er32(RNBC);
4189 	adapter->stats.ruc += er32(RUC);
4190 
4191 	adapter->stats.mptc += er32(MPTC);
4192 	adapter->stats.bptc += er32(BPTC);
4193 
4194 	/* used for adaptive IFS */
4195 
4196 	hw->mac.tx_packet_delta = er32(TPT);
4197 	adapter->stats.tpt += hw->mac.tx_packet_delta;
4198 
4199 	adapter->stats.algnerrc += er32(ALGNERRC);
4200 	adapter->stats.rxerrc += er32(RXERRC);
4201 	adapter->stats.cexterr += er32(CEXTERR);
4202 	adapter->stats.tsctc += er32(TSCTC);
4203 	adapter->stats.tsctfc += er32(TSCTFC);
4204 
4205 	/* Fill out the OS statistics structure */
4206 	netdev->stats.multicast = adapter->stats.mprc;
4207 	netdev->stats.collisions = adapter->stats.colc;
4208 
4209 	/* Rx Errors */
4210 
4211 	/*
4212 	 * RLEC on some newer hardware can be incorrect so build
4213 	 * our own version based on RUC and ROC
4214 	 */
4215 	netdev->stats.rx_errors = adapter->stats.rxerrc +
4216 		adapter->stats.crcerrs + adapter->stats.algnerrc +
4217 		adapter->stats.ruc + adapter->stats.roc +
4218 		adapter->stats.cexterr;
4219 	netdev->stats.rx_length_errors = adapter->stats.ruc +
4220 					      adapter->stats.roc;
4221 	netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4222 	netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4223 	netdev->stats.rx_missed_errors = adapter->stats.mpc;
4224 
4225 	/* Tx Errors */
4226 	netdev->stats.tx_errors = adapter->stats.ecol +
4227 				       adapter->stats.latecol;
4228 	netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4229 	netdev->stats.tx_window_errors = adapter->stats.latecol;
4230 	netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4231 
4232 	/* Tx Dropped needs to be maintained elsewhere */
4233 
4234 	/* Management Stats */
4235 	adapter->stats.mgptc += er32(MGTPTC);
4236 	adapter->stats.mgprc += er32(MGTPRC);
4237 	adapter->stats.mgpdc += er32(MGTPDC);
4238 }
4239 
4240 /**
4241  * e1000_phy_read_status - Update the PHY register status snapshot
4242  * @adapter: board private structure
4243  **/
4244 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4245 {
4246 	struct e1000_hw *hw = &adapter->hw;
4247 	struct e1000_phy_regs *phy = &adapter->phy_regs;
4248 
4249 	if ((er32(STATUS) & E1000_STATUS_LU) &&
4250 	    (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4251 		int ret_val;
4252 
4253 		ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4254 		ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4255 		ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4256 		ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4257 		ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4258 		ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4259 		ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4260 		ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4261 		if (ret_val)
4262 			e_warn("Error reading PHY register\n");
4263 	} else {
4264 		/*
4265 		 * Do not read PHY registers if link is not up
4266 		 * Set values to typical power-on defaults
4267 		 */
4268 		phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4269 		phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4270 			     BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4271 			     BMSR_ERCAP);
4272 		phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4273 				  ADVERTISE_ALL | ADVERTISE_CSMA);
4274 		phy->lpa = 0;
4275 		phy->expansion = EXPANSION_ENABLENPAGE;
4276 		phy->ctrl1000 = ADVERTISE_1000FULL;
4277 		phy->stat1000 = 0;
4278 		phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4279 	}
4280 }
4281 
4282 static void e1000_print_link_info(struct e1000_adapter *adapter)
4283 {
4284 	struct e1000_hw *hw = &adapter->hw;
4285 	u32 ctrl = er32(CTRL);
4286 
4287 	/* Link status message must follow this format for user tools */
4288 	printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4289 		adapter->netdev->name,
4290 		adapter->link_speed,
4291 		adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
4292 		(ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
4293 		(ctrl & E1000_CTRL_RFCE) ? "Rx" :
4294 		(ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
4295 }
4296 
4297 static bool e1000e_has_link(struct e1000_adapter *adapter)
4298 {
4299 	struct e1000_hw *hw = &adapter->hw;
4300 	bool link_active = false;
4301 	s32 ret_val = 0;
4302 
4303 	/*
4304 	 * get_link_status is set on LSC (link status) interrupt or
4305 	 * Rx sequence error interrupt.  get_link_status will stay
4306 	 * false until the check_for_link establishes link
4307 	 * for copper adapters ONLY
4308 	 */
4309 	switch (hw->phy.media_type) {
4310 	case e1000_media_type_copper:
4311 		if (hw->mac.get_link_status) {
4312 			ret_val = hw->mac.ops.check_for_link(hw);
4313 			link_active = !hw->mac.get_link_status;
4314 		} else {
4315 			link_active = true;
4316 		}
4317 		break;
4318 	case e1000_media_type_fiber:
4319 		ret_val = hw->mac.ops.check_for_link(hw);
4320 		link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4321 		break;
4322 	case e1000_media_type_internal_serdes:
4323 		ret_val = hw->mac.ops.check_for_link(hw);
4324 		link_active = adapter->hw.mac.serdes_has_link;
4325 		break;
4326 	default:
4327 	case e1000_media_type_unknown:
4328 		break;
4329 	}
4330 
4331 	if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4332 	    (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4333 		/* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4334 		e_info("Gigabit has been disabled, downgrading speed\n");
4335 	}
4336 
4337 	return link_active;
4338 }
4339 
4340 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4341 {
4342 	/* make sure the receive unit is started */
4343 	if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4344 	    (adapter->flags & FLAG_RX_RESTART_NOW)) {
4345 		struct e1000_hw *hw = &adapter->hw;
4346 		u32 rctl = er32(RCTL);
4347 		ew32(RCTL, rctl | E1000_RCTL_EN);
4348 		adapter->flags &= ~FLAG_RX_RESTART_NOW;
4349 	}
4350 }
4351 
4352 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4353 {
4354 	struct e1000_hw *hw = &adapter->hw;
4355 
4356 	/*
4357 	 * With 82574 controllers, PHY needs to be checked periodically
4358 	 * for hung state and reset, if two calls return true
4359 	 */
4360 	if (e1000_check_phy_82574(hw))
4361 		adapter->phy_hang_count++;
4362 	else
4363 		adapter->phy_hang_count = 0;
4364 
4365 	if (adapter->phy_hang_count > 1) {
4366 		adapter->phy_hang_count = 0;
4367 		schedule_work(&adapter->reset_task);
4368 	}
4369 }
4370 
4371 /**
4372  * e1000_watchdog - Timer Call-back
4373  * @data: pointer to adapter cast into an unsigned long
4374  **/
4375 static void e1000_watchdog(unsigned long data)
4376 {
4377 	struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4378 
4379 	/* Do the rest outside of interrupt context */
4380 	schedule_work(&adapter->watchdog_task);
4381 
4382 	/* TODO: make this use queue_delayed_work() */
4383 }
4384 
4385 static void e1000_watchdog_task(struct work_struct *work)
4386 {
4387 	struct e1000_adapter *adapter = container_of(work,
4388 					struct e1000_adapter, watchdog_task);
4389 	struct net_device *netdev = adapter->netdev;
4390 	struct e1000_mac_info *mac = &adapter->hw.mac;
4391 	struct e1000_phy_info *phy = &adapter->hw.phy;
4392 	struct e1000_ring *tx_ring = adapter->tx_ring;
4393 	struct e1000_hw *hw = &adapter->hw;
4394 	u32 link, tctl;
4395 
4396 	if (test_bit(__E1000_DOWN, &adapter->state))
4397 		return;
4398 
4399 	link = e1000e_has_link(adapter);
4400 	if ((netif_carrier_ok(netdev)) && link) {
4401 		/* Cancel scheduled suspend requests. */
4402 		pm_runtime_resume(netdev->dev.parent);
4403 
4404 		e1000e_enable_receives(adapter);
4405 		goto link_up;
4406 	}
4407 
4408 	if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4409 	    (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4410 		e1000_update_mng_vlan(adapter);
4411 
4412 	if (link) {
4413 		if (!netif_carrier_ok(netdev)) {
4414 			bool txb2b = true;
4415 
4416 			/* Cancel scheduled suspend requests. */
4417 			pm_runtime_resume(netdev->dev.parent);
4418 
4419 			/* update snapshot of PHY registers on LSC */
4420 			e1000_phy_read_status(adapter);
4421 			mac->ops.get_link_up_info(&adapter->hw,
4422 						   &adapter->link_speed,
4423 						   &adapter->link_duplex);
4424 			e1000_print_link_info(adapter);
4425 			/*
4426 			 * On supported PHYs, check for duplex mismatch only
4427 			 * if link has autonegotiated at 10/100 half
4428 			 */
4429 			if ((hw->phy.type == e1000_phy_igp_3 ||
4430 			     hw->phy.type == e1000_phy_bm) &&
4431 			    (hw->mac.autoneg == true) &&
4432 			    (adapter->link_speed == SPEED_10 ||
4433 			     adapter->link_speed == SPEED_100) &&
4434 			    (adapter->link_duplex == HALF_DUPLEX)) {
4435 				u16 autoneg_exp;
4436 
4437 				e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4438 
4439 				if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4440 					e_info("Autonegotiated half duplex but link partner cannot autoneg.  Try forcing full duplex if link gets many collisions.\n");
4441 			}
4442 
4443 			/* adjust timeout factor according to speed/duplex */
4444 			adapter->tx_timeout_factor = 1;
4445 			switch (adapter->link_speed) {
4446 			case SPEED_10:
4447 				txb2b = false;
4448 				adapter->tx_timeout_factor = 16;
4449 				break;
4450 			case SPEED_100:
4451 				txb2b = false;
4452 				adapter->tx_timeout_factor = 10;
4453 				break;
4454 			}
4455 
4456 			/*
4457 			 * workaround: re-program speed mode bit after
4458 			 * link-up event
4459 			 */
4460 			if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4461 			    !txb2b) {
4462 				u32 tarc0;
4463 				tarc0 = er32(TARC(0));
4464 				tarc0 &= ~SPEED_MODE_BIT;
4465 				ew32(TARC(0), tarc0);
4466 			}
4467 
4468 			/*
4469 			 * disable TSO for pcie and 10/100 speeds, to avoid
4470 			 * some hardware issues
4471 			 */
4472 			if (!(adapter->flags & FLAG_TSO_FORCE)) {
4473 				switch (adapter->link_speed) {
4474 				case SPEED_10:
4475 				case SPEED_100:
4476 					e_info("10/100 speed: disabling TSO\n");
4477 					netdev->features &= ~NETIF_F_TSO;
4478 					netdev->features &= ~NETIF_F_TSO6;
4479 					break;
4480 				case SPEED_1000:
4481 					netdev->features |= NETIF_F_TSO;
4482 					netdev->features |= NETIF_F_TSO6;
4483 					break;
4484 				default:
4485 					/* oops */
4486 					break;
4487 				}
4488 			}
4489 
4490 			/*
4491 			 * enable transmits in the hardware, need to do this
4492 			 * after setting TARC(0)
4493 			 */
4494 			tctl = er32(TCTL);
4495 			tctl |= E1000_TCTL_EN;
4496 			ew32(TCTL, tctl);
4497 
4498                         /*
4499 			 * Perform any post-link-up configuration before
4500 			 * reporting link up.
4501 			 */
4502 			if (phy->ops.cfg_on_link_up)
4503 				phy->ops.cfg_on_link_up(hw);
4504 
4505 			netif_carrier_on(netdev);
4506 
4507 			if (!test_bit(__E1000_DOWN, &adapter->state))
4508 				mod_timer(&adapter->phy_info_timer,
4509 					  round_jiffies(jiffies + 2 * HZ));
4510 		}
4511 	} else {
4512 		if (netif_carrier_ok(netdev)) {
4513 			adapter->link_speed = 0;
4514 			adapter->link_duplex = 0;
4515 			/* Link status message must follow this format */
4516 			printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4517 			       adapter->netdev->name);
4518 			netif_carrier_off(netdev);
4519 			if (!test_bit(__E1000_DOWN, &adapter->state))
4520 				mod_timer(&adapter->phy_info_timer,
4521 					  round_jiffies(jiffies + 2 * HZ));
4522 
4523 			if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4524 				schedule_work(&adapter->reset_task);
4525 			else
4526 				pm_schedule_suspend(netdev->dev.parent,
4527 							LINK_TIMEOUT);
4528 		}
4529 	}
4530 
4531 link_up:
4532 	spin_lock(&adapter->stats64_lock);
4533 	e1000e_update_stats(adapter);
4534 
4535 	mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4536 	adapter->tpt_old = adapter->stats.tpt;
4537 	mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4538 	adapter->colc_old = adapter->stats.colc;
4539 
4540 	adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4541 	adapter->gorc_old = adapter->stats.gorc;
4542 	adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4543 	adapter->gotc_old = adapter->stats.gotc;
4544 	spin_unlock(&adapter->stats64_lock);
4545 
4546 	e1000e_update_adaptive(&adapter->hw);
4547 
4548 	if (!netif_carrier_ok(netdev) &&
4549 	    (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4550 		/*
4551 		 * We've lost link, so the controller stops DMA,
4552 		 * but we've got queued Tx work that's never going
4553 		 * to get done, so reset controller to flush Tx.
4554 		 * (Do the reset outside of interrupt context).
4555 		 */
4556 		schedule_work(&adapter->reset_task);
4557 		/* return immediately since reset is imminent */
4558 		return;
4559 	}
4560 
4561 	/* Simple mode for Interrupt Throttle Rate (ITR) */
4562 	if (adapter->itr_setting == 4) {
4563 		/*
4564 		 * Symmetric Tx/Rx gets a reduced ITR=2000;
4565 		 * Total asymmetrical Tx or Rx gets ITR=8000;
4566 		 * everyone else is between 2000-8000.
4567 		 */
4568 		u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4569 		u32 dif = (adapter->gotc > adapter->gorc ?
4570 			    adapter->gotc - adapter->gorc :
4571 			    adapter->gorc - adapter->gotc) / 10000;
4572 		u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4573 
4574 		ew32(ITR, 1000000000 / (itr * 256));
4575 	}
4576 
4577 	/* Cause software interrupt to ensure Rx ring is cleaned */
4578 	if (adapter->msix_entries)
4579 		ew32(ICS, adapter->rx_ring->ims_val);
4580 	else
4581 		ew32(ICS, E1000_ICS_RXDMT0);
4582 
4583 	/* flush pending descriptors to memory before detecting Tx hang */
4584 	e1000e_flush_descriptors(adapter);
4585 
4586 	/* Force detection of hung controller every watchdog period */
4587 	adapter->detect_tx_hung = true;
4588 
4589 	/*
4590 	 * With 82571 controllers, LAA may be overwritten due to controller
4591 	 * reset from the other port. Set the appropriate LAA in RAR[0]
4592 	 */
4593 	if (e1000e_get_laa_state_82571(hw))
4594 		e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4595 
4596 	if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4597 		e1000e_check_82574_phy_workaround(adapter);
4598 
4599 	/* Reset the timer */
4600 	if (!test_bit(__E1000_DOWN, &adapter->state))
4601 		mod_timer(&adapter->watchdog_timer,
4602 			  round_jiffies(jiffies + 2 * HZ));
4603 }
4604 
4605 #define E1000_TX_FLAGS_CSUM		0x00000001
4606 #define E1000_TX_FLAGS_VLAN		0x00000002
4607 #define E1000_TX_FLAGS_TSO		0x00000004
4608 #define E1000_TX_FLAGS_IPV4		0x00000008
4609 #define E1000_TX_FLAGS_VLAN_MASK	0xffff0000
4610 #define E1000_TX_FLAGS_VLAN_SHIFT	16
4611 
4612 static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb)
4613 {
4614 	struct e1000_context_desc *context_desc;
4615 	struct e1000_buffer *buffer_info;
4616 	unsigned int i;
4617 	u32 cmd_length = 0;
4618 	u16 ipcse = 0, tucse, mss;
4619 	u8 ipcss, ipcso, tucss, tucso, hdr_len;
4620 
4621 	if (!skb_is_gso(skb))
4622 		return 0;
4623 
4624 	if (skb_header_cloned(skb)) {
4625 		int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4626 
4627 		if (err)
4628 			return err;
4629 	}
4630 
4631 	hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4632 	mss = skb_shinfo(skb)->gso_size;
4633 	if (skb->protocol == htons(ETH_P_IP)) {
4634 		struct iphdr *iph = ip_hdr(skb);
4635 		iph->tot_len = 0;
4636 		iph->check = 0;
4637 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4638 		                                         0, IPPROTO_TCP, 0);
4639 		cmd_length = E1000_TXD_CMD_IP;
4640 		ipcse = skb_transport_offset(skb) - 1;
4641 	} else if (skb_is_gso_v6(skb)) {
4642 		ipv6_hdr(skb)->payload_len = 0;
4643 		tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4644 		                                       &ipv6_hdr(skb)->daddr,
4645 		                                       0, IPPROTO_TCP, 0);
4646 		ipcse = 0;
4647 	}
4648 	ipcss = skb_network_offset(skb);
4649 	ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4650 	tucss = skb_transport_offset(skb);
4651 	tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4652 	tucse = 0;
4653 
4654 	cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4655 	               E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4656 
4657 	i = tx_ring->next_to_use;
4658 	context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4659 	buffer_info = &tx_ring->buffer_info[i];
4660 
4661 	context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4662 	context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4663 	context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4664 	context_desc->upper_setup.tcp_fields.tucss = tucss;
4665 	context_desc->upper_setup.tcp_fields.tucso = tucso;
4666 	context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4667 	context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4668 	context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4669 	context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4670 
4671 	buffer_info->time_stamp = jiffies;
4672 	buffer_info->next_to_watch = i;
4673 
4674 	i++;
4675 	if (i == tx_ring->count)
4676 		i = 0;
4677 	tx_ring->next_to_use = i;
4678 
4679 	return 1;
4680 }
4681 
4682 static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb)
4683 {
4684 	struct e1000_adapter *adapter = tx_ring->adapter;
4685 	struct e1000_context_desc *context_desc;
4686 	struct e1000_buffer *buffer_info;
4687 	unsigned int i;
4688 	u8 css;
4689 	u32 cmd_len = E1000_TXD_CMD_DEXT;
4690 	__be16 protocol;
4691 
4692 	if (skb->ip_summed != CHECKSUM_PARTIAL)
4693 		return 0;
4694 
4695 	if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4696 		protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4697 	else
4698 		protocol = skb->protocol;
4699 
4700 	switch (protocol) {
4701 	case cpu_to_be16(ETH_P_IP):
4702 		if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4703 			cmd_len |= E1000_TXD_CMD_TCP;
4704 		break;
4705 	case cpu_to_be16(ETH_P_IPV6):
4706 		/* XXX not handling all IPV6 headers */
4707 		if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4708 			cmd_len |= E1000_TXD_CMD_TCP;
4709 		break;
4710 	default:
4711 		if (unlikely(net_ratelimit()))
4712 			e_warn("checksum_partial proto=%x!\n",
4713 			       be16_to_cpu(protocol));
4714 		break;
4715 	}
4716 
4717 	css = skb_checksum_start_offset(skb);
4718 
4719 	i = tx_ring->next_to_use;
4720 	buffer_info = &tx_ring->buffer_info[i];
4721 	context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4722 
4723 	context_desc->lower_setup.ip_config = 0;
4724 	context_desc->upper_setup.tcp_fields.tucss = css;
4725 	context_desc->upper_setup.tcp_fields.tucso =
4726 				css + skb->csum_offset;
4727 	context_desc->upper_setup.tcp_fields.tucse = 0;
4728 	context_desc->tcp_seg_setup.data = 0;
4729 	context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4730 
4731 	buffer_info->time_stamp = jiffies;
4732 	buffer_info->next_to_watch = i;
4733 
4734 	i++;
4735 	if (i == tx_ring->count)
4736 		i = 0;
4737 	tx_ring->next_to_use = i;
4738 
4739 	return 1;
4740 }
4741 
4742 #define E1000_MAX_PER_TXD	8192
4743 #define E1000_MAX_TXD_PWR	12
4744 
4745 static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
4746 			unsigned int first, unsigned int max_per_txd,
4747 			unsigned int nr_frags, unsigned int mss)
4748 {
4749 	struct e1000_adapter *adapter = tx_ring->adapter;
4750 	struct pci_dev *pdev = adapter->pdev;
4751 	struct e1000_buffer *buffer_info;
4752 	unsigned int len = skb_headlen(skb);
4753 	unsigned int offset = 0, size, count = 0, i;
4754 	unsigned int f, bytecount, segs;
4755 
4756 	i = tx_ring->next_to_use;
4757 
4758 	while (len) {
4759 		buffer_info = &tx_ring->buffer_info[i];
4760 		size = min(len, max_per_txd);
4761 
4762 		buffer_info->length = size;
4763 		buffer_info->time_stamp = jiffies;
4764 		buffer_info->next_to_watch = i;
4765 		buffer_info->dma = dma_map_single(&pdev->dev,
4766 						  skb->data + offset,
4767 						  size, DMA_TO_DEVICE);
4768 		buffer_info->mapped_as_page = false;
4769 		if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4770 			goto dma_error;
4771 
4772 		len -= size;
4773 		offset += size;
4774 		count++;
4775 
4776 		if (len) {
4777 			i++;
4778 			if (i == tx_ring->count)
4779 				i = 0;
4780 		}
4781 	}
4782 
4783 	for (f = 0; f < nr_frags; f++) {
4784 		const struct skb_frag_struct *frag;
4785 
4786 		frag = &skb_shinfo(skb)->frags[f];
4787 		len = skb_frag_size(frag);
4788 		offset = 0;
4789 
4790 		while (len) {
4791 			i++;
4792 			if (i == tx_ring->count)
4793 				i = 0;
4794 
4795 			buffer_info = &tx_ring->buffer_info[i];
4796 			size = min(len, max_per_txd);
4797 
4798 			buffer_info->length = size;
4799 			buffer_info->time_stamp = jiffies;
4800 			buffer_info->next_to_watch = i;
4801 			buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
4802 						offset, size, DMA_TO_DEVICE);
4803 			buffer_info->mapped_as_page = true;
4804 			if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4805 				goto dma_error;
4806 
4807 			len -= size;
4808 			offset += size;
4809 			count++;
4810 		}
4811 	}
4812 
4813 	segs = skb_shinfo(skb)->gso_segs ? : 1;
4814 	/* multiply data chunks by size of headers */
4815 	bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4816 
4817 	tx_ring->buffer_info[i].skb = skb;
4818 	tx_ring->buffer_info[i].segs = segs;
4819 	tx_ring->buffer_info[i].bytecount = bytecount;
4820 	tx_ring->buffer_info[first].next_to_watch = i;
4821 
4822 	return count;
4823 
4824 dma_error:
4825 	dev_err(&pdev->dev, "Tx DMA map failed\n");
4826 	buffer_info->dma = 0;
4827 	if (count)
4828 		count--;
4829 
4830 	while (count--) {
4831 		if (i == 0)
4832 			i += tx_ring->count;
4833 		i--;
4834 		buffer_info = &tx_ring->buffer_info[i];
4835 		e1000_put_txbuf(tx_ring, buffer_info);
4836 	}
4837 
4838 	return 0;
4839 }
4840 
4841 static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
4842 {
4843 	struct e1000_adapter *adapter = tx_ring->adapter;
4844 	struct e1000_tx_desc *tx_desc = NULL;
4845 	struct e1000_buffer *buffer_info;
4846 	u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4847 	unsigned int i;
4848 
4849 	if (tx_flags & E1000_TX_FLAGS_TSO) {
4850 		txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4851 			     E1000_TXD_CMD_TSE;
4852 		txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4853 
4854 		if (tx_flags & E1000_TX_FLAGS_IPV4)
4855 			txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4856 	}
4857 
4858 	if (tx_flags & E1000_TX_FLAGS_CSUM) {
4859 		txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4860 		txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4861 	}
4862 
4863 	if (tx_flags & E1000_TX_FLAGS_VLAN) {
4864 		txd_lower |= E1000_TXD_CMD_VLE;
4865 		txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4866 	}
4867 
4868 	i = tx_ring->next_to_use;
4869 
4870 	do {
4871 		buffer_info = &tx_ring->buffer_info[i];
4872 		tx_desc = E1000_TX_DESC(*tx_ring, i);
4873 		tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4874 		tx_desc->lower.data =
4875 			cpu_to_le32(txd_lower | buffer_info->length);
4876 		tx_desc->upper.data = cpu_to_le32(txd_upper);
4877 
4878 		i++;
4879 		if (i == tx_ring->count)
4880 			i = 0;
4881 	} while (--count > 0);
4882 
4883 	tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4884 
4885 	/*
4886 	 * Force memory writes to complete before letting h/w
4887 	 * know there are new descriptors to fetch.  (Only
4888 	 * applicable for weak-ordered memory model archs,
4889 	 * such as IA-64).
4890 	 */
4891 	wmb();
4892 
4893 	tx_ring->next_to_use = i;
4894 
4895 	if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
4896 		e1000e_update_tdt_wa(tx_ring, i);
4897 	else
4898 		writel(i, tx_ring->tail);
4899 
4900 	/*
4901 	 * we need this if more than one processor can write to our tail
4902 	 * at a time, it synchronizes IO on IA64/Altix systems
4903 	 */
4904 	mmiowb();
4905 }
4906 
4907 #define MINIMUM_DHCP_PACKET_SIZE 282
4908 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4909 				    struct sk_buff *skb)
4910 {
4911 	struct e1000_hw *hw =  &adapter->hw;
4912 	u16 length, offset;
4913 
4914 	if (vlan_tx_tag_present(skb)) {
4915 		if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4916 		    (adapter->hw.mng_cookie.status &
4917 			E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4918 			return 0;
4919 	}
4920 
4921 	if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4922 		return 0;
4923 
4924 	if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4925 		return 0;
4926 
4927 	{
4928 		const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4929 		struct udphdr *udp;
4930 
4931 		if (ip->protocol != IPPROTO_UDP)
4932 			return 0;
4933 
4934 		udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4935 		if (ntohs(udp->dest) != 67)
4936 			return 0;
4937 
4938 		offset = (u8 *)udp + 8 - skb->data;
4939 		length = skb->len - offset;
4940 		return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4941 	}
4942 
4943 	return 0;
4944 }
4945 
4946 static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
4947 {
4948 	struct e1000_adapter *adapter = tx_ring->adapter;
4949 
4950 	netif_stop_queue(adapter->netdev);
4951 	/*
4952 	 * Herbert's original patch had:
4953 	 *  smp_mb__after_netif_stop_queue();
4954 	 * but since that doesn't exist yet, just open code it.
4955 	 */
4956 	smp_mb();
4957 
4958 	/*
4959 	 * We need to check again in a case another CPU has just
4960 	 * made room available.
4961 	 */
4962 	if (e1000_desc_unused(tx_ring) < size)
4963 		return -EBUSY;
4964 
4965 	/* A reprieve! */
4966 	netif_start_queue(adapter->netdev);
4967 	++adapter->restart_queue;
4968 	return 0;
4969 }
4970 
4971 static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
4972 {
4973 	if (e1000_desc_unused(tx_ring) >= size)
4974 		return 0;
4975 	return __e1000_maybe_stop_tx(tx_ring, size);
4976 }
4977 
4978 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1)
4979 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4980 				    struct net_device *netdev)
4981 {
4982 	struct e1000_adapter *adapter = netdev_priv(netdev);
4983 	struct e1000_ring *tx_ring = adapter->tx_ring;
4984 	unsigned int first;
4985 	unsigned int max_per_txd = E1000_MAX_PER_TXD;
4986 	unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4987 	unsigned int tx_flags = 0;
4988 	unsigned int len = skb_headlen(skb);
4989 	unsigned int nr_frags;
4990 	unsigned int mss;
4991 	int count = 0;
4992 	int tso;
4993 	unsigned int f;
4994 
4995 	if (test_bit(__E1000_DOWN, &adapter->state)) {
4996 		dev_kfree_skb_any(skb);
4997 		return NETDEV_TX_OK;
4998 	}
4999 
5000 	if (skb->len <= 0) {
5001 		dev_kfree_skb_any(skb);
5002 		return NETDEV_TX_OK;
5003 	}
5004 
5005 	mss = skb_shinfo(skb)->gso_size;
5006 	/*
5007 	 * The controller does a simple calculation to
5008 	 * make sure there is enough room in the FIFO before
5009 	 * initiating the DMA for each buffer.  The calc is:
5010 	 * 4 = ceil(buffer len/mss).  To make sure we don't
5011 	 * overrun the FIFO, adjust the max buffer len if mss
5012 	 * drops.
5013 	 */
5014 	if (mss) {
5015 		u8 hdr_len;
5016 		max_per_txd = min(mss << 2, max_per_txd);
5017 		max_txd_pwr = fls(max_per_txd) - 1;
5018 
5019 		/*
5020 		 * TSO Workaround for 82571/2/3 Controllers -- if skb->data
5021 		 * points to just header, pull a few bytes of payload from
5022 		 * frags into skb->data
5023 		 */
5024 		hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5025 		/*
5026 		 * we do this workaround for ES2LAN, but it is un-necessary,
5027 		 * avoiding it could save a lot of cycles
5028 		 */
5029 		if (skb->data_len && (hdr_len == len)) {
5030 			unsigned int pull_size;
5031 
5032 			pull_size = min_t(unsigned int, 4, skb->data_len);
5033 			if (!__pskb_pull_tail(skb, pull_size)) {
5034 				e_err("__pskb_pull_tail failed.\n");
5035 				dev_kfree_skb_any(skb);
5036 				return NETDEV_TX_OK;
5037 			}
5038 			len = skb_headlen(skb);
5039 		}
5040 	}
5041 
5042 	/* reserve a descriptor for the offload context */
5043 	if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5044 		count++;
5045 	count++;
5046 
5047 	count += TXD_USE_COUNT(len, max_txd_pwr);
5048 
5049 	nr_frags = skb_shinfo(skb)->nr_frags;
5050 	for (f = 0; f < nr_frags; f++)
5051 		count += TXD_USE_COUNT(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5052 				       max_txd_pwr);
5053 
5054 	if (adapter->hw.mac.tx_pkt_filtering)
5055 		e1000_transfer_dhcp_info(adapter, skb);
5056 
5057 	/*
5058 	 * need: count + 2 desc gap to keep tail from touching
5059 	 * head, otherwise try next time
5060 	 */
5061 	if (e1000_maybe_stop_tx(tx_ring, count + 2))
5062 		return NETDEV_TX_BUSY;
5063 
5064 	if (vlan_tx_tag_present(skb)) {
5065 		tx_flags |= E1000_TX_FLAGS_VLAN;
5066 		tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
5067 	}
5068 
5069 	first = tx_ring->next_to_use;
5070 
5071 	tso = e1000_tso(tx_ring, skb);
5072 	if (tso < 0) {
5073 		dev_kfree_skb_any(skb);
5074 		return NETDEV_TX_OK;
5075 	}
5076 
5077 	if (tso)
5078 		tx_flags |= E1000_TX_FLAGS_TSO;
5079 	else if (e1000_tx_csum(tx_ring, skb))
5080 		tx_flags |= E1000_TX_FLAGS_CSUM;
5081 
5082 	/*
5083 	 * Old method was to assume IPv4 packet by default if TSO was enabled.
5084 	 * 82571 hardware supports TSO capabilities for IPv6 as well...
5085 	 * no longer assume, we must.
5086 	 */
5087 	if (skb->protocol == htons(ETH_P_IP))
5088 		tx_flags |= E1000_TX_FLAGS_IPV4;
5089 
5090 	/* if count is 0 then mapping error has occurred */
5091 	count = e1000_tx_map(tx_ring, skb, first, max_per_txd, nr_frags, mss);
5092 	if (count) {
5093 		netdev_sent_queue(netdev, skb->len);
5094 		e1000_tx_queue(tx_ring, tx_flags, count);
5095 		/* Make sure there is space in the ring for the next send. */
5096 		e1000_maybe_stop_tx(tx_ring, MAX_SKB_FRAGS + 2);
5097 
5098 	} else {
5099 		dev_kfree_skb_any(skb);
5100 		tx_ring->buffer_info[first].time_stamp = 0;
5101 		tx_ring->next_to_use = first;
5102 	}
5103 
5104 	return NETDEV_TX_OK;
5105 }
5106 
5107 /**
5108  * e1000_tx_timeout - Respond to a Tx Hang
5109  * @netdev: network interface device structure
5110  **/
5111 static void e1000_tx_timeout(struct net_device *netdev)
5112 {
5113 	struct e1000_adapter *adapter = netdev_priv(netdev);
5114 
5115 	/* Do the reset outside of interrupt context */
5116 	adapter->tx_timeout_count++;
5117 	schedule_work(&adapter->reset_task);
5118 }
5119 
5120 static void e1000_reset_task(struct work_struct *work)
5121 {
5122 	struct e1000_adapter *adapter;
5123 	adapter = container_of(work, struct e1000_adapter, reset_task);
5124 
5125 	/* don't run the task if already down */
5126 	if (test_bit(__E1000_DOWN, &adapter->state))
5127 		return;
5128 
5129 	if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5130 	      (adapter->flags & FLAG_RX_RESTART_NOW))) {
5131 		e1000e_dump(adapter);
5132 		e_err("Reset adapter\n");
5133 	}
5134 	e1000e_reinit_locked(adapter);
5135 }
5136 
5137 /**
5138  * e1000_get_stats64 - Get System Network Statistics
5139  * @netdev: network interface device structure
5140  * @stats: rtnl_link_stats64 pointer
5141  *
5142  * Returns the address of the device statistics structure.
5143  **/
5144 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
5145                                              struct rtnl_link_stats64 *stats)
5146 {
5147 	struct e1000_adapter *adapter = netdev_priv(netdev);
5148 
5149 	memset(stats, 0, sizeof(struct rtnl_link_stats64));
5150 	spin_lock(&adapter->stats64_lock);
5151 	e1000e_update_stats(adapter);
5152 	/* Fill out the OS statistics structure */
5153 	stats->rx_bytes = adapter->stats.gorc;
5154 	stats->rx_packets = adapter->stats.gprc;
5155 	stats->tx_bytes = adapter->stats.gotc;
5156 	stats->tx_packets = adapter->stats.gptc;
5157 	stats->multicast = adapter->stats.mprc;
5158 	stats->collisions = adapter->stats.colc;
5159 
5160 	/* Rx Errors */
5161 
5162 	/*
5163 	 * RLEC on some newer hardware can be incorrect so build
5164 	 * our own version based on RUC and ROC
5165 	 */
5166 	stats->rx_errors = adapter->stats.rxerrc +
5167 		adapter->stats.crcerrs + adapter->stats.algnerrc +
5168 		adapter->stats.ruc + adapter->stats.roc +
5169 		adapter->stats.cexterr;
5170 	stats->rx_length_errors = adapter->stats.ruc +
5171 					      adapter->stats.roc;
5172 	stats->rx_crc_errors = adapter->stats.crcerrs;
5173 	stats->rx_frame_errors = adapter->stats.algnerrc;
5174 	stats->rx_missed_errors = adapter->stats.mpc;
5175 
5176 	/* Tx Errors */
5177 	stats->tx_errors = adapter->stats.ecol +
5178 				       adapter->stats.latecol;
5179 	stats->tx_aborted_errors = adapter->stats.ecol;
5180 	stats->tx_window_errors = adapter->stats.latecol;
5181 	stats->tx_carrier_errors = adapter->stats.tncrs;
5182 
5183 	/* Tx Dropped needs to be maintained elsewhere */
5184 
5185 	spin_unlock(&adapter->stats64_lock);
5186 	return stats;
5187 }
5188 
5189 /**
5190  * e1000_change_mtu - Change the Maximum Transfer Unit
5191  * @netdev: network interface device structure
5192  * @new_mtu: new value for maximum frame size
5193  *
5194  * Returns 0 on success, negative on failure
5195  **/
5196 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5197 {
5198 	struct e1000_adapter *adapter = netdev_priv(netdev);
5199 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5200 
5201 	/* Jumbo frame support */
5202 	if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
5203 		if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5204 			e_err("Jumbo Frames not supported.\n");
5205 			return -EINVAL;
5206 		}
5207 
5208 		/*
5209 		 * IP payload checksum (enabled with jumbos/packet-split when
5210 		 * Rx checksum is enabled) and generation of RSS hash is
5211 		 * mutually exclusive in the hardware.
5212 		 */
5213 		if ((netdev->features & NETIF_F_RXCSUM) &&
5214 		    (netdev->features & NETIF_F_RXHASH)) {
5215 			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");
5216 			return -EINVAL;
5217 		}
5218 	}
5219 
5220 	/* Supported frame sizes */
5221 	if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5222 	    (max_frame > adapter->max_hw_frame_size)) {
5223 		e_err("Unsupported MTU setting\n");
5224 		return -EINVAL;
5225 	}
5226 
5227 	/* Jumbo frame workaround on 82579 requires CRC be stripped */
5228 	if ((adapter->hw.mac.type == e1000_pch2lan) &&
5229 	    !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5230 	    (new_mtu > ETH_DATA_LEN)) {
5231 		e_err("Jumbo Frames not supported on 82579 when CRC stripping is disabled.\n");
5232 		return -EINVAL;
5233 	}
5234 
5235 	/* 82573 Errata 17 */
5236 	if (((adapter->hw.mac.type == e1000_82573) ||
5237 	     (adapter->hw.mac.type == e1000_82574)) &&
5238 	    (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
5239 		adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
5240 		e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
5241 	}
5242 
5243 	while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5244 		usleep_range(1000, 2000);
5245 	/* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5246 	adapter->max_frame_size = max_frame;
5247 	e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5248 	netdev->mtu = new_mtu;
5249 	if (netif_running(netdev))
5250 		e1000e_down(adapter);
5251 
5252 	/*
5253 	 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5254 	 * means we reserve 2 more, this pushes us to allocate from the next
5255 	 * larger slab size.
5256 	 * i.e. RXBUFFER_2048 --> size-4096 slab
5257 	 * However with the new *_jumbo_rx* routines, jumbo receives will use
5258 	 * fragmented skbs
5259 	 */
5260 
5261 	if (max_frame <= 2048)
5262 		adapter->rx_buffer_len = 2048;
5263 	else
5264 		adapter->rx_buffer_len = 4096;
5265 
5266 	/* adjust allocation if LPE protects us, and we aren't using SBP */
5267 	if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5268 	     (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5269 		adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5270 					 + ETH_FCS_LEN;
5271 
5272 	if (netif_running(netdev))
5273 		e1000e_up(adapter);
5274 	else
5275 		e1000e_reset(adapter);
5276 
5277 	clear_bit(__E1000_RESETTING, &adapter->state);
5278 
5279 	return 0;
5280 }
5281 
5282 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5283 			   int cmd)
5284 {
5285 	struct e1000_adapter *adapter = netdev_priv(netdev);
5286 	struct mii_ioctl_data *data = if_mii(ifr);
5287 
5288 	if (adapter->hw.phy.media_type != e1000_media_type_copper)
5289 		return -EOPNOTSUPP;
5290 
5291 	switch (cmd) {
5292 	case SIOCGMIIPHY:
5293 		data->phy_id = adapter->hw.phy.addr;
5294 		break;
5295 	case SIOCGMIIREG:
5296 		e1000_phy_read_status(adapter);
5297 
5298 		switch (data->reg_num & 0x1F) {
5299 		case MII_BMCR:
5300 			data->val_out = adapter->phy_regs.bmcr;
5301 			break;
5302 		case MII_BMSR:
5303 			data->val_out = adapter->phy_regs.bmsr;
5304 			break;
5305 		case MII_PHYSID1:
5306 			data->val_out = (adapter->hw.phy.id >> 16);
5307 			break;
5308 		case MII_PHYSID2:
5309 			data->val_out = (adapter->hw.phy.id & 0xFFFF);
5310 			break;
5311 		case MII_ADVERTISE:
5312 			data->val_out = adapter->phy_regs.advertise;
5313 			break;
5314 		case MII_LPA:
5315 			data->val_out = adapter->phy_regs.lpa;
5316 			break;
5317 		case MII_EXPANSION:
5318 			data->val_out = adapter->phy_regs.expansion;
5319 			break;
5320 		case MII_CTRL1000:
5321 			data->val_out = adapter->phy_regs.ctrl1000;
5322 			break;
5323 		case MII_STAT1000:
5324 			data->val_out = adapter->phy_regs.stat1000;
5325 			break;
5326 		case MII_ESTATUS:
5327 			data->val_out = adapter->phy_regs.estatus;
5328 			break;
5329 		default:
5330 			return -EIO;
5331 		}
5332 		break;
5333 	case SIOCSMIIREG:
5334 	default:
5335 		return -EOPNOTSUPP;
5336 	}
5337 	return 0;
5338 }
5339 
5340 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5341 {
5342 	switch (cmd) {
5343 	case SIOCGMIIPHY:
5344 	case SIOCGMIIREG:
5345 	case SIOCSMIIREG:
5346 		return e1000_mii_ioctl(netdev, ifr, cmd);
5347 	default:
5348 		return -EOPNOTSUPP;
5349 	}
5350 }
5351 
5352 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5353 {
5354 	struct e1000_hw *hw = &adapter->hw;
5355 	u32 i, mac_reg;
5356 	u16 phy_reg, wuc_enable;
5357 	int retval = 0;
5358 
5359 	/* copy MAC RARs to PHY RARs */
5360 	e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5361 
5362 	retval = hw->phy.ops.acquire(hw);
5363 	if (retval) {
5364 		e_err("Could not acquire PHY\n");
5365 		return retval;
5366 	}
5367 
5368 	/* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
5369 	retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5370 	if (retval)
5371 		goto out;
5372 
5373 	/* copy MAC MTA to PHY MTA - only needed for pchlan */
5374 	for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5375 		mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5376 		hw->phy.ops.write_reg_page(hw, BM_MTA(i),
5377 					   (u16)(mac_reg & 0xFFFF));
5378 		hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
5379 					   (u16)((mac_reg >> 16) & 0xFFFF));
5380 	}
5381 
5382 	/* configure PHY Rx Control register */
5383 	hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
5384 	mac_reg = er32(RCTL);
5385 	if (mac_reg & E1000_RCTL_UPE)
5386 		phy_reg |= BM_RCTL_UPE;
5387 	if (mac_reg & E1000_RCTL_MPE)
5388 		phy_reg |= BM_RCTL_MPE;
5389 	phy_reg &= ~(BM_RCTL_MO_MASK);
5390 	if (mac_reg & E1000_RCTL_MO_3)
5391 		phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5392 				<< BM_RCTL_MO_SHIFT);
5393 	if (mac_reg & E1000_RCTL_BAM)
5394 		phy_reg |= BM_RCTL_BAM;
5395 	if (mac_reg & E1000_RCTL_PMCF)
5396 		phy_reg |= BM_RCTL_PMCF;
5397 	mac_reg = er32(CTRL);
5398 	if (mac_reg & E1000_CTRL_RFCE)
5399 		phy_reg |= BM_RCTL_RFCE;
5400 	hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
5401 
5402 	/* enable PHY wakeup in MAC register */
5403 	ew32(WUFC, wufc);
5404 	ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5405 
5406 	/* configure and enable PHY wakeup in PHY registers */
5407 	hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
5408 	hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5409 
5410 	/* activate PHY wakeup */
5411 	wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5412 	retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5413 	if (retval)
5414 		e_err("Could not set PHY Host Wakeup bit\n");
5415 out:
5416 	hw->phy.ops.release(hw);
5417 
5418 	return retval;
5419 }
5420 
5421 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5422 			    bool runtime)
5423 {
5424 	struct net_device *netdev = pci_get_drvdata(pdev);
5425 	struct e1000_adapter *adapter = netdev_priv(netdev);
5426 	struct e1000_hw *hw = &adapter->hw;
5427 	u32 ctrl, ctrl_ext, rctl, status;
5428 	/* Runtime suspend should only enable wakeup for link changes */
5429 	u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5430 	int retval = 0;
5431 
5432 	netif_device_detach(netdev);
5433 
5434 	if (netif_running(netdev)) {
5435 		WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5436 		e1000e_down(adapter);
5437 		e1000_free_irq(adapter);
5438 	}
5439 	e1000e_reset_interrupt_capability(adapter);
5440 
5441 	retval = pci_save_state(pdev);
5442 	if (retval)
5443 		return retval;
5444 
5445 	status = er32(STATUS);
5446 	if (status & E1000_STATUS_LU)
5447 		wufc &= ~E1000_WUFC_LNKC;
5448 
5449 	if (wufc) {
5450 		e1000_setup_rctl(adapter);
5451 		e1000e_set_rx_mode(netdev);
5452 
5453 		/* turn on all-multi mode if wake on multicast is enabled */
5454 		if (wufc & E1000_WUFC_MC) {
5455 			rctl = er32(RCTL);
5456 			rctl |= E1000_RCTL_MPE;
5457 			ew32(RCTL, rctl);
5458 		}
5459 
5460 		ctrl = er32(CTRL);
5461 		/* advertise wake from D3Cold */
5462 		#define E1000_CTRL_ADVD3WUC 0x00100000
5463 		/* phy power management enable */
5464 		#define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5465 		ctrl |= E1000_CTRL_ADVD3WUC;
5466 		if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5467 			ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5468 		ew32(CTRL, ctrl);
5469 
5470 		if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5471 		    adapter->hw.phy.media_type ==
5472 		    e1000_media_type_internal_serdes) {
5473 			/* keep the laser running in D3 */
5474 			ctrl_ext = er32(CTRL_EXT);
5475 			ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5476 			ew32(CTRL_EXT, ctrl_ext);
5477 		}
5478 
5479 		if (adapter->flags & FLAG_IS_ICH)
5480 			e1000_suspend_workarounds_ich8lan(&adapter->hw);
5481 
5482 		/* Allow time for pending master requests to run */
5483 		e1000e_disable_pcie_master(&adapter->hw);
5484 
5485 		if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5486 			/* enable wakeup by the PHY */
5487 			retval = e1000_init_phy_wakeup(adapter, wufc);
5488 			if (retval)
5489 				return retval;
5490 		} else {
5491 			/* enable wakeup by the MAC */
5492 			ew32(WUFC, wufc);
5493 			ew32(WUC, E1000_WUC_PME_EN);
5494 		}
5495 	} else {
5496 		ew32(WUC, 0);
5497 		ew32(WUFC, 0);
5498 	}
5499 
5500 	*enable_wake = !!wufc;
5501 
5502 	/* make sure adapter isn't asleep if manageability is enabled */
5503 	if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5504 	    (hw->mac.ops.check_mng_mode(hw)))
5505 		*enable_wake = true;
5506 
5507 	if (adapter->hw.phy.type == e1000_phy_igp_3)
5508 		e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5509 
5510 	/*
5511 	 * Release control of h/w to f/w.  If f/w is AMT enabled, this
5512 	 * would have already happened in close and is redundant.
5513 	 */
5514 	e1000e_release_hw_control(adapter);
5515 
5516 	pci_disable_device(pdev);
5517 
5518 	return 0;
5519 }
5520 
5521 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5522 {
5523 	if (sleep && wake) {
5524 		pci_prepare_to_sleep(pdev);
5525 		return;
5526 	}
5527 
5528 	pci_wake_from_d3(pdev, wake);
5529 	pci_set_power_state(pdev, PCI_D3hot);
5530 }
5531 
5532 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5533                                     bool wake)
5534 {
5535 	struct net_device *netdev = pci_get_drvdata(pdev);
5536 	struct e1000_adapter *adapter = netdev_priv(netdev);
5537 
5538 	/*
5539 	 * The pci-e switch on some quad port adapters will report a
5540 	 * correctable error when the MAC transitions from D0 to D3.  To
5541 	 * prevent this we need to mask off the correctable errors on the
5542 	 * downstream port of the pci-e switch.
5543 	 */
5544 	if (adapter->flags & FLAG_IS_QUAD_PORT) {
5545 		struct pci_dev *us_dev = pdev->bus->self;
5546 		int pos = pci_pcie_cap(us_dev);
5547 		u16 devctl;
5548 
5549 		pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5550 		pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5551 		                      (devctl & ~PCI_EXP_DEVCTL_CERE));
5552 
5553 		e1000_power_off(pdev, sleep, wake);
5554 
5555 		pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5556 	} else {
5557 		e1000_power_off(pdev, sleep, wake);
5558 	}
5559 }
5560 
5561 #ifdef CONFIG_PCIEASPM
5562 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5563 {
5564 	pci_disable_link_state_locked(pdev, state);
5565 }
5566 #else
5567 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5568 {
5569 	int pos;
5570 	u16 reg16;
5571 
5572 	/*
5573 	 * Both device and parent should have the same ASPM setting.
5574 	 * Disable ASPM in downstream component first and then upstream.
5575 	 */
5576 	pos = pci_pcie_cap(pdev);
5577 	pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5578 	reg16 &= ~state;
5579 	pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5580 
5581 	if (!pdev->bus->self)
5582 		return;
5583 
5584 	pos = pci_pcie_cap(pdev->bus->self);
5585 	pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5586 	reg16 &= ~state;
5587 	pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5588 }
5589 #endif
5590 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5591 {
5592 	dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5593 		 (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5594 		 (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5595 
5596 	__e1000e_disable_aspm(pdev, state);
5597 }
5598 
5599 #ifdef CONFIG_PM
5600 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5601 {
5602 	return !!adapter->tx_ring->buffer_info;
5603 }
5604 
5605 static int __e1000_resume(struct pci_dev *pdev)
5606 {
5607 	struct net_device *netdev = pci_get_drvdata(pdev);
5608 	struct e1000_adapter *adapter = netdev_priv(netdev);
5609 	struct e1000_hw *hw = &adapter->hw;
5610 	u16 aspm_disable_flag = 0;
5611 	u32 err;
5612 
5613 	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5614 		aspm_disable_flag = PCIE_LINK_STATE_L0S;
5615 	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5616 		aspm_disable_flag |= PCIE_LINK_STATE_L1;
5617 	if (aspm_disable_flag)
5618 		e1000e_disable_aspm(pdev, aspm_disable_flag);
5619 
5620 	pci_set_power_state(pdev, PCI_D0);
5621 	pci_restore_state(pdev);
5622 	pci_save_state(pdev);
5623 
5624 	e1000e_set_interrupt_capability(adapter);
5625 	if (netif_running(netdev)) {
5626 		err = e1000_request_irq(adapter);
5627 		if (err)
5628 			return err;
5629 	}
5630 
5631 	if (hw->mac.type == e1000_pch2lan)
5632 		e1000_resume_workarounds_pchlan(&adapter->hw);
5633 
5634 	e1000e_power_up_phy(adapter);
5635 
5636 	/* report the system wakeup cause from S3/S4 */
5637 	if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5638 		u16 phy_data;
5639 
5640 		e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5641 		if (phy_data) {
5642 			e_info("PHY Wakeup cause - %s\n",
5643 				phy_data & E1000_WUS_EX ? "Unicast Packet" :
5644 				phy_data & E1000_WUS_MC ? "Multicast Packet" :
5645 				phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5646 				phy_data & E1000_WUS_MAG ? "Magic Packet" :
5647 				phy_data & E1000_WUS_LNKC ?
5648 				"Link Status Change" : "other");
5649 		}
5650 		e1e_wphy(&adapter->hw, BM_WUS, ~0);
5651 	} else {
5652 		u32 wus = er32(WUS);
5653 		if (wus) {
5654 			e_info("MAC Wakeup cause - %s\n",
5655 				wus & E1000_WUS_EX ? "Unicast Packet" :
5656 				wus & E1000_WUS_MC ? "Multicast Packet" :
5657 				wus & E1000_WUS_BC ? "Broadcast Packet" :
5658 				wus & E1000_WUS_MAG ? "Magic Packet" :
5659 				wus & E1000_WUS_LNKC ? "Link Status Change" :
5660 				"other");
5661 		}
5662 		ew32(WUS, ~0);
5663 	}
5664 
5665 	e1000e_reset(adapter);
5666 
5667 	e1000_init_manageability_pt(adapter);
5668 
5669 	if (netif_running(netdev))
5670 		e1000e_up(adapter);
5671 
5672 	netif_device_attach(netdev);
5673 
5674 	/*
5675 	 * If the controller has AMT, do not set DRV_LOAD until the interface
5676 	 * is up.  For all other cases, let the f/w know that the h/w is now
5677 	 * under the control of the driver.
5678 	 */
5679 	if (!(adapter->flags & FLAG_HAS_AMT))
5680 		e1000e_get_hw_control(adapter);
5681 
5682 	return 0;
5683 }
5684 
5685 #ifdef CONFIG_PM_SLEEP
5686 static int e1000_suspend(struct device *dev)
5687 {
5688 	struct pci_dev *pdev = to_pci_dev(dev);
5689 	int retval;
5690 	bool wake;
5691 
5692 	retval = __e1000_shutdown(pdev, &wake, false);
5693 	if (!retval)
5694 		e1000_complete_shutdown(pdev, true, wake);
5695 
5696 	return retval;
5697 }
5698 
5699 static int e1000_resume(struct device *dev)
5700 {
5701 	struct pci_dev *pdev = to_pci_dev(dev);
5702 	struct net_device *netdev = pci_get_drvdata(pdev);
5703 	struct e1000_adapter *adapter = netdev_priv(netdev);
5704 
5705 	if (e1000e_pm_ready(adapter))
5706 		adapter->idle_check = true;
5707 
5708 	return __e1000_resume(pdev);
5709 }
5710 #endif /* CONFIG_PM_SLEEP */
5711 
5712 #ifdef CONFIG_PM_RUNTIME
5713 static int e1000_runtime_suspend(struct device *dev)
5714 {
5715 	struct pci_dev *pdev = to_pci_dev(dev);
5716 	struct net_device *netdev = pci_get_drvdata(pdev);
5717 	struct e1000_adapter *adapter = netdev_priv(netdev);
5718 
5719 	if (e1000e_pm_ready(adapter)) {
5720 		bool wake;
5721 
5722 		__e1000_shutdown(pdev, &wake, true);
5723 	}
5724 
5725 	return 0;
5726 }
5727 
5728 static int e1000_idle(struct device *dev)
5729 {
5730 	struct pci_dev *pdev = to_pci_dev(dev);
5731 	struct net_device *netdev = pci_get_drvdata(pdev);
5732 	struct e1000_adapter *adapter = netdev_priv(netdev);
5733 
5734 	if (!e1000e_pm_ready(adapter))
5735 		return 0;
5736 
5737 	if (adapter->idle_check) {
5738 		adapter->idle_check = false;
5739 		if (!e1000e_has_link(adapter))
5740 			pm_schedule_suspend(dev, MSEC_PER_SEC);
5741 	}
5742 
5743 	return -EBUSY;
5744 }
5745 
5746 static int e1000_runtime_resume(struct device *dev)
5747 {
5748 	struct pci_dev *pdev = to_pci_dev(dev);
5749 	struct net_device *netdev = pci_get_drvdata(pdev);
5750 	struct e1000_adapter *adapter = netdev_priv(netdev);
5751 
5752 	if (!e1000e_pm_ready(adapter))
5753 		return 0;
5754 
5755 	adapter->idle_check = !dev->power.runtime_auto;
5756 	return __e1000_resume(pdev);
5757 }
5758 #endif /* CONFIG_PM_RUNTIME */
5759 #endif /* CONFIG_PM */
5760 
5761 static void e1000_shutdown(struct pci_dev *pdev)
5762 {
5763 	bool wake = false;
5764 
5765 	__e1000_shutdown(pdev, &wake, false);
5766 
5767 	if (system_state == SYSTEM_POWER_OFF)
5768 		e1000_complete_shutdown(pdev, false, wake);
5769 }
5770 
5771 #ifdef CONFIG_NET_POLL_CONTROLLER
5772 
5773 static irqreturn_t e1000_intr_msix(int irq, void *data)
5774 {
5775 	struct net_device *netdev = data;
5776 	struct e1000_adapter *adapter = netdev_priv(netdev);
5777 
5778 	if (adapter->msix_entries) {
5779 		int vector, msix_irq;
5780 
5781 		vector = 0;
5782 		msix_irq = adapter->msix_entries[vector].vector;
5783 		disable_irq(msix_irq);
5784 		e1000_intr_msix_rx(msix_irq, netdev);
5785 		enable_irq(msix_irq);
5786 
5787 		vector++;
5788 		msix_irq = adapter->msix_entries[vector].vector;
5789 		disable_irq(msix_irq);
5790 		e1000_intr_msix_tx(msix_irq, netdev);
5791 		enable_irq(msix_irq);
5792 
5793 		vector++;
5794 		msix_irq = adapter->msix_entries[vector].vector;
5795 		disable_irq(msix_irq);
5796 		e1000_msix_other(msix_irq, netdev);
5797 		enable_irq(msix_irq);
5798 	}
5799 
5800 	return IRQ_HANDLED;
5801 }
5802 
5803 /*
5804  * Polling 'interrupt' - used by things like netconsole to send skbs
5805  * without having to re-enable interrupts. It's not called while
5806  * the interrupt routine is executing.
5807  */
5808 static void e1000_netpoll(struct net_device *netdev)
5809 {
5810 	struct e1000_adapter *adapter = netdev_priv(netdev);
5811 
5812 	switch (adapter->int_mode) {
5813 	case E1000E_INT_MODE_MSIX:
5814 		e1000_intr_msix(adapter->pdev->irq, netdev);
5815 		break;
5816 	case E1000E_INT_MODE_MSI:
5817 		disable_irq(adapter->pdev->irq);
5818 		e1000_intr_msi(adapter->pdev->irq, netdev);
5819 		enable_irq(adapter->pdev->irq);
5820 		break;
5821 	default: /* E1000E_INT_MODE_LEGACY */
5822 		disable_irq(adapter->pdev->irq);
5823 		e1000_intr(adapter->pdev->irq, netdev);
5824 		enable_irq(adapter->pdev->irq);
5825 		break;
5826 	}
5827 }
5828 #endif
5829 
5830 /**
5831  * e1000_io_error_detected - called when PCI error is detected
5832  * @pdev: Pointer to PCI device
5833  * @state: The current pci connection state
5834  *
5835  * This function is called after a PCI bus error affecting
5836  * this device has been detected.
5837  */
5838 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5839 						pci_channel_state_t state)
5840 {
5841 	struct net_device *netdev = pci_get_drvdata(pdev);
5842 	struct e1000_adapter *adapter = netdev_priv(netdev);
5843 
5844 	netif_device_detach(netdev);
5845 
5846 	if (state == pci_channel_io_perm_failure)
5847 		return PCI_ERS_RESULT_DISCONNECT;
5848 
5849 	if (netif_running(netdev))
5850 		e1000e_down(adapter);
5851 	pci_disable_device(pdev);
5852 
5853 	/* Request a slot slot reset. */
5854 	return PCI_ERS_RESULT_NEED_RESET;
5855 }
5856 
5857 /**
5858  * e1000_io_slot_reset - called after the pci bus has been reset.
5859  * @pdev: Pointer to PCI device
5860  *
5861  * Restart the card from scratch, as if from a cold-boot. Implementation
5862  * resembles the first-half of the e1000_resume routine.
5863  */
5864 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5865 {
5866 	struct net_device *netdev = pci_get_drvdata(pdev);
5867 	struct e1000_adapter *adapter = netdev_priv(netdev);
5868 	struct e1000_hw *hw = &adapter->hw;
5869 	u16 aspm_disable_flag = 0;
5870 	int err;
5871 	pci_ers_result_t result;
5872 
5873 	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5874 		aspm_disable_flag = PCIE_LINK_STATE_L0S;
5875 	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5876 		aspm_disable_flag |= PCIE_LINK_STATE_L1;
5877 	if (aspm_disable_flag)
5878 		e1000e_disable_aspm(pdev, aspm_disable_flag);
5879 
5880 	err = pci_enable_device_mem(pdev);
5881 	if (err) {
5882 		dev_err(&pdev->dev,
5883 			"Cannot re-enable PCI device after reset.\n");
5884 		result = PCI_ERS_RESULT_DISCONNECT;
5885 	} else {
5886 		pci_set_master(pdev);
5887 		pdev->state_saved = true;
5888 		pci_restore_state(pdev);
5889 
5890 		pci_enable_wake(pdev, PCI_D3hot, 0);
5891 		pci_enable_wake(pdev, PCI_D3cold, 0);
5892 
5893 		e1000e_reset(adapter);
5894 		ew32(WUS, ~0);
5895 		result = PCI_ERS_RESULT_RECOVERED;
5896 	}
5897 
5898 	pci_cleanup_aer_uncorrect_error_status(pdev);
5899 
5900 	return result;
5901 }
5902 
5903 /**
5904  * e1000_io_resume - called when traffic can start flowing again.
5905  * @pdev: Pointer to PCI device
5906  *
5907  * This callback is called when the error recovery driver tells us that
5908  * its OK to resume normal operation. Implementation resembles the
5909  * second-half of the e1000_resume routine.
5910  */
5911 static void e1000_io_resume(struct pci_dev *pdev)
5912 {
5913 	struct net_device *netdev = pci_get_drvdata(pdev);
5914 	struct e1000_adapter *adapter = netdev_priv(netdev);
5915 
5916 	e1000_init_manageability_pt(adapter);
5917 
5918 	if (netif_running(netdev)) {
5919 		if (e1000e_up(adapter)) {
5920 			dev_err(&pdev->dev,
5921 				"can't bring device back up after reset\n");
5922 			return;
5923 		}
5924 	}
5925 
5926 	netif_device_attach(netdev);
5927 
5928 	/*
5929 	 * If the controller has AMT, do not set DRV_LOAD until the interface
5930 	 * is up.  For all other cases, let the f/w know that the h/w is now
5931 	 * under the control of the driver.
5932 	 */
5933 	if (!(adapter->flags & FLAG_HAS_AMT))
5934 		e1000e_get_hw_control(adapter);
5935 
5936 }
5937 
5938 static void e1000_print_device_info(struct e1000_adapter *adapter)
5939 {
5940 	struct e1000_hw *hw = &adapter->hw;
5941 	struct net_device *netdev = adapter->netdev;
5942 	u32 ret_val;
5943 	u8 pba_str[E1000_PBANUM_LENGTH];
5944 
5945 	/* print bus type/speed/width info */
5946 	e_info("(PCI Express:2.5GT/s:%s) %pM\n",
5947 	       /* bus width */
5948 	       ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5949 	        "Width x1"),
5950 	       /* MAC address */
5951 	       netdev->dev_addr);
5952 	e_info("Intel(R) PRO/%s Network Connection\n",
5953 	       (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5954 	ret_val = e1000_read_pba_string_generic(hw, pba_str,
5955 						E1000_PBANUM_LENGTH);
5956 	if (ret_val)
5957 		strlcpy((char *)pba_str, "Unknown", sizeof(pba_str));
5958 	e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5959 	       hw->mac.type, hw->phy.type, pba_str);
5960 }
5961 
5962 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5963 {
5964 	struct e1000_hw *hw = &adapter->hw;
5965 	int ret_val;
5966 	u16 buf = 0;
5967 
5968 	if (hw->mac.type != e1000_82573)
5969 		return;
5970 
5971 	ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5972 	le16_to_cpus(&buf);
5973 	if (!ret_val && (!(buf & (1 << 0)))) {
5974 		/* Deep Smart Power Down (DSPD) */
5975 		dev_warn(&adapter->pdev->dev,
5976 			 "Warning: detected DSPD enabled in EEPROM\n");
5977 	}
5978 }
5979 
5980 static int e1000_set_features(struct net_device *netdev,
5981 			      netdev_features_t features)
5982 {
5983 	struct e1000_adapter *adapter = netdev_priv(netdev);
5984 	netdev_features_t changed = features ^ netdev->features;
5985 
5986 	if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
5987 		adapter->flags |= FLAG_TSO_FORCE;
5988 
5989 	if (!(changed & (NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX |
5990 			 NETIF_F_RXCSUM | NETIF_F_RXHASH)))
5991 		return 0;
5992 
5993 	/*
5994 	 * IP payload checksum (enabled with jumbos/packet-split when Rx
5995 	 * checksum is enabled) and generation of RSS hash is mutually
5996 	 * exclusive in the hardware.
5997 	 */
5998 	if (adapter->rx_ps_pages &&
5999 	    (features & NETIF_F_RXCSUM) && (features & NETIF_F_RXHASH)) {
6000 		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");
6001 		return -EINVAL;
6002 	}
6003 
6004 	netdev->features = features;
6005 
6006 	if (netif_running(netdev))
6007 		e1000e_reinit_locked(adapter);
6008 	else
6009 		e1000e_reset(adapter);
6010 
6011 	return 0;
6012 }
6013 
6014 static const struct net_device_ops e1000e_netdev_ops = {
6015 	.ndo_open		= e1000_open,
6016 	.ndo_stop		= e1000_close,
6017 	.ndo_start_xmit		= e1000_xmit_frame,
6018 	.ndo_get_stats64	= e1000e_get_stats64,
6019 	.ndo_set_rx_mode	= e1000e_set_rx_mode,
6020 	.ndo_set_mac_address	= e1000_set_mac,
6021 	.ndo_change_mtu		= e1000_change_mtu,
6022 	.ndo_do_ioctl		= e1000_ioctl,
6023 	.ndo_tx_timeout		= e1000_tx_timeout,
6024 	.ndo_validate_addr	= eth_validate_addr,
6025 
6026 	.ndo_vlan_rx_add_vid	= e1000_vlan_rx_add_vid,
6027 	.ndo_vlan_rx_kill_vid	= e1000_vlan_rx_kill_vid,
6028 #ifdef CONFIG_NET_POLL_CONTROLLER
6029 	.ndo_poll_controller	= e1000_netpoll,
6030 #endif
6031 	.ndo_set_features = e1000_set_features,
6032 };
6033 
6034 /**
6035  * e1000_probe - Device Initialization Routine
6036  * @pdev: PCI device information struct
6037  * @ent: entry in e1000_pci_tbl
6038  *
6039  * Returns 0 on success, negative on failure
6040  *
6041  * e1000_probe initializes an adapter identified by a pci_dev structure.
6042  * The OS initialization, configuring of the adapter private structure,
6043  * and a hardware reset occur.
6044  **/
6045 static int __devinit e1000_probe(struct pci_dev *pdev,
6046 				 const struct pci_device_id *ent)
6047 {
6048 	struct net_device *netdev;
6049 	struct e1000_adapter *adapter;
6050 	struct e1000_hw *hw;
6051 	const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
6052 	resource_size_t mmio_start, mmio_len;
6053 	resource_size_t flash_start, flash_len;
6054 
6055 	static int cards_found;
6056 	u16 aspm_disable_flag = 0;
6057 	int i, err, pci_using_dac;
6058 	u16 eeprom_data = 0;
6059 	u16 eeprom_apme_mask = E1000_EEPROM_APME;
6060 
6061 	if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
6062 		aspm_disable_flag = PCIE_LINK_STATE_L0S;
6063 	if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
6064 		aspm_disable_flag |= PCIE_LINK_STATE_L1;
6065 	if (aspm_disable_flag)
6066 		e1000e_disable_aspm(pdev, aspm_disable_flag);
6067 
6068 	err = pci_enable_device_mem(pdev);
6069 	if (err)
6070 		return err;
6071 
6072 	pci_using_dac = 0;
6073 	err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
6074 	if (!err) {
6075 		err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
6076 		if (!err)
6077 			pci_using_dac = 1;
6078 	} else {
6079 		err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6080 		if (err) {
6081 			err = dma_set_coherent_mask(&pdev->dev,
6082 						    DMA_BIT_MASK(32));
6083 			if (err) {
6084 				dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
6085 				goto err_dma;
6086 			}
6087 		}
6088 	}
6089 
6090 	err = pci_request_selected_regions_exclusive(pdev,
6091 	                                  pci_select_bars(pdev, IORESOURCE_MEM),
6092 	                                  e1000e_driver_name);
6093 	if (err)
6094 		goto err_pci_reg;
6095 
6096 	/* AER (Advanced Error Reporting) hooks */
6097 	pci_enable_pcie_error_reporting(pdev);
6098 
6099 	pci_set_master(pdev);
6100 	/* PCI config space info */
6101 	err = pci_save_state(pdev);
6102 	if (err)
6103 		goto err_alloc_etherdev;
6104 
6105 	err = -ENOMEM;
6106 	netdev = alloc_etherdev(sizeof(struct e1000_adapter));
6107 	if (!netdev)
6108 		goto err_alloc_etherdev;
6109 
6110 	SET_NETDEV_DEV(netdev, &pdev->dev);
6111 
6112 	netdev->irq = pdev->irq;
6113 
6114 	pci_set_drvdata(pdev, netdev);
6115 	adapter = netdev_priv(netdev);
6116 	hw = &adapter->hw;
6117 	adapter->netdev = netdev;
6118 	adapter->pdev = pdev;
6119 	adapter->ei = ei;
6120 	adapter->pba = ei->pba;
6121 	adapter->flags = ei->flags;
6122 	adapter->flags2 = ei->flags2;
6123 	adapter->hw.adapter = adapter;
6124 	adapter->hw.mac.type = ei->mac;
6125 	adapter->max_hw_frame_size = ei->max_hw_frame_size;
6126 	adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
6127 
6128 	mmio_start = pci_resource_start(pdev, 0);
6129 	mmio_len = pci_resource_len(pdev, 0);
6130 
6131 	err = -EIO;
6132 	adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
6133 	if (!adapter->hw.hw_addr)
6134 		goto err_ioremap;
6135 
6136 	if ((adapter->flags & FLAG_HAS_FLASH) &&
6137 	    (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
6138 		flash_start = pci_resource_start(pdev, 1);
6139 		flash_len = pci_resource_len(pdev, 1);
6140 		adapter->hw.flash_address = ioremap(flash_start, flash_len);
6141 		if (!adapter->hw.flash_address)
6142 			goto err_flashmap;
6143 	}
6144 
6145 	/* construct the net_device struct */
6146 	netdev->netdev_ops		= &e1000e_netdev_ops;
6147 	e1000e_set_ethtool_ops(netdev);
6148 	netdev->watchdog_timeo		= 5 * HZ;
6149 	netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
6150 	strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
6151 
6152 	netdev->mem_start = mmio_start;
6153 	netdev->mem_end = mmio_start + mmio_len;
6154 
6155 	adapter->bd_number = cards_found++;
6156 
6157 	e1000e_check_options(adapter);
6158 
6159 	/* setup adapter struct */
6160 	err = e1000_sw_init(adapter);
6161 	if (err)
6162 		goto err_sw_init;
6163 
6164 	memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
6165 	memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
6166 	memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
6167 
6168 	err = ei->get_variants(adapter);
6169 	if (err)
6170 		goto err_hw_init;
6171 
6172 	if ((adapter->flags & FLAG_IS_ICH) &&
6173 	    (adapter->flags & FLAG_READ_ONLY_NVM))
6174 		e1000e_write_protect_nvm_ich8lan(&adapter->hw);
6175 
6176 	hw->mac.ops.get_bus_info(&adapter->hw);
6177 
6178 	adapter->hw.phy.autoneg_wait_to_complete = 0;
6179 
6180 	/* Copper options */
6181 	if (adapter->hw.phy.media_type == e1000_media_type_copper) {
6182 		adapter->hw.phy.mdix = AUTO_ALL_MODES;
6183 		adapter->hw.phy.disable_polarity_correction = 0;
6184 		adapter->hw.phy.ms_type = e1000_ms_hw_default;
6185 	}
6186 
6187 	if (e1000_check_reset_block(&adapter->hw))
6188 		e_info("PHY reset is blocked due to SOL/IDER session.\n");
6189 
6190 	/* Set initial default active device features */
6191 	netdev->features = (NETIF_F_SG |
6192 			    NETIF_F_HW_VLAN_RX |
6193 			    NETIF_F_HW_VLAN_TX |
6194 			    NETIF_F_TSO |
6195 			    NETIF_F_TSO6 |
6196 			    NETIF_F_RXHASH |
6197 			    NETIF_F_RXCSUM |
6198 			    NETIF_F_HW_CSUM);
6199 
6200 	/* Set user-changeable features (subset of all device features) */
6201 	netdev->hw_features = netdev->features;
6202 
6203 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
6204 		netdev->features |= NETIF_F_HW_VLAN_FILTER;
6205 
6206 	netdev->vlan_features |= (NETIF_F_SG |
6207 				  NETIF_F_TSO |
6208 				  NETIF_F_TSO6 |
6209 				  NETIF_F_HW_CSUM);
6210 
6211 	netdev->priv_flags |= IFF_UNICAST_FLT;
6212 
6213 	if (pci_using_dac) {
6214 		netdev->features |= NETIF_F_HIGHDMA;
6215 		netdev->vlan_features |= NETIF_F_HIGHDMA;
6216 	}
6217 
6218 	if (e1000e_enable_mng_pass_thru(&adapter->hw))
6219 		adapter->flags |= FLAG_MNG_PT_ENABLED;
6220 
6221 	/*
6222 	 * before reading the NVM, reset the controller to
6223 	 * put the device in a known good starting state
6224 	 */
6225 	adapter->hw.mac.ops.reset_hw(&adapter->hw);
6226 
6227 	/*
6228 	 * systems with ASPM and others may see the checksum fail on the first
6229 	 * attempt. Let's give it a few tries
6230 	 */
6231 	for (i = 0;; i++) {
6232 		if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
6233 			break;
6234 		if (i == 2) {
6235 			e_err("The NVM Checksum Is Not Valid\n");
6236 			err = -EIO;
6237 			goto err_eeprom;
6238 		}
6239 	}
6240 
6241 	e1000_eeprom_checks(adapter);
6242 
6243 	/* copy the MAC address */
6244 	if (e1000e_read_mac_addr(&adapter->hw))
6245 		e_err("NVM Read Error while reading MAC address\n");
6246 
6247 	memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
6248 	memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
6249 
6250 	if (!is_valid_ether_addr(netdev->perm_addr)) {
6251 		e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
6252 		err = -EIO;
6253 		goto err_eeprom;
6254 	}
6255 
6256 	init_timer(&adapter->watchdog_timer);
6257 	adapter->watchdog_timer.function = e1000_watchdog;
6258 	adapter->watchdog_timer.data = (unsigned long) adapter;
6259 
6260 	init_timer(&adapter->phy_info_timer);
6261 	adapter->phy_info_timer.function = e1000_update_phy_info;
6262 	adapter->phy_info_timer.data = (unsigned long) adapter;
6263 
6264 	INIT_WORK(&adapter->reset_task, e1000_reset_task);
6265 	INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
6266 	INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
6267 	INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
6268 	INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
6269 
6270 	/* Initialize link parameters. User can change them with ethtool */
6271 	adapter->hw.mac.autoneg = 1;
6272 	adapter->fc_autoneg = true;
6273 	adapter->hw.fc.requested_mode = e1000_fc_default;
6274 	adapter->hw.fc.current_mode = e1000_fc_default;
6275 	adapter->hw.phy.autoneg_advertised = 0x2f;
6276 
6277 	/* ring size defaults */
6278 	adapter->rx_ring->count = 256;
6279 	adapter->tx_ring->count = 256;
6280 
6281 	/*
6282 	 * Initial Wake on LAN setting - If APM wake is enabled in
6283 	 * the EEPROM, enable the ACPI Magic Packet filter
6284 	 */
6285 	if (adapter->flags & FLAG_APME_IN_WUC) {
6286 		/* APME bit in EEPROM is mapped to WUC.APME */
6287 		eeprom_data = er32(WUC);
6288 		eeprom_apme_mask = E1000_WUC_APME;
6289 		if ((hw->mac.type > e1000_ich10lan) &&
6290 		    (eeprom_data & E1000_WUC_PHY_WAKE))
6291 			adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
6292 	} else if (adapter->flags & FLAG_APME_IN_CTRL3) {
6293 		if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
6294 		    (adapter->hw.bus.func == 1))
6295 			e1000_read_nvm(&adapter->hw,
6296 				NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
6297 		else
6298 			e1000_read_nvm(&adapter->hw,
6299 				NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
6300 	}
6301 
6302 	/* fetch WoL from EEPROM */
6303 	if (eeprom_data & eeprom_apme_mask)
6304 		adapter->eeprom_wol |= E1000_WUFC_MAG;
6305 
6306 	/*
6307 	 * now that we have the eeprom settings, apply the special cases
6308 	 * where the eeprom may be wrong or the board simply won't support
6309 	 * wake on lan on a particular port
6310 	 */
6311 	if (!(adapter->flags & FLAG_HAS_WOL))
6312 		adapter->eeprom_wol = 0;
6313 
6314 	/* initialize the wol settings based on the eeprom settings */
6315 	adapter->wol = adapter->eeprom_wol;
6316 	device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6317 
6318 	/* save off EEPROM version number */
6319 	e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6320 
6321 	/* reset the hardware with the new settings */
6322 	e1000e_reset(adapter);
6323 
6324 	/*
6325 	 * If the controller has AMT, do not set DRV_LOAD until the interface
6326 	 * is up.  For all other cases, let the f/w know that the h/w is now
6327 	 * under the control of the driver.
6328 	 */
6329 	if (!(adapter->flags & FLAG_HAS_AMT))
6330 		e1000e_get_hw_control(adapter);
6331 
6332 	strlcpy(netdev->name, "eth%d", sizeof(netdev->name));
6333 	err = register_netdev(netdev);
6334 	if (err)
6335 		goto err_register;
6336 
6337 	/* carrier off reporting is important to ethtool even BEFORE open */
6338 	netif_carrier_off(netdev);
6339 
6340 	e1000_print_device_info(adapter);
6341 
6342 	if (pci_dev_run_wake(pdev))
6343 		pm_runtime_put_noidle(&pdev->dev);
6344 
6345 	return 0;
6346 
6347 err_register:
6348 	if (!(adapter->flags & FLAG_HAS_AMT))
6349 		e1000e_release_hw_control(adapter);
6350 err_eeprom:
6351 	if (!e1000_check_reset_block(&adapter->hw))
6352 		e1000_phy_hw_reset(&adapter->hw);
6353 err_hw_init:
6354 	kfree(adapter->tx_ring);
6355 	kfree(adapter->rx_ring);
6356 err_sw_init:
6357 	if (adapter->hw.flash_address)
6358 		iounmap(adapter->hw.flash_address);
6359 	e1000e_reset_interrupt_capability(adapter);
6360 err_flashmap:
6361 	iounmap(adapter->hw.hw_addr);
6362 err_ioremap:
6363 	free_netdev(netdev);
6364 err_alloc_etherdev:
6365 	pci_release_selected_regions(pdev,
6366 	                             pci_select_bars(pdev, IORESOURCE_MEM));
6367 err_pci_reg:
6368 err_dma:
6369 	pci_disable_device(pdev);
6370 	return err;
6371 }
6372 
6373 /**
6374  * e1000_remove - Device Removal Routine
6375  * @pdev: PCI device information struct
6376  *
6377  * e1000_remove is called by the PCI subsystem to alert the driver
6378  * that it should release a PCI device.  The could be caused by a
6379  * Hot-Plug event, or because the driver is going to be removed from
6380  * memory.
6381  **/
6382 static void __devexit e1000_remove(struct pci_dev *pdev)
6383 {
6384 	struct net_device *netdev = pci_get_drvdata(pdev);
6385 	struct e1000_adapter *adapter = netdev_priv(netdev);
6386 	bool down = test_bit(__E1000_DOWN, &adapter->state);
6387 
6388 	/*
6389 	 * The timers may be rescheduled, so explicitly disable them
6390 	 * from being rescheduled.
6391 	 */
6392 	if (!down)
6393 		set_bit(__E1000_DOWN, &adapter->state);
6394 	del_timer_sync(&adapter->watchdog_timer);
6395 	del_timer_sync(&adapter->phy_info_timer);
6396 
6397 	cancel_work_sync(&adapter->reset_task);
6398 	cancel_work_sync(&adapter->watchdog_task);
6399 	cancel_work_sync(&adapter->downshift_task);
6400 	cancel_work_sync(&adapter->update_phy_task);
6401 	cancel_work_sync(&adapter->print_hang_task);
6402 
6403 	if (!(netdev->flags & IFF_UP))
6404 		e1000_power_down_phy(adapter);
6405 
6406 	/* Don't lie to e1000_close() down the road. */
6407 	if (!down)
6408 		clear_bit(__E1000_DOWN, &adapter->state);
6409 	unregister_netdev(netdev);
6410 
6411 	if (pci_dev_run_wake(pdev))
6412 		pm_runtime_get_noresume(&pdev->dev);
6413 
6414 	/*
6415 	 * Release control of h/w to f/w.  If f/w is AMT enabled, this
6416 	 * would have already happened in close and is redundant.
6417 	 */
6418 	e1000e_release_hw_control(adapter);
6419 
6420 	e1000e_reset_interrupt_capability(adapter);
6421 	kfree(adapter->tx_ring);
6422 	kfree(adapter->rx_ring);
6423 
6424 	iounmap(adapter->hw.hw_addr);
6425 	if (adapter->hw.flash_address)
6426 		iounmap(adapter->hw.flash_address);
6427 	pci_release_selected_regions(pdev,
6428 	                             pci_select_bars(pdev, IORESOURCE_MEM));
6429 
6430 	free_netdev(netdev);
6431 
6432 	/* AER disable */
6433 	pci_disable_pcie_error_reporting(pdev);
6434 
6435 	pci_disable_device(pdev);
6436 }
6437 
6438 /* PCI Error Recovery (ERS) */
6439 static struct pci_error_handlers e1000_err_handler = {
6440 	.error_detected = e1000_io_error_detected,
6441 	.slot_reset = e1000_io_slot_reset,
6442 	.resume = e1000_io_resume,
6443 };
6444 
6445 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6446 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6447 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6448 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6449 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6450 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6451 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6452 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6453 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6454 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6455 
6456 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6457 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6458 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6459 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6460 
6461 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6462 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6463 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6464 
6465 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6466 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6467 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6468 
6469 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6470 	  board_80003es2lan },
6471 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6472 	  board_80003es2lan },
6473 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6474 	  board_80003es2lan },
6475 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6476 	  board_80003es2lan },
6477 
6478 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6479 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6480 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6481 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6482 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6483 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6484 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6485 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6486 
6487 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6488 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6489 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6490 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6491 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6492 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6493 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6494 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6495 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6496 
6497 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6498 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6499 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6500 
6501 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6502 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6503 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6504 
6505 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6506 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6507 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6508 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6509 
6510 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6511 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6512 
6513 	{ 0, 0, 0, 0, 0, 0, 0 }	/* terminate list */
6514 };
6515 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6516 
6517 #ifdef CONFIG_PM
6518 static const struct dev_pm_ops e1000_pm_ops = {
6519 	SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6520 	SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6521 				e1000_runtime_resume, e1000_idle)
6522 };
6523 #endif
6524 
6525 /* PCI Device API Driver */
6526 static struct pci_driver e1000_driver = {
6527 	.name     = e1000e_driver_name,
6528 	.id_table = e1000_pci_tbl,
6529 	.probe    = e1000_probe,
6530 	.remove   = __devexit_p(e1000_remove),
6531 #ifdef CONFIG_PM
6532 	.driver   = {
6533 		.pm = &e1000_pm_ops,
6534 	},
6535 #endif
6536 	.shutdown = e1000_shutdown,
6537 	.err_handler = &e1000_err_handler
6538 };
6539 
6540 /**
6541  * e1000_init_module - Driver Registration Routine
6542  *
6543  * e1000_init_module is the first routine called when the driver is
6544  * loaded. All it does is register with the PCI subsystem.
6545  **/
6546 static int __init e1000_init_module(void)
6547 {
6548 	int ret;
6549 	pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6550 		e1000e_driver_version);
6551 	pr_info("Copyright(c) 1999 - 2012 Intel Corporation.\n");
6552 	ret = pci_register_driver(&e1000_driver);
6553 
6554 	return ret;
6555 }
6556 module_init(e1000_init_module);
6557 
6558 /**
6559  * e1000_exit_module - Driver Exit Cleanup Routine
6560  *
6561  * e1000_exit_module is called just before the driver is removed
6562  * from memory.
6563  **/
6564 static void __exit e1000_exit_module(void)
6565 {
6566 	pci_unregister_driver(&e1000_driver);
6567 }
6568 module_exit(e1000_exit_module);
6569 
6570 
6571 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6572 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6573 MODULE_LICENSE("GPL");
6574 MODULE_VERSION(DRV_VERSION);
6575 
6576 /* e1000_main.c */
6577