xref: /linux/drivers/net/ethernet/intel/e1000e/netdev.c (revision 41de8d4cff21a2e81e3d9ff66f5f7c903f9c3ab1)
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 	{}
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 		u64 a;
187 		u64 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 		u64 a;
195 		u64 b;
196 		u64 c;
197 		u64 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 it
1272 		 * 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 packet
1278 		 * throughput, so unsplit small packets and save the alloc/put
1279 		 * only valid in softirq (napi) context to call kmap_*
1280 		 */
1281 		if (l1 && (l1 <= copybreak) &&
1282 		    ((length + l1) <= adapter->rx_ps_bsize0)) {
1283 			u8 *vaddr;
1284 
1285 			ps_page = &buffer_info->ps_pages[0];
1286 
1287 			/*
1288 			 * there is no documentation about how to call
1289 			 * kmap_atomic, so we can't hold the mapping
1290 			 * very long
1291 			 */
1292 			dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1293 						PAGE_SIZE, DMA_FROM_DEVICE);
1294 			vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1295 			memcpy(skb_tail_pointer(skb), vaddr, l1);
1296 			kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1297 			dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1298 						   PAGE_SIZE, DMA_FROM_DEVICE);
1299 
1300 			/* remove the CRC */
1301 			if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1302 				l1 -= 4;
1303 
1304 			skb_put(skb, l1);
1305 			goto copydone;
1306 		} /* if */
1307 		}
1308 
1309 		for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1310 			length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1311 			if (!length)
1312 				break;
1313 
1314 			ps_page = &buffer_info->ps_pages[j];
1315 			dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1316 				       DMA_FROM_DEVICE);
1317 			ps_page->dma = 0;
1318 			skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1319 			ps_page->page = NULL;
1320 			skb->len += length;
1321 			skb->data_len += length;
1322 			skb->truesize += PAGE_SIZE;
1323 		}
1324 
1325 		/* strip the ethernet crc, problem is we're using pages now so
1326 		 * this whole operation can get a little cpu intensive
1327 		 */
1328 		if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1329 			pskb_trim(skb, skb->len - 4);
1330 
1331 copydone:
1332 		total_rx_bytes += skb->len;
1333 		total_rx_packets++;
1334 
1335 		e1000_rx_checksum(adapter, staterr,
1336 				  rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
1337 
1338 		e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1339 
1340 		if (rx_desc->wb.upper.header_status &
1341 			   cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1342 			adapter->rx_hdr_split++;
1343 
1344 		e1000_receive_skb(adapter, netdev, skb,
1345 				  staterr, rx_desc->wb.middle.vlan);
1346 
1347 next_desc:
1348 		rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1349 		buffer_info->skb = NULL;
1350 
1351 		/* return some buffers to hardware, one at a time is too slow */
1352 		if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1353 			adapter->alloc_rx_buf(rx_ring, cleaned_count,
1354 					      GFP_ATOMIC);
1355 			cleaned_count = 0;
1356 		}
1357 
1358 		/* use prefetched values */
1359 		rx_desc = next_rxd;
1360 		buffer_info = next_buffer;
1361 
1362 		staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1363 	}
1364 	rx_ring->next_to_clean = i;
1365 
1366 	cleaned_count = e1000_desc_unused(rx_ring);
1367 	if (cleaned_count)
1368 		adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1369 
1370 	adapter->total_rx_bytes += total_rx_bytes;
1371 	adapter->total_rx_packets += total_rx_packets;
1372 	return cleaned;
1373 }
1374 
1375 /**
1376  * e1000_consume_page - helper function
1377  **/
1378 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1379                                u16 length)
1380 {
1381 	bi->page = NULL;
1382 	skb->len += length;
1383 	skb->data_len += length;
1384 	skb->truesize += PAGE_SIZE;
1385 }
1386 
1387 /**
1388  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1389  * @adapter: board private structure
1390  *
1391  * the return value indicates whether actual cleaning was done, there
1392  * is no guarantee that everything was cleaned
1393  **/
1394 static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1395 				     int work_to_do)
1396 {
1397 	struct e1000_adapter *adapter = rx_ring->adapter;
1398 	struct net_device *netdev = adapter->netdev;
1399 	struct pci_dev *pdev = adapter->pdev;
1400 	union e1000_rx_desc_extended *rx_desc, *next_rxd;
1401 	struct e1000_buffer *buffer_info, *next_buffer;
1402 	u32 length, staterr;
1403 	unsigned int i;
1404 	int cleaned_count = 0;
1405 	bool cleaned = false;
1406 	unsigned int total_rx_bytes=0, total_rx_packets=0;
1407 
1408 	i = rx_ring->next_to_clean;
1409 	rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1410 	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1411 	buffer_info = &rx_ring->buffer_info[i];
1412 
1413 	while (staterr & E1000_RXD_STAT_DD) {
1414 		struct sk_buff *skb;
1415 
1416 		if (*work_done >= work_to_do)
1417 			break;
1418 		(*work_done)++;
1419 		rmb();	/* read descriptor and rx_buffer_info after status DD */
1420 
1421 		skb = buffer_info->skb;
1422 		buffer_info->skb = NULL;
1423 
1424 		++i;
1425 		if (i == rx_ring->count)
1426 			i = 0;
1427 		next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1428 		prefetch(next_rxd);
1429 
1430 		next_buffer = &rx_ring->buffer_info[i];
1431 
1432 		cleaned = true;
1433 		cleaned_count++;
1434 		dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1435 			       DMA_FROM_DEVICE);
1436 		buffer_info->dma = 0;
1437 
1438 		length = le16_to_cpu(rx_desc->wb.upper.length);
1439 
1440 		/* errors is only valid for DD + EOP descriptors */
1441 		if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1442 			     (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK))) {
1443 			/* recycle both page and skb */
1444 			buffer_info->skb = skb;
1445 			/* an error means any chain goes out the window too */
1446 			if (rx_ring->rx_skb_top)
1447 				dev_kfree_skb_irq(rx_ring->rx_skb_top);
1448 			rx_ring->rx_skb_top = NULL;
1449 			goto next_desc;
1450 		}
1451 
1452 #define rxtop (rx_ring->rx_skb_top)
1453 		if (!(staterr & E1000_RXD_STAT_EOP)) {
1454 			/* this descriptor is only the beginning (or middle) */
1455 			if (!rxtop) {
1456 				/* this is the beginning of a chain */
1457 				rxtop = skb;
1458 				skb_fill_page_desc(rxtop, 0, buffer_info->page,
1459 				                   0, length);
1460 			} else {
1461 				/* this is the middle of a chain */
1462 				skb_fill_page_desc(rxtop,
1463 				    skb_shinfo(rxtop)->nr_frags,
1464 				    buffer_info->page, 0, length);
1465 				/* re-use the skb, only consumed the page */
1466 				buffer_info->skb = skb;
1467 			}
1468 			e1000_consume_page(buffer_info, rxtop, length);
1469 			goto next_desc;
1470 		} else {
1471 			if (rxtop) {
1472 				/* end of the chain */
1473 				skb_fill_page_desc(rxtop,
1474 				    skb_shinfo(rxtop)->nr_frags,
1475 				    buffer_info->page, 0, length);
1476 				/* re-use the current skb, we only consumed the
1477 				 * page */
1478 				buffer_info->skb = skb;
1479 				skb = rxtop;
1480 				rxtop = NULL;
1481 				e1000_consume_page(buffer_info, skb, length);
1482 			} else {
1483 				/* no chain, got EOP, this buf is the packet
1484 				 * copybreak to save the put_page/alloc_page */
1485 				if (length <= copybreak &&
1486 				    skb_tailroom(skb) >= length) {
1487 					u8 *vaddr;
1488 					vaddr = kmap_atomic(buffer_info->page,
1489 					                   KM_SKB_DATA_SOFTIRQ);
1490 					memcpy(skb_tail_pointer(skb), vaddr,
1491 					       length);
1492 					kunmap_atomic(vaddr,
1493 					              KM_SKB_DATA_SOFTIRQ);
1494 					/* re-use the page, so don't erase
1495 					 * buffer_info->page */
1496 					skb_put(skb, length);
1497 				} else {
1498 					skb_fill_page_desc(skb, 0,
1499 					                   buffer_info->page, 0,
1500 				                           length);
1501 					e1000_consume_page(buffer_info, skb,
1502 					                   length);
1503 				}
1504 			}
1505 		}
1506 
1507 		/* Receive Checksum Offload XXX recompute due to CRC strip? */
1508 		e1000_rx_checksum(adapter, staterr,
1509 				  rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
1510 
1511 		e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1512 
1513 		/* probably a little skewed due to removing CRC */
1514 		total_rx_bytes += skb->len;
1515 		total_rx_packets++;
1516 
1517 		/* eth type trans needs skb->data to point to something */
1518 		if (!pskb_may_pull(skb, ETH_HLEN)) {
1519 			e_err("pskb_may_pull failed.\n");
1520 			dev_kfree_skb_irq(skb);
1521 			goto next_desc;
1522 		}
1523 
1524 		e1000_receive_skb(adapter, netdev, skb, staterr,
1525 				  rx_desc->wb.upper.vlan);
1526 
1527 next_desc:
1528 		rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1529 
1530 		/* return some buffers to hardware, one at a time is too slow */
1531 		if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1532 			adapter->alloc_rx_buf(rx_ring, cleaned_count,
1533 					      GFP_ATOMIC);
1534 			cleaned_count = 0;
1535 		}
1536 
1537 		/* use prefetched values */
1538 		rx_desc = next_rxd;
1539 		buffer_info = next_buffer;
1540 
1541 		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1542 	}
1543 	rx_ring->next_to_clean = i;
1544 
1545 	cleaned_count = e1000_desc_unused(rx_ring);
1546 	if (cleaned_count)
1547 		adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1548 
1549 	adapter->total_rx_bytes += total_rx_bytes;
1550 	adapter->total_rx_packets += total_rx_packets;
1551 	return cleaned;
1552 }
1553 
1554 /**
1555  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1556  * @rx_ring: Rx descriptor ring
1557  **/
1558 static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1559 {
1560 	struct e1000_adapter *adapter = rx_ring->adapter;
1561 	struct e1000_buffer *buffer_info;
1562 	struct e1000_ps_page *ps_page;
1563 	struct pci_dev *pdev = adapter->pdev;
1564 	unsigned int i, j;
1565 
1566 	/* Free all the Rx ring sk_buffs */
1567 	for (i = 0; i < rx_ring->count; i++) {
1568 		buffer_info = &rx_ring->buffer_info[i];
1569 		if (buffer_info->dma) {
1570 			if (adapter->clean_rx == e1000_clean_rx_irq)
1571 				dma_unmap_single(&pdev->dev, buffer_info->dma,
1572 						 adapter->rx_buffer_len,
1573 						 DMA_FROM_DEVICE);
1574 			else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1575 				dma_unmap_page(&pdev->dev, buffer_info->dma,
1576 				               PAGE_SIZE,
1577 					       DMA_FROM_DEVICE);
1578 			else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1579 				dma_unmap_single(&pdev->dev, buffer_info->dma,
1580 						 adapter->rx_ps_bsize0,
1581 						 DMA_FROM_DEVICE);
1582 			buffer_info->dma = 0;
1583 		}
1584 
1585 		if (buffer_info->page) {
1586 			put_page(buffer_info->page);
1587 			buffer_info->page = NULL;
1588 		}
1589 
1590 		if (buffer_info->skb) {
1591 			dev_kfree_skb(buffer_info->skb);
1592 			buffer_info->skb = NULL;
1593 		}
1594 
1595 		for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1596 			ps_page = &buffer_info->ps_pages[j];
1597 			if (!ps_page->page)
1598 				break;
1599 			dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1600 				       DMA_FROM_DEVICE);
1601 			ps_page->dma = 0;
1602 			put_page(ps_page->page);
1603 			ps_page->page = NULL;
1604 		}
1605 	}
1606 
1607 	/* there also may be some cached data from a chained receive */
1608 	if (rx_ring->rx_skb_top) {
1609 		dev_kfree_skb(rx_ring->rx_skb_top);
1610 		rx_ring->rx_skb_top = NULL;
1611 	}
1612 
1613 	/* Zero out the descriptor ring */
1614 	memset(rx_ring->desc, 0, rx_ring->size);
1615 
1616 	rx_ring->next_to_clean = 0;
1617 	rx_ring->next_to_use = 0;
1618 	adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1619 
1620 	writel(0, rx_ring->head);
1621 	writel(0, rx_ring->tail);
1622 }
1623 
1624 static void e1000e_downshift_workaround(struct work_struct *work)
1625 {
1626 	struct e1000_adapter *adapter = container_of(work,
1627 					struct e1000_adapter, downshift_task);
1628 
1629 	if (test_bit(__E1000_DOWN, &adapter->state))
1630 		return;
1631 
1632 	e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1633 }
1634 
1635 /**
1636  * e1000_intr_msi - Interrupt Handler
1637  * @irq: interrupt number
1638  * @data: pointer to a network interface device structure
1639  **/
1640 static irqreturn_t e1000_intr_msi(int irq, void *data)
1641 {
1642 	struct net_device *netdev = data;
1643 	struct e1000_adapter *adapter = netdev_priv(netdev);
1644 	struct e1000_hw *hw = &adapter->hw;
1645 	u32 icr = er32(ICR);
1646 
1647 	/*
1648 	 * read ICR disables interrupts using IAM
1649 	 */
1650 
1651 	if (icr & E1000_ICR_LSC) {
1652 		hw->mac.get_link_status = 1;
1653 		/*
1654 		 * ICH8 workaround-- Call gig speed drop workaround on cable
1655 		 * disconnect (LSC) before accessing any PHY registers
1656 		 */
1657 		if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1658 		    (!(er32(STATUS) & E1000_STATUS_LU)))
1659 			schedule_work(&adapter->downshift_task);
1660 
1661 		/*
1662 		 * 80003ES2LAN workaround-- For packet buffer work-around on
1663 		 * link down event; disable receives here in the ISR and reset
1664 		 * adapter in watchdog
1665 		 */
1666 		if (netif_carrier_ok(netdev) &&
1667 		    adapter->flags & FLAG_RX_NEEDS_RESTART) {
1668 			/* disable receives */
1669 			u32 rctl = er32(RCTL);
1670 			ew32(RCTL, rctl & ~E1000_RCTL_EN);
1671 			adapter->flags |= FLAG_RX_RESTART_NOW;
1672 		}
1673 		/* guard against interrupt when we're going down */
1674 		if (!test_bit(__E1000_DOWN, &adapter->state))
1675 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
1676 	}
1677 
1678 	if (napi_schedule_prep(&adapter->napi)) {
1679 		adapter->total_tx_bytes = 0;
1680 		adapter->total_tx_packets = 0;
1681 		adapter->total_rx_bytes = 0;
1682 		adapter->total_rx_packets = 0;
1683 		__napi_schedule(&adapter->napi);
1684 	}
1685 
1686 	return IRQ_HANDLED;
1687 }
1688 
1689 /**
1690  * e1000_intr - Interrupt Handler
1691  * @irq: interrupt number
1692  * @data: pointer to a network interface device structure
1693  **/
1694 static irqreturn_t e1000_intr(int irq, void *data)
1695 {
1696 	struct net_device *netdev = data;
1697 	struct e1000_adapter *adapter = netdev_priv(netdev);
1698 	struct e1000_hw *hw = &adapter->hw;
1699 	u32 rctl, icr = er32(ICR);
1700 
1701 	if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1702 		return IRQ_NONE;  /* Not our interrupt */
1703 
1704 	/*
1705 	 * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1706 	 * not set, then the adapter didn't send an interrupt
1707 	 */
1708 	if (!(icr & E1000_ICR_INT_ASSERTED))
1709 		return IRQ_NONE;
1710 
1711 	/*
1712 	 * Interrupt Auto-Mask...upon reading ICR,
1713 	 * interrupts are masked.  No need for the
1714 	 * IMC write
1715 	 */
1716 
1717 	if (icr & E1000_ICR_LSC) {
1718 		hw->mac.get_link_status = 1;
1719 		/*
1720 		 * ICH8 workaround-- Call gig speed drop workaround on cable
1721 		 * disconnect (LSC) before accessing any PHY registers
1722 		 */
1723 		if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1724 		    (!(er32(STATUS) & E1000_STATUS_LU)))
1725 			schedule_work(&adapter->downshift_task);
1726 
1727 		/*
1728 		 * 80003ES2LAN workaround--
1729 		 * For packet buffer work-around on link down event;
1730 		 * disable receives here in the ISR and
1731 		 * reset adapter in watchdog
1732 		 */
1733 		if (netif_carrier_ok(netdev) &&
1734 		    (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1735 			/* disable receives */
1736 			rctl = er32(RCTL);
1737 			ew32(RCTL, rctl & ~E1000_RCTL_EN);
1738 			adapter->flags |= FLAG_RX_RESTART_NOW;
1739 		}
1740 		/* guard against interrupt when we're going down */
1741 		if (!test_bit(__E1000_DOWN, &adapter->state))
1742 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
1743 	}
1744 
1745 	if (napi_schedule_prep(&adapter->napi)) {
1746 		adapter->total_tx_bytes = 0;
1747 		adapter->total_tx_packets = 0;
1748 		adapter->total_rx_bytes = 0;
1749 		adapter->total_rx_packets = 0;
1750 		__napi_schedule(&adapter->napi);
1751 	}
1752 
1753 	return IRQ_HANDLED;
1754 }
1755 
1756 static irqreturn_t e1000_msix_other(int irq, void *data)
1757 {
1758 	struct net_device *netdev = data;
1759 	struct e1000_adapter *adapter = netdev_priv(netdev);
1760 	struct e1000_hw *hw = &adapter->hw;
1761 	u32 icr = er32(ICR);
1762 
1763 	if (!(icr & E1000_ICR_INT_ASSERTED)) {
1764 		if (!test_bit(__E1000_DOWN, &adapter->state))
1765 			ew32(IMS, E1000_IMS_OTHER);
1766 		return IRQ_NONE;
1767 	}
1768 
1769 	if (icr & adapter->eiac_mask)
1770 		ew32(ICS, (icr & adapter->eiac_mask));
1771 
1772 	if (icr & E1000_ICR_OTHER) {
1773 		if (!(icr & E1000_ICR_LSC))
1774 			goto no_link_interrupt;
1775 		hw->mac.get_link_status = 1;
1776 		/* guard against interrupt when we're going down */
1777 		if (!test_bit(__E1000_DOWN, &adapter->state))
1778 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
1779 	}
1780 
1781 no_link_interrupt:
1782 	if (!test_bit(__E1000_DOWN, &adapter->state))
1783 		ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1784 
1785 	return IRQ_HANDLED;
1786 }
1787 
1788 
1789 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1790 {
1791 	struct net_device *netdev = data;
1792 	struct e1000_adapter *adapter = netdev_priv(netdev);
1793 	struct e1000_hw *hw = &adapter->hw;
1794 	struct e1000_ring *tx_ring = adapter->tx_ring;
1795 
1796 
1797 	adapter->total_tx_bytes = 0;
1798 	adapter->total_tx_packets = 0;
1799 
1800 	if (!e1000_clean_tx_irq(tx_ring))
1801 		/* Ring was not completely cleaned, so fire another interrupt */
1802 		ew32(ICS, tx_ring->ims_val);
1803 
1804 	return IRQ_HANDLED;
1805 }
1806 
1807 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1808 {
1809 	struct net_device *netdev = data;
1810 	struct e1000_adapter *adapter = netdev_priv(netdev);
1811 	struct e1000_ring *rx_ring = adapter->rx_ring;
1812 
1813 	/* Write the ITR value calculated at the end of the
1814 	 * previous interrupt.
1815 	 */
1816 	if (rx_ring->set_itr) {
1817 		writel(1000000000 / (rx_ring->itr_val * 256),
1818 		       rx_ring->itr_register);
1819 		rx_ring->set_itr = 0;
1820 	}
1821 
1822 	if (napi_schedule_prep(&adapter->napi)) {
1823 		adapter->total_rx_bytes = 0;
1824 		adapter->total_rx_packets = 0;
1825 		__napi_schedule(&adapter->napi);
1826 	}
1827 	return IRQ_HANDLED;
1828 }
1829 
1830 /**
1831  * e1000_configure_msix - Configure MSI-X hardware
1832  *
1833  * e1000_configure_msix sets up the hardware to properly
1834  * generate MSI-X interrupts.
1835  **/
1836 static void e1000_configure_msix(struct e1000_adapter *adapter)
1837 {
1838 	struct e1000_hw *hw = &adapter->hw;
1839 	struct e1000_ring *rx_ring = adapter->rx_ring;
1840 	struct e1000_ring *tx_ring = adapter->tx_ring;
1841 	int vector = 0;
1842 	u32 ctrl_ext, ivar = 0;
1843 
1844 	adapter->eiac_mask = 0;
1845 
1846 	/* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1847 	if (hw->mac.type == e1000_82574) {
1848 		u32 rfctl = er32(RFCTL);
1849 		rfctl |= E1000_RFCTL_ACK_DIS;
1850 		ew32(RFCTL, rfctl);
1851 	}
1852 
1853 #define E1000_IVAR_INT_ALLOC_VALID	0x8
1854 	/* Configure Rx vector */
1855 	rx_ring->ims_val = E1000_IMS_RXQ0;
1856 	adapter->eiac_mask |= rx_ring->ims_val;
1857 	if (rx_ring->itr_val)
1858 		writel(1000000000 / (rx_ring->itr_val * 256),
1859 		       rx_ring->itr_register);
1860 	else
1861 		writel(1, rx_ring->itr_register);
1862 	ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1863 
1864 	/* Configure Tx vector */
1865 	tx_ring->ims_val = E1000_IMS_TXQ0;
1866 	vector++;
1867 	if (tx_ring->itr_val)
1868 		writel(1000000000 / (tx_ring->itr_val * 256),
1869 		       tx_ring->itr_register);
1870 	else
1871 		writel(1, tx_ring->itr_register);
1872 	adapter->eiac_mask |= tx_ring->ims_val;
1873 	ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1874 
1875 	/* set vector for Other Causes, e.g. link changes */
1876 	vector++;
1877 	ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1878 	if (rx_ring->itr_val)
1879 		writel(1000000000 / (rx_ring->itr_val * 256),
1880 		       hw->hw_addr + E1000_EITR_82574(vector));
1881 	else
1882 		writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1883 
1884 	/* Cause Tx interrupts on every write back */
1885 	ivar |= (1 << 31);
1886 
1887 	ew32(IVAR, ivar);
1888 
1889 	/* enable MSI-X PBA support */
1890 	ctrl_ext = er32(CTRL_EXT);
1891 	ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1892 
1893 	/* Auto-Mask Other interrupts upon ICR read */
1894 #define E1000_EIAC_MASK_82574   0x01F00000
1895 	ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1896 	ctrl_ext |= E1000_CTRL_EXT_EIAME;
1897 	ew32(CTRL_EXT, ctrl_ext);
1898 	e1e_flush();
1899 }
1900 
1901 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1902 {
1903 	if (adapter->msix_entries) {
1904 		pci_disable_msix(adapter->pdev);
1905 		kfree(adapter->msix_entries);
1906 		adapter->msix_entries = NULL;
1907 	} else if (adapter->flags & FLAG_MSI_ENABLED) {
1908 		pci_disable_msi(adapter->pdev);
1909 		adapter->flags &= ~FLAG_MSI_ENABLED;
1910 	}
1911 }
1912 
1913 /**
1914  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1915  *
1916  * Attempt to configure interrupts using the best available
1917  * capabilities of the hardware and kernel.
1918  **/
1919 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1920 {
1921 	int err;
1922 	int i;
1923 
1924 	switch (adapter->int_mode) {
1925 	case E1000E_INT_MODE_MSIX:
1926 		if (adapter->flags & FLAG_HAS_MSIX) {
1927 			adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1928 			adapter->msix_entries = kcalloc(adapter->num_vectors,
1929 						      sizeof(struct msix_entry),
1930 						      GFP_KERNEL);
1931 			if (adapter->msix_entries) {
1932 				for (i = 0; i < adapter->num_vectors; i++)
1933 					adapter->msix_entries[i].entry = i;
1934 
1935 				err = pci_enable_msix(adapter->pdev,
1936 						      adapter->msix_entries,
1937 						      adapter->num_vectors);
1938 				if (err == 0)
1939 					return;
1940 			}
1941 			/* MSI-X failed, so fall through and try MSI */
1942 			e_err("Failed to initialize MSI-X interrupts.  Falling back to MSI interrupts.\n");
1943 			e1000e_reset_interrupt_capability(adapter);
1944 		}
1945 		adapter->int_mode = E1000E_INT_MODE_MSI;
1946 		/* Fall through */
1947 	case E1000E_INT_MODE_MSI:
1948 		if (!pci_enable_msi(adapter->pdev)) {
1949 			adapter->flags |= FLAG_MSI_ENABLED;
1950 		} else {
1951 			adapter->int_mode = E1000E_INT_MODE_LEGACY;
1952 			e_err("Failed to initialize MSI interrupts.  Falling back to legacy interrupts.\n");
1953 		}
1954 		/* Fall through */
1955 	case E1000E_INT_MODE_LEGACY:
1956 		/* Don't do anything; this is the system default */
1957 		break;
1958 	}
1959 
1960 	/* store the number of vectors being used */
1961 	adapter->num_vectors = 1;
1962 }
1963 
1964 /**
1965  * e1000_request_msix - Initialize MSI-X interrupts
1966  *
1967  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1968  * kernel.
1969  **/
1970 static int e1000_request_msix(struct e1000_adapter *adapter)
1971 {
1972 	struct net_device *netdev = adapter->netdev;
1973 	int err = 0, vector = 0;
1974 
1975 	if (strlen(netdev->name) < (IFNAMSIZ - 5))
1976 		snprintf(adapter->rx_ring->name,
1977 			 sizeof(adapter->rx_ring->name) - 1,
1978 			 "%s-rx-0", netdev->name);
1979 	else
1980 		memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1981 	err = request_irq(adapter->msix_entries[vector].vector,
1982 			  e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1983 			  netdev);
1984 	if (err)
1985 		goto out;
1986 	adapter->rx_ring->itr_register = adapter->hw.hw_addr +
1987 	    E1000_EITR_82574(vector);
1988 	adapter->rx_ring->itr_val = adapter->itr;
1989 	vector++;
1990 
1991 	if (strlen(netdev->name) < (IFNAMSIZ - 5))
1992 		snprintf(adapter->tx_ring->name,
1993 			 sizeof(adapter->tx_ring->name) - 1,
1994 			 "%s-tx-0", netdev->name);
1995 	else
1996 		memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1997 	err = request_irq(adapter->msix_entries[vector].vector,
1998 			  e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1999 			  netdev);
2000 	if (err)
2001 		goto out;
2002 	adapter->tx_ring->itr_register = adapter->hw.hw_addr +
2003 	    E1000_EITR_82574(vector);
2004 	adapter->tx_ring->itr_val = adapter->itr;
2005 	vector++;
2006 
2007 	err = request_irq(adapter->msix_entries[vector].vector,
2008 			  e1000_msix_other, 0, netdev->name, netdev);
2009 	if (err)
2010 		goto out;
2011 
2012 	e1000_configure_msix(adapter);
2013 	return 0;
2014 out:
2015 	return err;
2016 }
2017 
2018 /**
2019  * e1000_request_irq - initialize interrupts
2020  *
2021  * Attempts to configure interrupts using the best available
2022  * capabilities of the hardware and kernel.
2023  **/
2024 static int e1000_request_irq(struct e1000_adapter *adapter)
2025 {
2026 	struct net_device *netdev = adapter->netdev;
2027 	int err;
2028 
2029 	if (adapter->msix_entries) {
2030 		err = e1000_request_msix(adapter);
2031 		if (!err)
2032 			return err;
2033 		/* fall back to MSI */
2034 		e1000e_reset_interrupt_capability(adapter);
2035 		adapter->int_mode = E1000E_INT_MODE_MSI;
2036 		e1000e_set_interrupt_capability(adapter);
2037 	}
2038 	if (adapter->flags & FLAG_MSI_ENABLED) {
2039 		err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2040 				  netdev->name, netdev);
2041 		if (!err)
2042 			return err;
2043 
2044 		/* fall back to legacy interrupt */
2045 		e1000e_reset_interrupt_capability(adapter);
2046 		adapter->int_mode = E1000E_INT_MODE_LEGACY;
2047 	}
2048 
2049 	err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2050 			  netdev->name, netdev);
2051 	if (err)
2052 		e_err("Unable to allocate interrupt, Error: %d\n", err);
2053 
2054 	return err;
2055 }
2056 
2057 static void e1000_free_irq(struct e1000_adapter *adapter)
2058 {
2059 	struct net_device *netdev = adapter->netdev;
2060 
2061 	if (adapter->msix_entries) {
2062 		int vector = 0;
2063 
2064 		free_irq(adapter->msix_entries[vector].vector, netdev);
2065 		vector++;
2066 
2067 		free_irq(adapter->msix_entries[vector].vector, netdev);
2068 		vector++;
2069 
2070 		/* Other Causes interrupt vector */
2071 		free_irq(adapter->msix_entries[vector].vector, netdev);
2072 		return;
2073 	}
2074 
2075 	free_irq(adapter->pdev->irq, netdev);
2076 }
2077 
2078 /**
2079  * e1000_irq_disable - Mask off interrupt generation on the NIC
2080  **/
2081 static void e1000_irq_disable(struct e1000_adapter *adapter)
2082 {
2083 	struct e1000_hw *hw = &adapter->hw;
2084 
2085 	ew32(IMC, ~0);
2086 	if (adapter->msix_entries)
2087 		ew32(EIAC_82574, 0);
2088 	e1e_flush();
2089 
2090 	if (adapter->msix_entries) {
2091 		int i;
2092 		for (i = 0; i < adapter->num_vectors; i++)
2093 			synchronize_irq(adapter->msix_entries[i].vector);
2094 	} else {
2095 		synchronize_irq(adapter->pdev->irq);
2096 	}
2097 }
2098 
2099 /**
2100  * e1000_irq_enable - Enable default interrupt generation settings
2101  **/
2102 static void e1000_irq_enable(struct e1000_adapter *adapter)
2103 {
2104 	struct e1000_hw *hw = &adapter->hw;
2105 
2106 	if (adapter->msix_entries) {
2107 		ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2108 		ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
2109 	} else {
2110 		ew32(IMS, IMS_ENABLE_MASK);
2111 	}
2112 	e1e_flush();
2113 }
2114 
2115 /**
2116  * e1000e_get_hw_control - get control of the h/w from f/w
2117  * @adapter: address of board private structure
2118  *
2119  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2120  * For ASF and Pass Through versions of f/w this means that
2121  * the driver is loaded. For AMT version (only with 82573)
2122  * of the f/w this means that the network i/f is open.
2123  **/
2124 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2125 {
2126 	struct e1000_hw *hw = &adapter->hw;
2127 	u32 ctrl_ext;
2128 	u32 swsm;
2129 
2130 	/* Let firmware know the driver has taken over */
2131 	if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2132 		swsm = er32(SWSM);
2133 		ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2134 	} else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2135 		ctrl_ext = er32(CTRL_EXT);
2136 		ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2137 	}
2138 }
2139 
2140 /**
2141  * e1000e_release_hw_control - release control of the h/w to f/w
2142  * @adapter: address of board private structure
2143  *
2144  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2145  * For ASF and Pass Through versions of f/w this means that the
2146  * driver is no longer loaded. For AMT version (only with 82573) i
2147  * of the f/w this means that the network i/f is closed.
2148  *
2149  **/
2150 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2151 {
2152 	struct e1000_hw *hw = &adapter->hw;
2153 	u32 ctrl_ext;
2154 	u32 swsm;
2155 
2156 	/* Let firmware taken over control of h/w */
2157 	if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2158 		swsm = er32(SWSM);
2159 		ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2160 	} else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2161 		ctrl_ext = er32(CTRL_EXT);
2162 		ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2163 	}
2164 }
2165 
2166 /**
2167  * @e1000_alloc_ring - allocate memory for a ring structure
2168  **/
2169 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2170 				struct e1000_ring *ring)
2171 {
2172 	struct pci_dev *pdev = adapter->pdev;
2173 
2174 	ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2175 					GFP_KERNEL);
2176 	if (!ring->desc)
2177 		return -ENOMEM;
2178 
2179 	return 0;
2180 }
2181 
2182 /**
2183  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2184  * @tx_ring: Tx descriptor ring
2185  *
2186  * Return 0 on success, negative on failure
2187  **/
2188 int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2189 {
2190 	struct e1000_adapter *adapter = tx_ring->adapter;
2191 	int err = -ENOMEM, size;
2192 
2193 	size = sizeof(struct e1000_buffer) * tx_ring->count;
2194 	tx_ring->buffer_info = vzalloc(size);
2195 	if (!tx_ring->buffer_info)
2196 		goto err;
2197 
2198 	/* round up to nearest 4K */
2199 	tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2200 	tx_ring->size = ALIGN(tx_ring->size, 4096);
2201 
2202 	err = e1000_alloc_ring_dma(adapter, tx_ring);
2203 	if (err)
2204 		goto err;
2205 
2206 	tx_ring->next_to_use = 0;
2207 	tx_ring->next_to_clean = 0;
2208 
2209 	return 0;
2210 err:
2211 	vfree(tx_ring->buffer_info);
2212 	e_err("Unable to allocate memory for the transmit descriptor ring\n");
2213 	return err;
2214 }
2215 
2216 /**
2217  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2218  * @rx_ring: Rx descriptor ring
2219  *
2220  * Returns 0 on success, negative on failure
2221  **/
2222 int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2223 {
2224 	struct e1000_adapter *adapter = rx_ring->adapter;
2225 	struct e1000_buffer *buffer_info;
2226 	int i, size, desc_len, err = -ENOMEM;
2227 
2228 	size = sizeof(struct e1000_buffer) * rx_ring->count;
2229 	rx_ring->buffer_info = vzalloc(size);
2230 	if (!rx_ring->buffer_info)
2231 		goto err;
2232 
2233 	for (i = 0; i < rx_ring->count; i++) {
2234 		buffer_info = &rx_ring->buffer_info[i];
2235 		buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2236 						sizeof(struct e1000_ps_page),
2237 						GFP_KERNEL);
2238 		if (!buffer_info->ps_pages)
2239 			goto err_pages;
2240 	}
2241 
2242 	desc_len = sizeof(union e1000_rx_desc_packet_split);
2243 
2244 	/* Round up to nearest 4K */
2245 	rx_ring->size = rx_ring->count * desc_len;
2246 	rx_ring->size = ALIGN(rx_ring->size, 4096);
2247 
2248 	err = e1000_alloc_ring_dma(adapter, rx_ring);
2249 	if (err)
2250 		goto err_pages;
2251 
2252 	rx_ring->next_to_clean = 0;
2253 	rx_ring->next_to_use = 0;
2254 	rx_ring->rx_skb_top = NULL;
2255 
2256 	return 0;
2257 
2258 err_pages:
2259 	for (i = 0; i < rx_ring->count; i++) {
2260 		buffer_info = &rx_ring->buffer_info[i];
2261 		kfree(buffer_info->ps_pages);
2262 	}
2263 err:
2264 	vfree(rx_ring->buffer_info);
2265 	e_err("Unable to allocate memory for the receive descriptor ring\n");
2266 	return err;
2267 }
2268 
2269 /**
2270  * e1000_clean_tx_ring - Free Tx Buffers
2271  * @tx_ring: Tx descriptor ring
2272  **/
2273 static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2274 {
2275 	struct e1000_adapter *adapter = tx_ring->adapter;
2276 	struct e1000_buffer *buffer_info;
2277 	unsigned long size;
2278 	unsigned int i;
2279 
2280 	for (i = 0; i < tx_ring->count; i++) {
2281 		buffer_info = &tx_ring->buffer_info[i];
2282 		e1000_put_txbuf(tx_ring, buffer_info);
2283 	}
2284 
2285 	netdev_reset_queue(adapter->netdev);
2286 	size = sizeof(struct e1000_buffer) * tx_ring->count;
2287 	memset(tx_ring->buffer_info, 0, size);
2288 
2289 	memset(tx_ring->desc, 0, tx_ring->size);
2290 
2291 	tx_ring->next_to_use = 0;
2292 	tx_ring->next_to_clean = 0;
2293 
2294 	writel(0, tx_ring->head);
2295 	writel(0, tx_ring->tail);
2296 }
2297 
2298 /**
2299  * e1000e_free_tx_resources - Free Tx Resources per Queue
2300  * @tx_ring: Tx descriptor ring
2301  *
2302  * Free all transmit software resources
2303  **/
2304 void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2305 {
2306 	struct e1000_adapter *adapter = tx_ring->adapter;
2307 	struct pci_dev *pdev = adapter->pdev;
2308 
2309 	e1000_clean_tx_ring(tx_ring);
2310 
2311 	vfree(tx_ring->buffer_info);
2312 	tx_ring->buffer_info = NULL;
2313 
2314 	dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2315 			  tx_ring->dma);
2316 	tx_ring->desc = NULL;
2317 }
2318 
2319 /**
2320  * e1000e_free_rx_resources - Free Rx Resources
2321  * @rx_ring: Rx descriptor ring
2322  *
2323  * Free all receive software resources
2324  **/
2325 void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2326 {
2327 	struct e1000_adapter *adapter = rx_ring->adapter;
2328 	struct pci_dev *pdev = adapter->pdev;
2329 	int i;
2330 
2331 	e1000_clean_rx_ring(rx_ring);
2332 
2333 	for (i = 0; i < rx_ring->count; i++)
2334 		kfree(rx_ring->buffer_info[i].ps_pages);
2335 
2336 	vfree(rx_ring->buffer_info);
2337 	rx_ring->buffer_info = NULL;
2338 
2339 	dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2340 			  rx_ring->dma);
2341 	rx_ring->desc = NULL;
2342 }
2343 
2344 /**
2345  * e1000_update_itr - update the dynamic ITR value based on statistics
2346  * @adapter: pointer to adapter
2347  * @itr_setting: current adapter->itr
2348  * @packets: the number of packets during this measurement interval
2349  * @bytes: the number of bytes during this measurement interval
2350  *
2351  *      Stores a new ITR value based on packets and byte
2352  *      counts during the last interrupt.  The advantage of per interrupt
2353  *      computation is faster updates and more accurate ITR for the current
2354  *      traffic pattern.  Constants in this function were computed
2355  *      based on theoretical maximum wire speed and thresholds were set based
2356  *      on testing data as well as attempting to minimize response time
2357  *      while increasing bulk throughput.  This functionality is controlled
2358  *      by the InterruptThrottleRate module parameter.
2359  **/
2360 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2361 				     u16 itr_setting, int packets,
2362 				     int bytes)
2363 {
2364 	unsigned int retval = itr_setting;
2365 
2366 	if (packets == 0)
2367 		goto update_itr_done;
2368 
2369 	switch (itr_setting) {
2370 	case lowest_latency:
2371 		/* handle TSO and jumbo frames */
2372 		if (bytes/packets > 8000)
2373 			retval = bulk_latency;
2374 		else if ((packets < 5) && (bytes > 512))
2375 			retval = low_latency;
2376 		break;
2377 	case low_latency:  /* 50 usec aka 20000 ints/s */
2378 		if (bytes > 10000) {
2379 			/* this if handles the TSO accounting */
2380 			if (bytes/packets > 8000)
2381 				retval = bulk_latency;
2382 			else if ((packets < 10) || ((bytes/packets) > 1200))
2383 				retval = bulk_latency;
2384 			else if ((packets > 35))
2385 				retval = lowest_latency;
2386 		} else if (bytes/packets > 2000) {
2387 			retval = bulk_latency;
2388 		} else if (packets <= 2 && bytes < 512) {
2389 			retval = lowest_latency;
2390 		}
2391 		break;
2392 	case bulk_latency: /* 250 usec aka 4000 ints/s */
2393 		if (bytes > 25000) {
2394 			if (packets > 35)
2395 				retval = low_latency;
2396 		} else if (bytes < 6000) {
2397 			retval = low_latency;
2398 		}
2399 		break;
2400 	}
2401 
2402 update_itr_done:
2403 	return retval;
2404 }
2405 
2406 static void e1000_set_itr(struct e1000_adapter *adapter)
2407 {
2408 	struct e1000_hw *hw = &adapter->hw;
2409 	u16 current_itr;
2410 	u32 new_itr = adapter->itr;
2411 
2412 	/* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2413 	if (adapter->link_speed != SPEED_1000) {
2414 		current_itr = 0;
2415 		new_itr = 4000;
2416 		goto set_itr_now;
2417 	}
2418 
2419 	if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2420 		new_itr = 0;
2421 		goto set_itr_now;
2422 	}
2423 
2424 	adapter->tx_itr = e1000_update_itr(adapter,
2425 				    adapter->tx_itr,
2426 				    adapter->total_tx_packets,
2427 				    adapter->total_tx_bytes);
2428 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
2429 	if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2430 		adapter->tx_itr = low_latency;
2431 
2432 	adapter->rx_itr = e1000_update_itr(adapter,
2433 				    adapter->rx_itr,
2434 				    adapter->total_rx_packets,
2435 				    adapter->total_rx_bytes);
2436 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
2437 	if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2438 		adapter->rx_itr = low_latency;
2439 
2440 	current_itr = max(adapter->rx_itr, adapter->tx_itr);
2441 
2442 	switch (current_itr) {
2443 	/* counts and packets in update_itr are dependent on these numbers */
2444 	case lowest_latency:
2445 		new_itr = 70000;
2446 		break;
2447 	case low_latency:
2448 		new_itr = 20000; /* aka hwitr = ~200 */
2449 		break;
2450 	case bulk_latency:
2451 		new_itr = 4000;
2452 		break;
2453 	default:
2454 		break;
2455 	}
2456 
2457 set_itr_now:
2458 	if (new_itr != adapter->itr) {
2459 		/*
2460 		 * this attempts to bias the interrupt rate towards Bulk
2461 		 * by adding intermediate steps when interrupt rate is
2462 		 * increasing
2463 		 */
2464 		new_itr = new_itr > adapter->itr ?
2465 			     min(adapter->itr + (new_itr >> 2), new_itr) :
2466 			     new_itr;
2467 		adapter->itr = new_itr;
2468 		adapter->rx_ring->itr_val = new_itr;
2469 		if (adapter->msix_entries)
2470 			adapter->rx_ring->set_itr = 1;
2471 		else
2472 			if (new_itr)
2473 				ew32(ITR, 1000000000 / (new_itr * 256));
2474 			else
2475 				ew32(ITR, 0);
2476 	}
2477 }
2478 
2479 /**
2480  * e1000_alloc_queues - Allocate memory for all rings
2481  * @adapter: board private structure to initialize
2482  **/
2483 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2484 {
2485 	int size = sizeof(struct e1000_ring);
2486 
2487 	adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2488 	if (!adapter->tx_ring)
2489 		goto err;
2490 	adapter->tx_ring->count = adapter->tx_ring_count;
2491 	adapter->tx_ring->adapter = adapter;
2492 
2493 	adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2494 	if (!adapter->rx_ring)
2495 		goto err;
2496 	adapter->rx_ring->count = adapter->rx_ring_count;
2497 	adapter->rx_ring->adapter = adapter;
2498 
2499 	return 0;
2500 err:
2501 	e_err("Unable to allocate memory for queues\n");
2502 	kfree(adapter->rx_ring);
2503 	kfree(adapter->tx_ring);
2504 	return -ENOMEM;
2505 }
2506 
2507 /**
2508  * e1000_clean - NAPI Rx polling callback
2509  * @napi: struct associated with this polling callback
2510  * @budget: amount of packets driver is allowed to process this poll
2511  **/
2512 static int e1000_clean(struct napi_struct *napi, int budget)
2513 {
2514 	struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2515 	struct e1000_hw *hw = &adapter->hw;
2516 	struct net_device *poll_dev = adapter->netdev;
2517 	int tx_cleaned = 1, work_done = 0;
2518 
2519 	adapter = netdev_priv(poll_dev);
2520 
2521 	if (adapter->msix_entries &&
2522 	    !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2523 		goto clean_rx;
2524 
2525 	tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2526 
2527 clean_rx:
2528 	adapter->clean_rx(adapter->rx_ring, &work_done, budget);
2529 
2530 	if (!tx_cleaned)
2531 		work_done = budget;
2532 
2533 	/* If budget not fully consumed, exit the polling mode */
2534 	if (work_done < budget) {
2535 		if (adapter->itr_setting & 3)
2536 			e1000_set_itr(adapter);
2537 		napi_complete(napi);
2538 		if (!test_bit(__E1000_DOWN, &adapter->state)) {
2539 			if (adapter->msix_entries)
2540 				ew32(IMS, adapter->rx_ring->ims_val);
2541 			else
2542 				e1000_irq_enable(adapter);
2543 		}
2544 	}
2545 
2546 	return work_done;
2547 }
2548 
2549 static int e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2550 {
2551 	struct e1000_adapter *adapter = netdev_priv(netdev);
2552 	struct e1000_hw *hw = &adapter->hw;
2553 	u32 vfta, index;
2554 
2555 	/* don't update vlan cookie if already programmed */
2556 	if ((adapter->hw.mng_cookie.status &
2557 	     E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2558 	    (vid == adapter->mng_vlan_id))
2559 		return 0;
2560 
2561 	/* add VID to filter table */
2562 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2563 		index = (vid >> 5) & 0x7F;
2564 		vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2565 		vfta |= (1 << (vid & 0x1F));
2566 		hw->mac.ops.write_vfta(hw, index, vfta);
2567 	}
2568 
2569 	set_bit(vid, adapter->active_vlans);
2570 
2571 	return 0;
2572 }
2573 
2574 static int e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2575 {
2576 	struct e1000_adapter *adapter = netdev_priv(netdev);
2577 	struct e1000_hw *hw = &adapter->hw;
2578 	u32 vfta, index;
2579 
2580 	if ((adapter->hw.mng_cookie.status &
2581 	     E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2582 	    (vid == adapter->mng_vlan_id)) {
2583 		/* release control to f/w */
2584 		e1000e_release_hw_control(adapter);
2585 		return 0;
2586 	}
2587 
2588 	/* remove VID from filter table */
2589 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2590 		index = (vid >> 5) & 0x7F;
2591 		vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2592 		vfta &= ~(1 << (vid & 0x1F));
2593 		hw->mac.ops.write_vfta(hw, index, vfta);
2594 	}
2595 
2596 	clear_bit(vid, adapter->active_vlans);
2597 
2598 	return 0;
2599 }
2600 
2601 /**
2602  * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2603  * @adapter: board private structure to initialize
2604  **/
2605 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2606 {
2607 	struct net_device *netdev = adapter->netdev;
2608 	struct e1000_hw *hw = &adapter->hw;
2609 	u32 rctl;
2610 
2611 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2612 		/* disable VLAN receive filtering */
2613 		rctl = er32(RCTL);
2614 		rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2615 		ew32(RCTL, rctl);
2616 
2617 		if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2618 			e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2619 			adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2620 		}
2621 	}
2622 }
2623 
2624 /**
2625  * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2626  * @adapter: board private structure to initialize
2627  **/
2628 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2629 {
2630 	struct e1000_hw *hw = &adapter->hw;
2631 	u32 rctl;
2632 
2633 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2634 		/* enable VLAN receive filtering */
2635 		rctl = er32(RCTL);
2636 		rctl |= E1000_RCTL_VFE;
2637 		rctl &= ~E1000_RCTL_CFIEN;
2638 		ew32(RCTL, rctl);
2639 	}
2640 }
2641 
2642 /**
2643  * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2644  * @adapter: board private structure to initialize
2645  **/
2646 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2647 {
2648 	struct e1000_hw *hw = &adapter->hw;
2649 	u32 ctrl;
2650 
2651 	/* disable VLAN tag insert/strip */
2652 	ctrl = er32(CTRL);
2653 	ctrl &= ~E1000_CTRL_VME;
2654 	ew32(CTRL, ctrl);
2655 }
2656 
2657 /**
2658  * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2659  * @adapter: board private structure to initialize
2660  **/
2661 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2662 {
2663 	struct e1000_hw *hw = &adapter->hw;
2664 	u32 ctrl;
2665 
2666 	/* enable VLAN tag insert/strip */
2667 	ctrl = er32(CTRL);
2668 	ctrl |= E1000_CTRL_VME;
2669 	ew32(CTRL, ctrl);
2670 }
2671 
2672 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2673 {
2674 	struct net_device *netdev = adapter->netdev;
2675 	u16 vid = adapter->hw.mng_cookie.vlan_id;
2676 	u16 old_vid = adapter->mng_vlan_id;
2677 
2678 	if (adapter->hw.mng_cookie.status &
2679 	    E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2680 		e1000_vlan_rx_add_vid(netdev, vid);
2681 		adapter->mng_vlan_id = vid;
2682 	}
2683 
2684 	if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2685 		e1000_vlan_rx_kill_vid(netdev, old_vid);
2686 }
2687 
2688 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2689 {
2690 	u16 vid;
2691 
2692 	e1000_vlan_rx_add_vid(adapter->netdev, 0);
2693 
2694 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2695 		e1000_vlan_rx_add_vid(adapter->netdev, vid);
2696 }
2697 
2698 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2699 {
2700 	struct e1000_hw *hw = &adapter->hw;
2701 	u32 manc, manc2h, mdef, i, j;
2702 
2703 	if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2704 		return;
2705 
2706 	manc = er32(MANC);
2707 
2708 	/*
2709 	 * enable receiving management packets to the host. this will probably
2710 	 * generate destination unreachable messages from the host OS, but
2711 	 * the packets will be handled on SMBUS
2712 	 */
2713 	manc |= E1000_MANC_EN_MNG2HOST;
2714 	manc2h = er32(MANC2H);
2715 
2716 	switch (hw->mac.type) {
2717 	default:
2718 		manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2719 		break;
2720 	case e1000_82574:
2721 	case e1000_82583:
2722 		/*
2723 		 * Check if IPMI pass-through decision filter already exists;
2724 		 * if so, enable it.
2725 		 */
2726 		for (i = 0, j = 0; i < 8; i++) {
2727 			mdef = er32(MDEF(i));
2728 
2729 			/* Ignore filters with anything other than IPMI ports */
2730 			if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2731 				continue;
2732 
2733 			/* Enable this decision filter in MANC2H */
2734 			if (mdef)
2735 				manc2h |= (1 << i);
2736 
2737 			j |= mdef;
2738 		}
2739 
2740 		if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2741 			break;
2742 
2743 		/* Create new decision filter in an empty filter */
2744 		for (i = 0, j = 0; i < 8; i++)
2745 			if (er32(MDEF(i)) == 0) {
2746 				ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2747 					       E1000_MDEF_PORT_664));
2748 				manc2h |= (1 << 1);
2749 				j++;
2750 				break;
2751 			}
2752 
2753 		if (!j)
2754 			e_warn("Unable to create IPMI pass-through filter\n");
2755 		break;
2756 	}
2757 
2758 	ew32(MANC2H, manc2h);
2759 	ew32(MANC, manc);
2760 }
2761 
2762 /**
2763  * e1000_configure_tx - Configure Transmit Unit after Reset
2764  * @adapter: board private structure
2765  *
2766  * Configure the Tx unit of the MAC after a reset.
2767  **/
2768 static void e1000_configure_tx(struct e1000_adapter *adapter)
2769 {
2770 	struct e1000_hw *hw = &adapter->hw;
2771 	struct e1000_ring *tx_ring = adapter->tx_ring;
2772 	u64 tdba;
2773 	u32 tdlen, tarc;
2774 
2775 	/* Setup the HW Tx Head and Tail descriptor pointers */
2776 	tdba = tx_ring->dma;
2777 	tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2778 	ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2779 	ew32(TDBAH, (tdba >> 32));
2780 	ew32(TDLEN, tdlen);
2781 	ew32(TDH, 0);
2782 	ew32(TDT, 0);
2783 	tx_ring->head = adapter->hw.hw_addr + E1000_TDH;
2784 	tx_ring->tail = adapter->hw.hw_addr + E1000_TDT;
2785 
2786 	/* Set the Tx Interrupt Delay register */
2787 	ew32(TIDV, adapter->tx_int_delay);
2788 	/* Tx irq moderation */
2789 	ew32(TADV, adapter->tx_abs_int_delay);
2790 
2791 	if (adapter->flags2 & FLAG2_DMA_BURST) {
2792 		u32 txdctl = er32(TXDCTL(0));
2793 		txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2794 			    E1000_TXDCTL_WTHRESH);
2795 		/*
2796 		 * set up some performance related parameters to encourage the
2797 		 * hardware to use the bus more efficiently in bursts, depends
2798 		 * on the tx_int_delay to be enabled,
2799 		 * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2800 		 * hthresh = 1 ==> prefetch when one or more available
2801 		 * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2802 		 * BEWARE: this seems to work but should be considered first if
2803 		 * there are Tx hangs or other Tx related bugs
2804 		 */
2805 		txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2806 		ew32(TXDCTL(0), txdctl);
2807 	}
2808 	/* erratum work around: set txdctl the same for both queues */
2809 	ew32(TXDCTL(1), er32(TXDCTL(0)));
2810 
2811 	if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2812 		tarc = er32(TARC(0));
2813 		/*
2814 		 * set the speed mode bit, we'll clear it if we're not at
2815 		 * gigabit link later
2816 		 */
2817 #define SPEED_MODE_BIT (1 << 21)
2818 		tarc |= SPEED_MODE_BIT;
2819 		ew32(TARC(0), tarc);
2820 	}
2821 
2822 	/* errata: program both queues to unweighted RR */
2823 	if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2824 		tarc = er32(TARC(0));
2825 		tarc |= 1;
2826 		ew32(TARC(0), tarc);
2827 		tarc = er32(TARC(1));
2828 		tarc |= 1;
2829 		ew32(TARC(1), tarc);
2830 	}
2831 
2832 	/* Setup Transmit Descriptor Settings for eop descriptor */
2833 	adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2834 
2835 	/* only set IDE if we are delaying interrupts using the timers */
2836 	if (adapter->tx_int_delay)
2837 		adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2838 
2839 	/* enable Report Status bit */
2840 	adapter->txd_cmd |= E1000_TXD_CMD_RS;
2841 
2842 	e1000e_config_collision_dist(hw);
2843 }
2844 
2845 /**
2846  * e1000_setup_rctl - configure the receive control registers
2847  * @adapter: Board private structure
2848  **/
2849 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2850 			   (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2851 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2852 {
2853 	struct e1000_hw *hw = &adapter->hw;
2854 	u32 rctl, rfctl;
2855 	u32 pages = 0;
2856 
2857 	/* Workaround Si errata on 82579 - configure jumbo frame flow */
2858 	if (hw->mac.type == e1000_pch2lan) {
2859 		s32 ret_val;
2860 
2861 		if (adapter->netdev->mtu > ETH_DATA_LEN)
2862 			ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2863 		else
2864 			ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2865 
2866 		if (ret_val)
2867 			e_dbg("failed to enable jumbo frame workaround mode\n");
2868 	}
2869 
2870 	/* Program MC offset vector base */
2871 	rctl = er32(RCTL);
2872 	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2873 	rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2874 		E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2875 		(adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2876 
2877 	/* Do not Store bad packets */
2878 	rctl &= ~E1000_RCTL_SBP;
2879 
2880 	/* Enable Long Packet receive */
2881 	if (adapter->netdev->mtu <= ETH_DATA_LEN)
2882 		rctl &= ~E1000_RCTL_LPE;
2883 	else
2884 		rctl |= E1000_RCTL_LPE;
2885 
2886 	/* Some systems expect that the CRC is included in SMBUS traffic. The
2887 	 * hardware strips the CRC before sending to both SMBUS (BMC) and to
2888 	 * host memory when this is enabled
2889 	 */
2890 	if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2891 		rctl |= E1000_RCTL_SECRC;
2892 
2893 	/* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2894 	if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2895 		u16 phy_data;
2896 
2897 		e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2898 		phy_data &= 0xfff8;
2899 		phy_data |= (1 << 2);
2900 		e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2901 
2902 		e1e_rphy(hw, 22, &phy_data);
2903 		phy_data &= 0x0fff;
2904 		phy_data |= (1 << 14);
2905 		e1e_wphy(hw, 0x10, 0x2823);
2906 		e1e_wphy(hw, 0x11, 0x0003);
2907 		e1e_wphy(hw, 22, phy_data);
2908 	}
2909 
2910 	/* Setup buffer sizes */
2911 	rctl &= ~E1000_RCTL_SZ_4096;
2912 	rctl |= E1000_RCTL_BSEX;
2913 	switch (adapter->rx_buffer_len) {
2914 	case 2048:
2915 	default:
2916 		rctl |= E1000_RCTL_SZ_2048;
2917 		rctl &= ~E1000_RCTL_BSEX;
2918 		break;
2919 	case 4096:
2920 		rctl |= E1000_RCTL_SZ_4096;
2921 		break;
2922 	case 8192:
2923 		rctl |= E1000_RCTL_SZ_8192;
2924 		break;
2925 	case 16384:
2926 		rctl |= E1000_RCTL_SZ_16384;
2927 		break;
2928 	}
2929 
2930 	/* Enable Extended Status in all Receive Descriptors */
2931 	rfctl = er32(RFCTL);
2932 	rfctl |= E1000_RFCTL_EXTEN;
2933 
2934 	/*
2935 	 * 82571 and greater support packet-split where the protocol
2936 	 * header is placed in skb->data and the packet data is
2937 	 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2938 	 * In the case of a non-split, skb->data is linearly filled,
2939 	 * followed by the page buffers.  Therefore, skb->data is
2940 	 * sized to hold the largest protocol header.
2941 	 *
2942 	 * allocations using alloc_page take too long for regular MTU
2943 	 * so only enable packet split for jumbo frames
2944 	 *
2945 	 * Using pages when the page size is greater than 16k wastes
2946 	 * a lot of memory, since we allocate 3 pages at all times
2947 	 * per packet.
2948 	 */
2949 	pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2950 	if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2951 		adapter->rx_ps_pages = pages;
2952 	else
2953 		adapter->rx_ps_pages = 0;
2954 
2955 	if (adapter->rx_ps_pages) {
2956 		u32 psrctl = 0;
2957 
2958 		/*
2959 		 * disable packet split support for IPv6 extension headers,
2960 		 * because some malformed IPv6 headers can hang the Rx
2961 		 */
2962 		rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2963 			  E1000_RFCTL_NEW_IPV6_EXT_DIS);
2964 
2965 		/* Enable Packet split descriptors */
2966 		rctl |= E1000_RCTL_DTYP_PS;
2967 
2968 		psrctl |= adapter->rx_ps_bsize0 >>
2969 			E1000_PSRCTL_BSIZE0_SHIFT;
2970 
2971 		switch (adapter->rx_ps_pages) {
2972 		case 3:
2973 			psrctl |= PAGE_SIZE <<
2974 				E1000_PSRCTL_BSIZE3_SHIFT;
2975 		case 2:
2976 			psrctl |= PAGE_SIZE <<
2977 				E1000_PSRCTL_BSIZE2_SHIFT;
2978 		case 1:
2979 			psrctl |= PAGE_SIZE >>
2980 				E1000_PSRCTL_BSIZE1_SHIFT;
2981 			break;
2982 		}
2983 
2984 		ew32(PSRCTL, psrctl);
2985 	}
2986 
2987 	ew32(RFCTL, rfctl);
2988 	ew32(RCTL, rctl);
2989 	/* just started the receive unit, no need to restart */
2990 	adapter->flags &= ~FLAG_RX_RESTART_NOW;
2991 }
2992 
2993 /**
2994  * e1000_configure_rx - Configure Receive Unit after Reset
2995  * @adapter: board private structure
2996  *
2997  * Configure the Rx unit of the MAC after a reset.
2998  **/
2999 static void e1000_configure_rx(struct e1000_adapter *adapter)
3000 {
3001 	struct e1000_hw *hw = &adapter->hw;
3002 	struct e1000_ring *rx_ring = adapter->rx_ring;
3003 	u64 rdba;
3004 	u32 rdlen, rctl, rxcsum, ctrl_ext;
3005 
3006 	if (adapter->rx_ps_pages) {
3007 		/* this is a 32 byte descriptor */
3008 		rdlen = rx_ring->count *
3009 		    sizeof(union e1000_rx_desc_packet_split);
3010 		adapter->clean_rx = e1000_clean_rx_irq_ps;
3011 		adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3012 	} else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3013 		rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3014 		adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3015 		adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3016 	} else {
3017 		rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3018 		adapter->clean_rx = e1000_clean_rx_irq;
3019 		adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3020 	}
3021 
3022 	/* disable receives while setting up the descriptors */
3023 	rctl = er32(RCTL);
3024 	if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3025 		ew32(RCTL, rctl & ~E1000_RCTL_EN);
3026 	e1e_flush();
3027 	usleep_range(10000, 20000);
3028 
3029 	if (adapter->flags2 & FLAG2_DMA_BURST) {
3030 		/*
3031 		 * set the writeback threshold (only takes effect if the RDTR
3032 		 * is set). set GRAN=1 and write back up to 0x4 worth, and
3033 		 * enable prefetching of 0x20 Rx descriptors
3034 		 * granularity = 01
3035 		 * wthresh = 04,
3036 		 * hthresh = 04,
3037 		 * pthresh = 0x20
3038 		 */
3039 		ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3040 		ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3041 
3042 		/*
3043 		 * override the delay timers for enabling bursting, only if
3044 		 * the value was not set by the user via module options
3045 		 */
3046 		if (adapter->rx_int_delay == DEFAULT_RDTR)
3047 			adapter->rx_int_delay = BURST_RDTR;
3048 		if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3049 			adapter->rx_abs_int_delay = BURST_RADV;
3050 	}
3051 
3052 	/* set the Receive Delay Timer Register */
3053 	ew32(RDTR, adapter->rx_int_delay);
3054 
3055 	/* irq moderation */
3056 	ew32(RADV, adapter->rx_abs_int_delay);
3057 	if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3058 		ew32(ITR, 1000000000 / (adapter->itr * 256));
3059 
3060 	ctrl_ext = er32(CTRL_EXT);
3061 	/* Auto-Mask interrupts upon ICR access */
3062 	ctrl_ext |= E1000_CTRL_EXT_IAME;
3063 	ew32(IAM, 0xffffffff);
3064 	ew32(CTRL_EXT, ctrl_ext);
3065 	e1e_flush();
3066 
3067 	/*
3068 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
3069 	 * the Base and Length of the Rx Descriptor Ring
3070 	 */
3071 	rdba = rx_ring->dma;
3072 	ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
3073 	ew32(RDBAH, (rdba >> 32));
3074 	ew32(RDLEN, rdlen);
3075 	ew32(RDH, 0);
3076 	ew32(RDT, 0);
3077 	rx_ring->head = adapter->hw.hw_addr + E1000_RDH;
3078 	rx_ring->tail = adapter->hw.hw_addr + E1000_RDT;
3079 
3080 	/* Enable Receive Checksum Offload for TCP and UDP */
3081 	rxcsum = er32(RXCSUM);
3082 	if (adapter->netdev->features & NETIF_F_RXCSUM) {
3083 		rxcsum |= E1000_RXCSUM_TUOFL;
3084 
3085 		/*
3086 		 * IPv4 payload checksum for UDP fragments must be
3087 		 * used in conjunction with packet-split.
3088 		 */
3089 		if (adapter->rx_ps_pages)
3090 			rxcsum |= E1000_RXCSUM_IPPCSE;
3091 	} else {
3092 		rxcsum &= ~E1000_RXCSUM_TUOFL;
3093 		/* no need to clear IPPCSE as it defaults to 0 */
3094 	}
3095 	ew32(RXCSUM, rxcsum);
3096 
3097 	if (adapter->hw.mac.type == e1000_pch2lan) {
3098 		/*
3099 		 * With jumbo frames, excessive C-state transition
3100 		 * latencies result in dropped transactions.
3101 		 */
3102 		if (adapter->netdev->mtu > ETH_DATA_LEN) {
3103 			u32 rxdctl = er32(RXDCTL(0));
3104 			ew32(RXDCTL(0), rxdctl | 0x3);
3105 			pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
3106 		} else {
3107 			pm_qos_update_request(&adapter->netdev->pm_qos_req,
3108 					      PM_QOS_DEFAULT_VALUE);
3109 		}
3110 	}
3111 
3112 	/* Enable Receives */
3113 	ew32(RCTL, rctl);
3114 }
3115 
3116 /**
3117  * e1000e_write_mc_addr_list - write multicast addresses to MTA
3118  * @netdev: network interface device structure
3119  *
3120  * Writes multicast address list to the MTA hash table.
3121  * Returns: -ENOMEM on failure
3122  *                0 on no addresses written
3123  *                X on writing X addresses to MTA
3124  */
3125 static int e1000e_write_mc_addr_list(struct net_device *netdev)
3126 {
3127 	struct e1000_adapter *adapter = netdev_priv(netdev);
3128 	struct e1000_hw *hw = &adapter->hw;
3129 	struct netdev_hw_addr *ha;
3130 	u8 *mta_list;
3131 	int i;
3132 
3133 	if (netdev_mc_empty(netdev)) {
3134 		/* nothing to program, so clear mc list */
3135 		hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3136 		return 0;
3137 	}
3138 
3139 	mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3140 	if (!mta_list)
3141 		return -ENOMEM;
3142 
3143 	/* update_mc_addr_list expects a packed array of only addresses. */
3144 	i = 0;
3145 	netdev_for_each_mc_addr(ha, netdev)
3146 		memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3147 
3148 	hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3149 	kfree(mta_list);
3150 
3151 	return netdev_mc_count(netdev);
3152 }
3153 
3154 /**
3155  * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3156  * @netdev: network interface device structure
3157  *
3158  * Writes unicast address list to the RAR table.
3159  * Returns: -ENOMEM on failure/insufficient address space
3160  *                0 on no addresses written
3161  *                X on writing X addresses to the RAR table
3162  **/
3163 static int e1000e_write_uc_addr_list(struct net_device *netdev)
3164 {
3165 	struct e1000_adapter *adapter = netdev_priv(netdev);
3166 	struct e1000_hw *hw = &adapter->hw;
3167 	unsigned int rar_entries = hw->mac.rar_entry_count;
3168 	int count = 0;
3169 
3170 	/* save a rar entry for our hardware address */
3171 	rar_entries--;
3172 
3173 	/* save a rar entry for the LAA workaround */
3174 	if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3175 		rar_entries--;
3176 
3177 	/* return ENOMEM indicating insufficient memory for addresses */
3178 	if (netdev_uc_count(netdev) > rar_entries)
3179 		return -ENOMEM;
3180 
3181 	if (!netdev_uc_empty(netdev) && rar_entries) {
3182 		struct netdev_hw_addr *ha;
3183 
3184 		/*
3185 		 * write the addresses in reverse order to avoid write
3186 		 * combining
3187 		 */
3188 		netdev_for_each_uc_addr(ha, netdev) {
3189 			if (!rar_entries)
3190 				break;
3191 			e1000e_rar_set(hw, ha->addr, rar_entries--);
3192 			count++;
3193 		}
3194 	}
3195 
3196 	/* zero out the remaining RAR entries not used above */
3197 	for (; rar_entries > 0; rar_entries--) {
3198 		ew32(RAH(rar_entries), 0);
3199 		ew32(RAL(rar_entries), 0);
3200 	}
3201 	e1e_flush();
3202 
3203 	return count;
3204 }
3205 
3206 /**
3207  * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3208  * @netdev: network interface device structure
3209  *
3210  * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3211  * address list or the network interface flags are updated.  This routine is
3212  * responsible for configuring the hardware for proper unicast, multicast,
3213  * promiscuous mode, and all-multi behavior.
3214  **/
3215 static void e1000e_set_rx_mode(struct net_device *netdev)
3216 {
3217 	struct e1000_adapter *adapter = netdev_priv(netdev);
3218 	struct e1000_hw *hw = &adapter->hw;
3219 	u32 rctl;
3220 
3221 	/* Check for Promiscuous and All Multicast modes */
3222 	rctl = er32(RCTL);
3223 
3224 	/* clear the affected bits */
3225 	rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3226 
3227 	if (netdev->flags & IFF_PROMISC) {
3228 		rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3229 		/* Do not hardware filter VLANs in promisc mode */
3230 		e1000e_vlan_filter_disable(adapter);
3231 	} else {
3232 		int count;
3233 		if (netdev->flags & IFF_ALLMULTI) {
3234 			rctl |= E1000_RCTL_MPE;
3235 		} else {
3236 			/*
3237 			 * Write addresses to the MTA, if the attempt fails
3238 			 * then we should just turn on promiscuous mode so
3239 			 * that we can at least receive multicast traffic
3240 			 */
3241 			count = e1000e_write_mc_addr_list(netdev);
3242 			if (count < 0)
3243 				rctl |= E1000_RCTL_MPE;
3244 		}
3245 		e1000e_vlan_filter_enable(adapter);
3246 		/*
3247 		 * Write addresses to available RAR registers, if there is not
3248 		 * sufficient space to store all the addresses then enable
3249 		 * unicast promiscuous mode
3250 		 */
3251 		count = e1000e_write_uc_addr_list(netdev);
3252 		if (count < 0)
3253 			rctl |= E1000_RCTL_UPE;
3254 	}
3255 
3256 	ew32(RCTL, rctl);
3257 
3258 	if (netdev->features & NETIF_F_HW_VLAN_RX)
3259 		e1000e_vlan_strip_enable(adapter);
3260 	else
3261 		e1000e_vlan_strip_disable(adapter);
3262 }
3263 
3264 static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3265 {
3266 	struct e1000_hw *hw = &adapter->hw;
3267 	u32 mrqc, rxcsum;
3268 	int i;
3269 	static const u32 rsskey[10] = {
3270 		0xda565a6d, 0xc20e5b25, 0x3d256741, 0xb08fa343, 0xcb2bcad0,
3271 		0xb4307bae, 0xa32dcb77, 0x0cf23080, 0x3bb7426a, 0xfa01acbe
3272 	};
3273 
3274 	/* Fill out hash function seed */
3275 	for (i = 0; i < 10; i++)
3276 		ew32(RSSRK(i), rsskey[i]);
3277 
3278 	/* Direct all traffic to queue 0 */
3279 	for (i = 0; i < 32; i++)
3280 		ew32(RETA(i), 0);
3281 
3282 	/*
3283 	 * Disable raw packet checksumming so that RSS hash is placed in
3284 	 * descriptor on writeback.
3285 	 */
3286 	rxcsum = er32(RXCSUM);
3287 	rxcsum |= E1000_RXCSUM_PCSD;
3288 
3289 	ew32(RXCSUM, rxcsum);
3290 
3291 	mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3292 		E1000_MRQC_RSS_FIELD_IPV4_TCP |
3293 		E1000_MRQC_RSS_FIELD_IPV6 |
3294 		E1000_MRQC_RSS_FIELD_IPV6_TCP |
3295 		E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3296 
3297 	ew32(MRQC, mrqc);
3298 }
3299 
3300 /**
3301  * e1000_configure - configure the hardware for Rx and Tx
3302  * @adapter: private board structure
3303  **/
3304 static void e1000_configure(struct e1000_adapter *adapter)
3305 {
3306 	struct e1000_ring *rx_ring = adapter->rx_ring;
3307 
3308 	e1000e_set_rx_mode(adapter->netdev);
3309 
3310 	e1000_restore_vlan(adapter);
3311 	e1000_init_manageability_pt(adapter);
3312 
3313 	e1000_configure_tx(adapter);
3314 
3315 	if (adapter->netdev->features & NETIF_F_RXHASH)
3316 		e1000e_setup_rss_hash(adapter);
3317 	e1000_setup_rctl(adapter);
3318 	e1000_configure_rx(adapter);
3319 	adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3320 }
3321 
3322 /**
3323  * e1000e_power_up_phy - restore link in case the phy was powered down
3324  * @adapter: address of board private structure
3325  *
3326  * The phy may be powered down to save power and turn off link when the
3327  * driver is unloaded and wake on lan is not enabled (among others)
3328  * *** this routine MUST be followed by a call to e1000e_reset ***
3329  **/
3330 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3331 {
3332 	if (adapter->hw.phy.ops.power_up)
3333 		adapter->hw.phy.ops.power_up(&adapter->hw);
3334 
3335 	adapter->hw.mac.ops.setup_link(&adapter->hw);
3336 }
3337 
3338 /**
3339  * e1000_power_down_phy - Power down the PHY
3340  *
3341  * Power down the PHY so no link is implied when interface is down.
3342  * The PHY cannot be powered down if management or WoL is active.
3343  */
3344 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3345 {
3346 	/* WoL is enabled */
3347 	if (adapter->wol)
3348 		return;
3349 
3350 	if (adapter->hw.phy.ops.power_down)
3351 		adapter->hw.phy.ops.power_down(&adapter->hw);
3352 }
3353 
3354 /**
3355  * e1000e_reset - bring the hardware into a known good state
3356  *
3357  * This function boots the hardware and enables some settings that
3358  * require a configuration cycle of the hardware - those cannot be
3359  * set/changed during runtime. After reset the device needs to be
3360  * properly configured for Rx, Tx etc.
3361  */
3362 void e1000e_reset(struct e1000_adapter *adapter)
3363 {
3364 	struct e1000_mac_info *mac = &adapter->hw.mac;
3365 	struct e1000_fc_info *fc = &adapter->hw.fc;
3366 	struct e1000_hw *hw = &adapter->hw;
3367 	u32 tx_space, min_tx_space, min_rx_space;
3368 	u32 pba = adapter->pba;
3369 	u16 hwm;
3370 
3371 	/* reset Packet Buffer Allocation to default */
3372 	ew32(PBA, pba);
3373 
3374 	if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3375 		/*
3376 		 * To maintain wire speed transmits, the Tx FIFO should be
3377 		 * large enough to accommodate two full transmit packets,
3378 		 * rounded up to the next 1KB and expressed in KB.  Likewise,
3379 		 * the Rx FIFO should be large enough to accommodate at least
3380 		 * one full receive packet and is similarly rounded up and
3381 		 * expressed in KB.
3382 		 */
3383 		pba = er32(PBA);
3384 		/* upper 16 bits has Tx packet buffer allocation size in KB */
3385 		tx_space = pba >> 16;
3386 		/* lower 16 bits has Rx packet buffer allocation size in KB */
3387 		pba &= 0xffff;
3388 		/*
3389 		 * the Tx fifo also stores 16 bytes of information about the Tx
3390 		 * but don't include ethernet FCS because hardware appends it
3391 		 */
3392 		min_tx_space = (adapter->max_frame_size +
3393 				sizeof(struct e1000_tx_desc) -
3394 				ETH_FCS_LEN) * 2;
3395 		min_tx_space = ALIGN(min_tx_space, 1024);
3396 		min_tx_space >>= 10;
3397 		/* software strips receive CRC, so leave room for it */
3398 		min_rx_space = adapter->max_frame_size;
3399 		min_rx_space = ALIGN(min_rx_space, 1024);
3400 		min_rx_space >>= 10;
3401 
3402 		/*
3403 		 * If current Tx allocation is less than the min Tx FIFO size,
3404 		 * and the min Tx FIFO size is less than the current Rx FIFO
3405 		 * allocation, take space away from current Rx allocation
3406 		 */
3407 		if ((tx_space < min_tx_space) &&
3408 		    ((min_tx_space - tx_space) < pba)) {
3409 			pba -= min_tx_space - tx_space;
3410 
3411 			/*
3412 			 * if short on Rx space, Rx wins and must trump Tx
3413 			 * adjustment or use Early Receive if available
3414 			 */
3415 			if (pba < min_rx_space)
3416 				pba = min_rx_space;
3417 		}
3418 
3419 		ew32(PBA, pba);
3420 	}
3421 
3422 	/*
3423 	 * flow control settings
3424 	 *
3425 	 * The high water mark must be low enough to fit one full frame
3426 	 * (or the size used for early receive) above it in the Rx FIFO.
3427 	 * Set it to the lower of:
3428 	 * - 90% of the Rx FIFO size, and
3429 	 * - the full Rx FIFO size minus one full frame
3430 	 */
3431 	if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3432 		fc->pause_time = 0xFFFF;
3433 	else
3434 		fc->pause_time = E1000_FC_PAUSE_TIME;
3435 	fc->send_xon = 1;
3436 	fc->current_mode = fc->requested_mode;
3437 
3438 	switch (hw->mac.type) {
3439 	case e1000_ich9lan:
3440 	case e1000_ich10lan:
3441 		if (adapter->netdev->mtu > ETH_DATA_LEN) {
3442 			pba = 14;
3443 			ew32(PBA, pba);
3444 			fc->high_water = 0x2800;
3445 			fc->low_water = fc->high_water - 8;
3446 			break;
3447 		}
3448 		/* fall-through */
3449 	default:
3450 		hwm = min(((pba << 10) * 9 / 10),
3451 			  ((pba << 10) - adapter->max_frame_size));
3452 
3453 		fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3454 		fc->low_water = fc->high_water - 8;
3455 		break;
3456 	case e1000_pchlan:
3457 		/*
3458 		 * Workaround PCH LOM adapter hangs with certain network
3459 		 * loads.  If hangs persist, try disabling Tx flow control.
3460 		 */
3461 		if (adapter->netdev->mtu > ETH_DATA_LEN) {
3462 			fc->high_water = 0x3500;
3463 			fc->low_water  = 0x1500;
3464 		} else {
3465 			fc->high_water = 0x5000;
3466 			fc->low_water  = 0x3000;
3467 		}
3468 		fc->refresh_time = 0x1000;
3469 		break;
3470 	case e1000_pch2lan:
3471 		fc->high_water = 0x05C20;
3472 		fc->low_water = 0x05048;
3473 		fc->pause_time = 0x0650;
3474 		fc->refresh_time = 0x0400;
3475 		if (adapter->netdev->mtu > ETH_DATA_LEN) {
3476 			pba = 14;
3477 			ew32(PBA, pba);
3478 		}
3479 		break;
3480 	}
3481 
3482 	/*
3483 	 * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3484 	 * fit in receive buffer.
3485 	 */
3486 	if (adapter->itr_setting & 0x3) {
3487 		if ((adapter->max_frame_size * 2) > (pba << 10)) {
3488 			if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3489 				dev_info(&adapter->pdev->dev,
3490 					"Interrupt Throttle Rate turned off\n");
3491 				adapter->flags2 |= FLAG2_DISABLE_AIM;
3492 				ew32(ITR, 0);
3493 			}
3494 		} else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3495 			dev_info(&adapter->pdev->dev,
3496 				 "Interrupt Throttle Rate turned on\n");
3497 			adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3498 			adapter->itr = 20000;
3499 			ew32(ITR, 1000000000 / (adapter->itr * 256));
3500 		}
3501 	}
3502 
3503 	/* Allow time for pending master requests to run */
3504 	mac->ops.reset_hw(hw);
3505 
3506 	/*
3507 	 * For parts with AMT enabled, let the firmware know
3508 	 * that the network interface is in control
3509 	 */
3510 	if (adapter->flags & FLAG_HAS_AMT)
3511 		e1000e_get_hw_control(adapter);
3512 
3513 	ew32(WUC, 0);
3514 
3515 	if (mac->ops.init_hw(hw))
3516 		e_err("Hardware Error\n");
3517 
3518 	e1000_update_mng_vlan(adapter);
3519 
3520 	/* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3521 	ew32(VET, ETH_P_8021Q);
3522 
3523 	e1000e_reset_adaptive(hw);
3524 
3525 	if (!netif_running(adapter->netdev) &&
3526 	    !test_bit(__E1000_TESTING, &adapter->state)) {
3527 		e1000_power_down_phy(adapter);
3528 		return;
3529 	}
3530 
3531 	e1000_get_phy_info(hw);
3532 
3533 	if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3534 	    !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3535 		u16 phy_data = 0;
3536 		/*
3537 		 * speed up time to link by disabling smart power down, ignore
3538 		 * the return value of this function because there is nothing
3539 		 * different we would do if it failed
3540 		 */
3541 		e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3542 		phy_data &= ~IGP02E1000_PM_SPD;
3543 		e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3544 	}
3545 }
3546 
3547 int e1000e_up(struct e1000_adapter *adapter)
3548 {
3549 	struct e1000_hw *hw = &adapter->hw;
3550 
3551 	/* hardware has been reset, we need to reload some things */
3552 	e1000_configure(adapter);
3553 
3554 	clear_bit(__E1000_DOWN, &adapter->state);
3555 
3556 	if (adapter->msix_entries)
3557 		e1000_configure_msix(adapter);
3558 	e1000_irq_enable(adapter);
3559 
3560 	netif_start_queue(adapter->netdev);
3561 
3562 	/* fire a link change interrupt to start the watchdog */
3563 	if (adapter->msix_entries)
3564 		ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3565 	else
3566 		ew32(ICS, E1000_ICS_LSC);
3567 
3568 	return 0;
3569 }
3570 
3571 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3572 {
3573 	struct e1000_hw *hw = &adapter->hw;
3574 
3575 	if (!(adapter->flags2 & FLAG2_DMA_BURST))
3576 		return;
3577 
3578 	/* flush pending descriptor writebacks to memory */
3579 	ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3580 	ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3581 
3582 	/* execute the writes immediately */
3583 	e1e_flush();
3584 }
3585 
3586 static void e1000e_update_stats(struct e1000_adapter *adapter);
3587 
3588 void e1000e_down(struct e1000_adapter *adapter)
3589 {
3590 	struct net_device *netdev = adapter->netdev;
3591 	struct e1000_hw *hw = &adapter->hw;
3592 	u32 tctl, rctl;
3593 
3594 	/*
3595 	 * signal that we're down so the interrupt handler does not
3596 	 * reschedule our watchdog timer
3597 	 */
3598 	set_bit(__E1000_DOWN, &adapter->state);
3599 
3600 	/* disable receives in the hardware */
3601 	rctl = er32(RCTL);
3602 	if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3603 		ew32(RCTL, rctl & ~E1000_RCTL_EN);
3604 	/* flush and sleep below */
3605 
3606 	netif_stop_queue(netdev);
3607 
3608 	/* disable transmits in the hardware */
3609 	tctl = er32(TCTL);
3610 	tctl &= ~E1000_TCTL_EN;
3611 	ew32(TCTL, tctl);
3612 
3613 	/* flush both disables and wait for them to finish */
3614 	e1e_flush();
3615 	usleep_range(10000, 20000);
3616 
3617 	e1000_irq_disable(adapter);
3618 
3619 	del_timer_sync(&adapter->watchdog_timer);
3620 	del_timer_sync(&adapter->phy_info_timer);
3621 
3622 	netif_carrier_off(netdev);
3623 
3624 	spin_lock(&adapter->stats64_lock);
3625 	e1000e_update_stats(adapter);
3626 	spin_unlock(&adapter->stats64_lock);
3627 
3628 	e1000e_flush_descriptors(adapter);
3629 	e1000_clean_tx_ring(adapter->tx_ring);
3630 	e1000_clean_rx_ring(adapter->rx_ring);
3631 
3632 	adapter->link_speed = 0;
3633 	adapter->link_duplex = 0;
3634 
3635 	if (!pci_channel_offline(adapter->pdev))
3636 		e1000e_reset(adapter);
3637 
3638 	/*
3639 	 * TODO: for power management, we could drop the link and
3640 	 * pci_disable_device here.
3641 	 */
3642 }
3643 
3644 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3645 {
3646 	might_sleep();
3647 	while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3648 		usleep_range(1000, 2000);
3649 	e1000e_down(adapter);
3650 	e1000e_up(adapter);
3651 	clear_bit(__E1000_RESETTING, &adapter->state);
3652 }
3653 
3654 /**
3655  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3656  * @adapter: board private structure to initialize
3657  *
3658  * e1000_sw_init initializes the Adapter private data structure.
3659  * Fields are initialized based on PCI device information and
3660  * OS network device settings (MTU size).
3661  **/
3662 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3663 {
3664 	struct net_device *netdev = adapter->netdev;
3665 
3666 	adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3667 	adapter->rx_ps_bsize0 = 128;
3668 	adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3669 	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3670 	adapter->tx_ring_count = E1000_DEFAULT_TXD;
3671 	adapter->rx_ring_count = E1000_DEFAULT_RXD;
3672 
3673 	spin_lock_init(&adapter->stats64_lock);
3674 
3675 	e1000e_set_interrupt_capability(adapter);
3676 
3677 	if (e1000_alloc_queues(adapter))
3678 		return -ENOMEM;
3679 
3680 	/* Explicitly disable IRQ since the NIC can be in any state. */
3681 	e1000_irq_disable(adapter);
3682 
3683 	set_bit(__E1000_DOWN, &adapter->state);
3684 	return 0;
3685 }
3686 
3687 /**
3688  * e1000_intr_msi_test - Interrupt Handler
3689  * @irq: interrupt number
3690  * @data: pointer to a network interface device structure
3691  **/
3692 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3693 {
3694 	struct net_device *netdev = data;
3695 	struct e1000_adapter *adapter = netdev_priv(netdev);
3696 	struct e1000_hw *hw = &adapter->hw;
3697 	u32 icr = er32(ICR);
3698 
3699 	e_dbg("icr is %08X\n", icr);
3700 	if (icr & E1000_ICR_RXSEQ) {
3701 		adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3702 		wmb();
3703 	}
3704 
3705 	return IRQ_HANDLED;
3706 }
3707 
3708 /**
3709  * e1000_test_msi_interrupt - Returns 0 for successful test
3710  * @adapter: board private struct
3711  *
3712  * code flow taken from tg3.c
3713  **/
3714 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3715 {
3716 	struct net_device *netdev = adapter->netdev;
3717 	struct e1000_hw *hw = &adapter->hw;
3718 	int err;
3719 
3720 	/* poll_enable hasn't been called yet, so don't need disable */
3721 	/* clear any pending events */
3722 	er32(ICR);
3723 
3724 	/* free the real vector and request a test handler */
3725 	e1000_free_irq(adapter);
3726 	e1000e_reset_interrupt_capability(adapter);
3727 
3728 	/* Assume that the test fails, if it succeeds then the test
3729 	 * MSI irq handler will unset this flag */
3730 	adapter->flags |= FLAG_MSI_TEST_FAILED;
3731 
3732 	err = pci_enable_msi(adapter->pdev);
3733 	if (err)
3734 		goto msi_test_failed;
3735 
3736 	err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3737 			  netdev->name, netdev);
3738 	if (err) {
3739 		pci_disable_msi(adapter->pdev);
3740 		goto msi_test_failed;
3741 	}
3742 
3743 	wmb();
3744 
3745 	e1000_irq_enable(adapter);
3746 
3747 	/* fire an unusual interrupt on the test handler */
3748 	ew32(ICS, E1000_ICS_RXSEQ);
3749 	e1e_flush();
3750 	msleep(50);
3751 
3752 	e1000_irq_disable(adapter);
3753 
3754 	rmb();
3755 
3756 	if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3757 		adapter->int_mode = E1000E_INT_MODE_LEGACY;
3758 		e_info("MSI interrupt test failed, using legacy interrupt.\n");
3759 	} else
3760 		e_dbg("MSI interrupt test succeeded!\n");
3761 
3762 	free_irq(adapter->pdev->irq, netdev);
3763 	pci_disable_msi(adapter->pdev);
3764 
3765 msi_test_failed:
3766 	e1000e_set_interrupt_capability(adapter);
3767 	return e1000_request_irq(adapter);
3768 }
3769 
3770 /**
3771  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3772  * @adapter: board private struct
3773  *
3774  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3775  **/
3776 static int e1000_test_msi(struct e1000_adapter *adapter)
3777 {
3778 	int err;
3779 	u16 pci_cmd;
3780 
3781 	if (!(adapter->flags & FLAG_MSI_ENABLED))
3782 		return 0;
3783 
3784 	/* disable SERR in case the MSI write causes a master abort */
3785 	pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3786 	if (pci_cmd & PCI_COMMAND_SERR)
3787 		pci_write_config_word(adapter->pdev, PCI_COMMAND,
3788 				      pci_cmd & ~PCI_COMMAND_SERR);
3789 
3790 	err = e1000_test_msi_interrupt(adapter);
3791 
3792 	/* re-enable SERR */
3793 	if (pci_cmd & PCI_COMMAND_SERR) {
3794 		pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3795 		pci_cmd |= PCI_COMMAND_SERR;
3796 		pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3797 	}
3798 
3799 	return err;
3800 }
3801 
3802 /**
3803  * e1000_open - Called when a network interface is made active
3804  * @netdev: network interface device structure
3805  *
3806  * Returns 0 on success, negative value on failure
3807  *
3808  * The open entry point is called when a network interface is made
3809  * active by the system (IFF_UP).  At this point all resources needed
3810  * for transmit and receive operations are allocated, the interrupt
3811  * handler is registered with the OS, the watchdog timer is started,
3812  * and the stack is notified that the interface is ready.
3813  **/
3814 static int e1000_open(struct net_device *netdev)
3815 {
3816 	struct e1000_adapter *adapter = netdev_priv(netdev);
3817 	struct e1000_hw *hw = &adapter->hw;
3818 	struct pci_dev *pdev = adapter->pdev;
3819 	int err;
3820 
3821 	/* disallow open during test */
3822 	if (test_bit(__E1000_TESTING, &adapter->state))
3823 		return -EBUSY;
3824 
3825 	pm_runtime_get_sync(&pdev->dev);
3826 
3827 	netif_carrier_off(netdev);
3828 
3829 	/* allocate transmit descriptors */
3830 	err = e1000e_setup_tx_resources(adapter->tx_ring);
3831 	if (err)
3832 		goto err_setup_tx;
3833 
3834 	/* allocate receive descriptors */
3835 	err = e1000e_setup_rx_resources(adapter->rx_ring);
3836 	if (err)
3837 		goto err_setup_rx;
3838 
3839 	/*
3840 	 * If AMT is enabled, let the firmware know that the network
3841 	 * interface is now open and reset the part to a known state.
3842 	 */
3843 	if (adapter->flags & FLAG_HAS_AMT) {
3844 		e1000e_get_hw_control(adapter);
3845 		e1000e_reset(adapter);
3846 	}
3847 
3848 	e1000e_power_up_phy(adapter);
3849 
3850 	adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3851 	if ((adapter->hw.mng_cookie.status &
3852 	     E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3853 		e1000_update_mng_vlan(adapter);
3854 
3855 	/* DMA latency requirement to workaround jumbo issue */
3856 	if (adapter->hw.mac.type == e1000_pch2lan)
3857 		pm_qos_add_request(&adapter->netdev->pm_qos_req,
3858 				   PM_QOS_CPU_DMA_LATENCY,
3859 				   PM_QOS_DEFAULT_VALUE);
3860 
3861 	/*
3862 	 * before we allocate an interrupt, we must be ready to handle it.
3863 	 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3864 	 * as soon as we call pci_request_irq, so we have to setup our
3865 	 * clean_rx handler before we do so.
3866 	 */
3867 	e1000_configure(adapter);
3868 
3869 	err = e1000_request_irq(adapter);
3870 	if (err)
3871 		goto err_req_irq;
3872 
3873 	/*
3874 	 * Work around PCIe errata with MSI interrupts causing some chipsets to
3875 	 * ignore e1000e MSI messages, which means we need to test our MSI
3876 	 * interrupt now
3877 	 */
3878 	if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3879 		err = e1000_test_msi(adapter);
3880 		if (err) {
3881 			e_err("Interrupt allocation failed\n");
3882 			goto err_req_irq;
3883 		}
3884 	}
3885 
3886 	/* From here on the code is the same as e1000e_up() */
3887 	clear_bit(__E1000_DOWN, &adapter->state);
3888 
3889 	napi_enable(&adapter->napi);
3890 
3891 	e1000_irq_enable(adapter);
3892 
3893 	adapter->tx_hang_recheck = false;
3894 	netif_start_queue(netdev);
3895 
3896 	adapter->idle_check = true;
3897 	pm_runtime_put(&pdev->dev);
3898 
3899 	/* fire a link status change interrupt to start the watchdog */
3900 	if (adapter->msix_entries)
3901 		ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3902 	else
3903 		ew32(ICS, E1000_ICS_LSC);
3904 
3905 	return 0;
3906 
3907 err_req_irq:
3908 	e1000e_release_hw_control(adapter);
3909 	e1000_power_down_phy(adapter);
3910 	e1000e_free_rx_resources(adapter->rx_ring);
3911 err_setup_rx:
3912 	e1000e_free_tx_resources(adapter->tx_ring);
3913 err_setup_tx:
3914 	e1000e_reset(adapter);
3915 	pm_runtime_put_sync(&pdev->dev);
3916 
3917 	return err;
3918 }
3919 
3920 /**
3921  * e1000_close - Disables a network interface
3922  * @netdev: network interface device structure
3923  *
3924  * Returns 0, this is not allowed to fail
3925  *
3926  * The close entry point is called when an interface is de-activated
3927  * by the OS.  The hardware is still under the drivers control, but
3928  * needs to be disabled.  A global MAC reset is issued to stop the
3929  * hardware, and all transmit and receive resources are freed.
3930  **/
3931 static int e1000_close(struct net_device *netdev)
3932 {
3933 	struct e1000_adapter *adapter = netdev_priv(netdev);
3934 	struct pci_dev *pdev = adapter->pdev;
3935 
3936 	WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3937 
3938 	pm_runtime_get_sync(&pdev->dev);
3939 
3940 	napi_disable(&adapter->napi);
3941 
3942 	if (!test_bit(__E1000_DOWN, &adapter->state)) {
3943 		e1000e_down(adapter);
3944 		e1000_free_irq(adapter);
3945 	}
3946 	e1000_power_down_phy(adapter);
3947 
3948 	e1000e_free_tx_resources(adapter->tx_ring);
3949 	e1000e_free_rx_resources(adapter->rx_ring);
3950 
3951 	/*
3952 	 * kill manageability vlan ID if supported, but not if a vlan with
3953 	 * the same ID is registered on the host OS (let 8021q kill it)
3954 	 */
3955 	if (adapter->hw.mng_cookie.status &
3956 	    E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
3957 		e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3958 
3959 	/*
3960 	 * If AMT is enabled, let the firmware know that the network
3961 	 * interface is now closed
3962 	 */
3963 	if ((adapter->flags & FLAG_HAS_AMT) &&
3964 	    !test_bit(__E1000_TESTING, &adapter->state))
3965 		e1000e_release_hw_control(adapter);
3966 
3967 	if (adapter->hw.mac.type == e1000_pch2lan)
3968 		pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3969 
3970 	pm_runtime_put_sync(&pdev->dev);
3971 
3972 	return 0;
3973 }
3974 /**
3975  * e1000_set_mac - Change the Ethernet Address of the NIC
3976  * @netdev: network interface device structure
3977  * @p: pointer to an address structure
3978  *
3979  * Returns 0 on success, negative on failure
3980  **/
3981 static int e1000_set_mac(struct net_device *netdev, void *p)
3982 {
3983 	struct e1000_adapter *adapter = netdev_priv(netdev);
3984 	struct sockaddr *addr = p;
3985 
3986 	if (!is_valid_ether_addr(addr->sa_data))
3987 		return -EADDRNOTAVAIL;
3988 
3989 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3990 	memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3991 
3992 	e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3993 
3994 	if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3995 		/* activate the work around */
3996 		e1000e_set_laa_state_82571(&adapter->hw, 1);
3997 
3998 		/*
3999 		 * Hold a copy of the LAA in RAR[14] This is done so that
4000 		 * between the time RAR[0] gets clobbered  and the time it
4001 		 * gets fixed (in e1000_watchdog), the actual LAA is in one
4002 		 * of the RARs and no incoming packets directed to this port
4003 		 * are dropped. Eventually the LAA will be in RAR[0] and
4004 		 * RAR[14]
4005 		 */
4006 		e1000e_rar_set(&adapter->hw,
4007 			      adapter->hw.mac.addr,
4008 			      adapter->hw.mac.rar_entry_count - 1);
4009 	}
4010 
4011 	return 0;
4012 }
4013 
4014 /**
4015  * e1000e_update_phy_task - work thread to update phy
4016  * @work: pointer to our work struct
4017  *
4018  * this worker thread exists because we must acquire a
4019  * semaphore to read the phy, which we could msleep while
4020  * waiting for it, and we can't msleep in a timer.
4021  **/
4022 static void e1000e_update_phy_task(struct work_struct *work)
4023 {
4024 	struct e1000_adapter *adapter = container_of(work,
4025 					struct e1000_adapter, update_phy_task);
4026 
4027 	if (test_bit(__E1000_DOWN, &adapter->state))
4028 		return;
4029 
4030 	e1000_get_phy_info(&adapter->hw);
4031 }
4032 
4033 /*
4034  * Need to wait a few seconds after link up to get diagnostic information from
4035  * the phy
4036  */
4037 static void e1000_update_phy_info(unsigned long data)
4038 {
4039 	struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4040 
4041 	if (test_bit(__E1000_DOWN, &adapter->state))
4042 		return;
4043 
4044 	schedule_work(&adapter->update_phy_task);
4045 }
4046 
4047 /**
4048  * e1000e_update_phy_stats - Update the PHY statistics counters
4049  * @adapter: board private structure
4050  *
4051  * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4052  **/
4053 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4054 {
4055 	struct e1000_hw *hw = &adapter->hw;
4056 	s32 ret_val;
4057 	u16 phy_data;
4058 
4059 	ret_val = hw->phy.ops.acquire(hw);
4060 	if (ret_val)
4061 		return;
4062 
4063 	/*
4064 	 * A page set is expensive so check if already on desired page.
4065 	 * If not, set to the page with the PHY status registers.
4066 	 */
4067 	hw->phy.addr = 1;
4068 	ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4069 					   &phy_data);
4070 	if (ret_val)
4071 		goto release;
4072 	if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4073 		ret_val = hw->phy.ops.set_page(hw,
4074 					       HV_STATS_PAGE << IGP_PAGE_SHIFT);
4075 		if (ret_val)
4076 			goto release;
4077 	}
4078 
4079 	/* Single Collision Count */
4080 	hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4081 	ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4082 	if (!ret_val)
4083 		adapter->stats.scc += phy_data;
4084 
4085 	/* Excessive Collision Count */
4086 	hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4087 	ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4088 	if (!ret_val)
4089 		adapter->stats.ecol += phy_data;
4090 
4091 	/* Multiple Collision Count */
4092 	hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4093 	ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4094 	if (!ret_val)
4095 		adapter->stats.mcc += phy_data;
4096 
4097 	/* Late Collision Count */
4098 	hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4099 	ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4100 	if (!ret_val)
4101 		adapter->stats.latecol += phy_data;
4102 
4103 	/* Collision Count - also used for adaptive IFS */
4104 	hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4105 	ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4106 	if (!ret_val)
4107 		hw->mac.collision_delta = phy_data;
4108 
4109 	/* Defer Count */
4110 	hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4111 	ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4112 	if (!ret_val)
4113 		adapter->stats.dc += phy_data;
4114 
4115 	/* Transmit with no CRS */
4116 	hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4117 	ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4118 	if (!ret_val)
4119 		adapter->stats.tncrs += phy_data;
4120 
4121 release:
4122 	hw->phy.ops.release(hw);
4123 }
4124 
4125 /**
4126  * e1000e_update_stats - Update the board statistics counters
4127  * @adapter: board private structure
4128  **/
4129 static void e1000e_update_stats(struct e1000_adapter *adapter)
4130 {
4131 	struct net_device *netdev = adapter->netdev;
4132 	struct e1000_hw *hw = &adapter->hw;
4133 	struct pci_dev *pdev = adapter->pdev;
4134 
4135 	/*
4136 	 * Prevent stats update while adapter is being reset, or if the pci
4137 	 * connection is down.
4138 	 */
4139 	if (adapter->link_speed == 0)
4140 		return;
4141 	if (pci_channel_offline(pdev))
4142 		return;
4143 
4144 	adapter->stats.crcerrs += er32(CRCERRS);
4145 	adapter->stats.gprc += er32(GPRC);
4146 	adapter->stats.gorc += er32(GORCL);
4147 	er32(GORCH); /* Clear gorc */
4148 	adapter->stats.bprc += er32(BPRC);
4149 	adapter->stats.mprc += er32(MPRC);
4150 	adapter->stats.roc += er32(ROC);
4151 
4152 	adapter->stats.mpc += er32(MPC);
4153 
4154 	/* Half-duplex statistics */
4155 	if (adapter->link_duplex == HALF_DUPLEX) {
4156 		if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4157 			e1000e_update_phy_stats(adapter);
4158 		} else {
4159 			adapter->stats.scc += er32(SCC);
4160 			adapter->stats.ecol += er32(ECOL);
4161 			adapter->stats.mcc += er32(MCC);
4162 			adapter->stats.latecol += er32(LATECOL);
4163 			adapter->stats.dc += er32(DC);
4164 
4165 			hw->mac.collision_delta = er32(COLC);
4166 
4167 			if ((hw->mac.type != e1000_82574) &&
4168 			    (hw->mac.type != e1000_82583))
4169 				adapter->stats.tncrs += er32(TNCRS);
4170 		}
4171 		adapter->stats.colc += hw->mac.collision_delta;
4172 	}
4173 
4174 	adapter->stats.xonrxc += er32(XONRXC);
4175 	adapter->stats.xontxc += er32(XONTXC);
4176 	adapter->stats.xoffrxc += er32(XOFFRXC);
4177 	adapter->stats.xofftxc += er32(XOFFTXC);
4178 	adapter->stats.gptc += er32(GPTC);
4179 	adapter->stats.gotc += er32(GOTCL);
4180 	er32(GOTCH); /* Clear gotc */
4181 	adapter->stats.rnbc += er32(RNBC);
4182 	adapter->stats.ruc += er32(RUC);
4183 
4184 	adapter->stats.mptc += er32(MPTC);
4185 	adapter->stats.bptc += er32(BPTC);
4186 
4187 	/* used for adaptive IFS */
4188 
4189 	hw->mac.tx_packet_delta = er32(TPT);
4190 	adapter->stats.tpt += hw->mac.tx_packet_delta;
4191 
4192 	adapter->stats.algnerrc += er32(ALGNERRC);
4193 	adapter->stats.rxerrc += er32(RXERRC);
4194 	adapter->stats.cexterr += er32(CEXTERR);
4195 	adapter->stats.tsctc += er32(TSCTC);
4196 	adapter->stats.tsctfc += er32(TSCTFC);
4197 
4198 	/* Fill out the OS statistics structure */
4199 	netdev->stats.multicast = adapter->stats.mprc;
4200 	netdev->stats.collisions = adapter->stats.colc;
4201 
4202 	/* Rx Errors */
4203 
4204 	/*
4205 	 * RLEC on some newer hardware can be incorrect so build
4206 	 * our own version based on RUC and ROC
4207 	 */
4208 	netdev->stats.rx_errors = adapter->stats.rxerrc +
4209 		adapter->stats.crcerrs + adapter->stats.algnerrc +
4210 		adapter->stats.ruc + adapter->stats.roc +
4211 		adapter->stats.cexterr;
4212 	netdev->stats.rx_length_errors = adapter->stats.ruc +
4213 					      adapter->stats.roc;
4214 	netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4215 	netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4216 	netdev->stats.rx_missed_errors = adapter->stats.mpc;
4217 
4218 	/* Tx Errors */
4219 	netdev->stats.tx_errors = adapter->stats.ecol +
4220 				       adapter->stats.latecol;
4221 	netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4222 	netdev->stats.tx_window_errors = adapter->stats.latecol;
4223 	netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4224 
4225 	/* Tx Dropped needs to be maintained elsewhere */
4226 
4227 	/* Management Stats */
4228 	adapter->stats.mgptc += er32(MGTPTC);
4229 	adapter->stats.mgprc += er32(MGTPRC);
4230 	adapter->stats.mgpdc += er32(MGTPDC);
4231 }
4232 
4233 /**
4234  * e1000_phy_read_status - Update the PHY register status snapshot
4235  * @adapter: board private structure
4236  **/
4237 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4238 {
4239 	struct e1000_hw *hw = &adapter->hw;
4240 	struct e1000_phy_regs *phy = &adapter->phy_regs;
4241 
4242 	if ((er32(STATUS) & E1000_STATUS_LU) &&
4243 	    (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4244 		int ret_val;
4245 
4246 		ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4247 		ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4248 		ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4249 		ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4250 		ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4251 		ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4252 		ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4253 		ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4254 		if (ret_val)
4255 			e_warn("Error reading PHY register\n");
4256 	} else {
4257 		/*
4258 		 * Do not read PHY registers if link is not up
4259 		 * Set values to typical power-on defaults
4260 		 */
4261 		phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4262 		phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4263 			     BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4264 			     BMSR_ERCAP);
4265 		phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4266 				  ADVERTISE_ALL | ADVERTISE_CSMA);
4267 		phy->lpa = 0;
4268 		phy->expansion = EXPANSION_ENABLENPAGE;
4269 		phy->ctrl1000 = ADVERTISE_1000FULL;
4270 		phy->stat1000 = 0;
4271 		phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4272 	}
4273 }
4274 
4275 static void e1000_print_link_info(struct e1000_adapter *adapter)
4276 {
4277 	struct e1000_hw *hw = &adapter->hw;
4278 	u32 ctrl = er32(CTRL);
4279 
4280 	/* Link status message must follow this format for user tools */
4281 	printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4282 		adapter->netdev->name,
4283 		adapter->link_speed,
4284 		adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
4285 		(ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
4286 		(ctrl & E1000_CTRL_RFCE) ? "Rx" :
4287 		(ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
4288 }
4289 
4290 static bool e1000e_has_link(struct e1000_adapter *adapter)
4291 {
4292 	struct e1000_hw *hw = &adapter->hw;
4293 	bool link_active = false;
4294 	s32 ret_val = 0;
4295 
4296 	/*
4297 	 * get_link_status is set on LSC (link status) interrupt or
4298 	 * Rx sequence error interrupt.  get_link_status will stay
4299 	 * false until the check_for_link establishes link
4300 	 * for copper adapters ONLY
4301 	 */
4302 	switch (hw->phy.media_type) {
4303 	case e1000_media_type_copper:
4304 		if (hw->mac.get_link_status) {
4305 			ret_val = hw->mac.ops.check_for_link(hw);
4306 			link_active = !hw->mac.get_link_status;
4307 		} else {
4308 			link_active = true;
4309 		}
4310 		break;
4311 	case e1000_media_type_fiber:
4312 		ret_val = hw->mac.ops.check_for_link(hw);
4313 		link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4314 		break;
4315 	case e1000_media_type_internal_serdes:
4316 		ret_val = hw->mac.ops.check_for_link(hw);
4317 		link_active = adapter->hw.mac.serdes_has_link;
4318 		break;
4319 	default:
4320 	case e1000_media_type_unknown:
4321 		break;
4322 	}
4323 
4324 	if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4325 	    (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4326 		/* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4327 		e_info("Gigabit has been disabled, downgrading speed\n");
4328 	}
4329 
4330 	return link_active;
4331 }
4332 
4333 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4334 {
4335 	/* make sure the receive unit is started */
4336 	if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4337 	    (adapter->flags & FLAG_RX_RESTART_NOW)) {
4338 		struct e1000_hw *hw = &adapter->hw;
4339 		u32 rctl = er32(RCTL);
4340 		ew32(RCTL, rctl | E1000_RCTL_EN);
4341 		adapter->flags &= ~FLAG_RX_RESTART_NOW;
4342 	}
4343 }
4344 
4345 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4346 {
4347 	struct e1000_hw *hw = &adapter->hw;
4348 
4349 	/*
4350 	 * With 82574 controllers, PHY needs to be checked periodically
4351 	 * for hung state and reset, if two calls return true
4352 	 */
4353 	if (e1000_check_phy_82574(hw))
4354 		adapter->phy_hang_count++;
4355 	else
4356 		adapter->phy_hang_count = 0;
4357 
4358 	if (adapter->phy_hang_count > 1) {
4359 		adapter->phy_hang_count = 0;
4360 		schedule_work(&adapter->reset_task);
4361 	}
4362 }
4363 
4364 /**
4365  * e1000_watchdog - Timer Call-back
4366  * @data: pointer to adapter cast into an unsigned long
4367  **/
4368 static void e1000_watchdog(unsigned long data)
4369 {
4370 	struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4371 
4372 	/* Do the rest outside of interrupt context */
4373 	schedule_work(&adapter->watchdog_task);
4374 
4375 	/* TODO: make this use queue_delayed_work() */
4376 }
4377 
4378 static void e1000_watchdog_task(struct work_struct *work)
4379 {
4380 	struct e1000_adapter *adapter = container_of(work,
4381 					struct e1000_adapter, watchdog_task);
4382 	struct net_device *netdev = adapter->netdev;
4383 	struct e1000_mac_info *mac = &adapter->hw.mac;
4384 	struct e1000_phy_info *phy = &adapter->hw.phy;
4385 	struct e1000_ring *tx_ring = adapter->tx_ring;
4386 	struct e1000_hw *hw = &adapter->hw;
4387 	u32 link, tctl;
4388 
4389 	if (test_bit(__E1000_DOWN, &adapter->state))
4390 		return;
4391 
4392 	link = e1000e_has_link(adapter);
4393 	if ((netif_carrier_ok(netdev)) && link) {
4394 		/* Cancel scheduled suspend requests. */
4395 		pm_runtime_resume(netdev->dev.parent);
4396 
4397 		e1000e_enable_receives(adapter);
4398 		goto link_up;
4399 	}
4400 
4401 	if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4402 	    (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4403 		e1000_update_mng_vlan(adapter);
4404 
4405 	if (link) {
4406 		if (!netif_carrier_ok(netdev)) {
4407 			bool txb2b = true;
4408 
4409 			/* Cancel scheduled suspend requests. */
4410 			pm_runtime_resume(netdev->dev.parent);
4411 
4412 			/* update snapshot of PHY registers on LSC */
4413 			e1000_phy_read_status(adapter);
4414 			mac->ops.get_link_up_info(&adapter->hw,
4415 						   &adapter->link_speed,
4416 						   &adapter->link_duplex);
4417 			e1000_print_link_info(adapter);
4418 			/*
4419 			 * On supported PHYs, check for duplex mismatch only
4420 			 * if link has autonegotiated at 10/100 half
4421 			 */
4422 			if ((hw->phy.type == e1000_phy_igp_3 ||
4423 			     hw->phy.type == e1000_phy_bm) &&
4424 			    (hw->mac.autoneg == true) &&
4425 			    (adapter->link_speed == SPEED_10 ||
4426 			     adapter->link_speed == SPEED_100) &&
4427 			    (adapter->link_duplex == HALF_DUPLEX)) {
4428 				u16 autoneg_exp;
4429 
4430 				e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4431 
4432 				if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4433 					e_info("Autonegotiated half duplex but link partner cannot autoneg.  Try forcing full duplex if link gets many collisions.\n");
4434 			}
4435 
4436 			/* adjust timeout factor according to speed/duplex */
4437 			adapter->tx_timeout_factor = 1;
4438 			switch (adapter->link_speed) {
4439 			case SPEED_10:
4440 				txb2b = false;
4441 				adapter->tx_timeout_factor = 16;
4442 				break;
4443 			case SPEED_100:
4444 				txb2b = false;
4445 				adapter->tx_timeout_factor = 10;
4446 				break;
4447 			}
4448 
4449 			/*
4450 			 * workaround: re-program speed mode bit after
4451 			 * link-up event
4452 			 */
4453 			if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4454 			    !txb2b) {
4455 				u32 tarc0;
4456 				tarc0 = er32(TARC(0));
4457 				tarc0 &= ~SPEED_MODE_BIT;
4458 				ew32(TARC(0), tarc0);
4459 			}
4460 
4461 			/*
4462 			 * disable TSO for pcie and 10/100 speeds, to avoid
4463 			 * some hardware issues
4464 			 */
4465 			if (!(adapter->flags & FLAG_TSO_FORCE)) {
4466 				switch (adapter->link_speed) {
4467 				case SPEED_10:
4468 				case SPEED_100:
4469 					e_info("10/100 speed: disabling TSO\n");
4470 					netdev->features &= ~NETIF_F_TSO;
4471 					netdev->features &= ~NETIF_F_TSO6;
4472 					break;
4473 				case SPEED_1000:
4474 					netdev->features |= NETIF_F_TSO;
4475 					netdev->features |= NETIF_F_TSO6;
4476 					break;
4477 				default:
4478 					/* oops */
4479 					break;
4480 				}
4481 			}
4482 
4483 			/*
4484 			 * enable transmits in the hardware, need to do this
4485 			 * after setting TARC(0)
4486 			 */
4487 			tctl = er32(TCTL);
4488 			tctl |= E1000_TCTL_EN;
4489 			ew32(TCTL, tctl);
4490 
4491                         /*
4492 			 * Perform any post-link-up configuration before
4493 			 * reporting link up.
4494 			 */
4495 			if (phy->ops.cfg_on_link_up)
4496 				phy->ops.cfg_on_link_up(hw);
4497 
4498 			netif_carrier_on(netdev);
4499 
4500 			if (!test_bit(__E1000_DOWN, &adapter->state))
4501 				mod_timer(&adapter->phy_info_timer,
4502 					  round_jiffies(jiffies + 2 * HZ));
4503 		}
4504 	} else {
4505 		if (netif_carrier_ok(netdev)) {
4506 			adapter->link_speed = 0;
4507 			adapter->link_duplex = 0;
4508 			/* Link status message must follow this format */
4509 			printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4510 			       adapter->netdev->name);
4511 			netif_carrier_off(netdev);
4512 			if (!test_bit(__E1000_DOWN, &adapter->state))
4513 				mod_timer(&adapter->phy_info_timer,
4514 					  round_jiffies(jiffies + 2 * HZ));
4515 
4516 			if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4517 				schedule_work(&adapter->reset_task);
4518 			else
4519 				pm_schedule_suspend(netdev->dev.parent,
4520 							LINK_TIMEOUT);
4521 		}
4522 	}
4523 
4524 link_up:
4525 	spin_lock(&adapter->stats64_lock);
4526 	e1000e_update_stats(adapter);
4527 
4528 	mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4529 	adapter->tpt_old = adapter->stats.tpt;
4530 	mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4531 	adapter->colc_old = adapter->stats.colc;
4532 
4533 	adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4534 	adapter->gorc_old = adapter->stats.gorc;
4535 	adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4536 	adapter->gotc_old = adapter->stats.gotc;
4537 	spin_unlock(&adapter->stats64_lock);
4538 
4539 	e1000e_update_adaptive(&adapter->hw);
4540 
4541 	if (!netif_carrier_ok(netdev) &&
4542 	    (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4543 		/*
4544 		 * We've lost link, so the controller stops DMA,
4545 		 * but we've got queued Tx work that's never going
4546 		 * to get done, so reset controller to flush Tx.
4547 		 * (Do the reset outside of interrupt context).
4548 		 */
4549 		schedule_work(&adapter->reset_task);
4550 		/* return immediately since reset is imminent */
4551 		return;
4552 	}
4553 
4554 	/* Simple mode for Interrupt Throttle Rate (ITR) */
4555 	if (adapter->itr_setting == 4) {
4556 		/*
4557 		 * Symmetric Tx/Rx gets a reduced ITR=2000;
4558 		 * Total asymmetrical Tx or Rx gets ITR=8000;
4559 		 * everyone else is between 2000-8000.
4560 		 */
4561 		u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4562 		u32 dif = (adapter->gotc > adapter->gorc ?
4563 			    adapter->gotc - adapter->gorc :
4564 			    adapter->gorc - adapter->gotc) / 10000;
4565 		u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4566 
4567 		ew32(ITR, 1000000000 / (itr * 256));
4568 	}
4569 
4570 	/* Cause software interrupt to ensure Rx ring is cleaned */
4571 	if (adapter->msix_entries)
4572 		ew32(ICS, adapter->rx_ring->ims_val);
4573 	else
4574 		ew32(ICS, E1000_ICS_RXDMT0);
4575 
4576 	/* flush pending descriptors to memory before detecting Tx hang */
4577 	e1000e_flush_descriptors(adapter);
4578 
4579 	/* Force detection of hung controller every watchdog period */
4580 	adapter->detect_tx_hung = true;
4581 
4582 	/*
4583 	 * With 82571 controllers, LAA may be overwritten due to controller
4584 	 * reset from the other port. Set the appropriate LAA in RAR[0]
4585 	 */
4586 	if (e1000e_get_laa_state_82571(hw))
4587 		e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4588 
4589 	if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4590 		e1000e_check_82574_phy_workaround(adapter);
4591 
4592 	/* Reset the timer */
4593 	if (!test_bit(__E1000_DOWN, &adapter->state))
4594 		mod_timer(&adapter->watchdog_timer,
4595 			  round_jiffies(jiffies + 2 * HZ));
4596 }
4597 
4598 #define E1000_TX_FLAGS_CSUM		0x00000001
4599 #define E1000_TX_FLAGS_VLAN		0x00000002
4600 #define E1000_TX_FLAGS_TSO		0x00000004
4601 #define E1000_TX_FLAGS_IPV4		0x00000008
4602 #define E1000_TX_FLAGS_VLAN_MASK	0xffff0000
4603 #define E1000_TX_FLAGS_VLAN_SHIFT	16
4604 
4605 static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb)
4606 {
4607 	struct e1000_context_desc *context_desc;
4608 	struct e1000_buffer *buffer_info;
4609 	unsigned int i;
4610 	u32 cmd_length = 0;
4611 	u16 ipcse = 0, tucse, mss;
4612 	u8 ipcss, ipcso, tucss, tucso, hdr_len;
4613 
4614 	if (!skb_is_gso(skb))
4615 		return 0;
4616 
4617 	if (skb_header_cloned(skb)) {
4618 		int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4619 
4620 		if (err)
4621 			return err;
4622 	}
4623 
4624 	hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4625 	mss = skb_shinfo(skb)->gso_size;
4626 	if (skb->protocol == htons(ETH_P_IP)) {
4627 		struct iphdr *iph = ip_hdr(skb);
4628 		iph->tot_len = 0;
4629 		iph->check = 0;
4630 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4631 		                                         0, IPPROTO_TCP, 0);
4632 		cmd_length = E1000_TXD_CMD_IP;
4633 		ipcse = skb_transport_offset(skb) - 1;
4634 	} else if (skb_is_gso_v6(skb)) {
4635 		ipv6_hdr(skb)->payload_len = 0;
4636 		tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4637 		                                       &ipv6_hdr(skb)->daddr,
4638 		                                       0, IPPROTO_TCP, 0);
4639 		ipcse = 0;
4640 	}
4641 	ipcss = skb_network_offset(skb);
4642 	ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4643 	tucss = skb_transport_offset(skb);
4644 	tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4645 	tucse = 0;
4646 
4647 	cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4648 	               E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4649 
4650 	i = tx_ring->next_to_use;
4651 	context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4652 	buffer_info = &tx_ring->buffer_info[i];
4653 
4654 	context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4655 	context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4656 	context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4657 	context_desc->upper_setup.tcp_fields.tucss = tucss;
4658 	context_desc->upper_setup.tcp_fields.tucso = tucso;
4659 	context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4660 	context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4661 	context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4662 	context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4663 
4664 	buffer_info->time_stamp = jiffies;
4665 	buffer_info->next_to_watch = i;
4666 
4667 	i++;
4668 	if (i == tx_ring->count)
4669 		i = 0;
4670 	tx_ring->next_to_use = i;
4671 
4672 	return 1;
4673 }
4674 
4675 static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb)
4676 {
4677 	struct e1000_adapter *adapter = tx_ring->adapter;
4678 	struct e1000_context_desc *context_desc;
4679 	struct e1000_buffer *buffer_info;
4680 	unsigned int i;
4681 	u8 css;
4682 	u32 cmd_len = E1000_TXD_CMD_DEXT;
4683 	__be16 protocol;
4684 
4685 	if (skb->ip_summed != CHECKSUM_PARTIAL)
4686 		return 0;
4687 
4688 	if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4689 		protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4690 	else
4691 		protocol = skb->protocol;
4692 
4693 	switch (protocol) {
4694 	case cpu_to_be16(ETH_P_IP):
4695 		if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4696 			cmd_len |= E1000_TXD_CMD_TCP;
4697 		break;
4698 	case cpu_to_be16(ETH_P_IPV6):
4699 		/* XXX not handling all IPV6 headers */
4700 		if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4701 			cmd_len |= E1000_TXD_CMD_TCP;
4702 		break;
4703 	default:
4704 		if (unlikely(net_ratelimit()))
4705 			e_warn("checksum_partial proto=%x!\n",
4706 			       be16_to_cpu(protocol));
4707 		break;
4708 	}
4709 
4710 	css = skb_checksum_start_offset(skb);
4711 
4712 	i = tx_ring->next_to_use;
4713 	buffer_info = &tx_ring->buffer_info[i];
4714 	context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4715 
4716 	context_desc->lower_setup.ip_config = 0;
4717 	context_desc->upper_setup.tcp_fields.tucss = css;
4718 	context_desc->upper_setup.tcp_fields.tucso =
4719 				css + skb->csum_offset;
4720 	context_desc->upper_setup.tcp_fields.tucse = 0;
4721 	context_desc->tcp_seg_setup.data = 0;
4722 	context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4723 
4724 	buffer_info->time_stamp = jiffies;
4725 	buffer_info->next_to_watch = i;
4726 
4727 	i++;
4728 	if (i == tx_ring->count)
4729 		i = 0;
4730 	tx_ring->next_to_use = i;
4731 
4732 	return 1;
4733 }
4734 
4735 #define E1000_MAX_PER_TXD	8192
4736 #define E1000_MAX_TXD_PWR	12
4737 
4738 static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
4739 			unsigned int first, unsigned int max_per_txd,
4740 			unsigned int nr_frags, unsigned int mss)
4741 {
4742 	struct e1000_adapter *adapter = tx_ring->adapter;
4743 	struct pci_dev *pdev = adapter->pdev;
4744 	struct e1000_buffer *buffer_info;
4745 	unsigned int len = skb_headlen(skb);
4746 	unsigned int offset = 0, size, count = 0, i;
4747 	unsigned int f, bytecount, segs;
4748 
4749 	i = tx_ring->next_to_use;
4750 
4751 	while (len) {
4752 		buffer_info = &tx_ring->buffer_info[i];
4753 		size = min(len, max_per_txd);
4754 
4755 		buffer_info->length = size;
4756 		buffer_info->time_stamp = jiffies;
4757 		buffer_info->next_to_watch = i;
4758 		buffer_info->dma = dma_map_single(&pdev->dev,
4759 						  skb->data + offset,
4760 						  size, DMA_TO_DEVICE);
4761 		buffer_info->mapped_as_page = false;
4762 		if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4763 			goto dma_error;
4764 
4765 		len -= size;
4766 		offset += size;
4767 		count++;
4768 
4769 		if (len) {
4770 			i++;
4771 			if (i == tx_ring->count)
4772 				i = 0;
4773 		}
4774 	}
4775 
4776 	for (f = 0; f < nr_frags; f++) {
4777 		const struct skb_frag_struct *frag;
4778 
4779 		frag = &skb_shinfo(skb)->frags[f];
4780 		len = skb_frag_size(frag);
4781 		offset = 0;
4782 
4783 		while (len) {
4784 			i++;
4785 			if (i == tx_ring->count)
4786 				i = 0;
4787 
4788 			buffer_info = &tx_ring->buffer_info[i];
4789 			size = min(len, max_per_txd);
4790 
4791 			buffer_info->length = size;
4792 			buffer_info->time_stamp = jiffies;
4793 			buffer_info->next_to_watch = i;
4794 			buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
4795 						offset, size, DMA_TO_DEVICE);
4796 			buffer_info->mapped_as_page = true;
4797 			if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4798 				goto dma_error;
4799 
4800 			len -= size;
4801 			offset += size;
4802 			count++;
4803 		}
4804 	}
4805 
4806 	segs = skb_shinfo(skb)->gso_segs ? : 1;
4807 	/* multiply data chunks by size of headers */
4808 	bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4809 
4810 	tx_ring->buffer_info[i].skb = skb;
4811 	tx_ring->buffer_info[i].segs = segs;
4812 	tx_ring->buffer_info[i].bytecount = bytecount;
4813 	tx_ring->buffer_info[first].next_to_watch = i;
4814 
4815 	return count;
4816 
4817 dma_error:
4818 	dev_err(&pdev->dev, "Tx DMA map failed\n");
4819 	buffer_info->dma = 0;
4820 	if (count)
4821 		count--;
4822 
4823 	while (count--) {
4824 		if (i == 0)
4825 			i += tx_ring->count;
4826 		i--;
4827 		buffer_info = &tx_ring->buffer_info[i];
4828 		e1000_put_txbuf(tx_ring, buffer_info);
4829 	}
4830 
4831 	return 0;
4832 }
4833 
4834 static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
4835 {
4836 	struct e1000_adapter *adapter = tx_ring->adapter;
4837 	struct e1000_tx_desc *tx_desc = NULL;
4838 	struct e1000_buffer *buffer_info;
4839 	u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4840 	unsigned int i;
4841 
4842 	if (tx_flags & E1000_TX_FLAGS_TSO) {
4843 		txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4844 			     E1000_TXD_CMD_TSE;
4845 		txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4846 
4847 		if (tx_flags & E1000_TX_FLAGS_IPV4)
4848 			txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4849 	}
4850 
4851 	if (tx_flags & E1000_TX_FLAGS_CSUM) {
4852 		txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4853 		txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4854 	}
4855 
4856 	if (tx_flags & E1000_TX_FLAGS_VLAN) {
4857 		txd_lower |= E1000_TXD_CMD_VLE;
4858 		txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4859 	}
4860 
4861 	i = tx_ring->next_to_use;
4862 
4863 	do {
4864 		buffer_info = &tx_ring->buffer_info[i];
4865 		tx_desc = E1000_TX_DESC(*tx_ring, i);
4866 		tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4867 		tx_desc->lower.data =
4868 			cpu_to_le32(txd_lower | buffer_info->length);
4869 		tx_desc->upper.data = cpu_to_le32(txd_upper);
4870 
4871 		i++;
4872 		if (i == tx_ring->count)
4873 			i = 0;
4874 	} while (--count > 0);
4875 
4876 	tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4877 
4878 	/*
4879 	 * Force memory writes to complete before letting h/w
4880 	 * know there are new descriptors to fetch.  (Only
4881 	 * applicable for weak-ordered memory model archs,
4882 	 * such as IA-64).
4883 	 */
4884 	wmb();
4885 
4886 	tx_ring->next_to_use = i;
4887 
4888 	if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
4889 		e1000e_update_tdt_wa(tx_ring, i);
4890 	else
4891 		writel(i, tx_ring->tail);
4892 
4893 	/*
4894 	 * we need this if more than one processor can write to our tail
4895 	 * at a time, it synchronizes IO on IA64/Altix systems
4896 	 */
4897 	mmiowb();
4898 }
4899 
4900 #define MINIMUM_DHCP_PACKET_SIZE 282
4901 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4902 				    struct sk_buff *skb)
4903 {
4904 	struct e1000_hw *hw =  &adapter->hw;
4905 	u16 length, offset;
4906 
4907 	if (vlan_tx_tag_present(skb)) {
4908 		if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4909 		    (adapter->hw.mng_cookie.status &
4910 			E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4911 			return 0;
4912 	}
4913 
4914 	if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4915 		return 0;
4916 
4917 	if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4918 		return 0;
4919 
4920 	{
4921 		const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4922 		struct udphdr *udp;
4923 
4924 		if (ip->protocol != IPPROTO_UDP)
4925 			return 0;
4926 
4927 		udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4928 		if (ntohs(udp->dest) != 67)
4929 			return 0;
4930 
4931 		offset = (u8 *)udp + 8 - skb->data;
4932 		length = skb->len - offset;
4933 		return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4934 	}
4935 
4936 	return 0;
4937 }
4938 
4939 static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
4940 {
4941 	struct e1000_adapter *adapter = tx_ring->adapter;
4942 
4943 	netif_stop_queue(adapter->netdev);
4944 	/*
4945 	 * Herbert's original patch had:
4946 	 *  smp_mb__after_netif_stop_queue();
4947 	 * but since that doesn't exist yet, just open code it.
4948 	 */
4949 	smp_mb();
4950 
4951 	/*
4952 	 * We need to check again in a case another CPU has just
4953 	 * made room available.
4954 	 */
4955 	if (e1000_desc_unused(tx_ring) < size)
4956 		return -EBUSY;
4957 
4958 	/* A reprieve! */
4959 	netif_start_queue(adapter->netdev);
4960 	++adapter->restart_queue;
4961 	return 0;
4962 }
4963 
4964 static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
4965 {
4966 	if (e1000_desc_unused(tx_ring) >= size)
4967 		return 0;
4968 	return __e1000_maybe_stop_tx(tx_ring, size);
4969 }
4970 
4971 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4972 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4973 				    struct net_device *netdev)
4974 {
4975 	struct e1000_adapter *adapter = netdev_priv(netdev);
4976 	struct e1000_ring *tx_ring = adapter->tx_ring;
4977 	unsigned int first;
4978 	unsigned int max_per_txd = E1000_MAX_PER_TXD;
4979 	unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4980 	unsigned int tx_flags = 0;
4981 	unsigned int len = skb_headlen(skb);
4982 	unsigned int nr_frags;
4983 	unsigned int mss;
4984 	int count = 0;
4985 	int tso;
4986 	unsigned int f;
4987 
4988 	if (test_bit(__E1000_DOWN, &adapter->state)) {
4989 		dev_kfree_skb_any(skb);
4990 		return NETDEV_TX_OK;
4991 	}
4992 
4993 	if (skb->len <= 0) {
4994 		dev_kfree_skb_any(skb);
4995 		return NETDEV_TX_OK;
4996 	}
4997 
4998 	mss = skb_shinfo(skb)->gso_size;
4999 	/*
5000 	 * The controller does a simple calculation to
5001 	 * make sure there is enough room in the FIFO before
5002 	 * initiating the DMA for each buffer.  The calc is:
5003 	 * 4 = ceil(buffer len/mss).  To make sure we don't
5004 	 * overrun the FIFO, adjust the max buffer len if mss
5005 	 * drops.
5006 	 */
5007 	if (mss) {
5008 		u8 hdr_len;
5009 		max_per_txd = min(mss << 2, max_per_txd);
5010 		max_txd_pwr = fls(max_per_txd) - 1;
5011 
5012 		/*
5013 		 * TSO Workaround for 82571/2/3 Controllers -- if skb->data
5014 		 * points to just header, pull a few bytes of payload from
5015 		 * frags into skb->data
5016 		 */
5017 		hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5018 		/*
5019 		 * we do this workaround for ES2LAN, but it is un-necessary,
5020 		 * avoiding it could save a lot of cycles
5021 		 */
5022 		if (skb->data_len && (hdr_len == len)) {
5023 			unsigned int pull_size;
5024 
5025 			pull_size = min((unsigned int)4, skb->data_len);
5026 			if (!__pskb_pull_tail(skb, pull_size)) {
5027 				e_err("__pskb_pull_tail failed.\n");
5028 				dev_kfree_skb_any(skb);
5029 				return NETDEV_TX_OK;
5030 			}
5031 			len = skb_headlen(skb);
5032 		}
5033 	}
5034 
5035 	/* reserve a descriptor for the offload context */
5036 	if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5037 		count++;
5038 	count++;
5039 
5040 	count += TXD_USE_COUNT(len, max_txd_pwr);
5041 
5042 	nr_frags = skb_shinfo(skb)->nr_frags;
5043 	for (f = 0; f < nr_frags; f++)
5044 		count += TXD_USE_COUNT(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5045 				       max_txd_pwr);
5046 
5047 	if (adapter->hw.mac.tx_pkt_filtering)
5048 		e1000_transfer_dhcp_info(adapter, skb);
5049 
5050 	/*
5051 	 * need: count + 2 desc gap to keep tail from touching
5052 	 * head, otherwise try next time
5053 	 */
5054 	if (e1000_maybe_stop_tx(tx_ring, count + 2))
5055 		return NETDEV_TX_BUSY;
5056 
5057 	if (vlan_tx_tag_present(skb)) {
5058 		tx_flags |= E1000_TX_FLAGS_VLAN;
5059 		tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
5060 	}
5061 
5062 	first = tx_ring->next_to_use;
5063 
5064 	tso = e1000_tso(tx_ring, skb);
5065 	if (tso < 0) {
5066 		dev_kfree_skb_any(skb);
5067 		return NETDEV_TX_OK;
5068 	}
5069 
5070 	if (tso)
5071 		tx_flags |= E1000_TX_FLAGS_TSO;
5072 	else if (e1000_tx_csum(tx_ring, skb))
5073 		tx_flags |= E1000_TX_FLAGS_CSUM;
5074 
5075 	/*
5076 	 * Old method was to assume IPv4 packet by default if TSO was enabled.
5077 	 * 82571 hardware supports TSO capabilities for IPv6 as well...
5078 	 * no longer assume, we must.
5079 	 */
5080 	if (skb->protocol == htons(ETH_P_IP))
5081 		tx_flags |= E1000_TX_FLAGS_IPV4;
5082 
5083 	/* if count is 0 then mapping error has occurred */
5084 	count = e1000_tx_map(tx_ring, skb, first, max_per_txd, nr_frags, mss);
5085 	if (count) {
5086 		netdev_sent_queue(netdev, skb->len);
5087 		e1000_tx_queue(tx_ring, tx_flags, count);
5088 		/* Make sure there is space in the ring for the next send. */
5089 		e1000_maybe_stop_tx(tx_ring, MAX_SKB_FRAGS + 2);
5090 
5091 	} else {
5092 		dev_kfree_skb_any(skb);
5093 		tx_ring->buffer_info[first].time_stamp = 0;
5094 		tx_ring->next_to_use = first;
5095 	}
5096 
5097 	return NETDEV_TX_OK;
5098 }
5099 
5100 /**
5101  * e1000_tx_timeout - Respond to a Tx Hang
5102  * @netdev: network interface device structure
5103  **/
5104 static void e1000_tx_timeout(struct net_device *netdev)
5105 {
5106 	struct e1000_adapter *adapter = netdev_priv(netdev);
5107 
5108 	/* Do the reset outside of interrupt context */
5109 	adapter->tx_timeout_count++;
5110 	schedule_work(&adapter->reset_task);
5111 }
5112 
5113 static void e1000_reset_task(struct work_struct *work)
5114 {
5115 	struct e1000_adapter *adapter;
5116 	adapter = container_of(work, struct e1000_adapter, reset_task);
5117 
5118 	/* don't run the task if already down */
5119 	if (test_bit(__E1000_DOWN, &adapter->state))
5120 		return;
5121 
5122 	if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5123 	      (adapter->flags & FLAG_RX_RESTART_NOW))) {
5124 		e1000e_dump(adapter);
5125 		e_err("Reset adapter\n");
5126 	}
5127 	e1000e_reinit_locked(adapter);
5128 }
5129 
5130 /**
5131  * e1000_get_stats64 - Get System Network Statistics
5132  * @netdev: network interface device structure
5133  * @stats: rtnl_link_stats64 pointer
5134  *
5135  * Returns the address of the device statistics structure.
5136  **/
5137 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
5138                                              struct rtnl_link_stats64 *stats)
5139 {
5140 	struct e1000_adapter *adapter = netdev_priv(netdev);
5141 
5142 	memset(stats, 0, sizeof(struct rtnl_link_stats64));
5143 	spin_lock(&adapter->stats64_lock);
5144 	e1000e_update_stats(adapter);
5145 	/* Fill out the OS statistics structure */
5146 	stats->rx_bytes = adapter->stats.gorc;
5147 	stats->rx_packets = adapter->stats.gprc;
5148 	stats->tx_bytes = adapter->stats.gotc;
5149 	stats->tx_packets = adapter->stats.gptc;
5150 	stats->multicast = adapter->stats.mprc;
5151 	stats->collisions = adapter->stats.colc;
5152 
5153 	/* Rx Errors */
5154 
5155 	/*
5156 	 * RLEC on some newer hardware can be incorrect so build
5157 	 * our own version based on RUC and ROC
5158 	 */
5159 	stats->rx_errors = adapter->stats.rxerrc +
5160 		adapter->stats.crcerrs + adapter->stats.algnerrc +
5161 		adapter->stats.ruc + adapter->stats.roc +
5162 		adapter->stats.cexterr;
5163 	stats->rx_length_errors = adapter->stats.ruc +
5164 					      adapter->stats.roc;
5165 	stats->rx_crc_errors = adapter->stats.crcerrs;
5166 	stats->rx_frame_errors = adapter->stats.algnerrc;
5167 	stats->rx_missed_errors = adapter->stats.mpc;
5168 
5169 	/* Tx Errors */
5170 	stats->tx_errors = adapter->stats.ecol +
5171 				       adapter->stats.latecol;
5172 	stats->tx_aborted_errors = adapter->stats.ecol;
5173 	stats->tx_window_errors = adapter->stats.latecol;
5174 	stats->tx_carrier_errors = adapter->stats.tncrs;
5175 
5176 	/* Tx Dropped needs to be maintained elsewhere */
5177 
5178 	spin_unlock(&adapter->stats64_lock);
5179 	return stats;
5180 }
5181 
5182 /**
5183  * e1000_change_mtu - Change the Maximum Transfer Unit
5184  * @netdev: network interface device structure
5185  * @new_mtu: new value for maximum frame size
5186  *
5187  * Returns 0 on success, negative on failure
5188  **/
5189 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5190 {
5191 	struct e1000_adapter *adapter = netdev_priv(netdev);
5192 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5193 
5194 	/* Jumbo frame support */
5195 	if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
5196 		if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5197 			e_err("Jumbo Frames not supported.\n");
5198 			return -EINVAL;
5199 		}
5200 
5201 		/*
5202 		 * IP payload checksum (enabled with jumbos/packet-split when
5203 		 * Rx checksum is enabled) and generation of RSS hash is
5204 		 * mutually exclusive in the hardware.
5205 		 */
5206 		if ((netdev->features & NETIF_F_RXCSUM) &&
5207 		    (netdev->features & NETIF_F_RXHASH)) {
5208 			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");
5209 			return -EINVAL;
5210 		}
5211 	}
5212 
5213 	/* Supported frame sizes */
5214 	if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5215 	    (max_frame > adapter->max_hw_frame_size)) {
5216 		e_err("Unsupported MTU setting\n");
5217 		return -EINVAL;
5218 	}
5219 
5220 	/* Jumbo frame workaround on 82579 requires CRC be stripped */
5221 	if ((adapter->hw.mac.type == e1000_pch2lan) &&
5222 	    !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5223 	    (new_mtu > ETH_DATA_LEN)) {
5224 		e_err("Jumbo Frames not supported on 82579 when CRC stripping is disabled.\n");
5225 		return -EINVAL;
5226 	}
5227 
5228 	/* 82573 Errata 17 */
5229 	if (((adapter->hw.mac.type == e1000_82573) ||
5230 	     (adapter->hw.mac.type == e1000_82574)) &&
5231 	    (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
5232 		adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
5233 		e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
5234 	}
5235 
5236 	while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5237 		usleep_range(1000, 2000);
5238 	/* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5239 	adapter->max_frame_size = max_frame;
5240 	e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5241 	netdev->mtu = new_mtu;
5242 	if (netif_running(netdev))
5243 		e1000e_down(adapter);
5244 
5245 	/*
5246 	 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5247 	 * means we reserve 2 more, this pushes us to allocate from the next
5248 	 * larger slab size.
5249 	 * i.e. RXBUFFER_2048 --> size-4096 slab
5250 	 * However with the new *_jumbo_rx* routines, jumbo receives will use
5251 	 * fragmented skbs
5252 	 */
5253 
5254 	if (max_frame <= 2048)
5255 		adapter->rx_buffer_len = 2048;
5256 	else
5257 		adapter->rx_buffer_len = 4096;
5258 
5259 	/* adjust allocation if LPE protects us, and we aren't using SBP */
5260 	if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5261 	     (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5262 		adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5263 					 + ETH_FCS_LEN;
5264 
5265 	if (netif_running(netdev))
5266 		e1000e_up(adapter);
5267 	else
5268 		e1000e_reset(adapter);
5269 
5270 	clear_bit(__E1000_RESETTING, &adapter->state);
5271 
5272 	return 0;
5273 }
5274 
5275 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5276 			   int cmd)
5277 {
5278 	struct e1000_adapter *adapter = netdev_priv(netdev);
5279 	struct mii_ioctl_data *data = if_mii(ifr);
5280 
5281 	if (adapter->hw.phy.media_type != e1000_media_type_copper)
5282 		return -EOPNOTSUPP;
5283 
5284 	switch (cmd) {
5285 	case SIOCGMIIPHY:
5286 		data->phy_id = adapter->hw.phy.addr;
5287 		break;
5288 	case SIOCGMIIREG:
5289 		e1000_phy_read_status(adapter);
5290 
5291 		switch (data->reg_num & 0x1F) {
5292 		case MII_BMCR:
5293 			data->val_out = adapter->phy_regs.bmcr;
5294 			break;
5295 		case MII_BMSR:
5296 			data->val_out = adapter->phy_regs.bmsr;
5297 			break;
5298 		case MII_PHYSID1:
5299 			data->val_out = (adapter->hw.phy.id >> 16);
5300 			break;
5301 		case MII_PHYSID2:
5302 			data->val_out = (adapter->hw.phy.id & 0xFFFF);
5303 			break;
5304 		case MII_ADVERTISE:
5305 			data->val_out = adapter->phy_regs.advertise;
5306 			break;
5307 		case MII_LPA:
5308 			data->val_out = adapter->phy_regs.lpa;
5309 			break;
5310 		case MII_EXPANSION:
5311 			data->val_out = adapter->phy_regs.expansion;
5312 			break;
5313 		case MII_CTRL1000:
5314 			data->val_out = adapter->phy_regs.ctrl1000;
5315 			break;
5316 		case MII_STAT1000:
5317 			data->val_out = adapter->phy_regs.stat1000;
5318 			break;
5319 		case MII_ESTATUS:
5320 			data->val_out = adapter->phy_regs.estatus;
5321 			break;
5322 		default:
5323 			return -EIO;
5324 		}
5325 		break;
5326 	case SIOCSMIIREG:
5327 	default:
5328 		return -EOPNOTSUPP;
5329 	}
5330 	return 0;
5331 }
5332 
5333 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5334 {
5335 	switch (cmd) {
5336 	case SIOCGMIIPHY:
5337 	case SIOCGMIIREG:
5338 	case SIOCSMIIREG:
5339 		return e1000_mii_ioctl(netdev, ifr, cmd);
5340 	default:
5341 		return -EOPNOTSUPP;
5342 	}
5343 }
5344 
5345 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5346 {
5347 	struct e1000_hw *hw = &adapter->hw;
5348 	u32 i, mac_reg;
5349 	u16 phy_reg, wuc_enable;
5350 	int retval = 0;
5351 
5352 	/* copy MAC RARs to PHY RARs */
5353 	e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5354 
5355 	retval = hw->phy.ops.acquire(hw);
5356 	if (retval) {
5357 		e_err("Could not acquire PHY\n");
5358 		return retval;
5359 	}
5360 
5361 	/* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
5362 	retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5363 	if (retval)
5364 		goto out;
5365 
5366 	/* copy MAC MTA to PHY MTA - only needed for pchlan */
5367 	for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5368 		mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5369 		hw->phy.ops.write_reg_page(hw, BM_MTA(i),
5370 					   (u16)(mac_reg & 0xFFFF));
5371 		hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
5372 					   (u16)((mac_reg >> 16) & 0xFFFF));
5373 	}
5374 
5375 	/* configure PHY Rx Control register */
5376 	hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
5377 	mac_reg = er32(RCTL);
5378 	if (mac_reg & E1000_RCTL_UPE)
5379 		phy_reg |= BM_RCTL_UPE;
5380 	if (mac_reg & E1000_RCTL_MPE)
5381 		phy_reg |= BM_RCTL_MPE;
5382 	phy_reg &= ~(BM_RCTL_MO_MASK);
5383 	if (mac_reg & E1000_RCTL_MO_3)
5384 		phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5385 				<< BM_RCTL_MO_SHIFT);
5386 	if (mac_reg & E1000_RCTL_BAM)
5387 		phy_reg |= BM_RCTL_BAM;
5388 	if (mac_reg & E1000_RCTL_PMCF)
5389 		phy_reg |= BM_RCTL_PMCF;
5390 	mac_reg = er32(CTRL);
5391 	if (mac_reg & E1000_CTRL_RFCE)
5392 		phy_reg |= BM_RCTL_RFCE;
5393 	hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
5394 
5395 	/* enable PHY wakeup in MAC register */
5396 	ew32(WUFC, wufc);
5397 	ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5398 
5399 	/* configure and enable PHY wakeup in PHY registers */
5400 	hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
5401 	hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5402 
5403 	/* activate PHY wakeup */
5404 	wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5405 	retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5406 	if (retval)
5407 		e_err("Could not set PHY Host Wakeup bit\n");
5408 out:
5409 	hw->phy.ops.release(hw);
5410 
5411 	return retval;
5412 }
5413 
5414 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5415 			    bool runtime)
5416 {
5417 	struct net_device *netdev = pci_get_drvdata(pdev);
5418 	struct e1000_adapter *adapter = netdev_priv(netdev);
5419 	struct e1000_hw *hw = &adapter->hw;
5420 	u32 ctrl, ctrl_ext, rctl, status;
5421 	/* Runtime suspend should only enable wakeup for link changes */
5422 	u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5423 	int retval = 0;
5424 
5425 	netif_device_detach(netdev);
5426 
5427 	if (netif_running(netdev)) {
5428 		WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5429 		e1000e_down(adapter);
5430 		e1000_free_irq(adapter);
5431 	}
5432 	e1000e_reset_interrupt_capability(adapter);
5433 
5434 	retval = pci_save_state(pdev);
5435 	if (retval)
5436 		return retval;
5437 
5438 	status = er32(STATUS);
5439 	if (status & E1000_STATUS_LU)
5440 		wufc &= ~E1000_WUFC_LNKC;
5441 
5442 	if (wufc) {
5443 		e1000_setup_rctl(adapter);
5444 		e1000e_set_rx_mode(netdev);
5445 
5446 		/* turn on all-multi mode if wake on multicast is enabled */
5447 		if (wufc & E1000_WUFC_MC) {
5448 			rctl = er32(RCTL);
5449 			rctl |= E1000_RCTL_MPE;
5450 			ew32(RCTL, rctl);
5451 		}
5452 
5453 		ctrl = er32(CTRL);
5454 		/* advertise wake from D3Cold */
5455 		#define E1000_CTRL_ADVD3WUC 0x00100000
5456 		/* phy power management enable */
5457 		#define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5458 		ctrl |= E1000_CTRL_ADVD3WUC;
5459 		if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5460 			ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5461 		ew32(CTRL, ctrl);
5462 
5463 		if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5464 		    adapter->hw.phy.media_type ==
5465 		    e1000_media_type_internal_serdes) {
5466 			/* keep the laser running in D3 */
5467 			ctrl_ext = er32(CTRL_EXT);
5468 			ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5469 			ew32(CTRL_EXT, ctrl_ext);
5470 		}
5471 
5472 		if (adapter->flags & FLAG_IS_ICH)
5473 			e1000_suspend_workarounds_ich8lan(&adapter->hw);
5474 
5475 		/* Allow time for pending master requests to run */
5476 		e1000e_disable_pcie_master(&adapter->hw);
5477 
5478 		if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5479 			/* enable wakeup by the PHY */
5480 			retval = e1000_init_phy_wakeup(adapter, wufc);
5481 			if (retval)
5482 				return retval;
5483 		} else {
5484 			/* enable wakeup by the MAC */
5485 			ew32(WUFC, wufc);
5486 			ew32(WUC, E1000_WUC_PME_EN);
5487 		}
5488 	} else {
5489 		ew32(WUC, 0);
5490 		ew32(WUFC, 0);
5491 	}
5492 
5493 	*enable_wake = !!wufc;
5494 
5495 	/* make sure adapter isn't asleep if manageability is enabled */
5496 	if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5497 	    (hw->mac.ops.check_mng_mode(hw)))
5498 		*enable_wake = true;
5499 
5500 	if (adapter->hw.phy.type == e1000_phy_igp_3)
5501 		e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5502 
5503 	/*
5504 	 * Release control of h/w to f/w.  If f/w is AMT enabled, this
5505 	 * would have already happened in close and is redundant.
5506 	 */
5507 	e1000e_release_hw_control(adapter);
5508 
5509 	pci_disable_device(pdev);
5510 
5511 	return 0;
5512 }
5513 
5514 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5515 {
5516 	if (sleep && wake) {
5517 		pci_prepare_to_sleep(pdev);
5518 		return;
5519 	}
5520 
5521 	pci_wake_from_d3(pdev, wake);
5522 	pci_set_power_state(pdev, PCI_D3hot);
5523 }
5524 
5525 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5526                                     bool wake)
5527 {
5528 	struct net_device *netdev = pci_get_drvdata(pdev);
5529 	struct e1000_adapter *adapter = netdev_priv(netdev);
5530 
5531 	/*
5532 	 * The pci-e switch on some quad port adapters will report a
5533 	 * correctable error when the MAC transitions from D0 to D3.  To
5534 	 * prevent this we need to mask off the correctable errors on the
5535 	 * downstream port of the pci-e switch.
5536 	 */
5537 	if (adapter->flags & FLAG_IS_QUAD_PORT) {
5538 		struct pci_dev *us_dev = pdev->bus->self;
5539 		int pos = pci_pcie_cap(us_dev);
5540 		u16 devctl;
5541 
5542 		pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5543 		pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5544 		                      (devctl & ~PCI_EXP_DEVCTL_CERE));
5545 
5546 		e1000_power_off(pdev, sleep, wake);
5547 
5548 		pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5549 	} else {
5550 		e1000_power_off(pdev, sleep, wake);
5551 	}
5552 }
5553 
5554 #ifdef CONFIG_PCIEASPM
5555 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5556 {
5557 	pci_disable_link_state_locked(pdev, state);
5558 }
5559 #else
5560 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5561 {
5562 	int pos;
5563 	u16 reg16;
5564 
5565 	/*
5566 	 * Both device and parent should have the same ASPM setting.
5567 	 * Disable ASPM in downstream component first and then upstream.
5568 	 */
5569 	pos = pci_pcie_cap(pdev);
5570 	pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5571 	reg16 &= ~state;
5572 	pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5573 
5574 	if (!pdev->bus->self)
5575 		return;
5576 
5577 	pos = pci_pcie_cap(pdev->bus->self);
5578 	pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5579 	reg16 &= ~state;
5580 	pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5581 }
5582 #endif
5583 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5584 {
5585 	dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5586 		 (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5587 		 (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5588 
5589 	__e1000e_disable_aspm(pdev, state);
5590 }
5591 
5592 #ifdef CONFIG_PM
5593 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5594 {
5595 	return !!adapter->tx_ring->buffer_info;
5596 }
5597 
5598 static int __e1000_resume(struct pci_dev *pdev)
5599 {
5600 	struct net_device *netdev = pci_get_drvdata(pdev);
5601 	struct e1000_adapter *adapter = netdev_priv(netdev);
5602 	struct e1000_hw *hw = &adapter->hw;
5603 	u16 aspm_disable_flag = 0;
5604 	u32 err;
5605 
5606 	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5607 		aspm_disable_flag = PCIE_LINK_STATE_L0S;
5608 	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5609 		aspm_disable_flag |= PCIE_LINK_STATE_L1;
5610 	if (aspm_disable_flag)
5611 		e1000e_disable_aspm(pdev, aspm_disable_flag);
5612 
5613 	pci_set_power_state(pdev, PCI_D0);
5614 	pci_restore_state(pdev);
5615 	pci_save_state(pdev);
5616 
5617 	e1000e_set_interrupt_capability(adapter);
5618 	if (netif_running(netdev)) {
5619 		err = e1000_request_irq(adapter);
5620 		if (err)
5621 			return err;
5622 	}
5623 
5624 	if (hw->mac.type == e1000_pch2lan)
5625 		e1000_resume_workarounds_pchlan(&adapter->hw);
5626 
5627 	e1000e_power_up_phy(adapter);
5628 
5629 	/* report the system wakeup cause from S3/S4 */
5630 	if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5631 		u16 phy_data;
5632 
5633 		e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5634 		if (phy_data) {
5635 			e_info("PHY Wakeup cause - %s\n",
5636 				phy_data & E1000_WUS_EX ? "Unicast Packet" :
5637 				phy_data & E1000_WUS_MC ? "Multicast Packet" :
5638 				phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5639 				phy_data & E1000_WUS_MAG ? "Magic Packet" :
5640 				phy_data & E1000_WUS_LNKC ?
5641 				"Link Status Change" : "other");
5642 		}
5643 		e1e_wphy(&adapter->hw, BM_WUS, ~0);
5644 	} else {
5645 		u32 wus = er32(WUS);
5646 		if (wus) {
5647 			e_info("MAC Wakeup cause - %s\n",
5648 				wus & E1000_WUS_EX ? "Unicast Packet" :
5649 				wus & E1000_WUS_MC ? "Multicast Packet" :
5650 				wus & E1000_WUS_BC ? "Broadcast Packet" :
5651 				wus & E1000_WUS_MAG ? "Magic Packet" :
5652 				wus & E1000_WUS_LNKC ? "Link Status Change" :
5653 				"other");
5654 		}
5655 		ew32(WUS, ~0);
5656 	}
5657 
5658 	e1000e_reset(adapter);
5659 
5660 	e1000_init_manageability_pt(adapter);
5661 
5662 	if (netif_running(netdev))
5663 		e1000e_up(adapter);
5664 
5665 	netif_device_attach(netdev);
5666 
5667 	/*
5668 	 * If the controller has AMT, do not set DRV_LOAD until the interface
5669 	 * is up.  For all other cases, let the f/w know that the h/w is now
5670 	 * under the control of the driver.
5671 	 */
5672 	if (!(adapter->flags & FLAG_HAS_AMT))
5673 		e1000e_get_hw_control(adapter);
5674 
5675 	return 0;
5676 }
5677 
5678 #ifdef CONFIG_PM_SLEEP
5679 static int e1000_suspend(struct device *dev)
5680 {
5681 	struct pci_dev *pdev = to_pci_dev(dev);
5682 	int retval;
5683 	bool wake;
5684 
5685 	retval = __e1000_shutdown(pdev, &wake, false);
5686 	if (!retval)
5687 		e1000_complete_shutdown(pdev, true, wake);
5688 
5689 	return retval;
5690 }
5691 
5692 static int e1000_resume(struct device *dev)
5693 {
5694 	struct pci_dev *pdev = to_pci_dev(dev);
5695 	struct net_device *netdev = pci_get_drvdata(pdev);
5696 	struct e1000_adapter *adapter = netdev_priv(netdev);
5697 
5698 	if (e1000e_pm_ready(adapter))
5699 		adapter->idle_check = true;
5700 
5701 	return __e1000_resume(pdev);
5702 }
5703 #endif /* CONFIG_PM_SLEEP */
5704 
5705 #ifdef CONFIG_PM_RUNTIME
5706 static int e1000_runtime_suspend(struct device *dev)
5707 {
5708 	struct pci_dev *pdev = to_pci_dev(dev);
5709 	struct net_device *netdev = pci_get_drvdata(pdev);
5710 	struct e1000_adapter *adapter = netdev_priv(netdev);
5711 
5712 	if (e1000e_pm_ready(adapter)) {
5713 		bool wake;
5714 
5715 		__e1000_shutdown(pdev, &wake, true);
5716 	}
5717 
5718 	return 0;
5719 }
5720 
5721 static int e1000_idle(struct device *dev)
5722 {
5723 	struct pci_dev *pdev = to_pci_dev(dev);
5724 	struct net_device *netdev = pci_get_drvdata(pdev);
5725 	struct e1000_adapter *adapter = netdev_priv(netdev);
5726 
5727 	if (!e1000e_pm_ready(adapter))
5728 		return 0;
5729 
5730 	if (adapter->idle_check) {
5731 		adapter->idle_check = false;
5732 		if (!e1000e_has_link(adapter))
5733 			pm_schedule_suspend(dev, MSEC_PER_SEC);
5734 	}
5735 
5736 	return -EBUSY;
5737 }
5738 
5739 static int e1000_runtime_resume(struct device *dev)
5740 {
5741 	struct pci_dev *pdev = to_pci_dev(dev);
5742 	struct net_device *netdev = pci_get_drvdata(pdev);
5743 	struct e1000_adapter *adapter = netdev_priv(netdev);
5744 
5745 	if (!e1000e_pm_ready(adapter))
5746 		return 0;
5747 
5748 	adapter->idle_check = !dev->power.runtime_auto;
5749 	return __e1000_resume(pdev);
5750 }
5751 #endif /* CONFIG_PM_RUNTIME */
5752 #endif /* CONFIG_PM */
5753 
5754 static void e1000_shutdown(struct pci_dev *pdev)
5755 {
5756 	bool wake = false;
5757 
5758 	__e1000_shutdown(pdev, &wake, false);
5759 
5760 	if (system_state == SYSTEM_POWER_OFF)
5761 		e1000_complete_shutdown(pdev, false, wake);
5762 }
5763 
5764 #ifdef CONFIG_NET_POLL_CONTROLLER
5765 
5766 static irqreturn_t e1000_intr_msix(int irq, void *data)
5767 {
5768 	struct net_device *netdev = data;
5769 	struct e1000_adapter *adapter = netdev_priv(netdev);
5770 
5771 	if (adapter->msix_entries) {
5772 		int vector, msix_irq;
5773 
5774 		vector = 0;
5775 		msix_irq = adapter->msix_entries[vector].vector;
5776 		disable_irq(msix_irq);
5777 		e1000_intr_msix_rx(msix_irq, netdev);
5778 		enable_irq(msix_irq);
5779 
5780 		vector++;
5781 		msix_irq = adapter->msix_entries[vector].vector;
5782 		disable_irq(msix_irq);
5783 		e1000_intr_msix_tx(msix_irq, netdev);
5784 		enable_irq(msix_irq);
5785 
5786 		vector++;
5787 		msix_irq = adapter->msix_entries[vector].vector;
5788 		disable_irq(msix_irq);
5789 		e1000_msix_other(msix_irq, netdev);
5790 		enable_irq(msix_irq);
5791 	}
5792 
5793 	return IRQ_HANDLED;
5794 }
5795 
5796 /*
5797  * Polling 'interrupt' - used by things like netconsole to send skbs
5798  * without having to re-enable interrupts. It's not called while
5799  * the interrupt routine is executing.
5800  */
5801 static void e1000_netpoll(struct net_device *netdev)
5802 {
5803 	struct e1000_adapter *adapter = netdev_priv(netdev);
5804 
5805 	switch (adapter->int_mode) {
5806 	case E1000E_INT_MODE_MSIX:
5807 		e1000_intr_msix(adapter->pdev->irq, netdev);
5808 		break;
5809 	case E1000E_INT_MODE_MSI:
5810 		disable_irq(adapter->pdev->irq);
5811 		e1000_intr_msi(adapter->pdev->irq, netdev);
5812 		enable_irq(adapter->pdev->irq);
5813 		break;
5814 	default: /* E1000E_INT_MODE_LEGACY */
5815 		disable_irq(adapter->pdev->irq);
5816 		e1000_intr(adapter->pdev->irq, netdev);
5817 		enable_irq(adapter->pdev->irq);
5818 		break;
5819 	}
5820 }
5821 #endif
5822 
5823 /**
5824  * e1000_io_error_detected - called when PCI error is detected
5825  * @pdev: Pointer to PCI device
5826  * @state: The current pci connection state
5827  *
5828  * This function is called after a PCI bus error affecting
5829  * this device has been detected.
5830  */
5831 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5832 						pci_channel_state_t state)
5833 {
5834 	struct net_device *netdev = pci_get_drvdata(pdev);
5835 	struct e1000_adapter *adapter = netdev_priv(netdev);
5836 
5837 	netif_device_detach(netdev);
5838 
5839 	if (state == pci_channel_io_perm_failure)
5840 		return PCI_ERS_RESULT_DISCONNECT;
5841 
5842 	if (netif_running(netdev))
5843 		e1000e_down(adapter);
5844 	pci_disable_device(pdev);
5845 
5846 	/* Request a slot slot reset. */
5847 	return PCI_ERS_RESULT_NEED_RESET;
5848 }
5849 
5850 /**
5851  * e1000_io_slot_reset - called after the pci bus has been reset.
5852  * @pdev: Pointer to PCI device
5853  *
5854  * Restart the card from scratch, as if from a cold-boot. Implementation
5855  * resembles the first-half of the e1000_resume routine.
5856  */
5857 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5858 {
5859 	struct net_device *netdev = pci_get_drvdata(pdev);
5860 	struct e1000_adapter *adapter = netdev_priv(netdev);
5861 	struct e1000_hw *hw = &adapter->hw;
5862 	u16 aspm_disable_flag = 0;
5863 	int err;
5864 	pci_ers_result_t result;
5865 
5866 	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5867 		aspm_disable_flag = PCIE_LINK_STATE_L0S;
5868 	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5869 		aspm_disable_flag |= PCIE_LINK_STATE_L1;
5870 	if (aspm_disable_flag)
5871 		e1000e_disable_aspm(pdev, aspm_disable_flag);
5872 
5873 	err = pci_enable_device_mem(pdev);
5874 	if (err) {
5875 		dev_err(&pdev->dev,
5876 			"Cannot re-enable PCI device after reset.\n");
5877 		result = PCI_ERS_RESULT_DISCONNECT;
5878 	} else {
5879 		pci_set_master(pdev);
5880 		pdev->state_saved = true;
5881 		pci_restore_state(pdev);
5882 
5883 		pci_enable_wake(pdev, PCI_D3hot, 0);
5884 		pci_enable_wake(pdev, PCI_D3cold, 0);
5885 
5886 		e1000e_reset(adapter);
5887 		ew32(WUS, ~0);
5888 		result = PCI_ERS_RESULT_RECOVERED;
5889 	}
5890 
5891 	pci_cleanup_aer_uncorrect_error_status(pdev);
5892 
5893 	return result;
5894 }
5895 
5896 /**
5897  * e1000_io_resume - called when traffic can start flowing again.
5898  * @pdev: Pointer to PCI device
5899  *
5900  * This callback is called when the error recovery driver tells us that
5901  * its OK to resume normal operation. Implementation resembles the
5902  * second-half of the e1000_resume routine.
5903  */
5904 static void e1000_io_resume(struct pci_dev *pdev)
5905 {
5906 	struct net_device *netdev = pci_get_drvdata(pdev);
5907 	struct e1000_adapter *adapter = netdev_priv(netdev);
5908 
5909 	e1000_init_manageability_pt(adapter);
5910 
5911 	if (netif_running(netdev)) {
5912 		if (e1000e_up(adapter)) {
5913 			dev_err(&pdev->dev,
5914 				"can't bring device back up after reset\n");
5915 			return;
5916 		}
5917 	}
5918 
5919 	netif_device_attach(netdev);
5920 
5921 	/*
5922 	 * If the controller has AMT, do not set DRV_LOAD until the interface
5923 	 * is up.  For all other cases, let the f/w know that the h/w is now
5924 	 * under the control of the driver.
5925 	 */
5926 	if (!(adapter->flags & FLAG_HAS_AMT))
5927 		e1000e_get_hw_control(adapter);
5928 
5929 }
5930 
5931 static void e1000_print_device_info(struct e1000_adapter *adapter)
5932 {
5933 	struct e1000_hw *hw = &adapter->hw;
5934 	struct net_device *netdev = adapter->netdev;
5935 	u32 ret_val;
5936 	u8 pba_str[E1000_PBANUM_LENGTH];
5937 
5938 	/* print bus type/speed/width info */
5939 	e_info("(PCI Express:2.5GT/s:%s) %pM\n",
5940 	       /* bus width */
5941 	       ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5942 	        "Width x1"),
5943 	       /* MAC address */
5944 	       netdev->dev_addr);
5945 	e_info("Intel(R) PRO/%s Network Connection\n",
5946 	       (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5947 	ret_val = e1000_read_pba_string_generic(hw, pba_str,
5948 						E1000_PBANUM_LENGTH);
5949 	if (ret_val)
5950 		strlcpy((char *)pba_str, "Unknown", sizeof(pba_str));
5951 	e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5952 	       hw->mac.type, hw->phy.type, pba_str);
5953 }
5954 
5955 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5956 {
5957 	struct e1000_hw *hw = &adapter->hw;
5958 	int ret_val;
5959 	u16 buf = 0;
5960 
5961 	if (hw->mac.type != e1000_82573)
5962 		return;
5963 
5964 	ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5965 	if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5966 		/* Deep Smart Power Down (DSPD) */
5967 		dev_warn(&adapter->pdev->dev,
5968 			 "Warning: detected DSPD enabled in EEPROM\n");
5969 	}
5970 }
5971 
5972 static int e1000_set_features(struct net_device *netdev,
5973 			      netdev_features_t features)
5974 {
5975 	struct e1000_adapter *adapter = netdev_priv(netdev);
5976 	netdev_features_t changed = features ^ netdev->features;
5977 
5978 	if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
5979 		adapter->flags |= FLAG_TSO_FORCE;
5980 
5981 	if (!(changed & (NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX |
5982 			 NETIF_F_RXCSUM | NETIF_F_RXHASH)))
5983 		return 0;
5984 
5985 	/*
5986 	 * IP payload checksum (enabled with jumbos/packet-split when Rx
5987 	 * checksum is enabled) and generation of RSS hash is mutually
5988 	 * exclusive in the hardware.
5989 	 */
5990 	if (adapter->rx_ps_pages &&
5991 	    (features & NETIF_F_RXCSUM) && (features & NETIF_F_RXHASH)) {
5992 		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");
5993 		return -EINVAL;
5994 	}
5995 
5996 	netdev->features = features;
5997 
5998 	if (netif_running(netdev))
5999 		e1000e_reinit_locked(adapter);
6000 	else
6001 		e1000e_reset(adapter);
6002 
6003 	return 0;
6004 }
6005 
6006 static const struct net_device_ops e1000e_netdev_ops = {
6007 	.ndo_open		= e1000_open,
6008 	.ndo_stop		= e1000_close,
6009 	.ndo_start_xmit		= e1000_xmit_frame,
6010 	.ndo_get_stats64	= e1000e_get_stats64,
6011 	.ndo_set_rx_mode	= e1000e_set_rx_mode,
6012 	.ndo_set_mac_address	= e1000_set_mac,
6013 	.ndo_change_mtu		= e1000_change_mtu,
6014 	.ndo_do_ioctl		= e1000_ioctl,
6015 	.ndo_tx_timeout		= e1000_tx_timeout,
6016 	.ndo_validate_addr	= eth_validate_addr,
6017 
6018 	.ndo_vlan_rx_add_vid	= e1000_vlan_rx_add_vid,
6019 	.ndo_vlan_rx_kill_vid	= e1000_vlan_rx_kill_vid,
6020 #ifdef CONFIG_NET_POLL_CONTROLLER
6021 	.ndo_poll_controller	= e1000_netpoll,
6022 #endif
6023 	.ndo_set_features = e1000_set_features,
6024 };
6025 
6026 /**
6027  * e1000_probe - Device Initialization Routine
6028  * @pdev: PCI device information struct
6029  * @ent: entry in e1000_pci_tbl
6030  *
6031  * Returns 0 on success, negative on failure
6032  *
6033  * e1000_probe initializes an adapter identified by a pci_dev structure.
6034  * The OS initialization, configuring of the adapter private structure,
6035  * and a hardware reset occur.
6036  **/
6037 static int __devinit e1000_probe(struct pci_dev *pdev,
6038 				 const struct pci_device_id *ent)
6039 {
6040 	struct net_device *netdev;
6041 	struct e1000_adapter *adapter;
6042 	struct e1000_hw *hw;
6043 	const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
6044 	resource_size_t mmio_start, mmio_len;
6045 	resource_size_t flash_start, flash_len;
6046 
6047 	static int cards_found;
6048 	u16 aspm_disable_flag = 0;
6049 	int i, err, pci_using_dac;
6050 	u16 eeprom_data = 0;
6051 	u16 eeprom_apme_mask = E1000_EEPROM_APME;
6052 
6053 	if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
6054 		aspm_disable_flag = PCIE_LINK_STATE_L0S;
6055 	if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
6056 		aspm_disable_flag |= PCIE_LINK_STATE_L1;
6057 	if (aspm_disable_flag)
6058 		e1000e_disable_aspm(pdev, aspm_disable_flag);
6059 
6060 	err = pci_enable_device_mem(pdev);
6061 	if (err)
6062 		return err;
6063 
6064 	pci_using_dac = 0;
6065 	err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
6066 	if (!err) {
6067 		err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
6068 		if (!err)
6069 			pci_using_dac = 1;
6070 	} else {
6071 		err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6072 		if (err) {
6073 			err = dma_set_coherent_mask(&pdev->dev,
6074 						    DMA_BIT_MASK(32));
6075 			if (err) {
6076 				dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
6077 				goto err_dma;
6078 			}
6079 		}
6080 	}
6081 
6082 	err = pci_request_selected_regions_exclusive(pdev,
6083 	                                  pci_select_bars(pdev, IORESOURCE_MEM),
6084 	                                  e1000e_driver_name);
6085 	if (err)
6086 		goto err_pci_reg;
6087 
6088 	/* AER (Advanced Error Reporting) hooks */
6089 	pci_enable_pcie_error_reporting(pdev);
6090 
6091 	pci_set_master(pdev);
6092 	/* PCI config space info */
6093 	err = pci_save_state(pdev);
6094 	if (err)
6095 		goto err_alloc_etherdev;
6096 
6097 	err = -ENOMEM;
6098 	netdev = alloc_etherdev(sizeof(struct e1000_adapter));
6099 	if (!netdev)
6100 		goto err_alloc_etherdev;
6101 
6102 	SET_NETDEV_DEV(netdev, &pdev->dev);
6103 
6104 	netdev->irq = pdev->irq;
6105 
6106 	pci_set_drvdata(pdev, netdev);
6107 	adapter = netdev_priv(netdev);
6108 	hw = &adapter->hw;
6109 	adapter->netdev = netdev;
6110 	adapter->pdev = pdev;
6111 	adapter->ei = ei;
6112 	adapter->pba = ei->pba;
6113 	adapter->flags = ei->flags;
6114 	adapter->flags2 = ei->flags2;
6115 	adapter->hw.adapter = adapter;
6116 	adapter->hw.mac.type = ei->mac;
6117 	adapter->max_hw_frame_size = ei->max_hw_frame_size;
6118 	adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
6119 
6120 	mmio_start = pci_resource_start(pdev, 0);
6121 	mmio_len = pci_resource_len(pdev, 0);
6122 
6123 	err = -EIO;
6124 	adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
6125 	if (!adapter->hw.hw_addr)
6126 		goto err_ioremap;
6127 
6128 	if ((adapter->flags & FLAG_HAS_FLASH) &&
6129 	    (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
6130 		flash_start = pci_resource_start(pdev, 1);
6131 		flash_len = pci_resource_len(pdev, 1);
6132 		adapter->hw.flash_address = ioremap(flash_start, flash_len);
6133 		if (!adapter->hw.flash_address)
6134 			goto err_flashmap;
6135 	}
6136 
6137 	/* construct the net_device struct */
6138 	netdev->netdev_ops		= &e1000e_netdev_ops;
6139 	e1000e_set_ethtool_ops(netdev);
6140 	netdev->watchdog_timeo		= 5 * HZ;
6141 	netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
6142 	strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
6143 
6144 	netdev->mem_start = mmio_start;
6145 	netdev->mem_end = mmio_start + mmio_len;
6146 
6147 	adapter->bd_number = cards_found++;
6148 
6149 	e1000e_check_options(adapter);
6150 
6151 	/* setup adapter struct */
6152 	err = e1000_sw_init(adapter);
6153 	if (err)
6154 		goto err_sw_init;
6155 
6156 	memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
6157 	memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
6158 	memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
6159 
6160 	err = ei->get_variants(adapter);
6161 	if (err)
6162 		goto err_hw_init;
6163 
6164 	if ((adapter->flags & FLAG_IS_ICH) &&
6165 	    (adapter->flags & FLAG_READ_ONLY_NVM))
6166 		e1000e_write_protect_nvm_ich8lan(&adapter->hw);
6167 
6168 	hw->mac.ops.get_bus_info(&adapter->hw);
6169 
6170 	adapter->hw.phy.autoneg_wait_to_complete = 0;
6171 
6172 	/* Copper options */
6173 	if (adapter->hw.phy.media_type == e1000_media_type_copper) {
6174 		adapter->hw.phy.mdix = AUTO_ALL_MODES;
6175 		adapter->hw.phy.disable_polarity_correction = 0;
6176 		adapter->hw.phy.ms_type = e1000_ms_hw_default;
6177 	}
6178 
6179 	if (e1000_check_reset_block(&adapter->hw))
6180 		e_info("PHY reset is blocked due to SOL/IDER session.\n");
6181 
6182 	/* Set initial default active device features */
6183 	netdev->features = (NETIF_F_SG |
6184 			    NETIF_F_HW_VLAN_RX |
6185 			    NETIF_F_HW_VLAN_TX |
6186 			    NETIF_F_TSO |
6187 			    NETIF_F_TSO6 |
6188 			    NETIF_F_RXHASH |
6189 			    NETIF_F_RXCSUM |
6190 			    NETIF_F_HW_CSUM);
6191 
6192 	/* Set user-changeable features (subset of all device features) */
6193 	netdev->hw_features = netdev->features;
6194 
6195 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
6196 		netdev->features |= NETIF_F_HW_VLAN_FILTER;
6197 
6198 	netdev->vlan_features |= (NETIF_F_SG |
6199 				  NETIF_F_TSO |
6200 				  NETIF_F_TSO6 |
6201 				  NETIF_F_HW_CSUM);
6202 
6203 	netdev->priv_flags |= IFF_UNICAST_FLT;
6204 
6205 	if (pci_using_dac) {
6206 		netdev->features |= NETIF_F_HIGHDMA;
6207 		netdev->vlan_features |= NETIF_F_HIGHDMA;
6208 	}
6209 
6210 	if (e1000e_enable_mng_pass_thru(&adapter->hw))
6211 		adapter->flags |= FLAG_MNG_PT_ENABLED;
6212 
6213 	/*
6214 	 * before reading the NVM, reset the controller to
6215 	 * put the device in a known good starting state
6216 	 */
6217 	adapter->hw.mac.ops.reset_hw(&adapter->hw);
6218 
6219 	/*
6220 	 * systems with ASPM and others may see the checksum fail on the first
6221 	 * attempt. Let's give it a few tries
6222 	 */
6223 	for (i = 0;; i++) {
6224 		if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
6225 			break;
6226 		if (i == 2) {
6227 			e_err("The NVM Checksum Is Not Valid\n");
6228 			err = -EIO;
6229 			goto err_eeprom;
6230 		}
6231 	}
6232 
6233 	e1000_eeprom_checks(adapter);
6234 
6235 	/* copy the MAC address */
6236 	if (e1000e_read_mac_addr(&adapter->hw))
6237 		e_err("NVM Read Error while reading MAC address\n");
6238 
6239 	memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
6240 	memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
6241 
6242 	if (!is_valid_ether_addr(netdev->perm_addr)) {
6243 		e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
6244 		err = -EIO;
6245 		goto err_eeprom;
6246 	}
6247 
6248 	init_timer(&adapter->watchdog_timer);
6249 	adapter->watchdog_timer.function = e1000_watchdog;
6250 	adapter->watchdog_timer.data = (unsigned long) adapter;
6251 
6252 	init_timer(&adapter->phy_info_timer);
6253 	adapter->phy_info_timer.function = e1000_update_phy_info;
6254 	adapter->phy_info_timer.data = (unsigned long) adapter;
6255 
6256 	INIT_WORK(&adapter->reset_task, e1000_reset_task);
6257 	INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
6258 	INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
6259 	INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
6260 	INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
6261 
6262 	/* Initialize link parameters. User can change them with ethtool */
6263 	adapter->hw.mac.autoneg = 1;
6264 	adapter->fc_autoneg = true;
6265 	adapter->hw.fc.requested_mode = e1000_fc_default;
6266 	adapter->hw.fc.current_mode = e1000_fc_default;
6267 	adapter->hw.phy.autoneg_advertised = 0x2f;
6268 
6269 	/* ring size defaults */
6270 	adapter->rx_ring->count = 256;
6271 	adapter->tx_ring->count = 256;
6272 
6273 	/*
6274 	 * Initial Wake on LAN setting - If APM wake is enabled in
6275 	 * the EEPROM, enable the ACPI Magic Packet filter
6276 	 */
6277 	if (adapter->flags & FLAG_APME_IN_WUC) {
6278 		/* APME bit in EEPROM is mapped to WUC.APME */
6279 		eeprom_data = er32(WUC);
6280 		eeprom_apme_mask = E1000_WUC_APME;
6281 		if ((hw->mac.type > e1000_ich10lan) &&
6282 		    (eeprom_data & E1000_WUC_PHY_WAKE))
6283 			adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
6284 	} else if (adapter->flags & FLAG_APME_IN_CTRL3) {
6285 		if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
6286 		    (adapter->hw.bus.func == 1))
6287 			e1000_read_nvm(&adapter->hw,
6288 				NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
6289 		else
6290 			e1000_read_nvm(&adapter->hw,
6291 				NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
6292 	}
6293 
6294 	/* fetch WoL from EEPROM */
6295 	if (eeprom_data & eeprom_apme_mask)
6296 		adapter->eeprom_wol |= E1000_WUFC_MAG;
6297 
6298 	/*
6299 	 * now that we have the eeprom settings, apply the special cases
6300 	 * where the eeprom may be wrong or the board simply won't support
6301 	 * wake on lan on a particular port
6302 	 */
6303 	if (!(adapter->flags & FLAG_HAS_WOL))
6304 		adapter->eeprom_wol = 0;
6305 
6306 	/* initialize the wol settings based on the eeprom settings */
6307 	adapter->wol = adapter->eeprom_wol;
6308 	device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6309 
6310 	/* save off EEPROM version number */
6311 	e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6312 
6313 	/* reset the hardware with the new settings */
6314 	e1000e_reset(adapter);
6315 
6316 	/*
6317 	 * If the controller has AMT, do not set DRV_LOAD until the interface
6318 	 * is up.  For all other cases, let the f/w know that the h/w is now
6319 	 * under the control of the driver.
6320 	 */
6321 	if (!(adapter->flags & FLAG_HAS_AMT))
6322 		e1000e_get_hw_control(adapter);
6323 
6324 	strlcpy(netdev->name, "eth%d", sizeof(netdev->name));
6325 	err = register_netdev(netdev);
6326 	if (err)
6327 		goto err_register;
6328 
6329 	/* carrier off reporting is important to ethtool even BEFORE open */
6330 	netif_carrier_off(netdev);
6331 
6332 	e1000_print_device_info(adapter);
6333 
6334 	if (pci_dev_run_wake(pdev))
6335 		pm_runtime_put_noidle(&pdev->dev);
6336 
6337 	return 0;
6338 
6339 err_register:
6340 	if (!(adapter->flags & FLAG_HAS_AMT))
6341 		e1000e_release_hw_control(adapter);
6342 err_eeprom:
6343 	if (!e1000_check_reset_block(&adapter->hw))
6344 		e1000_phy_hw_reset(&adapter->hw);
6345 err_hw_init:
6346 	kfree(adapter->tx_ring);
6347 	kfree(adapter->rx_ring);
6348 err_sw_init:
6349 	if (adapter->hw.flash_address)
6350 		iounmap(adapter->hw.flash_address);
6351 	e1000e_reset_interrupt_capability(adapter);
6352 err_flashmap:
6353 	iounmap(adapter->hw.hw_addr);
6354 err_ioremap:
6355 	free_netdev(netdev);
6356 err_alloc_etherdev:
6357 	pci_release_selected_regions(pdev,
6358 	                             pci_select_bars(pdev, IORESOURCE_MEM));
6359 err_pci_reg:
6360 err_dma:
6361 	pci_disable_device(pdev);
6362 	return err;
6363 }
6364 
6365 /**
6366  * e1000_remove - Device Removal Routine
6367  * @pdev: PCI device information struct
6368  *
6369  * e1000_remove is called by the PCI subsystem to alert the driver
6370  * that it should release a PCI device.  The could be caused by a
6371  * Hot-Plug event, or because the driver is going to be removed from
6372  * memory.
6373  **/
6374 static void __devexit e1000_remove(struct pci_dev *pdev)
6375 {
6376 	struct net_device *netdev = pci_get_drvdata(pdev);
6377 	struct e1000_adapter *adapter = netdev_priv(netdev);
6378 	bool down = test_bit(__E1000_DOWN, &adapter->state);
6379 
6380 	/*
6381 	 * The timers may be rescheduled, so explicitly disable them
6382 	 * from being rescheduled.
6383 	 */
6384 	if (!down)
6385 		set_bit(__E1000_DOWN, &adapter->state);
6386 	del_timer_sync(&adapter->watchdog_timer);
6387 	del_timer_sync(&adapter->phy_info_timer);
6388 
6389 	cancel_work_sync(&adapter->reset_task);
6390 	cancel_work_sync(&adapter->watchdog_task);
6391 	cancel_work_sync(&adapter->downshift_task);
6392 	cancel_work_sync(&adapter->update_phy_task);
6393 	cancel_work_sync(&adapter->print_hang_task);
6394 
6395 	if (!(netdev->flags & IFF_UP))
6396 		e1000_power_down_phy(adapter);
6397 
6398 	/* Don't lie to e1000_close() down the road. */
6399 	if (!down)
6400 		clear_bit(__E1000_DOWN, &adapter->state);
6401 	unregister_netdev(netdev);
6402 
6403 	if (pci_dev_run_wake(pdev))
6404 		pm_runtime_get_noresume(&pdev->dev);
6405 
6406 	/*
6407 	 * Release control of h/w to f/w.  If f/w is AMT enabled, this
6408 	 * would have already happened in close and is redundant.
6409 	 */
6410 	e1000e_release_hw_control(adapter);
6411 
6412 	e1000e_reset_interrupt_capability(adapter);
6413 	kfree(adapter->tx_ring);
6414 	kfree(adapter->rx_ring);
6415 
6416 	iounmap(adapter->hw.hw_addr);
6417 	if (adapter->hw.flash_address)
6418 		iounmap(adapter->hw.flash_address);
6419 	pci_release_selected_regions(pdev,
6420 	                             pci_select_bars(pdev, IORESOURCE_MEM));
6421 
6422 	free_netdev(netdev);
6423 
6424 	/* AER disable */
6425 	pci_disable_pcie_error_reporting(pdev);
6426 
6427 	pci_disable_device(pdev);
6428 }
6429 
6430 /* PCI Error Recovery (ERS) */
6431 static struct pci_error_handlers e1000_err_handler = {
6432 	.error_detected = e1000_io_error_detected,
6433 	.slot_reset = e1000_io_slot_reset,
6434 	.resume = e1000_io_resume,
6435 };
6436 
6437 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6438 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6439 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6440 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6441 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6442 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6443 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6444 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6445 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6446 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6447 
6448 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6449 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6450 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6451 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6452 
6453 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6454 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6455 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6456 
6457 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6458 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6459 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6460 
6461 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6462 	  board_80003es2lan },
6463 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6464 	  board_80003es2lan },
6465 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6466 	  board_80003es2lan },
6467 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6468 	  board_80003es2lan },
6469 
6470 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6471 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6472 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6473 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6474 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6475 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6476 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6477 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6478 
6479 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6480 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6481 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6482 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6483 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6484 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6485 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6486 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6487 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6488 
6489 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6490 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6491 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6492 
6493 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6494 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6495 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6496 
6497 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6498 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6499 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6500 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6501 
6502 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6503 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6504 
6505 	{ }	/* terminate list */
6506 };
6507 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6508 
6509 #ifdef CONFIG_PM
6510 static const struct dev_pm_ops e1000_pm_ops = {
6511 	SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6512 	SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6513 				e1000_runtime_resume, e1000_idle)
6514 };
6515 #endif
6516 
6517 /* PCI Device API Driver */
6518 static struct pci_driver e1000_driver = {
6519 	.name     = e1000e_driver_name,
6520 	.id_table = e1000_pci_tbl,
6521 	.probe    = e1000_probe,
6522 	.remove   = __devexit_p(e1000_remove),
6523 #ifdef CONFIG_PM
6524 	.driver.pm = &e1000_pm_ops,
6525 #endif
6526 	.shutdown = e1000_shutdown,
6527 	.err_handler = &e1000_err_handler
6528 };
6529 
6530 /**
6531  * e1000_init_module - Driver Registration Routine
6532  *
6533  * e1000_init_module is the first routine called when the driver is
6534  * loaded. All it does is register with the PCI subsystem.
6535  **/
6536 static int __init e1000_init_module(void)
6537 {
6538 	int ret;
6539 	pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6540 		e1000e_driver_version);
6541 	pr_info("Copyright(c) 1999 - 2012 Intel Corporation.\n");
6542 	ret = pci_register_driver(&e1000_driver);
6543 
6544 	return ret;
6545 }
6546 module_init(e1000_init_module);
6547 
6548 /**
6549  * e1000_exit_module - Driver Exit Cleanup Routine
6550  *
6551  * e1000_exit_module is called just before the driver is removed
6552  * from memory.
6553  **/
6554 static void __exit e1000_exit_module(void)
6555 {
6556 	pci_unregister_driver(&e1000_driver);
6557 }
6558 module_exit(e1000_exit_module);
6559 
6560 
6561 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6562 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6563 MODULE_LICENSE("GPL");
6564 MODULE_VERSION(DRV_VERSION);
6565 
6566 /* e1000_main.c */
6567