xref: /linux/drivers/net/ethernet/intel/e1000e/netdev.c (revision 7ad24ea4bf620a32631d7b3069c3e30c078b0c3e)
1 /* Intel PRO/1000 Linux driver
2  * Copyright(c) 1999 - 2014 Intel Corporation.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms and conditions of the GNU General Public License,
6  * version 2, as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope it will be useful, but WITHOUT
9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
11  * more details.
12  *
13  * The full GNU General Public License is included in this distribution in
14  * the file called "COPYING".
15  *
16  * Contact Information:
17  * Linux NICS <linux.nics@intel.com>
18  * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
19  * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
20  */
21 
22 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
23 
24 #include <linux/module.h>
25 #include <linux/types.h>
26 #include <linux/init.h>
27 #include <linux/pci.h>
28 #include <linux/vmalloc.h>
29 #include <linux/pagemap.h>
30 #include <linux/delay.h>
31 #include <linux/netdevice.h>
32 #include <linux/interrupt.h>
33 #include <linux/tcp.h>
34 #include <linux/ipv6.h>
35 #include <linux/slab.h>
36 #include <net/checksum.h>
37 #include <net/ip6_checksum.h>
38 #include <linux/ethtool.h>
39 #include <linux/if_vlan.h>
40 #include <linux/cpu.h>
41 #include <linux/smp.h>
42 #include <linux/pm_qos.h>
43 #include <linux/pm_runtime.h>
44 #include <linux/aer.h>
45 #include <linux/prefetch.h>
46 
47 #include "e1000.h"
48 
49 #define DRV_EXTRAVERSION "-k"
50 
51 #define DRV_VERSION "2.3.2" DRV_EXTRAVERSION
52 char e1000e_driver_name[] = "e1000e";
53 const char e1000e_driver_version[] = DRV_VERSION;
54 
55 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
56 static int debug = -1;
57 module_param(debug, int, 0);
58 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
59 
60 static const struct e1000_info *e1000_info_tbl[] = {
61 	[board_82571]		= &e1000_82571_info,
62 	[board_82572]		= &e1000_82572_info,
63 	[board_82573]		= &e1000_82573_info,
64 	[board_82574]		= &e1000_82574_info,
65 	[board_82583]		= &e1000_82583_info,
66 	[board_80003es2lan]	= &e1000_es2_info,
67 	[board_ich8lan]		= &e1000_ich8_info,
68 	[board_ich9lan]		= &e1000_ich9_info,
69 	[board_ich10lan]	= &e1000_ich10_info,
70 	[board_pchlan]		= &e1000_pch_info,
71 	[board_pch2lan]		= &e1000_pch2_info,
72 	[board_pch_lpt]		= &e1000_pch_lpt_info,
73 };
74 
75 struct e1000_reg_info {
76 	u32 ofs;
77 	char *name;
78 };
79 
80 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
81 	/* General Registers */
82 	{E1000_CTRL, "CTRL"},
83 	{E1000_STATUS, "STATUS"},
84 	{E1000_CTRL_EXT, "CTRL_EXT"},
85 
86 	/* Interrupt Registers */
87 	{E1000_ICR, "ICR"},
88 
89 	/* Rx Registers */
90 	{E1000_RCTL, "RCTL"},
91 	{E1000_RDLEN(0), "RDLEN"},
92 	{E1000_RDH(0), "RDH"},
93 	{E1000_RDT(0), "RDT"},
94 	{E1000_RDTR, "RDTR"},
95 	{E1000_RXDCTL(0), "RXDCTL"},
96 	{E1000_ERT, "ERT"},
97 	{E1000_RDBAL(0), "RDBAL"},
98 	{E1000_RDBAH(0), "RDBAH"},
99 	{E1000_RDFH, "RDFH"},
100 	{E1000_RDFT, "RDFT"},
101 	{E1000_RDFHS, "RDFHS"},
102 	{E1000_RDFTS, "RDFTS"},
103 	{E1000_RDFPC, "RDFPC"},
104 
105 	/* Tx Registers */
106 	{E1000_TCTL, "TCTL"},
107 	{E1000_TDBAL(0), "TDBAL"},
108 	{E1000_TDBAH(0), "TDBAH"},
109 	{E1000_TDLEN(0), "TDLEN"},
110 	{E1000_TDH(0), "TDH"},
111 	{E1000_TDT(0), "TDT"},
112 	{E1000_TIDV, "TIDV"},
113 	{E1000_TXDCTL(0), "TXDCTL"},
114 	{E1000_TADV, "TADV"},
115 	{E1000_TARC(0), "TARC"},
116 	{E1000_TDFH, "TDFH"},
117 	{E1000_TDFT, "TDFT"},
118 	{E1000_TDFHS, "TDFHS"},
119 	{E1000_TDFTS, "TDFTS"},
120 	{E1000_TDFPC, "TDFPC"},
121 
122 	/* List Terminator */
123 	{0, NULL}
124 };
125 
126 /**
127  * e1000_regdump - register printout routine
128  * @hw: pointer to the HW structure
129  * @reginfo: pointer to the register info table
130  **/
131 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
132 {
133 	int n = 0;
134 	char rname[16];
135 	u32 regs[8];
136 
137 	switch (reginfo->ofs) {
138 	case E1000_RXDCTL(0):
139 		for (n = 0; n < 2; n++)
140 			regs[n] = __er32(hw, E1000_RXDCTL(n));
141 		break;
142 	case E1000_TXDCTL(0):
143 		for (n = 0; n < 2; n++)
144 			regs[n] = __er32(hw, E1000_TXDCTL(n));
145 		break;
146 	case E1000_TARC(0):
147 		for (n = 0; n < 2; n++)
148 			regs[n] = __er32(hw, E1000_TARC(n));
149 		break;
150 	default:
151 		pr_info("%-15s %08x\n",
152 			reginfo->name, __er32(hw, reginfo->ofs));
153 		return;
154 	}
155 
156 	snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
157 	pr_info("%-15s %08x %08x\n", rname, regs[0], regs[1]);
158 }
159 
160 static void e1000e_dump_ps_pages(struct e1000_adapter *adapter,
161 				 struct e1000_buffer *bi)
162 {
163 	int i;
164 	struct e1000_ps_page *ps_page;
165 
166 	for (i = 0; i < adapter->rx_ps_pages; i++) {
167 		ps_page = &bi->ps_pages[i];
168 
169 		if (ps_page->page) {
170 			pr_info("packet dump for ps_page %d:\n", i);
171 			print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
172 				       16, 1, page_address(ps_page->page),
173 				       PAGE_SIZE, true);
174 		}
175 	}
176 }
177 
178 /**
179  * e1000e_dump - Print registers, Tx-ring and Rx-ring
180  * @adapter: board private structure
181  **/
182 static void e1000e_dump(struct e1000_adapter *adapter)
183 {
184 	struct net_device *netdev = adapter->netdev;
185 	struct e1000_hw *hw = &adapter->hw;
186 	struct e1000_reg_info *reginfo;
187 	struct e1000_ring *tx_ring = adapter->tx_ring;
188 	struct e1000_tx_desc *tx_desc;
189 	struct my_u0 {
190 		__le64 a;
191 		__le64 b;
192 	} *u0;
193 	struct e1000_buffer *buffer_info;
194 	struct e1000_ring *rx_ring = adapter->rx_ring;
195 	union e1000_rx_desc_packet_split *rx_desc_ps;
196 	union e1000_rx_desc_extended *rx_desc;
197 	struct my_u1 {
198 		__le64 a;
199 		__le64 b;
200 		__le64 c;
201 		__le64 d;
202 	} *u1;
203 	u32 staterr;
204 	int i = 0;
205 
206 	if (!netif_msg_hw(adapter))
207 		return;
208 
209 	/* Print netdevice Info */
210 	if (netdev) {
211 		dev_info(&adapter->pdev->dev, "Net device Info\n");
212 		pr_info("Device Name     state            trans_start      last_rx\n");
213 		pr_info("%-15s %016lX %016lX %016lX\n", netdev->name,
214 			netdev->state, netdev->trans_start, netdev->last_rx);
215 	}
216 
217 	/* Print Registers */
218 	dev_info(&adapter->pdev->dev, "Register Dump\n");
219 	pr_info(" Register Name   Value\n");
220 	for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
221 	     reginfo->name; reginfo++) {
222 		e1000_regdump(hw, reginfo);
223 	}
224 
225 	/* Print Tx Ring Summary */
226 	if (!netdev || !netif_running(netdev))
227 		return;
228 
229 	dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
230 	pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
231 	buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
232 	pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
233 		0, tx_ring->next_to_use, tx_ring->next_to_clean,
234 		(unsigned long long)buffer_info->dma,
235 		buffer_info->length,
236 		buffer_info->next_to_watch,
237 		(unsigned long long)buffer_info->time_stamp);
238 
239 	/* Print Tx Ring */
240 	if (!netif_msg_tx_done(adapter))
241 		goto rx_ring_summary;
242 
243 	dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
244 
245 	/* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
246 	 *
247 	 * Legacy Transmit Descriptor
248 	 *   +--------------------------------------------------------------+
249 	 * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
250 	 *   +--------------------------------------------------------------+
251 	 * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
252 	 *   +--------------------------------------------------------------+
253 	 *   63       48 47        36 35    32 31     24 23    16 15        0
254 	 *
255 	 * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
256 	 *   63      48 47    40 39       32 31             16 15    8 7      0
257 	 *   +----------------------------------------------------------------+
258 	 * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
259 	 *   +----------------------------------------------------------------+
260 	 * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
261 	 *   +----------------------------------------------------------------+
262 	 *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
263 	 *
264 	 * Extended Data Descriptor (DTYP=0x1)
265 	 *   +----------------------------------------------------------------+
266 	 * 0 |                     Buffer Address [63:0]                      |
267 	 *   +----------------------------------------------------------------+
268 	 * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
269 	 *   +----------------------------------------------------------------+
270 	 *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
271 	 */
272 	pr_info("Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Legacy format\n");
273 	pr_info("Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Context format\n");
274 	pr_info("Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Data format\n");
275 	for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
276 		const char *next_desc;
277 		tx_desc = E1000_TX_DESC(*tx_ring, i);
278 		buffer_info = &tx_ring->buffer_info[i];
279 		u0 = (struct my_u0 *)tx_desc;
280 		if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
281 			next_desc = " NTC/U";
282 		else if (i == tx_ring->next_to_use)
283 			next_desc = " NTU";
284 		else if (i == tx_ring->next_to_clean)
285 			next_desc = " NTC";
286 		else
287 			next_desc = "";
288 		pr_info("T%c[0x%03X]    %016llX %016llX %016llX %04X  %3X %016llX %p%s\n",
289 			(!(le64_to_cpu(u0->b) & (1 << 29)) ? 'l' :
290 			 ((le64_to_cpu(u0->b) & (1 << 20)) ? 'd' : 'c')),
291 			i,
292 			(unsigned long long)le64_to_cpu(u0->a),
293 			(unsigned long long)le64_to_cpu(u0->b),
294 			(unsigned long long)buffer_info->dma,
295 			buffer_info->length, buffer_info->next_to_watch,
296 			(unsigned long long)buffer_info->time_stamp,
297 			buffer_info->skb, next_desc);
298 
299 		if (netif_msg_pktdata(adapter) && buffer_info->skb)
300 			print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
301 				       16, 1, buffer_info->skb->data,
302 				       buffer_info->skb->len, true);
303 	}
304 
305 	/* Print Rx Ring Summary */
306 rx_ring_summary:
307 	dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
308 	pr_info("Queue [NTU] [NTC]\n");
309 	pr_info(" %5d %5X %5X\n",
310 		0, rx_ring->next_to_use, rx_ring->next_to_clean);
311 
312 	/* Print Rx Ring */
313 	if (!netif_msg_rx_status(adapter))
314 		return;
315 
316 	dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
317 	switch (adapter->rx_ps_pages) {
318 	case 1:
319 	case 2:
320 	case 3:
321 		/* [Extended] Packet Split Receive Descriptor Format
322 		 *
323 		 *    +-----------------------------------------------------+
324 		 *  0 |                Buffer Address 0 [63:0]              |
325 		 *    +-----------------------------------------------------+
326 		 *  8 |                Buffer Address 1 [63:0]              |
327 		 *    +-----------------------------------------------------+
328 		 * 16 |                Buffer Address 2 [63:0]              |
329 		 *    +-----------------------------------------------------+
330 		 * 24 |                Buffer Address 3 [63:0]              |
331 		 *    +-----------------------------------------------------+
332 		 */
333 		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");
334 		/* [Extended] Receive Descriptor (Write-Back) Format
335 		 *
336 		 *   63       48 47    32 31     13 12    8 7    4 3        0
337 		 *   +------------------------------------------------------+
338 		 * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
339 		 *   | Checksum | Ident  |         | Queue |      |  Type   |
340 		 *   +------------------------------------------------------+
341 		 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
342 		 *   +------------------------------------------------------+
343 		 *   63       48 47    32 31            20 19               0
344 		 */
345 		pr_info("RWB[desc]      [ck ipid mrqhsh] [vl   l0 ee  es] [ l3  l2  l1 hs] [reserved      ] ---------------- [bi->skb] <-- Ext Rx Write-Back format\n");
346 		for (i = 0; i < rx_ring->count; i++) {
347 			const char *next_desc;
348 			buffer_info = &rx_ring->buffer_info[i];
349 			rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
350 			u1 = (struct my_u1 *)rx_desc_ps;
351 			staterr =
352 			    le32_to_cpu(rx_desc_ps->wb.middle.status_error);
353 
354 			if (i == rx_ring->next_to_use)
355 				next_desc = " NTU";
356 			else if (i == rx_ring->next_to_clean)
357 				next_desc = " NTC";
358 			else
359 				next_desc = "";
360 
361 			if (staterr & E1000_RXD_STAT_DD) {
362 				/* Descriptor Done */
363 				pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX ---------------- %p%s\n",
364 					"RWB", i,
365 					(unsigned long long)le64_to_cpu(u1->a),
366 					(unsigned long long)le64_to_cpu(u1->b),
367 					(unsigned long long)le64_to_cpu(u1->c),
368 					(unsigned long long)le64_to_cpu(u1->d),
369 					buffer_info->skb, next_desc);
370 			} else {
371 				pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX %016llX %p%s\n",
372 					"R  ", i,
373 					(unsigned long long)le64_to_cpu(u1->a),
374 					(unsigned long long)le64_to_cpu(u1->b),
375 					(unsigned long long)le64_to_cpu(u1->c),
376 					(unsigned long long)le64_to_cpu(u1->d),
377 					(unsigned long long)buffer_info->dma,
378 					buffer_info->skb, next_desc);
379 
380 				if (netif_msg_pktdata(adapter))
381 					e1000e_dump_ps_pages(adapter,
382 							     buffer_info);
383 			}
384 		}
385 		break;
386 	default:
387 	case 0:
388 		/* Extended Receive Descriptor (Read) Format
389 		 *
390 		 *   +-----------------------------------------------------+
391 		 * 0 |                Buffer Address [63:0]                |
392 		 *   +-----------------------------------------------------+
393 		 * 8 |                      Reserved                       |
394 		 *   +-----------------------------------------------------+
395 		 */
396 		pr_info("R  [desc]      [buf addr 63:0 ] [reserved 63:0 ] [bi->dma       ] [bi->skb] <-- Ext (Read) format\n");
397 		/* Extended Receive Descriptor (Write-Back) Format
398 		 *
399 		 *   63       48 47    32 31    24 23            4 3        0
400 		 *   +------------------------------------------------------+
401 		 *   |     RSS Hash      |        |               |         |
402 		 * 0 +-------------------+  Rsvd  |   Reserved    | MRQ RSS |
403 		 *   | Packet   | IP     |        |               |  Type   |
404 		 *   | Checksum | Ident  |        |               |         |
405 		 *   +------------------------------------------------------+
406 		 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
407 		 *   +------------------------------------------------------+
408 		 *   63       48 47    32 31            20 19               0
409 		 */
410 		pr_info("RWB[desc]      [cs ipid    mrq] [vt   ln xe  xs] [bi->skb] <-- Ext (Write-Back) format\n");
411 
412 		for (i = 0; i < rx_ring->count; i++) {
413 			const char *next_desc;
414 
415 			buffer_info = &rx_ring->buffer_info[i];
416 			rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
417 			u1 = (struct my_u1 *)rx_desc;
418 			staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
419 
420 			if (i == rx_ring->next_to_use)
421 				next_desc = " NTU";
422 			else if (i == rx_ring->next_to_clean)
423 				next_desc = " NTC";
424 			else
425 				next_desc = "";
426 
427 			if (staterr & E1000_RXD_STAT_DD) {
428 				/* Descriptor Done */
429 				pr_info("%s[0x%03X]     %016llX %016llX ---------------- %p%s\n",
430 					"RWB", i,
431 					(unsigned long long)le64_to_cpu(u1->a),
432 					(unsigned long long)le64_to_cpu(u1->b),
433 					buffer_info->skb, next_desc);
434 			} else {
435 				pr_info("%s[0x%03X]     %016llX %016llX %016llX %p%s\n",
436 					"R  ", i,
437 					(unsigned long long)le64_to_cpu(u1->a),
438 					(unsigned long long)le64_to_cpu(u1->b),
439 					(unsigned long long)buffer_info->dma,
440 					buffer_info->skb, next_desc);
441 
442 				if (netif_msg_pktdata(adapter) &&
443 				    buffer_info->skb)
444 					print_hex_dump(KERN_INFO, "",
445 						       DUMP_PREFIX_ADDRESS, 16,
446 						       1,
447 						       buffer_info->skb->data,
448 						       adapter->rx_buffer_len,
449 						       true);
450 			}
451 		}
452 	}
453 }
454 
455 /**
456  * e1000_desc_unused - calculate if we have unused descriptors
457  **/
458 static int e1000_desc_unused(struct e1000_ring *ring)
459 {
460 	if (ring->next_to_clean > ring->next_to_use)
461 		return ring->next_to_clean - ring->next_to_use - 1;
462 
463 	return ring->count + ring->next_to_clean - ring->next_to_use - 1;
464 }
465 
466 /**
467  * e1000e_systim_to_hwtstamp - convert system time value to hw time stamp
468  * @adapter: board private structure
469  * @hwtstamps: time stamp structure to update
470  * @systim: unsigned 64bit system time value.
471  *
472  * Convert the system time value stored in the RX/TXSTMP registers into a
473  * hwtstamp which can be used by the upper level time stamping functions.
474  *
475  * The 'systim_lock' spinlock is used to protect the consistency of the
476  * system time value. This is needed because reading the 64 bit time
477  * value involves reading two 32 bit registers. The first read latches the
478  * value.
479  **/
480 static void e1000e_systim_to_hwtstamp(struct e1000_adapter *adapter,
481 				      struct skb_shared_hwtstamps *hwtstamps,
482 				      u64 systim)
483 {
484 	u64 ns;
485 	unsigned long flags;
486 
487 	spin_lock_irqsave(&adapter->systim_lock, flags);
488 	ns = timecounter_cyc2time(&adapter->tc, systim);
489 	spin_unlock_irqrestore(&adapter->systim_lock, flags);
490 
491 	memset(hwtstamps, 0, sizeof(*hwtstamps));
492 	hwtstamps->hwtstamp = ns_to_ktime(ns);
493 }
494 
495 /**
496  * e1000e_rx_hwtstamp - utility function which checks for Rx time stamp
497  * @adapter: board private structure
498  * @status: descriptor extended error and status field
499  * @skb: particular skb to include time stamp
500  *
501  * If the time stamp is valid, convert it into the timecounter ns value
502  * and store that result into the shhwtstamps structure which is passed
503  * up the network stack.
504  **/
505 static void e1000e_rx_hwtstamp(struct e1000_adapter *adapter, u32 status,
506 			       struct sk_buff *skb)
507 {
508 	struct e1000_hw *hw = &adapter->hw;
509 	u64 rxstmp;
510 
511 	if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP) ||
512 	    !(status & E1000_RXDEXT_STATERR_TST) ||
513 	    !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID))
514 		return;
515 
516 	/* The Rx time stamp registers contain the time stamp.  No other
517 	 * received packet will be time stamped until the Rx time stamp
518 	 * registers are read.  Because only one packet can be time stamped
519 	 * at a time, the register values must belong to this packet and
520 	 * therefore none of the other additional attributes need to be
521 	 * compared.
522 	 */
523 	rxstmp = (u64)er32(RXSTMPL);
524 	rxstmp |= (u64)er32(RXSTMPH) << 32;
525 	e1000e_systim_to_hwtstamp(adapter, skb_hwtstamps(skb), rxstmp);
526 
527 	adapter->flags2 &= ~FLAG2_CHECK_RX_HWTSTAMP;
528 }
529 
530 /**
531  * e1000_receive_skb - helper function to handle Rx indications
532  * @adapter: board private structure
533  * @staterr: descriptor extended error and status field as written by hardware
534  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
535  * @skb: pointer to sk_buff to be indicated to stack
536  **/
537 static void e1000_receive_skb(struct e1000_adapter *adapter,
538 			      struct net_device *netdev, struct sk_buff *skb,
539 			      u32 staterr, __le16 vlan)
540 {
541 	u16 tag = le16_to_cpu(vlan);
542 
543 	e1000e_rx_hwtstamp(adapter, staterr, skb);
544 
545 	skb->protocol = eth_type_trans(skb, netdev);
546 
547 	if (staterr & E1000_RXD_STAT_VP)
548 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tag);
549 
550 	napi_gro_receive(&adapter->napi, skb);
551 }
552 
553 /**
554  * e1000_rx_checksum - Receive Checksum Offload
555  * @adapter: board private structure
556  * @status_err: receive descriptor status and error fields
557  * @csum: receive descriptor csum field
558  * @sk_buff: socket buffer with received data
559  **/
560 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
561 			      struct sk_buff *skb)
562 {
563 	u16 status = (u16)status_err;
564 	u8 errors = (u8)(status_err >> 24);
565 
566 	skb_checksum_none_assert(skb);
567 
568 	/* Rx checksum disabled */
569 	if (!(adapter->netdev->features & NETIF_F_RXCSUM))
570 		return;
571 
572 	/* Ignore Checksum bit is set */
573 	if (status & E1000_RXD_STAT_IXSM)
574 		return;
575 
576 	/* TCP/UDP checksum error bit or IP checksum error bit is set */
577 	if (errors & (E1000_RXD_ERR_TCPE | E1000_RXD_ERR_IPE)) {
578 		/* let the stack verify checksum errors */
579 		adapter->hw_csum_err++;
580 		return;
581 	}
582 
583 	/* TCP/UDP Checksum has not been calculated */
584 	if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
585 		return;
586 
587 	/* It must be a TCP or UDP packet with a valid checksum */
588 	skb->ip_summed = CHECKSUM_UNNECESSARY;
589 	adapter->hw_csum_good++;
590 }
591 
592 static void e1000e_update_rdt_wa(struct e1000_ring *rx_ring, unsigned int i)
593 {
594 	struct e1000_adapter *adapter = rx_ring->adapter;
595 	struct e1000_hw *hw = &adapter->hw;
596 	s32 ret_val = __ew32_prepare(hw);
597 
598 	writel(i, rx_ring->tail);
599 
600 	if (unlikely(!ret_val && (i != readl(rx_ring->tail)))) {
601 		u32 rctl = er32(RCTL);
602 
603 		ew32(RCTL, rctl & ~E1000_RCTL_EN);
604 		e_err("ME firmware caused invalid RDT - resetting\n");
605 		schedule_work(&adapter->reset_task);
606 	}
607 }
608 
609 static void e1000e_update_tdt_wa(struct e1000_ring *tx_ring, unsigned int i)
610 {
611 	struct e1000_adapter *adapter = tx_ring->adapter;
612 	struct e1000_hw *hw = &adapter->hw;
613 	s32 ret_val = __ew32_prepare(hw);
614 
615 	writel(i, tx_ring->tail);
616 
617 	if (unlikely(!ret_val && (i != readl(tx_ring->tail)))) {
618 		u32 tctl = er32(TCTL);
619 
620 		ew32(TCTL, tctl & ~E1000_TCTL_EN);
621 		e_err("ME firmware caused invalid TDT - resetting\n");
622 		schedule_work(&adapter->reset_task);
623 	}
624 }
625 
626 /**
627  * e1000_alloc_rx_buffers - Replace used receive buffers
628  * @rx_ring: Rx descriptor ring
629  **/
630 static void e1000_alloc_rx_buffers(struct e1000_ring *rx_ring,
631 				   int cleaned_count, gfp_t gfp)
632 {
633 	struct e1000_adapter *adapter = rx_ring->adapter;
634 	struct net_device *netdev = adapter->netdev;
635 	struct pci_dev *pdev = adapter->pdev;
636 	union e1000_rx_desc_extended *rx_desc;
637 	struct e1000_buffer *buffer_info;
638 	struct sk_buff *skb;
639 	unsigned int i;
640 	unsigned int bufsz = adapter->rx_buffer_len;
641 
642 	i = rx_ring->next_to_use;
643 	buffer_info = &rx_ring->buffer_info[i];
644 
645 	while (cleaned_count--) {
646 		skb = buffer_info->skb;
647 		if (skb) {
648 			skb_trim(skb, 0);
649 			goto map_skb;
650 		}
651 
652 		skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
653 		if (!skb) {
654 			/* Better luck next round */
655 			adapter->alloc_rx_buff_failed++;
656 			break;
657 		}
658 
659 		buffer_info->skb = skb;
660 map_skb:
661 		buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
662 						  adapter->rx_buffer_len,
663 						  DMA_FROM_DEVICE);
664 		if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
665 			dev_err(&pdev->dev, "Rx DMA map failed\n");
666 			adapter->rx_dma_failed++;
667 			break;
668 		}
669 
670 		rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
671 		rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
672 
673 		if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
674 			/* Force memory writes to complete before letting h/w
675 			 * know there are new descriptors to fetch.  (Only
676 			 * applicable for weak-ordered memory model archs,
677 			 * such as IA-64).
678 			 */
679 			wmb();
680 			if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
681 				e1000e_update_rdt_wa(rx_ring, i);
682 			else
683 				writel(i, rx_ring->tail);
684 		}
685 		i++;
686 		if (i == rx_ring->count)
687 			i = 0;
688 		buffer_info = &rx_ring->buffer_info[i];
689 	}
690 
691 	rx_ring->next_to_use = i;
692 }
693 
694 /**
695  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
696  * @rx_ring: Rx descriptor ring
697  **/
698 static void e1000_alloc_rx_buffers_ps(struct e1000_ring *rx_ring,
699 				      int cleaned_count, gfp_t gfp)
700 {
701 	struct e1000_adapter *adapter = rx_ring->adapter;
702 	struct net_device *netdev = adapter->netdev;
703 	struct pci_dev *pdev = adapter->pdev;
704 	union e1000_rx_desc_packet_split *rx_desc;
705 	struct e1000_buffer *buffer_info;
706 	struct e1000_ps_page *ps_page;
707 	struct sk_buff *skb;
708 	unsigned int i, j;
709 
710 	i = rx_ring->next_to_use;
711 	buffer_info = &rx_ring->buffer_info[i];
712 
713 	while (cleaned_count--) {
714 		rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
715 
716 		for (j = 0; j < PS_PAGE_BUFFERS; j++) {
717 			ps_page = &buffer_info->ps_pages[j];
718 			if (j >= adapter->rx_ps_pages) {
719 				/* all unused desc entries get hw null ptr */
720 				rx_desc->read.buffer_addr[j + 1] =
721 				    ~cpu_to_le64(0);
722 				continue;
723 			}
724 			if (!ps_page->page) {
725 				ps_page->page = alloc_page(gfp);
726 				if (!ps_page->page) {
727 					adapter->alloc_rx_buff_failed++;
728 					goto no_buffers;
729 				}
730 				ps_page->dma = dma_map_page(&pdev->dev,
731 							    ps_page->page,
732 							    0, PAGE_SIZE,
733 							    DMA_FROM_DEVICE);
734 				if (dma_mapping_error(&pdev->dev,
735 						      ps_page->dma)) {
736 					dev_err(&adapter->pdev->dev,
737 						"Rx DMA page map failed\n");
738 					adapter->rx_dma_failed++;
739 					goto no_buffers;
740 				}
741 			}
742 			/* Refresh the desc even if buffer_addrs
743 			 * didn't change because each write-back
744 			 * erases this info.
745 			 */
746 			rx_desc->read.buffer_addr[j + 1] =
747 			    cpu_to_le64(ps_page->dma);
748 		}
749 
750 		skb = __netdev_alloc_skb_ip_align(netdev, adapter->rx_ps_bsize0,
751 						  gfp);
752 
753 		if (!skb) {
754 			adapter->alloc_rx_buff_failed++;
755 			break;
756 		}
757 
758 		buffer_info->skb = skb;
759 		buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
760 						  adapter->rx_ps_bsize0,
761 						  DMA_FROM_DEVICE);
762 		if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
763 			dev_err(&pdev->dev, "Rx DMA map failed\n");
764 			adapter->rx_dma_failed++;
765 			/* cleanup skb */
766 			dev_kfree_skb_any(skb);
767 			buffer_info->skb = NULL;
768 			break;
769 		}
770 
771 		rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
772 
773 		if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
774 			/* Force memory writes to complete before letting h/w
775 			 * know there are new descriptors to fetch.  (Only
776 			 * applicable for weak-ordered memory model archs,
777 			 * such as IA-64).
778 			 */
779 			wmb();
780 			if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
781 				e1000e_update_rdt_wa(rx_ring, i << 1);
782 			else
783 				writel(i << 1, rx_ring->tail);
784 		}
785 
786 		i++;
787 		if (i == rx_ring->count)
788 			i = 0;
789 		buffer_info = &rx_ring->buffer_info[i];
790 	}
791 
792 no_buffers:
793 	rx_ring->next_to_use = i;
794 }
795 
796 /**
797  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
798  * @rx_ring: Rx descriptor ring
799  * @cleaned_count: number of buffers to allocate this pass
800  **/
801 
802 static void e1000_alloc_jumbo_rx_buffers(struct e1000_ring *rx_ring,
803 					 int cleaned_count, gfp_t gfp)
804 {
805 	struct e1000_adapter *adapter = rx_ring->adapter;
806 	struct net_device *netdev = adapter->netdev;
807 	struct pci_dev *pdev = adapter->pdev;
808 	union e1000_rx_desc_extended *rx_desc;
809 	struct e1000_buffer *buffer_info;
810 	struct sk_buff *skb;
811 	unsigned int i;
812 	unsigned int bufsz = 256 - 16;	/* for skb_reserve */
813 
814 	i = rx_ring->next_to_use;
815 	buffer_info = &rx_ring->buffer_info[i];
816 
817 	while (cleaned_count--) {
818 		skb = buffer_info->skb;
819 		if (skb) {
820 			skb_trim(skb, 0);
821 			goto check_page;
822 		}
823 
824 		skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
825 		if (unlikely(!skb)) {
826 			/* Better luck next round */
827 			adapter->alloc_rx_buff_failed++;
828 			break;
829 		}
830 
831 		buffer_info->skb = skb;
832 check_page:
833 		/* allocate a new page if necessary */
834 		if (!buffer_info->page) {
835 			buffer_info->page = alloc_page(gfp);
836 			if (unlikely(!buffer_info->page)) {
837 				adapter->alloc_rx_buff_failed++;
838 				break;
839 			}
840 		}
841 
842 		if (!buffer_info->dma) {
843 			buffer_info->dma = dma_map_page(&pdev->dev,
844 							buffer_info->page, 0,
845 							PAGE_SIZE,
846 							DMA_FROM_DEVICE);
847 			if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
848 				adapter->alloc_rx_buff_failed++;
849 				break;
850 			}
851 		}
852 
853 		rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
854 		rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
855 
856 		if (unlikely(++i == rx_ring->count))
857 			i = 0;
858 		buffer_info = &rx_ring->buffer_info[i];
859 	}
860 
861 	if (likely(rx_ring->next_to_use != i)) {
862 		rx_ring->next_to_use = i;
863 		if (unlikely(i-- == 0))
864 			i = (rx_ring->count - 1);
865 
866 		/* Force memory writes to complete before letting h/w
867 		 * know there are new descriptors to fetch.  (Only
868 		 * applicable for weak-ordered memory model archs,
869 		 * such as IA-64).
870 		 */
871 		wmb();
872 		if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
873 			e1000e_update_rdt_wa(rx_ring, i);
874 		else
875 			writel(i, rx_ring->tail);
876 	}
877 }
878 
879 static inline void e1000_rx_hash(struct net_device *netdev, __le32 rss,
880 				 struct sk_buff *skb)
881 {
882 	if (netdev->features & NETIF_F_RXHASH)
883 		skb_set_hash(skb, le32_to_cpu(rss), PKT_HASH_TYPE_L3);
884 }
885 
886 /**
887  * e1000_clean_rx_irq - Send received data up the network stack
888  * @rx_ring: Rx descriptor ring
889  *
890  * the return value indicates whether actual cleaning was done, there
891  * is no guarantee that everything was cleaned
892  **/
893 static bool e1000_clean_rx_irq(struct e1000_ring *rx_ring, int *work_done,
894 			       int work_to_do)
895 {
896 	struct e1000_adapter *adapter = rx_ring->adapter;
897 	struct net_device *netdev = adapter->netdev;
898 	struct pci_dev *pdev = adapter->pdev;
899 	struct e1000_hw *hw = &adapter->hw;
900 	union e1000_rx_desc_extended *rx_desc, *next_rxd;
901 	struct e1000_buffer *buffer_info, *next_buffer;
902 	u32 length, staterr;
903 	unsigned int i;
904 	int cleaned_count = 0;
905 	bool cleaned = false;
906 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
907 
908 	i = rx_ring->next_to_clean;
909 	rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
910 	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
911 	buffer_info = &rx_ring->buffer_info[i];
912 
913 	while (staterr & E1000_RXD_STAT_DD) {
914 		struct sk_buff *skb;
915 
916 		if (*work_done >= work_to_do)
917 			break;
918 		(*work_done)++;
919 		rmb();	/* read descriptor and rx_buffer_info after status DD */
920 
921 		skb = buffer_info->skb;
922 		buffer_info->skb = NULL;
923 
924 		prefetch(skb->data - NET_IP_ALIGN);
925 
926 		i++;
927 		if (i == rx_ring->count)
928 			i = 0;
929 		next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
930 		prefetch(next_rxd);
931 
932 		next_buffer = &rx_ring->buffer_info[i];
933 
934 		cleaned = true;
935 		cleaned_count++;
936 		dma_unmap_single(&pdev->dev, buffer_info->dma,
937 				 adapter->rx_buffer_len, DMA_FROM_DEVICE);
938 		buffer_info->dma = 0;
939 
940 		length = le16_to_cpu(rx_desc->wb.upper.length);
941 
942 		/* !EOP means multiple descriptors were used to store a single
943 		 * packet, if that's the case we need to toss it.  In fact, we
944 		 * need to toss every packet with the EOP bit clear and the
945 		 * next frame that _does_ have the EOP bit set, as it is by
946 		 * definition only a frame fragment
947 		 */
948 		if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
949 			adapter->flags2 |= FLAG2_IS_DISCARDING;
950 
951 		if (adapter->flags2 & FLAG2_IS_DISCARDING) {
952 			/* All receives must fit into a single buffer */
953 			e_dbg("Receive packet consumed multiple buffers\n");
954 			/* recycle */
955 			buffer_info->skb = skb;
956 			if (staterr & E1000_RXD_STAT_EOP)
957 				adapter->flags2 &= ~FLAG2_IS_DISCARDING;
958 			goto next_desc;
959 		}
960 
961 		if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
962 			     !(netdev->features & NETIF_F_RXALL))) {
963 			/* recycle */
964 			buffer_info->skb = skb;
965 			goto next_desc;
966 		}
967 
968 		/* adjust length to remove Ethernet CRC */
969 		if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
970 			/* If configured to store CRC, don't subtract FCS,
971 			 * but keep the FCS bytes out of the total_rx_bytes
972 			 * counter
973 			 */
974 			if (netdev->features & NETIF_F_RXFCS)
975 				total_rx_bytes -= 4;
976 			else
977 				length -= 4;
978 		}
979 
980 		total_rx_bytes += length;
981 		total_rx_packets++;
982 
983 		/* code added for copybreak, this should improve
984 		 * performance for small packets with large amounts
985 		 * of reassembly being done in the stack
986 		 */
987 		if (length < copybreak) {
988 			struct sk_buff *new_skb =
989 			    netdev_alloc_skb_ip_align(netdev, length);
990 			if (new_skb) {
991 				skb_copy_to_linear_data_offset(new_skb,
992 							       -NET_IP_ALIGN,
993 							       (skb->data -
994 								NET_IP_ALIGN),
995 							       (length +
996 								NET_IP_ALIGN));
997 				/* save the skb in buffer_info as good */
998 				buffer_info->skb = skb;
999 				skb = new_skb;
1000 			}
1001 			/* else just continue with the old one */
1002 		}
1003 		/* end copybreak code */
1004 		skb_put(skb, length);
1005 
1006 		/* Receive Checksum Offload */
1007 		e1000_rx_checksum(adapter, staterr, skb);
1008 
1009 		e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1010 
1011 		e1000_receive_skb(adapter, netdev, skb, staterr,
1012 				  rx_desc->wb.upper.vlan);
1013 
1014 next_desc:
1015 		rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1016 
1017 		/* return some buffers to hardware, one at a time is too slow */
1018 		if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1019 			adapter->alloc_rx_buf(rx_ring, cleaned_count,
1020 					      GFP_ATOMIC);
1021 			cleaned_count = 0;
1022 		}
1023 
1024 		/* use prefetched values */
1025 		rx_desc = next_rxd;
1026 		buffer_info = next_buffer;
1027 
1028 		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1029 	}
1030 	rx_ring->next_to_clean = i;
1031 
1032 	cleaned_count = e1000_desc_unused(rx_ring);
1033 	if (cleaned_count)
1034 		adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1035 
1036 	adapter->total_rx_bytes += total_rx_bytes;
1037 	adapter->total_rx_packets += total_rx_packets;
1038 	return cleaned;
1039 }
1040 
1041 static void e1000_put_txbuf(struct e1000_ring *tx_ring,
1042 			    struct e1000_buffer *buffer_info)
1043 {
1044 	struct e1000_adapter *adapter = tx_ring->adapter;
1045 
1046 	if (buffer_info->dma) {
1047 		if (buffer_info->mapped_as_page)
1048 			dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1049 				       buffer_info->length, DMA_TO_DEVICE);
1050 		else
1051 			dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1052 					 buffer_info->length, DMA_TO_DEVICE);
1053 		buffer_info->dma = 0;
1054 	}
1055 	if (buffer_info->skb) {
1056 		dev_kfree_skb_any(buffer_info->skb);
1057 		buffer_info->skb = NULL;
1058 	}
1059 	buffer_info->time_stamp = 0;
1060 }
1061 
1062 static void e1000_print_hw_hang(struct work_struct *work)
1063 {
1064 	struct e1000_adapter *adapter = container_of(work,
1065 						     struct e1000_adapter,
1066 						     print_hang_task);
1067 	struct net_device *netdev = adapter->netdev;
1068 	struct e1000_ring *tx_ring = adapter->tx_ring;
1069 	unsigned int i = tx_ring->next_to_clean;
1070 	unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1071 	struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1072 	struct e1000_hw *hw = &adapter->hw;
1073 	u16 phy_status, phy_1000t_status, phy_ext_status;
1074 	u16 pci_status;
1075 
1076 	if (test_bit(__E1000_DOWN, &adapter->state))
1077 		return;
1078 
1079 	if (!adapter->tx_hang_recheck && (adapter->flags2 & FLAG2_DMA_BURST)) {
1080 		/* May be block on write-back, flush and detect again
1081 		 * flush pending descriptor writebacks to memory
1082 		 */
1083 		ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1084 		/* execute the writes immediately */
1085 		e1e_flush();
1086 		/* Due to rare timing issues, write to TIDV again to ensure
1087 		 * the write is successful
1088 		 */
1089 		ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1090 		/* execute the writes immediately */
1091 		e1e_flush();
1092 		adapter->tx_hang_recheck = true;
1093 		return;
1094 	}
1095 	adapter->tx_hang_recheck = false;
1096 
1097 	if (er32(TDH(0)) == er32(TDT(0))) {
1098 		e_dbg("false hang detected, ignoring\n");
1099 		return;
1100 	}
1101 
1102 	/* Real hang detected */
1103 	netif_stop_queue(netdev);
1104 
1105 	e1e_rphy(hw, MII_BMSR, &phy_status);
1106 	e1e_rphy(hw, MII_STAT1000, &phy_1000t_status);
1107 	e1e_rphy(hw, MII_ESTATUS, &phy_ext_status);
1108 
1109 	pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1110 
1111 	/* detected Hardware unit hang */
1112 	e_err("Detected Hardware Unit Hang:\n"
1113 	      "  TDH                  <%x>\n"
1114 	      "  TDT                  <%x>\n"
1115 	      "  next_to_use          <%x>\n"
1116 	      "  next_to_clean        <%x>\n"
1117 	      "buffer_info[next_to_clean]:\n"
1118 	      "  time_stamp           <%lx>\n"
1119 	      "  next_to_watch        <%x>\n"
1120 	      "  jiffies              <%lx>\n"
1121 	      "  next_to_watch.status <%x>\n"
1122 	      "MAC Status             <%x>\n"
1123 	      "PHY Status             <%x>\n"
1124 	      "PHY 1000BASE-T Status  <%x>\n"
1125 	      "PHY Extended Status    <%x>\n"
1126 	      "PCI Status             <%x>\n",
1127 	      readl(tx_ring->head), readl(tx_ring->tail), tx_ring->next_to_use,
1128 	      tx_ring->next_to_clean, tx_ring->buffer_info[eop].time_stamp,
1129 	      eop, jiffies, eop_desc->upper.fields.status, er32(STATUS),
1130 	      phy_status, phy_1000t_status, phy_ext_status, pci_status);
1131 
1132 	e1000e_dump(adapter);
1133 
1134 	/* Suggest workaround for known h/w issue */
1135 	if ((hw->mac.type == e1000_pchlan) && (er32(CTRL) & E1000_CTRL_TFCE))
1136 		e_err("Try turning off Tx pause (flow control) via ethtool\n");
1137 }
1138 
1139 /**
1140  * e1000e_tx_hwtstamp_work - check for Tx time stamp
1141  * @work: pointer to work struct
1142  *
1143  * This work function polls the TSYNCTXCTL valid bit to determine when a
1144  * timestamp has been taken for the current stored skb.  The timestamp must
1145  * be for this skb because only one such packet is allowed in the queue.
1146  */
1147 static void e1000e_tx_hwtstamp_work(struct work_struct *work)
1148 {
1149 	struct e1000_adapter *adapter = container_of(work, struct e1000_adapter,
1150 						     tx_hwtstamp_work);
1151 	struct e1000_hw *hw = &adapter->hw;
1152 
1153 	if (er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID) {
1154 		struct skb_shared_hwtstamps shhwtstamps;
1155 		u64 txstmp;
1156 
1157 		txstmp = er32(TXSTMPL);
1158 		txstmp |= (u64)er32(TXSTMPH) << 32;
1159 
1160 		e1000e_systim_to_hwtstamp(adapter, &shhwtstamps, txstmp);
1161 
1162 		skb_tstamp_tx(adapter->tx_hwtstamp_skb, &shhwtstamps);
1163 		dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
1164 		adapter->tx_hwtstamp_skb = NULL;
1165 	} else if (time_after(jiffies, adapter->tx_hwtstamp_start
1166 			      + adapter->tx_timeout_factor * HZ)) {
1167 		dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
1168 		adapter->tx_hwtstamp_skb = NULL;
1169 		adapter->tx_hwtstamp_timeouts++;
1170 		e_warn("clearing Tx timestamp hang\n");
1171 	} else {
1172 		/* reschedule to check later */
1173 		schedule_work(&adapter->tx_hwtstamp_work);
1174 	}
1175 }
1176 
1177 /**
1178  * e1000_clean_tx_irq - Reclaim resources after transmit completes
1179  * @tx_ring: Tx descriptor ring
1180  *
1181  * the return value indicates whether actual cleaning was done, there
1182  * is no guarantee that everything was cleaned
1183  **/
1184 static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
1185 {
1186 	struct e1000_adapter *adapter = tx_ring->adapter;
1187 	struct net_device *netdev = adapter->netdev;
1188 	struct e1000_hw *hw = &adapter->hw;
1189 	struct e1000_tx_desc *tx_desc, *eop_desc;
1190 	struct e1000_buffer *buffer_info;
1191 	unsigned int i, eop;
1192 	unsigned int count = 0;
1193 	unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1194 	unsigned int bytes_compl = 0, pkts_compl = 0;
1195 
1196 	i = tx_ring->next_to_clean;
1197 	eop = tx_ring->buffer_info[i].next_to_watch;
1198 	eop_desc = E1000_TX_DESC(*tx_ring, eop);
1199 
1200 	while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1201 	       (count < tx_ring->count)) {
1202 		bool cleaned = false;
1203 
1204 		rmb();		/* read buffer_info after eop_desc */
1205 		for (; !cleaned; count++) {
1206 			tx_desc = E1000_TX_DESC(*tx_ring, i);
1207 			buffer_info = &tx_ring->buffer_info[i];
1208 			cleaned = (i == eop);
1209 
1210 			if (cleaned) {
1211 				total_tx_packets += buffer_info->segs;
1212 				total_tx_bytes += buffer_info->bytecount;
1213 				if (buffer_info->skb) {
1214 					bytes_compl += buffer_info->skb->len;
1215 					pkts_compl++;
1216 				}
1217 			}
1218 
1219 			e1000_put_txbuf(tx_ring, buffer_info);
1220 			tx_desc->upper.data = 0;
1221 
1222 			i++;
1223 			if (i == tx_ring->count)
1224 				i = 0;
1225 		}
1226 
1227 		if (i == tx_ring->next_to_use)
1228 			break;
1229 		eop = tx_ring->buffer_info[i].next_to_watch;
1230 		eop_desc = E1000_TX_DESC(*tx_ring, eop);
1231 	}
1232 
1233 	tx_ring->next_to_clean = i;
1234 
1235 	netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1236 
1237 #define TX_WAKE_THRESHOLD 32
1238 	if (count && netif_carrier_ok(netdev) &&
1239 	    e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1240 		/* Make sure that anybody stopping the queue after this
1241 		 * sees the new next_to_clean.
1242 		 */
1243 		smp_mb();
1244 
1245 		if (netif_queue_stopped(netdev) &&
1246 		    !(test_bit(__E1000_DOWN, &adapter->state))) {
1247 			netif_wake_queue(netdev);
1248 			++adapter->restart_queue;
1249 		}
1250 	}
1251 
1252 	if (adapter->detect_tx_hung) {
1253 		/* Detect a transmit hang in hardware, this serializes the
1254 		 * check with the clearing of time_stamp and movement of i
1255 		 */
1256 		adapter->detect_tx_hung = false;
1257 		if (tx_ring->buffer_info[i].time_stamp &&
1258 		    time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1259 			       + (adapter->tx_timeout_factor * HZ)) &&
1260 		    !(er32(STATUS) & E1000_STATUS_TXOFF))
1261 			schedule_work(&adapter->print_hang_task);
1262 		else
1263 			adapter->tx_hang_recheck = false;
1264 	}
1265 	adapter->total_tx_bytes += total_tx_bytes;
1266 	adapter->total_tx_packets += total_tx_packets;
1267 	return count < tx_ring->count;
1268 }
1269 
1270 /**
1271  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1272  * @rx_ring: Rx descriptor ring
1273  *
1274  * the return value indicates whether actual cleaning was done, there
1275  * is no guarantee that everything was cleaned
1276  **/
1277 static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
1278 				  int work_to_do)
1279 {
1280 	struct e1000_adapter *adapter = rx_ring->adapter;
1281 	struct e1000_hw *hw = &adapter->hw;
1282 	union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1283 	struct net_device *netdev = adapter->netdev;
1284 	struct pci_dev *pdev = adapter->pdev;
1285 	struct e1000_buffer *buffer_info, *next_buffer;
1286 	struct e1000_ps_page *ps_page;
1287 	struct sk_buff *skb;
1288 	unsigned int i, j;
1289 	u32 length, staterr;
1290 	int cleaned_count = 0;
1291 	bool cleaned = false;
1292 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1293 
1294 	i = rx_ring->next_to_clean;
1295 	rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1296 	staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1297 	buffer_info = &rx_ring->buffer_info[i];
1298 
1299 	while (staterr & E1000_RXD_STAT_DD) {
1300 		if (*work_done >= work_to_do)
1301 			break;
1302 		(*work_done)++;
1303 		skb = buffer_info->skb;
1304 		rmb();	/* read descriptor and rx_buffer_info after status DD */
1305 
1306 		/* in the packet split case this is header only */
1307 		prefetch(skb->data - NET_IP_ALIGN);
1308 
1309 		i++;
1310 		if (i == rx_ring->count)
1311 			i = 0;
1312 		next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1313 		prefetch(next_rxd);
1314 
1315 		next_buffer = &rx_ring->buffer_info[i];
1316 
1317 		cleaned = true;
1318 		cleaned_count++;
1319 		dma_unmap_single(&pdev->dev, buffer_info->dma,
1320 				 adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1321 		buffer_info->dma = 0;
1322 
1323 		/* see !EOP comment in other Rx routine */
1324 		if (!(staterr & E1000_RXD_STAT_EOP))
1325 			adapter->flags2 |= FLAG2_IS_DISCARDING;
1326 
1327 		if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1328 			e_dbg("Packet Split buffers didn't pick up the full packet\n");
1329 			dev_kfree_skb_irq(skb);
1330 			if (staterr & E1000_RXD_STAT_EOP)
1331 				adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1332 			goto next_desc;
1333 		}
1334 
1335 		if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1336 			     !(netdev->features & NETIF_F_RXALL))) {
1337 			dev_kfree_skb_irq(skb);
1338 			goto next_desc;
1339 		}
1340 
1341 		length = le16_to_cpu(rx_desc->wb.middle.length0);
1342 
1343 		if (!length) {
1344 			e_dbg("Last part of the packet spanning multiple descriptors\n");
1345 			dev_kfree_skb_irq(skb);
1346 			goto next_desc;
1347 		}
1348 
1349 		/* Good Receive */
1350 		skb_put(skb, length);
1351 
1352 		{
1353 			/* this looks ugly, but it seems compiler issues make
1354 			 * it more efficient than reusing j
1355 			 */
1356 			int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1357 
1358 			/* page alloc/put takes too long and effects small
1359 			 * packet throughput, so unsplit small packets and
1360 			 * save the alloc/put only valid in softirq (napi)
1361 			 * context to call kmap_*
1362 			 */
1363 			if (l1 && (l1 <= copybreak) &&
1364 			    ((length + l1) <= adapter->rx_ps_bsize0)) {
1365 				u8 *vaddr;
1366 
1367 				ps_page = &buffer_info->ps_pages[0];
1368 
1369 				/* there is no documentation about how to call
1370 				 * kmap_atomic, so we can't hold the mapping
1371 				 * very long
1372 				 */
1373 				dma_sync_single_for_cpu(&pdev->dev,
1374 							ps_page->dma,
1375 							PAGE_SIZE,
1376 							DMA_FROM_DEVICE);
1377 				vaddr = kmap_atomic(ps_page->page);
1378 				memcpy(skb_tail_pointer(skb), vaddr, l1);
1379 				kunmap_atomic(vaddr);
1380 				dma_sync_single_for_device(&pdev->dev,
1381 							   ps_page->dma,
1382 							   PAGE_SIZE,
1383 							   DMA_FROM_DEVICE);
1384 
1385 				/* remove the CRC */
1386 				if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1387 					if (!(netdev->features & NETIF_F_RXFCS))
1388 						l1 -= 4;
1389 				}
1390 
1391 				skb_put(skb, l1);
1392 				goto copydone;
1393 			}	/* if */
1394 		}
1395 
1396 		for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1397 			length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1398 			if (!length)
1399 				break;
1400 
1401 			ps_page = &buffer_info->ps_pages[j];
1402 			dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1403 				       DMA_FROM_DEVICE);
1404 			ps_page->dma = 0;
1405 			skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1406 			ps_page->page = NULL;
1407 			skb->len += length;
1408 			skb->data_len += length;
1409 			skb->truesize += PAGE_SIZE;
1410 		}
1411 
1412 		/* strip the ethernet crc, problem is we're using pages now so
1413 		 * this whole operation can get a little cpu intensive
1414 		 */
1415 		if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1416 			if (!(netdev->features & NETIF_F_RXFCS))
1417 				pskb_trim(skb, skb->len - 4);
1418 		}
1419 
1420 copydone:
1421 		total_rx_bytes += skb->len;
1422 		total_rx_packets++;
1423 
1424 		e1000_rx_checksum(adapter, staterr, skb);
1425 
1426 		e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1427 
1428 		if (rx_desc->wb.upper.header_status &
1429 		    cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1430 			adapter->rx_hdr_split++;
1431 
1432 		e1000_receive_skb(adapter, netdev, skb, staterr,
1433 				  rx_desc->wb.middle.vlan);
1434 
1435 next_desc:
1436 		rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1437 		buffer_info->skb = NULL;
1438 
1439 		/* return some buffers to hardware, one at a time is too slow */
1440 		if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1441 			adapter->alloc_rx_buf(rx_ring, cleaned_count,
1442 					      GFP_ATOMIC);
1443 			cleaned_count = 0;
1444 		}
1445 
1446 		/* use prefetched values */
1447 		rx_desc = next_rxd;
1448 		buffer_info = next_buffer;
1449 
1450 		staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1451 	}
1452 	rx_ring->next_to_clean = i;
1453 
1454 	cleaned_count = e1000_desc_unused(rx_ring);
1455 	if (cleaned_count)
1456 		adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1457 
1458 	adapter->total_rx_bytes += total_rx_bytes;
1459 	adapter->total_rx_packets += total_rx_packets;
1460 	return cleaned;
1461 }
1462 
1463 /**
1464  * e1000_consume_page - helper function
1465  **/
1466 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1467 			       u16 length)
1468 {
1469 	bi->page = NULL;
1470 	skb->len += length;
1471 	skb->data_len += length;
1472 	skb->truesize += PAGE_SIZE;
1473 }
1474 
1475 /**
1476  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1477  * @adapter: board private structure
1478  *
1479  * the return value indicates whether actual cleaning was done, there
1480  * is no guarantee that everything was cleaned
1481  **/
1482 static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1483 				     int work_to_do)
1484 {
1485 	struct e1000_adapter *adapter = rx_ring->adapter;
1486 	struct net_device *netdev = adapter->netdev;
1487 	struct pci_dev *pdev = adapter->pdev;
1488 	union e1000_rx_desc_extended *rx_desc, *next_rxd;
1489 	struct e1000_buffer *buffer_info, *next_buffer;
1490 	u32 length, staterr;
1491 	unsigned int i;
1492 	int cleaned_count = 0;
1493 	bool cleaned = false;
1494 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1495 	struct skb_shared_info *shinfo;
1496 
1497 	i = rx_ring->next_to_clean;
1498 	rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1499 	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1500 	buffer_info = &rx_ring->buffer_info[i];
1501 
1502 	while (staterr & E1000_RXD_STAT_DD) {
1503 		struct sk_buff *skb;
1504 
1505 		if (*work_done >= work_to_do)
1506 			break;
1507 		(*work_done)++;
1508 		rmb();	/* read descriptor and rx_buffer_info after status DD */
1509 
1510 		skb = buffer_info->skb;
1511 		buffer_info->skb = NULL;
1512 
1513 		++i;
1514 		if (i == rx_ring->count)
1515 			i = 0;
1516 		next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1517 		prefetch(next_rxd);
1518 
1519 		next_buffer = &rx_ring->buffer_info[i];
1520 
1521 		cleaned = true;
1522 		cleaned_count++;
1523 		dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1524 			       DMA_FROM_DEVICE);
1525 		buffer_info->dma = 0;
1526 
1527 		length = le16_to_cpu(rx_desc->wb.upper.length);
1528 
1529 		/* errors is only valid for DD + EOP descriptors */
1530 		if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1531 			     ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1532 			      !(netdev->features & NETIF_F_RXALL)))) {
1533 			/* recycle both page and skb */
1534 			buffer_info->skb = skb;
1535 			/* an error means any chain goes out the window too */
1536 			if (rx_ring->rx_skb_top)
1537 				dev_kfree_skb_irq(rx_ring->rx_skb_top);
1538 			rx_ring->rx_skb_top = NULL;
1539 			goto next_desc;
1540 		}
1541 #define rxtop (rx_ring->rx_skb_top)
1542 		if (!(staterr & E1000_RXD_STAT_EOP)) {
1543 			/* this descriptor is only the beginning (or middle) */
1544 			if (!rxtop) {
1545 				/* this is the beginning of a chain */
1546 				rxtop = skb;
1547 				skb_fill_page_desc(rxtop, 0, buffer_info->page,
1548 						   0, length);
1549 			} else {
1550 				/* this is the middle of a chain */
1551 				shinfo = skb_shinfo(rxtop);
1552 				skb_fill_page_desc(rxtop, shinfo->nr_frags,
1553 						   buffer_info->page, 0,
1554 						   length);
1555 				/* re-use the skb, only consumed the page */
1556 				buffer_info->skb = skb;
1557 			}
1558 			e1000_consume_page(buffer_info, rxtop, length);
1559 			goto next_desc;
1560 		} else {
1561 			if (rxtop) {
1562 				/* end of the chain */
1563 				shinfo = skb_shinfo(rxtop);
1564 				skb_fill_page_desc(rxtop, shinfo->nr_frags,
1565 						   buffer_info->page, 0,
1566 						   length);
1567 				/* re-use the current skb, we only consumed the
1568 				 * page
1569 				 */
1570 				buffer_info->skb = skb;
1571 				skb = rxtop;
1572 				rxtop = NULL;
1573 				e1000_consume_page(buffer_info, skb, length);
1574 			} else {
1575 				/* no chain, got EOP, this buf is the packet
1576 				 * copybreak to save the put_page/alloc_page
1577 				 */
1578 				if (length <= copybreak &&
1579 				    skb_tailroom(skb) >= length) {
1580 					u8 *vaddr;
1581 					vaddr = kmap_atomic(buffer_info->page);
1582 					memcpy(skb_tail_pointer(skb), vaddr,
1583 					       length);
1584 					kunmap_atomic(vaddr);
1585 					/* re-use the page, so don't erase
1586 					 * buffer_info->page
1587 					 */
1588 					skb_put(skb, length);
1589 				} else {
1590 					skb_fill_page_desc(skb, 0,
1591 							   buffer_info->page, 0,
1592 							   length);
1593 					e1000_consume_page(buffer_info, skb,
1594 							   length);
1595 				}
1596 			}
1597 		}
1598 
1599 		/* Receive Checksum Offload */
1600 		e1000_rx_checksum(adapter, staterr, skb);
1601 
1602 		e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1603 
1604 		/* probably a little skewed due to removing CRC */
1605 		total_rx_bytes += skb->len;
1606 		total_rx_packets++;
1607 
1608 		/* eth type trans needs skb->data to point to something */
1609 		if (!pskb_may_pull(skb, ETH_HLEN)) {
1610 			e_err("pskb_may_pull failed.\n");
1611 			dev_kfree_skb_irq(skb);
1612 			goto next_desc;
1613 		}
1614 
1615 		e1000_receive_skb(adapter, netdev, skb, staterr,
1616 				  rx_desc->wb.upper.vlan);
1617 
1618 next_desc:
1619 		rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1620 
1621 		/* return some buffers to hardware, one at a time is too slow */
1622 		if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1623 			adapter->alloc_rx_buf(rx_ring, cleaned_count,
1624 					      GFP_ATOMIC);
1625 			cleaned_count = 0;
1626 		}
1627 
1628 		/* use prefetched values */
1629 		rx_desc = next_rxd;
1630 		buffer_info = next_buffer;
1631 
1632 		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1633 	}
1634 	rx_ring->next_to_clean = i;
1635 
1636 	cleaned_count = e1000_desc_unused(rx_ring);
1637 	if (cleaned_count)
1638 		adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1639 
1640 	adapter->total_rx_bytes += total_rx_bytes;
1641 	adapter->total_rx_packets += total_rx_packets;
1642 	return cleaned;
1643 }
1644 
1645 /**
1646  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1647  * @rx_ring: Rx descriptor ring
1648  **/
1649 static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1650 {
1651 	struct e1000_adapter *adapter = rx_ring->adapter;
1652 	struct e1000_buffer *buffer_info;
1653 	struct e1000_ps_page *ps_page;
1654 	struct pci_dev *pdev = adapter->pdev;
1655 	unsigned int i, j;
1656 
1657 	/* Free all the Rx ring sk_buffs */
1658 	for (i = 0; i < rx_ring->count; i++) {
1659 		buffer_info = &rx_ring->buffer_info[i];
1660 		if (buffer_info->dma) {
1661 			if (adapter->clean_rx == e1000_clean_rx_irq)
1662 				dma_unmap_single(&pdev->dev, buffer_info->dma,
1663 						 adapter->rx_buffer_len,
1664 						 DMA_FROM_DEVICE);
1665 			else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1666 				dma_unmap_page(&pdev->dev, buffer_info->dma,
1667 					       PAGE_SIZE, DMA_FROM_DEVICE);
1668 			else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1669 				dma_unmap_single(&pdev->dev, buffer_info->dma,
1670 						 adapter->rx_ps_bsize0,
1671 						 DMA_FROM_DEVICE);
1672 			buffer_info->dma = 0;
1673 		}
1674 
1675 		if (buffer_info->page) {
1676 			put_page(buffer_info->page);
1677 			buffer_info->page = NULL;
1678 		}
1679 
1680 		if (buffer_info->skb) {
1681 			dev_kfree_skb(buffer_info->skb);
1682 			buffer_info->skb = NULL;
1683 		}
1684 
1685 		for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1686 			ps_page = &buffer_info->ps_pages[j];
1687 			if (!ps_page->page)
1688 				break;
1689 			dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1690 				       DMA_FROM_DEVICE);
1691 			ps_page->dma = 0;
1692 			put_page(ps_page->page);
1693 			ps_page->page = NULL;
1694 		}
1695 	}
1696 
1697 	/* there also may be some cached data from a chained receive */
1698 	if (rx_ring->rx_skb_top) {
1699 		dev_kfree_skb(rx_ring->rx_skb_top);
1700 		rx_ring->rx_skb_top = NULL;
1701 	}
1702 
1703 	/* Zero out the descriptor ring */
1704 	memset(rx_ring->desc, 0, rx_ring->size);
1705 
1706 	rx_ring->next_to_clean = 0;
1707 	rx_ring->next_to_use = 0;
1708 	adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1709 
1710 	writel(0, rx_ring->head);
1711 	if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
1712 		e1000e_update_rdt_wa(rx_ring, 0);
1713 	else
1714 		writel(0, rx_ring->tail);
1715 }
1716 
1717 static void e1000e_downshift_workaround(struct work_struct *work)
1718 {
1719 	struct e1000_adapter *adapter = container_of(work,
1720 						     struct e1000_adapter,
1721 						     downshift_task);
1722 
1723 	if (test_bit(__E1000_DOWN, &adapter->state))
1724 		return;
1725 
1726 	e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1727 }
1728 
1729 /**
1730  * e1000_intr_msi - Interrupt Handler
1731  * @irq: interrupt number
1732  * @data: pointer to a network interface device structure
1733  **/
1734 static irqreturn_t e1000_intr_msi(int __always_unused irq, void *data)
1735 {
1736 	struct net_device *netdev = data;
1737 	struct e1000_adapter *adapter = netdev_priv(netdev);
1738 	struct e1000_hw *hw = &adapter->hw;
1739 	u32 icr = er32(ICR);
1740 
1741 	/* read ICR disables interrupts using IAM */
1742 	if (icr & E1000_ICR_LSC) {
1743 		hw->mac.get_link_status = true;
1744 		/* ICH8 workaround-- Call gig speed drop workaround on cable
1745 		 * disconnect (LSC) before accessing any PHY registers
1746 		 */
1747 		if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1748 		    (!(er32(STATUS) & E1000_STATUS_LU)))
1749 			schedule_work(&adapter->downshift_task);
1750 
1751 		/* 80003ES2LAN workaround-- For packet buffer work-around on
1752 		 * link down event; disable receives here in the ISR and reset
1753 		 * adapter in watchdog
1754 		 */
1755 		if (netif_carrier_ok(netdev) &&
1756 		    adapter->flags & FLAG_RX_NEEDS_RESTART) {
1757 			/* disable receives */
1758 			u32 rctl = er32(RCTL);
1759 
1760 			ew32(RCTL, rctl & ~E1000_RCTL_EN);
1761 			adapter->flags |= FLAG_RESTART_NOW;
1762 		}
1763 		/* guard against interrupt when we're going down */
1764 		if (!test_bit(__E1000_DOWN, &adapter->state))
1765 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
1766 	}
1767 
1768 	/* Reset on uncorrectable ECC error */
1769 	if ((icr & E1000_ICR_ECCER) && (hw->mac.type == e1000_pch_lpt)) {
1770 		u32 pbeccsts = er32(PBECCSTS);
1771 
1772 		adapter->corr_errors +=
1773 		    pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1774 		adapter->uncorr_errors +=
1775 		    (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
1776 		    E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
1777 
1778 		/* Do the reset outside of interrupt context */
1779 		schedule_work(&adapter->reset_task);
1780 
1781 		/* return immediately since reset is imminent */
1782 		return IRQ_HANDLED;
1783 	}
1784 
1785 	if (napi_schedule_prep(&adapter->napi)) {
1786 		adapter->total_tx_bytes = 0;
1787 		adapter->total_tx_packets = 0;
1788 		adapter->total_rx_bytes = 0;
1789 		adapter->total_rx_packets = 0;
1790 		__napi_schedule(&adapter->napi);
1791 	}
1792 
1793 	return IRQ_HANDLED;
1794 }
1795 
1796 /**
1797  * e1000_intr - Interrupt Handler
1798  * @irq: interrupt number
1799  * @data: pointer to a network interface device structure
1800  **/
1801 static irqreturn_t e1000_intr(int __always_unused irq, void *data)
1802 {
1803 	struct net_device *netdev = data;
1804 	struct e1000_adapter *adapter = netdev_priv(netdev);
1805 	struct e1000_hw *hw = &adapter->hw;
1806 	u32 rctl, icr = er32(ICR);
1807 
1808 	if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1809 		return IRQ_NONE;	/* Not our interrupt */
1810 
1811 	/* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1812 	 * not set, then the adapter didn't send an interrupt
1813 	 */
1814 	if (!(icr & E1000_ICR_INT_ASSERTED))
1815 		return IRQ_NONE;
1816 
1817 	/* Interrupt Auto-Mask...upon reading ICR,
1818 	 * interrupts are masked.  No need for the
1819 	 * IMC write
1820 	 */
1821 
1822 	if (icr & E1000_ICR_LSC) {
1823 		hw->mac.get_link_status = true;
1824 		/* ICH8 workaround-- Call gig speed drop workaround on cable
1825 		 * disconnect (LSC) before accessing any PHY registers
1826 		 */
1827 		if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1828 		    (!(er32(STATUS) & E1000_STATUS_LU)))
1829 			schedule_work(&adapter->downshift_task);
1830 
1831 		/* 80003ES2LAN workaround--
1832 		 * For packet buffer work-around on link down event;
1833 		 * disable receives here in the ISR and
1834 		 * reset adapter in watchdog
1835 		 */
1836 		if (netif_carrier_ok(netdev) &&
1837 		    (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1838 			/* disable receives */
1839 			rctl = er32(RCTL);
1840 			ew32(RCTL, rctl & ~E1000_RCTL_EN);
1841 			adapter->flags |= FLAG_RESTART_NOW;
1842 		}
1843 		/* guard against interrupt when we're going down */
1844 		if (!test_bit(__E1000_DOWN, &adapter->state))
1845 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
1846 	}
1847 
1848 	/* Reset on uncorrectable ECC error */
1849 	if ((icr & E1000_ICR_ECCER) && (hw->mac.type == e1000_pch_lpt)) {
1850 		u32 pbeccsts = er32(PBECCSTS);
1851 
1852 		adapter->corr_errors +=
1853 		    pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1854 		adapter->uncorr_errors +=
1855 		    (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
1856 		    E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
1857 
1858 		/* Do the reset outside of interrupt context */
1859 		schedule_work(&adapter->reset_task);
1860 
1861 		/* return immediately since reset is imminent */
1862 		return IRQ_HANDLED;
1863 	}
1864 
1865 	if (napi_schedule_prep(&adapter->napi)) {
1866 		adapter->total_tx_bytes = 0;
1867 		adapter->total_tx_packets = 0;
1868 		adapter->total_rx_bytes = 0;
1869 		adapter->total_rx_packets = 0;
1870 		__napi_schedule(&adapter->napi);
1871 	}
1872 
1873 	return IRQ_HANDLED;
1874 }
1875 
1876 static irqreturn_t e1000_msix_other(int __always_unused irq, void *data)
1877 {
1878 	struct net_device *netdev = data;
1879 	struct e1000_adapter *adapter = netdev_priv(netdev);
1880 	struct e1000_hw *hw = &adapter->hw;
1881 	u32 icr = er32(ICR);
1882 
1883 	if (!(icr & E1000_ICR_INT_ASSERTED)) {
1884 		if (!test_bit(__E1000_DOWN, &adapter->state))
1885 			ew32(IMS, E1000_IMS_OTHER);
1886 		return IRQ_NONE;
1887 	}
1888 
1889 	if (icr & adapter->eiac_mask)
1890 		ew32(ICS, (icr & adapter->eiac_mask));
1891 
1892 	if (icr & E1000_ICR_OTHER) {
1893 		if (!(icr & E1000_ICR_LSC))
1894 			goto no_link_interrupt;
1895 		hw->mac.get_link_status = true;
1896 		/* guard against interrupt when we're going down */
1897 		if (!test_bit(__E1000_DOWN, &adapter->state))
1898 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
1899 	}
1900 
1901 no_link_interrupt:
1902 	if (!test_bit(__E1000_DOWN, &adapter->state))
1903 		ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1904 
1905 	return IRQ_HANDLED;
1906 }
1907 
1908 static irqreturn_t e1000_intr_msix_tx(int __always_unused irq, void *data)
1909 {
1910 	struct net_device *netdev = data;
1911 	struct e1000_adapter *adapter = netdev_priv(netdev);
1912 	struct e1000_hw *hw = &adapter->hw;
1913 	struct e1000_ring *tx_ring = adapter->tx_ring;
1914 
1915 	adapter->total_tx_bytes = 0;
1916 	adapter->total_tx_packets = 0;
1917 
1918 	if (!e1000_clean_tx_irq(tx_ring))
1919 		/* Ring was not completely cleaned, so fire another interrupt */
1920 		ew32(ICS, tx_ring->ims_val);
1921 
1922 	return IRQ_HANDLED;
1923 }
1924 
1925 static irqreturn_t e1000_intr_msix_rx(int __always_unused irq, void *data)
1926 {
1927 	struct net_device *netdev = data;
1928 	struct e1000_adapter *adapter = netdev_priv(netdev);
1929 	struct e1000_ring *rx_ring = adapter->rx_ring;
1930 
1931 	/* Write the ITR value calculated at the end of the
1932 	 * previous interrupt.
1933 	 */
1934 	if (rx_ring->set_itr) {
1935 		writel(1000000000 / (rx_ring->itr_val * 256),
1936 		       rx_ring->itr_register);
1937 		rx_ring->set_itr = 0;
1938 	}
1939 
1940 	if (napi_schedule_prep(&adapter->napi)) {
1941 		adapter->total_rx_bytes = 0;
1942 		adapter->total_rx_packets = 0;
1943 		__napi_schedule(&adapter->napi);
1944 	}
1945 	return IRQ_HANDLED;
1946 }
1947 
1948 /**
1949  * e1000_configure_msix - Configure MSI-X hardware
1950  *
1951  * e1000_configure_msix sets up the hardware to properly
1952  * generate MSI-X interrupts.
1953  **/
1954 static void e1000_configure_msix(struct e1000_adapter *adapter)
1955 {
1956 	struct e1000_hw *hw = &adapter->hw;
1957 	struct e1000_ring *rx_ring = adapter->rx_ring;
1958 	struct e1000_ring *tx_ring = adapter->tx_ring;
1959 	int vector = 0;
1960 	u32 ctrl_ext, ivar = 0;
1961 
1962 	adapter->eiac_mask = 0;
1963 
1964 	/* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1965 	if (hw->mac.type == e1000_82574) {
1966 		u32 rfctl = er32(RFCTL);
1967 
1968 		rfctl |= E1000_RFCTL_ACK_DIS;
1969 		ew32(RFCTL, rfctl);
1970 	}
1971 
1972 	/* Configure Rx vector */
1973 	rx_ring->ims_val = E1000_IMS_RXQ0;
1974 	adapter->eiac_mask |= rx_ring->ims_val;
1975 	if (rx_ring->itr_val)
1976 		writel(1000000000 / (rx_ring->itr_val * 256),
1977 		       rx_ring->itr_register);
1978 	else
1979 		writel(1, rx_ring->itr_register);
1980 	ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1981 
1982 	/* Configure Tx vector */
1983 	tx_ring->ims_val = E1000_IMS_TXQ0;
1984 	vector++;
1985 	if (tx_ring->itr_val)
1986 		writel(1000000000 / (tx_ring->itr_val * 256),
1987 		       tx_ring->itr_register);
1988 	else
1989 		writel(1, tx_ring->itr_register);
1990 	adapter->eiac_mask |= tx_ring->ims_val;
1991 	ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1992 
1993 	/* set vector for Other Causes, e.g. link changes */
1994 	vector++;
1995 	ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1996 	if (rx_ring->itr_val)
1997 		writel(1000000000 / (rx_ring->itr_val * 256),
1998 		       hw->hw_addr + E1000_EITR_82574(vector));
1999 	else
2000 		writel(1, hw->hw_addr + E1000_EITR_82574(vector));
2001 
2002 	/* Cause Tx interrupts on every write back */
2003 	ivar |= (1 << 31);
2004 
2005 	ew32(IVAR, ivar);
2006 
2007 	/* enable MSI-X PBA support */
2008 	ctrl_ext = er32(CTRL_EXT);
2009 	ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
2010 
2011 	/* Auto-Mask Other interrupts upon ICR read */
2012 	ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
2013 	ctrl_ext |= E1000_CTRL_EXT_EIAME;
2014 	ew32(CTRL_EXT, ctrl_ext);
2015 	e1e_flush();
2016 }
2017 
2018 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
2019 {
2020 	if (adapter->msix_entries) {
2021 		pci_disable_msix(adapter->pdev);
2022 		kfree(adapter->msix_entries);
2023 		adapter->msix_entries = NULL;
2024 	} else if (adapter->flags & FLAG_MSI_ENABLED) {
2025 		pci_disable_msi(adapter->pdev);
2026 		adapter->flags &= ~FLAG_MSI_ENABLED;
2027 	}
2028 }
2029 
2030 /**
2031  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
2032  *
2033  * Attempt to configure interrupts using the best available
2034  * capabilities of the hardware and kernel.
2035  **/
2036 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
2037 {
2038 	int err;
2039 	int i;
2040 
2041 	switch (adapter->int_mode) {
2042 	case E1000E_INT_MODE_MSIX:
2043 		if (adapter->flags & FLAG_HAS_MSIX) {
2044 			adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
2045 			adapter->msix_entries = kcalloc(adapter->num_vectors,
2046 							sizeof(struct
2047 							       msix_entry),
2048 							GFP_KERNEL);
2049 			if (adapter->msix_entries) {
2050 				struct e1000_adapter *a = adapter;
2051 
2052 				for (i = 0; i < adapter->num_vectors; i++)
2053 					adapter->msix_entries[i].entry = i;
2054 
2055 				err = pci_enable_msix_range(a->pdev,
2056 							    a->msix_entries,
2057 							    a->num_vectors,
2058 							    a->num_vectors);
2059 				if (err > 0)
2060 					return;
2061 			}
2062 			/* MSI-X failed, so fall through and try MSI */
2063 			e_err("Failed to initialize MSI-X interrupts.  Falling back to MSI interrupts.\n");
2064 			e1000e_reset_interrupt_capability(adapter);
2065 		}
2066 		adapter->int_mode = E1000E_INT_MODE_MSI;
2067 		/* Fall through */
2068 	case E1000E_INT_MODE_MSI:
2069 		if (!pci_enable_msi(adapter->pdev)) {
2070 			adapter->flags |= FLAG_MSI_ENABLED;
2071 		} else {
2072 			adapter->int_mode = E1000E_INT_MODE_LEGACY;
2073 			e_err("Failed to initialize MSI interrupts.  Falling back to legacy interrupts.\n");
2074 		}
2075 		/* Fall through */
2076 	case E1000E_INT_MODE_LEGACY:
2077 		/* Don't do anything; this is the system default */
2078 		break;
2079 	}
2080 
2081 	/* store the number of vectors being used */
2082 	adapter->num_vectors = 1;
2083 }
2084 
2085 /**
2086  * e1000_request_msix - Initialize MSI-X interrupts
2087  *
2088  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
2089  * kernel.
2090  **/
2091 static int e1000_request_msix(struct e1000_adapter *adapter)
2092 {
2093 	struct net_device *netdev = adapter->netdev;
2094 	int err = 0, vector = 0;
2095 
2096 	if (strlen(netdev->name) < (IFNAMSIZ - 5))
2097 		snprintf(adapter->rx_ring->name,
2098 			 sizeof(adapter->rx_ring->name) - 1,
2099 			 "%s-rx-0", netdev->name);
2100 	else
2101 		memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
2102 	err = request_irq(adapter->msix_entries[vector].vector,
2103 			  e1000_intr_msix_rx, 0, adapter->rx_ring->name,
2104 			  netdev);
2105 	if (err)
2106 		return err;
2107 	adapter->rx_ring->itr_register = adapter->hw.hw_addr +
2108 	    E1000_EITR_82574(vector);
2109 	adapter->rx_ring->itr_val = adapter->itr;
2110 	vector++;
2111 
2112 	if (strlen(netdev->name) < (IFNAMSIZ - 5))
2113 		snprintf(adapter->tx_ring->name,
2114 			 sizeof(adapter->tx_ring->name) - 1,
2115 			 "%s-tx-0", netdev->name);
2116 	else
2117 		memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
2118 	err = request_irq(adapter->msix_entries[vector].vector,
2119 			  e1000_intr_msix_tx, 0, adapter->tx_ring->name,
2120 			  netdev);
2121 	if (err)
2122 		return err;
2123 	adapter->tx_ring->itr_register = adapter->hw.hw_addr +
2124 	    E1000_EITR_82574(vector);
2125 	adapter->tx_ring->itr_val = adapter->itr;
2126 	vector++;
2127 
2128 	err = request_irq(adapter->msix_entries[vector].vector,
2129 			  e1000_msix_other, 0, netdev->name, netdev);
2130 	if (err)
2131 		return err;
2132 
2133 	e1000_configure_msix(adapter);
2134 
2135 	return 0;
2136 }
2137 
2138 /**
2139  * e1000_request_irq - initialize interrupts
2140  *
2141  * Attempts to configure interrupts using the best available
2142  * capabilities of the hardware and kernel.
2143  **/
2144 static int e1000_request_irq(struct e1000_adapter *adapter)
2145 {
2146 	struct net_device *netdev = adapter->netdev;
2147 	int err;
2148 
2149 	if (adapter->msix_entries) {
2150 		err = e1000_request_msix(adapter);
2151 		if (!err)
2152 			return err;
2153 		/* fall back to MSI */
2154 		e1000e_reset_interrupt_capability(adapter);
2155 		adapter->int_mode = E1000E_INT_MODE_MSI;
2156 		e1000e_set_interrupt_capability(adapter);
2157 	}
2158 	if (adapter->flags & FLAG_MSI_ENABLED) {
2159 		err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2160 				  netdev->name, netdev);
2161 		if (!err)
2162 			return err;
2163 
2164 		/* fall back to legacy interrupt */
2165 		e1000e_reset_interrupt_capability(adapter);
2166 		adapter->int_mode = E1000E_INT_MODE_LEGACY;
2167 	}
2168 
2169 	err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2170 			  netdev->name, netdev);
2171 	if (err)
2172 		e_err("Unable to allocate interrupt, Error: %d\n", err);
2173 
2174 	return err;
2175 }
2176 
2177 static void e1000_free_irq(struct e1000_adapter *adapter)
2178 {
2179 	struct net_device *netdev = adapter->netdev;
2180 
2181 	if (adapter->msix_entries) {
2182 		int vector = 0;
2183 
2184 		free_irq(adapter->msix_entries[vector].vector, netdev);
2185 		vector++;
2186 
2187 		free_irq(adapter->msix_entries[vector].vector, netdev);
2188 		vector++;
2189 
2190 		/* Other Causes interrupt vector */
2191 		free_irq(adapter->msix_entries[vector].vector, netdev);
2192 		return;
2193 	}
2194 
2195 	free_irq(adapter->pdev->irq, netdev);
2196 }
2197 
2198 /**
2199  * e1000_irq_disable - Mask off interrupt generation on the NIC
2200  **/
2201 static void e1000_irq_disable(struct e1000_adapter *adapter)
2202 {
2203 	struct e1000_hw *hw = &adapter->hw;
2204 
2205 	ew32(IMC, ~0);
2206 	if (adapter->msix_entries)
2207 		ew32(EIAC_82574, 0);
2208 	e1e_flush();
2209 
2210 	if (adapter->msix_entries) {
2211 		int i;
2212 
2213 		for (i = 0; i < adapter->num_vectors; i++)
2214 			synchronize_irq(adapter->msix_entries[i].vector);
2215 	} else {
2216 		synchronize_irq(adapter->pdev->irq);
2217 	}
2218 }
2219 
2220 /**
2221  * e1000_irq_enable - Enable default interrupt generation settings
2222  **/
2223 static void e1000_irq_enable(struct e1000_adapter *adapter)
2224 {
2225 	struct e1000_hw *hw = &adapter->hw;
2226 
2227 	if (adapter->msix_entries) {
2228 		ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2229 		ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
2230 	} else if (hw->mac.type == e1000_pch_lpt) {
2231 		ew32(IMS, IMS_ENABLE_MASK | E1000_IMS_ECCER);
2232 	} else {
2233 		ew32(IMS, IMS_ENABLE_MASK);
2234 	}
2235 	e1e_flush();
2236 }
2237 
2238 /**
2239  * e1000e_get_hw_control - get control of the h/w from f/w
2240  * @adapter: address of board private structure
2241  *
2242  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2243  * For ASF and Pass Through versions of f/w this means that
2244  * the driver is loaded. For AMT version (only with 82573)
2245  * of the f/w this means that the network i/f is open.
2246  **/
2247 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2248 {
2249 	struct e1000_hw *hw = &adapter->hw;
2250 	u32 ctrl_ext;
2251 	u32 swsm;
2252 
2253 	/* Let firmware know the driver has taken over */
2254 	if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2255 		swsm = er32(SWSM);
2256 		ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2257 	} else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2258 		ctrl_ext = er32(CTRL_EXT);
2259 		ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2260 	}
2261 }
2262 
2263 /**
2264  * e1000e_release_hw_control - release control of the h/w to f/w
2265  * @adapter: address of board private structure
2266  *
2267  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2268  * For ASF and Pass Through versions of f/w this means that the
2269  * driver is no longer loaded. For AMT version (only with 82573) i
2270  * of the f/w this means that the network i/f is closed.
2271  *
2272  **/
2273 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2274 {
2275 	struct e1000_hw *hw = &adapter->hw;
2276 	u32 ctrl_ext;
2277 	u32 swsm;
2278 
2279 	/* Let firmware taken over control of h/w */
2280 	if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2281 		swsm = er32(SWSM);
2282 		ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2283 	} else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2284 		ctrl_ext = er32(CTRL_EXT);
2285 		ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2286 	}
2287 }
2288 
2289 /**
2290  * e1000_alloc_ring_dma - allocate memory for a ring structure
2291  **/
2292 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2293 				struct e1000_ring *ring)
2294 {
2295 	struct pci_dev *pdev = adapter->pdev;
2296 
2297 	ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2298 					GFP_KERNEL);
2299 	if (!ring->desc)
2300 		return -ENOMEM;
2301 
2302 	return 0;
2303 }
2304 
2305 /**
2306  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2307  * @tx_ring: Tx descriptor ring
2308  *
2309  * Return 0 on success, negative on failure
2310  **/
2311 int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2312 {
2313 	struct e1000_adapter *adapter = tx_ring->adapter;
2314 	int err = -ENOMEM, size;
2315 
2316 	size = sizeof(struct e1000_buffer) * tx_ring->count;
2317 	tx_ring->buffer_info = vzalloc(size);
2318 	if (!tx_ring->buffer_info)
2319 		goto err;
2320 
2321 	/* round up to nearest 4K */
2322 	tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2323 	tx_ring->size = ALIGN(tx_ring->size, 4096);
2324 
2325 	err = e1000_alloc_ring_dma(adapter, tx_ring);
2326 	if (err)
2327 		goto err;
2328 
2329 	tx_ring->next_to_use = 0;
2330 	tx_ring->next_to_clean = 0;
2331 
2332 	return 0;
2333 err:
2334 	vfree(tx_ring->buffer_info);
2335 	e_err("Unable to allocate memory for the transmit descriptor ring\n");
2336 	return err;
2337 }
2338 
2339 /**
2340  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2341  * @rx_ring: Rx descriptor ring
2342  *
2343  * Returns 0 on success, negative on failure
2344  **/
2345 int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2346 {
2347 	struct e1000_adapter *adapter = rx_ring->adapter;
2348 	struct e1000_buffer *buffer_info;
2349 	int i, size, desc_len, err = -ENOMEM;
2350 
2351 	size = sizeof(struct e1000_buffer) * rx_ring->count;
2352 	rx_ring->buffer_info = vzalloc(size);
2353 	if (!rx_ring->buffer_info)
2354 		goto err;
2355 
2356 	for (i = 0; i < rx_ring->count; i++) {
2357 		buffer_info = &rx_ring->buffer_info[i];
2358 		buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2359 						sizeof(struct e1000_ps_page),
2360 						GFP_KERNEL);
2361 		if (!buffer_info->ps_pages)
2362 			goto err_pages;
2363 	}
2364 
2365 	desc_len = sizeof(union e1000_rx_desc_packet_split);
2366 
2367 	/* Round up to nearest 4K */
2368 	rx_ring->size = rx_ring->count * desc_len;
2369 	rx_ring->size = ALIGN(rx_ring->size, 4096);
2370 
2371 	err = e1000_alloc_ring_dma(adapter, rx_ring);
2372 	if (err)
2373 		goto err_pages;
2374 
2375 	rx_ring->next_to_clean = 0;
2376 	rx_ring->next_to_use = 0;
2377 	rx_ring->rx_skb_top = NULL;
2378 
2379 	return 0;
2380 
2381 err_pages:
2382 	for (i = 0; i < rx_ring->count; i++) {
2383 		buffer_info = &rx_ring->buffer_info[i];
2384 		kfree(buffer_info->ps_pages);
2385 	}
2386 err:
2387 	vfree(rx_ring->buffer_info);
2388 	e_err("Unable to allocate memory for the receive descriptor ring\n");
2389 	return err;
2390 }
2391 
2392 /**
2393  * e1000_clean_tx_ring - Free Tx Buffers
2394  * @tx_ring: Tx descriptor ring
2395  **/
2396 static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2397 {
2398 	struct e1000_adapter *adapter = tx_ring->adapter;
2399 	struct e1000_buffer *buffer_info;
2400 	unsigned long size;
2401 	unsigned int i;
2402 
2403 	for (i = 0; i < tx_ring->count; i++) {
2404 		buffer_info = &tx_ring->buffer_info[i];
2405 		e1000_put_txbuf(tx_ring, buffer_info);
2406 	}
2407 
2408 	netdev_reset_queue(adapter->netdev);
2409 	size = sizeof(struct e1000_buffer) * tx_ring->count;
2410 	memset(tx_ring->buffer_info, 0, size);
2411 
2412 	memset(tx_ring->desc, 0, tx_ring->size);
2413 
2414 	tx_ring->next_to_use = 0;
2415 	tx_ring->next_to_clean = 0;
2416 
2417 	writel(0, tx_ring->head);
2418 	if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
2419 		e1000e_update_tdt_wa(tx_ring, 0);
2420 	else
2421 		writel(0, tx_ring->tail);
2422 }
2423 
2424 /**
2425  * e1000e_free_tx_resources - Free Tx Resources per Queue
2426  * @tx_ring: Tx descriptor ring
2427  *
2428  * Free all transmit software resources
2429  **/
2430 void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2431 {
2432 	struct e1000_adapter *adapter = tx_ring->adapter;
2433 	struct pci_dev *pdev = adapter->pdev;
2434 
2435 	e1000_clean_tx_ring(tx_ring);
2436 
2437 	vfree(tx_ring->buffer_info);
2438 	tx_ring->buffer_info = NULL;
2439 
2440 	dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2441 			  tx_ring->dma);
2442 	tx_ring->desc = NULL;
2443 }
2444 
2445 /**
2446  * e1000e_free_rx_resources - Free Rx Resources
2447  * @rx_ring: Rx descriptor ring
2448  *
2449  * Free all receive software resources
2450  **/
2451 void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2452 {
2453 	struct e1000_adapter *adapter = rx_ring->adapter;
2454 	struct pci_dev *pdev = adapter->pdev;
2455 	int i;
2456 
2457 	e1000_clean_rx_ring(rx_ring);
2458 
2459 	for (i = 0; i < rx_ring->count; i++)
2460 		kfree(rx_ring->buffer_info[i].ps_pages);
2461 
2462 	vfree(rx_ring->buffer_info);
2463 	rx_ring->buffer_info = NULL;
2464 
2465 	dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2466 			  rx_ring->dma);
2467 	rx_ring->desc = NULL;
2468 }
2469 
2470 /**
2471  * e1000_update_itr - update the dynamic ITR value based on statistics
2472  * @adapter: pointer to adapter
2473  * @itr_setting: current adapter->itr
2474  * @packets: the number of packets during this measurement interval
2475  * @bytes: the number of bytes during this measurement interval
2476  *
2477  *      Stores a new ITR value based on packets and byte
2478  *      counts during the last interrupt.  The advantage of per interrupt
2479  *      computation is faster updates and more accurate ITR for the current
2480  *      traffic pattern.  Constants in this function were computed
2481  *      based on theoretical maximum wire speed and thresholds were set based
2482  *      on testing data as well as attempting to minimize response time
2483  *      while increasing bulk throughput.  This functionality is controlled
2484  *      by the InterruptThrottleRate module parameter.
2485  **/
2486 static unsigned int e1000_update_itr(u16 itr_setting, int packets, int bytes)
2487 {
2488 	unsigned int retval = itr_setting;
2489 
2490 	if (packets == 0)
2491 		return itr_setting;
2492 
2493 	switch (itr_setting) {
2494 	case lowest_latency:
2495 		/* handle TSO and jumbo frames */
2496 		if (bytes / packets > 8000)
2497 			retval = bulk_latency;
2498 		else if ((packets < 5) && (bytes > 512))
2499 			retval = low_latency;
2500 		break;
2501 	case low_latency:	/* 50 usec aka 20000 ints/s */
2502 		if (bytes > 10000) {
2503 			/* this if handles the TSO accounting */
2504 			if (bytes / packets > 8000)
2505 				retval = bulk_latency;
2506 			else if ((packets < 10) || ((bytes / packets) > 1200))
2507 				retval = bulk_latency;
2508 			else if ((packets > 35))
2509 				retval = lowest_latency;
2510 		} else if (bytes / packets > 2000) {
2511 			retval = bulk_latency;
2512 		} else if (packets <= 2 && bytes < 512) {
2513 			retval = lowest_latency;
2514 		}
2515 		break;
2516 	case bulk_latency:	/* 250 usec aka 4000 ints/s */
2517 		if (bytes > 25000) {
2518 			if (packets > 35)
2519 				retval = low_latency;
2520 		} else if (bytes < 6000) {
2521 			retval = low_latency;
2522 		}
2523 		break;
2524 	}
2525 
2526 	return retval;
2527 }
2528 
2529 static void e1000_set_itr(struct e1000_adapter *adapter)
2530 {
2531 	u16 current_itr;
2532 	u32 new_itr = adapter->itr;
2533 
2534 	/* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2535 	if (adapter->link_speed != SPEED_1000) {
2536 		current_itr = 0;
2537 		new_itr = 4000;
2538 		goto set_itr_now;
2539 	}
2540 
2541 	if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2542 		new_itr = 0;
2543 		goto set_itr_now;
2544 	}
2545 
2546 	adapter->tx_itr = e1000_update_itr(adapter->tx_itr,
2547 					   adapter->total_tx_packets,
2548 					   adapter->total_tx_bytes);
2549 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
2550 	if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2551 		adapter->tx_itr = low_latency;
2552 
2553 	adapter->rx_itr = e1000_update_itr(adapter->rx_itr,
2554 					   adapter->total_rx_packets,
2555 					   adapter->total_rx_bytes);
2556 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
2557 	if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2558 		adapter->rx_itr = low_latency;
2559 
2560 	current_itr = max(adapter->rx_itr, adapter->tx_itr);
2561 
2562 	/* counts and packets in update_itr are dependent on these numbers */
2563 	switch (current_itr) {
2564 	case lowest_latency:
2565 		new_itr = 70000;
2566 		break;
2567 	case low_latency:
2568 		new_itr = 20000;	/* aka hwitr = ~200 */
2569 		break;
2570 	case bulk_latency:
2571 		new_itr = 4000;
2572 		break;
2573 	default:
2574 		break;
2575 	}
2576 
2577 set_itr_now:
2578 	if (new_itr != adapter->itr) {
2579 		/* this attempts to bias the interrupt rate towards Bulk
2580 		 * by adding intermediate steps when interrupt rate is
2581 		 * increasing
2582 		 */
2583 		new_itr = new_itr > adapter->itr ?
2584 		    min(adapter->itr + (new_itr >> 2), new_itr) : new_itr;
2585 		adapter->itr = new_itr;
2586 		adapter->rx_ring->itr_val = new_itr;
2587 		if (adapter->msix_entries)
2588 			adapter->rx_ring->set_itr = 1;
2589 		else
2590 			e1000e_write_itr(adapter, new_itr);
2591 	}
2592 }
2593 
2594 /**
2595  * e1000e_write_itr - write the ITR value to the appropriate registers
2596  * @adapter: address of board private structure
2597  * @itr: new ITR value to program
2598  *
2599  * e1000e_write_itr determines if the adapter is in MSI-X mode
2600  * and, if so, writes the EITR registers with the ITR value.
2601  * Otherwise, it writes the ITR value into the ITR register.
2602  **/
2603 void e1000e_write_itr(struct e1000_adapter *adapter, u32 itr)
2604 {
2605 	struct e1000_hw *hw = &adapter->hw;
2606 	u32 new_itr = itr ? 1000000000 / (itr * 256) : 0;
2607 
2608 	if (adapter->msix_entries) {
2609 		int vector;
2610 
2611 		for (vector = 0; vector < adapter->num_vectors; vector++)
2612 			writel(new_itr, hw->hw_addr + E1000_EITR_82574(vector));
2613 	} else {
2614 		ew32(ITR, new_itr);
2615 	}
2616 }
2617 
2618 /**
2619  * e1000_alloc_queues - Allocate memory for all rings
2620  * @adapter: board private structure to initialize
2621  **/
2622 static int e1000_alloc_queues(struct e1000_adapter *adapter)
2623 {
2624 	int size = sizeof(struct e1000_ring);
2625 
2626 	adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2627 	if (!adapter->tx_ring)
2628 		goto err;
2629 	adapter->tx_ring->count = adapter->tx_ring_count;
2630 	adapter->tx_ring->adapter = adapter;
2631 
2632 	adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2633 	if (!adapter->rx_ring)
2634 		goto err;
2635 	adapter->rx_ring->count = adapter->rx_ring_count;
2636 	adapter->rx_ring->adapter = adapter;
2637 
2638 	return 0;
2639 err:
2640 	e_err("Unable to allocate memory for queues\n");
2641 	kfree(adapter->rx_ring);
2642 	kfree(adapter->tx_ring);
2643 	return -ENOMEM;
2644 }
2645 
2646 /**
2647  * e1000e_poll - NAPI Rx polling callback
2648  * @napi: struct associated with this polling callback
2649  * @weight: number of packets driver is allowed to process this poll
2650  **/
2651 static int e1000e_poll(struct napi_struct *napi, int weight)
2652 {
2653 	struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
2654 						     napi);
2655 	struct e1000_hw *hw = &adapter->hw;
2656 	struct net_device *poll_dev = adapter->netdev;
2657 	int tx_cleaned = 1, work_done = 0;
2658 
2659 	adapter = netdev_priv(poll_dev);
2660 
2661 	if (!adapter->msix_entries ||
2662 	    (adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2663 		tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2664 
2665 	adapter->clean_rx(adapter->rx_ring, &work_done, weight);
2666 
2667 	if (!tx_cleaned)
2668 		work_done = weight;
2669 
2670 	/* If weight not fully consumed, exit the polling mode */
2671 	if (work_done < weight) {
2672 		if (adapter->itr_setting & 3)
2673 			e1000_set_itr(adapter);
2674 		napi_complete(napi);
2675 		if (!test_bit(__E1000_DOWN, &adapter->state)) {
2676 			if (adapter->msix_entries)
2677 				ew32(IMS, adapter->rx_ring->ims_val);
2678 			else
2679 				e1000_irq_enable(adapter);
2680 		}
2681 	}
2682 
2683 	return work_done;
2684 }
2685 
2686 static int e1000_vlan_rx_add_vid(struct net_device *netdev,
2687 				 __always_unused __be16 proto, u16 vid)
2688 {
2689 	struct e1000_adapter *adapter = netdev_priv(netdev);
2690 	struct e1000_hw *hw = &adapter->hw;
2691 	u32 vfta, index;
2692 
2693 	/* don't update vlan cookie if already programmed */
2694 	if ((adapter->hw.mng_cookie.status &
2695 	     E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2696 	    (vid == adapter->mng_vlan_id))
2697 		return 0;
2698 
2699 	/* add VID to filter table */
2700 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2701 		index = (vid >> 5) & 0x7F;
2702 		vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2703 		vfta |= (1 << (vid & 0x1F));
2704 		hw->mac.ops.write_vfta(hw, index, vfta);
2705 	}
2706 
2707 	set_bit(vid, adapter->active_vlans);
2708 
2709 	return 0;
2710 }
2711 
2712 static int e1000_vlan_rx_kill_vid(struct net_device *netdev,
2713 				  __always_unused __be16 proto, u16 vid)
2714 {
2715 	struct e1000_adapter *adapter = netdev_priv(netdev);
2716 	struct e1000_hw *hw = &adapter->hw;
2717 	u32 vfta, index;
2718 
2719 	if ((adapter->hw.mng_cookie.status &
2720 	     E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2721 	    (vid == adapter->mng_vlan_id)) {
2722 		/* release control to f/w */
2723 		e1000e_release_hw_control(adapter);
2724 		return 0;
2725 	}
2726 
2727 	/* remove VID from filter table */
2728 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2729 		index = (vid >> 5) & 0x7F;
2730 		vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2731 		vfta &= ~(1 << (vid & 0x1F));
2732 		hw->mac.ops.write_vfta(hw, index, vfta);
2733 	}
2734 
2735 	clear_bit(vid, adapter->active_vlans);
2736 
2737 	return 0;
2738 }
2739 
2740 /**
2741  * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2742  * @adapter: board private structure to initialize
2743  **/
2744 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2745 {
2746 	struct net_device *netdev = adapter->netdev;
2747 	struct e1000_hw *hw = &adapter->hw;
2748 	u32 rctl;
2749 
2750 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2751 		/* disable VLAN receive filtering */
2752 		rctl = er32(RCTL);
2753 		rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2754 		ew32(RCTL, rctl);
2755 
2756 		if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2757 			e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
2758 					       adapter->mng_vlan_id);
2759 			adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2760 		}
2761 	}
2762 }
2763 
2764 /**
2765  * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2766  * @adapter: board private structure to initialize
2767  **/
2768 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2769 {
2770 	struct e1000_hw *hw = &adapter->hw;
2771 	u32 rctl;
2772 
2773 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2774 		/* enable VLAN receive filtering */
2775 		rctl = er32(RCTL);
2776 		rctl |= E1000_RCTL_VFE;
2777 		rctl &= ~E1000_RCTL_CFIEN;
2778 		ew32(RCTL, rctl);
2779 	}
2780 }
2781 
2782 /**
2783  * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2784  * @adapter: board private structure to initialize
2785  **/
2786 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2787 {
2788 	struct e1000_hw *hw = &adapter->hw;
2789 	u32 ctrl;
2790 
2791 	/* disable VLAN tag insert/strip */
2792 	ctrl = er32(CTRL);
2793 	ctrl &= ~E1000_CTRL_VME;
2794 	ew32(CTRL, ctrl);
2795 }
2796 
2797 /**
2798  * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2799  * @adapter: board private structure to initialize
2800  **/
2801 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2802 {
2803 	struct e1000_hw *hw = &adapter->hw;
2804 	u32 ctrl;
2805 
2806 	/* enable VLAN tag insert/strip */
2807 	ctrl = er32(CTRL);
2808 	ctrl |= E1000_CTRL_VME;
2809 	ew32(CTRL, ctrl);
2810 }
2811 
2812 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2813 {
2814 	struct net_device *netdev = adapter->netdev;
2815 	u16 vid = adapter->hw.mng_cookie.vlan_id;
2816 	u16 old_vid = adapter->mng_vlan_id;
2817 
2818 	if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2819 		e1000_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
2820 		adapter->mng_vlan_id = vid;
2821 	}
2822 
2823 	if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2824 		e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q), old_vid);
2825 }
2826 
2827 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2828 {
2829 	u16 vid;
2830 
2831 	e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
2832 
2833 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2834 	    e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
2835 }
2836 
2837 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2838 {
2839 	struct e1000_hw *hw = &adapter->hw;
2840 	u32 manc, manc2h, mdef, i, j;
2841 
2842 	if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2843 		return;
2844 
2845 	manc = er32(MANC);
2846 
2847 	/* enable receiving management packets to the host. this will probably
2848 	 * generate destination unreachable messages from the host OS, but
2849 	 * the packets will be handled on SMBUS
2850 	 */
2851 	manc |= E1000_MANC_EN_MNG2HOST;
2852 	manc2h = er32(MANC2H);
2853 
2854 	switch (hw->mac.type) {
2855 	default:
2856 		manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2857 		break;
2858 	case e1000_82574:
2859 	case e1000_82583:
2860 		/* Check if IPMI pass-through decision filter already exists;
2861 		 * if so, enable it.
2862 		 */
2863 		for (i = 0, j = 0; i < 8; i++) {
2864 			mdef = er32(MDEF(i));
2865 
2866 			/* Ignore filters with anything other than IPMI ports */
2867 			if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2868 				continue;
2869 
2870 			/* Enable this decision filter in MANC2H */
2871 			if (mdef)
2872 				manc2h |= (1 << i);
2873 
2874 			j |= mdef;
2875 		}
2876 
2877 		if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2878 			break;
2879 
2880 		/* Create new decision filter in an empty filter */
2881 		for (i = 0, j = 0; i < 8; i++)
2882 			if (er32(MDEF(i)) == 0) {
2883 				ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2884 					       E1000_MDEF_PORT_664));
2885 				manc2h |= (1 << 1);
2886 				j++;
2887 				break;
2888 			}
2889 
2890 		if (!j)
2891 			e_warn("Unable to create IPMI pass-through filter\n");
2892 		break;
2893 	}
2894 
2895 	ew32(MANC2H, manc2h);
2896 	ew32(MANC, manc);
2897 }
2898 
2899 /**
2900  * e1000_configure_tx - Configure Transmit Unit after Reset
2901  * @adapter: board private structure
2902  *
2903  * Configure the Tx unit of the MAC after a reset.
2904  **/
2905 static void e1000_configure_tx(struct e1000_adapter *adapter)
2906 {
2907 	struct e1000_hw *hw = &adapter->hw;
2908 	struct e1000_ring *tx_ring = adapter->tx_ring;
2909 	u64 tdba;
2910 	u32 tdlen, tctl, tarc;
2911 
2912 	/* Setup the HW Tx Head and Tail descriptor pointers */
2913 	tdba = tx_ring->dma;
2914 	tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2915 	ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
2916 	ew32(TDBAH(0), (tdba >> 32));
2917 	ew32(TDLEN(0), tdlen);
2918 	ew32(TDH(0), 0);
2919 	ew32(TDT(0), 0);
2920 	tx_ring->head = adapter->hw.hw_addr + E1000_TDH(0);
2921 	tx_ring->tail = adapter->hw.hw_addr + E1000_TDT(0);
2922 
2923 	/* Set the Tx Interrupt Delay register */
2924 	ew32(TIDV, adapter->tx_int_delay);
2925 	/* Tx irq moderation */
2926 	ew32(TADV, adapter->tx_abs_int_delay);
2927 
2928 	if (adapter->flags2 & FLAG2_DMA_BURST) {
2929 		u32 txdctl = er32(TXDCTL(0));
2930 
2931 		txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2932 			    E1000_TXDCTL_WTHRESH);
2933 		/* set up some performance related parameters to encourage the
2934 		 * hardware to use the bus more efficiently in bursts, depends
2935 		 * on the tx_int_delay to be enabled,
2936 		 * wthresh = 1 ==> burst write is disabled to avoid Tx stalls
2937 		 * hthresh = 1 ==> prefetch when one or more available
2938 		 * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2939 		 * BEWARE: this seems to work but should be considered first if
2940 		 * there are Tx hangs or other Tx related bugs
2941 		 */
2942 		txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2943 		ew32(TXDCTL(0), txdctl);
2944 	}
2945 	/* erratum work around: set txdctl the same for both queues */
2946 	ew32(TXDCTL(1), er32(TXDCTL(0)));
2947 
2948 	/* Program the Transmit Control Register */
2949 	tctl = er32(TCTL);
2950 	tctl &= ~E1000_TCTL_CT;
2951 	tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2952 		(E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2953 
2954 	if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2955 		tarc = er32(TARC(0));
2956 		/* set the speed mode bit, we'll clear it if we're not at
2957 		 * gigabit link later
2958 		 */
2959 #define SPEED_MODE_BIT (1 << 21)
2960 		tarc |= SPEED_MODE_BIT;
2961 		ew32(TARC(0), tarc);
2962 	}
2963 
2964 	/* errata: program both queues to unweighted RR */
2965 	if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2966 		tarc = er32(TARC(0));
2967 		tarc |= 1;
2968 		ew32(TARC(0), tarc);
2969 		tarc = er32(TARC(1));
2970 		tarc |= 1;
2971 		ew32(TARC(1), tarc);
2972 	}
2973 
2974 	/* Setup Transmit Descriptor Settings for eop descriptor */
2975 	adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2976 
2977 	/* only set IDE if we are delaying interrupts using the timers */
2978 	if (adapter->tx_int_delay)
2979 		adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2980 
2981 	/* enable Report Status bit */
2982 	adapter->txd_cmd |= E1000_TXD_CMD_RS;
2983 
2984 	ew32(TCTL, tctl);
2985 
2986 	hw->mac.ops.config_collision_dist(hw);
2987 }
2988 
2989 /**
2990  * e1000_setup_rctl - configure the receive control registers
2991  * @adapter: Board private structure
2992  **/
2993 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2994 			   (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2995 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2996 {
2997 	struct e1000_hw *hw = &adapter->hw;
2998 	u32 rctl, rfctl;
2999 	u32 pages = 0;
3000 
3001 	/* Workaround Si errata on PCHx - configure jumbo frame flow.
3002 	 * If jumbo frames not set, program related MAC/PHY registers
3003 	 * to h/w defaults
3004 	 */
3005 	if (hw->mac.type >= e1000_pch2lan) {
3006 		s32 ret_val;
3007 
3008 		if (adapter->netdev->mtu > ETH_DATA_LEN)
3009 			ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
3010 		else
3011 			ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
3012 
3013 		if (ret_val)
3014 			e_dbg("failed to enable|disable jumbo frame workaround mode\n");
3015 	}
3016 
3017 	/* Program MC offset vector base */
3018 	rctl = er32(RCTL);
3019 	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3020 	rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
3021 	    E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
3022 	    (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3023 
3024 	/* Do not Store bad packets */
3025 	rctl &= ~E1000_RCTL_SBP;
3026 
3027 	/* Enable Long Packet receive */
3028 	if (adapter->netdev->mtu <= ETH_DATA_LEN)
3029 		rctl &= ~E1000_RCTL_LPE;
3030 	else
3031 		rctl |= E1000_RCTL_LPE;
3032 
3033 	/* Some systems expect that the CRC is included in SMBUS traffic. The
3034 	 * hardware strips the CRC before sending to both SMBUS (BMC) and to
3035 	 * host memory when this is enabled
3036 	 */
3037 	if (adapter->flags2 & FLAG2_CRC_STRIPPING)
3038 		rctl |= E1000_RCTL_SECRC;
3039 
3040 	/* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
3041 	if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
3042 		u16 phy_data;
3043 
3044 		e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
3045 		phy_data &= 0xfff8;
3046 		phy_data |= (1 << 2);
3047 		e1e_wphy(hw, PHY_REG(770, 26), phy_data);
3048 
3049 		e1e_rphy(hw, 22, &phy_data);
3050 		phy_data &= 0x0fff;
3051 		phy_data |= (1 << 14);
3052 		e1e_wphy(hw, 0x10, 0x2823);
3053 		e1e_wphy(hw, 0x11, 0x0003);
3054 		e1e_wphy(hw, 22, phy_data);
3055 	}
3056 
3057 	/* Setup buffer sizes */
3058 	rctl &= ~E1000_RCTL_SZ_4096;
3059 	rctl |= E1000_RCTL_BSEX;
3060 	switch (adapter->rx_buffer_len) {
3061 	case 2048:
3062 	default:
3063 		rctl |= E1000_RCTL_SZ_2048;
3064 		rctl &= ~E1000_RCTL_BSEX;
3065 		break;
3066 	case 4096:
3067 		rctl |= E1000_RCTL_SZ_4096;
3068 		break;
3069 	case 8192:
3070 		rctl |= E1000_RCTL_SZ_8192;
3071 		break;
3072 	case 16384:
3073 		rctl |= E1000_RCTL_SZ_16384;
3074 		break;
3075 	}
3076 
3077 	/* Enable Extended Status in all Receive Descriptors */
3078 	rfctl = er32(RFCTL);
3079 	rfctl |= E1000_RFCTL_EXTEN;
3080 	ew32(RFCTL, rfctl);
3081 
3082 	/* 82571 and greater support packet-split where the protocol
3083 	 * header is placed in skb->data and the packet data is
3084 	 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
3085 	 * In the case of a non-split, skb->data is linearly filled,
3086 	 * followed by the page buffers.  Therefore, skb->data is
3087 	 * sized to hold the largest protocol header.
3088 	 *
3089 	 * allocations using alloc_page take too long for regular MTU
3090 	 * so only enable packet split for jumbo frames
3091 	 *
3092 	 * Using pages when the page size is greater than 16k wastes
3093 	 * a lot of memory, since we allocate 3 pages at all times
3094 	 * per packet.
3095 	 */
3096 	pages = PAGE_USE_COUNT(adapter->netdev->mtu);
3097 	if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
3098 		adapter->rx_ps_pages = pages;
3099 	else
3100 		adapter->rx_ps_pages = 0;
3101 
3102 	if (adapter->rx_ps_pages) {
3103 		u32 psrctl = 0;
3104 
3105 		/* Enable Packet split descriptors */
3106 		rctl |= E1000_RCTL_DTYP_PS;
3107 
3108 		psrctl |= adapter->rx_ps_bsize0 >> E1000_PSRCTL_BSIZE0_SHIFT;
3109 
3110 		switch (adapter->rx_ps_pages) {
3111 		case 3:
3112 			psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE3_SHIFT;
3113 			/* fall-through */
3114 		case 2:
3115 			psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE2_SHIFT;
3116 			/* fall-through */
3117 		case 1:
3118 			psrctl |= PAGE_SIZE >> E1000_PSRCTL_BSIZE1_SHIFT;
3119 			break;
3120 		}
3121 
3122 		ew32(PSRCTL, psrctl);
3123 	}
3124 
3125 	/* This is useful for sniffing bad packets. */
3126 	if (adapter->netdev->features & NETIF_F_RXALL) {
3127 		/* UPE and MPE will be handled by normal PROMISC logic
3128 		 * in e1000e_set_rx_mode
3129 		 */
3130 		rctl |= (E1000_RCTL_SBP |	/* Receive bad packets */
3131 			 E1000_RCTL_BAM |	/* RX All Bcast Pkts */
3132 			 E1000_RCTL_PMCF);	/* RX All MAC Ctrl Pkts */
3133 
3134 		rctl &= ~(E1000_RCTL_VFE |	/* Disable VLAN filter */
3135 			  E1000_RCTL_DPF |	/* Allow filtered pause */
3136 			  E1000_RCTL_CFIEN);	/* Dis VLAN CFIEN Filter */
3137 		/* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3138 		 * and that breaks VLANs.
3139 		 */
3140 	}
3141 
3142 	ew32(RCTL, rctl);
3143 	/* just started the receive unit, no need to restart */
3144 	adapter->flags &= ~FLAG_RESTART_NOW;
3145 }
3146 
3147 /**
3148  * e1000_configure_rx - Configure Receive Unit after Reset
3149  * @adapter: board private structure
3150  *
3151  * Configure the Rx unit of the MAC after a reset.
3152  **/
3153 static void e1000_configure_rx(struct e1000_adapter *adapter)
3154 {
3155 	struct e1000_hw *hw = &adapter->hw;
3156 	struct e1000_ring *rx_ring = adapter->rx_ring;
3157 	u64 rdba;
3158 	u32 rdlen, rctl, rxcsum, ctrl_ext;
3159 
3160 	if (adapter->rx_ps_pages) {
3161 		/* this is a 32 byte descriptor */
3162 		rdlen = rx_ring->count *
3163 		    sizeof(union e1000_rx_desc_packet_split);
3164 		adapter->clean_rx = e1000_clean_rx_irq_ps;
3165 		adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3166 	} else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3167 		rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3168 		adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3169 		adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3170 	} else {
3171 		rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3172 		adapter->clean_rx = e1000_clean_rx_irq;
3173 		adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3174 	}
3175 
3176 	/* disable receives while setting up the descriptors */
3177 	rctl = er32(RCTL);
3178 	if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3179 		ew32(RCTL, rctl & ~E1000_RCTL_EN);
3180 	e1e_flush();
3181 	usleep_range(10000, 20000);
3182 
3183 	if (adapter->flags2 & FLAG2_DMA_BURST) {
3184 		/* set the writeback threshold (only takes effect if the RDTR
3185 		 * is set). set GRAN=1 and write back up to 0x4 worth, and
3186 		 * enable prefetching of 0x20 Rx descriptors
3187 		 * granularity = 01
3188 		 * wthresh = 04,
3189 		 * hthresh = 04,
3190 		 * pthresh = 0x20
3191 		 */
3192 		ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3193 		ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3194 
3195 		/* override the delay timers for enabling bursting, only if
3196 		 * the value was not set by the user via module options
3197 		 */
3198 		if (adapter->rx_int_delay == DEFAULT_RDTR)
3199 			adapter->rx_int_delay = BURST_RDTR;
3200 		if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3201 			adapter->rx_abs_int_delay = BURST_RADV;
3202 	}
3203 
3204 	/* set the Receive Delay Timer Register */
3205 	ew32(RDTR, adapter->rx_int_delay);
3206 
3207 	/* irq moderation */
3208 	ew32(RADV, adapter->rx_abs_int_delay);
3209 	if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3210 		e1000e_write_itr(adapter, adapter->itr);
3211 
3212 	ctrl_ext = er32(CTRL_EXT);
3213 	/* Auto-Mask interrupts upon ICR access */
3214 	ctrl_ext |= E1000_CTRL_EXT_IAME;
3215 	ew32(IAM, 0xffffffff);
3216 	ew32(CTRL_EXT, ctrl_ext);
3217 	e1e_flush();
3218 
3219 	/* Setup the HW Rx Head and Tail Descriptor Pointers and
3220 	 * the Base and Length of the Rx Descriptor Ring
3221 	 */
3222 	rdba = rx_ring->dma;
3223 	ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
3224 	ew32(RDBAH(0), (rdba >> 32));
3225 	ew32(RDLEN(0), rdlen);
3226 	ew32(RDH(0), 0);
3227 	ew32(RDT(0), 0);
3228 	rx_ring->head = adapter->hw.hw_addr + E1000_RDH(0);
3229 	rx_ring->tail = adapter->hw.hw_addr + E1000_RDT(0);
3230 
3231 	/* Enable Receive Checksum Offload for TCP and UDP */
3232 	rxcsum = er32(RXCSUM);
3233 	if (adapter->netdev->features & NETIF_F_RXCSUM)
3234 		rxcsum |= E1000_RXCSUM_TUOFL;
3235 	else
3236 		rxcsum &= ~E1000_RXCSUM_TUOFL;
3237 	ew32(RXCSUM, rxcsum);
3238 
3239 	/* With jumbo frames, excessive C-state transition latencies result
3240 	 * in dropped transactions.
3241 	 */
3242 	if (adapter->netdev->mtu > ETH_DATA_LEN) {
3243 		u32 lat =
3244 		    ((er32(PBA) & E1000_PBA_RXA_MASK) * 1024 -
3245 		     adapter->max_frame_size) * 8 / 1000;
3246 
3247 		if (adapter->flags & FLAG_IS_ICH) {
3248 			u32 rxdctl = er32(RXDCTL(0));
3249 
3250 			ew32(RXDCTL(0), rxdctl | 0x3);
3251 		}
3252 
3253 		pm_qos_update_request(&adapter->netdev->pm_qos_req, lat);
3254 	} else {
3255 		pm_qos_update_request(&adapter->netdev->pm_qos_req,
3256 				      PM_QOS_DEFAULT_VALUE);
3257 	}
3258 
3259 	/* Enable Receives */
3260 	ew32(RCTL, rctl);
3261 }
3262 
3263 /**
3264  * e1000e_write_mc_addr_list - write multicast addresses to MTA
3265  * @netdev: network interface device structure
3266  *
3267  * Writes multicast address list to the MTA hash table.
3268  * Returns: -ENOMEM on failure
3269  *                0 on no addresses written
3270  *                X on writing X addresses to MTA
3271  */
3272 static int e1000e_write_mc_addr_list(struct net_device *netdev)
3273 {
3274 	struct e1000_adapter *adapter = netdev_priv(netdev);
3275 	struct e1000_hw *hw = &adapter->hw;
3276 	struct netdev_hw_addr *ha;
3277 	u8 *mta_list;
3278 	int i;
3279 
3280 	if (netdev_mc_empty(netdev)) {
3281 		/* nothing to program, so clear mc list */
3282 		hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3283 		return 0;
3284 	}
3285 
3286 	mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3287 	if (!mta_list)
3288 		return -ENOMEM;
3289 
3290 	/* update_mc_addr_list expects a packed array of only addresses. */
3291 	i = 0;
3292 	netdev_for_each_mc_addr(ha, netdev)
3293 	    memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3294 
3295 	hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3296 	kfree(mta_list);
3297 
3298 	return netdev_mc_count(netdev);
3299 }
3300 
3301 /**
3302  * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3303  * @netdev: network interface device structure
3304  *
3305  * Writes unicast address list to the RAR table.
3306  * Returns: -ENOMEM on failure/insufficient address space
3307  *                0 on no addresses written
3308  *                X on writing X addresses to the RAR table
3309  **/
3310 static int e1000e_write_uc_addr_list(struct net_device *netdev)
3311 {
3312 	struct e1000_adapter *adapter = netdev_priv(netdev);
3313 	struct e1000_hw *hw = &adapter->hw;
3314 	unsigned int rar_entries = hw->mac.rar_entry_count;
3315 	int count = 0;
3316 
3317 	/* save a rar entry for our hardware address */
3318 	rar_entries--;
3319 
3320 	/* save a rar entry for the LAA workaround */
3321 	if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3322 		rar_entries--;
3323 
3324 	/* return ENOMEM indicating insufficient memory for addresses */
3325 	if (netdev_uc_count(netdev) > rar_entries)
3326 		return -ENOMEM;
3327 
3328 	if (!netdev_uc_empty(netdev) && rar_entries) {
3329 		struct netdev_hw_addr *ha;
3330 
3331 		/* write the addresses in reverse order to avoid write
3332 		 * combining
3333 		 */
3334 		netdev_for_each_uc_addr(ha, netdev) {
3335 			if (!rar_entries)
3336 				break;
3337 			hw->mac.ops.rar_set(hw, ha->addr, rar_entries--);
3338 			count++;
3339 		}
3340 	}
3341 
3342 	/* zero out the remaining RAR entries not used above */
3343 	for (; rar_entries > 0; rar_entries--) {
3344 		ew32(RAH(rar_entries), 0);
3345 		ew32(RAL(rar_entries), 0);
3346 	}
3347 	e1e_flush();
3348 
3349 	return count;
3350 }
3351 
3352 /**
3353  * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3354  * @netdev: network interface device structure
3355  *
3356  * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3357  * address list or the network interface flags are updated.  This routine is
3358  * responsible for configuring the hardware for proper unicast, multicast,
3359  * promiscuous mode, and all-multi behavior.
3360  **/
3361 static void e1000e_set_rx_mode(struct net_device *netdev)
3362 {
3363 	struct e1000_adapter *adapter = netdev_priv(netdev);
3364 	struct e1000_hw *hw = &adapter->hw;
3365 	u32 rctl;
3366 
3367 	if (pm_runtime_suspended(netdev->dev.parent))
3368 		return;
3369 
3370 	/* Check for Promiscuous and All Multicast modes */
3371 	rctl = er32(RCTL);
3372 
3373 	/* clear the affected bits */
3374 	rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3375 
3376 	if (netdev->flags & IFF_PROMISC) {
3377 		rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3378 		/* Do not hardware filter VLANs in promisc mode */
3379 		e1000e_vlan_filter_disable(adapter);
3380 	} else {
3381 		int count;
3382 
3383 		if (netdev->flags & IFF_ALLMULTI) {
3384 			rctl |= E1000_RCTL_MPE;
3385 		} else {
3386 			/* Write addresses to the MTA, if the attempt fails
3387 			 * then we should just turn on promiscuous mode so
3388 			 * that we can at least receive multicast traffic
3389 			 */
3390 			count = e1000e_write_mc_addr_list(netdev);
3391 			if (count < 0)
3392 				rctl |= E1000_RCTL_MPE;
3393 		}
3394 		e1000e_vlan_filter_enable(adapter);
3395 		/* Write addresses to available RAR registers, if there is not
3396 		 * sufficient space to store all the addresses then enable
3397 		 * unicast promiscuous mode
3398 		 */
3399 		count = e1000e_write_uc_addr_list(netdev);
3400 		if (count < 0)
3401 			rctl |= E1000_RCTL_UPE;
3402 	}
3403 
3404 	ew32(RCTL, rctl);
3405 
3406 	if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
3407 		e1000e_vlan_strip_enable(adapter);
3408 	else
3409 		e1000e_vlan_strip_disable(adapter);
3410 }
3411 
3412 static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3413 {
3414 	struct e1000_hw *hw = &adapter->hw;
3415 	u32 mrqc, rxcsum;
3416 	int i;
3417 	static const u32 rsskey[10] = {
3418 		0xda565a6d, 0xc20e5b25, 0x3d256741, 0xb08fa343, 0xcb2bcad0,
3419 		0xb4307bae, 0xa32dcb77, 0x0cf23080, 0x3bb7426a, 0xfa01acbe
3420 	};
3421 
3422 	/* Fill out hash function seed */
3423 	for (i = 0; i < 10; i++)
3424 		ew32(RSSRK(i), rsskey[i]);
3425 
3426 	/* Direct all traffic to queue 0 */
3427 	for (i = 0; i < 32; i++)
3428 		ew32(RETA(i), 0);
3429 
3430 	/* Disable raw packet checksumming so that RSS hash is placed in
3431 	 * descriptor on writeback.
3432 	 */
3433 	rxcsum = er32(RXCSUM);
3434 	rxcsum |= E1000_RXCSUM_PCSD;
3435 
3436 	ew32(RXCSUM, rxcsum);
3437 
3438 	mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3439 		E1000_MRQC_RSS_FIELD_IPV4_TCP |
3440 		E1000_MRQC_RSS_FIELD_IPV6 |
3441 		E1000_MRQC_RSS_FIELD_IPV6_TCP |
3442 		E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3443 
3444 	ew32(MRQC, mrqc);
3445 }
3446 
3447 /**
3448  * e1000e_get_base_timinca - get default SYSTIM time increment attributes
3449  * @adapter: board private structure
3450  * @timinca: pointer to returned time increment attributes
3451  *
3452  * Get attributes for incrementing the System Time Register SYSTIML/H at
3453  * the default base frequency, and set the cyclecounter shift value.
3454  **/
3455 s32 e1000e_get_base_timinca(struct e1000_adapter *adapter, u32 *timinca)
3456 {
3457 	struct e1000_hw *hw = &adapter->hw;
3458 	u32 incvalue, incperiod, shift;
3459 
3460 	/* Make sure clock is enabled on I217 before checking the frequency */
3461 	if ((hw->mac.type == e1000_pch_lpt) &&
3462 	    !(er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) &&
3463 	    !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_ENABLED)) {
3464 		u32 fextnvm7 = er32(FEXTNVM7);
3465 
3466 		if (!(fextnvm7 & (1 << 0))) {
3467 			ew32(FEXTNVM7, fextnvm7 | (1 << 0));
3468 			e1e_flush();
3469 		}
3470 	}
3471 
3472 	switch (hw->mac.type) {
3473 	case e1000_pch2lan:
3474 	case e1000_pch_lpt:
3475 		/* On I217, the clock frequency is 25MHz or 96MHz as
3476 		 * indicated by the System Clock Frequency Indication
3477 		 */
3478 		if ((hw->mac.type != e1000_pch_lpt) ||
3479 		    (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI)) {
3480 			/* Stable 96MHz frequency */
3481 			incperiod = INCPERIOD_96MHz;
3482 			incvalue = INCVALUE_96MHz;
3483 			shift = INCVALUE_SHIFT_96MHz;
3484 			adapter->cc.shift = shift + INCPERIOD_SHIFT_96MHz;
3485 			break;
3486 		}
3487 		/* fall-through */
3488 	case e1000_82574:
3489 	case e1000_82583:
3490 		/* Stable 25MHz frequency */
3491 		incperiod = INCPERIOD_25MHz;
3492 		incvalue = INCVALUE_25MHz;
3493 		shift = INCVALUE_SHIFT_25MHz;
3494 		adapter->cc.shift = shift;
3495 		break;
3496 	default:
3497 		return -EINVAL;
3498 	}
3499 
3500 	*timinca = ((incperiod << E1000_TIMINCA_INCPERIOD_SHIFT) |
3501 		    ((incvalue << shift) & E1000_TIMINCA_INCVALUE_MASK));
3502 
3503 	return 0;
3504 }
3505 
3506 /**
3507  * e1000e_config_hwtstamp - configure the hwtstamp registers and enable/disable
3508  * @adapter: board private structure
3509  *
3510  * Outgoing time stamping can be enabled and disabled. Play nice and
3511  * disable it when requested, although it shouldn't cause any overhead
3512  * when no packet needs it. At most one packet in the queue may be
3513  * marked for time stamping, otherwise it would be impossible to tell
3514  * for sure to which packet the hardware time stamp belongs.
3515  *
3516  * Incoming time stamping has to be configured via the hardware filters.
3517  * Not all combinations are supported, in particular event type has to be
3518  * specified. Matching the kind of event packet is not supported, with the
3519  * exception of "all V2 events regardless of level 2 or 4".
3520  **/
3521 static int e1000e_config_hwtstamp(struct e1000_adapter *adapter,
3522 				  struct hwtstamp_config *config)
3523 {
3524 	struct e1000_hw *hw = &adapter->hw;
3525 	u32 tsync_tx_ctl = E1000_TSYNCTXCTL_ENABLED;
3526 	u32 tsync_rx_ctl = E1000_TSYNCRXCTL_ENABLED;
3527 	u32 rxmtrl = 0;
3528 	u16 rxudp = 0;
3529 	bool is_l4 = false;
3530 	bool is_l2 = false;
3531 	u32 regval;
3532 	s32 ret_val;
3533 
3534 	if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP))
3535 		return -EINVAL;
3536 
3537 	/* flags reserved for future extensions - must be zero */
3538 	if (config->flags)
3539 		return -EINVAL;
3540 
3541 	switch (config->tx_type) {
3542 	case HWTSTAMP_TX_OFF:
3543 		tsync_tx_ctl = 0;
3544 		break;
3545 	case HWTSTAMP_TX_ON:
3546 		break;
3547 	default:
3548 		return -ERANGE;
3549 	}
3550 
3551 	switch (config->rx_filter) {
3552 	case HWTSTAMP_FILTER_NONE:
3553 		tsync_rx_ctl = 0;
3554 		break;
3555 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3556 		tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3557 		rxmtrl = E1000_RXMTRL_PTP_V1_SYNC_MESSAGE;
3558 		is_l4 = true;
3559 		break;
3560 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3561 		tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3562 		rxmtrl = E1000_RXMTRL_PTP_V1_DELAY_REQ_MESSAGE;
3563 		is_l4 = true;
3564 		break;
3565 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3566 		/* Also time stamps V2 L2 Path Delay Request/Response */
3567 		tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3568 		rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3569 		is_l2 = true;
3570 		break;
3571 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3572 		/* Also time stamps V2 L2 Path Delay Request/Response. */
3573 		tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3574 		rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3575 		is_l2 = true;
3576 		break;
3577 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3578 		/* Hardware cannot filter just V2 L4 Sync messages;
3579 		 * fall-through to V2 (both L2 and L4) Sync.
3580 		 */
3581 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
3582 		/* Also time stamps V2 Path Delay Request/Response. */
3583 		tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3584 		rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3585 		is_l2 = true;
3586 		is_l4 = true;
3587 		break;
3588 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3589 		/* Hardware cannot filter just V2 L4 Delay Request messages;
3590 		 * fall-through to V2 (both L2 and L4) Delay Request.
3591 		 */
3592 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3593 		/* Also time stamps V2 Path Delay Request/Response. */
3594 		tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3595 		rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3596 		is_l2 = true;
3597 		is_l4 = true;
3598 		break;
3599 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3600 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3601 		/* Hardware cannot filter just V2 L4 or L2 Event messages;
3602 		 * fall-through to all V2 (both L2 and L4) Events.
3603 		 */
3604 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
3605 		tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_EVENT_V2;
3606 		config->rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
3607 		is_l2 = true;
3608 		is_l4 = true;
3609 		break;
3610 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3611 		/* For V1, the hardware can only filter Sync messages or
3612 		 * Delay Request messages but not both so fall-through to
3613 		 * time stamp all packets.
3614 		 */
3615 	case HWTSTAMP_FILTER_ALL:
3616 		is_l2 = true;
3617 		is_l4 = true;
3618 		tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_ALL;
3619 		config->rx_filter = HWTSTAMP_FILTER_ALL;
3620 		break;
3621 	default:
3622 		return -ERANGE;
3623 	}
3624 
3625 	adapter->hwtstamp_config = *config;
3626 
3627 	/* enable/disable Tx h/w time stamping */
3628 	regval = er32(TSYNCTXCTL);
3629 	regval &= ~E1000_TSYNCTXCTL_ENABLED;
3630 	regval |= tsync_tx_ctl;
3631 	ew32(TSYNCTXCTL, regval);
3632 	if ((er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) !=
3633 	    (regval & E1000_TSYNCTXCTL_ENABLED)) {
3634 		e_err("Timesync Tx Control register not set as expected\n");
3635 		return -EAGAIN;
3636 	}
3637 
3638 	/* enable/disable Rx h/w time stamping */
3639 	regval = er32(TSYNCRXCTL);
3640 	regval &= ~(E1000_TSYNCRXCTL_ENABLED | E1000_TSYNCRXCTL_TYPE_MASK);
3641 	regval |= tsync_rx_ctl;
3642 	ew32(TSYNCRXCTL, regval);
3643 	if ((er32(TSYNCRXCTL) & (E1000_TSYNCRXCTL_ENABLED |
3644 				 E1000_TSYNCRXCTL_TYPE_MASK)) !=
3645 	    (regval & (E1000_TSYNCRXCTL_ENABLED |
3646 		       E1000_TSYNCRXCTL_TYPE_MASK))) {
3647 		e_err("Timesync Rx Control register not set as expected\n");
3648 		return -EAGAIN;
3649 	}
3650 
3651 	/* L2: define ethertype filter for time stamped packets */
3652 	if (is_l2)
3653 		rxmtrl |= ETH_P_1588;
3654 
3655 	/* define which PTP packets get time stamped */
3656 	ew32(RXMTRL, rxmtrl);
3657 
3658 	/* Filter by destination port */
3659 	if (is_l4) {
3660 		rxudp = PTP_EV_PORT;
3661 		cpu_to_be16s(&rxudp);
3662 	}
3663 	ew32(RXUDP, rxudp);
3664 
3665 	e1e_flush();
3666 
3667 	/* Clear TSYNCRXCTL_VALID & TSYNCTXCTL_VALID bit */
3668 	er32(RXSTMPH);
3669 	er32(TXSTMPH);
3670 
3671 	/* Get and set the System Time Register SYSTIM base frequency */
3672 	ret_val = e1000e_get_base_timinca(adapter, &regval);
3673 	if (ret_val)
3674 		return ret_val;
3675 	ew32(TIMINCA, regval);
3676 
3677 	/* reset the ns time counter */
3678 	timecounter_init(&adapter->tc, &adapter->cc,
3679 			 ktime_to_ns(ktime_get_real()));
3680 
3681 	return 0;
3682 }
3683 
3684 /**
3685  * e1000_configure - configure the hardware for Rx and Tx
3686  * @adapter: private board structure
3687  **/
3688 static void e1000_configure(struct e1000_adapter *adapter)
3689 {
3690 	struct e1000_ring *rx_ring = adapter->rx_ring;
3691 
3692 	e1000e_set_rx_mode(adapter->netdev);
3693 
3694 	e1000_restore_vlan(adapter);
3695 	e1000_init_manageability_pt(adapter);
3696 
3697 	e1000_configure_tx(adapter);
3698 
3699 	if (adapter->netdev->features & NETIF_F_RXHASH)
3700 		e1000e_setup_rss_hash(adapter);
3701 	e1000_setup_rctl(adapter);
3702 	e1000_configure_rx(adapter);
3703 	adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3704 }
3705 
3706 /**
3707  * e1000e_power_up_phy - restore link in case the phy was powered down
3708  * @adapter: address of board private structure
3709  *
3710  * The phy may be powered down to save power and turn off link when the
3711  * driver is unloaded and wake on lan is not enabled (among others)
3712  * *** this routine MUST be followed by a call to e1000e_reset ***
3713  **/
3714 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3715 {
3716 	if (adapter->hw.phy.ops.power_up)
3717 		adapter->hw.phy.ops.power_up(&adapter->hw);
3718 
3719 	adapter->hw.mac.ops.setup_link(&adapter->hw);
3720 }
3721 
3722 /**
3723  * e1000_power_down_phy - Power down the PHY
3724  *
3725  * Power down the PHY so no link is implied when interface is down.
3726  * The PHY cannot be powered down if management or WoL is active.
3727  */
3728 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3729 {
3730 	if (adapter->hw.phy.ops.power_down)
3731 		adapter->hw.phy.ops.power_down(&adapter->hw);
3732 }
3733 
3734 /**
3735  * e1000e_reset - bring the hardware into a known good state
3736  *
3737  * This function boots the hardware and enables some settings that
3738  * require a configuration cycle of the hardware - those cannot be
3739  * set/changed during runtime. After reset the device needs to be
3740  * properly configured for Rx, Tx etc.
3741  */
3742 void e1000e_reset(struct e1000_adapter *adapter)
3743 {
3744 	struct e1000_mac_info *mac = &adapter->hw.mac;
3745 	struct e1000_fc_info *fc = &adapter->hw.fc;
3746 	struct e1000_hw *hw = &adapter->hw;
3747 	u32 tx_space, min_tx_space, min_rx_space;
3748 	u32 pba = adapter->pba;
3749 	u16 hwm;
3750 
3751 	/* reset Packet Buffer Allocation to default */
3752 	ew32(PBA, pba);
3753 
3754 	if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3755 		/* To maintain wire speed transmits, the Tx FIFO should be
3756 		 * large enough to accommodate two full transmit packets,
3757 		 * rounded up to the next 1KB and expressed in KB.  Likewise,
3758 		 * the Rx FIFO should be large enough to accommodate at least
3759 		 * one full receive packet and is similarly rounded up and
3760 		 * expressed in KB.
3761 		 */
3762 		pba = er32(PBA);
3763 		/* upper 16 bits has Tx packet buffer allocation size in KB */
3764 		tx_space = pba >> 16;
3765 		/* lower 16 bits has Rx packet buffer allocation size in KB */
3766 		pba &= 0xffff;
3767 		/* the Tx fifo also stores 16 bytes of information about the Tx
3768 		 * but don't include ethernet FCS because hardware appends it
3769 		 */
3770 		min_tx_space = (adapter->max_frame_size +
3771 				sizeof(struct e1000_tx_desc) - ETH_FCS_LEN) * 2;
3772 		min_tx_space = ALIGN(min_tx_space, 1024);
3773 		min_tx_space >>= 10;
3774 		/* software strips receive CRC, so leave room for it */
3775 		min_rx_space = adapter->max_frame_size;
3776 		min_rx_space = ALIGN(min_rx_space, 1024);
3777 		min_rx_space >>= 10;
3778 
3779 		/* If current Tx allocation is less than the min Tx FIFO size,
3780 		 * and the min Tx FIFO size is less than the current Rx FIFO
3781 		 * allocation, take space away from current Rx allocation
3782 		 */
3783 		if ((tx_space < min_tx_space) &&
3784 		    ((min_tx_space - tx_space) < pba)) {
3785 			pba -= min_tx_space - tx_space;
3786 
3787 			/* if short on Rx space, Rx wins and must trump Tx
3788 			 * adjustment
3789 			 */
3790 			if (pba < min_rx_space)
3791 				pba = min_rx_space;
3792 		}
3793 
3794 		ew32(PBA, pba);
3795 	}
3796 
3797 	/* flow control settings
3798 	 *
3799 	 * The high water mark must be low enough to fit one full frame
3800 	 * (or the size used for early receive) above it in the Rx FIFO.
3801 	 * Set it to the lower of:
3802 	 * - 90% of the Rx FIFO size, and
3803 	 * - the full Rx FIFO size minus one full frame
3804 	 */
3805 	if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3806 		fc->pause_time = 0xFFFF;
3807 	else
3808 		fc->pause_time = E1000_FC_PAUSE_TIME;
3809 	fc->send_xon = true;
3810 	fc->current_mode = fc->requested_mode;
3811 
3812 	switch (hw->mac.type) {
3813 	case e1000_ich9lan:
3814 	case e1000_ich10lan:
3815 		if (adapter->netdev->mtu > ETH_DATA_LEN) {
3816 			pba = 14;
3817 			ew32(PBA, pba);
3818 			fc->high_water = 0x2800;
3819 			fc->low_water = fc->high_water - 8;
3820 			break;
3821 		}
3822 		/* fall-through */
3823 	default:
3824 		hwm = min(((pba << 10) * 9 / 10),
3825 			  ((pba << 10) - adapter->max_frame_size));
3826 
3827 		fc->high_water = hwm & E1000_FCRTH_RTH;	/* 8-byte granularity */
3828 		fc->low_water = fc->high_water - 8;
3829 		break;
3830 	case e1000_pchlan:
3831 		/* Workaround PCH LOM adapter hangs with certain network
3832 		 * loads.  If hangs persist, try disabling Tx flow control.
3833 		 */
3834 		if (adapter->netdev->mtu > ETH_DATA_LEN) {
3835 			fc->high_water = 0x3500;
3836 			fc->low_water = 0x1500;
3837 		} else {
3838 			fc->high_water = 0x5000;
3839 			fc->low_water = 0x3000;
3840 		}
3841 		fc->refresh_time = 0x1000;
3842 		break;
3843 	case e1000_pch2lan:
3844 	case e1000_pch_lpt:
3845 		fc->refresh_time = 0x0400;
3846 
3847 		if (adapter->netdev->mtu <= ETH_DATA_LEN) {
3848 			fc->high_water = 0x05C20;
3849 			fc->low_water = 0x05048;
3850 			fc->pause_time = 0x0650;
3851 			break;
3852 		}
3853 
3854 		pba = 14;
3855 		ew32(PBA, pba);
3856 		fc->high_water = ((pba << 10) * 9 / 10) & E1000_FCRTH_RTH;
3857 		fc->low_water = ((pba << 10) * 8 / 10) & E1000_FCRTL_RTL;
3858 		break;
3859 	}
3860 
3861 	/* Alignment of Tx data is on an arbitrary byte boundary with the
3862 	 * maximum size per Tx descriptor limited only to the transmit
3863 	 * allocation of the packet buffer minus 96 bytes with an upper
3864 	 * limit of 24KB due to receive synchronization limitations.
3865 	 */
3866 	adapter->tx_fifo_limit = min_t(u32, ((er32(PBA) >> 16) << 10) - 96,
3867 				       24 << 10);
3868 
3869 	/* Disable Adaptive Interrupt Moderation if 2 full packets cannot
3870 	 * fit in receive buffer.
3871 	 */
3872 	if (adapter->itr_setting & 0x3) {
3873 		if ((adapter->max_frame_size * 2) > (pba << 10)) {
3874 			if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3875 				dev_info(&adapter->pdev->dev,
3876 					 "Interrupt Throttle Rate off\n");
3877 				adapter->flags2 |= FLAG2_DISABLE_AIM;
3878 				e1000e_write_itr(adapter, 0);
3879 			}
3880 		} else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3881 			dev_info(&adapter->pdev->dev,
3882 				 "Interrupt Throttle Rate on\n");
3883 			adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3884 			adapter->itr = 20000;
3885 			e1000e_write_itr(adapter, adapter->itr);
3886 		}
3887 	}
3888 
3889 	/* Allow time for pending master requests to run */
3890 	mac->ops.reset_hw(hw);
3891 
3892 	/* For parts with AMT enabled, let the firmware know
3893 	 * that the network interface is in control
3894 	 */
3895 	if (adapter->flags & FLAG_HAS_AMT)
3896 		e1000e_get_hw_control(adapter);
3897 
3898 	ew32(WUC, 0);
3899 
3900 	if (mac->ops.init_hw(hw))
3901 		e_err("Hardware Error\n");
3902 
3903 	e1000_update_mng_vlan(adapter);
3904 
3905 	/* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3906 	ew32(VET, ETH_P_8021Q);
3907 
3908 	e1000e_reset_adaptive(hw);
3909 
3910 	/* initialize systim and reset the ns time counter */
3911 	e1000e_config_hwtstamp(adapter, &adapter->hwtstamp_config);
3912 
3913 	/* Set EEE advertisement as appropriate */
3914 	if (adapter->flags2 & FLAG2_HAS_EEE) {
3915 		s32 ret_val;
3916 		u16 adv_addr;
3917 
3918 		switch (hw->phy.type) {
3919 		case e1000_phy_82579:
3920 			adv_addr = I82579_EEE_ADVERTISEMENT;
3921 			break;
3922 		case e1000_phy_i217:
3923 			adv_addr = I217_EEE_ADVERTISEMENT;
3924 			break;
3925 		default:
3926 			dev_err(&adapter->pdev->dev,
3927 				"Invalid PHY type setting EEE advertisement\n");
3928 			return;
3929 		}
3930 
3931 		ret_val = hw->phy.ops.acquire(hw);
3932 		if (ret_val) {
3933 			dev_err(&adapter->pdev->dev,
3934 				"EEE advertisement - unable to acquire PHY\n");
3935 			return;
3936 		}
3937 
3938 		e1000_write_emi_reg_locked(hw, adv_addr,
3939 					   hw->dev_spec.ich8lan.eee_disable ?
3940 					   0 : adapter->eee_advert);
3941 
3942 		hw->phy.ops.release(hw);
3943 	}
3944 
3945 	if (!netif_running(adapter->netdev) &&
3946 	    !test_bit(__E1000_TESTING, &adapter->state))
3947 		e1000_power_down_phy(adapter);
3948 
3949 	e1000_get_phy_info(hw);
3950 
3951 	if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3952 	    !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3953 		u16 phy_data = 0;
3954 		/* speed up time to link by disabling smart power down, ignore
3955 		 * the return value of this function because there is nothing
3956 		 * different we would do if it failed
3957 		 */
3958 		e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3959 		phy_data &= ~IGP02E1000_PM_SPD;
3960 		e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3961 	}
3962 }
3963 
3964 int e1000e_up(struct e1000_adapter *adapter)
3965 {
3966 	struct e1000_hw *hw = &adapter->hw;
3967 
3968 	/* hardware has been reset, we need to reload some things */
3969 	e1000_configure(adapter);
3970 
3971 	clear_bit(__E1000_DOWN, &adapter->state);
3972 
3973 	if (adapter->msix_entries)
3974 		e1000_configure_msix(adapter);
3975 	e1000_irq_enable(adapter);
3976 
3977 	netif_start_queue(adapter->netdev);
3978 
3979 	/* fire a link change interrupt to start the watchdog */
3980 	if (adapter->msix_entries)
3981 		ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3982 	else
3983 		ew32(ICS, E1000_ICS_LSC);
3984 
3985 	return 0;
3986 }
3987 
3988 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3989 {
3990 	struct e1000_hw *hw = &adapter->hw;
3991 
3992 	if (!(adapter->flags2 & FLAG2_DMA_BURST))
3993 		return;
3994 
3995 	/* flush pending descriptor writebacks to memory */
3996 	ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3997 	ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3998 
3999 	/* execute the writes immediately */
4000 	e1e_flush();
4001 
4002 	/* due to rare timing issues, write to TIDV/RDTR again to ensure the
4003 	 * write is successful
4004 	 */
4005 	ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4006 	ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4007 
4008 	/* execute the writes immediately */
4009 	e1e_flush();
4010 }
4011 
4012 static void e1000e_update_stats(struct e1000_adapter *adapter);
4013 
4014 /**
4015  * e1000e_down - quiesce the device and optionally reset the hardware
4016  * @adapter: board private structure
4017  * @reset: boolean flag to reset the hardware or not
4018  */
4019 void e1000e_down(struct e1000_adapter *adapter, bool reset)
4020 {
4021 	struct net_device *netdev = adapter->netdev;
4022 	struct e1000_hw *hw = &adapter->hw;
4023 	u32 tctl, rctl;
4024 
4025 	/* signal that we're down so the interrupt handler does not
4026 	 * reschedule our watchdog timer
4027 	 */
4028 	set_bit(__E1000_DOWN, &adapter->state);
4029 
4030 	/* disable receives in the hardware */
4031 	rctl = er32(RCTL);
4032 	if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
4033 		ew32(RCTL, rctl & ~E1000_RCTL_EN);
4034 	/* flush and sleep below */
4035 
4036 	netif_stop_queue(netdev);
4037 
4038 	/* disable transmits in the hardware */
4039 	tctl = er32(TCTL);
4040 	tctl &= ~E1000_TCTL_EN;
4041 	ew32(TCTL, tctl);
4042 
4043 	/* flush both disables and wait for them to finish */
4044 	e1e_flush();
4045 	usleep_range(10000, 20000);
4046 
4047 	e1000_irq_disable(adapter);
4048 
4049 	napi_synchronize(&adapter->napi);
4050 
4051 	del_timer_sync(&adapter->watchdog_timer);
4052 	del_timer_sync(&adapter->phy_info_timer);
4053 
4054 	netif_carrier_off(netdev);
4055 
4056 	spin_lock(&adapter->stats64_lock);
4057 	e1000e_update_stats(adapter);
4058 	spin_unlock(&adapter->stats64_lock);
4059 
4060 	e1000e_flush_descriptors(adapter);
4061 	e1000_clean_tx_ring(adapter->tx_ring);
4062 	e1000_clean_rx_ring(adapter->rx_ring);
4063 
4064 	adapter->link_speed = 0;
4065 	adapter->link_duplex = 0;
4066 
4067 	/* Disable Si errata workaround on PCHx for jumbo frame flow */
4068 	if ((hw->mac.type >= e1000_pch2lan) &&
4069 	    (adapter->netdev->mtu > ETH_DATA_LEN) &&
4070 	    e1000_lv_jumbo_workaround_ich8lan(hw, false))
4071 		e_dbg("failed to disable jumbo frame workaround mode\n");
4072 
4073 	if (reset && !pci_channel_offline(adapter->pdev))
4074 		e1000e_reset(adapter);
4075 }
4076 
4077 void e1000e_reinit_locked(struct e1000_adapter *adapter)
4078 {
4079 	might_sleep();
4080 	while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4081 		usleep_range(1000, 2000);
4082 	e1000e_down(adapter, true);
4083 	e1000e_up(adapter);
4084 	clear_bit(__E1000_RESETTING, &adapter->state);
4085 }
4086 
4087 /**
4088  * e1000e_cyclecounter_read - read raw cycle counter (used by time counter)
4089  * @cc: cyclecounter structure
4090  **/
4091 static cycle_t e1000e_cyclecounter_read(const struct cyclecounter *cc)
4092 {
4093 	struct e1000_adapter *adapter = container_of(cc, struct e1000_adapter,
4094 						     cc);
4095 	struct e1000_hw *hw = &adapter->hw;
4096 	cycle_t systim;
4097 
4098 	/* latch SYSTIMH on read of SYSTIML */
4099 	systim = (cycle_t)er32(SYSTIML);
4100 	systim |= (cycle_t)er32(SYSTIMH) << 32;
4101 
4102 	return systim;
4103 }
4104 
4105 /**
4106  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
4107  * @adapter: board private structure to initialize
4108  *
4109  * e1000_sw_init initializes the Adapter private data structure.
4110  * Fields are initialized based on PCI device information and
4111  * OS network device settings (MTU size).
4112  **/
4113 static int e1000_sw_init(struct e1000_adapter *adapter)
4114 {
4115 	struct net_device *netdev = adapter->netdev;
4116 
4117 	adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
4118 	adapter->rx_ps_bsize0 = 128;
4119 	adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4120 	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
4121 	adapter->tx_ring_count = E1000_DEFAULT_TXD;
4122 	adapter->rx_ring_count = E1000_DEFAULT_RXD;
4123 
4124 	spin_lock_init(&adapter->stats64_lock);
4125 
4126 	e1000e_set_interrupt_capability(adapter);
4127 
4128 	if (e1000_alloc_queues(adapter))
4129 		return -ENOMEM;
4130 
4131 	/* Setup hardware time stamping cyclecounter */
4132 	if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
4133 		adapter->cc.read = e1000e_cyclecounter_read;
4134 		adapter->cc.mask = CLOCKSOURCE_MASK(64);
4135 		adapter->cc.mult = 1;
4136 		/* cc.shift set in e1000e_get_base_tininca() */
4137 
4138 		spin_lock_init(&adapter->systim_lock);
4139 		INIT_WORK(&adapter->tx_hwtstamp_work, e1000e_tx_hwtstamp_work);
4140 	}
4141 
4142 	/* Explicitly disable IRQ since the NIC can be in any state. */
4143 	e1000_irq_disable(adapter);
4144 
4145 	set_bit(__E1000_DOWN, &adapter->state);
4146 	return 0;
4147 }
4148 
4149 /**
4150  * e1000_intr_msi_test - Interrupt Handler
4151  * @irq: interrupt number
4152  * @data: pointer to a network interface device structure
4153  **/
4154 static irqreturn_t e1000_intr_msi_test(int __always_unused irq, void *data)
4155 {
4156 	struct net_device *netdev = data;
4157 	struct e1000_adapter *adapter = netdev_priv(netdev);
4158 	struct e1000_hw *hw = &adapter->hw;
4159 	u32 icr = er32(ICR);
4160 
4161 	e_dbg("icr is %08X\n", icr);
4162 	if (icr & E1000_ICR_RXSEQ) {
4163 		adapter->flags &= ~FLAG_MSI_TEST_FAILED;
4164 		/* Force memory writes to complete before acknowledging the
4165 		 * interrupt is handled.
4166 		 */
4167 		wmb();
4168 	}
4169 
4170 	return IRQ_HANDLED;
4171 }
4172 
4173 /**
4174  * e1000_test_msi_interrupt - Returns 0 for successful test
4175  * @adapter: board private struct
4176  *
4177  * code flow taken from tg3.c
4178  **/
4179 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
4180 {
4181 	struct net_device *netdev = adapter->netdev;
4182 	struct e1000_hw *hw = &adapter->hw;
4183 	int err;
4184 
4185 	/* poll_enable hasn't been called yet, so don't need disable */
4186 	/* clear any pending events */
4187 	er32(ICR);
4188 
4189 	/* free the real vector and request a test handler */
4190 	e1000_free_irq(adapter);
4191 	e1000e_reset_interrupt_capability(adapter);
4192 
4193 	/* Assume that the test fails, if it succeeds then the test
4194 	 * MSI irq handler will unset this flag
4195 	 */
4196 	adapter->flags |= FLAG_MSI_TEST_FAILED;
4197 
4198 	err = pci_enable_msi(adapter->pdev);
4199 	if (err)
4200 		goto msi_test_failed;
4201 
4202 	err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
4203 			  netdev->name, netdev);
4204 	if (err) {
4205 		pci_disable_msi(adapter->pdev);
4206 		goto msi_test_failed;
4207 	}
4208 
4209 	/* Force memory writes to complete before enabling and firing an
4210 	 * interrupt.
4211 	 */
4212 	wmb();
4213 
4214 	e1000_irq_enable(adapter);
4215 
4216 	/* fire an unusual interrupt on the test handler */
4217 	ew32(ICS, E1000_ICS_RXSEQ);
4218 	e1e_flush();
4219 	msleep(100);
4220 
4221 	e1000_irq_disable(adapter);
4222 
4223 	rmb();			/* read flags after interrupt has been fired */
4224 
4225 	if (adapter->flags & FLAG_MSI_TEST_FAILED) {
4226 		adapter->int_mode = E1000E_INT_MODE_LEGACY;
4227 		e_info("MSI interrupt test failed, using legacy interrupt.\n");
4228 	} else {
4229 		e_dbg("MSI interrupt test succeeded!\n");
4230 	}
4231 
4232 	free_irq(adapter->pdev->irq, netdev);
4233 	pci_disable_msi(adapter->pdev);
4234 
4235 msi_test_failed:
4236 	e1000e_set_interrupt_capability(adapter);
4237 	return e1000_request_irq(adapter);
4238 }
4239 
4240 /**
4241  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
4242  * @adapter: board private struct
4243  *
4244  * code flow taken from tg3.c, called with e1000 interrupts disabled.
4245  **/
4246 static int e1000_test_msi(struct e1000_adapter *adapter)
4247 {
4248 	int err;
4249 	u16 pci_cmd;
4250 
4251 	if (!(adapter->flags & FLAG_MSI_ENABLED))
4252 		return 0;
4253 
4254 	/* disable SERR in case the MSI write causes a master abort */
4255 	pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4256 	if (pci_cmd & PCI_COMMAND_SERR)
4257 		pci_write_config_word(adapter->pdev, PCI_COMMAND,
4258 				      pci_cmd & ~PCI_COMMAND_SERR);
4259 
4260 	err = e1000_test_msi_interrupt(adapter);
4261 
4262 	/* re-enable SERR */
4263 	if (pci_cmd & PCI_COMMAND_SERR) {
4264 		pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4265 		pci_cmd |= PCI_COMMAND_SERR;
4266 		pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
4267 	}
4268 
4269 	return err;
4270 }
4271 
4272 /**
4273  * e1000_open - Called when a network interface is made active
4274  * @netdev: network interface device structure
4275  *
4276  * Returns 0 on success, negative value on failure
4277  *
4278  * The open entry point is called when a network interface is made
4279  * active by the system (IFF_UP).  At this point all resources needed
4280  * for transmit and receive operations are allocated, the interrupt
4281  * handler is registered with the OS, the watchdog timer is started,
4282  * and the stack is notified that the interface is ready.
4283  **/
4284 static int e1000_open(struct net_device *netdev)
4285 {
4286 	struct e1000_adapter *adapter = netdev_priv(netdev);
4287 	struct e1000_hw *hw = &adapter->hw;
4288 	struct pci_dev *pdev = adapter->pdev;
4289 	int err;
4290 
4291 	/* disallow open during test */
4292 	if (test_bit(__E1000_TESTING, &adapter->state))
4293 		return -EBUSY;
4294 
4295 	pm_runtime_get_sync(&pdev->dev);
4296 
4297 	netif_carrier_off(netdev);
4298 
4299 	/* allocate transmit descriptors */
4300 	err = e1000e_setup_tx_resources(adapter->tx_ring);
4301 	if (err)
4302 		goto err_setup_tx;
4303 
4304 	/* allocate receive descriptors */
4305 	err = e1000e_setup_rx_resources(adapter->rx_ring);
4306 	if (err)
4307 		goto err_setup_rx;
4308 
4309 	/* If AMT is enabled, let the firmware know that the network
4310 	 * interface is now open and reset the part to a known state.
4311 	 */
4312 	if (adapter->flags & FLAG_HAS_AMT) {
4313 		e1000e_get_hw_control(adapter);
4314 		e1000e_reset(adapter);
4315 	}
4316 
4317 	e1000e_power_up_phy(adapter);
4318 
4319 	adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4320 	if ((adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
4321 		e1000_update_mng_vlan(adapter);
4322 
4323 	/* DMA latency requirement to workaround jumbo issue */
4324 	pm_qos_add_request(&adapter->netdev->pm_qos_req, PM_QOS_CPU_DMA_LATENCY,
4325 			   PM_QOS_DEFAULT_VALUE);
4326 
4327 	/* before we allocate an interrupt, we must be ready to handle it.
4328 	 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
4329 	 * as soon as we call pci_request_irq, so we have to setup our
4330 	 * clean_rx handler before we do so.
4331 	 */
4332 	e1000_configure(adapter);
4333 
4334 	err = e1000_request_irq(adapter);
4335 	if (err)
4336 		goto err_req_irq;
4337 
4338 	/* Work around PCIe errata with MSI interrupts causing some chipsets to
4339 	 * ignore e1000e MSI messages, which means we need to test our MSI
4340 	 * interrupt now
4341 	 */
4342 	if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
4343 		err = e1000_test_msi(adapter);
4344 		if (err) {
4345 			e_err("Interrupt allocation failed\n");
4346 			goto err_req_irq;
4347 		}
4348 	}
4349 
4350 	/* From here on the code is the same as e1000e_up() */
4351 	clear_bit(__E1000_DOWN, &adapter->state);
4352 
4353 	napi_enable(&adapter->napi);
4354 
4355 	e1000_irq_enable(adapter);
4356 
4357 	adapter->tx_hang_recheck = false;
4358 	netif_start_queue(netdev);
4359 
4360 	hw->mac.get_link_status = true;
4361 	pm_runtime_put(&pdev->dev);
4362 
4363 	/* fire a link status change interrupt to start the watchdog */
4364 	if (adapter->msix_entries)
4365 		ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
4366 	else
4367 		ew32(ICS, E1000_ICS_LSC);
4368 
4369 	return 0;
4370 
4371 err_req_irq:
4372 	e1000e_release_hw_control(adapter);
4373 	e1000_power_down_phy(adapter);
4374 	e1000e_free_rx_resources(adapter->rx_ring);
4375 err_setup_rx:
4376 	e1000e_free_tx_resources(adapter->tx_ring);
4377 err_setup_tx:
4378 	e1000e_reset(adapter);
4379 	pm_runtime_put_sync(&pdev->dev);
4380 
4381 	return err;
4382 }
4383 
4384 /**
4385  * e1000_close - Disables a network interface
4386  * @netdev: network interface device structure
4387  *
4388  * Returns 0, this is not allowed to fail
4389  *
4390  * The close entry point is called when an interface is de-activated
4391  * by the OS.  The hardware is still under the drivers control, but
4392  * needs to be disabled.  A global MAC reset is issued to stop the
4393  * hardware, and all transmit and receive resources are freed.
4394  **/
4395 static int e1000_close(struct net_device *netdev)
4396 {
4397 	struct e1000_adapter *adapter = netdev_priv(netdev);
4398 	struct pci_dev *pdev = adapter->pdev;
4399 	int count = E1000_CHECK_RESET_COUNT;
4400 
4401 	while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
4402 		usleep_range(10000, 20000);
4403 
4404 	WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4405 
4406 	pm_runtime_get_sync(&pdev->dev);
4407 
4408 	if (!test_bit(__E1000_DOWN, &adapter->state)) {
4409 		e1000e_down(adapter, true);
4410 		e1000_free_irq(adapter);
4411 
4412 		/* Link status message must follow this format */
4413 		pr_info("%s NIC Link is Down\n", adapter->netdev->name);
4414 	}
4415 
4416 	napi_disable(&adapter->napi);
4417 
4418 	e1000e_free_tx_resources(adapter->tx_ring);
4419 	e1000e_free_rx_resources(adapter->rx_ring);
4420 
4421 	/* kill manageability vlan ID if supported, but not if a vlan with
4422 	 * the same ID is registered on the host OS (let 8021q kill it)
4423 	 */
4424 	if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
4425 		e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
4426 				       adapter->mng_vlan_id);
4427 
4428 	/* If AMT is enabled, let the firmware know that the network
4429 	 * interface is now closed
4430 	 */
4431 	if ((adapter->flags & FLAG_HAS_AMT) &&
4432 	    !test_bit(__E1000_TESTING, &adapter->state))
4433 		e1000e_release_hw_control(adapter);
4434 
4435 	pm_qos_remove_request(&adapter->netdev->pm_qos_req);
4436 
4437 	pm_runtime_put_sync(&pdev->dev);
4438 
4439 	return 0;
4440 }
4441 
4442 /**
4443  * e1000_set_mac - Change the Ethernet Address of the NIC
4444  * @netdev: network interface device structure
4445  * @p: pointer to an address structure
4446  *
4447  * Returns 0 on success, negative on failure
4448  **/
4449 static int e1000_set_mac(struct net_device *netdev, void *p)
4450 {
4451 	struct e1000_adapter *adapter = netdev_priv(netdev);
4452 	struct e1000_hw *hw = &adapter->hw;
4453 	struct sockaddr *addr = p;
4454 
4455 	if (!is_valid_ether_addr(addr->sa_data))
4456 		return -EADDRNOTAVAIL;
4457 
4458 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4459 	memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
4460 
4461 	hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
4462 
4463 	if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
4464 		/* activate the work around */
4465 		e1000e_set_laa_state_82571(&adapter->hw, 1);
4466 
4467 		/* Hold a copy of the LAA in RAR[14] This is done so that
4468 		 * between the time RAR[0] gets clobbered  and the time it
4469 		 * gets fixed (in e1000_watchdog), the actual LAA is in one
4470 		 * of the RARs and no incoming packets directed to this port
4471 		 * are dropped. Eventually the LAA will be in RAR[0] and
4472 		 * RAR[14]
4473 		 */
4474 		hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr,
4475 				    adapter->hw.mac.rar_entry_count - 1);
4476 	}
4477 
4478 	return 0;
4479 }
4480 
4481 /**
4482  * e1000e_update_phy_task - work thread to update phy
4483  * @work: pointer to our work struct
4484  *
4485  * this worker thread exists because we must acquire a
4486  * semaphore to read the phy, which we could msleep while
4487  * waiting for it, and we can't msleep in a timer.
4488  **/
4489 static void e1000e_update_phy_task(struct work_struct *work)
4490 {
4491 	struct e1000_adapter *adapter = container_of(work,
4492 						     struct e1000_adapter,
4493 						     update_phy_task);
4494 	struct e1000_hw *hw = &adapter->hw;
4495 
4496 	if (test_bit(__E1000_DOWN, &adapter->state))
4497 		return;
4498 
4499 	e1000_get_phy_info(hw);
4500 
4501 	/* Enable EEE on 82579 after link up */
4502 	if (hw->phy.type == e1000_phy_82579)
4503 		e1000_set_eee_pchlan(hw);
4504 }
4505 
4506 /**
4507  * e1000_update_phy_info - timre call-back to update PHY info
4508  * @data: pointer to adapter cast into an unsigned long
4509  *
4510  * Need to wait a few seconds after link up to get diagnostic information from
4511  * the phy
4512  **/
4513 static void e1000_update_phy_info(unsigned long data)
4514 {
4515 	struct e1000_adapter *adapter = (struct e1000_adapter *)data;
4516 
4517 	if (test_bit(__E1000_DOWN, &adapter->state))
4518 		return;
4519 
4520 	schedule_work(&adapter->update_phy_task);
4521 }
4522 
4523 /**
4524  * e1000e_update_phy_stats - Update the PHY statistics counters
4525  * @adapter: board private structure
4526  *
4527  * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4528  **/
4529 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4530 {
4531 	struct e1000_hw *hw = &adapter->hw;
4532 	s32 ret_val;
4533 	u16 phy_data;
4534 
4535 	ret_val = hw->phy.ops.acquire(hw);
4536 	if (ret_val)
4537 		return;
4538 
4539 	/* A page set is expensive so check if already on desired page.
4540 	 * If not, set to the page with the PHY status registers.
4541 	 */
4542 	hw->phy.addr = 1;
4543 	ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4544 					   &phy_data);
4545 	if (ret_val)
4546 		goto release;
4547 	if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4548 		ret_val = hw->phy.ops.set_page(hw,
4549 					       HV_STATS_PAGE << IGP_PAGE_SHIFT);
4550 		if (ret_val)
4551 			goto release;
4552 	}
4553 
4554 	/* Single Collision Count */
4555 	hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4556 	ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4557 	if (!ret_val)
4558 		adapter->stats.scc += phy_data;
4559 
4560 	/* Excessive Collision Count */
4561 	hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4562 	ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4563 	if (!ret_val)
4564 		adapter->stats.ecol += phy_data;
4565 
4566 	/* Multiple Collision Count */
4567 	hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4568 	ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4569 	if (!ret_val)
4570 		adapter->stats.mcc += phy_data;
4571 
4572 	/* Late Collision Count */
4573 	hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4574 	ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4575 	if (!ret_val)
4576 		adapter->stats.latecol += phy_data;
4577 
4578 	/* Collision Count - also used for adaptive IFS */
4579 	hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4580 	ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4581 	if (!ret_val)
4582 		hw->mac.collision_delta = phy_data;
4583 
4584 	/* Defer Count */
4585 	hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4586 	ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4587 	if (!ret_val)
4588 		adapter->stats.dc += phy_data;
4589 
4590 	/* Transmit with no CRS */
4591 	hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4592 	ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4593 	if (!ret_val)
4594 		adapter->stats.tncrs += phy_data;
4595 
4596 release:
4597 	hw->phy.ops.release(hw);
4598 }
4599 
4600 /**
4601  * e1000e_update_stats - Update the board statistics counters
4602  * @adapter: board private structure
4603  **/
4604 static void e1000e_update_stats(struct e1000_adapter *adapter)
4605 {
4606 	struct net_device *netdev = adapter->netdev;
4607 	struct e1000_hw *hw = &adapter->hw;
4608 	struct pci_dev *pdev = adapter->pdev;
4609 
4610 	/* Prevent stats update while adapter is being reset, or if the pci
4611 	 * connection is down.
4612 	 */
4613 	if (adapter->link_speed == 0)
4614 		return;
4615 	if (pci_channel_offline(pdev))
4616 		return;
4617 
4618 	adapter->stats.crcerrs += er32(CRCERRS);
4619 	adapter->stats.gprc += er32(GPRC);
4620 	adapter->stats.gorc += er32(GORCL);
4621 	er32(GORCH);		/* Clear gorc */
4622 	adapter->stats.bprc += er32(BPRC);
4623 	adapter->stats.mprc += er32(MPRC);
4624 	adapter->stats.roc += er32(ROC);
4625 
4626 	adapter->stats.mpc += er32(MPC);
4627 
4628 	/* Half-duplex statistics */
4629 	if (adapter->link_duplex == HALF_DUPLEX) {
4630 		if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4631 			e1000e_update_phy_stats(adapter);
4632 		} else {
4633 			adapter->stats.scc += er32(SCC);
4634 			adapter->stats.ecol += er32(ECOL);
4635 			adapter->stats.mcc += er32(MCC);
4636 			adapter->stats.latecol += er32(LATECOL);
4637 			adapter->stats.dc += er32(DC);
4638 
4639 			hw->mac.collision_delta = er32(COLC);
4640 
4641 			if ((hw->mac.type != e1000_82574) &&
4642 			    (hw->mac.type != e1000_82583))
4643 				adapter->stats.tncrs += er32(TNCRS);
4644 		}
4645 		adapter->stats.colc += hw->mac.collision_delta;
4646 	}
4647 
4648 	adapter->stats.xonrxc += er32(XONRXC);
4649 	adapter->stats.xontxc += er32(XONTXC);
4650 	adapter->stats.xoffrxc += er32(XOFFRXC);
4651 	adapter->stats.xofftxc += er32(XOFFTXC);
4652 	adapter->stats.gptc += er32(GPTC);
4653 	adapter->stats.gotc += er32(GOTCL);
4654 	er32(GOTCH);		/* Clear gotc */
4655 	adapter->stats.rnbc += er32(RNBC);
4656 	adapter->stats.ruc += er32(RUC);
4657 
4658 	adapter->stats.mptc += er32(MPTC);
4659 	adapter->stats.bptc += er32(BPTC);
4660 
4661 	/* used for adaptive IFS */
4662 
4663 	hw->mac.tx_packet_delta = er32(TPT);
4664 	adapter->stats.tpt += hw->mac.tx_packet_delta;
4665 
4666 	adapter->stats.algnerrc += er32(ALGNERRC);
4667 	adapter->stats.rxerrc += er32(RXERRC);
4668 	adapter->stats.cexterr += er32(CEXTERR);
4669 	adapter->stats.tsctc += er32(TSCTC);
4670 	adapter->stats.tsctfc += er32(TSCTFC);
4671 
4672 	/* Fill out the OS statistics structure */
4673 	netdev->stats.multicast = adapter->stats.mprc;
4674 	netdev->stats.collisions = adapter->stats.colc;
4675 
4676 	/* Rx Errors */
4677 
4678 	/* RLEC on some newer hardware can be incorrect so build
4679 	 * our own version based on RUC and ROC
4680 	 */
4681 	netdev->stats.rx_errors = adapter->stats.rxerrc +
4682 	    adapter->stats.crcerrs + adapter->stats.algnerrc +
4683 	    adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
4684 	netdev->stats.rx_length_errors = adapter->stats.ruc +
4685 	    adapter->stats.roc;
4686 	netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4687 	netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4688 	netdev->stats.rx_missed_errors = adapter->stats.mpc;
4689 
4690 	/* Tx Errors */
4691 	netdev->stats.tx_errors = adapter->stats.ecol + adapter->stats.latecol;
4692 	netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4693 	netdev->stats.tx_window_errors = adapter->stats.latecol;
4694 	netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4695 
4696 	/* Tx Dropped needs to be maintained elsewhere */
4697 
4698 	/* Management Stats */
4699 	adapter->stats.mgptc += er32(MGTPTC);
4700 	adapter->stats.mgprc += er32(MGTPRC);
4701 	adapter->stats.mgpdc += er32(MGTPDC);
4702 
4703 	/* Correctable ECC Errors */
4704 	if (hw->mac.type == e1000_pch_lpt) {
4705 		u32 pbeccsts = er32(PBECCSTS);
4706 
4707 		adapter->corr_errors +=
4708 		    pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
4709 		adapter->uncorr_errors +=
4710 		    (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
4711 		    E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
4712 	}
4713 }
4714 
4715 /**
4716  * e1000_phy_read_status - Update the PHY register status snapshot
4717  * @adapter: board private structure
4718  **/
4719 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4720 {
4721 	struct e1000_hw *hw = &adapter->hw;
4722 	struct e1000_phy_regs *phy = &adapter->phy_regs;
4723 
4724 	if (!pm_runtime_suspended((&adapter->pdev->dev)->parent) &&
4725 	    (er32(STATUS) & E1000_STATUS_LU) &&
4726 	    (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4727 		int ret_val;
4728 
4729 		ret_val = e1e_rphy(hw, MII_BMCR, &phy->bmcr);
4730 		ret_val |= e1e_rphy(hw, MII_BMSR, &phy->bmsr);
4731 		ret_val |= e1e_rphy(hw, MII_ADVERTISE, &phy->advertise);
4732 		ret_val |= e1e_rphy(hw, MII_LPA, &phy->lpa);
4733 		ret_val |= e1e_rphy(hw, MII_EXPANSION, &phy->expansion);
4734 		ret_val |= e1e_rphy(hw, MII_CTRL1000, &phy->ctrl1000);
4735 		ret_val |= e1e_rphy(hw, MII_STAT1000, &phy->stat1000);
4736 		ret_val |= e1e_rphy(hw, MII_ESTATUS, &phy->estatus);
4737 		if (ret_val)
4738 			e_warn("Error reading PHY register\n");
4739 	} else {
4740 		/* Do not read PHY registers if link is not up
4741 		 * Set values to typical power-on defaults
4742 		 */
4743 		phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4744 		phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4745 			     BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4746 			     BMSR_ERCAP);
4747 		phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4748 				  ADVERTISE_ALL | ADVERTISE_CSMA);
4749 		phy->lpa = 0;
4750 		phy->expansion = EXPANSION_ENABLENPAGE;
4751 		phy->ctrl1000 = ADVERTISE_1000FULL;
4752 		phy->stat1000 = 0;
4753 		phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4754 	}
4755 }
4756 
4757 static void e1000_print_link_info(struct e1000_adapter *adapter)
4758 {
4759 	struct e1000_hw *hw = &adapter->hw;
4760 	u32 ctrl = er32(CTRL);
4761 
4762 	/* Link status message must follow this format for user tools */
4763 	pr_info("%s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4764 		adapter->netdev->name, adapter->link_speed,
4765 		adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
4766 		(ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
4767 		(ctrl & E1000_CTRL_RFCE) ? "Rx" :
4768 		(ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
4769 }
4770 
4771 static bool e1000e_has_link(struct e1000_adapter *adapter)
4772 {
4773 	struct e1000_hw *hw = &adapter->hw;
4774 	bool link_active = false;
4775 	s32 ret_val = 0;
4776 
4777 	/* get_link_status is set on LSC (link status) interrupt or
4778 	 * Rx sequence error interrupt.  get_link_status will stay
4779 	 * false until the check_for_link establishes link
4780 	 * for copper adapters ONLY
4781 	 */
4782 	switch (hw->phy.media_type) {
4783 	case e1000_media_type_copper:
4784 		if (hw->mac.get_link_status) {
4785 			ret_val = hw->mac.ops.check_for_link(hw);
4786 			link_active = !hw->mac.get_link_status;
4787 		} else {
4788 			link_active = true;
4789 		}
4790 		break;
4791 	case e1000_media_type_fiber:
4792 		ret_val = hw->mac.ops.check_for_link(hw);
4793 		link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4794 		break;
4795 	case e1000_media_type_internal_serdes:
4796 		ret_val = hw->mac.ops.check_for_link(hw);
4797 		link_active = adapter->hw.mac.serdes_has_link;
4798 		break;
4799 	default:
4800 	case e1000_media_type_unknown:
4801 		break;
4802 	}
4803 
4804 	if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4805 	    (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4806 		/* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4807 		e_info("Gigabit has been disabled, downgrading speed\n");
4808 	}
4809 
4810 	return link_active;
4811 }
4812 
4813 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4814 {
4815 	/* make sure the receive unit is started */
4816 	if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4817 	    (adapter->flags & FLAG_RESTART_NOW)) {
4818 		struct e1000_hw *hw = &adapter->hw;
4819 		u32 rctl = er32(RCTL);
4820 
4821 		ew32(RCTL, rctl | E1000_RCTL_EN);
4822 		adapter->flags &= ~FLAG_RESTART_NOW;
4823 	}
4824 }
4825 
4826 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4827 {
4828 	struct e1000_hw *hw = &adapter->hw;
4829 
4830 	/* With 82574 controllers, PHY needs to be checked periodically
4831 	 * for hung state and reset, if two calls return true
4832 	 */
4833 	if (e1000_check_phy_82574(hw))
4834 		adapter->phy_hang_count++;
4835 	else
4836 		adapter->phy_hang_count = 0;
4837 
4838 	if (adapter->phy_hang_count > 1) {
4839 		adapter->phy_hang_count = 0;
4840 		e_dbg("PHY appears hung - resetting\n");
4841 		schedule_work(&adapter->reset_task);
4842 	}
4843 }
4844 
4845 /**
4846  * e1000_watchdog - Timer Call-back
4847  * @data: pointer to adapter cast into an unsigned long
4848  **/
4849 static void e1000_watchdog(unsigned long data)
4850 {
4851 	struct e1000_adapter *adapter = (struct e1000_adapter *)data;
4852 
4853 	/* Do the rest outside of interrupt context */
4854 	schedule_work(&adapter->watchdog_task);
4855 
4856 	/* TODO: make this use queue_delayed_work() */
4857 }
4858 
4859 static void e1000_watchdog_task(struct work_struct *work)
4860 {
4861 	struct e1000_adapter *adapter = container_of(work,
4862 						     struct e1000_adapter,
4863 						     watchdog_task);
4864 	struct net_device *netdev = adapter->netdev;
4865 	struct e1000_mac_info *mac = &adapter->hw.mac;
4866 	struct e1000_phy_info *phy = &adapter->hw.phy;
4867 	struct e1000_ring *tx_ring = adapter->tx_ring;
4868 	struct e1000_hw *hw = &adapter->hw;
4869 	u32 link, tctl;
4870 
4871 	if (test_bit(__E1000_DOWN, &adapter->state))
4872 		return;
4873 
4874 	link = e1000e_has_link(adapter);
4875 	if ((netif_carrier_ok(netdev)) && link) {
4876 		/* Cancel scheduled suspend requests. */
4877 		pm_runtime_resume(netdev->dev.parent);
4878 
4879 		e1000e_enable_receives(adapter);
4880 		goto link_up;
4881 	}
4882 
4883 	if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4884 	    (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4885 		e1000_update_mng_vlan(adapter);
4886 
4887 	if (link) {
4888 		if (!netif_carrier_ok(netdev)) {
4889 			bool txb2b = true;
4890 
4891 			/* Cancel scheduled suspend requests. */
4892 			pm_runtime_resume(netdev->dev.parent);
4893 
4894 			/* update snapshot of PHY registers on LSC */
4895 			e1000_phy_read_status(adapter);
4896 			mac->ops.get_link_up_info(&adapter->hw,
4897 						  &adapter->link_speed,
4898 						  &adapter->link_duplex);
4899 			e1000_print_link_info(adapter);
4900 
4901 			/* check if SmartSpeed worked */
4902 			e1000e_check_downshift(hw);
4903 			if (phy->speed_downgraded)
4904 				netdev_warn(netdev,
4905 					    "Link Speed was downgraded by SmartSpeed\n");
4906 
4907 			/* On supported PHYs, check for duplex mismatch only
4908 			 * if link has autonegotiated at 10/100 half
4909 			 */
4910 			if ((hw->phy.type == e1000_phy_igp_3 ||
4911 			     hw->phy.type == e1000_phy_bm) &&
4912 			    hw->mac.autoneg &&
4913 			    (adapter->link_speed == SPEED_10 ||
4914 			     adapter->link_speed == SPEED_100) &&
4915 			    (adapter->link_duplex == HALF_DUPLEX)) {
4916 				u16 autoneg_exp;
4917 
4918 				e1e_rphy(hw, MII_EXPANSION, &autoneg_exp);
4919 
4920 				if (!(autoneg_exp & EXPANSION_NWAY))
4921 					e_info("Autonegotiated half duplex but link partner cannot autoneg.  Try forcing full duplex if link gets many collisions.\n");
4922 			}
4923 
4924 			/* adjust timeout factor according to speed/duplex */
4925 			adapter->tx_timeout_factor = 1;
4926 			switch (adapter->link_speed) {
4927 			case SPEED_10:
4928 				txb2b = false;
4929 				adapter->tx_timeout_factor = 16;
4930 				break;
4931 			case SPEED_100:
4932 				txb2b = false;
4933 				adapter->tx_timeout_factor = 10;
4934 				break;
4935 			}
4936 
4937 			/* workaround: re-program speed mode bit after
4938 			 * link-up event
4939 			 */
4940 			if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4941 			    !txb2b) {
4942 				u32 tarc0;
4943 
4944 				tarc0 = er32(TARC(0));
4945 				tarc0 &= ~SPEED_MODE_BIT;
4946 				ew32(TARC(0), tarc0);
4947 			}
4948 
4949 			/* disable TSO for pcie and 10/100 speeds, to avoid
4950 			 * some hardware issues
4951 			 */
4952 			if (!(adapter->flags & FLAG_TSO_FORCE)) {
4953 				switch (adapter->link_speed) {
4954 				case SPEED_10:
4955 				case SPEED_100:
4956 					e_info("10/100 speed: disabling TSO\n");
4957 					netdev->features &= ~NETIF_F_TSO;
4958 					netdev->features &= ~NETIF_F_TSO6;
4959 					break;
4960 				case SPEED_1000:
4961 					netdev->features |= NETIF_F_TSO;
4962 					netdev->features |= NETIF_F_TSO6;
4963 					break;
4964 				default:
4965 					/* oops */
4966 					break;
4967 				}
4968 			}
4969 
4970 			/* enable transmits in the hardware, need to do this
4971 			 * after setting TARC(0)
4972 			 */
4973 			tctl = er32(TCTL);
4974 			tctl |= E1000_TCTL_EN;
4975 			ew32(TCTL, tctl);
4976 
4977 			/* Perform any post-link-up configuration before
4978 			 * reporting link up.
4979 			 */
4980 			if (phy->ops.cfg_on_link_up)
4981 				phy->ops.cfg_on_link_up(hw);
4982 
4983 			netif_carrier_on(netdev);
4984 
4985 			if (!test_bit(__E1000_DOWN, &adapter->state))
4986 				mod_timer(&adapter->phy_info_timer,
4987 					  round_jiffies(jiffies + 2 * HZ));
4988 		}
4989 	} else {
4990 		if (netif_carrier_ok(netdev)) {
4991 			adapter->link_speed = 0;
4992 			adapter->link_duplex = 0;
4993 			/* Link status message must follow this format */
4994 			pr_info("%s NIC Link is Down\n", adapter->netdev->name);
4995 			netif_carrier_off(netdev);
4996 			if (!test_bit(__E1000_DOWN, &adapter->state))
4997 				mod_timer(&adapter->phy_info_timer,
4998 					  round_jiffies(jiffies + 2 * HZ));
4999 
5000 			/* 8000ES2LAN requires a Rx packet buffer work-around
5001 			 * on link down event; reset the controller to flush
5002 			 * the Rx packet buffer.
5003 			 */
5004 			if (adapter->flags & FLAG_RX_NEEDS_RESTART)
5005 				adapter->flags |= FLAG_RESTART_NOW;
5006 			else
5007 				pm_schedule_suspend(netdev->dev.parent,
5008 						    LINK_TIMEOUT);
5009 		}
5010 	}
5011 
5012 link_up:
5013 	spin_lock(&adapter->stats64_lock);
5014 	e1000e_update_stats(adapter);
5015 
5016 	mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
5017 	adapter->tpt_old = adapter->stats.tpt;
5018 	mac->collision_delta = adapter->stats.colc - adapter->colc_old;
5019 	adapter->colc_old = adapter->stats.colc;
5020 
5021 	adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
5022 	adapter->gorc_old = adapter->stats.gorc;
5023 	adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
5024 	adapter->gotc_old = adapter->stats.gotc;
5025 	spin_unlock(&adapter->stats64_lock);
5026 
5027 	/* If the link is lost the controller stops DMA, but
5028 	 * if there is queued Tx work it cannot be done.  So
5029 	 * reset the controller to flush the Tx packet buffers.
5030 	 */
5031 	if (!netif_carrier_ok(netdev) &&
5032 	    (e1000_desc_unused(tx_ring) + 1 < tx_ring->count))
5033 		adapter->flags |= FLAG_RESTART_NOW;
5034 
5035 	/* If reset is necessary, do it outside of interrupt context. */
5036 	if (adapter->flags & FLAG_RESTART_NOW) {
5037 		schedule_work(&adapter->reset_task);
5038 		/* return immediately since reset is imminent */
5039 		return;
5040 	}
5041 
5042 	e1000e_update_adaptive(&adapter->hw);
5043 
5044 	/* Simple mode for Interrupt Throttle Rate (ITR) */
5045 	if (adapter->itr_setting == 4) {
5046 		/* Symmetric Tx/Rx gets a reduced ITR=2000;
5047 		 * Total asymmetrical Tx or Rx gets ITR=8000;
5048 		 * everyone else is between 2000-8000.
5049 		 */
5050 		u32 goc = (adapter->gotc + adapter->gorc) / 10000;
5051 		u32 dif = (adapter->gotc > adapter->gorc ?
5052 			   adapter->gotc - adapter->gorc :
5053 			   adapter->gorc - adapter->gotc) / 10000;
5054 		u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
5055 
5056 		e1000e_write_itr(adapter, itr);
5057 	}
5058 
5059 	/* Cause software interrupt to ensure Rx ring is cleaned */
5060 	if (adapter->msix_entries)
5061 		ew32(ICS, adapter->rx_ring->ims_val);
5062 	else
5063 		ew32(ICS, E1000_ICS_RXDMT0);
5064 
5065 	/* flush pending descriptors to memory before detecting Tx hang */
5066 	e1000e_flush_descriptors(adapter);
5067 
5068 	/* Force detection of hung controller every watchdog period */
5069 	adapter->detect_tx_hung = true;
5070 
5071 	/* With 82571 controllers, LAA may be overwritten due to controller
5072 	 * reset from the other port. Set the appropriate LAA in RAR[0]
5073 	 */
5074 	if (e1000e_get_laa_state_82571(hw))
5075 		hw->mac.ops.rar_set(hw, adapter->hw.mac.addr, 0);
5076 
5077 	if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
5078 		e1000e_check_82574_phy_workaround(adapter);
5079 
5080 	/* Clear valid timestamp stuck in RXSTMPL/H due to a Rx error */
5081 	if (adapter->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE) {
5082 		if ((adapter->flags2 & FLAG2_CHECK_RX_HWTSTAMP) &&
5083 		    (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID)) {
5084 			er32(RXSTMPH);
5085 			adapter->rx_hwtstamp_cleared++;
5086 		} else {
5087 			adapter->flags2 |= FLAG2_CHECK_RX_HWTSTAMP;
5088 		}
5089 	}
5090 
5091 	/* Reset the timer */
5092 	if (!test_bit(__E1000_DOWN, &adapter->state))
5093 		mod_timer(&adapter->watchdog_timer,
5094 			  round_jiffies(jiffies + 2 * HZ));
5095 }
5096 
5097 #define E1000_TX_FLAGS_CSUM		0x00000001
5098 #define E1000_TX_FLAGS_VLAN		0x00000002
5099 #define E1000_TX_FLAGS_TSO		0x00000004
5100 #define E1000_TX_FLAGS_IPV4		0x00000008
5101 #define E1000_TX_FLAGS_NO_FCS		0x00000010
5102 #define E1000_TX_FLAGS_HWTSTAMP		0x00000020
5103 #define E1000_TX_FLAGS_VLAN_MASK	0xffff0000
5104 #define E1000_TX_FLAGS_VLAN_SHIFT	16
5105 
5106 static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb)
5107 {
5108 	struct e1000_context_desc *context_desc;
5109 	struct e1000_buffer *buffer_info;
5110 	unsigned int i;
5111 	u32 cmd_length = 0;
5112 	u16 ipcse = 0, mss;
5113 	u8 ipcss, ipcso, tucss, tucso, hdr_len;
5114 	int err;
5115 
5116 	if (!skb_is_gso(skb))
5117 		return 0;
5118 
5119 	err = skb_cow_head(skb, 0);
5120 	if (err < 0)
5121 		return err;
5122 
5123 	hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5124 	mss = skb_shinfo(skb)->gso_size;
5125 	if (skb->protocol == htons(ETH_P_IP)) {
5126 		struct iphdr *iph = ip_hdr(skb);
5127 		iph->tot_len = 0;
5128 		iph->check = 0;
5129 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
5130 							 0, IPPROTO_TCP, 0);
5131 		cmd_length = E1000_TXD_CMD_IP;
5132 		ipcse = skb_transport_offset(skb) - 1;
5133 	} else if (skb_is_gso_v6(skb)) {
5134 		ipv6_hdr(skb)->payload_len = 0;
5135 		tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5136 						       &ipv6_hdr(skb)->daddr,
5137 						       0, IPPROTO_TCP, 0);
5138 		ipcse = 0;
5139 	}
5140 	ipcss = skb_network_offset(skb);
5141 	ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
5142 	tucss = skb_transport_offset(skb);
5143 	tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
5144 
5145 	cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
5146 		       E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
5147 
5148 	i = tx_ring->next_to_use;
5149 	context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5150 	buffer_info = &tx_ring->buffer_info[i];
5151 
5152 	context_desc->lower_setup.ip_fields.ipcss = ipcss;
5153 	context_desc->lower_setup.ip_fields.ipcso = ipcso;
5154 	context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
5155 	context_desc->upper_setup.tcp_fields.tucss = tucss;
5156 	context_desc->upper_setup.tcp_fields.tucso = tucso;
5157 	context_desc->upper_setup.tcp_fields.tucse = 0;
5158 	context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
5159 	context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
5160 	context_desc->cmd_and_length = cpu_to_le32(cmd_length);
5161 
5162 	buffer_info->time_stamp = jiffies;
5163 	buffer_info->next_to_watch = i;
5164 
5165 	i++;
5166 	if (i == tx_ring->count)
5167 		i = 0;
5168 	tx_ring->next_to_use = i;
5169 
5170 	return 1;
5171 }
5172 
5173 static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb)
5174 {
5175 	struct e1000_adapter *adapter = tx_ring->adapter;
5176 	struct e1000_context_desc *context_desc;
5177 	struct e1000_buffer *buffer_info;
5178 	unsigned int i;
5179 	u8 css;
5180 	u32 cmd_len = E1000_TXD_CMD_DEXT;
5181 	__be16 protocol;
5182 
5183 	if (skb->ip_summed != CHECKSUM_PARTIAL)
5184 		return false;
5185 
5186 	if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
5187 		protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
5188 	else
5189 		protocol = skb->protocol;
5190 
5191 	switch (protocol) {
5192 	case cpu_to_be16(ETH_P_IP):
5193 		if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5194 			cmd_len |= E1000_TXD_CMD_TCP;
5195 		break;
5196 	case cpu_to_be16(ETH_P_IPV6):
5197 		/* XXX not handling all IPV6 headers */
5198 		if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5199 			cmd_len |= E1000_TXD_CMD_TCP;
5200 		break;
5201 	default:
5202 		if (unlikely(net_ratelimit()))
5203 			e_warn("checksum_partial proto=%x!\n",
5204 			       be16_to_cpu(protocol));
5205 		break;
5206 	}
5207 
5208 	css = skb_checksum_start_offset(skb);
5209 
5210 	i = tx_ring->next_to_use;
5211 	buffer_info = &tx_ring->buffer_info[i];
5212 	context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5213 
5214 	context_desc->lower_setup.ip_config = 0;
5215 	context_desc->upper_setup.tcp_fields.tucss = css;
5216 	context_desc->upper_setup.tcp_fields.tucso = css + skb->csum_offset;
5217 	context_desc->upper_setup.tcp_fields.tucse = 0;
5218 	context_desc->tcp_seg_setup.data = 0;
5219 	context_desc->cmd_and_length = cpu_to_le32(cmd_len);
5220 
5221 	buffer_info->time_stamp = jiffies;
5222 	buffer_info->next_to_watch = i;
5223 
5224 	i++;
5225 	if (i == tx_ring->count)
5226 		i = 0;
5227 	tx_ring->next_to_use = i;
5228 
5229 	return true;
5230 }
5231 
5232 static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
5233 			unsigned int first, unsigned int max_per_txd,
5234 			unsigned int nr_frags)
5235 {
5236 	struct e1000_adapter *adapter = tx_ring->adapter;
5237 	struct pci_dev *pdev = adapter->pdev;
5238 	struct e1000_buffer *buffer_info;
5239 	unsigned int len = skb_headlen(skb);
5240 	unsigned int offset = 0, size, count = 0, i;
5241 	unsigned int f, bytecount, segs;
5242 
5243 	i = tx_ring->next_to_use;
5244 
5245 	while (len) {
5246 		buffer_info = &tx_ring->buffer_info[i];
5247 		size = min(len, max_per_txd);
5248 
5249 		buffer_info->length = size;
5250 		buffer_info->time_stamp = jiffies;
5251 		buffer_info->next_to_watch = i;
5252 		buffer_info->dma = dma_map_single(&pdev->dev,
5253 						  skb->data + offset,
5254 						  size, DMA_TO_DEVICE);
5255 		buffer_info->mapped_as_page = false;
5256 		if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5257 			goto dma_error;
5258 
5259 		len -= size;
5260 		offset += size;
5261 		count++;
5262 
5263 		if (len) {
5264 			i++;
5265 			if (i == tx_ring->count)
5266 				i = 0;
5267 		}
5268 	}
5269 
5270 	for (f = 0; f < nr_frags; f++) {
5271 		const struct skb_frag_struct *frag;
5272 
5273 		frag = &skb_shinfo(skb)->frags[f];
5274 		len = skb_frag_size(frag);
5275 		offset = 0;
5276 
5277 		while (len) {
5278 			i++;
5279 			if (i == tx_ring->count)
5280 				i = 0;
5281 
5282 			buffer_info = &tx_ring->buffer_info[i];
5283 			size = min(len, max_per_txd);
5284 
5285 			buffer_info->length = size;
5286 			buffer_info->time_stamp = jiffies;
5287 			buffer_info->next_to_watch = i;
5288 			buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
5289 							    offset, size,
5290 							    DMA_TO_DEVICE);
5291 			buffer_info->mapped_as_page = true;
5292 			if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5293 				goto dma_error;
5294 
5295 			len -= size;
5296 			offset += size;
5297 			count++;
5298 		}
5299 	}
5300 
5301 	segs = skb_shinfo(skb)->gso_segs ? : 1;
5302 	/* multiply data chunks by size of headers */
5303 	bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
5304 
5305 	tx_ring->buffer_info[i].skb = skb;
5306 	tx_ring->buffer_info[i].segs = segs;
5307 	tx_ring->buffer_info[i].bytecount = bytecount;
5308 	tx_ring->buffer_info[first].next_to_watch = i;
5309 
5310 	return count;
5311 
5312 dma_error:
5313 	dev_err(&pdev->dev, "Tx DMA map failed\n");
5314 	buffer_info->dma = 0;
5315 	if (count)
5316 		count--;
5317 
5318 	while (count--) {
5319 		if (i == 0)
5320 			i += tx_ring->count;
5321 		i--;
5322 		buffer_info = &tx_ring->buffer_info[i];
5323 		e1000_put_txbuf(tx_ring, buffer_info);
5324 	}
5325 
5326 	return 0;
5327 }
5328 
5329 static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
5330 {
5331 	struct e1000_adapter *adapter = tx_ring->adapter;
5332 	struct e1000_tx_desc *tx_desc = NULL;
5333 	struct e1000_buffer *buffer_info;
5334 	u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
5335 	unsigned int i;
5336 
5337 	if (tx_flags & E1000_TX_FLAGS_TSO) {
5338 		txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
5339 		    E1000_TXD_CMD_TSE;
5340 		txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5341 
5342 		if (tx_flags & E1000_TX_FLAGS_IPV4)
5343 			txd_upper |= E1000_TXD_POPTS_IXSM << 8;
5344 	}
5345 
5346 	if (tx_flags & E1000_TX_FLAGS_CSUM) {
5347 		txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5348 		txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5349 	}
5350 
5351 	if (tx_flags & E1000_TX_FLAGS_VLAN) {
5352 		txd_lower |= E1000_TXD_CMD_VLE;
5353 		txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
5354 	}
5355 
5356 	if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5357 		txd_lower &= ~(E1000_TXD_CMD_IFCS);
5358 
5359 	if (unlikely(tx_flags & E1000_TX_FLAGS_HWTSTAMP)) {
5360 		txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5361 		txd_upper |= E1000_TXD_EXTCMD_TSTAMP;
5362 	}
5363 
5364 	i = tx_ring->next_to_use;
5365 
5366 	do {
5367 		buffer_info = &tx_ring->buffer_info[i];
5368 		tx_desc = E1000_TX_DESC(*tx_ring, i);
5369 		tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
5370 		tx_desc->lower.data = cpu_to_le32(txd_lower |
5371 						  buffer_info->length);
5372 		tx_desc->upper.data = cpu_to_le32(txd_upper);
5373 
5374 		i++;
5375 		if (i == tx_ring->count)
5376 			i = 0;
5377 	} while (--count > 0);
5378 
5379 	tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
5380 
5381 	/* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
5382 	if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5383 		tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
5384 
5385 	/* Force memory writes to complete before letting h/w
5386 	 * know there are new descriptors to fetch.  (Only
5387 	 * applicable for weak-ordered memory model archs,
5388 	 * such as IA-64).
5389 	 */
5390 	wmb();
5391 
5392 	tx_ring->next_to_use = i;
5393 
5394 	if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
5395 		e1000e_update_tdt_wa(tx_ring, i);
5396 	else
5397 		writel(i, tx_ring->tail);
5398 
5399 	/* we need this if more than one processor can write to our tail
5400 	 * at a time, it synchronizes IO on IA64/Altix systems
5401 	 */
5402 	mmiowb();
5403 }
5404 
5405 #define MINIMUM_DHCP_PACKET_SIZE 282
5406 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
5407 				    struct sk_buff *skb)
5408 {
5409 	struct e1000_hw *hw = &adapter->hw;
5410 	u16 length, offset;
5411 
5412 	if (vlan_tx_tag_present(skb) &&
5413 	    !((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
5414 	      (adapter->hw.mng_cookie.status &
5415 	       E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
5416 		return 0;
5417 
5418 	if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
5419 		return 0;
5420 
5421 	if (((struct ethhdr *)skb->data)->h_proto != htons(ETH_P_IP))
5422 		return 0;
5423 
5424 	{
5425 		const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data + 14);
5426 		struct udphdr *udp;
5427 
5428 		if (ip->protocol != IPPROTO_UDP)
5429 			return 0;
5430 
5431 		udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
5432 		if (ntohs(udp->dest) != 67)
5433 			return 0;
5434 
5435 		offset = (u8 *)udp + 8 - skb->data;
5436 		length = skb->len - offset;
5437 		return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
5438 	}
5439 
5440 	return 0;
5441 }
5442 
5443 static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5444 {
5445 	struct e1000_adapter *adapter = tx_ring->adapter;
5446 
5447 	netif_stop_queue(adapter->netdev);
5448 	/* Herbert's original patch had:
5449 	 *  smp_mb__after_netif_stop_queue();
5450 	 * but since that doesn't exist yet, just open code it.
5451 	 */
5452 	smp_mb();
5453 
5454 	/* We need to check again in a case another CPU has just
5455 	 * made room available.
5456 	 */
5457 	if (e1000_desc_unused(tx_ring) < size)
5458 		return -EBUSY;
5459 
5460 	/* A reprieve! */
5461 	netif_start_queue(adapter->netdev);
5462 	++adapter->restart_queue;
5463 	return 0;
5464 }
5465 
5466 static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5467 {
5468 	BUG_ON(size > tx_ring->count);
5469 
5470 	if (e1000_desc_unused(tx_ring) >= size)
5471 		return 0;
5472 	return __e1000_maybe_stop_tx(tx_ring, size);
5473 }
5474 
5475 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
5476 				    struct net_device *netdev)
5477 {
5478 	struct e1000_adapter *adapter = netdev_priv(netdev);
5479 	struct e1000_ring *tx_ring = adapter->tx_ring;
5480 	unsigned int first;
5481 	unsigned int tx_flags = 0;
5482 	unsigned int len = skb_headlen(skb);
5483 	unsigned int nr_frags;
5484 	unsigned int mss;
5485 	int count = 0;
5486 	int tso;
5487 	unsigned int f;
5488 
5489 	if (test_bit(__E1000_DOWN, &adapter->state)) {
5490 		dev_kfree_skb_any(skb);
5491 		return NETDEV_TX_OK;
5492 	}
5493 
5494 	if (skb->len <= 0) {
5495 		dev_kfree_skb_any(skb);
5496 		return NETDEV_TX_OK;
5497 	}
5498 
5499 	/* The minimum packet size with TCTL.PSP set is 17 bytes so
5500 	 * pad skb in order to meet this minimum size requirement
5501 	 */
5502 	if (unlikely(skb->len < 17)) {
5503 		if (skb_pad(skb, 17 - skb->len))
5504 			return NETDEV_TX_OK;
5505 		skb->len = 17;
5506 		skb_set_tail_pointer(skb, 17);
5507 	}
5508 
5509 	mss = skb_shinfo(skb)->gso_size;
5510 	if (mss) {
5511 		u8 hdr_len;
5512 
5513 		/* TSO Workaround for 82571/2/3 Controllers -- if skb->data
5514 		 * points to just header, pull a few bytes of payload from
5515 		 * frags into skb->data
5516 		 */
5517 		hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5518 		/* we do this workaround for ES2LAN, but it is un-necessary,
5519 		 * avoiding it could save a lot of cycles
5520 		 */
5521 		if (skb->data_len && (hdr_len == len)) {
5522 			unsigned int pull_size;
5523 
5524 			pull_size = min_t(unsigned int, 4, skb->data_len);
5525 			if (!__pskb_pull_tail(skb, pull_size)) {
5526 				e_err("__pskb_pull_tail failed.\n");
5527 				dev_kfree_skb_any(skb);
5528 				return NETDEV_TX_OK;
5529 			}
5530 			len = skb_headlen(skb);
5531 		}
5532 	}
5533 
5534 	/* reserve a descriptor for the offload context */
5535 	if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5536 		count++;
5537 	count++;
5538 
5539 	count += DIV_ROUND_UP(len, adapter->tx_fifo_limit);
5540 
5541 	nr_frags = skb_shinfo(skb)->nr_frags;
5542 	for (f = 0; f < nr_frags; f++)
5543 		count += DIV_ROUND_UP(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5544 				      adapter->tx_fifo_limit);
5545 
5546 	if (adapter->hw.mac.tx_pkt_filtering)
5547 		e1000_transfer_dhcp_info(adapter, skb);
5548 
5549 	/* need: count + 2 desc gap to keep tail from touching
5550 	 * head, otherwise try next time
5551 	 */
5552 	if (e1000_maybe_stop_tx(tx_ring, count + 2))
5553 		return NETDEV_TX_BUSY;
5554 
5555 	if (vlan_tx_tag_present(skb)) {
5556 		tx_flags |= E1000_TX_FLAGS_VLAN;
5557 		tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
5558 	}
5559 
5560 	first = tx_ring->next_to_use;
5561 
5562 	tso = e1000_tso(tx_ring, skb);
5563 	if (tso < 0) {
5564 		dev_kfree_skb_any(skb);
5565 		return NETDEV_TX_OK;
5566 	}
5567 
5568 	if (tso)
5569 		tx_flags |= E1000_TX_FLAGS_TSO;
5570 	else if (e1000_tx_csum(tx_ring, skb))
5571 		tx_flags |= E1000_TX_FLAGS_CSUM;
5572 
5573 	/* Old method was to assume IPv4 packet by default if TSO was enabled.
5574 	 * 82571 hardware supports TSO capabilities for IPv6 as well...
5575 	 * no longer assume, we must.
5576 	 */
5577 	if (skb->protocol == htons(ETH_P_IP))
5578 		tx_flags |= E1000_TX_FLAGS_IPV4;
5579 
5580 	if (unlikely(skb->no_fcs))
5581 		tx_flags |= E1000_TX_FLAGS_NO_FCS;
5582 
5583 	/* if count is 0 then mapping error has occurred */
5584 	count = e1000_tx_map(tx_ring, skb, first, adapter->tx_fifo_limit,
5585 			     nr_frags);
5586 	if (count) {
5587 		if (unlikely((skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
5588 			     !adapter->tx_hwtstamp_skb)) {
5589 			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
5590 			tx_flags |= E1000_TX_FLAGS_HWTSTAMP;
5591 			adapter->tx_hwtstamp_skb = skb_get(skb);
5592 			adapter->tx_hwtstamp_start = jiffies;
5593 			schedule_work(&adapter->tx_hwtstamp_work);
5594 		} else {
5595 			skb_tx_timestamp(skb);
5596 		}
5597 
5598 		netdev_sent_queue(netdev, skb->len);
5599 		e1000_tx_queue(tx_ring, tx_flags, count);
5600 		/* Make sure there is space in the ring for the next send. */
5601 		e1000_maybe_stop_tx(tx_ring,
5602 				    (MAX_SKB_FRAGS *
5603 				     DIV_ROUND_UP(PAGE_SIZE,
5604 						  adapter->tx_fifo_limit) + 2));
5605 	} else {
5606 		dev_kfree_skb_any(skb);
5607 		tx_ring->buffer_info[first].time_stamp = 0;
5608 		tx_ring->next_to_use = first;
5609 	}
5610 
5611 	return NETDEV_TX_OK;
5612 }
5613 
5614 /**
5615  * e1000_tx_timeout - Respond to a Tx Hang
5616  * @netdev: network interface device structure
5617  **/
5618 static void e1000_tx_timeout(struct net_device *netdev)
5619 {
5620 	struct e1000_adapter *adapter = netdev_priv(netdev);
5621 
5622 	/* Do the reset outside of interrupt context */
5623 	adapter->tx_timeout_count++;
5624 	schedule_work(&adapter->reset_task);
5625 }
5626 
5627 static void e1000_reset_task(struct work_struct *work)
5628 {
5629 	struct e1000_adapter *adapter;
5630 	adapter = container_of(work, struct e1000_adapter, reset_task);
5631 
5632 	/* don't run the task if already down */
5633 	if (test_bit(__E1000_DOWN, &adapter->state))
5634 		return;
5635 
5636 	if (!(adapter->flags & FLAG_RESTART_NOW)) {
5637 		e1000e_dump(adapter);
5638 		e_err("Reset adapter unexpectedly\n");
5639 	}
5640 	e1000e_reinit_locked(adapter);
5641 }
5642 
5643 /**
5644  * e1000_get_stats64 - Get System Network Statistics
5645  * @netdev: network interface device structure
5646  * @stats: rtnl_link_stats64 pointer
5647  *
5648  * Returns the address of the device statistics structure.
5649  **/
5650 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
5651 					     struct rtnl_link_stats64 *stats)
5652 {
5653 	struct e1000_adapter *adapter = netdev_priv(netdev);
5654 
5655 	memset(stats, 0, sizeof(struct rtnl_link_stats64));
5656 	spin_lock(&adapter->stats64_lock);
5657 	e1000e_update_stats(adapter);
5658 	/* Fill out the OS statistics structure */
5659 	stats->rx_bytes = adapter->stats.gorc;
5660 	stats->rx_packets = adapter->stats.gprc;
5661 	stats->tx_bytes = adapter->stats.gotc;
5662 	stats->tx_packets = adapter->stats.gptc;
5663 	stats->multicast = adapter->stats.mprc;
5664 	stats->collisions = adapter->stats.colc;
5665 
5666 	/* Rx Errors */
5667 
5668 	/* RLEC on some newer hardware can be incorrect so build
5669 	 * our own version based on RUC and ROC
5670 	 */
5671 	stats->rx_errors = adapter->stats.rxerrc +
5672 	    adapter->stats.crcerrs + adapter->stats.algnerrc +
5673 	    adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
5674 	stats->rx_length_errors = adapter->stats.ruc + adapter->stats.roc;
5675 	stats->rx_crc_errors = adapter->stats.crcerrs;
5676 	stats->rx_frame_errors = adapter->stats.algnerrc;
5677 	stats->rx_missed_errors = adapter->stats.mpc;
5678 
5679 	/* Tx Errors */
5680 	stats->tx_errors = adapter->stats.ecol + adapter->stats.latecol;
5681 	stats->tx_aborted_errors = adapter->stats.ecol;
5682 	stats->tx_window_errors = adapter->stats.latecol;
5683 	stats->tx_carrier_errors = adapter->stats.tncrs;
5684 
5685 	/* Tx Dropped needs to be maintained elsewhere */
5686 
5687 	spin_unlock(&adapter->stats64_lock);
5688 	return stats;
5689 }
5690 
5691 /**
5692  * e1000_change_mtu - Change the Maximum Transfer Unit
5693  * @netdev: network interface device structure
5694  * @new_mtu: new value for maximum frame size
5695  *
5696  * Returns 0 on success, negative on failure
5697  **/
5698 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5699 {
5700 	struct e1000_adapter *adapter = netdev_priv(netdev);
5701 	int max_frame = new_mtu + VLAN_HLEN + ETH_HLEN + ETH_FCS_LEN;
5702 
5703 	/* Jumbo frame support */
5704 	if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
5705 	    !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5706 		e_err("Jumbo Frames not supported.\n");
5707 		return -EINVAL;
5708 	}
5709 
5710 	/* Supported frame sizes */
5711 	if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5712 	    (max_frame > adapter->max_hw_frame_size)) {
5713 		e_err("Unsupported MTU setting\n");
5714 		return -EINVAL;
5715 	}
5716 
5717 	/* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
5718 	if ((adapter->hw.mac.type >= e1000_pch2lan) &&
5719 	    !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5720 	    (new_mtu > ETH_DATA_LEN)) {
5721 		e_err("Jumbo Frames not supported on this device when CRC stripping is disabled.\n");
5722 		return -EINVAL;
5723 	}
5724 
5725 	while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5726 		usleep_range(1000, 2000);
5727 	/* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5728 	adapter->max_frame_size = max_frame;
5729 	e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5730 	netdev->mtu = new_mtu;
5731 
5732 	pm_runtime_get_sync(netdev->dev.parent);
5733 
5734 	if (netif_running(netdev))
5735 		e1000e_down(adapter, true);
5736 
5737 	/* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5738 	 * means we reserve 2 more, this pushes us to allocate from the next
5739 	 * larger slab size.
5740 	 * i.e. RXBUFFER_2048 --> size-4096 slab
5741 	 * However with the new *_jumbo_rx* routines, jumbo receives will use
5742 	 * fragmented skbs
5743 	 */
5744 
5745 	if (max_frame <= 2048)
5746 		adapter->rx_buffer_len = 2048;
5747 	else
5748 		adapter->rx_buffer_len = 4096;
5749 
5750 	/* adjust allocation if LPE protects us, and we aren't using SBP */
5751 	if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5752 	    (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5753 		adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5754 		    + ETH_FCS_LEN;
5755 
5756 	if (netif_running(netdev))
5757 		e1000e_up(adapter);
5758 	else
5759 		e1000e_reset(adapter);
5760 
5761 	pm_runtime_put_sync(netdev->dev.parent);
5762 
5763 	clear_bit(__E1000_RESETTING, &adapter->state);
5764 
5765 	return 0;
5766 }
5767 
5768 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5769 			   int cmd)
5770 {
5771 	struct e1000_adapter *adapter = netdev_priv(netdev);
5772 	struct mii_ioctl_data *data = if_mii(ifr);
5773 
5774 	if (adapter->hw.phy.media_type != e1000_media_type_copper)
5775 		return -EOPNOTSUPP;
5776 
5777 	switch (cmd) {
5778 	case SIOCGMIIPHY:
5779 		data->phy_id = adapter->hw.phy.addr;
5780 		break;
5781 	case SIOCGMIIREG:
5782 		e1000_phy_read_status(adapter);
5783 
5784 		switch (data->reg_num & 0x1F) {
5785 		case MII_BMCR:
5786 			data->val_out = adapter->phy_regs.bmcr;
5787 			break;
5788 		case MII_BMSR:
5789 			data->val_out = adapter->phy_regs.bmsr;
5790 			break;
5791 		case MII_PHYSID1:
5792 			data->val_out = (adapter->hw.phy.id >> 16);
5793 			break;
5794 		case MII_PHYSID2:
5795 			data->val_out = (adapter->hw.phy.id & 0xFFFF);
5796 			break;
5797 		case MII_ADVERTISE:
5798 			data->val_out = adapter->phy_regs.advertise;
5799 			break;
5800 		case MII_LPA:
5801 			data->val_out = adapter->phy_regs.lpa;
5802 			break;
5803 		case MII_EXPANSION:
5804 			data->val_out = adapter->phy_regs.expansion;
5805 			break;
5806 		case MII_CTRL1000:
5807 			data->val_out = adapter->phy_regs.ctrl1000;
5808 			break;
5809 		case MII_STAT1000:
5810 			data->val_out = adapter->phy_regs.stat1000;
5811 			break;
5812 		case MII_ESTATUS:
5813 			data->val_out = adapter->phy_regs.estatus;
5814 			break;
5815 		default:
5816 			return -EIO;
5817 		}
5818 		break;
5819 	case SIOCSMIIREG:
5820 	default:
5821 		return -EOPNOTSUPP;
5822 	}
5823 	return 0;
5824 }
5825 
5826 /**
5827  * e1000e_hwtstamp_ioctl - control hardware time stamping
5828  * @netdev: network interface device structure
5829  * @ifreq: interface request
5830  *
5831  * Outgoing time stamping can be enabled and disabled. Play nice and
5832  * disable it when requested, although it shouldn't cause any overhead
5833  * when no packet needs it. At most one packet in the queue may be
5834  * marked for time stamping, otherwise it would be impossible to tell
5835  * for sure to which packet the hardware time stamp belongs.
5836  *
5837  * Incoming time stamping has to be configured via the hardware filters.
5838  * Not all combinations are supported, in particular event type has to be
5839  * specified. Matching the kind of event packet is not supported, with the
5840  * exception of "all V2 events regardless of level 2 or 4".
5841  **/
5842 static int e1000e_hwtstamp_set(struct net_device *netdev, struct ifreq *ifr)
5843 {
5844 	struct e1000_adapter *adapter = netdev_priv(netdev);
5845 	struct hwtstamp_config config;
5846 	int ret_val;
5847 
5848 	if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
5849 		return -EFAULT;
5850 
5851 	ret_val = e1000e_config_hwtstamp(adapter, &config);
5852 	if (ret_val)
5853 		return ret_val;
5854 
5855 	switch (config.rx_filter) {
5856 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
5857 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
5858 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
5859 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
5860 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
5861 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
5862 		/* With V2 type filters which specify a Sync or Delay Request,
5863 		 * Path Delay Request/Response messages are also time stamped
5864 		 * by hardware so notify the caller the requested packets plus
5865 		 * some others are time stamped.
5866 		 */
5867 		config.rx_filter = HWTSTAMP_FILTER_SOME;
5868 		break;
5869 	default:
5870 		break;
5871 	}
5872 
5873 	return copy_to_user(ifr->ifr_data, &config,
5874 			    sizeof(config)) ? -EFAULT : 0;
5875 }
5876 
5877 static int e1000e_hwtstamp_get(struct net_device *netdev, struct ifreq *ifr)
5878 {
5879 	struct e1000_adapter *adapter = netdev_priv(netdev);
5880 
5881 	return copy_to_user(ifr->ifr_data, &adapter->hwtstamp_config,
5882 			    sizeof(adapter->hwtstamp_config)) ? -EFAULT : 0;
5883 }
5884 
5885 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5886 {
5887 	switch (cmd) {
5888 	case SIOCGMIIPHY:
5889 	case SIOCGMIIREG:
5890 	case SIOCSMIIREG:
5891 		return e1000_mii_ioctl(netdev, ifr, cmd);
5892 	case SIOCSHWTSTAMP:
5893 		return e1000e_hwtstamp_set(netdev, ifr);
5894 	case SIOCGHWTSTAMP:
5895 		return e1000e_hwtstamp_get(netdev, ifr);
5896 	default:
5897 		return -EOPNOTSUPP;
5898 	}
5899 }
5900 
5901 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5902 {
5903 	struct e1000_hw *hw = &adapter->hw;
5904 	u32 i, mac_reg, wuc;
5905 	u16 phy_reg, wuc_enable;
5906 	int retval;
5907 
5908 	/* copy MAC RARs to PHY RARs */
5909 	e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5910 
5911 	retval = hw->phy.ops.acquire(hw);
5912 	if (retval) {
5913 		e_err("Could not acquire PHY\n");
5914 		return retval;
5915 	}
5916 
5917 	/* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
5918 	retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5919 	if (retval)
5920 		goto release;
5921 
5922 	/* copy MAC MTA to PHY MTA - only needed for pchlan */
5923 	for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5924 		mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5925 		hw->phy.ops.write_reg_page(hw, BM_MTA(i),
5926 					   (u16)(mac_reg & 0xFFFF));
5927 		hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
5928 					   (u16)((mac_reg >> 16) & 0xFFFF));
5929 	}
5930 
5931 	/* configure PHY Rx Control register */
5932 	hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
5933 	mac_reg = er32(RCTL);
5934 	if (mac_reg & E1000_RCTL_UPE)
5935 		phy_reg |= BM_RCTL_UPE;
5936 	if (mac_reg & E1000_RCTL_MPE)
5937 		phy_reg |= BM_RCTL_MPE;
5938 	phy_reg &= ~(BM_RCTL_MO_MASK);
5939 	if (mac_reg & E1000_RCTL_MO_3)
5940 		phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5941 			    << BM_RCTL_MO_SHIFT);
5942 	if (mac_reg & E1000_RCTL_BAM)
5943 		phy_reg |= BM_RCTL_BAM;
5944 	if (mac_reg & E1000_RCTL_PMCF)
5945 		phy_reg |= BM_RCTL_PMCF;
5946 	mac_reg = er32(CTRL);
5947 	if (mac_reg & E1000_CTRL_RFCE)
5948 		phy_reg |= BM_RCTL_RFCE;
5949 	hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
5950 
5951 	wuc = E1000_WUC_PME_EN;
5952 	if (wufc & (E1000_WUFC_MAG | E1000_WUFC_LNKC))
5953 		wuc |= E1000_WUC_APME;
5954 
5955 	/* enable PHY wakeup in MAC register */
5956 	ew32(WUFC, wufc);
5957 	ew32(WUC, (E1000_WUC_PHY_WAKE | E1000_WUC_APMPME |
5958 		   E1000_WUC_PME_STATUS | wuc));
5959 
5960 	/* configure and enable PHY wakeup in PHY registers */
5961 	hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
5962 	hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, wuc);
5963 
5964 	/* activate PHY wakeup */
5965 	wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5966 	retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5967 	if (retval)
5968 		e_err("Could not set PHY Host Wakeup bit\n");
5969 release:
5970 	hw->phy.ops.release(hw);
5971 
5972 	return retval;
5973 }
5974 
5975 static int e1000e_pm_freeze(struct device *dev)
5976 {
5977 	struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
5978 	struct e1000_adapter *adapter = netdev_priv(netdev);
5979 
5980 	netif_device_detach(netdev);
5981 
5982 	if (netif_running(netdev)) {
5983 		int count = E1000_CHECK_RESET_COUNT;
5984 
5985 		while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
5986 			usleep_range(10000, 20000);
5987 
5988 		WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5989 
5990 		/* Quiesce the device without resetting the hardware */
5991 		e1000e_down(adapter, false);
5992 		e1000_free_irq(adapter);
5993 	}
5994 	e1000e_reset_interrupt_capability(adapter);
5995 
5996 	/* Allow time for pending master requests to run */
5997 	e1000e_disable_pcie_master(&adapter->hw);
5998 
5999 	return 0;
6000 }
6001 
6002 static int __e1000_shutdown(struct pci_dev *pdev, bool runtime)
6003 {
6004 	struct net_device *netdev = pci_get_drvdata(pdev);
6005 	struct e1000_adapter *adapter = netdev_priv(netdev);
6006 	struct e1000_hw *hw = &adapter->hw;
6007 	u32 ctrl, ctrl_ext, rctl, status;
6008 	/* Runtime suspend should only enable wakeup for link changes */
6009 	u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
6010 	int retval = 0;
6011 
6012 	status = er32(STATUS);
6013 	if (status & E1000_STATUS_LU)
6014 		wufc &= ~E1000_WUFC_LNKC;
6015 
6016 	if (wufc) {
6017 		e1000_setup_rctl(adapter);
6018 		e1000e_set_rx_mode(netdev);
6019 
6020 		/* turn on all-multi mode if wake on multicast is enabled */
6021 		if (wufc & E1000_WUFC_MC) {
6022 			rctl = er32(RCTL);
6023 			rctl |= E1000_RCTL_MPE;
6024 			ew32(RCTL, rctl);
6025 		}
6026 
6027 		ctrl = er32(CTRL);
6028 		ctrl |= E1000_CTRL_ADVD3WUC;
6029 		if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
6030 			ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
6031 		ew32(CTRL, ctrl);
6032 
6033 		if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
6034 		    adapter->hw.phy.media_type ==
6035 		    e1000_media_type_internal_serdes) {
6036 			/* keep the laser running in D3 */
6037 			ctrl_ext = er32(CTRL_EXT);
6038 			ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
6039 			ew32(CTRL_EXT, ctrl_ext);
6040 		}
6041 
6042 		if (!runtime)
6043 			e1000e_power_up_phy(adapter);
6044 
6045 		if (adapter->flags & FLAG_IS_ICH)
6046 			e1000_suspend_workarounds_ich8lan(&adapter->hw);
6047 
6048 		if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6049 			/* enable wakeup by the PHY */
6050 			retval = e1000_init_phy_wakeup(adapter, wufc);
6051 			if (retval)
6052 				return retval;
6053 		} else {
6054 			/* enable wakeup by the MAC */
6055 			ew32(WUFC, wufc);
6056 			ew32(WUC, E1000_WUC_PME_EN);
6057 		}
6058 	} else {
6059 		ew32(WUC, 0);
6060 		ew32(WUFC, 0);
6061 
6062 		e1000_power_down_phy(adapter);
6063 	}
6064 
6065 	if (adapter->hw.phy.type == e1000_phy_igp_3) {
6066 		e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
6067 	} else if (hw->mac.type == e1000_pch_lpt) {
6068 		if (!(wufc & (E1000_WUFC_EX | E1000_WUFC_MC | E1000_WUFC_BC)))
6069 			/* ULP does not support wake from unicast, multicast
6070 			 * or broadcast.
6071 			 */
6072 			retval = e1000_enable_ulp_lpt_lp(hw, !runtime);
6073 
6074 		if (retval)
6075 			return retval;
6076 	}
6077 
6078 
6079 	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
6080 	 * would have already happened in close and is redundant.
6081 	 */
6082 	e1000e_release_hw_control(adapter);
6083 
6084 	pci_clear_master(pdev);
6085 
6086 	/* The pci-e switch on some quad port adapters will report a
6087 	 * correctable error when the MAC transitions from D0 to D3.  To
6088 	 * prevent this we need to mask off the correctable errors on the
6089 	 * downstream port of the pci-e switch.
6090 	 *
6091 	 * We don't have the associated upstream bridge while assigning
6092 	 * the PCI device into guest. For example, the KVM on power is
6093 	 * one of the cases.
6094 	 */
6095 	if (adapter->flags & FLAG_IS_QUAD_PORT) {
6096 		struct pci_dev *us_dev = pdev->bus->self;
6097 		u16 devctl;
6098 
6099 		if (!us_dev)
6100 			return 0;
6101 
6102 		pcie_capability_read_word(us_dev, PCI_EXP_DEVCTL, &devctl);
6103 		pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL,
6104 					   (devctl & ~PCI_EXP_DEVCTL_CERE));
6105 
6106 		pci_save_state(pdev);
6107 		pci_prepare_to_sleep(pdev);
6108 
6109 		pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL, devctl);
6110 	}
6111 
6112 	return 0;
6113 }
6114 
6115 /**
6116  * e1000e_disable_aspm - Disable ASPM states
6117  * @pdev: pointer to PCI device struct
6118  * @state: bit-mask of ASPM states to disable
6119  *
6120  * Some devices *must* have certain ASPM states disabled per hardware errata.
6121  **/
6122 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
6123 {
6124 	struct pci_dev *parent = pdev->bus->self;
6125 	u16 aspm_dis_mask = 0;
6126 	u16 pdev_aspmc, parent_aspmc;
6127 
6128 	switch (state) {
6129 	case PCIE_LINK_STATE_L0S:
6130 	case PCIE_LINK_STATE_L0S | PCIE_LINK_STATE_L1:
6131 		aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L0S;
6132 		/* fall-through - can't have L1 without L0s */
6133 	case PCIE_LINK_STATE_L1:
6134 		aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L1;
6135 		break;
6136 	default:
6137 		return;
6138 	}
6139 
6140 	pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6141 	pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6142 
6143 	if (parent) {
6144 		pcie_capability_read_word(parent, PCI_EXP_LNKCTL,
6145 					  &parent_aspmc);
6146 		parent_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6147 	}
6148 
6149 	/* Nothing to do if the ASPM states to be disabled already are */
6150 	if (!(pdev_aspmc & aspm_dis_mask) &&
6151 	    (!parent || !(parent_aspmc & aspm_dis_mask)))
6152 		return;
6153 
6154 	dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
6155 		 (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L0S) ?
6156 		 "L0s" : "",
6157 		 (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L1) ?
6158 		 "L1" : "");
6159 
6160 #ifdef CONFIG_PCIEASPM
6161 	pci_disable_link_state_locked(pdev, state);
6162 
6163 	/* Double-check ASPM control.  If not disabled by the above, the
6164 	 * BIOS is preventing that from happening (or CONFIG_PCIEASPM is
6165 	 * not enabled); override by writing PCI config space directly.
6166 	 */
6167 	pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6168 	pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6169 
6170 	if (!(aspm_dis_mask & pdev_aspmc))
6171 		return;
6172 #endif
6173 
6174 	/* Both device and parent should have the same ASPM setting.
6175 	 * Disable ASPM in downstream component first and then upstream.
6176 	 */
6177 	pcie_capability_clear_word(pdev, PCI_EXP_LNKCTL, aspm_dis_mask);
6178 
6179 	if (parent)
6180 		pcie_capability_clear_word(parent, PCI_EXP_LNKCTL,
6181 					   aspm_dis_mask);
6182 }
6183 
6184 #ifdef CONFIG_PM
6185 static int __e1000_resume(struct pci_dev *pdev)
6186 {
6187 	struct net_device *netdev = pci_get_drvdata(pdev);
6188 	struct e1000_adapter *adapter = netdev_priv(netdev);
6189 	struct e1000_hw *hw = &adapter->hw;
6190 	u16 aspm_disable_flag = 0;
6191 
6192 	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
6193 		aspm_disable_flag = PCIE_LINK_STATE_L0S;
6194 	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
6195 		aspm_disable_flag |= PCIE_LINK_STATE_L1;
6196 	if (aspm_disable_flag)
6197 		e1000e_disable_aspm(pdev, aspm_disable_flag);
6198 
6199 	pci_set_master(pdev);
6200 
6201 	if (hw->mac.type >= e1000_pch2lan)
6202 		e1000_resume_workarounds_pchlan(&adapter->hw);
6203 
6204 	e1000e_power_up_phy(adapter);
6205 
6206 	/* report the system wakeup cause from S3/S4 */
6207 	if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6208 		u16 phy_data;
6209 
6210 		e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
6211 		if (phy_data) {
6212 			e_info("PHY Wakeup cause - %s\n",
6213 			       phy_data & E1000_WUS_EX ? "Unicast Packet" :
6214 			       phy_data & E1000_WUS_MC ? "Multicast Packet" :
6215 			       phy_data & E1000_WUS_BC ? "Broadcast Packet" :
6216 			       phy_data & E1000_WUS_MAG ? "Magic Packet" :
6217 			       phy_data & E1000_WUS_LNKC ?
6218 			       "Link Status Change" : "other");
6219 		}
6220 		e1e_wphy(&adapter->hw, BM_WUS, ~0);
6221 	} else {
6222 		u32 wus = er32(WUS);
6223 
6224 		if (wus) {
6225 			e_info("MAC Wakeup cause - %s\n",
6226 			       wus & E1000_WUS_EX ? "Unicast Packet" :
6227 			       wus & E1000_WUS_MC ? "Multicast Packet" :
6228 			       wus & E1000_WUS_BC ? "Broadcast Packet" :
6229 			       wus & E1000_WUS_MAG ? "Magic Packet" :
6230 			       wus & E1000_WUS_LNKC ? "Link Status Change" :
6231 			       "other");
6232 		}
6233 		ew32(WUS, ~0);
6234 	}
6235 
6236 	e1000e_reset(adapter);
6237 
6238 	e1000_init_manageability_pt(adapter);
6239 
6240 	/* If the controller has AMT, do not set DRV_LOAD until the interface
6241 	 * is up.  For all other cases, let the f/w know that the h/w is now
6242 	 * under the control of the driver.
6243 	 */
6244 	if (!(adapter->flags & FLAG_HAS_AMT))
6245 		e1000e_get_hw_control(adapter);
6246 
6247 	return 0;
6248 }
6249 
6250 #ifdef CONFIG_PM_SLEEP
6251 static int e1000e_pm_thaw(struct device *dev)
6252 {
6253 	struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
6254 	struct e1000_adapter *adapter = netdev_priv(netdev);
6255 
6256 	e1000e_set_interrupt_capability(adapter);
6257 	if (netif_running(netdev)) {
6258 		u32 err = e1000_request_irq(adapter);
6259 
6260 		if (err)
6261 			return err;
6262 
6263 		e1000e_up(adapter);
6264 	}
6265 
6266 	netif_device_attach(netdev);
6267 
6268 	return 0;
6269 }
6270 
6271 static int e1000e_pm_suspend(struct device *dev)
6272 {
6273 	struct pci_dev *pdev = to_pci_dev(dev);
6274 
6275 	e1000e_pm_freeze(dev);
6276 
6277 	return __e1000_shutdown(pdev, false);
6278 }
6279 
6280 static int e1000e_pm_resume(struct device *dev)
6281 {
6282 	struct pci_dev *pdev = to_pci_dev(dev);
6283 	int rc;
6284 
6285 	rc = __e1000_resume(pdev);
6286 	if (rc)
6287 		return rc;
6288 
6289 	return e1000e_pm_thaw(dev);
6290 }
6291 #endif /* CONFIG_PM_SLEEP */
6292 
6293 #ifdef CONFIG_PM_RUNTIME
6294 static int e1000e_pm_runtime_idle(struct device *dev)
6295 {
6296 	struct pci_dev *pdev = to_pci_dev(dev);
6297 	struct net_device *netdev = pci_get_drvdata(pdev);
6298 	struct e1000_adapter *adapter = netdev_priv(netdev);
6299 
6300 	if (!e1000e_has_link(adapter))
6301 		pm_schedule_suspend(dev, 5 * MSEC_PER_SEC);
6302 
6303 	return -EBUSY;
6304 }
6305 
6306 static int e1000e_pm_runtime_resume(struct device *dev)
6307 {
6308 	struct pci_dev *pdev = to_pci_dev(dev);
6309 	struct net_device *netdev = pci_get_drvdata(pdev);
6310 	struct e1000_adapter *adapter = netdev_priv(netdev);
6311 	int rc;
6312 
6313 	rc = __e1000_resume(pdev);
6314 	if (rc)
6315 		return rc;
6316 
6317 	if (netdev->flags & IFF_UP)
6318 		rc = e1000e_up(adapter);
6319 
6320 	return rc;
6321 }
6322 
6323 static int e1000e_pm_runtime_suspend(struct device *dev)
6324 {
6325 	struct pci_dev *pdev = to_pci_dev(dev);
6326 	struct net_device *netdev = pci_get_drvdata(pdev);
6327 	struct e1000_adapter *adapter = netdev_priv(netdev);
6328 
6329 	if (netdev->flags & IFF_UP) {
6330 		int count = E1000_CHECK_RESET_COUNT;
6331 
6332 		while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
6333 			usleep_range(10000, 20000);
6334 
6335 		WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
6336 
6337 		/* Down the device without resetting the hardware */
6338 		e1000e_down(adapter, false);
6339 	}
6340 
6341 	if (__e1000_shutdown(pdev, true)) {
6342 		e1000e_pm_runtime_resume(dev);
6343 		return -EBUSY;
6344 	}
6345 
6346 	return 0;
6347 }
6348 #endif /* CONFIG_PM_RUNTIME */
6349 #endif /* CONFIG_PM */
6350 
6351 static void e1000_shutdown(struct pci_dev *pdev)
6352 {
6353 	e1000e_pm_freeze(&pdev->dev);
6354 
6355 	__e1000_shutdown(pdev, false);
6356 }
6357 
6358 #ifdef CONFIG_NET_POLL_CONTROLLER
6359 
6360 static irqreturn_t e1000_intr_msix(int __always_unused irq, void *data)
6361 {
6362 	struct net_device *netdev = data;
6363 	struct e1000_adapter *adapter = netdev_priv(netdev);
6364 
6365 	if (adapter->msix_entries) {
6366 		int vector, msix_irq;
6367 
6368 		vector = 0;
6369 		msix_irq = adapter->msix_entries[vector].vector;
6370 		disable_irq(msix_irq);
6371 		e1000_intr_msix_rx(msix_irq, netdev);
6372 		enable_irq(msix_irq);
6373 
6374 		vector++;
6375 		msix_irq = adapter->msix_entries[vector].vector;
6376 		disable_irq(msix_irq);
6377 		e1000_intr_msix_tx(msix_irq, netdev);
6378 		enable_irq(msix_irq);
6379 
6380 		vector++;
6381 		msix_irq = adapter->msix_entries[vector].vector;
6382 		disable_irq(msix_irq);
6383 		e1000_msix_other(msix_irq, netdev);
6384 		enable_irq(msix_irq);
6385 	}
6386 
6387 	return IRQ_HANDLED;
6388 }
6389 
6390 /**
6391  * e1000_netpoll
6392  * @netdev: network interface device structure
6393  *
6394  * Polling 'interrupt' - used by things like netconsole to send skbs
6395  * without having to re-enable interrupts. It's not called while
6396  * the interrupt routine is executing.
6397  */
6398 static void e1000_netpoll(struct net_device *netdev)
6399 {
6400 	struct e1000_adapter *adapter = netdev_priv(netdev);
6401 
6402 	switch (adapter->int_mode) {
6403 	case E1000E_INT_MODE_MSIX:
6404 		e1000_intr_msix(adapter->pdev->irq, netdev);
6405 		break;
6406 	case E1000E_INT_MODE_MSI:
6407 		disable_irq(adapter->pdev->irq);
6408 		e1000_intr_msi(adapter->pdev->irq, netdev);
6409 		enable_irq(adapter->pdev->irq);
6410 		break;
6411 	default:		/* E1000E_INT_MODE_LEGACY */
6412 		disable_irq(adapter->pdev->irq);
6413 		e1000_intr(adapter->pdev->irq, netdev);
6414 		enable_irq(adapter->pdev->irq);
6415 		break;
6416 	}
6417 }
6418 #endif
6419 
6420 /**
6421  * e1000_io_error_detected - called when PCI error is detected
6422  * @pdev: Pointer to PCI device
6423  * @state: The current pci connection state
6424  *
6425  * This function is called after a PCI bus error affecting
6426  * this device has been detected.
6427  */
6428 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
6429 						pci_channel_state_t state)
6430 {
6431 	struct net_device *netdev = pci_get_drvdata(pdev);
6432 	struct e1000_adapter *adapter = netdev_priv(netdev);
6433 
6434 	netif_device_detach(netdev);
6435 
6436 	if (state == pci_channel_io_perm_failure)
6437 		return PCI_ERS_RESULT_DISCONNECT;
6438 
6439 	if (netif_running(netdev))
6440 		e1000e_down(adapter, true);
6441 	pci_disable_device(pdev);
6442 
6443 	/* Request a slot slot reset. */
6444 	return PCI_ERS_RESULT_NEED_RESET;
6445 }
6446 
6447 /**
6448  * e1000_io_slot_reset - called after the pci bus has been reset.
6449  * @pdev: Pointer to PCI device
6450  *
6451  * Restart the card from scratch, as if from a cold-boot. Implementation
6452  * resembles the first-half of the e1000e_pm_resume routine.
6453  */
6454 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
6455 {
6456 	struct net_device *netdev = pci_get_drvdata(pdev);
6457 	struct e1000_adapter *adapter = netdev_priv(netdev);
6458 	struct e1000_hw *hw = &adapter->hw;
6459 	u16 aspm_disable_flag = 0;
6460 	int err;
6461 	pci_ers_result_t result;
6462 
6463 	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
6464 		aspm_disable_flag = PCIE_LINK_STATE_L0S;
6465 	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
6466 		aspm_disable_flag |= PCIE_LINK_STATE_L1;
6467 	if (aspm_disable_flag)
6468 		e1000e_disable_aspm(pdev, aspm_disable_flag);
6469 
6470 	err = pci_enable_device_mem(pdev);
6471 	if (err) {
6472 		dev_err(&pdev->dev,
6473 			"Cannot re-enable PCI device after reset.\n");
6474 		result = PCI_ERS_RESULT_DISCONNECT;
6475 	} else {
6476 		pdev->state_saved = true;
6477 		pci_restore_state(pdev);
6478 		pci_set_master(pdev);
6479 
6480 		pci_enable_wake(pdev, PCI_D3hot, 0);
6481 		pci_enable_wake(pdev, PCI_D3cold, 0);
6482 
6483 		e1000e_reset(adapter);
6484 		ew32(WUS, ~0);
6485 		result = PCI_ERS_RESULT_RECOVERED;
6486 	}
6487 
6488 	pci_cleanup_aer_uncorrect_error_status(pdev);
6489 
6490 	return result;
6491 }
6492 
6493 /**
6494  * e1000_io_resume - called when traffic can start flowing again.
6495  * @pdev: Pointer to PCI device
6496  *
6497  * This callback is called when the error recovery driver tells us that
6498  * its OK to resume normal operation. Implementation resembles the
6499  * second-half of the e1000e_pm_resume routine.
6500  */
6501 static void e1000_io_resume(struct pci_dev *pdev)
6502 {
6503 	struct net_device *netdev = pci_get_drvdata(pdev);
6504 	struct e1000_adapter *adapter = netdev_priv(netdev);
6505 
6506 	e1000_init_manageability_pt(adapter);
6507 
6508 	if (netif_running(netdev)) {
6509 		if (e1000e_up(adapter)) {
6510 			dev_err(&pdev->dev,
6511 				"can't bring device back up after reset\n");
6512 			return;
6513 		}
6514 	}
6515 
6516 	netif_device_attach(netdev);
6517 
6518 	/* If the controller has AMT, do not set DRV_LOAD until the interface
6519 	 * is up.  For all other cases, let the f/w know that the h/w is now
6520 	 * under the control of the driver.
6521 	 */
6522 	if (!(adapter->flags & FLAG_HAS_AMT))
6523 		e1000e_get_hw_control(adapter);
6524 }
6525 
6526 static void e1000_print_device_info(struct e1000_adapter *adapter)
6527 {
6528 	struct e1000_hw *hw = &adapter->hw;
6529 	struct net_device *netdev = adapter->netdev;
6530 	u32 ret_val;
6531 	u8 pba_str[E1000_PBANUM_LENGTH];
6532 
6533 	/* print bus type/speed/width info */
6534 	e_info("(PCI Express:2.5GT/s:%s) %pM\n",
6535 	       /* bus width */
6536 	       ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
6537 		"Width x1"),
6538 	       /* MAC address */
6539 	       netdev->dev_addr);
6540 	e_info("Intel(R) PRO/%s Network Connection\n",
6541 	       (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
6542 	ret_val = e1000_read_pba_string_generic(hw, pba_str,
6543 						E1000_PBANUM_LENGTH);
6544 	if (ret_val)
6545 		strlcpy((char *)pba_str, "Unknown", sizeof(pba_str));
6546 	e_info("MAC: %d, PHY: %d, PBA No: %s\n",
6547 	       hw->mac.type, hw->phy.type, pba_str);
6548 }
6549 
6550 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
6551 {
6552 	struct e1000_hw *hw = &adapter->hw;
6553 	int ret_val;
6554 	u16 buf = 0;
6555 
6556 	if (hw->mac.type != e1000_82573)
6557 		return;
6558 
6559 	ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
6560 	le16_to_cpus(&buf);
6561 	if (!ret_val && (!(buf & (1 << 0)))) {
6562 		/* Deep Smart Power Down (DSPD) */
6563 		dev_warn(&adapter->pdev->dev,
6564 			 "Warning: detected DSPD enabled in EEPROM\n");
6565 	}
6566 }
6567 
6568 static int e1000_set_features(struct net_device *netdev,
6569 			      netdev_features_t features)
6570 {
6571 	struct e1000_adapter *adapter = netdev_priv(netdev);
6572 	netdev_features_t changed = features ^ netdev->features;
6573 
6574 	if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
6575 		adapter->flags |= FLAG_TSO_FORCE;
6576 
6577 	if (!(changed & (NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_TX |
6578 			 NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_RXFCS |
6579 			 NETIF_F_RXALL)))
6580 		return 0;
6581 
6582 	if (changed & NETIF_F_RXFCS) {
6583 		if (features & NETIF_F_RXFCS) {
6584 			adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6585 		} else {
6586 			/* We need to take it back to defaults, which might mean
6587 			 * stripping is still disabled at the adapter level.
6588 			 */
6589 			if (adapter->flags2 & FLAG2_DFLT_CRC_STRIPPING)
6590 				adapter->flags2 |= FLAG2_CRC_STRIPPING;
6591 			else
6592 				adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6593 		}
6594 	}
6595 
6596 	netdev->features = features;
6597 
6598 	if (netif_running(netdev))
6599 		e1000e_reinit_locked(adapter);
6600 	else
6601 		e1000e_reset(adapter);
6602 
6603 	return 0;
6604 }
6605 
6606 static const struct net_device_ops e1000e_netdev_ops = {
6607 	.ndo_open		= e1000_open,
6608 	.ndo_stop		= e1000_close,
6609 	.ndo_start_xmit		= e1000_xmit_frame,
6610 	.ndo_get_stats64	= e1000e_get_stats64,
6611 	.ndo_set_rx_mode	= e1000e_set_rx_mode,
6612 	.ndo_set_mac_address	= e1000_set_mac,
6613 	.ndo_change_mtu		= e1000_change_mtu,
6614 	.ndo_do_ioctl		= e1000_ioctl,
6615 	.ndo_tx_timeout		= e1000_tx_timeout,
6616 	.ndo_validate_addr	= eth_validate_addr,
6617 
6618 	.ndo_vlan_rx_add_vid	= e1000_vlan_rx_add_vid,
6619 	.ndo_vlan_rx_kill_vid	= e1000_vlan_rx_kill_vid,
6620 #ifdef CONFIG_NET_POLL_CONTROLLER
6621 	.ndo_poll_controller	= e1000_netpoll,
6622 #endif
6623 	.ndo_set_features = e1000_set_features,
6624 };
6625 
6626 /**
6627  * e1000_probe - Device Initialization Routine
6628  * @pdev: PCI device information struct
6629  * @ent: entry in e1000_pci_tbl
6630  *
6631  * Returns 0 on success, negative on failure
6632  *
6633  * e1000_probe initializes an adapter identified by a pci_dev structure.
6634  * The OS initialization, configuring of the adapter private structure,
6635  * and a hardware reset occur.
6636  **/
6637 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
6638 {
6639 	struct net_device *netdev;
6640 	struct e1000_adapter *adapter;
6641 	struct e1000_hw *hw;
6642 	const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
6643 	resource_size_t mmio_start, mmio_len;
6644 	resource_size_t flash_start, flash_len;
6645 	static int cards_found;
6646 	u16 aspm_disable_flag = 0;
6647 	int bars, i, err, pci_using_dac;
6648 	u16 eeprom_data = 0;
6649 	u16 eeprom_apme_mask = E1000_EEPROM_APME;
6650 
6651 	if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
6652 		aspm_disable_flag = PCIE_LINK_STATE_L0S;
6653 	if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
6654 		aspm_disable_flag |= PCIE_LINK_STATE_L1;
6655 	if (aspm_disable_flag)
6656 		e1000e_disable_aspm(pdev, aspm_disable_flag);
6657 
6658 	err = pci_enable_device_mem(pdev);
6659 	if (err)
6660 		return err;
6661 
6662 	pci_using_dac = 0;
6663 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
6664 	if (!err) {
6665 		pci_using_dac = 1;
6666 	} else {
6667 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
6668 		if (err) {
6669 			dev_err(&pdev->dev,
6670 				"No usable DMA configuration, aborting\n");
6671 			goto err_dma;
6672 		}
6673 	}
6674 
6675 	bars = pci_select_bars(pdev, IORESOURCE_MEM);
6676 	err = pci_request_selected_regions_exclusive(pdev, bars,
6677 						     e1000e_driver_name);
6678 	if (err)
6679 		goto err_pci_reg;
6680 
6681 	/* AER (Advanced Error Reporting) hooks */
6682 	pci_enable_pcie_error_reporting(pdev);
6683 
6684 	pci_set_master(pdev);
6685 	/* PCI config space info */
6686 	err = pci_save_state(pdev);
6687 	if (err)
6688 		goto err_alloc_etherdev;
6689 
6690 	err = -ENOMEM;
6691 	netdev = alloc_etherdev(sizeof(struct e1000_adapter));
6692 	if (!netdev)
6693 		goto err_alloc_etherdev;
6694 
6695 	SET_NETDEV_DEV(netdev, &pdev->dev);
6696 
6697 	netdev->irq = pdev->irq;
6698 
6699 	pci_set_drvdata(pdev, netdev);
6700 	adapter = netdev_priv(netdev);
6701 	hw = &adapter->hw;
6702 	adapter->netdev = netdev;
6703 	adapter->pdev = pdev;
6704 	adapter->ei = ei;
6705 	adapter->pba = ei->pba;
6706 	adapter->flags = ei->flags;
6707 	adapter->flags2 = ei->flags2;
6708 	adapter->hw.adapter = adapter;
6709 	adapter->hw.mac.type = ei->mac;
6710 	adapter->max_hw_frame_size = ei->max_hw_frame_size;
6711 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
6712 
6713 	mmio_start = pci_resource_start(pdev, 0);
6714 	mmio_len = pci_resource_len(pdev, 0);
6715 
6716 	err = -EIO;
6717 	adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
6718 	if (!adapter->hw.hw_addr)
6719 		goto err_ioremap;
6720 
6721 	if ((adapter->flags & FLAG_HAS_FLASH) &&
6722 	    (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
6723 		flash_start = pci_resource_start(pdev, 1);
6724 		flash_len = pci_resource_len(pdev, 1);
6725 		adapter->hw.flash_address = ioremap(flash_start, flash_len);
6726 		if (!adapter->hw.flash_address)
6727 			goto err_flashmap;
6728 	}
6729 
6730 	/* Set default EEE advertisement */
6731 	if (adapter->flags2 & FLAG2_HAS_EEE)
6732 		adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
6733 
6734 	/* construct the net_device struct */
6735 	netdev->netdev_ops = &e1000e_netdev_ops;
6736 	e1000e_set_ethtool_ops(netdev);
6737 	netdev->watchdog_timeo = 5 * HZ;
6738 	netif_napi_add(netdev, &adapter->napi, e1000e_poll, 64);
6739 	strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
6740 
6741 	netdev->mem_start = mmio_start;
6742 	netdev->mem_end = mmio_start + mmio_len;
6743 
6744 	adapter->bd_number = cards_found++;
6745 
6746 	e1000e_check_options(adapter);
6747 
6748 	/* setup adapter struct */
6749 	err = e1000_sw_init(adapter);
6750 	if (err)
6751 		goto err_sw_init;
6752 
6753 	memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
6754 	memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
6755 	memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
6756 
6757 	err = ei->get_variants(adapter);
6758 	if (err)
6759 		goto err_hw_init;
6760 
6761 	if ((adapter->flags & FLAG_IS_ICH) &&
6762 	    (adapter->flags & FLAG_READ_ONLY_NVM))
6763 		e1000e_write_protect_nvm_ich8lan(&adapter->hw);
6764 
6765 	hw->mac.ops.get_bus_info(&adapter->hw);
6766 
6767 	adapter->hw.phy.autoneg_wait_to_complete = 0;
6768 
6769 	/* Copper options */
6770 	if (adapter->hw.phy.media_type == e1000_media_type_copper) {
6771 		adapter->hw.phy.mdix = AUTO_ALL_MODES;
6772 		adapter->hw.phy.disable_polarity_correction = 0;
6773 		adapter->hw.phy.ms_type = e1000_ms_hw_default;
6774 	}
6775 
6776 	if (hw->phy.ops.check_reset_block && hw->phy.ops.check_reset_block(hw))
6777 		dev_info(&pdev->dev,
6778 			 "PHY reset is blocked due to SOL/IDER session.\n");
6779 
6780 	/* Set initial default active device features */
6781 	netdev->features = (NETIF_F_SG |
6782 			    NETIF_F_HW_VLAN_CTAG_RX |
6783 			    NETIF_F_HW_VLAN_CTAG_TX |
6784 			    NETIF_F_TSO |
6785 			    NETIF_F_TSO6 |
6786 			    NETIF_F_RXHASH |
6787 			    NETIF_F_RXCSUM |
6788 			    NETIF_F_HW_CSUM);
6789 
6790 	/* Set user-changeable features (subset of all device features) */
6791 	netdev->hw_features = netdev->features;
6792 	netdev->hw_features |= NETIF_F_RXFCS;
6793 	netdev->priv_flags |= IFF_SUPP_NOFCS;
6794 	netdev->hw_features |= NETIF_F_RXALL;
6795 
6796 	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
6797 		netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
6798 
6799 	netdev->vlan_features |= (NETIF_F_SG |
6800 				  NETIF_F_TSO |
6801 				  NETIF_F_TSO6 |
6802 				  NETIF_F_HW_CSUM);
6803 
6804 	netdev->priv_flags |= IFF_UNICAST_FLT;
6805 
6806 	if (pci_using_dac) {
6807 		netdev->features |= NETIF_F_HIGHDMA;
6808 		netdev->vlan_features |= NETIF_F_HIGHDMA;
6809 	}
6810 
6811 	if (e1000e_enable_mng_pass_thru(&adapter->hw))
6812 		adapter->flags |= FLAG_MNG_PT_ENABLED;
6813 
6814 	/* before reading the NVM, reset the controller to
6815 	 * put the device in a known good starting state
6816 	 */
6817 	adapter->hw.mac.ops.reset_hw(&adapter->hw);
6818 
6819 	/* systems with ASPM and others may see the checksum fail on the first
6820 	 * attempt. Let's give it a few tries
6821 	 */
6822 	for (i = 0;; i++) {
6823 		if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
6824 			break;
6825 		if (i == 2) {
6826 			dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
6827 			err = -EIO;
6828 			goto err_eeprom;
6829 		}
6830 	}
6831 
6832 	e1000_eeprom_checks(adapter);
6833 
6834 	/* copy the MAC address */
6835 	if (e1000e_read_mac_addr(&adapter->hw))
6836 		dev_err(&pdev->dev,
6837 			"NVM Read Error while reading MAC address\n");
6838 
6839 	memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
6840 
6841 	if (!is_valid_ether_addr(netdev->dev_addr)) {
6842 		dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
6843 			netdev->dev_addr);
6844 		err = -EIO;
6845 		goto err_eeprom;
6846 	}
6847 
6848 	init_timer(&adapter->watchdog_timer);
6849 	adapter->watchdog_timer.function = e1000_watchdog;
6850 	adapter->watchdog_timer.data = (unsigned long)adapter;
6851 
6852 	init_timer(&adapter->phy_info_timer);
6853 	adapter->phy_info_timer.function = e1000_update_phy_info;
6854 	adapter->phy_info_timer.data = (unsigned long)adapter;
6855 
6856 	INIT_WORK(&adapter->reset_task, e1000_reset_task);
6857 	INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
6858 	INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
6859 	INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
6860 	INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
6861 
6862 	/* Initialize link parameters. User can change them with ethtool */
6863 	adapter->hw.mac.autoneg = 1;
6864 	adapter->fc_autoneg = true;
6865 	adapter->hw.fc.requested_mode = e1000_fc_default;
6866 	adapter->hw.fc.current_mode = e1000_fc_default;
6867 	adapter->hw.phy.autoneg_advertised = 0x2f;
6868 
6869 	/* Initial Wake on LAN setting - If APM wake is enabled in
6870 	 * the EEPROM, enable the ACPI Magic Packet filter
6871 	 */
6872 	if (adapter->flags & FLAG_APME_IN_WUC) {
6873 		/* APME bit in EEPROM is mapped to WUC.APME */
6874 		eeprom_data = er32(WUC);
6875 		eeprom_apme_mask = E1000_WUC_APME;
6876 		if ((hw->mac.type > e1000_ich10lan) &&
6877 		    (eeprom_data & E1000_WUC_PHY_WAKE))
6878 			adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
6879 	} else if (adapter->flags & FLAG_APME_IN_CTRL3) {
6880 		if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
6881 		    (adapter->hw.bus.func == 1))
6882 			e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_B,
6883 				       1, &eeprom_data);
6884 		else
6885 			e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_A,
6886 				       1, &eeprom_data);
6887 	}
6888 
6889 	/* fetch WoL from EEPROM */
6890 	if (eeprom_data & eeprom_apme_mask)
6891 		adapter->eeprom_wol |= E1000_WUFC_MAG;
6892 
6893 	/* now that we have the eeprom settings, apply the special cases
6894 	 * where the eeprom may be wrong or the board simply won't support
6895 	 * wake on lan on a particular port
6896 	 */
6897 	if (!(adapter->flags & FLAG_HAS_WOL))
6898 		adapter->eeprom_wol = 0;
6899 
6900 	/* initialize the wol settings based on the eeprom settings */
6901 	adapter->wol = adapter->eeprom_wol;
6902 
6903 	/* make sure adapter isn't asleep if manageability is enabled */
6904 	if (adapter->wol || (adapter->flags & FLAG_MNG_PT_ENABLED) ||
6905 	    (hw->mac.ops.check_mng_mode(hw)))
6906 		device_wakeup_enable(&pdev->dev);
6907 
6908 	/* save off EEPROM version number */
6909 	e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6910 
6911 	/* reset the hardware with the new settings */
6912 	e1000e_reset(adapter);
6913 
6914 	/* If the controller has AMT, do not set DRV_LOAD until the interface
6915 	 * is up.  For all other cases, let the f/w know that the h/w is now
6916 	 * under the control of the driver.
6917 	 */
6918 	if (!(adapter->flags & FLAG_HAS_AMT))
6919 		e1000e_get_hw_control(adapter);
6920 
6921 	strlcpy(netdev->name, "eth%d", sizeof(netdev->name));
6922 	err = register_netdev(netdev);
6923 	if (err)
6924 		goto err_register;
6925 
6926 	/* carrier off reporting is important to ethtool even BEFORE open */
6927 	netif_carrier_off(netdev);
6928 
6929 	/* init PTP hardware clock */
6930 	e1000e_ptp_init(adapter);
6931 
6932 	e1000_print_device_info(adapter);
6933 
6934 	if (pci_dev_run_wake(pdev))
6935 		pm_runtime_put_noidle(&pdev->dev);
6936 
6937 	return 0;
6938 
6939 err_register:
6940 	if (!(adapter->flags & FLAG_HAS_AMT))
6941 		e1000e_release_hw_control(adapter);
6942 err_eeprom:
6943 	if (hw->phy.ops.check_reset_block && !hw->phy.ops.check_reset_block(hw))
6944 		e1000_phy_hw_reset(&adapter->hw);
6945 err_hw_init:
6946 	kfree(adapter->tx_ring);
6947 	kfree(adapter->rx_ring);
6948 err_sw_init:
6949 	if (adapter->hw.flash_address)
6950 		iounmap(adapter->hw.flash_address);
6951 	e1000e_reset_interrupt_capability(adapter);
6952 err_flashmap:
6953 	iounmap(adapter->hw.hw_addr);
6954 err_ioremap:
6955 	free_netdev(netdev);
6956 err_alloc_etherdev:
6957 	pci_release_selected_regions(pdev,
6958 				     pci_select_bars(pdev, IORESOURCE_MEM));
6959 err_pci_reg:
6960 err_dma:
6961 	pci_disable_device(pdev);
6962 	return err;
6963 }
6964 
6965 /**
6966  * e1000_remove - Device Removal Routine
6967  * @pdev: PCI device information struct
6968  *
6969  * e1000_remove is called by the PCI subsystem to alert the driver
6970  * that it should release a PCI device.  The could be caused by a
6971  * Hot-Plug event, or because the driver is going to be removed from
6972  * memory.
6973  **/
6974 static void e1000_remove(struct pci_dev *pdev)
6975 {
6976 	struct net_device *netdev = pci_get_drvdata(pdev);
6977 	struct e1000_adapter *adapter = netdev_priv(netdev);
6978 	bool down = test_bit(__E1000_DOWN, &adapter->state);
6979 
6980 	e1000e_ptp_remove(adapter);
6981 
6982 	/* The timers may be rescheduled, so explicitly disable them
6983 	 * from being rescheduled.
6984 	 */
6985 	if (!down)
6986 		set_bit(__E1000_DOWN, &adapter->state);
6987 	del_timer_sync(&adapter->watchdog_timer);
6988 	del_timer_sync(&adapter->phy_info_timer);
6989 
6990 	cancel_work_sync(&adapter->reset_task);
6991 	cancel_work_sync(&adapter->watchdog_task);
6992 	cancel_work_sync(&adapter->downshift_task);
6993 	cancel_work_sync(&adapter->update_phy_task);
6994 	cancel_work_sync(&adapter->print_hang_task);
6995 
6996 	if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
6997 		cancel_work_sync(&adapter->tx_hwtstamp_work);
6998 		if (adapter->tx_hwtstamp_skb) {
6999 			dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
7000 			adapter->tx_hwtstamp_skb = NULL;
7001 		}
7002 	}
7003 
7004 	/* Don't lie to e1000_close() down the road. */
7005 	if (!down)
7006 		clear_bit(__E1000_DOWN, &adapter->state);
7007 	unregister_netdev(netdev);
7008 
7009 	if (pci_dev_run_wake(pdev))
7010 		pm_runtime_get_noresume(&pdev->dev);
7011 
7012 	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
7013 	 * would have already happened in close and is redundant.
7014 	 */
7015 	e1000e_release_hw_control(adapter);
7016 
7017 	e1000e_reset_interrupt_capability(adapter);
7018 	kfree(adapter->tx_ring);
7019 	kfree(adapter->rx_ring);
7020 
7021 	iounmap(adapter->hw.hw_addr);
7022 	if (adapter->hw.flash_address)
7023 		iounmap(adapter->hw.flash_address);
7024 	pci_release_selected_regions(pdev,
7025 				     pci_select_bars(pdev, IORESOURCE_MEM));
7026 
7027 	free_netdev(netdev);
7028 
7029 	/* AER disable */
7030 	pci_disable_pcie_error_reporting(pdev);
7031 
7032 	pci_disable_device(pdev);
7033 }
7034 
7035 /* PCI Error Recovery (ERS) */
7036 static const struct pci_error_handlers e1000_err_handler = {
7037 	.error_detected = e1000_io_error_detected,
7038 	.slot_reset = e1000_io_slot_reset,
7039 	.resume = e1000_io_resume,
7040 };
7041 
7042 static const struct pci_device_id e1000_pci_tbl[] = {
7043 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
7044 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
7045 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
7046 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP),
7047 	  board_82571 },
7048 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
7049 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
7050 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
7051 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
7052 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
7053 
7054 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
7055 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
7056 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
7057 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
7058 
7059 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
7060 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
7061 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
7062 
7063 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
7064 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
7065 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
7066 
7067 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
7068 	  board_80003es2lan },
7069 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
7070 	  board_80003es2lan },
7071 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
7072 	  board_80003es2lan },
7073 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
7074 	  board_80003es2lan },
7075 
7076 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
7077 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
7078 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
7079 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
7080 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
7081 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
7082 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
7083 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
7084 
7085 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
7086 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
7087 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
7088 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
7089 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
7090 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
7091 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
7092 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
7093 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
7094 
7095 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
7096 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
7097 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
7098 
7099 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
7100 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
7101 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
7102 
7103 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
7104 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
7105 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
7106 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
7107 
7108 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
7109 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
7110 
7111 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_LM), board_pch_lpt },
7112 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_V), board_pch_lpt },
7113 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_LM), board_pch_lpt },
7114 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_V), board_pch_lpt },
7115 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM2), board_pch_lpt },
7116 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V2), board_pch_lpt },
7117 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM3), board_pch_lpt },
7118 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V3), board_pch_lpt },
7119 
7120 	{ 0, 0, 0, 0, 0, 0, 0 }	/* terminate list */
7121 };
7122 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
7123 
7124 static const struct dev_pm_ops e1000_pm_ops = {
7125 #ifdef CONFIG_PM_SLEEP
7126 	.suspend	= e1000e_pm_suspend,
7127 	.resume		= e1000e_pm_resume,
7128 	.freeze		= e1000e_pm_freeze,
7129 	.thaw		= e1000e_pm_thaw,
7130 	.poweroff	= e1000e_pm_suspend,
7131 	.restore	= e1000e_pm_resume,
7132 #endif
7133 	SET_RUNTIME_PM_OPS(e1000e_pm_runtime_suspend, e1000e_pm_runtime_resume,
7134 			   e1000e_pm_runtime_idle)
7135 };
7136 
7137 /* PCI Device API Driver */
7138 static struct pci_driver e1000_driver = {
7139 	.name     = e1000e_driver_name,
7140 	.id_table = e1000_pci_tbl,
7141 	.probe    = e1000_probe,
7142 	.remove   = e1000_remove,
7143 	.driver   = {
7144 		.pm = &e1000_pm_ops,
7145 	},
7146 	.shutdown = e1000_shutdown,
7147 	.err_handler = &e1000_err_handler
7148 };
7149 
7150 /**
7151  * e1000_init_module - Driver Registration Routine
7152  *
7153  * e1000_init_module is the first routine called when the driver is
7154  * loaded. All it does is register with the PCI subsystem.
7155  **/
7156 static int __init e1000_init_module(void)
7157 {
7158 	int ret;
7159 
7160 	pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
7161 		e1000e_driver_version);
7162 	pr_info("Copyright(c) 1999 - 2014 Intel Corporation.\n");
7163 	ret = pci_register_driver(&e1000_driver);
7164 
7165 	return ret;
7166 }
7167 module_init(e1000_init_module);
7168 
7169 /**
7170  * e1000_exit_module - Driver Exit Cleanup Routine
7171  *
7172  * e1000_exit_module is called just before the driver is removed
7173  * from memory.
7174  **/
7175 static void __exit e1000_exit_module(void)
7176 {
7177 	pci_unregister_driver(&e1000_driver);
7178 }
7179 module_exit(e1000_exit_module);
7180 
7181 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
7182 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
7183 MODULE_LICENSE("GPL");
7184 MODULE_VERSION(DRV_VERSION);
7185 
7186 /* netdev.c */
7187