xref: /linux/drivers/net/ethernet/intel/e1000e/netdev.c (revision baf69e275317602bb1f2c87e40d1e03c90b31e38)
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 "2.0.0" DRV_EXTRAVERSION
60 char e1000e_driver_name[] = "e1000e";
61 const char e1000e_driver_version[] = DRV_VERSION;
62 
63 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
64 static int debug = -1;
65 module_param(debug, int, 0);
66 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
67 
68 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state);
69 
70 static const struct e1000_info *e1000_info_tbl[] = {
71 	[board_82571]		= &e1000_82571_info,
72 	[board_82572]		= &e1000_82572_info,
73 	[board_82573]		= &e1000_82573_info,
74 	[board_82574]		= &e1000_82574_info,
75 	[board_82583]		= &e1000_82583_info,
76 	[board_80003es2lan]	= &e1000_es2_info,
77 	[board_ich8lan]		= &e1000_ich8_info,
78 	[board_ich9lan]		= &e1000_ich9_info,
79 	[board_ich10lan]	= &e1000_ich10_info,
80 	[board_pchlan]		= &e1000_pch_info,
81 	[board_pch2lan]		= &e1000_pch2_info,
82 	[board_pch_lpt]		= &e1000_pch_lpt_info,
83 };
84 
85 struct e1000_reg_info {
86 	u32 ofs;
87 	char *name;
88 };
89 
90 #define E1000_RDFH	0x02410	/* Rx Data FIFO Head - RW */
91 #define E1000_RDFT	0x02418	/* Rx Data FIFO Tail - RW */
92 #define E1000_RDFHS	0x02420	/* Rx Data FIFO Head Saved - RW */
93 #define E1000_RDFTS	0x02428	/* Rx Data FIFO Tail Saved - RW */
94 #define E1000_RDFPC	0x02430	/* Rx Data FIFO Packet Count - RW */
95 
96 #define E1000_TDFH	0x03410	/* Tx Data FIFO Head - RW */
97 #define E1000_TDFT	0x03418	/* Tx Data FIFO Tail - RW */
98 #define E1000_TDFHS	0x03420	/* Tx Data FIFO Head Saved - RW */
99 #define E1000_TDFTS	0x03428	/* Tx Data FIFO Tail Saved - RW */
100 #define E1000_TDFPC	0x03430	/* Tx Data FIFO Packet Count - RW */
101 
102 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
103 
104 	/* General Registers */
105 	{E1000_CTRL, "CTRL"},
106 	{E1000_STATUS, "STATUS"},
107 	{E1000_CTRL_EXT, "CTRL_EXT"},
108 
109 	/* Interrupt Registers */
110 	{E1000_ICR, "ICR"},
111 
112 	/* Rx Registers */
113 	{E1000_RCTL, "RCTL"},
114 	{E1000_RDLEN(0), "RDLEN"},
115 	{E1000_RDH(0), "RDH"},
116 	{E1000_RDT(0), "RDT"},
117 	{E1000_RDTR, "RDTR"},
118 	{E1000_RXDCTL(0), "RXDCTL"},
119 	{E1000_ERT, "ERT"},
120 	{E1000_RDBAL(0), "RDBAL"},
121 	{E1000_RDBAH(0), "RDBAH"},
122 	{E1000_RDFH, "RDFH"},
123 	{E1000_RDFT, "RDFT"},
124 	{E1000_RDFHS, "RDFHS"},
125 	{E1000_RDFTS, "RDFTS"},
126 	{E1000_RDFPC, "RDFPC"},
127 
128 	/* Tx Registers */
129 	{E1000_TCTL, "TCTL"},
130 	{E1000_TDBAL(0), "TDBAL"},
131 	{E1000_TDBAH(0), "TDBAH"},
132 	{E1000_TDLEN(0), "TDLEN"},
133 	{E1000_TDH(0), "TDH"},
134 	{E1000_TDT(0), "TDT"},
135 	{E1000_TIDV, "TIDV"},
136 	{E1000_TXDCTL(0), "TXDCTL"},
137 	{E1000_TADV, "TADV"},
138 	{E1000_TARC(0), "TARC"},
139 	{E1000_TDFH, "TDFH"},
140 	{E1000_TDFT, "TDFT"},
141 	{E1000_TDFHS, "TDFHS"},
142 	{E1000_TDFTS, "TDFTS"},
143 	{E1000_TDFPC, "TDFPC"},
144 
145 	/* List Terminator */
146 	{0, NULL}
147 };
148 
149 /*
150  * e1000_regdump - register printout routine
151  */
152 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
153 {
154 	int n = 0;
155 	char rname[16];
156 	u32 regs[8];
157 
158 	switch (reginfo->ofs) {
159 	case E1000_RXDCTL(0):
160 		for (n = 0; n < 2; n++)
161 			regs[n] = __er32(hw, E1000_RXDCTL(n));
162 		break;
163 	case E1000_TXDCTL(0):
164 		for (n = 0; n < 2; n++)
165 			regs[n] = __er32(hw, E1000_TXDCTL(n));
166 		break;
167 	case E1000_TARC(0):
168 		for (n = 0; n < 2; n++)
169 			regs[n] = __er32(hw, E1000_TARC(n));
170 		break;
171 	default:
172 		pr_info("%-15s %08x\n",
173 			reginfo->name, __er32(hw, reginfo->ofs));
174 		return;
175 	}
176 
177 	snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
178 	pr_info("%-15s %08x %08x\n", rname, regs[0], regs[1]);
179 }
180 
181 /*
182  * e1000e_dump - Print registers, Tx-ring and Rx-ring
183  */
184 static void e1000e_dump(struct e1000_adapter *adapter)
185 {
186 	struct net_device *netdev = adapter->netdev;
187 	struct e1000_hw *hw = &adapter->hw;
188 	struct e1000_reg_info *reginfo;
189 	struct e1000_ring *tx_ring = adapter->tx_ring;
190 	struct e1000_tx_desc *tx_desc;
191 	struct my_u0 {
192 		__le64 a;
193 		__le64 b;
194 	} *u0;
195 	struct e1000_buffer *buffer_info;
196 	struct e1000_ring *rx_ring = adapter->rx_ring;
197 	union e1000_rx_desc_packet_split *rx_desc_ps;
198 	union e1000_rx_desc_extended *rx_desc;
199 	struct my_u1 {
200 		__le64 a;
201 		__le64 b;
202 		__le64 c;
203 		__le64 d;
204 	} *u1;
205 	u32 staterr;
206 	int i = 0;
207 
208 	if (!netif_msg_hw(adapter))
209 		return;
210 
211 	/* Print netdevice Info */
212 	if (netdev) {
213 		dev_info(&adapter->pdev->dev, "Net device Info\n");
214 		pr_info("Device Name     state            trans_start      last_rx\n");
215 		pr_info("%-15s %016lX %016lX %016lX\n",
216 			netdev->name, netdev->state, netdev->trans_start,
217 			netdev->last_rx);
218 	}
219 
220 	/* Print Registers */
221 	dev_info(&adapter->pdev->dev, "Register Dump\n");
222 	pr_info(" Register Name   Value\n");
223 	for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
224 	     reginfo->name; reginfo++) {
225 		e1000_regdump(hw, reginfo);
226 	}
227 
228 	/* Print Tx Ring Summary */
229 	if (!netdev || !netif_running(netdev))
230 		return;
231 
232 	dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
233 	pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
234 	buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
235 	pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
236 		0, tx_ring->next_to_use, tx_ring->next_to_clean,
237 		(unsigned long long)buffer_info->dma,
238 		buffer_info->length,
239 		buffer_info->next_to_watch,
240 		(unsigned long long)buffer_info->time_stamp);
241 
242 	/* Print Tx Ring */
243 	if (!netif_msg_tx_done(adapter))
244 		goto rx_ring_summary;
245 
246 	dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
247 
248 	/* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
249 	 *
250 	 * Legacy Transmit Descriptor
251 	 *   +--------------------------------------------------------------+
252 	 * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
253 	 *   +--------------------------------------------------------------+
254 	 * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
255 	 *   +--------------------------------------------------------------+
256 	 *   63       48 47        36 35    32 31     24 23    16 15        0
257 	 *
258 	 * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
259 	 *   63      48 47    40 39       32 31             16 15    8 7      0
260 	 *   +----------------------------------------------------------------+
261 	 * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
262 	 *   +----------------------------------------------------------------+
263 	 * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
264 	 *   +----------------------------------------------------------------+
265 	 *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
266 	 *
267 	 * Extended Data Descriptor (DTYP=0x1)
268 	 *   +----------------------------------------------------------------+
269 	 * 0 |                     Buffer Address [63:0]                      |
270 	 *   +----------------------------------------------------------------+
271 	 * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
272 	 *   +----------------------------------------------------------------+
273 	 *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
274 	 */
275 	pr_info("Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Legacy format\n");
276 	pr_info("Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Context format\n");
277 	pr_info("Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Data format\n");
278 	for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
279 		const char *next_desc;
280 		tx_desc = E1000_TX_DESC(*tx_ring, i);
281 		buffer_info = &tx_ring->buffer_info[i];
282 		u0 = (struct my_u0 *)tx_desc;
283 		if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
284 			next_desc = " NTC/U";
285 		else if (i == tx_ring->next_to_use)
286 			next_desc = " NTU";
287 		else if (i == tx_ring->next_to_clean)
288 			next_desc = " NTC";
289 		else
290 			next_desc = "";
291 		pr_info("T%c[0x%03X]    %016llX %016llX %016llX %04X  %3X %016llX %p%s\n",
292 			(!(le64_to_cpu(u0->b) & (1 << 29)) ? 'l' :
293 			 ((le64_to_cpu(u0->b) & (1 << 20)) ? 'd' : 'c')),
294 			i,
295 			(unsigned long long)le64_to_cpu(u0->a),
296 			(unsigned long long)le64_to_cpu(u0->b),
297 			(unsigned long long)buffer_info->dma,
298 			buffer_info->length, buffer_info->next_to_watch,
299 			(unsigned long long)buffer_info->time_stamp,
300 			buffer_info->skb, next_desc);
301 
302 		if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
303 			print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
304 				       16, 1, phys_to_virt(buffer_info->dma),
305 				       buffer_info->length, true);
306 	}
307 
308 	/* Print Rx Ring Summary */
309 rx_ring_summary:
310 	dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
311 	pr_info("Queue [NTU] [NTC]\n");
312 	pr_info(" %5d %5X %5X\n",
313 		0, rx_ring->next_to_use, rx_ring->next_to_clean);
314 
315 	/* Print Rx Ring */
316 	if (!netif_msg_rx_status(adapter))
317 		return;
318 
319 	dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
320 	switch (adapter->rx_ps_pages) {
321 	case 1:
322 	case 2:
323 	case 3:
324 		/* [Extended] Packet Split Receive Descriptor Format
325 		 *
326 		 *    +-----------------------------------------------------+
327 		 *  0 |                Buffer Address 0 [63:0]              |
328 		 *    +-----------------------------------------------------+
329 		 *  8 |                Buffer Address 1 [63:0]              |
330 		 *    +-----------------------------------------------------+
331 		 * 16 |                Buffer Address 2 [63:0]              |
332 		 *    +-----------------------------------------------------+
333 		 * 24 |                Buffer Address 3 [63:0]              |
334 		 *    +-----------------------------------------------------+
335 		 */
336 		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");
337 		/* [Extended] Receive Descriptor (Write-Back) Format
338 		 *
339 		 *   63       48 47    32 31     13 12    8 7    4 3        0
340 		 *   +------------------------------------------------------+
341 		 * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
342 		 *   | Checksum | Ident  |         | Queue |      |  Type   |
343 		 *   +------------------------------------------------------+
344 		 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
345 		 *   +------------------------------------------------------+
346 		 *   63       48 47    32 31            20 19               0
347 		 */
348 		pr_info("RWB[desc]      [ck ipid mrqhsh] [vl   l0 ee  es] [ l3  l2  l1 hs] [reserved      ] ---------------- [bi->skb] <-- Ext Rx Write-Back format\n");
349 		for (i = 0; i < rx_ring->count; i++) {
350 			const char *next_desc;
351 			buffer_info = &rx_ring->buffer_info[i];
352 			rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
353 			u1 = (struct my_u1 *)rx_desc_ps;
354 			staterr =
355 			    le32_to_cpu(rx_desc_ps->wb.middle.status_error);
356 
357 			if (i == rx_ring->next_to_use)
358 				next_desc = " NTU";
359 			else if (i == rx_ring->next_to_clean)
360 				next_desc = " NTC";
361 			else
362 				next_desc = "";
363 
364 			if (staterr & E1000_RXD_STAT_DD) {
365 				/* Descriptor Done */
366 				pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX ---------------- %p%s\n",
367 					"RWB", i,
368 					(unsigned long long)le64_to_cpu(u1->a),
369 					(unsigned long long)le64_to_cpu(u1->b),
370 					(unsigned long long)le64_to_cpu(u1->c),
371 					(unsigned long long)le64_to_cpu(u1->d),
372 					buffer_info->skb, next_desc);
373 			} else {
374 				pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX %016llX %p%s\n",
375 					"R  ", i,
376 					(unsigned long long)le64_to_cpu(u1->a),
377 					(unsigned long long)le64_to_cpu(u1->b),
378 					(unsigned long long)le64_to_cpu(u1->c),
379 					(unsigned long long)le64_to_cpu(u1->d),
380 					(unsigned long long)buffer_info->dma,
381 					buffer_info->skb, next_desc);
382 
383 				if (netif_msg_pktdata(adapter))
384 					print_hex_dump(KERN_INFO, "",
385 						DUMP_PREFIX_ADDRESS, 16, 1,
386 						phys_to_virt(buffer_info->dma),
387 						adapter->rx_ps_bsize0, true);
388 			}
389 		}
390 		break;
391 	default:
392 	case 0:
393 		/* Extended Receive Descriptor (Read) Format
394 		 *
395 		 *   +-----------------------------------------------------+
396 		 * 0 |                Buffer Address [63:0]                |
397 		 *   +-----------------------------------------------------+
398 		 * 8 |                      Reserved                       |
399 		 *   +-----------------------------------------------------+
400 		 */
401 		pr_info("R  [desc]      [buf addr 63:0 ] [reserved 63:0 ] [bi->dma       ] [bi->skb] <-- Ext (Read) format\n");
402 		/* Extended Receive Descriptor (Write-Back) Format
403 		 *
404 		 *   63       48 47    32 31    24 23            4 3        0
405 		 *   +------------------------------------------------------+
406 		 *   |     RSS Hash      |        |               |         |
407 		 * 0 +-------------------+  Rsvd  |   Reserved    | MRQ RSS |
408 		 *   | Packet   | IP     |        |               |  Type   |
409 		 *   | Checksum | Ident  |        |               |         |
410 		 *   +------------------------------------------------------+
411 		 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
412 		 *   +------------------------------------------------------+
413 		 *   63       48 47    32 31            20 19               0
414 		 */
415 		pr_info("RWB[desc]      [cs ipid    mrq] [vt   ln xe  xs] [bi->skb] <-- Ext (Write-Back) format\n");
416 
417 		for (i = 0; i < rx_ring->count; i++) {
418 			const char *next_desc;
419 
420 			buffer_info = &rx_ring->buffer_info[i];
421 			rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
422 			u1 = (struct my_u1 *)rx_desc;
423 			staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
424 
425 			if (i == rx_ring->next_to_use)
426 				next_desc = " NTU";
427 			else if (i == rx_ring->next_to_clean)
428 				next_desc = " NTC";
429 			else
430 				next_desc = "";
431 
432 			if (staterr & E1000_RXD_STAT_DD) {
433 				/* Descriptor Done */
434 				pr_info("%s[0x%03X]     %016llX %016llX ---------------- %p%s\n",
435 					"RWB", i,
436 					(unsigned long long)le64_to_cpu(u1->a),
437 					(unsigned long long)le64_to_cpu(u1->b),
438 					buffer_info->skb, next_desc);
439 			} else {
440 				pr_info("%s[0x%03X]     %016llX %016llX %016llX %p%s\n",
441 					"R  ", i,
442 					(unsigned long long)le64_to_cpu(u1->a),
443 					(unsigned long long)le64_to_cpu(u1->b),
444 					(unsigned long long)buffer_info->dma,
445 					buffer_info->skb, next_desc);
446 
447 				if (netif_msg_pktdata(adapter))
448 					print_hex_dump(KERN_INFO, "",
449 						       DUMP_PREFIX_ADDRESS, 16,
450 						       1,
451 						       phys_to_virt
452 						       (buffer_info->dma),
453 						       adapter->rx_buffer_len,
454 						       true);
455 			}
456 		}
457 	}
458 }
459 
460 /**
461  * e1000_desc_unused - calculate if we have unused descriptors
462  **/
463 static int e1000_desc_unused(struct e1000_ring *ring)
464 {
465 	if (ring->next_to_clean > ring->next_to_use)
466 		return ring->next_to_clean - ring->next_to_use - 1;
467 
468 	return ring->count + ring->next_to_clean - ring->next_to_use - 1;
469 }
470 
471 /**
472  * e1000_receive_skb - helper function to handle Rx indications
473  * @adapter: board private structure
474  * @status: descriptor status field as written by hardware
475  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
476  * @skb: pointer to sk_buff to be indicated to stack
477  **/
478 static void e1000_receive_skb(struct e1000_adapter *adapter,
479 			      struct net_device *netdev, struct sk_buff *skb,
480 			      u8 status, __le16 vlan)
481 {
482 	u16 tag = le16_to_cpu(vlan);
483 	skb->protocol = eth_type_trans(skb, netdev);
484 
485 	if (status & E1000_RXD_STAT_VP)
486 		__vlan_hwaccel_put_tag(skb, tag);
487 
488 	napi_gro_receive(&adapter->napi, skb);
489 }
490 
491 /**
492  * e1000_rx_checksum - Receive Checksum Offload
493  * @adapter: board private structure
494  * @status_err: receive descriptor status and error fields
495  * @csum: receive descriptor csum field
496  * @sk_buff: socket buffer with received data
497  **/
498 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
499 			      struct sk_buff *skb)
500 {
501 	u16 status = (u16)status_err;
502 	u8 errors = (u8)(status_err >> 24);
503 
504 	skb_checksum_none_assert(skb);
505 
506 	/* Rx checksum disabled */
507 	if (!(adapter->netdev->features & NETIF_F_RXCSUM))
508 		return;
509 
510 	/* Ignore Checksum bit is set */
511 	if (status & E1000_RXD_STAT_IXSM)
512 		return;
513 
514 	/* TCP/UDP checksum error bit or IP checksum error bit is set */
515 	if (errors & (E1000_RXD_ERR_TCPE | E1000_RXD_ERR_IPE)) {
516 		/* let the stack verify checksum errors */
517 		adapter->hw_csum_err++;
518 		return;
519 	}
520 
521 	/* TCP/UDP Checksum has not been calculated */
522 	if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
523 		return;
524 
525 	/* It must be a TCP or UDP packet with a valid checksum */
526 	skb->ip_summed = CHECKSUM_UNNECESSARY;
527 	adapter->hw_csum_good++;
528 }
529 
530 static void e1000e_update_rdt_wa(struct e1000_ring *rx_ring, unsigned int i)
531 {
532 	struct e1000_adapter *adapter = rx_ring->adapter;
533 	struct e1000_hw *hw = &adapter->hw;
534 	s32 ret_val = __ew32_prepare(hw);
535 
536 	writel(i, rx_ring->tail);
537 
538 	if (unlikely(!ret_val && (i != readl(rx_ring->tail)))) {
539 		u32 rctl = er32(RCTL);
540 		ew32(RCTL, rctl & ~E1000_RCTL_EN);
541 		e_err("ME firmware caused invalid RDT - resetting\n");
542 		schedule_work(&adapter->reset_task);
543 	}
544 }
545 
546 static void e1000e_update_tdt_wa(struct e1000_ring *tx_ring, unsigned int i)
547 {
548 	struct e1000_adapter *adapter = tx_ring->adapter;
549 	struct e1000_hw *hw = &adapter->hw;
550 	s32 ret_val = __ew32_prepare(hw);
551 
552 	writel(i, tx_ring->tail);
553 
554 	if (unlikely(!ret_val && (i != readl(tx_ring->tail)))) {
555 		u32 tctl = er32(TCTL);
556 		ew32(TCTL, tctl & ~E1000_TCTL_EN);
557 		e_err("ME firmware caused invalid TDT - resetting\n");
558 		schedule_work(&adapter->reset_task);
559 	}
560 }
561 
562 /**
563  * e1000_alloc_rx_buffers - Replace used receive buffers
564  * @rx_ring: Rx descriptor ring
565  **/
566 static void e1000_alloc_rx_buffers(struct e1000_ring *rx_ring,
567 				   int cleaned_count, gfp_t gfp)
568 {
569 	struct e1000_adapter *adapter = rx_ring->adapter;
570 	struct net_device *netdev = adapter->netdev;
571 	struct pci_dev *pdev = adapter->pdev;
572 	union e1000_rx_desc_extended *rx_desc;
573 	struct e1000_buffer *buffer_info;
574 	struct sk_buff *skb;
575 	unsigned int i;
576 	unsigned int bufsz = adapter->rx_buffer_len;
577 
578 	i = rx_ring->next_to_use;
579 	buffer_info = &rx_ring->buffer_info[i];
580 
581 	while (cleaned_count--) {
582 		skb = buffer_info->skb;
583 		if (skb) {
584 			skb_trim(skb, 0);
585 			goto map_skb;
586 		}
587 
588 		skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
589 		if (!skb) {
590 			/* Better luck next round */
591 			adapter->alloc_rx_buff_failed++;
592 			break;
593 		}
594 
595 		buffer_info->skb = skb;
596 map_skb:
597 		buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
598 						  adapter->rx_buffer_len,
599 						  DMA_FROM_DEVICE);
600 		if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
601 			dev_err(&pdev->dev, "Rx DMA map failed\n");
602 			adapter->rx_dma_failed++;
603 			break;
604 		}
605 
606 		rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
607 		rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
608 
609 		if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
610 			/*
611 			 * Force memory writes to complete before letting h/w
612 			 * know there are new descriptors to fetch.  (Only
613 			 * applicable for weak-ordered memory model archs,
614 			 * such as IA-64).
615 			 */
616 			wmb();
617 			if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
618 				e1000e_update_rdt_wa(rx_ring, i);
619 			else
620 				writel(i, rx_ring->tail);
621 		}
622 		i++;
623 		if (i == rx_ring->count)
624 			i = 0;
625 		buffer_info = &rx_ring->buffer_info[i];
626 	}
627 
628 	rx_ring->next_to_use = i;
629 }
630 
631 /**
632  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
633  * @rx_ring: Rx descriptor ring
634  **/
635 static void e1000_alloc_rx_buffers_ps(struct e1000_ring *rx_ring,
636 				      int cleaned_count, gfp_t gfp)
637 {
638 	struct e1000_adapter *adapter = rx_ring->adapter;
639 	struct net_device *netdev = adapter->netdev;
640 	struct pci_dev *pdev = adapter->pdev;
641 	union e1000_rx_desc_packet_split *rx_desc;
642 	struct e1000_buffer *buffer_info;
643 	struct e1000_ps_page *ps_page;
644 	struct sk_buff *skb;
645 	unsigned int i, j;
646 
647 	i = rx_ring->next_to_use;
648 	buffer_info = &rx_ring->buffer_info[i];
649 
650 	while (cleaned_count--) {
651 		rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
652 
653 		for (j = 0; j < PS_PAGE_BUFFERS; j++) {
654 			ps_page = &buffer_info->ps_pages[j];
655 			if (j >= adapter->rx_ps_pages) {
656 				/* all unused desc entries get hw null ptr */
657 				rx_desc->read.buffer_addr[j + 1] =
658 				    ~cpu_to_le64(0);
659 				continue;
660 			}
661 			if (!ps_page->page) {
662 				ps_page->page = alloc_page(gfp);
663 				if (!ps_page->page) {
664 					adapter->alloc_rx_buff_failed++;
665 					goto no_buffers;
666 				}
667 				ps_page->dma = dma_map_page(&pdev->dev,
668 							    ps_page->page,
669 							    0, PAGE_SIZE,
670 							    DMA_FROM_DEVICE);
671 				if (dma_mapping_error(&pdev->dev,
672 						      ps_page->dma)) {
673 					dev_err(&adapter->pdev->dev,
674 						"Rx DMA page map failed\n");
675 					adapter->rx_dma_failed++;
676 					goto no_buffers;
677 				}
678 			}
679 			/*
680 			 * Refresh the desc even if buffer_addrs
681 			 * didn't change because each write-back
682 			 * erases this info.
683 			 */
684 			rx_desc->read.buffer_addr[j + 1] =
685 			    cpu_to_le64(ps_page->dma);
686 		}
687 
688 		skb = __netdev_alloc_skb_ip_align(netdev,
689 						  adapter->rx_ps_bsize0,
690 						  gfp);
691 
692 		if (!skb) {
693 			adapter->alloc_rx_buff_failed++;
694 			break;
695 		}
696 
697 		buffer_info->skb = skb;
698 		buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
699 						  adapter->rx_ps_bsize0,
700 						  DMA_FROM_DEVICE);
701 		if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
702 			dev_err(&pdev->dev, "Rx DMA map failed\n");
703 			adapter->rx_dma_failed++;
704 			/* cleanup skb */
705 			dev_kfree_skb_any(skb);
706 			buffer_info->skb = NULL;
707 			break;
708 		}
709 
710 		rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
711 
712 		if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
713 			/*
714 			 * Force memory writes to complete before letting h/w
715 			 * know there are new descriptors to fetch.  (Only
716 			 * applicable for weak-ordered memory model archs,
717 			 * such as IA-64).
718 			 */
719 			wmb();
720 			if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
721 				e1000e_update_rdt_wa(rx_ring, i << 1);
722 			else
723 				writel(i << 1, rx_ring->tail);
724 		}
725 
726 		i++;
727 		if (i == rx_ring->count)
728 			i = 0;
729 		buffer_info = &rx_ring->buffer_info[i];
730 	}
731 
732 no_buffers:
733 	rx_ring->next_to_use = i;
734 }
735 
736 /**
737  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
738  * @rx_ring: Rx descriptor ring
739  * @cleaned_count: number of buffers to allocate this pass
740  **/
741 
742 static void e1000_alloc_jumbo_rx_buffers(struct e1000_ring *rx_ring,
743 					 int cleaned_count, gfp_t gfp)
744 {
745 	struct e1000_adapter *adapter = rx_ring->adapter;
746 	struct net_device *netdev = adapter->netdev;
747 	struct pci_dev *pdev = adapter->pdev;
748 	union e1000_rx_desc_extended *rx_desc;
749 	struct e1000_buffer *buffer_info;
750 	struct sk_buff *skb;
751 	unsigned int i;
752 	unsigned int bufsz = 256 - 16 /* for skb_reserve */;
753 
754 	i = rx_ring->next_to_use;
755 	buffer_info = &rx_ring->buffer_info[i];
756 
757 	while (cleaned_count--) {
758 		skb = buffer_info->skb;
759 		if (skb) {
760 			skb_trim(skb, 0);
761 			goto check_page;
762 		}
763 
764 		skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
765 		if (unlikely(!skb)) {
766 			/* Better luck next round */
767 			adapter->alloc_rx_buff_failed++;
768 			break;
769 		}
770 
771 		buffer_info->skb = skb;
772 check_page:
773 		/* allocate a new page if necessary */
774 		if (!buffer_info->page) {
775 			buffer_info->page = alloc_page(gfp);
776 			if (unlikely(!buffer_info->page)) {
777 				adapter->alloc_rx_buff_failed++;
778 				break;
779 			}
780 		}
781 
782 		if (!buffer_info->dma)
783 			buffer_info->dma = dma_map_page(&pdev->dev,
784 			                                buffer_info->page, 0,
785 			                                PAGE_SIZE,
786 							DMA_FROM_DEVICE);
787 
788 		rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
789 		rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
790 
791 		if (unlikely(++i == rx_ring->count))
792 			i = 0;
793 		buffer_info = &rx_ring->buffer_info[i];
794 	}
795 
796 	if (likely(rx_ring->next_to_use != i)) {
797 		rx_ring->next_to_use = i;
798 		if (unlikely(i-- == 0))
799 			i = (rx_ring->count - 1);
800 
801 		/* Force memory writes to complete before letting h/w
802 		 * know there are new descriptors to fetch.  (Only
803 		 * applicable for weak-ordered memory model archs,
804 		 * such as IA-64). */
805 		wmb();
806 		if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
807 			e1000e_update_rdt_wa(rx_ring, i);
808 		else
809 			writel(i, rx_ring->tail);
810 	}
811 }
812 
813 static inline void e1000_rx_hash(struct net_device *netdev, __le32 rss,
814 				 struct sk_buff *skb)
815 {
816 	if (netdev->features & NETIF_F_RXHASH)
817 		skb->rxhash = le32_to_cpu(rss);
818 }
819 
820 /**
821  * e1000_clean_rx_irq - Send received data up the network stack
822  * @rx_ring: Rx descriptor ring
823  *
824  * the return value indicates whether actual cleaning was done, there
825  * is no guarantee that everything was cleaned
826  **/
827 static bool e1000_clean_rx_irq(struct e1000_ring *rx_ring, int *work_done,
828 			       int work_to_do)
829 {
830 	struct e1000_adapter *adapter = rx_ring->adapter;
831 	struct net_device *netdev = adapter->netdev;
832 	struct pci_dev *pdev = adapter->pdev;
833 	struct e1000_hw *hw = &adapter->hw;
834 	union e1000_rx_desc_extended *rx_desc, *next_rxd;
835 	struct e1000_buffer *buffer_info, *next_buffer;
836 	u32 length, staterr;
837 	unsigned int i;
838 	int cleaned_count = 0;
839 	bool cleaned = false;
840 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
841 
842 	i = rx_ring->next_to_clean;
843 	rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
844 	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
845 	buffer_info = &rx_ring->buffer_info[i];
846 
847 	while (staterr & E1000_RXD_STAT_DD) {
848 		struct sk_buff *skb;
849 
850 		if (*work_done >= work_to_do)
851 			break;
852 		(*work_done)++;
853 		rmb();	/* read descriptor and rx_buffer_info after status DD */
854 
855 		skb = buffer_info->skb;
856 		buffer_info->skb = NULL;
857 
858 		prefetch(skb->data - NET_IP_ALIGN);
859 
860 		i++;
861 		if (i == rx_ring->count)
862 			i = 0;
863 		next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
864 		prefetch(next_rxd);
865 
866 		next_buffer = &rx_ring->buffer_info[i];
867 
868 		cleaned = true;
869 		cleaned_count++;
870 		dma_unmap_single(&pdev->dev,
871 				 buffer_info->dma,
872 				 adapter->rx_buffer_len,
873 				 DMA_FROM_DEVICE);
874 		buffer_info->dma = 0;
875 
876 		length = le16_to_cpu(rx_desc->wb.upper.length);
877 
878 		/*
879 		 * !EOP means multiple descriptors were used to store a single
880 		 * packet, if that's the case we need to toss it.  In fact, we
881 		 * need to toss every packet with the EOP bit clear and the
882 		 * next frame that _does_ have the EOP bit set, as it is by
883 		 * definition only a frame fragment
884 		 */
885 		if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
886 			adapter->flags2 |= FLAG2_IS_DISCARDING;
887 
888 		if (adapter->flags2 & FLAG2_IS_DISCARDING) {
889 			/* All receives must fit into a single buffer */
890 			e_dbg("Receive packet consumed multiple buffers\n");
891 			/* recycle */
892 			buffer_info->skb = skb;
893 			if (staterr & E1000_RXD_STAT_EOP)
894 				adapter->flags2 &= ~FLAG2_IS_DISCARDING;
895 			goto next_desc;
896 		}
897 
898 		if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
899 			     !(netdev->features & NETIF_F_RXALL))) {
900 			/* recycle */
901 			buffer_info->skb = skb;
902 			goto next_desc;
903 		}
904 
905 		/* adjust length to remove Ethernet CRC */
906 		if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
907 			/* If configured to store CRC, don't subtract FCS,
908 			 * but keep the FCS bytes out of the total_rx_bytes
909 			 * counter
910 			 */
911 			if (netdev->features & NETIF_F_RXFCS)
912 				total_rx_bytes -= 4;
913 			else
914 				length -= 4;
915 		}
916 
917 		total_rx_bytes += length;
918 		total_rx_packets++;
919 
920 		/*
921 		 * code added for copybreak, this should improve
922 		 * performance for small packets with large amounts
923 		 * of reassembly being done in the stack
924 		 */
925 		if (length < copybreak) {
926 			struct sk_buff *new_skb =
927 			    netdev_alloc_skb_ip_align(netdev, length);
928 			if (new_skb) {
929 				skb_copy_to_linear_data_offset(new_skb,
930 							       -NET_IP_ALIGN,
931 							       (skb->data -
932 								NET_IP_ALIGN),
933 							       (length +
934 								NET_IP_ALIGN));
935 				/* save the skb in buffer_info as good */
936 				buffer_info->skb = skb;
937 				skb = new_skb;
938 			}
939 			/* else just continue with the old one */
940 		}
941 		/* end copybreak code */
942 		skb_put(skb, length);
943 
944 		/* Receive Checksum Offload */
945 		e1000_rx_checksum(adapter, staterr, skb);
946 
947 		e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
948 
949 		e1000_receive_skb(adapter, netdev, skb, staterr,
950 				  rx_desc->wb.upper.vlan);
951 
952 next_desc:
953 		rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
954 
955 		/* return some buffers to hardware, one at a time is too slow */
956 		if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
957 			adapter->alloc_rx_buf(rx_ring, cleaned_count,
958 					      GFP_ATOMIC);
959 			cleaned_count = 0;
960 		}
961 
962 		/* use prefetched values */
963 		rx_desc = next_rxd;
964 		buffer_info = next_buffer;
965 
966 		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
967 	}
968 	rx_ring->next_to_clean = i;
969 
970 	cleaned_count = e1000_desc_unused(rx_ring);
971 	if (cleaned_count)
972 		adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
973 
974 	adapter->total_rx_bytes += total_rx_bytes;
975 	adapter->total_rx_packets += total_rx_packets;
976 	return cleaned;
977 }
978 
979 static void e1000_put_txbuf(struct e1000_ring *tx_ring,
980 			    struct e1000_buffer *buffer_info)
981 {
982 	struct e1000_adapter *adapter = tx_ring->adapter;
983 
984 	if (buffer_info->dma) {
985 		if (buffer_info->mapped_as_page)
986 			dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
987 				       buffer_info->length, DMA_TO_DEVICE);
988 		else
989 			dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
990 					 buffer_info->length, DMA_TO_DEVICE);
991 		buffer_info->dma = 0;
992 	}
993 	if (buffer_info->skb) {
994 		dev_kfree_skb_any(buffer_info->skb);
995 		buffer_info->skb = NULL;
996 	}
997 	buffer_info->time_stamp = 0;
998 }
999 
1000 static void e1000_print_hw_hang(struct work_struct *work)
1001 {
1002 	struct e1000_adapter *adapter = container_of(work,
1003 	                                             struct e1000_adapter,
1004 	                                             print_hang_task);
1005 	struct net_device *netdev = adapter->netdev;
1006 	struct e1000_ring *tx_ring = adapter->tx_ring;
1007 	unsigned int i = tx_ring->next_to_clean;
1008 	unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1009 	struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1010 	struct e1000_hw *hw = &adapter->hw;
1011 	u16 phy_status, phy_1000t_status, phy_ext_status;
1012 	u16 pci_status;
1013 
1014 	if (test_bit(__E1000_DOWN, &adapter->state))
1015 		return;
1016 
1017 	if (!adapter->tx_hang_recheck &&
1018 	    (adapter->flags2 & FLAG2_DMA_BURST)) {
1019 		/*
1020 		 * May be block on write-back, flush and detect again
1021 		 * flush pending descriptor writebacks to memory
1022 		 */
1023 		ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1024 		/* execute the writes immediately */
1025 		e1e_flush();
1026 		/*
1027 		 * Due to rare timing issues, write to TIDV again to ensure
1028 		 * the write is successful
1029 		 */
1030 		ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1031 		/* execute the writes immediately */
1032 		e1e_flush();
1033 		adapter->tx_hang_recheck = true;
1034 		return;
1035 	}
1036 	/* Real hang detected */
1037 	adapter->tx_hang_recheck = false;
1038 	netif_stop_queue(netdev);
1039 
1040 	e1e_rphy(hw, PHY_STATUS, &phy_status);
1041 	e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
1042 	e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
1043 
1044 	pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1045 
1046 	/* detected Hardware unit hang */
1047 	e_err("Detected Hardware Unit Hang:\n"
1048 	      "  TDH                  <%x>\n"
1049 	      "  TDT                  <%x>\n"
1050 	      "  next_to_use          <%x>\n"
1051 	      "  next_to_clean        <%x>\n"
1052 	      "buffer_info[next_to_clean]:\n"
1053 	      "  time_stamp           <%lx>\n"
1054 	      "  next_to_watch        <%x>\n"
1055 	      "  jiffies              <%lx>\n"
1056 	      "  next_to_watch.status <%x>\n"
1057 	      "MAC Status             <%x>\n"
1058 	      "PHY Status             <%x>\n"
1059 	      "PHY 1000BASE-T Status  <%x>\n"
1060 	      "PHY Extended Status    <%x>\n"
1061 	      "PCI Status             <%x>\n",
1062 	      readl(tx_ring->head),
1063 	      readl(tx_ring->tail),
1064 	      tx_ring->next_to_use,
1065 	      tx_ring->next_to_clean,
1066 	      tx_ring->buffer_info[eop].time_stamp,
1067 	      eop,
1068 	      jiffies,
1069 	      eop_desc->upper.fields.status,
1070 	      er32(STATUS),
1071 	      phy_status,
1072 	      phy_1000t_status,
1073 	      phy_ext_status,
1074 	      pci_status);
1075 
1076 	/* Suggest workaround for known h/w issue */
1077 	if ((hw->mac.type == e1000_pchlan) && (er32(CTRL) & E1000_CTRL_TFCE))
1078 		e_err("Try turning off Tx pause (flow control) via ethtool\n");
1079 }
1080 
1081 /**
1082  * e1000_clean_tx_irq - Reclaim resources after transmit completes
1083  * @tx_ring: Tx descriptor ring
1084  *
1085  * the return value indicates whether actual cleaning was done, there
1086  * is no guarantee that everything was cleaned
1087  **/
1088 static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
1089 {
1090 	struct e1000_adapter *adapter = tx_ring->adapter;
1091 	struct net_device *netdev = adapter->netdev;
1092 	struct e1000_hw *hw = &adapter->hw;
1093 	struct e1000_tx_desc *tx_desc, *eop_desc;
1094 	struct e1000_buffer *buffer_info;
1095 	unsigned int i, eop;
1096 	unsigned int count = 0;
1097 	unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1098 	unsigned int bytes_compl = 0, pkts_compl = 0;
1099 
1100 	i = tx_ring->next_to_clean;
1101 	eop = tx_ring->buffer_info[i].next_to_watch;
1102 	eop_desc = E1000_TX_DESC(*tx_ring, eop);
1103 
1104 	while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1105 	       (count < tx_ring->count)) {
1106 		bool cleaned = false;
1107 		rmb(); /* read buffer_info after eop_desc */
1108 		for (; !cleaned; count++) {
1109 			tx_desc = E1000_TX_DESC(*tx_ring, i);
1110 			buffer_info = &tx_ring->buffer_info[i];
1111 			cleaned = (i == eop);
1112 
1113 			if (cleaned) {
1114 				total_tx_packets += buffer_info->segs;
1115 				total_tx_bytes += buffer_info->bytecount;
1116 				if (buffer_info->skb) {
1117 					bytes_compl += buffer_info->skb->len;
1118 					pkts_compl++;
1119 				}
1120 			}
1121 
1122 			e1000_put_txbuf(tx_ring, buffer_info);
1123 			tx_desc->upper.data = 0;
1124 
1125 			i++;
1126 			if (i == tx_ring->count)
1127 				i = 0;
1128 		}
1129 
1130 		if (i == tx_ring->next_to_use)
1131 			break;
1132 		eop = tx_ring->buffer_info[i].next_to_watch;
1133 		eop_desc = E1000_TX_DESC(*tx_ring, eop);
1134 	}
1135 
1136 	tx_ring->next_to_clean = i;
1137 
1138 	netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1139 
1140 #define TX_WAKE_THRESHOLD 32
1141 	if (count && netif_carrier_ok(netdev) &&
1142 	    e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1143 		/* Make sure that anybody stopping the queue after this
1144 		 * sees the new next_to_clean.
1145 		 */
1146 		smp_mb();
1147 
1148 		if (netif_queue_stopped(netdev) &&
1149 		    !(test_bit(__E1000_DOWN, &adapter->state))) {
1150 			netif_wake_queue(netdev);
1151 			++adapter->restart_queue;
1152 		}
1153 	}
1154 
1155 	if (adapter->detect_tx_hung) {
1156 		/*
1157 		 * Detect a transmit hang in hardware, this serializes the
1158 		 * check with the clearing of time_stamp and movement of i
1159 		 */
1160 		adapter->detect_tx_hung = false;
1161 		if (tx_ring->buffer_info[i].time_stamp &&
1162 		    time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1163 			       + (adapter->tx_timeout_factor * HZ)) &&
1164 		    !(er32(STATUS) & E1000_STATUS_TXOFF))
1165 			schedule_work(&adapter->print_hang_task);
1166 		else
1167 			adapter->tx_hang_recheck = false;
1168 	}
1169 	adapter->total_tx_bytes += total_tx_bytes;
1170 	adapter->total_tx_packets += total_tx_packets;
1171 	return count < tx_ring->count;
1172 }
1173 
1174 /**
1175  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1176  * @rx_ring: Rx descriptor ring
1177  *
1178  * the return value indicates whether actual cleaning was done, there
1179  * is no guarantee that everything was cleaned
1180  **/
1181 static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
1182 				  int work_to_do)
1183 {
1184 	struct e1000_adapter *adapter = rx_ring->adapter;
1185 	struct e1000_hw *hw = &adapter->hw;
1186 	union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1187 	struct net_device *netdev = adapter->netdev;
1188 	struct pci_dev *pdev = adapter->pdev;
1189 	struct e1000_buffer *buffer_info, *next_buffer;
1190 	struct e1000_ps_page *ps_page;
1191 	struct sk_buff *skb;
1192 	unsigned int i, j;
1193 	u32 length, staterr;
1194 	int cleaned_count = 0;
1195 	bool cleaned = false;
1196 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1197 
1198 	i = rx_ring->next_to_clean;
1199 	rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1200 	staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1201 	buffer_info = &rx_ring->buffer_info[i];
1202 
1203 	while (staterr & E1000_RXD_STAT_DD) {
1204 		if (*work_done >= work_to_do)
1205 			break;
1206 		(*work_done)++;
1207 		skb = buffer_info->skb;
1208 		rmb();	/* read descriptor and rx_buffer_info after status DD */
1209 
1210 		/* in the packet split case this is header only */
1211 		prefetch(skb->data - NET_IP_ALIGN);
1212 
1213 		i++;
1214 		if (i == rx_ring->count)
1215 			i = 0;
1216 		next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1217 		prefetch(next_rxd);
1218 
1219 		next_buffer = &rx_ring->buffer_info[i];
1220 
1221 		cleaned = true;
1222 		cleaned_count++;
1223 		dma_unmap_single(&pdev->dev, buffer_info->dma,
1224 				 adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1225 		buffer_info->dma = 0;
1226 
1227 		/* see !EOP comment in other Rx routine */
1228 		if (!(staterr & E1000_RXD_STAT_EOP))
1229 			adapter->flags2 |= FLAG2_IS_DISCARDING;
1230 
1231 		if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1232 			e_dbg("Packet Split buffers didn't pick up the full packet\n");
1233 			dev_kfree_skb_irq(skb);
1234 			if (staterr & E1000_RXD_STAT_EOP)
1235 				adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1236 			goto next_desc;
1237 		}
1238 
1239 		if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1240 			     !(netdev->features & NETIF_F_RXALL))) {
1241 			dev_kfree_skb_irq(skb);
1242 			goto next_desc;
1243 		}
1244 
1245 		length = le16_to_cpu(rx_desc->wb.middle.length0);
1246 
1247 		if (!length) {
1248 			e_dbg("Last part of the packet spanning multiple descriptors\n");
1249 			dev_kfree_skb_irq(skb);
1250 			goto next_desc;
1251 		}
1252 
1253 		/* Good Receive */
1254 		skb_put(skb, length);
1255 
1256 		{
1257 			/*
1258 			 * this looks ugly, but it seems compiler issues make
1259 			 * it more efficient than reusing j
1260 			 */
1261 			int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1262 
1263 			/*
1264 			 * page alloc/put takes too long and effects small
1265 			 * packet throughput, so unsplit small packets and
1266 			 * save the alloc/put only valid in softirq (napi)
1267 			 * context to call kmap_*
1268 			 */
1269 			if (l1 && (l1 <= copybreak) &&
1270 			    ((length + l1) <= adapter->rx_ps_bsize0)) {
1271 				u8 *vaddr;
1272 
1273 				ps_page = &buffer_info->ps_pages[0];
1274 
1275 				/*
1276 				 * there is no documentation about how to call
1277 				 * kmap_atomic, so we can't hold the mapping
1278 				 * very long
1279 				 */
1280 				dma_sync_single_for_cpu(&pdev->dev,
1281 							ps_page->dma,
1282 							PAGE_SIZE,
1283 							DMA_FROM_DEVICE);
1284 				vaddr = kmap_atomic(ps_page->page);
1285 				memcpy(skb_tail_pointer(skb), vaddr, l1);
1286 				kunmap_atomic(vaddr);
1287 				dma_sync_single_for_device(&pdev->dev,
1288 							   ps_page->dma,
1289 							   PAGE_SIZE,
1290 							   DMA_FROM_DEVICE);
1291 
1292 				/* remove the CRC */
1293 				if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1294 					if (!(netdev->features & NETIF_F_RXFCS))
1295 						l1 -= 4;
1296 				}
1297 
1298 				skb_put(skb, l1);
1299 				goto copydone;
1300 			} /* if */
1301 		}
1302 
1303 		for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1304 			length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1305 			if (!length)
1306 				break;
1307 
1308 			ps_page = &buffer_info->ps_pages[j];
1309 			dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1310 				       DMA_FROM_DEVICE);
1311 			ps_page->dma = 0;
1312 			skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1313 			ps_page->page = NULL;
1314 			skb->len += length;
1315 			skb->data_len += length;
1316 			skb->truesize += PAGE_SIZE;
1317 		}
1318 
1319 		/* strip the ethernet crc, problem is we're using pages now so
1320 		 * this whole operation can get a little cpu intensive
1321 		 */
1322 		if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1323 			if (!(netdev->features & NETIF_F_RXFCS))
1324 				pskb_trim(skb, skb->len - 4);
1325 		}
1326 
1327 copydone:
1328 		total_rx_bytes += skb->len;
1329 		total_rx_packets++;
1330 
1331 		e1000_rx_checksum(adapter, staterr, skb);
1332 
1333 		e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1334 
1335 		if (rx_desc->wb.upper.header_status &
1336 			   cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1337 			adapter->rx_hdr_split++;
1338 
1339 		e1000_receive_skb(adapter, netdev, skb,
1340 				  staterr, rx_desc->wb.middle.vlan);
1341 
1342 next_desc:
1343 		rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1344 		buffer_info->skb = NULL;
1345 
1346 		/* return some buffers to hardware, one at a time is too slow */
1347 		if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1348 			adapter->alloc_rx_buf(rx_ring, cleaned_count,
1349 					      GFP_ATOMIC);
1350 			cleaned_count = 0;
1351 		}
1352 
1353 		/* use prefetched values */
1354 		rx_desc = next_rxd;
1355 		buffer_info = next_buffer;
1356 
1357 		staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1358 	}
1359 	rx_ring->next_to_clean = i;
1360 
1361 	cleaned_count = e1000_desc_unused(rx_ring);
1362 	if (cleaned_count)
1363 		adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1364 
1365 	adapter->total_rx_bytes += total_rx_bytes;
1366 	adapter->total_rx_packets += total_rx_packets;
1367 	return cleaned;
1368 }
1369 
1370 /**
1371  * e1000_consume_page - helper function
1372  **/
1373 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1374                                u16 length)
1375 {
1376 	bi->page = NULL;
1377 	skb->len += length;
1378 	skb->data_len += length;
1379 	skb->truesize += PAGE_SIZE;
1380 }
1381 
1382 /**
1383  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1384  * @adapter: board private structure
1385  *
1386  * the return value indicates whether actual cleaning was done, there
1387  * is no guarantee that everything was cleaned
1388  **/
1389 static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1390 				     int work_to_do)
1391 {
1392 	struct e1000_adapter *adapter = rx_ring->adapter;
1393 	struct net_device *netdev = adapter->netdev;
1394 	struct pci_dev *pdev = adapter->pdev;
1395 	union e1000_rx_desc_extended *rx_desc, *next_rxd;
1396 	struct e1000_buffer *buffer_info, *next_buffer;
1397 	u32 length, staterr;
1398 	unsigned int i;
1399 	int cleaned_count = 0;
1400 	bool cleaned = false;
1401 	unsigned int total_rx_bytes=0, total_rx_packets=0;
1402 
1403 	i = rx_ring->next_to_clean;
1404 	rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1405 	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1406 	buffer_info = &rx_ring->buffer_info[i];
1407 
1408 	while (staterr & E1000_RXD_STAT_DD) {
1409 		struct sk_buff *skb;
1410 
1411 		if (*work_done >= work_to_do)
1412 			break;
1413 		(*work_done)++;
1414 		rmb();	/* read descriptor and rx_buffer_info after status DD */
1415 
1416 		skb = buffer_info->skb;
1417 		buffer_info->skb = NULL;
1418 
1419 		++i;
1420 		if (i == rx_ring->count)
1421 			i = 0;
1422 		next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1423 		prefetch(next_rxd);
1424 
1425 		next_buffer = &rx_ring->buffer_info[i];
1426 
1427 		cleaned = true;
1428 		cleaned_count++;
1429 		dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1430 			       DMA_FROM_DEVICE);
1431 		buffer_info->dma = 0;
1432 
1433 		length = le16_to_cpu(rx_desc->wb.upper.length);
1434 
1435 		/* errors is only valid for DD + EOP descriptors */
1436 		if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1437 			     ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1438 			      !(netdev->features & NETIF_F_RXALL)))) {
1439 			/* recycle both page and skb */
1440 			buffer_info->skb = skb;
1441 			/* an error means any chain goes out the window too */
1442 			if (rx_ring->rx_skb_top)
1443 				dev_kfree_skb_irq(rx_ring->rx_skb_top);
1444 			rx_ring->rx_skb_top = NULL;
1445 			goto next_desc;
1446 		}
1447 
1448 #define rxtop (rx_ring->rx_skb_top)
1449 		if (!(staterr & E1000_RXD_STAT_EOP)) {
1450 			/* this descriptor is only the beginning (or middle) */
1451 			if (!rxtop) {
1452 				/* this is the beginning of a chain */
1453 				rxtop = skb;
1454 				skb_fill_page_desc(rxtop, 0, buffer_info->page,
1455 				                   0, length);
1456 			} else {
1457 				/* this is the middle of a chain */
1458 				skb_fill_page_desc(rxtop,
1459 				    skb_shinfo(rxtop)->nr_frags,
1460 				    buffer_info->page, 0, length);
1461 				/* re-use the skb, only consumed the page */
1462 				buffer_info->skb = skb;
1463 			}
1464 			e1000_consume_page(buffer_info, rxtop, length);
1465 			goto next_desc;
1466 		} else {
1467 			if (rxtop) {
1468 				/* end of the chain */
1469 				skb_fill_page_desc(rxtop,
1470 				    skb_shinfo(rxtop)->nr_frags,
1471 				    buffer_info->page, 0, length);
1472 				/* re-use the current skb, we only consumed the
1473 				 * page */
1474 				buffer_info->skb = skb;
1475 				skb = rxtop;
1476 				rxtop = NULL;
1477 				e1000_consume_page(buffer_info, skb, length);
1478 			} else {
1479 				/* no chain, got EOP, this buf is the packet
1480 				 * copybreak to save the put_page/alloc_page */
1481 				if (length <= copybreak &&
1482 				    skb_tailroom(skb) >= length) {
1483 					u8 *vaddr;
1484 					vaddr = kmap_atomic(buffer_info->page);
1485 					memcpy(skb_tail_pointer(skb), vaddr,
1486 					       length);
1487 					kunmap_atomic(vaddr);
1488 					/* re-use the page, so don't erase
1489 					 * buffer_info->page */
1490 					skb_put(skb, length);
1491 				} else {
1492 					skb_fill_page_desc(skb, 0,
1493 					                   buffer_info->page, 0,
1494 				                           length);
1495 					e1000_consume_page(buffer_info, skb,
1496 					                   length);
1497 				}
1498 			}
1499 		}
1500 
1501 		/* Receive Checksum Offload */
1502 		e1000_rx_checksum(adapter, staterr, skb);
1503 
1504 		e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1505 
1506 		/* probably a little skewed due to removing CRC */
1507 		total_rx_bytes += skb->len;
1508 		total_rx_packets++;
1509 
1510 		/* eth type trans needs skb->data to point to something */
1511 		if (!pskb_may_pull(skb, ETH_HLEN)) {
1512 			e_err("pskb_may_pull failed.\n");
1513 			dev_kfree_skb_irq(skb);
1514 			goto next_desc;
1515 		}
1516 
1517 		e1000_receive_skb(adapter, netdev, skb, staterr,
1518 				  rx_desc->wb.upper.vlan);
1519 
1520 next_desc:
1521 		rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1522 
1523 		/* return some buffers to hardware, one at a time is too slow */
1524 		if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1525 			adapter->alloc_rx_buf(rx_ring, cleaned_count,
1526 					      GFP_ATOMIC);
1527 			cleaned_count = 0;
1528 		}
1529 
1530 		/* use prefetched values */
1531 		rx_desc = next_rxd;
1532 		buffer_info = next_buffer;
1533 
1534 		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1535 	}
1536 	rx_ring->next_to_clean = i;
1537 
1538 	cleaned_count = e1000_desc_unused(rx_ring);
1539 	if (cleaned_count)
1540 		adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1541 
1542 	adapter->total_rx_bytes += total_rx_bytes;
1543 	adapter->total_rx_packets += total_rx_packets;
1544 	return cleaned;
1545 }
1546 
1547 /**
1548  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1549  * @rx_ring: Rx descriptor ring
1550  **/
1551 static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1552 {
1553 	struct e1000_adapter *adapter = rx_ring->adapter;
1554 	struct e1000_buffer *buffer_info;
1555 	struct e1000_ps_page *ps_page;
1556 	struct pci_dev *pdev = adapter->pdev;
1557 	unsigned int i, j;
1558 
1559 	/* Free all the Rx ring sk_buffs */
1560 	for (i = 0; i < rx_ring->count; i++) {
1561 		buffer_info = &rx_ring->buffer_info[i];
1562 		if (buffer_info->dma) {
1563 			if (adapter->clean_rx == e1000_clean_rx_irq)
1564 				dma_unmap_single(&pdev->dev, buffer_info->dma,
1565 						 adapter->rx_buffer_len,
1566 						 DMA_FROM_DEVICE);
1567 			else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1568 				dma_unmap_page(&pdev->dev, buffer_info->dma,
1569 				               PAGE_SIZE,
1570 					       DMA_FROM_DEVICE);
1571 			else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1572 				dma_unmap_single(&pdev->dev, buffer_info->dma,
1573 						 adapter->rx_ps_bsize0,
1574 						 DMA_FROM_DEVICE);
1575 			buffer_info->dma = 0;
1576 		}
1577 
1578 		if (buffer_info->page) {
1579 			put_page(buffer_info->page);
1580 			buffer_info->page = NULL;
1581 		}
1582 
1583 		if (buffer_info->skb) {
1584 			dev_kfree_skb(buffer_info->skb);
1585 			buffer_info->skb = NULL;
1586 		}
1587 
1588 		for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1589 			ps_page = &buffer_info->ps_pages[j];
1590 			if (!ps_page->page)
1591 				break;
1592 			dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1593 				       DMA_FROM_DEVICE);
1594 			ps_page->dma = 0;
1595 			put_page(ps_page->page);
1596 			ps_page->page = NULL;
1597 		}
1598 	}
1599 
1600 	/* there also may be some cached data from a chained receive */
1601 	if (rx_ring->rx_skb_top) {
1602 		dev_kfree_skb(rx_ring->rx_skb_top);
1603 		rx_ring->rx_skb_top = NULL;
1604 	}
1605 
1606 	/* Zero out the descriptor ring */
1607 	memset(rx_ring->desc, 0, rx_ring->size);
1608 
1609 	rx_ring->next_to_clean = 0;
1610 	rx_ring->next_to_use = 0;
1611 	adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1612 
1613 	writel(0, rx_ring->head);
1614 	if (rx_ring->adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
1615 		e1000e_update_rdt_wa(rx_ring, 0);
1616 	else
1617 		writel(0, rx_ring->tail);
1618 }
1619 
1620 static void e1000e_downshift_workaround(struct work_struct *work)
1621 {
1622 	struct e1000_adapter *adapter = container_of(work,
1623 					struct e1000_adapter, downshift_task);
1624 
1625 	if (test_bit(__E1000_DOWN, &adapter->state))
1626 		return;
1627 
1628 	e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1629 }
1630 
1631 /**
1632  * e1000_intr_msi - Interrupt Handler
1633  * @irq: interrupt number
1634  * @data: pointer to a network interface device structure
1635  **/
1636 static irqreturn_t e1000_intr_msi(int irq, void *data)
1637 {
1638 	struct net_device *netdev = data;
1639 	struct e1000_adapter *adapter = netdev_priv(netdev);
1640 	struct e1000_hw *hw = &adapter->hw;
1641 	u32 icr = er32(ICR);
1642 
1643 	/*
1644 	 * read ICR disables interrupts using IAM
1645 	 */
1646 
1647 	if (icr & E1000_ICR_LSC) {
1648 		hw->mac.get_link_status = true;
1649 		/*
1650 		 * ICH8 workaround-- Call gig speed drop workaround on cable
1651 		 * disconnect (LSC) before accessing any PHY registers
1652 		 */
1653 		if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1654 		    (!(er32(STATUS) & E1000_STATUS_LU)))
1655 			schedule_work(&adapter->downshift_task);
1656 
1657 		/*
1658 		 * 80003ES2LAN workaround-- For packet buffer work-around on
1659 		 * link down event; disable receives here in the ISR and reset
1660 		 * adapter in watchdog
1661 		 */
1662 		if (netif_carrier_ok(netdev) &&
1663 		    adapter->flags & FLAG_RX_NEEDS_RESTART) {
1664 			/* disable receives */
1665 			u32 rctl = er32(RCTL);
1666 			ew32(RCTL, rctl & ~E1000_RCTL_EN);
1667 			adapter->flags |= FLAG_RX_RESTART_NOW;
1668 		}
1669 		/* guard against interrupt when we're going down */
1670 		if (!test_bit(__E1000_DOWN, &adapter->state))
1671 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
1672 	}
1673 
1674 	if (napi_schedule_prep(&adapter->napi)) {
1675 		adapter->total_tx_bytes = 0;
1676 		adapter->total_tx_packets = 0;
1677 		adapter->total_rx_bytes = 0;
1678 		adapter->total_rx_packets = 0;
1679 		__napi_schedule(&adapter->napi);
1680 	}
1681 
1682 	return IRQ_HANDLED;
1683 }
1684 
1685 /**
1686  * e1000_intr - Interrupt Handler
1687  * @irq: interrupt number
1688  * @data: pointer to a network interface device structure
1689  **/
1690 static irqreturn_t e1000_intr(int irq, void *data)
1691 {
1692 	struct net_device *netdev = data;
1693 	struct e1000_adapter *adapter = netdev_priv(netdev);
1694 	struct e1000_hw *hw = &adapter->hw;
1695 	u32 rctl, icr = er32(ICR);
1696 
1697 	if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1698 		return IRQ_NONE;  /* Not our interrupt */
1699 
1700 	/*
1701 	 * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1702 	 * not set, then the adapter didn't send an interrupt
1703 	 */
1704 	if (!(icr & E1000_ICR_INT_ASSERTED))
1705 		return IRQ_NONE;
1706 
1707 	/*
1708 	 * Interrupt Auto-Mask...upon reading ICR,
1709 	 * interrupts are masked.  No need for the
1710 	 * IMC write
1711 	 */
1712 
1713 	if (icr & E1000_ICR_LSC) {
1714 		hw->mac.get_link_status = true;
1715 		/*
1716 		 * ICH8 workaround-- Call gig speed drop workaround on cable
1717 		 * disconnect (LSC) before accessing any PHY registers
1718 		 */
1719 		if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1720 		    (!(er32(STATUS) & E1000_STATUS_LU)))
1721 			schedule_work(&adapter->downshift_task);
1722 
1723 		/*
1724 		 * 80003ES2LAN workaround--
1725 		 * For packet buffer work-around on link down event;
1726 		 * disable receives here in the ISR and
1727 		 * reset adapter in watchdog
1728 		 */
1729 		if (netif_carrier_ok(netdev) &&
1730 		    (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1731 			/* disable receives */
1732 			rctl = er32(RCTL);
1733 			ew32(RCTL, rctl & ~E1000_RCTL_EN);
1734 			adapter->flags |= FLAG_RX_RESTART_NOW;
1735 		}
1736 		/* guard against interrupt when we're going down */
1737 		if (!test_bit(__E1000_DOWN, &adapter->state))
1738 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
1739 	}
1740 
1741 	if (napi_schedule_prep(&adapter->napi)) {
1742 		adapter->total_tx_bytes = 0;
1743 		adapter->total_tx_packets = 0;
1744 		adapter->total_rx_bytes = 0;
1745 		adapter->total_rx_packets = 0;
1746 		__napi_schedule(&adapter->napi);
1747 	}
1748 
1749 	return IRQ_HANDLED;
1750 }
1751 
1752 static irqreturn_t e1000_msix_other(int irq, void *data)
1753 {
1754 	struct net_device *netdev = data;
1755 	struct e1000_adapter *adapter = netdev_priv(netdev);
1756 	struct e1000_hw *hw = &adapter->hw;
1757 	u32 icr = er32(ICR);
1758 
1759 	if (!(icr & E1000_ICR_INT_ASSERTED)) {
1760 		if (!test_bit(__E1000_DOWN, &adapter->state))
1761 			ew32(IMS, E1000_IMS_OTHER);
1762 		return IRQ_NONE;
1763 	}
1764 
1765 	if (icr & adapter->eiac_mask)
1766 		ew32(ICS, (icr & adapter->eiac_mask));
1767 
1768 	if (icr & E1000_ICR_OTHER) {
1769 		if (!(icr & E1000_ICR_LSC))
1770 			goto no_link_interrupt;
1771 		hw->mac.get_link_status = true;
1772 		/* guard against interrupt when we're going down */
1773 		if (!test_bit(__E1000_DOWN, &adapter->state))
1774 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
1775 	}
1776 
1777 no_link_interrupt:
1778 	if (!test_bit(__E1000_DOWN, &adapter->state))
1779 		ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1780 
1781 	return IRQ_HANDLED;
1782 }
1783 
1784 
1785 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1786 {
1787 	struct net_device *netdev = data;
1788 	struct e1000_adapter *adapter = netdev_priv(netdev);
1789 	struct e1000_hw *hw = &adapter->hw;
1790 	struct e1000_ring *tx_ring = adapter->tx_ring;
1791 
1792 
1793 	adapter->total_tx_bytes = 0;
1794 	adapter->total_tx_packets = 0;
1795 
1796 	if (!e1000_clean_tx_irq(tx_ring))
1797 		/* Ring was not completely cleaned, so fire another interrupt */
1798 		ew32(ICS, tx_ring->ims_val);
1799 
1800 	return IRQ_HANDLED;
1801 }
1802 
1803 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1804 {
1805 	struct net_device *netdev = data;
1806 	struct e1000_adapter *adapter = netdev_priv(netdev);
1807 	struct e1000_ring *rx_ring = adapter->rx_ring;
1808 
1809 	/* Write the ITR value calculated at the end of the
1810 	 * previous interrupt.
1811 	 */
1812 	if (rx_ring->set_itr) {
1813 		writel(1000000000 / (rx_ring->itr_val * 256),
1814 		       rx_ring->itr_register);
1815 		rx_ring->set_itr = 0;
1816 	}
1817 
1818 	if (napi_schedule_prep(&adapter->napi)) {
1819 		adapter->total_rx_bytes = 0;
1820 		adapter->total_rx_packets = 0;
1821 		__napi_schedule(&adapter->napi);
1822 	}
1823 	return IRQ_HANDLED;
1824 }
1825 
1826 /**
1827  * e1000_configure_msix - Configure MSI-X hardware
1828  *
1829  * e1000_configure_msix sets up the hardware to properly
1830  * generate MSI-X interrupts.
1831  **/
1832 static void e1000_configure_msix(struct e1000_adapter *adapter)
1833 {
1834 	struct e1000_hw *hw = &adapter->hw;
1835 	struct e1000_ring *rx_ring = adapter->rx_ring;
1836 	struct e1000_ring *tx_ring = adapter->tx_ring;
1837 	int vector = 0;
1838 	u32 ctrl_ext, ivar = 0;
1839 
1840 	adapter->eiac_mask = 0;
1841 
1842 	/* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1843 	if (hw->mac.type == e1000_82574) {
1844 		u32 rfctl = er32(RFCTL);
1845 		rfctl |= E1000_RFCTL_ACK_DIS;
1846 		ew32(RFCTL, rfctl);
1847 	}
1848 
1849 #define E1000_IVAR_INT_ALLOC_VALID	0x8
1850 	/* Configure Rx vector */
1851 	rx_ring->ims_val = E1000_IMS_RXQ0;
1852 	adapter->eiac_mask |= rx_ring->ims_val;
1853 	if (rx_ring->itr_val)
1854 		writel(1000000000 / (rx_ring->itr_val * 256),
1855 		       rx_ring->itr_register);
1856 	else
1857 		writel(1, rx_ring->itr_register);
1858 	ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1859 
1860 	/* Configure Tx vector */
1861 	tx_ring->ims_val = E1000_IMS_TXQ0;
1862 	vector++;
1863 	if (tx_ring->itr_val)
1864 		writel(1000000000 / (tx_ring->itr_val * 256),
1865 		       tx_ring->itr_register);
1866 	else
1867 		writel(1, tx_ring->itr_register);
1868 	adapter->eiac_mask |= tx_ring->ims_val;
1869 	ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1870 
1871 	/* set vector for Other Causes, e.g. link changes */
1872 	vector++;
1873 	ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1874 	if (rx_ring->itr_val)
1875 		writel(1000000000 / (rx_ring->itr_val * 256),
1876 		       hw->hw_addr + E1000_EITR_82574(vector));
1877 	else
1878 		writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1879 
1880 	/* Cause Tx interrupts on every write back */
1881 	ivar |= (1 << 31);
1882 
1883 	ew32(IVAR, ivar);
1884 
1885 	/* enable MSI-X PBA support */
1886 	ctrl_ext = er32(CTRL_EXT);
1887 	ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1888 
1889 	/* Auto-Mask Other interrupts upon ICR read */
1890 #define E1000_EIAC_MASK_82574   0x01F00000
1891 	ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1892 	ctrl_ext |= E1000_CTRL_EXT_EIAME;
1893 	ew32(CTRL_EXT, ctrl_ext);
1894 	e1e_flush();
1895 }
1896 
1897 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1898 {
1899 	if (adapter->msix_entries) {
1900 		pci_disable_msix(adapter->pdev);
1901 		kfree(adapter->msix_entries);
1902 		adapter->msix_entries = NULL;
1903 	} else if (adapter->flags & FLAG_MSI_ENABLED) {
1904 		pci_disable_msi(adapter->pdev);
1905 		adapter->flags &= ~FLAG_MSI_ENABLED;
1906 	}
1907 }
1908 
1909 /**
1910  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1911  *
1912  * Attempt to configure interrupts using the best available
1913  * capabilities of the hardware and kernel.
1914  **/
1915 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1916 {
1917 	int err;
1918 	int i;
1919 
1920 	switch (adapter->int_mode) {
1921 	case E1000E_INT_MODE_MSIX:
1922 		if (adapter->flags & FLAG_HAS_MSIX) {
1923 			adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1924 			adapter->msix_entries = kcalloc(adapter->num_vectors,
1925 						      sizeof(struct msix_entry),
1926 						      GFP_KERNEL);
1927 			if (adapter->msix_entries) {
1928 				for (i = 0; i < adapter->num_vectors; i++)
1929 					adapter->msix_entries[i].entry = i;
1930 
1931 				err = pci_enable_msix(adapter->pdev,
1932 						      adapter->msix_entries,
1933 						      adapter->num_vectors);
1934 				if (err == 0)
1935 					return;
1936 			}
1937 			/* MSI-X failed, so fall through and try MSI */
1938 			e_err("Failed to initialize MSI-X interrupts.  Falling back to MSI interrupts.\n");
1939 			e1000e_reset_interrupt_capability(adapter);
1940 		}
1941 		adapter->int_mode = E1000E_INT_MODE_MSI;
1942 		/* Fall through */
1943 	case E1000E_INT_MODE_MSI:
1944 		if (!pci_enable_msi(adapter->pdev)) {
1945 			adapter->flags |= FLAG_MSI_ENABLED;
1946 		} else {
1947 			adapter->int_mode = E1000E_INT_MODE_LEGACY;
1948 			e_err("Failed to initialize MSI interrupts.  Falling back to legacy interrupts.\n");
1949 		}
1950 		/* Fall through */
1951 	case E1000E_INT_MODE_LEGACY:
1952 		/* Don't do anything; this is the system default */
1953 		break;
1954 	}
1955 
1956 	/* store the number of vectors being used */
1957 	adapter->num_vectors = 1;
1958 }
1959 
1960 /**
1961  * e1000_request_msix - Initialize MSI-X interrupts
1962  *
1963  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1964  * kernel.
1965  **/
1966 static int e1000_request_msix(struct e1000_adapter *adapter)
1967 {
1968 	struct net_device *netdev = adapter->netdev;
1969 	int err = 0, vector = 0;
1970 
1971 	if (strlen(netdev->name) < (IFNAMSIZ - 5))
1972 		snprintf(adapter->rx_ring->name,
1973 			 sizeof(adapter->rx_ring->name) - 1,
1974 			 "%s-rx-0", netdev->name);
1975 	else
1976 		memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1977 	err = request_irq(adapter->msix_entries[vector].vector,
1978 			  e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1979 			  netdev);
1980 	if (err)
1981 		return err;
1982 	adapter->rx_ring->itr_register = adapter->hw.hw_addr +
1983 	    E1000_EITR_82574(vector);
1984 	adapter->rx_ring->itr_val = adapter->itr;
1985 	vector++;
1986 
1987 	if (strlen(netdev->name) < (IFNAMSIZ - 5))
1988 		snprintf(adapter->tx_ring->name,
1989 			 sizeof(adapter->tx_ring->name) - 1,
1990 			 "%s-tx-0", netdev->name);
1991 	else
1992 		memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1993 	err = request_irq(adapter->msix_entries[vector].vector,
1994 			  e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1995 			  netdev);
1996 	if (err)
1997 		return err;
1998 	adapter->tx_ring->itr_register = adapter->hw.hw_addr +
1999 	    E1000_EITR_82574(vector);
2000 	adapter->tx_ring->itr_val = adapter->itr;
2001 	vector++;
2002 
2003 	err = request_irq(adapter->msix_entries[vector].vector,
2004 			  e1000_msix_other, 0, netdev->name, netdev);
2005 	if (err)
2006 		return err;
2007 
2008 	e1000_configure_msix(adapter);
2009 
2010 	return 0;
2011 }
2012 
2013 /**
2014  * e1000_request_irq - initialize interrupts
2015  *
2016  * Attempts to configure interrupts using the best available
2017  * capabilities of the hardware and kernel.
2018  **/
2019 static int e1000_request_irq(struct e1000_adapter *adapter)
2020 {
2021 	struct net_device *netdev = adapter->netdev;
2022 	int err;
2023 
2024 	if (adapter->msix_entries) {
2025 		err = e1000_request_msix(adapter);
2026 		if (!err)
2027 			return err;
2028 		/* fall back to MSI */
2029 		e1000e_reset_interrupt_capability(adapter);
2030 		adapter->int_mode = E1000E_INT_MODE_MSI;
2031 		e1000e_set_interrupt_capability(adapter);
2032 	}
2033 	if (adapter->flags & FLAG_MSI_ENABLED) {
2034 		err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2035 				  netdev->name, netdev);
2036 		if (!err)
2037 			return err;
2038 
2039 		/* fall back to legacy interrupt */
2040 		e1000e_reset_interrupt_capability(adapter);
2041 		adapter->int_mode = E1000E_INT_MODE_LEGACY;
2042 	}
2043 
2044 	err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2045 			  netdev->name, netdev);
2046 	if (err)
2047 		e_err("Unable to allocate interrupt, Error: %d\n", err);
2048 
2049 	return err;
2050 }
2051 
2052 static void e1000_free_irq(struct e1000_adapter *adapter)
2053 {
2054 	struct net_device *netdev = adapter->netdev;
2055 
2056 	if (adapter->msix_entries) {
2057 		int vector = 0;
2058 
2059 		free_irq(adapter->msix_entries[vector].vector, netdev);
2060 		vector++;
2061 
2062 		free_irq(adapter->msix_entries[vector].vector, netdev);
2063 		vector++;
2064 
2065 		/* Other Causes interrupt vector */
2066 		free_irq(adapter->msix_entries[vector].vector, netdev);
2067 		return;
2068 	}
2069 
2070 	free_irq(adapter->pdev->irq, netdev);
2071 }
2072 
2073 /**
2074  * e1000_irq_disable - Mask off interrupt generation on the NIC
2075  **/
2076 static void e1000_irq_disable(struct e1000_adapter *adapter)
2077 {
2078 	struct e1000_hw *hw = &adapter->hw;
2079 
2080 	ew32(IMC, ~0);
2081 	if (adapter->msix_entries)
2082 		ew32(EIAC_82574, 0);
2083 	e1e_flush();
2084 
2085 	if (adapter->msix_entries) {
2086 		int i;
2087 		for (i = 0; i < adapter->num_vectors; i++)
2088 			synchronize_irq(adapter->msix_entries[i].vector);
2089 	} else {
2090 		synchronize_irq(adapter->pdev->irq);
2091 	}
2092 }
2093 
2094 /**
2095  * e1000_irq_enable - Enable default interrupt generation settings
2096  **/
2097 static void e1000_irq_enable(struct e1000_adapter *adapter)
2098 {
2099 	struct e1000_hw *hw = &adapter->hw;
2100 
2101 	if (adapter->msix_entries) {
2102 		ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2103 		ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
2104 	} else {
2105 		ew32(IMS, IMS_ENABLE_MASK);
2106 	}
2107 	e1e_flush();
2108 }
2109 
2110 /**
2111  * e1000e_get_hw_control - get control of the h/w from f/w
2112  * @adapter: address of board private structure
2113  *
2114  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2115  * For ASF and Pass Through versions of f/w this means that
2116  * the driver is loaded. For AMT version (only with 82573)
2117  * of the f/w this means that the network i/f is open.
2118  **/
2119 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2120 {
2121 	struct e1000_hw *hw = &adapter->hw;
2122 	u32 ctrl_ext;
2123 	u32 swsm;
2124 
2125 	/* Let firmware know the driver has taken over */
2126 	if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2127 		swsm = er32(SWSM);
2128 		ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2129 	} else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2130 		ctrl_ext = er32(CTRL_EXT);
2131 		ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2132 	}
2133 }
2134 
2135 /**
2136  * e1000e_release_hw_control - release control of the h/w to f/w
2137  * @adapter: address of board private structure
2138  *
2139  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2140  * For ASF and Pass Through versions of f/w this means that the
2141  * driver is no longer loaded. For AMT version (only with 82573) i
2142  * of the f/w this means that the network i/f is closed.
2143  *
2144  **/
2145 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2146 {
2147 	struct e1000_hw *hw = &adapter->hw;
2148 	u32 ctrl_ext;
2149 	u32 swsm;
2150 
2151 	/* Let firmware taken over control of h/w */
2152 	if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2153 		swsm = er32(SWSM);
2154 		ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2155 	} else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2156 		ctrl_ext = er32(CTRL_EXT);
2157 		ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2158 	}
2159 }
2160 
2161 /**
2162  * @e1000_alloc_ring - allocate memory for a ring structure
2163  **/
2164 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2165 				struct e1000_ring *ring)
2166 {
2167 	struct pci_dev *pdev = adapter->pdev;
2168 
2169 	ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2170 					GFP_KERNEL);
2171 	if (!ring->desc)
2172 		return -ENOMEM;
2173 
2174 	return 0;
2175 }
2176 
2177 /**
2178  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2179  * @tx_ring: Tx descriptor ring
2180  *
2181  * Return 0 on success, negative on failure
2182  **/
2183 int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2184 {
2185 	struct e1000_adapter *adapter = tx_ring->adapter;
2186 	int err = -ENOMEM, size;
2187 
2188 	size = sizeof(struct e1000_buffer) * tx_ring->count;
2189 	tx_ring->buffer_info = vzalloc(size);
2190 	if (!tx_ring->buffer_info)
2191 		goto err;
2192 
2193 	/* round up to nearest 4K */
2194 	tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2195 	tx_ring->size = ALIGN(tx_ring->size, 4096);
2196 
2197 	err = e1000_alloc_ring_dma(adapter, tx_ring);
2198 	if (err)
2199 		goto err;
2200 
2201 	tx_ring->next_to_use = 0;
2202 	tx_ring->next_to_clean = 0;
2203 
2204 	return 0;
2205 err:
2206 	vfree(tx_ring->buffer_info);
2207 	e_err("Unable to allocate memory for the transmit descriptor ring\n");
2208 	return err;
2209 }
2210 
2211 /**
2212  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2213  * @rx_ring: Rx descriptor ring
2214  *
2215  * Returns 0 on success, negative on failure
2216  **/
2217 int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2218 {
2219 	struct e1000_adapter *adapter = rx_ring->adapter;
2220 	struct e1000_buffer *buffer_info;
2221 	int i, size, desc_len, err = -ENOMEM;
2222 
2223 	size = sizeof(struct e1000_buffer) * rx_ring->count;
2224 	rx_ring->buffer_info = vzalloc(size);
2225 	if (!rx_ring->buffer_info)
2226 		goto err;
2227 
2228 	for (i = 0; i < rx_ring->count; i++) {
2229 		buffer_info = &rx_ring->buffer_info[i];
2230 		buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2231 						sizeof(struct e1000_ps_page),
2232 						GFP_KERNEL);
2233 		if (!buffer_info->ps_pages)
2234 			goto err_pages;
2235 	}
2236 
2237 	desc_len = sizeof(union e1000_rx_desc_packet_split);
2238 
2239 	/* Round up to nearest 4K */
2240 	rx_ring->size = rx_ring->count * desc_len;
2241 	rx_ring->size = ALIGN(rx_ring->size, 4096);
2242 
2243 	err = e1000_alloc_ring_dma(adapter, rx_ring);
2244 	if (err)
2245 		goto err_pages;
2246 
2247 	rx_ring->next_to_clean = 0;
2248 	rx_ring->next_to_use = 0;
2249 	rx_ring->rx_skb_top = NULL;
2250 
2251 	return 0;
2252 
2253 err_pages:
2254 	for (i = 0; i < rx_ring->count; i++) {
2255 		buffer_info = &rx_ring->buffer_info[i];
2256 		kfree(buffer_info->ps_pages);
2257 	}
2258 err:
2259 	vfree(rx_ring->buffer_info);
2260 	e_err("Unable to allocate memory for the receive descriptor ring\n");
2261 	return err;
2262 }
2263 
2264 /**
2265  * e1000_clean_tx_ring - Free Tx Buffers
2266  * @tx_ring: Tx descriptor ring
2267  **/
2268 static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2269 {
2270 	struct e1000_adapter *adapter = tx_ring->adapter;
2271 	struct e1000_buffer *buffer_info;
2272 	unsigned long size;
2273 	unsigned int i;
2274 
2275 	for (i = 0; i < tx_ring->count; i++) {
2276 		buffer_info = &tx_ring->buffer_info[i];
2277 		e1000_put_txbuf(tx_ring, buffer_info);
2278 	}
2279 
2280 	netdev_reset_queue(adapter->netdev);
2281 	size = sizeof(struct e1000_buffer) * tx_ring->count;
2282 	memset(tx_ring->buffer_info, 0, size);
2283 
2284 	memset(tx_ring->desc, 0, tx_ring->size);
2285 
2286 	tx_ring->next_to_use = 0;
2287 	tx_ring->next_to_clean = 0;
2288 
2289 	writel(0, tx_ring->head);
2290 	if (tx_ring->adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
2291 		e1000e_update_tdt_wa(tx_ring, 0);
2292 	else
2293 		writel(0, tx_ring->tail);
2294 }
2295 
2296 /**
2297  * e1000e_free_tx_resources - Free Tx Resources per Queue
2298  * @tx_ring: Tx descriptor ring
2299  *
2300  * Free all transmit software resources
2301  **/
2302 void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2303 {
2304 	struct e1000_adapter *adapter = tx_ring->adapter;
2305 	struct pci_dev *pdev = adapter->pdev;
2306 
2307 	e1000_clean_tx_ring(tx_ring);
2308 
2309 	vfree(tx_ring->buffer_info);
2310 	tx_ring->buffer_info = NULL;
2311 
2312 	dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2313 			  tx_ring->dma);
2314 	tx_ring->desc = NULL;
2315 }
2316 
2317 /**
2318  * e1000e_free_rx_resources - Free Rx Resources
2319  * @rx_ring: Rx descriptor ring
2320  *
2321  * Free all receive software resources
2322  **/
2323 void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2324 {
2325 	struct e1000_adapter *adapter = rx_ring->adapter;
2326 	struct pci_dev *pdev = adapter->pdev;
2327 	int i;
2328 
2329 	e1000_clean_rx_ring(rx_ring);
2330 
2331 	for (i = 0; i < rx_ring->count; i++)
2332 		kfree(rx_ring->buffer_info[i].ps_pages);
2333 
2334 	vfree(rx_ring->buffer_info);
2335 	rx_ring->buffer_info = NULL;
2336 
2337 	dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2338 			  rx_ring->dma);
2339 	rx_ring->desc = NULL;
2340 }
2341 
2342 /**
2343  * e1000_update_itr - update the dynamic ITR value based on statistics
2344  * @adapter: pointer to adapter
2345  * @itr_setting: current adapter->itr
2346  * @packets: the number of packets during this measurement interval
2347  * @bytes: the number of bytes during this measurement interval
2348  *
2349  *      Stores a new ITR value based on packets and byte
2350  *      counts during the last interrupt.  The advantage of per interrupt
2351  *      computation is faster updates and more accurate ITR for the current
2352  *      traffic pattern.  Constants in this function were computed
2353  *      based on theoretical maximum wire speed and thresholds were set based
2354  *      on testing data as well as attempting to minimize response time
2355  *      while increasing bulk throughput.  This functionality is controlled
2356  *      by the InterruptThrottleRate module parameter.
2357  **/
2358 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2359 				     u16 itr_setting, int packets,
2360 				     int bytes)
2361 {
2362 	unsigned int retval = itr_setting;
2363 
2364 	if (packets == 0)
2365 		return itr_setting;
2366 
2367 	switch (itr_setting) {
2368 	case lowest_latency:
2369 		/* handle TSO and jumbo frames */
2370 		if (bytes/packets > 8000)
2371 			retval = bulk_latency;
2372 		else if ((packets < 5) && (bytes > 512))
2373 			retval = low_latency;
2374 		break;
2375 	case low_latency:  /* 50 usec aka 20000 ints/s */
2376 		if (bytes > 10000) {
2377 			/* this if handles the TSO accounting */
2378 			if (bytes/packets > 8000)
2379 				retval = bulk_latency;
2380 			else if ((packets < 10) || ((bytes/packets) > 1200))
2381 				retval = bulk_latency;
2382 			else if ((packets > 35))
2383 				retval = lowest_latency;
2384 		} else if (bytes/packets > 2000) {
2385 			retval = bulk_latency;
2386 		} else if (packets <= 2 && bytes < 512) {
2387 			retval = lowest_latency;
2388 		}
2389 		break;
2390 	case bulk_latency: /* 250 usec aka 4000 ints/s */
2391 		if (bytes > 25000) {
2392 			if (packets > 35)
2393 				retval = low_latency;
2394 		} else if (bytes < 6000) {
2395 			retval = low_latency;
2396 		}
2397 		break;
2398 	}
2399 
2400 	return retval;
2401 }
2402 
2403 static void e1000_set_itr(struct e1000_adapter *adapter)
2404 {
2405 	struct e1000_hw *hw = &adapter->hw;
2406 	u16 current_itr;
2407 	u32 new_itr = adapter->itr;
2408 
2409 	/* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2410 	if (adapter->link_speed != SPEED_1000) {
2411 		current_itr = 0;
2412 		new_itr = 4000;
2413 		goto set_itr_now;
2414 	}
2415 
2416 	if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2417 		new_itr = 0;
2418 		goto set_itr_now;
2419 	}
2420 
2421 	adapter->tx_itr = e1000_update_itr(adapter,
2422 				    adapter->tx_itr,
2423 				    adapter->total_tx_packets,
2424 				    adapter->total_tx_bytes);
2425 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
2426 	if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2427 		adapter->tx_itr = low_latency;
2428 
2429 	adapter->rx_itr = e1000_update_itr(adapter,
2430 				    adapter->rx_itr,
2431 				    adapter->total_rx_packets,
2432 				    adapter->total_rx_bytes);
2433 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
2434 	if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2435 		adapter->rx_itr = low_latency;
2436 
2437 	current_itr = max(adapter->rx_itr, adapter->tx_itr);
2438 
2439 	switch (current_itr) {
2440 	/* counts and packets in update_itr are dependent on these numbers */
2441 	case lowest_latency:
2442 		new_itr = 70000;
2443 		break;
2444 	case low_latency:
2445 		new_itr = 20000; /* aka hwitr = ~200 */
2446 		break;
2447 	case bulk_latency:
2448 		new_itr = 4000;
2449 		break;
2450 	default:
2451 		break;
2452 	}
2453 
2454 set_itr_now:
2455 	if (new_itr != adapter->itr) {
2456 		/*
2457 		 * this attempts to bias the interrupt rate towards Bulk
2458 		 * by adding intermediate steps when interrupt rate is
2459 		 * increasing
2460 		 */
2461 		new_itr = new_itr > adapter->itr ?
2462 			     min(adapter->itr + (new_itr >> 2), new_itr) :
2463 			     new_itr;
2464 		adapter->itr = new_itr;
2465 		adapter->rx_ring->itr_val = new_itr;
2466 		if (adapter->msix_entries)
2467 			adapter->rx_ring->set_itr = 1;
2468 		else
2469 			if (new_itr)
2470 				ew32(ITR, 1000000000 / (new_itr * 256));
2471 			else
2472 				ew32(ITR, 0);
2473 	}
2474 }
2475 
2476 /**
2477  * e1000_alloc_queues - Allocate memory for all rings
2478  * @adapter: board private structure to initialize
2479  **/
2480 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2481 {
2482 	int size = sizeof(struct e1000_ring);
2483 
2484 	adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2485 	if (!adapter->tx_ring)
2486 		goto err;
2487 	adapter->tx_ring->count = adapter->tx_ring_count;
2488 	adapter->tx_ring->adapter = adapter;
2489 
2490 	adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2491 	if (!adapter->rx_ring)
2492 		goto err;
2493 	adapter->rx_ring->count = adapter->rx_ring_count;
2494 	adapter->rx_ring->adapter = adapter;
2495 
2496 	return 0;
2497 err:
2498 	e_err("Unable to allocate memory for queues\n");
2499 	kfree(adapter->rx_ring);
2500 	kfree(adapter->tx_ring);
2501 	return -ENOMEM;
2502 }
2503 
2504 /**
2505  * e1000e_poll - NAPI Rx polling callback
2506  * @napi: struct associated with this polling callback
2507  * @weight: number of packets driver is allowed to process this poll
2508  **/
2509 static int e1000e_poll(struct napi_struct *napi, int weight)
2510 {
2511 	struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
2512 						     napi);
2513 	struct e1000_hw *hw = &adapter->hw;
2514 	struct net_device *poll_dev = adapter->netdev;
2515 	int tx_cleaned = 1, work_done = 0;
2516 
2517 	adapter = netdev_priv(poll_dev);
2518 
2519 	if (!adapter->msix_entries ||
2520 	    (adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2521 		tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2522 
2523 	adapter->clean_rx(adapter->rx_ring, &work_done, weight);
2524 
2525 	if (!tx_cleaned)
2526 		work_done = weight;
2527 
2528 	/* If weight not fully consumed, exit the polling mode */
2529 	if (work_done < weight) {
2530 		if (adapter->itr_setting & 3)
2531 			e1000_set_itr(adapter);
2532 		napi_complete(napi);
2533 		if (!test_bit(__E1000_DOWN, &adapter->state)) {
2534 			if (adapter->msix_entries)
2535 				ew32(IMS, adapter->rx_ring->ims_val);
2536 			else
2537 				e1000_irq_enable(adapter);
2538 		}
2539 	}
2540 
2541 	return work_done;
2542 }
2543 
2544 static int e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2545 {
2546 	struct e1000_adapter *adapter = netdev_priv(netdev);
2547 	struct e1000_hw *hw = &adapter->hw;
2548 	u32 vfta, index;
2549 
2550 	/* don't update vlan cookie if already programmed */
2551 	if ((adapter->hw.mng_cookie.status &
2552 	     E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2553 	    (vid == adapter->mng_vlan_id))
2554 		return 0;
2555 
2556 	/* add VID to filter table */
2557 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2558 		index = (vid >> 5) & 0x7F;
2559 		vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2560 		vfta |= (1 << (vid & 0x1F));
2561 		hw->mac.ops.write_vfta(hw, index, vfta);
2562 	}
2563 
2564 	set_bit(vid, adapter->active_vlans);
2565 
2566 	return 0;
2567 }
2568 
2569 static int e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2570 {
2571 	struct e1000_adapter *adapter = netdev_priv(netdev);
2572 	struct e1000_hw *hw = &adapter->hw;
2573 	u32 vfta, index;
2574 
2575 	if ((adapter->hw.mng_cookie.status &
2576 	     E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2577 	    (vid == adapter->mng_vlan_id)) {
2578 		/* release control to f/w */
2579 		e1000e_release_hw_control(adapter);
2580 		return 0;
2581 	}
2582 
2583 	/* remove VID from filter table */
2584 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2585 		index = (vid >> 5) & 0x7F;
2586 		vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2587 		vfta &= ~(1 << (vid & 0x1F));
2588 		hw->mac.ops.write_vfta(hw, index, vfta);
2589 	}
2590 
2591 	clear_bit(vid, adapter->active_vlans);
2592 
2593 	return 0;
2594 }
2595 
2596 /**
2597  * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2598  * @adapter: board private structure to initialize
2599  **/
2600 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2601 {
2602 	struct net_device *netdev = adapter->netdev;
2603 	struct e1000_hw *hw = &adapter->hw;
2604 	u32 rctl;
2605 
2606 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2607 		/* disable VLAN receive filtering */
2608 		rctl = er32(RCTL);
2609 		rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2610 		ew32(RCTL, rctl);
2611 
2612 		if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2613 			e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2614 			adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2615 		}
2616 	}
2617 }
2618 
2619 /**
2620  * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2621  * @adapter: board private structure to initialize
2622  **/
2623 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2624 {
2625 	struct e1000_hw *hw = &adapter->hw;
2626 	u32 rctl;
2627 
2628 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2629 		/* enable VLAN receive filtering */
2630 		rctl = er32(RCTL);
2631 		rctl |= E1000_RCTL_VFE;
2632 		rctl &= ~E1000_RCTL_CFIEN;
2633 		ew32(RCTL, rctl);
2634 	}
2635 }
2636 
2637 /**
2638  * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2639  * @adapter: board private structure to initialize
2640  **/
2641 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2642 {
2643 	struct e1000_hw *hw = &adapter->hw;
2644 	u32 ctrl;
2645 
2646 	/* disable VLAN tag insert/strip */
2647 	ctrl = er32(CTRL);
2648 	ctrl &= ~E1000_CTRL_VME;
2649 	ew32(CTRL, ctrl);
2650 }
2651 
2652 /**
2653  * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2654  * @adapter: board private structure to initialize
2655  **/
2656 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2657 {
2658 	struct e1000_hw *hw = &adapter->hw;
2659 	u32 ctrl;
2660 
2661 	/* enable VLAN tag insert/strip */
2662 	ctrl = er32(CTRL);
2663 	ctrl |= E1000_CTRL_VME;
2664 	ew32(CTRL, ctrl);
2665 }
2666 
2667 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2668 {
2669 	struct net_device *netdev = adapter->netdev;
2670 	u16 vid = adapter->hw.mng_cookie.vlan_id;
2671 	u16 old_vid = adapter->mng_vlan_id;
2672 
2673 	if (adapter->hw.mng_cookie.status &
2674 	    E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2675 		e1000_vlan_rx_add_vid(netdev, vid);
2676 		adapter->mng_vlan_id = vid;
2677 	}
2678 
2679 	if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2680 		e1000_vlan_rx_kill_vid(netdev, old_vid);
2681 }
2682 
2683 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2684 {
2685 	u16 vid;
2686 
2687 	e1000_vlan_rx_add_vid(adapter->netdev, 0);
2688 
2689 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2690 		e1000_vlan_rx_add_vid(adapter->netdev, vid);
2691 }
2692 
2693 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2694 {
2695 	struct e1000_hw *hw = &adapter->hw;
2696 	u32 manc, manc2h, mdef, i, j;
2697 
2698 	if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2699 		return;
2700 
2701 	manc = er32(MANC);
2702 
2703 	/*
2704 	 * enable receiving management packets to the host. this will probably
2705 	 * generate destination unreachable messages from the host OS, but
2706 	 * the packets will be handled on SMBUS
2707 	 */
2708 	manc |= E1000_MANC_EN_MNG2HOST;
2709 	manc2h = er32(MANC2H);
2710 
2711 	switch (hw->mac.type) {
2712 	default:
2713 		manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2714 		break;
2715 	case e1000_82574:
2716 	case e1000_82583:
2717 		/*
2718 		 * Check if IPMI pass-through decision filter already exists;
2719 		 * if so, enable it.
2720 		 */
2721 		for (i = 0, j = 0; i < 8; i++) {
2722 			mdef = er32(MDEF(i));
2723 
2724 			/* Ignore filters with anything other than IPMI ports */
2725 			if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2726 				continue;
2727 
2728 			/* Enable this decision filter in MANC2H */
2729 			if (mdef)
2730 				manc2h |= (1 << i);
2731 
2732 			j |= mdef;
2733 		}
2734 
2735 		if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2736 			break;
2737 
2738 		/* Create new decision filter in an empty filter */
2739 		for (i = 0, j = 0; i < 8; i++)
2740 			if (er32(MDEF(i)) == 0) {
2741 				ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2742 					       E1000_MDEF_PORT_664));
2743 				manc2h |= (1 << 1);
2744 				j++;
2745 				break;
2746 			}
2747 
2748 		if (!j)
2749 			e_warn("Unable to create IPMI pass-through filter\n");
2750 		break;
2751 	}
2752 
2753 	ew32(MANC2H, manc2h);
2754 	ew32(MANC, manc);
2755 }
2756 
2757 /**
2758  * e1000_configure_tx - Configure Transmit Unit after Reset
2759  * @adapter: board private structure
2760  *
2761  * Configure the Tx unit of the MAC after a reset.
2762  **/
2763 static void e1000_configure_tx(struct e1000_adapter *adapter)
2764 {
2765 	struct e1000_hw *hw = &adapter->hw;
2766 	struct e1000_ring *tx_ring = adapter->tx_ring;
2767 	u64 tdba;
2768 	u32 tdlen, tarc;
2769 
2770 	/* Setup the HW Tx Head and Tail descriptor pointers */
2771 	tdba = tx_ring->dma;
2772 	tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2773 	ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
2774 	ew32(TDBAH(0), (tdba >> 32));
2775 	ew32(TDLEN(0), tdlen);
2776 	ew32(TDH(0), 0);
2777 	ew32(TDT(0), 0);
2778 	tx_ring->head = adapter->hw.hw_addr + E1000_TDH(0);
2779 	tx_ring->tail = adapter->hw.hw_addr + E1000_TDT(0);
2780 
2781 	/* Set the Tx Interrupt Delay register */
2782 	ew32(TIDV, adapter->tx_int_delay);
2783 	/* Tx irq moderation */
2784 	ew32(TADV, adapter->tx_abs_int_delay);
2785 
2786 	if (adapter->flags2 & FLAG2_DMA_BURST) {
2787 		u32 txdctl = er32(TXDCTL(0));
2788 		txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2789 			    E1000_TXDCTL_WTHRESH);
2790 		/*
2791 		 * set up some performance related parameters to encourage the
2792 		 * hardware to use the bus more efficiently in bursts, depends
2793 		 * on the tx_int_delay to be enabled,
2794 		 * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2795 		 * hthresh = 1 ==> prefetch when one or more available
2796 		 * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2797 		 * BEWARE: this seems to work but should be considered first if
2798 		 * there are Tx hangs or other Tx related bugs
2799 		 */
2800 		txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2801 		ew32(TXDCTL(0), txdctl);
2802 	}
2803 	/* erratum work around: set txdctl the same for both queues */
2804 	ew32(TXDCTL(1), er32(TXDCTL(0)));
2805 
2806 	if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2807 		tarc = er32(TARC(0));
2808 		/*
2809 		 * set the speed mode bit, we'll clear it if we're not at
2810 		 * gigabit link later
2811 		 */
2812 #define SPEED_MODE_BIT (1 << 21)
2813 		tarc |= SPEED_MODE_BIT;
2814 		ew32(TARC(0), tarc);
2815 	}
2816 
2817 	/* errata: program both queues to unweighted RR */
2818 	if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2819 		tarc = er32(TARC(0));
2820 		tarc |= 1;
2821 		ew32(TARC(0), tarc);
2822 		tarc = er32(TARC(1));
2823 		tarc |= 1;
2824 		ew32(TARC(1), tarc);
2825 	}
2826 
2827 	/* Setup Transmit Descriptor Settings for eop descriptor */
2828 	adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2829 
2830 	/* only set IDE if we are delaying interrupts using the timers */
2831 	if (adapter->tx_int_delay)
2832 		adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2833 
2834 	/* enable Report Status bit */
2835 	adapter->txd_cmd |= E1000_TXD_CMD_RS;
2836 
2837 	hw->mac.ops.config_collision_dist(hw);
2838 }
2839 
2840 /**
2841  * e1000_setup_rctl - configure the receive control registers
2842  * @adapter: Board private structure
2843  **/
2844 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2845 			   (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2846 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2847 {
2848 	struct e1000_hw *hw = &adapter->hw;
2849 	u32 rctl, rfctl;
2850 	u32 pages = 0;
2851 
2852 	/* Workaround Si errata on PCHx - configure jumbo frame flow */
2853 	if (hw->mac.type >= e1000_pch2lan) {
2854 		s32 ret_val;
2855 
2856 		if (adapter->netdev->mtu > ETH_DATA_LEN)
2857 			ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2858 		else
2859 			ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2860 
2861 		if (ret_val)
2862 			e_dbg("failed to enable jumbo frame workaround mode\n");
2863 	}
2864 
2865 	/* Program MC offset vector base */
2866 	rctl = er32(RCTL);
2867 	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2868 	rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2869 		E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2870 		(adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2871 
2872 	/* Do not Store bad packets */
2873 	rctl &= ~E1000_RCTL_SBP;
2874 
2875 	/* Enable Long Packet receive */
2876 	if (adapter->netdev->mtu <= ETH_DATA_LEN)
2877 		rctl &= ~E1000_RCTL_LPE;
2878 	else
2879 		rctl |= E1000_RCTL_LPE;
2880 
2881 	/* Some systems expect that the CRC is included in SMBUS traffic. The
2882 	 * hardware strips the CRC before sending to both SMBUS (BMC) and to
2883 	 * host memory when this is enabled
2884 	 */
2885 	if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2886 		rctl |= E1000_RCTL_SECRC;
2887 
2888 	/* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2889 	if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2890 		u16 phy_data;
2891 
2892 		e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2893 		phy_data &= 0xfff8;
2894 		phy_data |= (1 << 2);
2895 		e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2896 
2897 		e1e_rphy(hw, 22, &phy_data);
2898 		phy_data &= 0x0fff;
2899 		phy_data |= (1 << 14);
2900 		e1e_wphy(hw, 0x10, 0x2823);
2901 		e1e_wphy(hw, 0x11, 0x0003);
2902 		e1e_wphy(hw, 22, phy_data);
2903 	}
2904 
2905 	/* Setup buffer sizes */
2906 	rctl &= ~E1000_RCTL_SZ_4096;
2907 	rctl |= E1000_RCTL_BSEX;
2908 	switch (adapter->rx_buffer_len) {
2909 	case 2048:
2910 	default:
2911 		rctl |= E1000_RCTL_SZ_2048;
2912 		rctl &= ~E1000_RCTL_BSEX;
2913 		break;
2914 	case 4096:
2915 		rctl |= E1000_RCTL_SZ_4096;
2916 		break;
2917 	case 8192:
2918 		rctl |= E1000_RCTL_SZ_8192;
2919 		break;
2920 	case 16384:
2921 		rctl |= E1000_RCTL_SZ_16384;
2922 		break;
2923 	}
2924 
2925 	/* Enable Extended Status in all Receive Descriptors */
2926 	rfctl = er32(RFCTL);
2927 	rfctl |= E1000_RFCTL_EXTEN;
2928 	ew32(RFCTL, rfctl);
2929 
2930 	/*
2931 	 * 82571 and greater support packet-split where the protocol
2932 	 * header is placed in skb->data and the packet data is
2933 	 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2934 	 * In the case of a non-split, skb->data is linearly filled,
2935 	 * followed by the page buffers.  Therefore, skb->data is
2936 	 * sized to hold the largest protocol header.
2937 	 *
2938 	 * allocations using alloc_page take too long for regular MTU
2939 	 * so only enable packet split for jumbo frames
2940 	 *
2941 	 * Using pages when the page size is greater than 16k wastes
2942 	 * a lot of memory, since we allocate 3 pages at all times
2943 	 * per packet.
2944 	 */
2945 	pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2946 	if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2947 		adapter->rx_ps_pages = pages;
2948 	else
2949 		adapter->rx_ps_pages = 0;
2950 
2951 	if (adapter->rx_ps_pages) {
2952 		u32 psrctl = 0;
2953 
2954 		/* Enable Packet split descriptors */
2955 		rctl |= E1000_RCTL_DTYP_PS;
2956 
2957 		psrctl |= adapter->rx_ps_bsize0 >>
2958 			E1000_PSRCTL_BSIZE0_SHIFT;
2959 
2960 		switch (adapter->rx_ps_pages) {
2961 		case 3:
2962 			psrctl |= PAGE_SIZE <<
2963 				E1000_PSRCTL_BSIZE3_SHIFT;
2964 		case 2:
2965 			psrctl |= PAGE_SIZE <<
2966 				E1000_PSRCTL_BSIZE2_SHIFT;
2967 		case 1:
2968 			psrctl |= PAGE_SIZE >>
2969 				E1000_PSRCTL_BSIZE1_SHIFT;
2970 			break;
2971 		}
2972 
2973 		ew32(PSRCTL, psrctl);
2974 	}
2975 
2976 	/* This is useful for sniffing bad packets. */
2977 	if (adapter->netdev->features & NETIF_F_RXALL) {
2978 		/* UPE and MPE will be handled by normal PROMISC logic
2979 		 * in e1000e_set_rx_mode */
2980 		rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
2981 			 E1000_RCTL_BAM | /* RX All Bcast Pkts */
2982 			 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
2983 
2984 		rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
2985 			  E1000_RCTL_DPF | /* Allow filtered pause */
2986 			  E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
2987 		/* Do not mess with E1000_CTRL_VME, it affects transmit as well,
2988 		 * and that breaks VLANs.
2989 		 */
2990 	}
2991 
2992 	ew32(RCTL, rctl);
2993 	/* just started the receive unit, no need to restart */
2994 	adapter->flags &= ~FLAG_RX_RESTART_NOW;
2995 }
2996 
2997 /**
2998  * e1000_configure_rx - Configure Receive Unit after Reset
2999  * @adapter: board private structure
3000  *
3001  * Configure the Rx unit of the MAC after a reset.
3002  **/
3003 static void e1000_configure_rx(struct e1000_adapter *adapter)
3004 {
3005 	struct e1000_hw *hw = &adapter->hw;
3006 	struct e1000_ring *rx_ring = adapter->rx_ring;
3007 	u64 rdba;
3008 	u32 rdlen, rctl, rxcsum, ctrl_ext;
3009 
3010 	if (adapter->rx_ps_pages) {
3011 		/* this is a 32 byte descriptor */
3012 		rdlen = rx_ring->count *
3013 		    sizeof(union e1000_rx_desc_packet_split);
3014 		adapter->clean_rx = e1000_clean_rx_irq_ps;
3015 		adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3016 	} else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3017 		rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3018 		adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3019 		adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3020 	} else {
3021 		rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3022 		adapter->clean_rx = e1000_clean_rx_irq;
3023 		adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3024 	}
3025 
3026 	/* disable receives while setting up the descriptors */
3027 	rctl = er32(RCTL);
3028 	if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3029 		ew32(RCTL, rctl & ~E1000_RCTL_EN);
3030 	e1e_flush();
3031 	usleep_range(10000, 20000);
3032 
3033 	if (adapter->flags2 & FLAG2_DMA_BURST) {
3034 		/*
3035 		 * set the writeback threshold (only takes effect if the RDTR
3036 		 * is set). set GRAN=1 and write back up to 0x4 worth, and
3037 		 * enable prefetching of 0x20 Rx descriptors
3038 		 * granularity = 01
3039 		 * wthresh = 04,
3040 		 * hthresh = 04,
3041 		 * pthresh = 0x20
3042 		 */
3043 		ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3044 		ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3045 
3046 		/*
3047 		 * override the delay timers for enabling bursting, only if
3048 		 * the value was not set by the user via module options
3049 		 */
3050 		if (adapter->rx_int_delay == DEFAULT_RDTR)
3051 			adapter->rx_int_delay = BURST_RDTR;
3052 		if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3053 			adapter->rx_abs_int_delay = BURST_RADV;
3054 	}
3055 
3056 	/* set the Receive Delay Timer Register */
3057 	ew32(RDTR, adapter->rx_int_delay);
3058 
3059 	/* irq moderation */
3060 	ew32(RADV, adapter->rx_abs_int_delay);
3061 	if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3062 		ew32(ITR, 1000000000 / (adapter->itr * 256));
3063 
3064 	ctrl_ext = er32(CTRL_EXT);
3065 	/* Auto-Mask interrupts upon ICR access */
3066 	ctrl_ext |= E1000_CTRL_EXT_IAME;
3067 	ew32(IAM, 0xffffffff);
3068 	ew32(CTRL_EXT, ctrl_ext);
3069 	e1e_flush();
3070 
3071 	/*
3072 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
3073 	 * the Base and Length of the Rx Descriptor Ring
3074 	 */
3075 	rdba = rx_ring->dma;
3076 	ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
3077 	ew32(RDBAH(0), (rdba >> 32));
3078 	ew32(RDLEN(0), rdlen);
3079 	ew32(RDH(0), 0);
3080 	ew32(RDT(0), 0);
3081 	rx_ring->head = adapter->hw.hw_addr + E1000_RDH(0);
3082 	rx_ring->tail = adapter->hw.hw_addr + E1000_RDT(0);
3083 
3084 	/* Enable Receive Checksum Offload for TCP and UDP */
3085 	rxcsum = er32(RXCSUM);
3086 	if (adapter->netdev->features & NETIF_F_RXCSUM)
3087 		rxcsum |= E1000_RXCSUM_TUOFL;
3088 	else
3089 		rxcsum &= ~E1000_RXCSUM_TUOFL;
3090 	ew32(RXCSUM, rxcsum);
3091 
3092 	if (adapter->hw.mac.type == e1000_pch2lan) {
3093 		/*
3094 		 * With jumbo frames, excessive C-state transition
3095 		 * latencies result in dropped transactions.
3096 		 */
3097 		if (adapter->netdev->mtu > ETH_DATA_LEN) {
3098 			u32 rxdctl = er32(RXDCTL(0));
3099 			ew32(RXDCTL(0), rxdctl | 0x3);
3100 			pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
3101 		} else {
3102 			pm_qos_update_request(&adapter->netdev->pm_qos_req,
3103 					      PM_QOS_DEFAULT_VALUE);
3104 		}
3105 	}
3106 
3107 	/* Enable Receives */
3108 	ew32(RCTL, rctl);
3109 }
3110 
3111 /**
3112  * e1000e_write_mc_addr_list - write multicast addresses to MTA
3113  * @netdev: network interface device structure
3114  *
3115  * Writes multicast address list to the MTA hash table.
3116  * Returns: -ENOMEM on failure
3117  *                0 on no addresses written
3118  *                X on writing X addresses to MTA
3119  */
3120 static int e1000e_write_mc_addr_list(struct net_device *netdev)
3121 {
3122 	struct e1000_adapter *adapter = netdev_priv(netdev);
3123 	struct e1000_hw *hw = &adapter->hw;
3124 	struct netdev_hw_addr *ha;
3125 	u8 *mta_list;
3126 	int i;
3127 
3128 	if (netdev_mc_empty(netdev)) {
3129 		/* nothing to program, so clear mc list */
3130 		hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3131 		return 0;
3132 	}
3133 
3134 	mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3135 	if (!mta_list)
3136 		return -ENOMEM;
3137 
3138 	/* update_mc_addr_list expects a packed array of only addresses. */
3139 	i = 0;
3140 	netdev_for_each_mc_addr(ha, netdev)
3141 		memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3142 
3143 	hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3144 	kfree(mta_list);
3145 
3146 	return netdev_mc_count(netdev);
3147 }
3148 
3149 /**
3150  * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3151  * @netdev: network interface device structure
3152  *
3153  * Writes unicast address list to the RAR table.
3154  * Returns: -ENOMEM on failure/insufficient address space
3155  *                0 on no addresses written
3156  *                X on writing X addresses to the RAR table
3157  **/
3158 static int e1000e_write_uc_addr_list(struct net_device *netdev)
3159 {
3160 	struct e1000_adapter *adapter = netdev_priv(netdev);
3161 	struct e1000_hw *hw = &adapter->hw;
3162 	unsigned int rar_entries = hw->mac.rar_entry_count;
3163 	int count = 0;
3164 
3165 	/* save a rar entry for our hardware address */
3166 	rar_entries--;
3167 
3168 	/* save a rar entry for the LAA workaround */
3169 	if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3170 		rar_entries--;
3171 
3172 	/* return ENOMEM indicating insufficient memory for addresses */
3173 	if (netdev_uc_count(netdev) > rar_entries)
3174 		return -ENOMEM;
3175 
3176 	if (!netdev_uc_empty(netdev) && rar_entries) {
3177 		struct netdev_hw_addr *ha;
3178 
3179 		/*
3180 		 * write the addresses in reverse order to avoid write
3181 		 * combining
3182 		 */
3183 		netdev_for_each_uc_addr(ha, netdev) {
3184 			if (!rar_entries)
3185 				break;
3186 			hw->mac.ops.rar_set(hw, ha->addr, rar_entries--);
3187 			count++;
3188 		}
3189 	}
3190 
3191 	/* zero out the remaining RAR entries not used above */
3192 	for (; rar_entries > 0; rar_entries--) {
3193 		ew32(RAH(rar_entries), 0);
3194 		ew32(RAL(rar_entries), 0);
3195 	}
3196 	e1e_flush();
3197 
3198 	return count;
3199 }
3200 
3201 /**
3202  * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3203  * @netdev: network interface device structure
3204  *
3205  * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3206  * address list or the network interface flags are updated.  This routine is
3207  * responsible for configuring the hardware for proper unicast, multicast,
3208  * promiscuous mode, and all-multi behavior.
3209  **/
3210 static void e1000e_set_rx_mode(struct net_device *netdev)
3211 {
3212 	struct e1000_adapter *adapter = netdev_priv(netdev);
3213 	struct e1000_hw *hw = &adapter->hw;
3214 	u32 rctl;
3215 
3216 	/* Check for Promiscuous and All Multicast modes */
3217 	rctl = er32(RCTL);
3218 
3219 	/* clear the affected bits */
3220 	rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3221 
3222 	if (netdev->flags & IFF_PROMISC) {
3223 		rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3224 		/* Do not hardware filter VLANs in promisc mode */
3225 		e1000e_vlan_filter_disable(adapter);
3226 	} else {
3227 		int count;
3228 
3229 		if (netdev->flags & IFF_ALLMULTI) {
3230 			rctl |= E1000_RCTL_MPE;
3231 		} else {
3232 			/*
3233 			 * Write addresses to the MTA, if the attempt fails
3234 			 * then we should just turn on promiscuous mode so
3235 			 * that we can at least receive multicast traffic
3236 			 */
3237 			count = e1000e_write_mc_addr_list(netdev);
3238 			if (count < 0)
3239 				rctl |= E1000_RCTL_MPE;
3240 		}
3241 		e1000e_vlan_filter_enable(adapter);
3242 		/*
3243 		 * Write addresses to available RAR registers, if there is not
3244 		 * sufficient space to store all the addresses then enable
3245 		 * unicast promiscuous mode
3246 		 */
3247 		count = e1000e_write_uc_addr_list(netdev);
3248 		if (count < 0)
3249 			rctl |= E1000_RCTL_UPE;
3250 	}
3251 
3252 	ew32(RCTL, rctl);
3253 
3254 	if (netdev->features & NETIF_F_HW_VLAN_RX)
3255 		e1000e_vlan_strip_enable(adapter);
3256 	else
3257 		e1000e_vlan_strip_disable(adapter);
3258 }
3259 
3260 static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3261 {
3262 	struct e1000_hw *hw = &adapter->hw;
3263 	u32 mrqc, rxcsum;
3264 	int i;
3265 	static const u32 rsskey[10] = {
3266 		0xda565a6d, 0xc20e5b25, 0x3d256741, 0xb08fa343, 0xcb2bcad0,
3267 		0xb4307bae, 0xa32dcb77, 0x0cf23080, 0x3bb7426a, 0xfa01acbe
3268 	};
3269 
3270 	/* Fill out hash function seed */
3271 	for (i = 0; i < 10; i++)
3272 		ew32(RSSRK(i), rsskey[i]);
3273 
3274 	/* Direct all traffic to queue 0 */
3275 	for (i = 0; i < 32; i++)
3276 		ew32(RETA(i), 0);
3277 
3278 	/*
3279 	 * Disable raw packet checksumming so that RSS hash is placed in
3280 	 * descriptor on writeback.
3281 	 */
3282 	rxcsum = er32(RXCSUM);
3283 	rxcsum |= E1000_RXCSUM_PCSD;
3284 
3285 	ew32(RXCSUM, rxcsum);
3286 
3287 	mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3288 		E1000_MRQC_RSS_FIELD_IPV4_TCP |
3289 		E1000_MRQC_RSS_FIELD_IPV6 |
3290 		E1000_MRQC_RSS_FIELD_IPV6_TCP |
3291 		E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3292 
3293 	ew32(MRQC, mrqc);
3294 }
3295 
3296 /**
3297  * e1000_configure - configure the hardware for Rx and Tx
3298  * @adapter: private board structure
3299  **/
3300 static void e1000_configure(struct e1000_adapter *adapter)
3301 {
3302 	struct e1000_ring *rx_ring = adapter->rx_ring;
3303 
3304 	e1000e_set_rx_mode(adapter->netdev);
3305 
3306 	e1000_restore_vlan(adapter);
3307 	e1000_init_manageability_pt(adapter);
3308 
3309 	e1000_configure_tx(adapter);
3310 
3311 	if (adapter->netdev->features & NETIF_F_RXHASH)
3312 		e1000e_setup_rss_hash(adapter);
3313 	e1000_setup_rctl(adapter);
3314 	e1000_configure_rx(adapter);
3315 	adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3316 }
3317 
3318 /**
3319  * e1000e_power_up_phy - restore link in case the phy was powered down
3320  * @adapter: address of board private structure
3321  *
3322  * The phy may be powered down to save power and turn off link when the
3323  * driver is unloaded and wake on lan is not enabled (among others)
3324  * *** this routine MUST be followed by a call to e1000e_reset ***
3325  **/
3326 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3327 {
3328 	if (adapter->hw.phy.ops.power_up)
3329 		adapter->hw.phy.ops.power_up(&adapter->hw);
3330 
3331 	adapter->hw.mac.ops.setup_link(&adapter->hw);
3332 }
3333 
3334 /**
3335  * e1000_power_down_phy - Power down the PHY
3336  *
3337  * Power down the PHY so no link is implied when interface is down.
3338  * The PHY cannot be powered down if management or WoL is active.
3339  */
3340 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3341 {
3342 	/* WoL is enabled */
3343 	if (adapter->wol)
3344 		return;
3345 
3346 	if (adapter->hw.phy.ops.power_down)
3347 		adapter->hw.phy.ops.power_down(&adapter->hw);
3348 }
3349 
3350 /**
3351  * e1000e_reset - bring the hardware into a known good state
3352  *
3353  * This function boots the hardware and enables some settings that
3354  * require a configuration cycle of the hardware - those cannot be
3355  * set/changed during runtime. After reset the device needs to be
3356  * properly configured for Rx, Tx etc.
3357  */
3358 void e1000e_reset(struct e1000_adapter *adapter)
3359 {
3360 	struct e1000_mac_info *mac = &adapter->hw.mac;
3361 	struct e1000_fc_info *fc = &adapter->hw.fc;
3362 	struct e1000_hw *hw = &adapter->hw;
3363 	u32 tx_space, min_tx_space, min_rx_space;
3364 	u32 pba = adapter->pba;
3365 	u16 hwm;
3366 
3367 	/* reset Packet Buffer Allocation to default */
3368 	ew32(PBA, pba);
3369 
3370 	if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3371 		/*
3372 		 * To maintain wire speed transmits, the Tx FIFO should be
3373 		 * large enough to accommodate two full transmit packets,
3374 		 * rounded up to the next 1KB and expressed in KB.  Likewise,
3375 		 * the Rx FIFO should be large enough to accommodate at least
3376 		 * one full receive packet and is similarly rounded up and
3377 		 * expressed in KB.
3378 		 */
3379 		pba = er32(PBA);
3380 		/* upper 16 bits has Tx packet buffer allocation size in KB */
3381 		tx_space = pba >> 16;
3382 		/* lower 16 bits has Rx packet buffer allocation size in KB */
3383 		pba &= 0xffff;
3384 		/*
3385 		 * the Tx fifo also stores 16 bytes of information about the Tx
3386 		 * but don't include ethernet FCS because hardware appends it
3387 		 */
3388 		min_tx_space = (adapter->max_frame_size +
3389 				sizeof(struct e1000_tx_desc) -
3390 				ETH_FCS_LEN) * 2;
3391 		min_tx_space = ALIGN(min_tx_space, 1024);
3392 		min_tx_space >>= 10;
3393 		/* software strips receive CRC, so leave room for it */
3394 		min_rx_space = adapter->max_frame_size;
3395 		min_rx_space = ALIGN(min_rx_space, 1024);
3396 		min_rx_space >>= 10;
3397 
3398 		/*
3399 		 * If current Tx allocation is less than the min Tx FIFO size,
3400 		 * and the min Tx FIFO size is less than the current Rx FIFO
3401 		 * allocation, take space away from current Rx allocation
3402 		 */
3403 		if ((tx_space < min_tx_space) &&
3404 		    ((min_tx_space - tx_space) < pba)) {
3405 			pba -= min_tx_space - tx_space;
3406 
3407 			/*
3408 			 * if short on Rx space, Rx wins and must trump Tx
3409 			 * adjustment or use Early Receive if available
3410 			 */
3411 			if (pba < min_rx_space)
3412 				pba = min_rx_space;
3413 		}
3414 
3415 		ew32(PBA, pba);
3416 	}
3417 
3418 	/*
3419 	 * flow control settings
3420 	 *
3421 	 * The high water mark must be low enough to fit one full frame
3422 	 * (or the size used for early receive) above it in the Rx FIFO.
3423 	 * Set it to the lower of:
3424 	 * - 90% of the Rx FIFO size, and
3425 	 * - the full Rx FIFO size minus one full frame
3426 	 */
3427 	if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3428 		fc->pause_time = 0xFFFF;
3429 	else
3430 		fc->pause_time = E1000_FC_PAUSE_TIME;
3431 	fc->send_xon = true;
3432 	fc->current_mode = fc->requested_mode;
3433 
3434 	switch (hw->mac.type) {
3435 	case e1000_ich9lan:
3436 	case e1000_ich10lan:
3437 		if (adapter->netdev->mtu > ETH_DATA_LEN) {
3438 			pba = 14;
3439 			ew32(PBA, pba);
3440 			fc->high_water = 0x2800;
3441 			fc->low_water = fc->high_water - 8;
3442 			break;
3443 		}
3444 		/* fall-through */
3445 	default:
3446 		hwm = min(((pba << 10) * 9 / 10),
3447 			  ((pba << 10) - adapter->max_frame_size));
3448 
3449 		fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3450 		fc->low_water = fc->high_water - 8;
3451 		break;
3452 	case e1000_pchlan:
3453 		/*
3454 		 * Workaround PCH LOM adapter hangs with certain network
3455 		 * loads.  If hangs persist, try disabling Tx flow control.
3456 		 */
3457 		if (adapter->netdev->mtu > ETH_DATA_LEN) {
3458 			fc->high_water = 0x3500;
3459 			fc->low_water  = 0x1500;
3460 		} else {
3461 			fc->high_water = 0x5000;
3462 			fc->low_water  = 0x3000;
3463 		}
3464 		fc->refresh_time = 0x1000;
3465 		break;
3466 	case e1000_pch2lan:
3467 	case e1000_pch_lpt:
3468 		fc->high_water = 0x05C20;
3469 		fc->low_water = 0x05048;
3470 		fc->pause_time = 0x0650;
3471 		fc->refresh_time = 0x0400;
3472 		if (adapter->netdev->mtu > ETH_DATA_LEN) {
3473 			pba = 14;
3474 			ew32(PBA, pba);
3475 		}
3476 		break;
3477 	}
3478 
3479 	/*
3480 	 * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3481 	 * fit in receive buffer.
3482 	 */
3483 	if (adapter->itr_setting & 0x3) {
3484 		if ((adapter->max_frame_size * 2) > (pba << 10)) {
3485 			if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3486 				dev_info(&adapter->pdev->dev,
3487 					"Interrupt Throttle Rate turned off\n");
3488 				adapter->flags2 |= FLAG2_DISABLE_AIM;
3489 				ew32(ITR, 0);
3490 			}
3491 		} else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3492 			dev_info(&adapter->pdev->dev,
3493 				 "Interrupt Throttle Rate turned on\n");
3494 			adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3495 			adapter->itr = 20000;
3496 			ew32(ITR, 1000000000 / (adapter->itr * 256));
3497 		}
3498 	}
3499 
3500 	/* Allow time for pending master requests to run */
3501 	mac->ops.reset_hw(hw);
3502 
3503 	/*
3504 	 * For parts with AMT enabled, let the firmware know
3505 	 * that the network interface is in control
3506 	 */
3507 	if (adapter->flags & FLAG_HAS_AMT)
3508 		e1000e_get_hw_control(adapter);
3509 
3510 	ew32(WUC, 0);
3511 
3512 	if (mac->ops.init_hw(hw))
3513 		e_err("Hardware Error\n");
3514 
3515 	e1000_update_mng_vlan(adapter);
3516 
3517 	/* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3518 	ew32(VET, ETH_P_8021Q);
3519 
3520 	e1000e_reset_adaptive(hw);
3521 
3522 	if (!netif_running(adapter->netdev) &&
3523 	    !test_bit(__E1000_TESTING, &adapter->state)) {
3524 		e1000_power_down_phy(adapter);
3525 		return;
3526 	}
3527 
3528 	e1000_get_phy_info(hw);
3529 
3530 	if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3531 	    !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3532 		u16 phy_data = 0;
3533 		/*
3534 		 * speed up time to link by disabling smart power down, ignore
3535 		 * the return value of this function because there is nothing
3536 		 * different we would do if it failed
3537 		 */
3538 		e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3539 		phy_data &= ~IGP02E1000_PM_SPD;
3540 		e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3541 	}
3542 }
3543 
3544 int e1000e_up(struct e1000_adapter *adapter)
3545 {
3546 	struct e1000_hw *hw = &adapter->hw;
3547 
3548 	/* hardware has been reset, we need to reload some things */
3549 	e1000_configure(adapter);
3550 
3551 	clear_bit(__E1000_DOWN, &adapter->state);
3552 
3553 	if (adapter->msix_entries)
3554 		e1000_configure_msix(adapter);
3555 	e1000_irq_enable(adapter);
3556 
3557 	netif_start_queue(adapter->netdev);
3558 
3559 	/* fire a link change interrupt to start the watchdog */
3560 	if (adapter->msix_entries)
3561 		ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3562 	else
3563 		ew32(ICS, E1000_ICS_LSC);
3564 
3565 	return 0;
3566 }
3567 
3568 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3569 {
3570 	struct e1000_hw *hw = &adapter->hw;
3571 
3572 	if (!(adapter->flags2 & FLAG2_DMA_BURST))
3573 		return;
3574 
3575 	/* flush pending descriptor writebacks to memory */
3576 	ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3577 	ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3578 
3579 	/* execute the writes immediately */
3580 	e1e_flush();
3581 
3582 	/*
3583 	 * due to rare timing issues, write to TIDV/RDTR again to ensure the
3584 	 * write is successful
3585 	 */
3586 	ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3587 	ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3588 
3589 	/* execute the writes immediately */
3590 	e1e_flush();
3591 }
3592 
3593 static void e1000e_update_stats(struct e1000_adapter *adapter);
3594 
3595 void e1000e_down(struct e1000_adapter *adapter)
3596 {
3597 	struct net_device *netdev = adapter->netdev;
3598 	struct e1000_hw *hw = &adapter->hw;
3599 	u32 tctl, rctl;
3600 
3601 	/*
3602 	 * signal that we're down so the interrupt handler does not
3603 	 * reschedule our watchdog timer
3604 	 */
3605 	set_bit(__E1000_DOWN, &adapter->state);
3606 
3607 	/* disable receives in the hardware */
3608 	rctl = er32(RCTL);
3609 	if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3610 		ew32(RCTL, rctl & ~E1000_RCTL_EN);
3611 	/* flush and sleep below */
3612 
3613 	netif_stop_queue(netdev);
3614 
3615 	/* disable transmits in the hardware */
3616 	tctl = er32(TCTL);
3617 	tctl &= ~E1000_TCTL_EN;
3618 	ew32(TCTL, tctl);
3619 
3620 	/* flush both disables and wait for them to finish */
3621 	e1e_flush();
3622 	usleep_range(10000, 20000);
3623 
3624 	e1000_irq_disable(adapter);
3625 
3626 	del_timer_sync(&adapter->watchdog_timer);
3627 	del_timer_sync(&adapter->phy_info_timer);
3628 
3629 	netif_carrier_off(netdev);
3630 
3631 	spin_lock(&adapter->stats64_lock);
3632 	e1000e_update_stats(adapter);
3633 	spin_unlock(&adapter->stats64_lock);
3634 
3635 	e1000e_flush_descriptors(adapter);
3636 	e1000_clean_tx_ring(adapter->tx_ring);
3637 	e1000_clean_rx_ring(adapter->rx_ring);
3638 
3639 	adapter->link_speed = 0;
3640 	adapter->link_duplex = 0;
3641 
3642 	if (!pci_channel_offline(adapter->pdev))
3643 		e1000e_reset(adapter);
3644 
3645 	/*
3646 	 * TODO: for power management, we could drop the link and
3647 	 * pci_disable_device here.
3648 	 */
3649 }
3650 
3651 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3652 {
3653 	might_sleep();
3654 	while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3655 		usleep_range(1000, 2000);
3656 	e1000e_down(adapter);
3657 	e1000e_up(adapter);
3658 	clear_bit(__E1000_RESETTING, &adapter->state);
3659 }
3660 
3661 /**
3662  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3663  * @adapter: board private structure to initialize
3664  *
3665  * e1000_sw_init initializes the Adapter private data structure.
3666  * Fields are initialized based on PCI device information and
3667  * OS network device settings (MTU size).
3668  **/
3669 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3670 {
3671 	struct net_device *netdev = adapter->netdev;
3672 
3673 	adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3674 	adapter->rx_ps_bsize0 = 128;
3675 	adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3676 	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3677 	adapter->tx_ring_count = E1000_DEFAULT_TXD;
3678 	adapter->rx_ring_count = E1000_DEFAULT_RXD;
3679 
3680 	spin_lock_init(&adapter->stats64_lock);
3681 
3682 	e1000e_set_interrupt_capability(adapter);
3683 
3684 	if (e1000_alloc_queues(adapter))
3685 		return -ENOMEM;
3686 
3687 	/* Explicitly disable IRQ since the NIC can be in any state. */
3688 	e1000_irq_disable(adapter);
3689 
3690 	set_bit(__E1000_DOWN, &adapter->state);
3691 	return 0;
3692 }
3693 
3694 /**
3695  * e1000_intr_msi_test - Interrupt Handler
3696  * @irq: interrupt number
3697  * @data: pointer to a network interface device structure
3698  **/
3699 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3700 {
3701 	struct net_device *netdev = data;
3702 	struct e1000_adapter *adapter = netdev_priv(netdev);
3703 	struct e1000_hw *hw = &adapter->hw;
3704 	u32 icr = er32(ICR);
3705 
3706 	e_dbg("icr is %08X\n", icr);
3707 	if (icr & E1000_ICR_RXSEQ) {
3708 		adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3709 		wmb();
3710 	}
3711 
3712 	return IRQ_HANDLED;
3713 }
3714 
3715 /**
3716  * e1000_test_msi_interrupt - Returns 0 for successful test
3717  * @adapter: board private struct
3718  *
3719  * code flow taken from tg3.c
3720  **/
3721 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3722 {
3723 	struct net_device *netdev = adapter->netdev;
3724 	struct e1000_hw *hw = &adapter->hw;
3725 	int err;
3726 
3727 	/* poll_enable hasn't been called yet, so don't need disable */
3728 	/* clear any pending events */
3729 	er32(ICR);
3730 
3731 	/* free the real vector and request a test handler */
3732 	e1000_free_irq(adapter);
3733 	e1000e_reset_interrupt_capability(adapter);
3734 
3735 	/* Assume that the test fails, if it succeeds then the test
3736 	 * MSI irq handler will unset this flag */
3737 	adapter->flags |= FLAG_MSI_TEST_FAILED;
3738 
3739 	err = pci_enable_msi(adapter->pdev);
3740 	if (err)
3741 		goto msi_test_failed;
3742 
3743 	err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3744 			  netdev->name, netdev);
3745 	if (err) {
3746 		pci_disable_msi(adapter->pdev);
3747 		goto msi_test_failed;
3748 	}
3749 
3750 	wmb();
3751 
3752 	e1000_irq_enable(adapter);
3753 
3754 	/* fire an unusual interrupt on the test handler */
3755 	ew32(ICS, E1000_ICS_RXSEQ);
3756 	e1e_flush();
3757 	msleep(100);
3758 
3759 	e1000_irq_disable(adapter);
3760 
3761 	rmb();
3762 
3763 	if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3764 		adapter->int_mode = E1000E_INT_MODE_LEGACY;
3765 		e_info("MSI interrupt test failed, using legacy interrupt.\n");
3766 	} else {
3767 		e_dbg("MSI interrupt test succeeded!\n");
3768 	}
3769 
3770 	free_irq(adapter->pdev->irq, netdev);
3771 	pci_disable_msi(adapter->pdev);
3772 
3773 msi_test_failed:
3774 	e1000e_set_interrupt_capability(adapter);
3775 	return e1000_request_irq(adapter);
3776 }
3777 
3778 /**
3779  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3780  * @adapter: board private struct
3781  *
3782  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3783  **/
3784 static int e1000_test_msi(struct e1000_adapter *adapter)
3785 {
3786 	int err;
3787 	u16 pci_cmd;
3788 
3789 	if (!(adapter->flags & FLAG_MSI_ENABLED))
3790 		return 0;
3791 
3792 	/* disable SERR in case the MSI write causes a master abort */
3793 	pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3794 	if (pci_cmd & PCI_COMMAND_SERR)
3795 		pci_write_config_word(adapter->pdev, PCI_COMMAND,
3796 				      pci_cmd & ~PCI_COMMAND_SERR);
3797 
3798 	err = e1000_test_msi_interrupt(adapter);
3799 
3800 	/* re-enable SERR */
3801 	if (pci_cmd & PCI_COMMAND_SERR) {
3802 		pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3803 		pci_cmd |= PCI_COMMAND_SERR;
3804 		pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3805 	}
3806 
3807 	return err;
3808 }
3809 
3810 /**
3811  * e1000_open - Called when a network interface is made active
3812  * @netdev: network interface device structure
3813  *
3814  * Returns 0 on success, negative value on failure
3815  *
3816  * The open entry point is called when a network interface is made
3817  * active by the system (IFF_UP).  At this point all resources needed
3818  * for transmit and receive operations are allocated, the interrupt
3819  * handler is registered with the OS, the watchdog timer is started,
3820  * and the stack is notified that the interface is ready.
3821  **/
3822 static int e1000_open(struct net_device *netdev)
3823 {
3824 	struct e1000_adapter *adapter = netdev_priv(netdev);
3825 	struct e1000_hw *hw = &adapter->hw;
3826 	struct pci_dev *pdev = adapter->pdev;
3827 	int err;
3828 
3829 	/* disallow open during test */
3830 	if (test_bit(__E1000_TESTING, &adapter->state))
3831 		return -EBUSY;
3832 
3833 	pm_runtime_get_sync(&pdev->dev);
3834 
3835 	netif_carrier_off(netdev);
3836 
3837 	/* allocate transmit descriptors */
3838 	err = e1000e_setup_tx_resources(adapter->tx_ring);
3839 	if (err)
3840 		goto err_setup_tx;
3841 
3842 	/* allocate receive descriptors */
3843 	err = e1000e_setup_rx_resources(adapter->rx_ring);
3844 	if (err)
3845 		goto err_setup_rx;
3846 
3847 	/*
3848 	 * If AMT is enabled, let the firmware know that the network
3849 	 * interface is now open and reset the part to a known state.
3850 	 */
3851 	if (adapter->flags & FLAG_HAS_AMT) {
3852 		e1000e_get_hw_control(adapter);
3853 		e1000e_reset(adapter);
3854 	}
3855 
3856 	e1000e_power_up_phy(adapter);
3857 
3858 	adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3859 	if ((adapter->hw.mng_cookie.status &
3860 	     E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3861 		e1000_update_mng_vlan(adapter);
3862 
3863 	/* DMA latency requirement to workaround jumbo issue */
3864 	if (adapter->hw.mac.type == e1000_pch2lan)
3865 		pm_qos_add_request(&adapter->netdev->pm_qos_req,
3866 				   PM_QOS_CPU_DMA_LATENCY,
3867 				   PM_QOS_DEFAULT_VALUE);
3868 
3869 	/*
3870 	 * before we allocate an interrupt, we must be ready to handle it.
3871 	 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3872 	 * as soon as we call pci_request_irq, so we have to setup our
3873 	 * clean_rx handler before we do so.
3874 	 */
3875 	e1000_configure(adapter);
3876 
3877 	err = e1000_request_irq(adapter);
3878 	if (err)
3879 		goto err_req_irq;
3880 
3881 	/*
3882 	 * Work around PCIe errata with MSI interrupts causing some chipsets to
3883 	 * ignore e1000e MSI messages, which means we need to test our MSI
3884 	 * interrupt now
3885 	 */
3886 	if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3887 		err = e1000_test_msi(adapter);
3888 		if (err) {
3889 			e_err("Interrupt allocation failed\n");
3890 			goto err_req_irq;
3891 		}
3892 	}
3893 
3894 	/* From here on the code is the same as e1000e_up() */
3895 	clear_bit(__E1000_DOWN, &adapter->state);
3896 
3897 	napi_enable(&adapter->napi);
3898 
3899 	e1000_irq_enable(adapter);
3900 
3901 	adapter->tx_hang_recheck = false;
3902 	netif_start_queue(netdev);
3903 
3904 	adapter->idle_check = true;
3905 	pm_runtime_put(&pdev->dev);
3906 
3907 	/* fire a link status change interrupt to start the watchdog */
3908 	if (adapter->msix_entries)
3909 		ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3910 	else
3911 		ew32(ICS, E1000_ICS_LSC);
3912 
3913 	return 0;
3914 
3915 err_req_irq:
3916 	e1000e_release_hw_control(adapter);
3917 	e1000_power_down_phy(adapter);
3918 	e1000e_free_rx_resources(adapter->rx_ring);
3919 err_setup_rx:
3920 	e1000e_free_tx_resources(adapter->tx_ring);
3921 err_setup_tx:
3922 	e1000e_reset(adapter);
3923 	pm_runtime_put_sync(&pdev->dev);
3924 
3925 	return err;
3926 }
3927 
3928 /**
3929  * e1000_close - Disables a network interface
3930  * @netdev: network interface device structure
3931  *
3932  * Returns 0, this is not allowed to fail
3933  *
3934  * The close entry point is called when an interface is de-activated
3935  * by the OS.  The hardware is still under the drivers control, but
3936  * needs to be disabled.  A global MAC reset is issued to stop the
3937  * hardware, and all transmit and receive resources are freed.
3938  **/
3939 static int e1000_close(struct net_device *netdev)
3940 {
3941 	struct e1000_adapter *adapter = netdev_priv(netdev);
3942 	struct pci_dev *pdev = adapter->pdev;
3943 	int count = E1000_CHECK_RESET_COUNT;
3944 
3945 	while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
3946 		usleep_range(10000, 20000);
3947 
3948 	WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3949 
3950 	pm_runtime_get_sync(&pdev->dev);
3951 
3952 	napi_disable(&adapter->napi);
3953 
3954 	if (!test_bit(__E1000_DOWN, &adapter->state)) {
3955 		e1000e_down(adapter);
3956 		e1000_free_irq(adapter);
3957 	}
3958 	e1000_power_down_phy(adapter);
3959 
3960 	e1000e_free_tx_resources(adapter->tx_ring);
3961 	e1000e_free_rx_resources(adapter->rx_ring);
3962 
3963 	/*
3964 	 * kill manageability vlan ID if supported, but not if a vlan with
3965 	 * the same ID is registered on the host OS (let 8021q kill it)
3966 	 */
3967 	if (adapter->hw.mng_cookie.status &
3968 	    E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
3969 		e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3970 
3971 	/*
3972 	 * If AMT is enabled, let the firmware know that the network
3973 	 * interface is now closed
3974 	 */
3975 	if ((adapter->flags & FLAG_HAS_AMT) &&
3976 	    !test_bit(__E1000_TESTING, &adapter->state))
3977 		e1000e_release_hw_control(adapter);
3978 
3979 	if (adapter->hw.mac.type == e1000_pch2lan)
3980 		pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3981 
3982 	pm_runtime_put_sync(&pdev->dev);
3983 
3984 	return 0;
3985 }
3986 /**
3987  * e1000_set_mac - Change the Ethernet Address of the NIC
3988  * @netdev: network interface device structure
3989  * @p: pointer to an address structure
3990  *
3991  * Returns 0 on success, negative on failure
3992  **/
3993 static int e1000_set_mac(struct net_device *netdev, void *p)
3994 {
3995 	struct e1000_adapter *adapter = netdev_priv(netdev);
3996 	struct e1000_hw *hw = &adapter->hw;
3997 	struct sockaddr *addr = p;
3998 
3999 	if (!is_valid_ether_addr(addr->sa_data))
4000 		return -EADDRNOTAVAIL;
4001 
4002 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4003 	memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
4004 
4005 	hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
4006 
4007 	if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
4008 		/* activate the work around */
4009 		e1000e_set_laa_state_82571(&adapter->hw, 1);
4010 
4011 		/*
4012 		 * Hold a copy of the LAA in RAR[14] This is done so that
4013 		 * between the time RAR[0] gets clobbered  and the time it
4014 		 * gets fixed (in e1000_watchdog), the actual LAA is in one
4015 		 * of the RARs and no incoming packets directed to this port
4016 		 * are dropped. Eventually the LAA will be in RAR[0] and
4017 		 * RAR[14]
4018 		 */
4019 		hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr,
4020 				    adapter->hw.mac.rar_entry_count - 1);
4021 	}
4022 
4023 	return 0;
4024 }
4025 
4026 /**
4027  * e1000e_update_phy_task - work thread to update phy
4028  * @work: pointer to our work struct
4029  *
4030  * this worker thread exists because we must acquire a
4031  * semaphore to read the phy, which we could msleep while
4032  * waiting for it, and we can't msleep in a timer.
4033  **/
4034 static void e1000e_update_phy_task(struct work_struct *work)
4035 {
4036 	struct e1000_adapter *adapter = container_of(work,
4037 					struct e1000_adapter, update_phy_task);
4038 
4039 	if (test_bit(__E1000_DOWN, &adapter->state))
4040 		return;
4041 
4042 	e1000_get_phy_info(&adapter->hw);
4043 }
4044 
4045 /*
4046  * Need to wait a few seconds after link up to get diagnostic information from
4047  * the phy
4048  */
4049 static void e1000_update_phy_info(unsigned long data)
4050 {
4051 	struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4052 
4053 	if (test_bit(__E1000_DOWN, &adapter->state))
4054 		return;
4055 
4056 	schedule_work(&adapter->update_phy_task);
4057 }
4058 
4059 /**
4060  * e1000e_update_phy_stats - Update the PHY statistics counters
4061  * @adapter: board private structure
4062  *
4063  * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4064  **/
4065 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4066 {
4067 	struct e1000_hw *hw = &adapter->hw;
4068 	s32 ret_val;
4069 	u16 phy_data;
4070 
4071 	ret_val = hw->phy.ops.acquire(hw);
4072 	if (ret_val)
4073 		return;
4074 
4075 	/*
4076 	 * A page set is expensive so check if already on desired page.
4077 	 * If not, set to the page with the PHY status registers.
4078 	 */
4079 	hw->phy.addr = 1;
4080 	ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4081 					   &phy_data);
4082 	if (ret_val)
4083 		goto release;
4084 	if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4085 		ret_val = hw->phy.ops.set_page(hw,
4086 					       HV_STATS_PAGE << IGP_PAGE_SHIFT);
4087 		if (ret_val)
4088 			goto release;
4089 	}
4090 
4091 	/* Single Collision Count */
4092 	hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4093 	ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4094 	if (!ret_val)
4095 		adapter->stats.scc += phy_data;
4096 
4097 	/* Excessive Collision Count */
4098 	hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4099 	ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4100 	if (!ret_val)
4101 		adapter->stats.ecol += phy_data;
4102 
4103 	/* Multiple Collision Count */
4104 	hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4105 	ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4106 	if (!ret_val)
4107 		adapter->stats.mcc += phy_data;
4108 
4109 	/* Late Collision Count */
4110 	hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4111 	ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4112 	if (!ret_val)
4113 		adapter->stats.latecol += phy_data;
4114 
4115 	/* Collision Count - also used for adaptive IFS */
4116 	hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4117 	ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4118 	if (!ret_val)
4119 		hw->mac.collision_delta = phy_data;
4120 
4121 	/* Defer Count */
4122 	hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4123 	ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4124 	if (!ret_val)
4125 		adapter->stats.dc += phy_data;
4126 
4127 	/* Transmit with no CRS */
4128 	hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4129 	ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4130 	if (!ret_val)
4131 		adapter->stats.tncrs += phy_data;
4132 
4133 release:
4134 	hw->phy.ops.release(hw);
4135 }
4136 
4137 /**
4138  * e1000e_update_stats - Update the board statistics counters
4139  * @adapter: board private structure
4140  **/
4141 static void e1000e_update_stats(struct e1000_adapter *adapter)
4142 {
4143 	struct net_device *netdev = adapter->netdev;
4144 	struct e1000_hw *hw = &adapter->hw;
4145 	struct pci_dev *pdev = adapter->pdev;
4146 
4147 	/*
4148 	 * Prevent stats update while adapter is being reset, or if the pci
4149 	 * connection is down.
4150 	 */
4151 	if (adapter->link_speed == 0)
4152 		return;
4153 	if (pci_channel_offline(pdev))
4154 		return;
4155 
4156 	adapter->stats.crcerrs += er32(CRCERRS);
4157 	adapter->stats.gprc += er32(GPRC);
4158 	adapter->stats.gorc += er32(GORCL);
4159 	er32(GORCH); /* Clear gorc */
4160 	adapter->stats.bprc += er32(BPRC);
4161 	adapter->stats.mprc += er32(MPRC);
4162 	adapter->stats.roc += er32(ROC);
4163 
4164 	adapter->stats.mpc += er32(MPC);
4165 
4166 	/* Half-duplex statistics */
4167 	if (adapter->link_duplex == HALF_DUPLEX) {
4168 		if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4169 			e1000e_update_phy_stats(adapter);
4170 		} else {
4171 			adapter->stats.scc += er32(SCC);
4172 			adapter->stats.ecol += er32(ECOL);
4173 			adapter->stats.mcc += er32(MCC);
4174 			adapter->stats.latecol += er32(LATECOL);
4175 			adapter->stats.dc += er32(DC);
4176 
4177 			hw->mac.collision_delta = er32(COLC);
4178 
4179 			if ((hw->mac.type != e1000_82574) &&
4180 			    (hw->mac.type != e1000_82583))
4181 				adapter->stats.tncrs += er32(TNCRS);
4182 		}
4183 		adapter->stats.colc += hw->mac.collision_delta;
4184 	}
4185 
4186 	adapter->stats.xonrxc += er32(XONRXC);
4187 	adapter->stats.xontxc += er32(XONTXC);
4188 	adapter->stats.xoffrxc += er32(XOFFRXC);
4189 	adapter->stats.xofftxc += er32(XOFFTXC);
4190 	adapter->stats.gptc += er32(GPTC);
4191 	adapter->stats.gotc += er32(GOTCL);
4192 	er32(GOTCH); /* Clear gotc */
4193 	adapter->stats.rnbc += er32(RNBC);
4194 	adapter->stats.ruc += er32(RUC);
4195 
4196 	adapter->stats.mptc += er32(MPTC);
4197 	adapter->stats.bptc += er32(BPTC);
4198 
4199 	/* used for adaptive IFS */
4200 
4201 	hw->mac.tx_packet_delta = er32(TPT);
4202 	adapter->stats.tpt += hw->mac.tx_packet_delta;
4203 
4204 	adapter->stats.algnerrc += er32(ALGNERRC);
4205 	adapter->stats.rxerrc += er32(RXERRC);
4206 	adapter->stats.cexterr += er32(CEXTERR);
4207 	adapter->stats.tsctc += er32(TSCTC);
4208 	adapter->stats.tsctfc += er32(TSCTFC);
4209 
4210 	/* Fill out the OS statistics structure */
4211 	netdev->stats.multicast = adapter->stats.mprc;
4212 	netdev->stats.collisions = adapter->stats.colc;
4213 
4214 	/* Rx Errors */
4215 
4216 	/*
4217 	 * RLEC on some newer hardware can be incorrect so build
4218 	 * our own version based on RUC and ROC
4219 	 */
4220 	netdev->stats.rx_errors = adapter->stats.rxerrc +
4221 		adapter->stats.crcerrs + adapter->stats.algnerrc +
4222 		adapter->stats.ruc + adapter->stats.roc +
4223 		adapter->stats.cexterr;
4224 	netdev->stats.rx_length_errors = adapter->stats.ruc +
4225 					      adapter->stats.roc;
4226 	netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4227 	netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4228 	netdev->stats.rx_missed_errors = adapter->stats.mpc;
4229 
4230 	/* Tx Errors */
4231 	netdev->stats.tx_errors = adapter->stats.ecol +
4232 				       adapter->stats.latecol;
4233 	netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4234 	netdev->stats.tx_window_errors = adapter->stats.latecol;
4235 	netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4236 
4237 	/* Tx Dropped needs to be maintained elsewhere */
4238 
4239 	/* Management Stats */
4240 	adapter->stats.mgptc += er32(MGTPTC);
4241 	adapter->stats.mgprc += er32(MGTPRC);
4242 	adapter->stats.mgpdc += er32(MGTPDC);
4243 }
4244 
4245 /**
4246  * e1000_phy_read_status - Update the PHY register status snapshot
4247  * @adapter: board private structure
4248  **/
4249 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4250 {
4251 	struct e1000_hw *hw = &adapter->hw;
4252 	struct e1000_phy_regs *phy = &adapter->phy_regs;
4253 
4254 	if ((er32(STATUS) & E1000_STATUS_LU) &&
4255 	    (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4256 		int ret_val;
4257 
4258 		ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4259 		ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4260 		ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4261 		ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4262 		ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4263 		ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4264 		ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4265 		ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4266 		if (ret_val)
4267 			e_warn("Error reading PHY register\n");
4268 	} else {
4269 		/*
4270 		 * Do not read PHY registers if link is not up
4271 		 * Set values to typical power-on defaults
4272 		 */
4273 		phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4274 		phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4275 			     BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4276 			     BMSR_ERCAP);
4277 		phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4278 				  ADVERTISE_ALL | ADVERTISE_CSMA);
4279 		phy->lpa = 0;
4280 		phy->expansion = EXPANSION_ENABLENPAGE;
4281 		phy->ctrl1000 = ADVERTISE_1000FULL;
4282 		phy->stat1000 = 0;
4283 		phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4284 	}
4285 }
4286 
4287 static void e1000_print_link_info(struct e1000_adapter *adapter)
4288 {
4289 	struct e1000_hw *hw = &adapter->hw;
4290 	u32 ctrl = er32(CTRL);
4291 
4292 	/* Link status message must follow this format for user tools */
4293 	printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4294 		adapter->netdev->name,
4295 		adapter->link_speed,
4296 		adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
4297 		(ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
4298 		(ctrl & E1000_CTRL_RFCE) ? "Rx" :
4299 		(ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
4300 }
4301 
4302 static bool e1000e_has_link(struct e1000_adapter *adapter)
4303 {
4304 	struct e1000_hw *hw = &adapter->hw;
4305 	bool link_active = false;
4306 	s32 ret_val = 0;
4307 
4308 	/*
4309 	 * get_link_status is set on LSC (link status) interrupt or
4310 	 * Rx sequence error interrupt.  get_link_status will stay
4311 	 * false until the check_for_link establishes link
4312 	 * for copper adapters ONLY
4313 	 */
4314 	switch (hw->phy.media_type) {
4315 	case e1000_media_type_copper:
4316 		if (hw->mac.get_link_status) {
4317 			ret_val = hw->mac.ops.check_for_link(hw);
4318 			link_active = !hw->mac.get_link_status;
4319 		} else {
4320 			link_active = true;
4321 		}
4322 		break;
4323 	case e1000_media_type_fiber:
4324 		ret_val = hw->mac.ops.check_for_link(hw);
4325 		link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4326 		break;
4327 	case e1000_media_type_internal_serdes:
4328 		ret_val = hw->mac.ops.check_for_link(hw);
4329 		link_active = adapter->hw.mac.serdes_has_link;
4330 		break;
4331 	default:
4332 	case e1000_media_type_unknown:
4333 		break;
4334 	}
4335 
4336 	if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4337 	    (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4338 		/* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4339 		e_info("Gigabit has been disabled, downgrading speed\n");
4340 	}
4341 
4342 	return link_active;
4343 }
4344 
4345 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4346 {
4347 	/* make sure the receive unit is started */
4348 	if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4349 	    (adapter->flags & FLAG_RX_RESTART_NOW)) {
4350 		struct e1000_hw *hw = &adapter->hw;
4351 		u32 rctl = er32(RCTL);
4352 		ew32(RCTL, rctl | E1000_RCTL_EN);
4353 		adapter->flags &= ~FLAG_RX_RESTART_NOW;
4354 	}
4355 }
4356 
4357 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4358 {
4359 	struct e1000_hw *hw = &adapter->hw;
4360 
4361 	/*
4362 	 * With 82574 controllers, PHY needs to be checked periodically
4363 	 * for hung state and reset, if two calls return true
4364 	 */
4365 	if (e1000_check_phy_82574(hw))
4366 		adapter->phy_hang_count++;
4367 	else
4368 		adapter->phy_hang_count = 0;
4369 
4370 	if (adapter->phy_hang_count > 1) {
4371 		adapter->phy_hang_count = 0;
4372 		schedule_work(&adapter->reset_task);
4373 	}
4374 }
4375 
4376 /**
4377  * e1000_watchdog - Timer Call-back
4378  * @data: pointer to adapter cast into an unsigned long
4379  **/
4380 static void e1000_watchdog(unsigned long data)
4381 {
4382 	struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4383 
4384 	/* Do the rest outside of interrupt context */
4385 	schedule_work(&adapter->watchdog_task);
4386 
4387 	/* TODO: make this use queue_delayed_work() */
4388 }
4389 
4390 static void e1000_watchdog_task(struct work_struct *work)
4391 {
4392 	struct e1000_adapter *adapter = container_of(work,
4393 					struct e1000_adapter, watchdog_task);
4394 	struct net_device *netdev = adapter->netdev;
4395 	struct e1000_mac_info *mac = &adapter->hw.mac;
4396 	struct e1000_phy_info *phy = &adapter->hw.phy;
4397 	struct e1000_ring *tx_ring = adapter->tx_ring;
4398 	struct e1000_hw *hw = &adapter->hw;
4399 	u32 link, tctl;
4400 
4401 	if (test_bit(__E1000_DOWN, &adapter->state))
4402 		return;
4403 
4404 	link = e1000e_has_link(adapter);
4405 	if ((netif_carrier_ok(netdev)) && link) {
4406 		/* Cancel scheduled suspend requests. */
4407 		pm_runtime_resume(netdev->dev.parent);
4408 
4409 		e1000e_enable_receives(adapter);
4410 		goto link_up;
4411 	}
4412 
4413 	if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4414 	    (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4415 		e1000_update_mng_vlan(adapter);
4416 
4417 	if (link) {
4418 		if (!netif_carrier_ok(netdev)) {
4419 			bool txb2b = true;
4420 
4421 			/* Cancel scheduled suspend requests. */
4422 			pm_runtime_resume(netdev->dev.parent);
4423 
4424 			/* update snapshot of PHY registers on LSC */
4425 			e1000_phy_read_status(adapter);
4426 			mac->ops.get_link_up_info(&adapter->hw,
4427 						   &adapter->link_speed,
4428 						   &adapter->link_duplex);
4429 			e1000_print_link_info(adapter);
4430 			/*
4431 			 * On supported PHYs, check for duplex mismatch only
4432 			 * if link has autonegotiated at 10/100 half
4433 			 */
4434 			if ((hw->phy.type == e1000_phy_igp_3 ||
4435 			     hw->phy.type == e1000_phy_bm) &&
4436 			    (hw->mac.autoneg == true) &&
4437 			    (adapter->link_speed == SPEED_10 ||
4438 			     adapter->link_speed == SPEED_100) &&
4439 			    (adapter->link_duplex == HALF_DUPLEX)) {
4440 				u16 autoneg_exp;
4441 
4442 				e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4443 
4444 				if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4445 					e_info("Autonegotiated half duplex but link partner cannot autoneg.  Try forcing full duplex if link gets many collisions.\n");
4446 			}
4447 
4448 			/* adjust timeout factor according to speed/duplex */
4449 			adapter->tx_timeout_factor = 1;
4450 			switch (adapter->link_speed) {
4451 			case SPEED_10:
4452 				txb2b = false;
4453 				adapter->tx_timeout_factor = 16;
4454 				break;
4455 			case SPEED_100:
4456 				txb2b = false;
4457 				adapter->tx_timeout_factor = 10;
4458 				break;
4459 			}
4460 
4461 			/*
4462 			 * workaround: re-program speed mode bit after
4463 			 * link-up event
4464 			 */
4465 			if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4466 			    !txb2b) {
4467 				u32 tarc0;
4468 				tarc0 = er32(TARC(0));
4469 				tarc0 &= ~SPEED_MODE_BIT;
4470 				ew32(TARC(0), tarc0);
4471 			}
4472 
4473 			/*
4474 			 * disable TSO for pcie and 10/100 speeds, to avoid
4475 			 * some hardware issues
4476 			 */
4477 			if (!(adapter->flags & FLAG_TSO_FORCE)) {
4478 				switch (adapter->link_speed) {
4479 				case SPEED_10:
4480 				case SPEED_100:
4481 					e_info("10/100 speed: disabling TSO\n");
4482 					netdev->features &= ~NETIF_F_TSO;
4483 					netdev->features &= ~NETIF_F_TSO6;
4484 					break;
4485 				case SPEED_1000:
4486 					netdev->features |= NETIF_F_TSO;
4487 					netdev->features |= NETIF_F_TSO6;
4488 					break;
4489 				default:
4490 					/* oops */
4491 					break;
4492 				}
4493 			}
4494 
4495 			/*
4496 			 * enable transmits in the hardware, need to do this
4497 			 * after setting TARC(0)
4498 			 */
4499 			tctl = er32(TCTL);
4500 			tctl |= E1000_TCTL_EN;
4501 			ew32(TCTL, tctl);
4502 
4503                         /*
4504 			 * Perform any post-link-up configuration before
4505 			 * reporting link up.
4506 			 */
4507 			if (phy->ops.cfg_on_link_up)
4508 				phy->ops.cfg_on_link_up(hw);
4509 
4510 			netif_carrier_on(netdev);
4511 
4512 			if (!test_bit(__E1000_DOWN, &adapter->state))
4513 				mod_timer(&adapter->phy_info_timer,
4514 					  round_jiffies(jiffies + 2 * HZ));
4515 		}
4516 	} else {
4517 		if (netif_carrier_ok(netdev)) {
4518 			adapter->link_speed = 0;
4519 			adapter->link_duplex = 0;
4520 			/* Link status message must follow this format */
4521 			printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4522 			       adapter->netdev->name);
4523 			netif_carrier_off(netdev);
4524 			if (!test_bit(__E1000_DOWN, &adapter->state))
4525 				mod_timer(&adapter->phy_info_timer,
4526 					  round_jiffies(jiffies + 2 * HZ));
4527 
4528 			if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4529 				schedule_work(&adapter->reset_task);
4530 			else
4531 				pm_schedule_suspend(netdev->dev.parent,
4532 							LINK_TIMEOUT);
4533 		}
4534 	}
4535 
4536 link_up:
4537 	spin_lock(&adapter->stats64_lock);
4538 	e1000e_update_stats(adapter);
4539 
4540 	mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4541 	adapter->tpt_old = adapter->stats.tpt;
4542 	mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4543 	adapter->colc_old = adapter->stats.colc;
4544 
4545 	adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4546 	adapter->gorc_old = adapter->stats.gorc;
4547 	adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4548 	adapter->gotc_old = adapter->stats.gotc;
4549 	spin_unlock(&adapter->stats64_lock);
4550 
4551 	e1000e_update_adaptive(&adapter->hw);
4552 
4553 	if (!netif_carrier_ok(netdev) &&
4554 	    (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4555 		/*
4556 		 * We've lost link, so the controller stops DMA,
4557 		 * but we've got queued Tx work that's never going
4558 		 * to get done, so reset controller to flush Tx.
4559 		 * (Do the reset outside of interrupt context).
4560 		 */
4561 		schedule_work(&adapter->reset_task);
4562 		/* return immediately since reset is imminent */
4563 		return;
4564 	}
4565 
4566 	/* Simple mode for Interrupt Throttle Rate (ITR) */
4567 	if (adapter->itr_setting == 4) {
4568 		/*
4569 		 * Symmetric Tx/Rx gets a reduced ITR=2000;
4570 		 * Total asymmetrical Tx or Rx gets ITR=8000;
4571 		 * everyone else is between 2000-8000.
4572 		 */
4573 		u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4574 		u32 dif = (adapter->gotc > adapter->gorc ?
4575 			    adapter->gotc - adapter->gorc :
4576 			    adapter->gorc - adapter->gotc) / 10000;
4577 		u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4578 
4579 		ew32(ITR, 1000000000 / (itr * 256));
4580 	}
4581 
4582 	/* Cause software interrupt to ensure Rx ring is cleaned */
4583 	if (adapter->msix_entries)
4584 		ew32(ICS, adapter->rx_ring->ims_val);
4585 	else
4586 		ew32(ICS, E1000_ICS_RXDMT0);
4587 
4588 	/* flush pending descriptors to memory before detecting Tx hang */
4589 	e1000e_flush_descriptors(adapter);
4590 
4591 	/* Force detection of hung controller every watchdog period */
4592 	adapter->detect_tx_hung = true;
4593 
4594 	/*
4595 	 * With 82571 controllers, LAA may be overwritten due to controller
4596 	 * reset from the other port. Set the appropriate LAA in RAR[0]
4597 	 */
4598 	if (e1000e_get_laa_state_82571(hw))
4599 		hw->mac.ops.rar_set(hw, adapter->hw.mac.addr, 0);
4600 
4601 	if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4602 		e1000e_check_82574_phy_workaround(adapter);
4603 
4604 	/* Reset the timer */
4605 	if (!test_bit(__E1000_DOWN, &adapter->state))
4606 		mod_timer(&adapter->watchdog_timer,
4607 			  round_jiffies(jiffies + 2 * HZ));
4608 }
4609 
4610 #define E1000_TX_FLAGS_CSUM		0x00000001
4611 #define E1000_TX_FLAGS_VLAN		0x00000002
4612 #define E1000_TX_FLAGS_TSO		0x00000004
4613 #define E1000_TX_FLAGS_IPV4		0x00000008
4614 #define E1000_TX_FLAGS_NO_FCS		0x00000010
4615 #define E1000_TX_FLAGS_VLAN_MASK	0xffff0000
4616 #define E1000_TX_FLAGS_VLAN_SHIFT	16
4617 
4618 static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb)
4619 {
4620 	struct e1000_context_desc *context_desc;
4621 	struct e1000_buffer *buffer_info;
4622 	unsigned int i;
4623 	u32 cmd_length = 0;
4624 	u16 ipcse = 0, tucse, mss;
4625 	u8 ipcss, ipcso, tucss, tucso, hdr_len;
4626 
4627 	if (!skb_is_gso(skb))
4628 		return 0;
4629 
4630 	if (skb_header_cloned(skb)) {
4631 		int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4632 
4633 		if (err)
4634 			return err;
4635 	}
4636 
4637 	hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4638 	mss = skb_shinfo(skb)->gso_size;
4639 	if (skb->protocol == htons(ETH_P_IP)) {
4640 		struct iphdr *iph = ip_hdr(skb);
4641 		iph->tot_len = 0;
4642 		iph->check = 0;
4643 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4644 		                                         0, IPPROTO_TCP, 0);
4645 		cmd_length = E1000_TXD_CMD_IP;
4646 		ipcse = skb_transport_offset(skb) - 1;
4647 	} else if (skb_is_gso_v6(skb)) {
4648 		ipv6_hdr(skb)->payload_len = 0;
4649 		tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4650 		                                       &ipv6_hdr(skb)->daddr,
4651 		                                       0, IPPROTO_TCP, 0);
4652 		ipcse = 0;
4653 	}
4654 	ipcss = skb_network_offset(skb);
4655 	ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4656 	tucss = skb_transport_offset(skb);
4657 	tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4658 	tucse = 0;
4659 
4660 	cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4661 	               E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4662 
4663 	i = tx_ring->next_to_use;
4664 	context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4665 	buffer_info = &tx_ring->buffer_info[i];
4666 
4667 	context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4668 	context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4669 	context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4670 	context_desc->upper_setup.tcp_fields.tucss = tucss;
4671 	context_desc->upper_setup.tcp_fields.tucso = tucso;
4672 	context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4673 	context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4674 	context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4675 	context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4676 
4677 	buffer_info->time_stamp = jiffies;
4678 	buffer_info->next_to_watch = i;
4679 
4680 	i++;
4681 	if (i == tx_ring->count)
4682 		i = 0;
4683 	tx_ring->next_to_use = i;
4684 
4685 	return 1;
4686 }
4687 
4688 static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb)
4689 {
4690 	struct e1000_adapter *adapter = tx_ring->adapter;
4691 	struct e1000_context_desc *context_desc;
4692 	struct e1000_buffer *buffer_info;
4693 	unsigned int i;
4694 	u8 css;
4695 	u32 cmd_len = E1000_TXD_CMD_DEXT;
4696 	__be16 protocol;
4697 
4698 	if (skb->ip_summed != CHECKSUM_PARTIAL)
4699 		return 0;
4700 
4701 	if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4702 		protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4703 	else
4704 		protocol = skb->protocol;
4705 
4706 	switch (protocol) {
4707 	case cpu_to_be16(ETH_P_IP):
4708 		if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4709 			cmd_len |= E1000_TXD_CMD_TCP;
4710 		break;
4711 	case cpu_to_be16(ETH_P_IPV6):
4712 		/* XXX not handling all IPV6 headers */
4713 		if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4714 			cmd_len |= E1000_TXD_CMD_TCP;
4715 		break;
4716 	default:
4717 		if (unlikely(net_ratelimit()))
4718 			e_warn("checksum_partial proto=%x!\n",
4719 			       be16_to_cpu(protocol));
4720 		break;
4721 	}
4722 
4723 	css = skb_checksum_start_offset(skb);
4724 
4725 	i = tx_ring->next_to_use;
4726 	buffer_info = &tx_ring->buffer_info[i];
4727 	context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4728 
4729 	context_desc->lower_setup.ip_config = 0;
4730 	context_desc->upper_setup.tcp_fields.tucss = css;
4731 	context_desc->upper_setup.tcp_fields.tucso =
4732 				css + skb->csum_offset;
4733 	context_desc->upper_setup.tcp_fields.tucse = 0;
4734 	context_desc->tcp_seg_setup.data = 0;
4735 	context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4736 
4737 	buffer_info->time_stamp = jiffies;
4738 	buffer_info->next_to_watch = i;
4739 
4740 	i++;
4741 	if (i == tx_ring->count)
4742 		i = 0;
4743 	tx_ring->next_to_use = i;
4744 
4745 	return 1;
4746 }
4747 
4748 #define E1000_MAX_PER_TXD	8192
4749 #define E1000_MAX_TXD_PWR	12
4750 
4751 static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
4752 			unsigned int first, unsigned int max_per_txd,
4753 			unsigned int nr_frags, unsigned int mss)
4754 {
4755 	struct e1000_adapter *adapter = tx_ring->adapter;
4756 	struct pci_dev *pdev = adapter->pdev;
4757 	struct e1000_buffer *buffer_info;
4758 	unsigned int len = skb_headlen(skb);
4759 	unsigned int offset = 0, size, count = 0, i;
4760 	unsigned int f, bytecount, segs;
4761 
4762 	i = tx_ring->next_to_use;
4763 
4764 	while (len) {
4765 		buffer_info = &tx_ring->buffer_info[i];
4766 		size = min(len, max_per_txd);
4767 
4768 		buffer_info->length = size;
4769 		buffer_info->time_stamp = jiffies;
4770 		buffer_info->next_to_watch = i;
4771 		buffer_info->dma = dma_map_single(&pdev->dev,
4772 						  skb->data + offset,
4773 						  size, DMA_TO_DEVICE);
4774 		buffer_info->mapped_as_page = false;
4775 		if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4776 			goto dma_error;
4777 
4778 		len -= size;
4779 		offset += size;
4780 		count++;
4781 
4782 		if (len) {
4783 			i++;
4784 			if (i == tx_ring->count)
4785 				i = 0;
4786 		}
4787 	}
4788 
4789 	for (f = 0; f < nr_frags; f++) {
4790 		const struct skb_frag_struct *frag;
4791 
4792 		frag = &skb_shinfo(skb)->frags[f];
4793 		len = skb_frag_size(frag);
4794 		offset = 0;
4795 
4796 		while (len) {
4797 			i++;
4798 			if (i == tx_ring->count)
4799 				i = 0;
4800 
4801 			buffer_info = &tx_ring->buffer_info[i];
4802 			size = min(len, max_per_txd);
4803 
4804 			buffer_info->length = size;
4805 			buffer_info->time_stamp = jiffies;
4806 			buffer_info->next_to_watch = i;
4807 			buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
4808 						offset, size, DMA_TO_DEVICE);
4809 			buffer_info->mapped_as_page = true;
4810 			if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4811 				goto dma_error;
4812 
4813 			len -= size;
4814 			offset += size;
4815 			count++;
4816 		}
4817 	}
4818 
4819 	segs = skb_shinfo(skb)->gso_segs ? : 1;
4820 	/* multiply data chunks by size of headers */
4821 	bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4822 
4823 	tx_ring->buffer_info[i].skb = skb;
4824 	tx_ring->buffer_info[i].segs = segs;
4825 	tx_ring->buffer_info[i].bytecount = bytecount;
4826 	tx_ring->buffer_info[first].next_to_watch = i;
4827 
4828 	return count;
4829 
4830 dma_error:
4831 	dev_err(&pdev->dev, "Tx DMA map failed\n");
4832 	buffer_info->dma = 0;
4833 	if (count)
4834 		count--;
4835 
4836 	while (count--) {
4837 		if (i == 0)
4838 			i += tx_ring->count;
4839 		i--;
4840 		buffer_info = &tx_ring->buffer_info[i];
4841 		e1000_put_txbuf(tx_ring, buffer_info);
4842 	}
4843 
4844 	return 0;
4845 }
4846 
4847 static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
4848 {
4849 	struct e1000_adapter *adapter = tx_ring->adapter;
4850 	struct e1000_tx_desc *tx_desc = NULL;
4851 	struct e1000_buffer *buffer_info;
4852 	u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4853 	unsigned int i;
4854 
4855 	if (tx_flags & E1000_TX_FLAGS_TSO) {
4856 		txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4857 			     E1000_TXD_CMD_TSE;
4858 		txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4859 
4860 		if (tx_flags & E1000_TX_FLAGS_IPV4)
4861 			txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4862 	}
4863 
4864 	if (tx_flags & E1000_TX_FLAGS_CSUM) {
4865 		txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4866 		txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4867 	}
4868 
4869 	if (tx_flags & E1000_TX_FLAGS_VLAN) {
4870 		txd_lower |= E1000_TXD_CMD_VLE;
4871 		txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4872 	}
4873 
4874 	if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
4875 		txd_lower &= ~(E1000_TXD_CMD_IFCS);
4876 
4877 	i = tx_ring->next_to_use;
4878 
4879 	do {
4880 		buffer_info = &tx_ring->buffer_info[i];
4881 		tx_desc = E1000_TX_DESC(*tx_ring, i);
4882 		tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4883 		tx_desc->lower.data =
4884 			cpu_to_le32(txd_lower | buffer_info->length);
4885 		tx_desc->upper.data = cpu_to_le32(txd_upper);
4886 
4887 		i++;
4888 		if (i == tx_ring->count)
4889 			i = 0;
4890 	} while (--count > 0);
4891 
4892 	tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4893 
4894 	/* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
4895 	if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
4896 		tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
4897 
4898 	/*
4899 	 * Force memory writes to complete before letting h/w
4900 	 * know there are new descriptors to fetch.  (Only
4901 	 * applicable for weak-ordered memory model archs,
4902 	 * such as IA-64).
4903 	 */
4904 	wmb();
4905 
4906 	tx_ring->next_to_use = i;
4907 
4908 	if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
4909 		e1000e_update_tdt_wa(tx_ring, i);
4910 	else
4911 		writel(i, tx_ring->tail);
4912 
4913 	/*
4914 	 * we need this if more than one processor can write to our tail
4915 	 * at a time, it synchronizes IO on IA64/Altix systems
4916 	 */
4917 	mmiowb();
4918 }
4919 
4920 #define MINIMUM_DHCP_PACKET_SIZE 282
4921 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4922 				    struct sk_buff *skb)
4923 {
4924 	struct e1000_hw *hw =  &adapter->hw;
4925 	u16 length, offset;
4926 
4927 	if (vlan_tx_tag_present(skb)) {
4928 		if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4929 		    (adapter->hw.mng_cookie.status &
4930 			E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4931 			return 0;
4932 	}
4933 
4934 	if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4935 		return 0;
4936 
4937 	if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4938 		return 0;
4939 
4940 	{
4941 		const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4942 		struct udphdr *udp;
4943 
4944 		if (ip->protocol != IPPROTO_UDP)
4945 			return 0;
4946 
4947 		udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4948 		if (ntohs(udp->dest) != 67)
4949 			return 0;
4950 
4951 		offset = (u8 *)udp + 8 - skb->data;
4952 		length = skb->len - offset;
4953 		return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4954 	}
4955 
4956 	return 0;
4957 }
4958 
4959 static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
4960 {
4961 	struct e1000_adapter *adapter = tx_ring->adapter;
4962 
4963 	netif_stop_queue(adapter->netdev);
4964 	/*
4965 	 * Herbert's original patch had:
4966 	 *  smp_mb__after_netif_stop_queue();
4967 	 * but since that doesn't exist yet, just open code it.
4968 	 */
4969 	smp_mb();
4970 
4971 	/*
4972 	 * We need to check again in a case another CPU has just
4973 	 * made room available.
4974 	 */
4975 	if (e1000_desc_unused(tx_ring) < size)
4976 		return -EBUSY;
4977 
4978 	/* A reprieve! */
4979 	netif_start_queue(adapter->netdev);
4980 	++adapter->restart_queue;
4981 	return 0;
4982 }
4983 
4984 static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
4985 {
4986 	if (e1000_desc_unused(tx_ring) >= size)
4987 		return 0;
4988 	return __e1000_maybe_stop_tx(tx_ring, size);
4989 }
4990 
4991 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1)
4992 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4993 				    struct net_device *netdev)
4994 {
4995 	struct e1000_adapter *adapter = netdev_priv(netdev);
4996 	struct e1000_ring *tx_ring = adapter->tx_ring;
4997 	unsigned int first;
4998 	unsigned int max_per_txd = E1000_MAX_PER_TXD;
4999 	unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
5000 	unsigned int tx_flags = 0;
5001 	unsigned int len = skb_headlen(skb);
5002 	unsigned int nr_frags;
5003 	unsigned int mss;
5004 	int count = 0;
5005 	int tso;
5006 	unsigned int f;
5007 
5008 	if (test_bit(__E1000_DOWN, &adapter->state)) {
5009 		dev_kfree_skb_any(skb);
5010 		return NETDEV_TX_OK;
5011 	}
5012 
5013 	if (skb->len <= 0) {
5014 		dev_kfree_skb_any(skb);
5015 		return NETDEV_TX_OK;
5016 	}
5017 
5018 	mss = skb_shinfo(skb)->gso_size;
5019 	/*
5020 	 * The controller does a simple calculation to
5021 	 * make sure there is enough room in the FIFO before
5022 	 * initiating the DMA for each buffer.  The calc is:
5023 	 * 4 = ceil(buffer len/mss).  To make sure we don't
5024 	 * overrun the FIFO, adjust the max buffer len if mss
5025 	 * drops.
5026 	 */
5027 	if (mss) {
5028 		u8 hdr_len;
5029 		max_per_txd = min(mss << 2, max_per_txd);
5030 		max_txd_pwr = fls(max_per_txd) - 1;
5031 
5032 		/*
5033 		 * TSO Workaround for 82571/2/3 Controllers -- if skb->data
5034 		 * points to just header, pull a few bytes of payload from
5035 		 * frags into skb->data
5036 		 */
5037 		hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5038 		/*
5039 		 * we do this workaround for ES2LAN, but it is un-necessary,
5040 		 * avoiding it could save a lot of cycles
5041 		 */
5042 		if (skb->data_len && (hdr_len == len)) {
5043 			unsigned int pull_size;
5044 
5045 			pull_size = min_t(unsigned int, 4, skb->data_len);
5046 			if (!__pskb_pull_tail(skb, pull_size)) {
5047 				e_err("__pskb_pull_tail failed.\n");
5048 				dev_kfree_skb_any(skb);
5049 				return NETDEV_TX_OK;
5050 			}
5051 			len = skb_headlen(skb);
5052 		}
5053 	}
5054 
5055 	/* reserve a descriptor for the offload context */
5056 	if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5057 		count++;
5058 	count++;
5059 
5060 	count += TXD_USE_COUNT(len, max_txd_pwr);
5061 
5062 	nr_frags = skb_shinfo(skb)->nr_frags;
5063 	for (f = 0; f < nr_frags; f++)
5064 		count += TXD_USE_COUNT(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5065 				       max_txd_pwr);
5066 
5067 	if (adapter->hw.mac.tx_pkt_filtering)
5068 		e1000_transfer_dhcp_info(adapter, skb);
5069 
5070 	/*
5071 	 * need: count + 2 desc gap to keep tail from touching
5072 	 * head, otherwise try next time
5073 	 */
5074 	if (e1000_maybe_stop_tx(tx_ring, count + 2))
5075 		return NETDEV_TX_BUSY;
5076 
5077 	if (vlan_tx_tag_present(skb)) {
5078 		tx_flags |= E1000_TX_FLAGS_VLAN;
5079 		tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
5080 	}
5081 
5082 	first = tx_ring->next_to_use;
5083 
5084 	tso = e1000_tso(tx_ring, skb);
5085 	if (tso < 0) {
5086 		dev_kfree_skb_any(skb);
5087 		return NETDEV_TX_OK;
5088 	}
5089 
5090 	if (tso)
5091 		tx_flags |= E1000_TX_FLAGS_TSO;
5092 	else if (e1000_tx_csum(tx_ring, skb))
5093 		tx_flags |= E1000_TX_FLAGS_CSUM;
5094 
5095 	/*
5096 	 * Old method was to assume IPv4 packet by default if TSO was enabled.
5097 	 * 82571 hardware supports TSO capabilities for IPv6 as well...
5098 	 * no longer assume, we must.
5099 	 */
5100 	if (skb->protocol == htons(ETH_P_IP))
5101 		tx_flags |= E1000_TX_FLAGS_IPV4;
5102 
5103 	if (unlikely(skb->no_fcs))
5104 		tx_flags |= E1000_TX_FLAGS_NO_FCS;
5105 
5106 	/* if count is 0 then mapping error has occurred */
5107 	count = e1000_tx_map(tx_ring, skb, first, max_per_txd, nr_frags, mss);
5108 	if (count) {
5109 		skb_tx_timestamp(skb);
5110 
5111 		netdev_sent_queue(netdev, skb->len);
5112 		e1000_tx_queue(tx_ring, tx_flags, count);
5113 		/* Make sure there is space in the ring for the next send. */
5114 		e1000_maybe_stop_tx(tx_ring, MAX_SKB_FRAGS + 2);
5115 
5116 	} else {
5117 		dev_kfree_skb_any(skb);
5118 		tx_ring->buffer_info[first].time_stamp = 0;
5119 		tx_ring->next_to_use = first;
5120 	}
5121 
5122 	return NETDEV_TX_OK;
5123 }
5124 
5125 /**
5126  * e1000_tx_timeout - Respond to a Tx Hang
5127  * @netdev: network interface device structure
5128  **/
5129 static void e1000_tx_timeout(struct net_device *netdev)
5130 {
5131 	struct e1000_adapter *adapter = netdev_priv(netdev);
5132 
5133 	/* Do the reset outside of interrupt context */
5134 	adapter->tx_timeout_count++;
5135 	schedule_work(&adapter->reset_task);
5136 }
5137 
5138 static void e1000_reset_task(struct work_struct *work)
5139 {
5140 	struct e1000_adapter *adapter;
5141 	adapter = container_of(work, struct e1000_adapter, reset_task);
5142 
5143 	/* don't run the task if already down */
5144 	if (test_bit(__E1000_DOWN, &adapter->state))
5145 		return;
5146 
5147 	if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5148 	      (adapter->flags & FLAG_RX_RESTART_NOW))) {
5149 		e1000e_dump(adapter);
5150 		e_err("Reset adapter\n");
5151 	}
5152 	e1000e_reinit_locked(adapter);
5153 }
5154 
5155 /**
5156  * e1000_get_stats64 - Get System Network Statistics
5157  * @netdev: network interface device structure
5158  * @stats: rtnl_link_stats64 pointer
5159  *
5160  * Returns the address of the device statistics structure.
5161  **/
5162 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
5163                                              struct rtnl_link_stats64 *stats)
5164 {
5165 	struct e1000_adapter *adapter = netdev_priv(netdev);
5166 
5167 	memset(stats, 0, sizeof(struct rtnl_link_stats64));
5168 	spin_lock(&adapter->stats64_lock);
5169 	e1000e_update_stats(adapter);
5170 	/* Fill out the OS statistics structure */
5171 	stats->rx_bytes = adapter->stats.gorc;
5172 	stats->rx_packets = adapter->stats.gprc;
5173 	stats->tx_bytes = adapter->stats.gotc;
5174 	stats->tx_packets = adapter->stats.gptc;
5175 	stats->multicast = adapter->stats.mprc;
5176 	stats->collisions = adapter->stats.colc;
5177 
5178 	/* Rx Errors */
5179 
5180 	/*
5181 	 * RLEC on some newer hardware can be incorrect so build
5182 	 * our own version based on RUC and ROC
5183 	 */
5184 	stats->rx_errors = adapter->stats.rxerrc +
5185 		adapter->stats.crcerrs + adapter->stats.algnerrc +
5186 		adapter->stats.ruc + adapter->stats.roc +
5187 		adapter->stats.cexterr;
5188 	stats->rx_length_errors = adapter->stats.ruc +
5189 					      adapter->stats.roc;
5190 	stats->rx_crc_errors = adapter->stats.crcerrs;
5191 	stats->rx_frame_errors = adapter->stats.algnerrc;
5192 	stats->rx_missed_errors = adapter->stats.mpc;
5193 
5194 	/* Tx Errors */
5195 	stats->tx_errors = adapter->stats.ecol +
5196 				       adapter->stats.latecol;
5197 	stats->tx_aborted_errors = adapter->stats.ecol;
5198 	stats->tx_window_errors = adapter->stats.latecol;
5199 	stats->tx_carrier_errors = adapter->stats.tncrs;
5200 
5201 	/* Tx Dropped needs to be maintained elsewhere */
5202 
5203 	spin_unlock(&adapter->stats64_lock);
5204 	return stats;
5205 }
5206 
5207 /**
5208  * e1000_change_mtu - Change the Maximum Transfer Unit
5209  * @netdev: network interface device structure
5210  * @new_mtu: new value for maximum frame size
5211  *
5212  * Returns 0 on success, negative on failure
5213  **/
5214 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5215 {
5216 	struct e1000_adapter *adapter = netdev_priv(netdev);
5217 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5218 
5219 	/* Jumbo frame support */
5220 	if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
5221 	    !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5222 		e_err("Jumbo Frames not supported.\n");
5223 		return -EINVAL;
5224 	}
5225 
5226 	/* Supported frame sizes */
5227 	if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5228 	    (max_frame > adapter->max_hw_frame_size)) {
5229 		e_err("Unsupported MTU setting\n");
5230 		return -EINVAL;
5231 	}
5232 
5233 	/* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
5234 	if ((adapter->hw.mac.type >= e1000_pch2lan) &&
5235 	    !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5236 	    (new_mtu > ETH_DATA_LEN)) {
5237 		e_err("Jumbo Frames not supported on this device when CRC stripping is disabled.\n");
5238 		return -EINVAL;
5239 	}
5240 
5241 	while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5242 		usleep_range(1000, 2000);
5243 	/* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5244 	adapter->max_frame_size = max_frame;
5245 	e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5246 	netdev->mtu = new_mtu;
5247 	if (netif_running(netdev))
5248 		e1000e_down(adapter);
5249 
5250 	/*
5251 	 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5252 	 * means we reserve 2 more, this pushes us to allocate from the next
5253 	 * larger slab size.
5254 	 * i.e. RXBUFFER_2048 --> size-4096 slab
5255 	 * However with the new *_jumbo_rx* routines, jumbo receives will use
5256 	 * fragmented skbs
5257 	 */
5258 
5259 	if (max_frame <= 2048)
5260 		adapter->rx_buffer_len = 2048;
5261 	else
5262 		adapter->rx_buffer_len = 4096;
5263 
5264 	/* adjust allocation if LPE protects us, and we aren't using SBP */
5265 	if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5266 	     (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5267 		adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5268 					 + ETH_FCS_LEN;
5269 
5270 	if (netif_running(netdev))
5271 		e1000e_up(adapter);
5272 	else
5273 		e1000e_reset(adapter);
5274 
5275 	clear_bit(__E1000_RESETTING, &adapter->state);
5276 
5277 	return 0;
5278 }
5279 
5280 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5281 			   int cmd)
5282 {
5283 	struct e1000_adapter *adapter = netdev_priv(netdev);
5284 	struct mii_ioctl_data *data = if_mii(ifr);
5285 
5286 	if (adapter->hw.phy.media_type != e1000_media_type_copper)
5287 		return -EOPNOTSUPP;
5288 
5289 	switch (cmd) {
5290 	case SIOCGMIIPHY:
5291 		data->phy_id = adapter->hw.phy.addr;
5292 		break;
5293 	case SIOCGMIIREG:
5294 		e1000_phy_read_status(adapter);
5295 
5296 		switch (data->reg_num & 0x1F) {
5297 		case MII_BMCR:
5298 			data->val_out = adapter->phy_regs.bmcr;
5299 			break;
5300 		case MII_BMSR:
5301 			data->val_out = adapter->phy_regs.bmsr;
5302 			break;
5303 		case MII_PHYSID1:
5304 			data->val_out = (adapter->hw.phy.id >> 16);
5305 			break;
5306 		case MII_PHYSID2:
5307 			data->val_out = (adapter->hw.phy.id & 0xFFFF);
5308 			break;
5309 		case MII_ADVERTISE:
5310 			data->val_out = adapter->phy_regs.advertise;
5311 			break;
5312 		case MII_LPA:
5313 			data->val_out = adapter->phy_regs.lpa;
5314 			break;
5315 		case MII_EXPANSION:
5316 			data->val_out = adapter->phy_regs.expansion;
5317 			break;
5318 		case MII_CTRL1000:
5319 			data->val_out = adapter->phy_regs.ctrl1000;
5320 			break;
5321 		case MII_STAT1000:
5322 			data->val_out = adapter->phy_regs.stat1000;
5323 			break;
5324 		case MII_ESTATUS:
5325 			data->val_out = adapter->phy_regs.estatus;
5326 			break;
5327 		default:
5328 			return -EIO;
5329 		}
5330 		break;
5331 	case SIOCSMIIREG:
5332 	default:
5333 		return -EOPNOTSUPP;
5334 	}
5335 	return 0;
5336 }
5337 
5338 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5339 {
5340 	switch (cmd) {
5341 	case SIOCGMIIPHY:
5342 	case SIOCGMIIREG:
5343 	case SIOCSMIIREG:
5344 		return e1000_mii_ioctl(netdev, ifr, cmd);
5345 	default:
5346 		return -EOPNOTSUPP;
5347 	}
5348 }
5349 
5350 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5351 {
5352 	struct e1000_hw *hw = &adapter->hw;
5353 	u32 i, mac_reg;
5354 	u16 phy_reg, wuc_enable;
5355 	int retval = 0;
5356 
5357 	/* copy MAC RARs to PHY RARs */
5358 	e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5359 
5360 	retval = hw->phy.ops.acquire(hw);
5361 	if (retval) {
5362 		e_err("Could not acquire PHY\n");
5363 		return retval;
5364 	}
5365 
5366 	/* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
5367 	retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5368 	if (retval)
5369 		goto release;
5370 
5371 	/* copy MAC MTA to PHY MTA - only needed for pchlan */
5372 	for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5373 		mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5374 		hw->phy.ops.write_reg_page(hw, BM_MTA(i),
5375 					   (u16)(mac_reg & 0xFFFF));
5376 		hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
5377 					   (u16)((mac_reg >> 16) & 0xFFFF));
5378 	}
5379 
5380 	/* configure PHY Rx Control register */
5381 	hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
5382 	mac_reg = er32(RCTL);
5383 	if (mac_reg & E1000_RCTL_UPE)
5384 		phy_reg |= BM_RCTL_UPE;
5385 	if (mac_reg & E1000_RCTL_MPE)
5386 		phy_reg |= BM_RCTL_MPE;
5387 	phy_reg &= ~(BM_RCTL_MO_MASK);
5388 	if (mac_reg & E1000_RCTL_MO_3)
5389 		phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5390 				<< BM_RCTL_MO_SHIFT);
5391 	if (mac_reg & E1000_RCTL_BAM)
5392 		phy_reg |= BM_RCTL_BAM;
5393 	if (mac_reg & E1000_RCTL_PMCF)
5394 		phy_reg |= BM_RCTL_PMCF;
5395 	mac_reg = er32(CTRL);
5396 	if (mac_reg & E1000_CTRL_RFCE)
5397 		phy_reg |= BM_RCTL_RFCE;
5398 	hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
5399 
5400 	/* enable PHY wakeup in MAC register */
5401 	ew32(WUFC, wufc);
5402 	ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5403 
5404 	/* configure and enable PHY wakeup in PHY registers */
5405 	hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
5406 	hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5407 
5408 	/* activate PHY wakeup */
5409 	wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5410 	retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5411 	if (retval)
5412 		e_err("Could not set PHY Host Wakeup bit\n");
5413 release:
5414 	hw->phy.ops.release(hw);
5415 
5416 	return retval;
5417 }
5418 
5419 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5420 			    bool runtime)
5421 {
5422 	struct net_device *netdev = pci_get_drvdata(pdev);
5423 	struct e1000_adapter *adapter = netdev_priv(netdev);
5424 	struct e1000_hw *hw = &adapter->hw;
5425 	u32 ctrl, ctrl_ext, rctl, status;
5426 	/* Runtime suspend should only enable wakeup for link changes */
5427 	u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5428 	int retval = 0;
5429 
5430 	netif_device_detach(netdev);
5431 
5432 	if (netif_running(netdev)) {
5433 		int count = E1000_CHECK_RESET_COUNT;
5434 
5435 		while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
5436 			usleep_range(10000, 20000);
5437 
5438 		WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5439 		e1000e_down(adapter);
5440 		e1000_free_irq(adapter);
5441 	}
5442 	e1000e_reset_interrupt_capability(adapter);
5443 
5444 	retval = pci_save_state(pdev);
5445 	if (retval)
5446 		return retval;
5447 
5448 	status = er32(STATUS);
5449 	if (status & E1000_STATUS_LU)
5450 		wufc &= ~E1000_WUFC_LNKC;
5451 
5452 	if (wufc) {
5453 		e1000_setup_rctl(adapter);
5454 		e1000e_set_rx_mode(netdev);
5455 
5456 		/* turn on all-multi mode if wake on multicast is enabled */
5457 		if (wufc & E1000_WUFC_MC) {
5458 			rctl = er32(RCTL);
5459 			rctl |= E1000_RCTL_MPE;
5460 			ew32(RCTL, rctl);
5461 		}
5462 
5463 		ctrl = er32(CTRL);
5464 		/* advertise wake from D3Cold */
5465 		#define E1000_CTRL_ADVD3WUC 0x00100000
5466 		/* phy power management enable */
5467 		#define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5468 		ctrl |= E1000_CTRL_ADVD3WUC;
5469 		if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5470 			ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5471 		ew32(CTRL, ctrl);
5472 
5473 		if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5474 		    adapter->hw.phy.media_type ==
5475 		    e1000_media_type_internal_serdes) {
5476 			/* keep the laser running in D3 */
5477 			ctrl_ext = er32(CTRL_EXT);
5478 			ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5479 			ew32(CTRL_EXT, ctrl_ext);
5480 		}
5481 
5482 		if (adapter->flags & FLAG_IS_ICH)
5483 			e1000_suspend_workarounds_ich8lan(&adapter->hw);
5484 
5485 		/* Allow time for pending master requests to run */
5486 		e1000e_disable_pcie_master(&adapter->hw);
5487 
5488 		if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5489 			/* enable wakeup by the PHY */
5490 			retval = e1000_init_phy_wakeup(adapter, wufc);
5491 			if (retval)
5492 				return retval;
5493 		} else {
5494 			/* enable wakeup by the MAC */
5495 			ew32(WUFC, wufc);
5496 			ew32(WUC, E1000_WUC_PME_EN);
5497 		}
5498 	} else {
5499 		ew32(WUC, 0);
5500 		ew32(WUFC, 0);
5501 	}
5502 
5503 	*enable_wake = !!wufc;
5504 
5505 	/* make sure adapter isn't asleep if manageability is enabled */
5506 	if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5507 	    (hw->mac.ops.check_mng_mode(hw)))
5508 		*enable_wake = true;
5509 
5510 	if (adapter->hw.phy.type == e1000_phy_igp_3)
5511 		e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5512 
5513 	/*
5514 	 * Release control of h/w to f/w.  If f/w is AMT enabled, this
5515 	 * would have already happened in close and is redundant.
5516 	 */
5517 	e1000e_release_hw_control(adapter);
5518 
5519 	pci_disable_device(pdev);
5520 
5521 	return 0;
5522 }
5523 
5524 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5525 {
5526 	if (sleep && wake) {
5527 		pci_prepare_to_sleep(pdev);
5528 		return;
5529 	}
5530 
5531 	pci_wake_from_d3(pdev, wake);
5532 	pci_set_power_state(pdev, PCI_D3hot);
5533 }
5534 
5535 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5536                                     bool wake)
5537 {
5538 	struct net_device *netdev = pci_get_drvdata(pdev);
5539 	struct e1000_adapter *adapter = netdev_priv(netdev);
5540 
5541 	/*
5542 	 * The pci-e switch on some quad port adapters will report a
5543 	 * correctable error when the MAC transitions from D0 to D3.  To
5544 	 * prevent this we need to mask off the correctable errors on the
5545 	 * downstream port of the pci-e switch.
5546 	 */
5547 	if (adapter->flags & FLAG_IS_QUAD_PORT) {
5548 		struct pci_dev *us_dev = pdev->bus->self;
5549 		int pos = pci_pcie_cap(us_dev);
5550 		u16 devctl;
5551 
5552 		pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5553 		pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5554 		                      (devctl & ~PCI_EXP_DEVCTL_CERE));
5555 
5556 		e1000_power_off(pdev, sleep, wake);
5557 
5558 		pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5559 	} else {
5560 		e1000_power_off(pdev, sleep, wake);
5561 	}
5562 }
5563 
5564 #ifdef CONFIG_PCIEASPM
5565 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5566 {
5567 	pci_disable_link_state_locked(pdev, state);
5568 }
5569 #else
5570 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5571 {
5572 	int pos;
5573 	u16 reg16;
5574 
5575 	/*
5576 	 * Both device and parent should have the same ASPM setting.
5577 	 * Disable ASPM in downstream component first and then upstream.
5578 	 */
5579 	pos = pci_pcie_cap(pdev);
5580 	pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5581 	reg16 &= ~state;
5582 	pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5583 
5584 	if (!pdev->bus->self)
5585 		return;
5586 
5587 	pos = pci_pcie_cap(pdev->bus->self);
5588 	pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5589 	reg16 &= ~state;
5590 	pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5591 }
5592 #endif
5593 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5594 {
5595 	dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5596 		 (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5597 		 (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5598 
5599 	__e1000e_disable_aspm(pdev, state);
5600 }
5601 
5602 #ifdef CONFIG_PM
5603 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5604 {
5605 	return !!adapter->tx_ring->buffer_info;
5606 }
5607 
5608 static int __e1000_resume(struct pci_dev *pdev)
5609 {
5610 	struct net_device *netdev = pci_get_drvdata(pdev);
5611 	struct e1000_adapter *adapter = netdev_priv(netdev);
5612 	struct e1000_hw *hw = &adapter->hw;
5613 	u16 aspm_disable_flag = 0;
5614 	u32 err;
5615 
5616 	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5617 		aspm_disable_flag = PCIE_LINK_STATE_L0S;
5618 	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5619 		aspm_disable_flag |= PCIE_LINK_STATE_L1;
5620 	if (aspm_disable_flag)
5621 		e1000e_disable_aspm(pdev, aspm_disable_flag);
5622 
5623 	pci_set_power_state(pdev, PCI_D0);
5624 	pci_restore_state(pdev);
5625 	pci_save_state(pdev);
5626 
5627 	e1000e_set_interrupt_capability(adapter);
5628 	if (netif_running(netdev)) {
5629 		err = e1000_request_irq(adapter);
5630 		if (err)
5631 			return err;
5632 	}
5633 
5634 	if (hw->mac.type >= e1000_pch2lan)
5635 		e1000_resume_workarounds_pchlan(&adapter->hw);
5636 
5637 	e1000e_power_up_phy(adapter);
5638 
5639 	/* report the system wakeup cause from S3/S4 */
5640 	if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5641 		u16 phy_data;
5642 
5643 		e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5644 		if (phy_data) {
5645 			e_info("PHY Wakeup cause - %s\n",
5646 				phy_data & E1000_WUS_EX ? "Unicast Packet" :
5647 				phy_data & E1000_WUS_MC ? "Multicast Packet" :
5648 				phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5649 				phy_data & E1000_WUS_MAG ? "Magic Packet" :
5650 				phy_data & E1000_WUS_LNKC ?
5651 				"Link Status Change" : "other");
5652 		}
5653 		e1e_wphy(&adapter->hw, BM_WUS, ~0);
5654 	} else {
5655 		u32 wus = er32(WUS);
5656 		if (wus) {
5657 			e_info("MAC Wakeup cause - %s\n",
5658 				wus & E1000_WUS_EX ? "Unicast Packet" :
5659 				wus & E1000_WUS_MC ? "Multicast Packet" :
5660 				wus & E1000_WUS_BC ? "Broadcast Packet" :
5661 				wus & E1000_WUS_MAG ? "Magic Packet" :
5662 				wus & E1000_WUS_LNKC ? "Link Status Change" :
5663 				"other");
5664 		}
5665 		ew32(WUS, ~0);
5666 	}
5667 
5668 	e1000e_reset(adapter);
5669 
5670 	e1000_init_manageability_pt(adapter);
5671 
5672 	if (netif_running(netdev))
5673 		e1000e_up(adapter);
5674 
5675 	netif_device_attach(netdev);
5676 
5677 	/*
5678 	 * If the controller has AMT, do not set DRV_LOAD until the interface
5679 	 * is up.  For all other cases, let the f/w know that the h/w is now
5680 	 * under the control of the driver.
5681 	 */
5682 	if (!(adapter->flags & FLAG_HAS_AMT))
5683 		e1000e_get_hw_control(adapter);
5684 
5685 	return 0;
5686 }
5687 
5688 #ifdef CONFIG_PM_SLEEP
5689 static int e1000_suspend(struct device *dev)
5690 {
5691 	struct pci_dev *pdev = to_pci_dev(dev);
5692 	int retval;
5693 	bool wake;
5694 
5695 	retval = __e1000_shutdown(pdev, &wake, false);
5696 	if (!retval)
5697 		e1000_complete_shutdown(pdev, true, wake);
5698 
5699 	return retval;
5700 }
5701 
5702 static int e1000_resume(struct device *dev)
5703 {
5704 	struct pci_dev *pdev = to_pci_dev(dev);
5705 	struct net_device *netdev = pci_get_drvdata(pdev);
5706 	struct e1000_adapter *adapter = netdev_priv(netdev);
5707 
5708 	if (e1000e_pm_ready(adapter))
5709 		adapter->idle_check = true;
5710 
5711 	return __e1000_resume(pdev);
5712 }
5713 #endif /* CONFIG_PM_SLEEP */
5714 
5715 #ifdef CONFIG_PM_RUNTIME
5716 static int e1000_runtime_suspend(struct device *dev)
5717 {
5718 	struct pci_dev *pdev = to_pci_dev(dev);
5719 	struct net_device *netdev = pci_get_drvdata(pdev);
5720 	struct e1000_adapter *adapter = netdev_priv(netdev);
5721 
5722 	if (e1000e_pm_ready(adapter)) {
5723 		bool wake;
5724 
5725 		__e1000_shutdown(pdev, &wake, true);
5726 	}
5727 
5728 	return 0;
5729 }
5730 
5731 static int e1000_idle(struct device *dev)
5732 {
5733 	struct pci_dev *pdev = to_pci_dev(dev);
5734 	struct net_device *netdev = pci_get_drvdata(pdev);
5735 	struct e1000_adapter *adapter = netdev_priv(netdev);
5736 
5737 	if (!e1000e_pm_ready(adapter))
5738 		return 0;
5739 
5740 	if (adapter->idle_check) {
5741 		adapter->idle_check = false;
5742 		if (!e1000e_has_link(adapter))
5743 			pm_schedule_suspend(dev, MSEC_PER_SEC);
5744 	}
5745 
5746 	return -EBUSY;
5747 }
5748 
5749 static int e1000_runtime_resume(struct device *dev)
5750 {
5751 	struct pci_dev *pdev = to_pci_dev(dev);
5752 	struct net_device *netdev = pci_get_drvdata(pdev);
5753 	struct e1000_adapter *adapter = netdev_priv(netdev);
5754 
5755 	if (!e1000e_pm_ready(adapter))
5756 		return 0;
5757 
5758 	adapter->idle_check = !dev->power.runtime_auto;
5759 	return __e1000_resume(pdev);
5760 }
5761 #endif /* CONFIG_PM_RUNTIME */
5762 #endif /* CONFIG_PM */
5763 
5764 static void e1000_shutdown(struct pci_dev *pdev)
5765 {
5766 	bool wake = false;
5767 
5768 	__e1000_shutdown(pdev, &wake, false);
5769 
5770 	if (system_state == SYSTEM_POWER_OFF)
5771 		e1000_complete_shutdown(pdev, false, wake);
5772 }
5773 
5774 #ifdef CONFIG_NET_POLL_CONTROLLER
5775 
5776 static irqreturn_t e1000_intr_msix(int irq, void *data)
5777 {
5778 	struct net_device *netdev = data;
5779 	struct e1000_adapter *adapter = netdev_priv(netdev);
5780 
5781 	if (adapter->msix_entries) {
5782 		int vector, msix_irq;
5783 
5784 		vector = 0;
5785 		msix_irq = adapter->msix_entries[vector].vector;
5786 		disable_irq(msix_irq);
5787 		e1000_intr_msix_rx(msix_irq, netdev);
5788 		enable_irq(msix_irq);
5789 
5790 		vector++;
5791 		msix_irq = adapter->msix_entries[vector].vector;
5792 		disable_irq(msix_irq);
5793 		e1000_intr_msix_tx(msix_irq, netdev);
5794 		enable_irq(msix_irq);
5795 
5796 		vector++;
5797 		msix_irq = adapter->msix_entries[vector].vector;
5798 		disable_irq(msix_irq);
5799 		e1000_msix_other(msix_irq, netdev);
5800 		enable_irq(msix_irq);
5801 	}
5802 
5803 	return IRQ_HANDLED;
5804 }
5805 
5806 /*
5807  * Polling 'interrupt' - used by things like netconsole to send skbs
5808  * without having to re-enable interrupts. It's not called while
5809  * the interrupt routine is executing.
5810  */
5811 static void e1000_netpoll(struct net_device *netdev)
5812 {
5813 	struct e1000_adapter *adapter = netdev_priv(netdev);
5814 
5815 	switch (adapter->int_mode) {
5816 	case E1000E_INT_MODE_MSIX:
5817 		e1000_intr_msix(adapter->pdev->irq, netdev);
5818 		break;
5819 	case E1000E_INT_MODE_MSI:
5820 		disable_irq(adapter->pdev->irq);
5821 		e1000_intr_msi(adapter->pdev->irq, netdev);
5822 		enable_irq(adapter->pdev->irq);
5823 		break;
5824 	default: /* E1000E_INT_MODE_LEGACY */
5825 		disable_irq(adapter->pdev->irq);
5826 		e1000_intr(adapter->pdev->irq, netdev);
5827 		enable_irq(adapter->pdev->irq);
5828 		break;
5829 	}
5830 }
5831 #endif
5832 
5833 /**
5834  * e1000_io_error_detected - called when PCI error is detected
5835  * @pdev: Pointer to PCI device
5836  * @state: The current pci connection state
5837  *
5838  * This function is called after a PCI bus error affecting
5839  * this device has been detected.
5840  */
5841 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5842 						pci_channel_state_t state)
5843 {
5844 	struct net_device *netdev = pci_get_drvdata(pdev);
5845 	struct e1000_adapter *adapter = netdev_priv(netdev);
5846 
5847 	netif_device_detach(netdev);
5848 
5849 	if (state == pci_channel_io_perm_failure)
5850 		return PCI_ERS_RESULT_DISCONNECT;
5851 
5852 	if (netif_running(netdev))
5853 		e1000e_down(adapter);
5854 	pci_disable_device(pdev);
5855 
5856 	/* Request a slot slot reset. */
5857 	return PCI_ERS_RESULT_NEED_RESET;
5858 }
5859 
5860 /**
5861  * e1000_io_slot_reset - called after the pci bus has been reset.
5862  * @pdev: Pointer to PCI device
5863  *
5864  * Restart the card from scratch, as if from a cold-boot. Implementation
5865  * resembles the first-half of the e1000_resume routine.
5866  */
5867 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5868 {
5869 	struct net_device *netdev = pci_get_drvdata(pdev);
5870 	struct e1000_adapter *adapter = netdev_priv(netdev);
5871 	struct e1000_hw *hw = &adapter->hw;
5872 	u16 aspm_disable_flag = 0;
5873 	int err;
5874 	pci_ers_result_t result;
5875 
5876 	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5877 		aspm_disable_flag = PCIE_LINK_STATE_L0S;
5878 	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5879 		aspm_disable_flag |= PCIE_LINK_STATE_L1;
5880 	if (aspm_disable_flag)
5881 		e1000e_disable_aspm(pdev, aspm_disable_flag);
5882 
5883 	err = pci_enable_device_mem(pdev);
5884 	if (err) {
5885 		dev_err(&pdev->dev,
5886 			"Cannot re-enable PCI device after reset.\n");
5887 		result = PCI_ERS_RESULT_DISCONNECT;
5888 	} else {
5889 		pci_set_master(pdev);
5890 		pdev->state_saved = true;
5891 		pci_restore_state(pdev);
5892 
5893 		pci_enable_wake(pdev, PCI_D3hot, 0);
5894 		pci_enable_wake(pdev, PCI_D3cold, 0);
5895 
5896 		e1000e_reset(adapter);
5897 		ew32(WUS, ~0);
5898 		result = PCI_ERS_RESULT_RECOVERED;
5899 	}
5900 
5901 	pci_cleanup_aer_uncorrect_error_status(pdev);
5902 
5903 	return result;
5904 }
5905 
5906 /**
5907  * e1000_io_resume - called when traffic can start flowing again.
5908  * @pdev: Pointer to PCI device
5909  *
5910  * This callback is called when the error recovery driver tells us that
5911  * its OK to resume normal operation. Implementation resembles the
5912  * second-half of the e1000_resume routine.
5913  */
5914 static void e1000_io_resume(struct pci_dev *pdev)
5915 {
5916 	struct net_device *netdev = pci_get_drvdata(pdev);
5917 	struct e1000_adapter *adapter = netdev_priv(netdev);
5918 
5919 	e1000_init_manageability_pt(adapter);
5920 
5921 	if (netif_running(netdev)) {
5922 		if (e1000e_up(adapter)) {
5923 			dev_err(&pdev->dev,
5924 				"can't bring device back up after reset\n");
5925 			return;
5926 		}
5927 	}
5928 
5929 	netif_device_attach(netdev);
5930 
5931 	/*
5932 	 * If the controller has AMT, do not set DRV_LOAD until the interface
5933 	 * is up.  For all other cases, let the f/w know that the h/w is now
5934 	 * under the control of the driver.
5935 	 */
5936 	if (!(adapter->flags & FLAG_HAS_AMT))
5937 		e1000e_get_hw_control(adapter);
5938 
5939 }
5940 
5941 static void e1000_print_device_info(struct e1000_adapter *adapter)
5942 {
5943 	struct e1000_hw *hw = &adapter->hw;
5944 	struct net_device *netdev = adapter->netdev;
5945 	u32 ret_val;
5946 	u8 pba_str[E1000_PBANUM_LENGTH];
5947 
5948 	/* print bus type/speed/width info */
5949 	e_info("(PCI Express:2.5GT/s:%s) %pM\n",
5950 	       /* bus width */
5951 	       ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5952 	        "Width x1"),
5953 	       /* MAC address */
5954 	       netdev->dev_addr);
5955 	e_info("Intel(R) PRO/%s Network Connection\n",
5956 	       (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5957 	ret_val = e1000_read_pba_string_generic(hw, pba_str,
5958 						E1000_PBANUM_LENGTH);
5959 	if (ret_val)
5960 		strlcpy((char *)pba_str, "Unknown", sizeof(pba_str));
5961 	e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5962 	       hw->mac.type, hw->phy.type, pba_str);
5963 }
5964 
5965 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5966 {
5967 	struct e1000_hw *hw = &adapter->hw;
5968 	int ret_val;
5969 	u16 buf = 0;
5970 
5971 	if (hw->mac.type != e1000_82573)
5972 		return;
5973 
5974 	ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5975 	le16_to_cpus(&buf);
5976 	if (!ret_val && (!(buf & (1 << 0)))) {
5977 		/* Deep Smart Power Down (DSPD) */
5978 		dev_warn(&adapter->pdev->dev,
5979 			 "Warning: detected DSPD enabled in EEPROM\n");
5980 	}
5981 }
5982 
5983 static int e1000_set_features(struct net_device *netdev,
5984 			      netdev_features_t features)
5985 {
5986 	struct e1000_adapter *adapter = netdev_priv(netdev);
5987 	netdev_features_t changed = features ^ netdev->features;
5988 
5989 	if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
5990 		adapter->flags |= FLAG_TSO_FORCE;
5991 
5992 	if (!(changed & (NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX |
5993 			 NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_RXFCS |
5994 			 NETIF_F_RXALL)))
5995 		return 0;
5996 
5997 	if (changed & NETIF_F_RXFCS) {
5998 		if (features & NETIF_F_RXFCS) {
5999 			adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6000 		} else {
6001 			/* We need to take it back to defaults, which might mean
6002 			 * stripping is still disabled at the adapter level.
6003 			 */
6004 			if (adapter->flags2 & FLAG2_DFLT_CRC_STRIPPING)
6005 				adapter->flags2 |= FLAG2_CRC_STRIPPING;
6006 			else
6007 				adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6008 		}
6009 	}
6010 
6011 	netdev->features = features;
6012 
6013 	if (netif_running(netdev))
6014 		e1000e_reinit_locked(adapter);
6015 	else
6016 		e1000e_reset(adapter);
6017 
6018 	return 0;
6019 }
6020 
6021 static const struct net_device_ops e1000e_netdev_ops = {
6022 	.ndo_open		= e1000_open,
6023 	.ndo_stop		= e1000_close,
6024 	.ndo_start_xmit		= e1000_xmit_frame,
6025 	.ndo_get_stats64	= e1000e_get_stats64,
6026 	.ndo_set_rx_mode	= e1000e_set_rx_mode,
6027 	.ndo_set_mac_address	= e1000_set_mac,
6028 	.ndo_change_mtu		= e1000_change_mtu,
6029 	.ndo_do_ioctl		= e1000_ioctl,
6030 	.ndo_tx_timeout		= e1000_tx_timeout,
6031 	.ndo_validate_addr	= eth_validate_addr,
6032 
6033 	.ndo_vlan_rx_add_vid	= e1000_vlan_rx_add_vid,
6034 	.ndo_vlan_rx_kill_vid	= e1000_vlan_rx_kill_vid,
6035 #ifdef CONFIG_NET_POLL_CONTROLLER
6036 	.ndo_poll_controller	= e1000_netpoll,
6037 #endif
6038 	.ndo_set_features = e1000_set_features,
6039 };
6040 
6041 /**
6042  * e1000_probe - Device Initialization Routine
6043  * @pdev: PCI device information struct
6044  * @ent: entry in e1000_pci_tbl
6045  *
6046  * Returns 0 on success, negative on failure
6047  *
6048  * e1000_probe initializes an adapter identified by a pci_dev structure.
6049  * The OS initialization, configuring of the adapter private structure,
6050  * and a hardware reset occur.
6051  **/
6052 static int __devinit e1000_probe(struct pci_dev *pdev,
6053 				 const struct pci_device_id *ent)
6054 {
6055 	struct net_device *netdev;
6056 	struct e1000_adapter *adapter;
6057 	struct e1000_hw *hw;
6058 	const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
6059 	resource_size_t mmio_start, mmio_len;
6060 	resource_size_t flash_start, flash_len;
6061 	static int cards_found;
6062 	u16 aspm_disable_flag = 0;
6063 	int i, err, pci_using_dac;
6064 	u16 eeprom_data = 0;
6065 	u16 eeprom_apme_mask = E1000_EEPROM_APME;
6066 
6067 	if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
6068 		aspm_disable_flag = PCIE_LINK_STATE_L0S;
6069 	if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
6070 		aspm_disable_flag |= PCIE_LINK_STATE_L1;
6071 	if (aspm_disable_flag)
6072 		e1000e_disable_aspm(pdev, aspm_disable_flag);
6073 
6074 	err = pci_enable_device_mem(pdev);
6075 	if (err)
6076 		return err;
6077 
6078 	pci_using_dac = 0;
6079 	err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
6080 	if (!err) {
6081 		err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
6082 		if (!err)
6083 			pci_using_dac = 1;
6084 	} else {
6085 		err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6086 		if (err) {
6087 			err = dma_set_coherent_mask(&pdev->dev,
6088 						    DMA_BIT_MASK(32));
6089 			if (err) {
6090 				dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
6091 				goto err_dma;
6092 			}
6093 		}
6094 	}
6095 
6096 	err = pci_request_selected_regions_exclusive(pdev,
6097 	                                  pci_select_bars(pdev, IORESOURCE_MEM),
6098 	                                  e1000e_driver_name);
6099 	if (err)
6100 		goto err_pci_reg;
6101 
6102 	/* AER (Advanced Error Reporting) hooks */
6103 	pci_enable_pcie_error_reporting(pdev);
6104 
6105 	pci_set_master(pdev);
6106 	/* PCI config space info */
6107 	err = pci_save_state(pdev);
6108 	if (err)
6109 		goto err_alloc_etherdev;
6110 
6111 	err = -ENOMEM;
6112 	netdev = alloc_etherdev(sizeof(struct e1000_adapter));
6113 	if (!netdev)
6114 		goto err_alloc_etherdev;
6115 
6116 	SET_NETDEV_DEV(netdev, &pdev->dev);
6117 
6118 	netdev->irq = pdev->irq;
6119 
6120 	pci_set_drvdata(pdev, netdev);
6121 	adapter = netdev_priv(netdev);
6122 	hw = &adapter->hw;
6123 	adapter->netdev = netdev;
6124 	adapter->pdev = pdev;
6125 	adapter->ei = ei;
6126 	adapter->pba = ei->pba;
6127 	adapter->flags = ei->flags;
6128 	adapter->flags2 = ei->flags2;
6129 	adapter->hw.adapter = adapter;
6130 	adapter->hw.mac.type = ei->mac;
6131 	adapter->max_hw_frame_size = ei->max_hw_frame_size;
6132 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
6133 
6134 	mmio_start = pci_resource_start(pdev, 0);
6135 	mmio_len = pci_resource_len(pdev, 0);
6136 
6137 	err = -EIO;
6138 	adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
6139 	if (!adapter->hw.hw_addr)
6140 		goto err_ioremap;
6141 
6142 	if ((adapter->flags & FLAG_HAS_FLASH) &&
6143 	    (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
6144 		flash_start = pci_resource_start(pdev, 1);
6145 		flash_len = pci_resource_len(pdev, 1);
6146 		adapter->hw.flash_address = ioremap(flash_start, flash_len);
6147 		if (!adapter->hw.flash_address)
6148 			goto err_flashmap;
6149 	}
6150 
6151 	/* construct the net_device struct */
6152 	netdev->netdev_ops		= &e1000e_netdev_ops;
6153 	e1000e_set_ethtool_ops(netdev);
6154 	netdev->watchdog_timeo		= 5 * HZ;
6155 	netif_napi_add(netdev, &adapter->napi, e1000e_poll, 64);
6156 	strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
6157 
6158 	netdev->mem_start = mmio_start;
6159 	netdev->mem_end = mmio_start + mmio_len;
6160 
6161 	adapter->bd_number = cards_found++;
6162 
6163 	e1000e_check_options(adapter);
6164 
6165 	/* setup adapter struct */
6166 	err = e1000_sw_init(adapter);
6167 	if (err)
6168 		goto err_sw_init;
6169 
6170 	memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
6171 	memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
6172 	memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
6173 
6174 	err = ei->get_variants(adapter);
6175 	if (err)
6176 		goto err_hw_init;
6177 
6178 	if ((adapter->flags & FLAG_IS_ICH) &&
6179 	    (adapter->flags & FLAG_READ_ONLY_NVM))
6180 		e1000e_write_protect_nvm_ich8lan(&adapter->hw);
6181 
6182 	hw->mac.ops.get_bus_info(&adapter->hw);
6183 
6184 	adapter->hw.phy.autoneg_wait_to_complete = 0;
6185 
6186 	/* Copper options */
6187 	if (adapter->hw.phy.media_type == e1000_media_type_copper) {
6188 		adapter->hw.phy.mdix = AUTO_ALL_MODES;
6189 		adapter->hw.phy.disable_polarity_correction = 0;
6190 		adapter->hw.phy.ms_type = e1000_ms_hw_default;
6191 	}
6192 
6193 	if (hw->phy.ops.check_reset_block && hw->phy.ops.check_reset_block(hw))
6194 		e_info("PHY reset is blocked due to SOL/IDER session.\n");
6195 
6196 	/* Set initial default active device features */
6197 	netdev->features = (NETIF_F_SG |
6198 			    NETIF_F_HW_VLAN_RX |
6199 			    NETIF_F_HW_VLAN_TX |
6200 			    NETIF_F_TSO |
6201 			    NETIF_F_TSO6 |
6202 			    NETIF_F_RXHASH |
6203 			    NETIF_F_RXCSUM |
6204 			    NETIF_F_HW_CSUM);
6205 
6206 	/* Set user-changeable features (subset of all device features) */
6207 	netdev->hw_features = netdev->features;
6208 	netdev->hw_features |= NETIF_F_RXFCS;
6209 	netdev->priv_flags |= IFF_SUPP_NOFCS;
6210 	netdev->hw_features |= NETIF_F_RXALL;
6211 
6212 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
6213 		netdev->features |= NETIF_F_HW_VLAN_FILTER;
6214 
6215 	netdev->vlan_features |= (NETIF_F_SG |
6216 				  NETIF_F_TSO |
6217 				  NETIF_F_TSO6 |
6218 				  NETIF_F_HW_CSUM);
6219 
6220 	netdev->priv_flags |= IFF_UNICAST_FLT;
6221 
6222 	if (pci_using_dac) {
6223 		netdev->features |= NETIF_F_HIGHDMA;
6224 		netdev->vlan_features |= NETIF_F_HIGHDMA;
6225 	}
6226 
6227 	if (e1000e_enable_mng_pass_thru(&adapter->hw))
6228 		adapter->flags |= FLAG_MNG_PT_ENABLED;
6229 
6230 	/*
6231 	 * before reading the NVM, reset the controller to
6232 	 * put the device in a known good starting state
6233 	 */
6234 	adapter->hw.mac.ops.reset_hw(&adapter->hw);
6235 
6236 	/*
6237 	 * systems with ASPM and others may see the checksum fail on the first
6238 	 * attempt. Let's give it a few tries
6239 	 */
6240 	for (i = 0;; i++) {
6241 		if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
6242 			break;
6243 		if (i == 2) {
6244 			e_err("The NVM Checksum Is Not Valid\n");
6245 			err = -EIO;
6246 			goto err_eeprom;
6247 		}
6248 	}
6249 
6250 	e1000_eeprom_checks(adapter);
6251 
6252 	/* copy the MAC address */
6253 	if (e1000e_read_mac_addr(&adapter->hw))
6254 		e_err("NVM Read Error while reading MAC address\n");
6255 
6256 	memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
6257 	memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
6258 
6259 	if (!is_valid_ether_addr(netdev->perm_addr)) {
6260 		e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
6261 		err = -EIO;
6262 		goto err_eeprom;
6263 	}
6264 
6265 	init_timer(&adapter->watchdog_timer);
6266 	adapter->watchdog_timer.function = e1000_watchdog;
6267 	adapter->watchdog_timer.data = (unsigned long) adapter;
6268 
6269 	init_timer(&adapter->phy_info_timer);
6270 	adapter->phy_info_timer.function = e1000_update_phy_info;
6271 	adapter->phy_info_timer.data = (unsigned long) adapter;
6272 
6273 	INIT_WORK(&adapter->reset_task, e1000_reset_task);
6274 	INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
6275 	INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
6276 	INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
6277 	INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
6278 
6279 	/* Initialize link parameters. User can change them with ethtool */
6280 	adapter->hw.mac.autoneg = 1;
6281 	adapter->fc_autoneg = true;
6282 	adapter->hw.fc.requested_mode = e1000_fc_default;
6283 	adapter->hw.fc.current_mode = e1000_fc_default;
6284 	adapter->hw.phy.autoneg_advertised = 0x2f;
6285 
6286 	/* ring size defaults */
6287 	adapter->rx_ring->count = 256;
6288 	adapter->tx_ring->count = 256;
6289 
6290 	/*
6291 	 * Initial Wake on LAN setting - If APM wake is enabled in
6292 	 * the EEPROM, enable the ACPI Magic Packet filter
6293 	 */
6294 	if (adapter->flags & FLAG_APME_IN_WUC) {
6295 		/* APME bit in EEPROM is mapped to WUC.APME */
6296 		eeprom_data = er32(WUC);
6297 		eeprom_apme_mask = E1000_WUC_APME;
6298 		if ((hw->mac.type > e1000_ich10lan) &&
6299 		    (eeprom_data & E1000_WUC_PHY_WAKE))
6300 			adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
6301 	} else if (adapter->flags & FLAG_APME_IN_CTRL3) {
6302 		if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
6303 		    (adapter->hw.bus.func == 1))
6304 			e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_B,
6305 				       1, &eeprom_data);
6306 		else
6307 			e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_A,
6308 				       1, &eeprom_data);
6309 	}
6310 
6311 	/* fetch WoL from EEPROM */
6312 	if (eeprom_data & eeprom_apme_mask)
6313 		adapter->eeprom_wol |= E1000_WUFC_MAG;
6314 
6315 	/*
6316 	 * now that we have the eeprom settings, apply the special cases
6317 	 * where the eeprom may be wrong or the board simply won't support
6318 	 * wake on lan on a particular port
6319 	 */
6320 	if (!(adapter->flags & FLAG_HAS_WOL))
6321 		adapter->eeprom_wol = 0;
6322 
6323 	/* initialize the wol settings based on the eeprom settings */
6324 	adapter->wol = adapter->eeprom_wol;
6325 	device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6326 
6327 	/* save off EEPROM version number */
6328 	e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6329 
6330 	/* reset the hardware with the new settings */
6331 	e1000e_reset(adapter);
6332 
6333 	/*
6334 	 * If the controller has AMT, do not set DRV_LOAD until the interface
6335 	 * is up.  For all other cases, let the f/w know that the h/w is now
6336 	 * under the control of the driver.
6337 	 */
6338 	if (!(adapter->flags & FLAG_HAS_AMT))
6339 		e1000e_get_hw_control(adapter);
6340 
6341 	strlcpy(netdev->name, "eth%d", sizeof(netdev->name));
6342 	err = register_netdev(netdev);
6343 	if (err)
6344 		goto err_register;
6345 
6346 	/* carrier off reporting is important to ethtool even BEFORE open */
6347 	netif_carrier_off(netdev);
6348 
6349 	e1000_print_device_info(adapter);
6350 
6351 	if (pci_dev_run_wake(pdev))
6352 		pm_runtime_put_noidle(&pdev->dev);
6353 
6354 	return 0;
6355 
6356 err_register:
6357 	if (!(adapter->flags & FLAG_HAS_AMT))
6358 		e1000e_release_hw_control(adapter);
6359 err_eeprom:
6360 	if (hw->phy.ops.check_reset_block && !hw->phy.ops.check_reset_block(hw))
6361 		e1000_phy_hw_reset(&adapter->hw);
6362 err_hw_init:
6363 	kfree(adapter->tx_ring);
6364 	kfree(adapter->rx_ring);
6365 err_sw_init:
6366 	if (adapter->hw.flash_address)
6367 		iounmap(adapter->hw.flash_address);
6368 	e1000e_reset_interrupt_capability(adapter);
6369 err_flashmap:
6370 	iounmap(adapter->hw.hw_addr);
6371 err_ioremap:
6372 	free_netdev(netdev);
6373 err_alloc_etherdev:
6374 	pci_release_selected_regions(pdev,
6375 	                             pci_select_bars(pdev, IORESOURCE_MEM));
6376 err_pci_reg:
6377 err_dma:
6378 	pci_disable_device(pdev);
6379 	return err;
6380 }
6381 
6382 /**
6383  * e1000_remove - Device Removal Routine
6384  * @pdev: PCI device information struct
6385  *
6386  * e1000_remove is called by the PCI subsystem to alert the driver
6387  * that it should release a PCI device.  The could be caused by a
6388  * Hot-Plug event, or because the driver is going to be removed from
6389  * memory.
6390  **/
6391 static void __devexit e1000_remove(struct pci_dev *pdev)
6392 {
6393 	struct net_device *netdev = pci_get_drvdata(pdev);
6394 	struct e1000_adapter *adapter = netdev_priv(netdev);
6395 	bool down = test_bit(__E1000_DOWN, &adapter->state);
6396 
6397 	/*
6398 	 * The timers may be rescheduled, so explicitly disable them
6399 	 * from being rescheduled.
6400 	 */
6401 	if (!down)
6402 		set_bit(__E1000_DOWN, &adapter->state);
6403 	del_timer_sync(&adapter->watchdog_timer);
6404 	del_timer_sync(&adapter->phy_info_timer);
6405 
6406 	cancel_work_sync(&adapter->reset_task);
6407 	cancel_work_sync(&adapter->watchdog_task);
6408 	cancel_work_sync(&adapter->downshift_task);
6409 	cancel_work_sync(&adapter->update_phy_task);
6410 	cancel_work_sync(&adapter->print_hang_task);
6411 
6412 	if (!(netdev->flags & IFF_UP))
6413 		e1000_power_down_phy(adapter);
6414 
6415 	/* Don't lie to e1000_close() down the road. */
6416 	if (!down)
6417 		clear_bit(__E1000_DOWN, &adapter->state);
6418 	unregister_netdev(netdev);
6419 
6420 	if (pci_dev_run_wake(pdev))
6421 		pm_runtime_get_noresume(&pdev->dev);
6422 
6423 	/*
6424 	 * Release control of h/w to f/w.  If f/w is AMT enabled, this
6425 	 * would have already happened in close and is redundant.
6426 	 */
6427 	e1000e_release_hw_control(adapter);
6428 
6429 	e1000e_reset_interrupt_capability(adapter);
6430 	kfree(adapter->tx_ring);
6431 	kfree(adapter->rx_ring);
6432 
6433 	iounmap(adapter->hw.hw_addr);
6434 	if (adapter->hw.flash_address)
6435 		iounmap(adapter->hw.flash_address);
6436 	pci_release_selected_regions(pdev,
6437 	                             pci_select_bars(pdev, IORESOURCE_MEM));
6438 
6439 	free_netdev(netdev);
6440 
6441 	/* AER disable */
6442 	pci_disable_pcie_error_reporting(pdev);
6443 
6444 	pci_disable_device(pdev);
6445 }
6446 
6447 /* PCI Error Recovery (ERS) */
6448 static struct pci_error_handlers e1000_err_handler = {
6449 	.error_detected = e1000_io_error_detected,
6450 	.slot_reset = e1000_io_slot_reset,
6451 	.resume = e1000_io_resume,
6452 };
6453 
6454 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6455 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6456 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6457 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6458 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6459 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6460 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6461 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6462 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6463 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6464 
6465 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6466 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6467 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6468 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6469 
6470 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6471 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6472 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6473 
6474 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6475 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6476 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6477 
6478 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6479 	  board_80003es2lan },
6480 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6481 	  board_80003es2lan },
6482 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6483 	  board_80003es2lan },
6484 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6485 	  board_80003es2lan },
6486 
6487 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6488 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6489 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6490 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6491 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6492 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6493 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6494 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6495 
6496 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6497 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6498 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6499 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6500 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6501 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6502 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6503 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6504 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6505 
6506 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6507 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6508 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6509 
6510 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6511 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6512 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6513 
6514 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6515 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6516 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6517 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6518 
6519 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6520 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6521 
6522 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_LM), board_pch_lpt },
6523 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_V), board_pch_lpt },
6524 
6525 	{ 0, 0, 0, 0, 0, 0, 0 }	/* terminate list */
6526 };
6527 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6528 
6529 #ifdef CONFIG_PM
6530 static const struct dev_pm_ops e1000_pm_ops = {
6531 	SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6532 	SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6533 				e1000_runtime_resume, e1000_idle)
6534 };
6535 #endif
6536 
6537 /* PCI Device API Driver */
6538 static struct pci_driver e1000_driver = {
6539 	.name     = e1000e_driver_name,
6540 	.id_table = e1000_pci_tbl,
6541 	.probe    = e1000_probe,
6542 	.remove   = __devexit_p(e1000_remove),
6543 #ifdef CONFIG_PM
6544 	.driver   = {
6545 		.pm = &e1000_pm_ops,
6546 	},
6547 #endif
6548 	.shutdown = e1000_shutdown,
6549 	.err_handler = &e1000_err_handler
6550 };
6551 
6552 /**
6553  * e1000_init_module - Driver Registration Routine
6554  *
6555  * e1000_init_module is the first routine called when the driver is
6556  * loaded. All it does is register with the PCI subsystem.
6557  **/
6558 static int __init e1000_init_module(void)
6559 {
6560 	int ret;
6561 	pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6562 		e1000e_driver_version);
6563 	pr_info("Copyright(c) 1999 - 2012 Intel Corporation.\n");
6564 	ret = pci_register_driver(&e1000_driver);
6565 
6566 	return ret;
6567 }
6568 module_init(e1000_init_module);
6569 
6570 /**
6571  * e1000_exit_module - Driver Exit Cleanup Routine
6572  *
6573  * e1000_exit_module is called just before the driver is removed
6574  * from memory.
6575  **/
6576 static void __exit e1000_exit_module(void)
6577 {
6578 	pci_unregister_driver(&e1000_driver);
6579 }
6580 module_exit(e1000_exit_module);
6581 
6582 
6583 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6584 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6585 MODULE_LICENSE("GPL");
6586 MODULE_VERSION(DRV_VERSION);
6587 
6588 /* netdev.c */
6589