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