1/* Intel PRO/1000 Linux driver
2 * Copyright(c) 1999 - 2015 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 "3.2.6" DRV_EXTRAVERSION
52char e1000e_driver_name[] = "e1000e";
53const char e1000e_driver_version[] = DRV_VERSION;
54
55#define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
56static int debug = -1;
57module_param(debug, int, 0);
58MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
59
60static 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	[board_pch_spt]		= &e1000_pch_spt_info,
74};
75
76struct e1000_reg_info {
77	u32 ofs;
78	char *name;
79};
80
81static const struct e1000_reg_info e1000_reg_info_tbl[] = {
82	/* General Registers */
83	{E1000_CTRL, "CTRL"},
84	{E1000_STATUS, "STATUS"},
85	{E1000_CTRL_EXT, "CTRL_EXT"},
86
87	/* Interrupt Registers */
88	{E1000_ICR, "ICR"},
89
90	/* Rx Registers */
91	{E1000_RCTL, "RCTL"},
92	{E1000_RDLEN(0), "RDLEN"},
93	{E1000_RDH(0), "RDH"},
94	{E1000_RDT(0), "RDT"},
95	{E1000_RDTR, "RDTR"},
96	{E1000_RXDCTL(0), "RXDCTL"},
97	{E1000_ERT, "ERT"},
98	{E1000_RDBAL(0), "RDBAL"},
99	{E1000_RDBAH(0), "RDBAH"},
100	{E1000_RDFH, "RDFH"},
101	{E1000_RDFT, "RDFT"},
102	{E1000_RDFHS, "RDFHS"},
103	{E1000_RDFTS, "RDFTS"},
104	{E1000_RDFPC, "RDFPC"},
105
106	/* Tx Registers */
107	{E1000_TCTL, "TCTL"},
108	{E1000_TDBAL(0), "TDBAL"},
109	{E1000_TDBAH(0), "TDBAH"},
110	{E1000_TDLEN(0), "TDLEN"},
111	{E1000_TDH(0), "TDH"},
112	{E1000_TDT(0), "TDT"},
113	{E1000_TIDV, "TIDV"},
114	{E1000_TXDCTL(0), "TXDCTL"},
115	{E1000_TADV, "TADV"},
116	{E1000_TARC(0), "TARC"},
117	{E1000_TDFH, "TDFH"},
118	{E1000_TDFT, "TDFT"},
119	{E1000_TDFHS, "TDFHS"},
120	{E1000_TDFTS, "TDFTS"},
121	{E1000_TDFPC, "TDFPC"},
122
123	/* List Terminator */
124	{0, NULL}
125};
126
127/**
128 * __ew32_prepare - prepare to write to MAC CSR register on certain parts
129 * @hw: pointer to the HW structure
130 *
131 * When updating the MAC CSR registers, the Manageability Engine (ME) could
132 * be accessing the registers at the same time.  Normally, this is handled in
133 * h/w by an arbiter but on some parts there is a bug that acknowledges Host
134 * accesses later than it should which could result in the register to have
135 * an incorrect value.  Workaround this by checking the FWSM register which
136 * has bit 24 set while ME is accessing MAC CSR registers, wait if it is set
137 * and try again a number of times.
138 **/
139s32 __ew32_prepare(struct e1000_hw *hw)
140{
141	s32 i = E1000_ICH_FWSM_PCIM2PCI_COUNT;
142
143	while ((er32(FWSM) & E1000_ICH_FWSM_PCIM2PCI) && --i)
144		udelay(50);
145
146	return i;
147}
148
149void __ew32(struct e1000_hw *hw, unsigned long reg, u32 val)
150{
151	if (hw->adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
152		__ew32_prepare(hw);
153
154	writel(val, hw->hw_addr + reg);
155}
156
157/**
158 * e1000_regdump - register printout routine
159 * @hw: pointer to the HW structure
160 * @reginfo: pointer to the register info table
161 **/
162static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
163{
164	int n = 0;
165	char rname[16];
166	u32 regs[8];
167
168	switch (reginfo->ofs) {
169	case E1000_RXDCTL(0):
170		for (n = 0; n < 2; n++)
171			regs[n] = __er32(hw, E1000_RXDCTL(n));
172		break;
173	case E1000_TXDCTL(0):
174		for (n = 0; n < 2; n++)
175			regs[n] = __er32(hw, E1000_TXDCTL(n));
176		break;
177	case E1000_TARC(0):
178		for (n = 0; n < 2; n++)
179			regs[n] = __er32(hw, E1000_TARC(n));
180		break;
181	default:
182		pr_info("%-15s %08x\n",
183			reginfo->name, __er32(hw, reginfo->ofs));
184		return;
185	}
186
187	snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
188	pr_info("%-15s %08x %08x\n", rname, regs[0], regs[1]);
189}
190
191static void e1000e_dump_ps_pages(struct e1000_adapter *adapter,
192				 struct e1000_buffer *bi)
193{
194	int i;
195	struct e1000_ps_page *ps_page;
196
197	for (i = 0; i < adapter->rx_ps_pages; i++) {
198		ps_page = &bi->ps_pages[i];
199
200		if (ps_page->page) {
201			pr_info("packet dump for ps_page %d:\n", i);
202			print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
203				       16, 1, page_address(ps_page->page),
204				       PAGE_SIZE, true);
205		}
206	}
207}
208
209/**
210 * e1000e_dump - Print registers, Tx-ring and Rx-ring
211 * @adapter: board private structure
212 **/
213static void e1000e_dump(struct e1000_adapter *adapter)
214{
215	struct net_device *netdev = adapter->netdev;
216	struct e1000_hw *hw = &adapter->hw;
217	struct e1000_reg_info *reginfo;
218	struct e1000_ring *tx_ring = adapter->tx_ring;
219	struct e1000_tx_desc *tx_desc;
220	struct my_u0 {
221		__le64 a;
222		__le64 b;
223	} *u0;
224	struct e1000_buffer *buffer_info;
225	struct e1000_ring *rx_ring = adapter->rx_ring;
226	union e1000_rx_desc_packet_split *rx_desc_ps;
227	union e1000_rx_desc_extended *rx_desc;
228	struct my_u1 {
229		__le64 a;
230		__le64 b;
231		__le64 c;
232		__le64 d;
233	} *u1;
234	u32 staterr;
235	int i = 0;
236
237	if (!netif_msg_hw(adapter))
238		return;
239
240	/* Print netdevice Info */
241	if (netdev) {
242		dev_info(&adapter->pdev->dev, "Net device Info\n");
243		pr_info("Device Name     state            trans_start      last_rx\n");
244		pr_info("%-15s %016lX %016lX %016lX\n", netdev->name,
245			netdev->state, netdev->trans_start, netdev->last_rx);
246	}
247
248	/* Print Registers */
249	dev_info(&adapter->pdev->dev, "Register Dump\n");
250	pr_info(" Register Name   Value\n");
251	for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
252	     reginfo->name; reginfo++) {
253		e1000_regdump(hw, reginfo);
254	}
255
256	/* Print Tx Ring Summary */
257	if (!netdev || !netif_running(netdev))
258		return;
259
260	dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
261	pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
262	buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
263	pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
264		0, tx_ring->next_to_use, tx_ring->next_to_clean,
265		(unsigned long long)buffer_info->dma,
266		buffer_info->length,
267		buffer_info->next_to_watch,
268		(unsigned long long)buffer_info->time_stamp);
269
270	/* Print Tx Ring */
271	if (!netif_msg_tx_done(adapter))
272		goto rx_ring_summary;
273
274	dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
275
276	/* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
277	 *
278	 * Legacy Transmit Descriptor
279	 *   +--------------------------------------------------------------+
280	 * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
281	 *   +--------------------------------------------------------------+
282	 * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
283	 *   +--------------------------------------------------------------+
284	 *   63       48 47        36 35    32 31     24 23    16 15        0
285	 *
286	 * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
287	 *   63      48 47    40 39       32 31             16 15    8 7      0
288	 *   +----------------------------------------------------------------+
289	 * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
290	 *   +----------------------------------------------------------------+
291	 * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
292	 *   +----------------------------------------------------------------+
293	 *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
294	 *
295	 * Extended Data Descriptor (DTYP=0x1)
296	 *   +----------------------------------------------------------------+
297	 * 0 |                     Buffer Address [63:0]                      |
298	 *   +----------------------------------------------------------------+
299	 * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
300	 *   +----------------------------------------------------------------+
301	 *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
302	 */
303	pr_info("Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Legacy format\n");
304	pr_info("Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Context format\n");
305	pr_info("Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Data format\n");
306	for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
307		const char *next_desc;
308		tx_desc = E1000_TX_DESC(*tx_ring, i);
309		buffer_info = &tx_ring->buffer_info[i];
310		u0 = (struct my_u0 *)tx_desc;
311		if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
312			next_desc = " NTC/U";
313		else if (i == tx_ring->next_to_use)
314			next_desc = " NTU";
315		else if (i == tx_ring->next_to_clean)
316			next_desc = " NTC";
317		else
318			next_desc = "";
319		pr_info("T%c[0x%03X]    %016llX %016llX %016llX %04X  %3X %016llX %p%s\n",
320			(!(le64_to_cpu(u0->b) & (1 << 29)) ? 'l' :
321			 ((le64_to_cpu(u0->b) & (1 << 20)) ? 'd' : 'c')),
322			i,
323			(unsigned long long)le64_to_cpu(u0->a),
324			(unsigned long long)le64_to_cpu(u0->b),
325			(unsigned long long)buffer_info->dma,
326			buffer_info->length, buffer_info->next_to_watch,
327			(unsigned long long)buffer_info->time_stamp,
328			buffer_info->skb, next_desc);
329
330		if (netif_msg_pktdata(adapter) && buffer_info->skb)
331			print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
332				       16, 1, buffer_info->skb->data,
333				       buffer_info->skb->len, true);
334	}
335
336	/* Print Rx Ring Summary */
337rx_ring_summary:
338	dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
339	pr_info("Queue [NTU] [NTC]\n");
340	pr_info(" %5d %5X %5X\n",
341		0, rx_ring->next_to_use, rx_ring->next_to_clean);
342
343	/* Print Rx Ring */
344	if (!netif_msg_rx_status(adapter))
345		return;
346
347	dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
348	switch (adapter->rx_ps_pages) {
349	case 1:
350	case 2:
351	case 3:
352		/* [Extended] Packet Split Receive Descriptor Format
353		 *
354		 *    +-----------------------------------------------------+
355		 *  0 |                Buffer Address 0 [63:0]              |
356		 *    +-----------------------------------------------------+
357		 *  8 |                Buffer Address 1 [63:0]              |
358		 *    +-----------------------------------------------------+
359		 * 16 |                Buffer Address 2 [63:0]              |
360		 *    +-----------------------------------------------------+
361		 * 24 |                Buffer Address 3 [63:0]              |
362		 *    +-----------------------------------------------------+
363		 */
364		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");
365		/* [Extended] Receive Descriptor (Write-Back) Format
366		 *
367		 *   63       48 47    32 31     13 12    8 7    4 3        0
368		 *   +------------------------------------------------------+
369		 * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
370		 *   | Checksum | Ident  |         | Queue |      |  Type   |
371		 *   +------------------------------------------------------+
372		 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
373		 *   +------------------------------------------------------+
374		 *   63       48 47    32 31            20 19               0
375		 */
376		pr_info("RWB[desc]      [ck ipid mrqhsh] [vl   l0 ee  es] [ l3  l2  l1 hs] [reserved      ] ---------------- [bi->skb] <-- Ext Rx Write-Back format\n");
377		for (i = 0; i < rx_ring->count; i++) {
378			const char *next_desc;
379			buffer_info = &rx_ring->buffer_info[i];
380			rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
381			u1 = (struct my_u1 *)rx_desc_ps;
382			staterr =
383			    le32_to_cpu(rx_desc_ps->wb.middle.status_error);
384
385			if (i == rx_ring->next_to_use)
386				next_desc = " NTU";
387			else if (i == rx_ring->next_to_clean)
388				next_desc = " NTC";
389			else
390				next_desc = "";
391
392			if (staterr & E1000_RXD_STAT_DD) {
393				/* Descriptor Done */
394				pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX ---------------- %p%s\n",
395					"RWB", i,
396					(unsigned long long)le64_to_cpu(u1->a),
397					(unsigned long long)le64_to_cpu(u1->b),
398					(unsigned long long)le64_to_cpu(u1->c),
399					(unsigned long long)le64_to_cpu(u1->d),
400					buffer_info->skb, next_desc);
401			} else {
402				pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX %016llX %p%s\n",
403					"R  ", i,
404					(unsigned long long)le64_to_cpu(u1->a),
405					(unsigned long long)le64_to_cpu(u1->b),
406					(unsigned long long)le64_to_cpu(u1->c),
407					(unsigned long long)le64_to_cpu(u1->d),
408					(unsigned long long)buffer_info->dma,
409					buffer_info->skb, next_desc);
410
411				if (netif_msg_pktdata(adapter))
412					e1000e_dump_ps_pages(adapter,
413							     buffer_info);
414			}
415		}
416		break;
417	default:
418	case 0:
419		/* Extended Receive Descriptor (Read) Format
420		 *
421		 *   +-----------------------------------------------------+
422		 * 0 |                Buffer Address [63:0]                |
423		 *   +-----------------------------------------------------+
424		 * 8 |                      Reserved                       |
425		 *   +-----------------------------------------------------+
426		 */
427		pr_info("R  [desc]      [buf addr 63:0 ] [reserved 63:0 ] [bi->dma       ] [bi->skb] <-- Ext (Read) format\n");
428		/* Extended Receive Descriptor (Write-Back) Format
429		 *
430		 *   63       48 47    32 31    24 23            4 3        0
431		 *   +------------------------------------------------------+
432		 *   |     RSS Hash      |        |               |         |
433		 * 0 +-------------------+  Rsvd  |   Reserved    | MRQ RSS |
434		 *   | Packet   | IP     |        |               |  Type   |
435		 *   | Checksum | Ident  |        |               |         |
436		 *   +------------------------------------------------------+
437		 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
438		 *   +------------------------------------------------------+
439		 *   63       48 47    32 31            20 19               0
440		 */
441		pr_info("RWB[desc]      [cs ipid    mrq] [vt   ln xe  xs] [bi->skb] <-- Ext (Write-Back) format\n");
442
443		for (i = 0; i < rx_ring->count; i++) {
444			const char *next_desc;
445
446			buffer_info = &rx_ring->buffer_info[i];
447			rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
448			u1 = (struct my_u1 *)rx_desc;
449			staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
450
451			if (i == rx_ring->next_to_use)
452				next_desc = " NTU";
453			else if (i == rx_ring->next_to_clean)
454				next_desc = " NTC";
455			else
456				next_desc = "";
457
458			if (staterr & E1000_RXD_STAT_DD) {
459				/* Descriptor Done */
460				pr_info("%s[0x%03X]     %016llX %016llX ---------------- %p%s\n",
461					"RWB", i,
462					(unsigned long long)le64_to_cpu(u1->a),
463					(unsigned long long)le64_to_cpu(u1->b),
464					buffer_info->skb, next_desc);
465			} else {
466				pr_info("%s[0x%03X]     %016llX %016llX %016llX %p%s\n",
467					"R  ", i,
468					(unsigned long long)le64_to_cpu(u1->a),
469					(unsigned long long)le64_to_cpu(u1->b),
470					(unsigned long long)buffer_info->dma,
471					buffer_info->skb, next_desc);
472
473				if (netif_msg_pktdata(adapter) &&
474				    buffer_info->skb)
475					print_hex_dump(KERN_INFO, "",
476						       DUMP_PREFIX_ADDRESS, 16,
477						       1,
478						       buffer_info->skb->data,
479						       adapter->rx_buffer_len,
480						       true);
481			}
482		}
483	}
484}
485
486/**
487 * e1000_desc_unused - calculate if we have unused descriptors
488 **/
489static int e1000_desc_unused(struct e1000_ring *ring)
490{
491	if (ring->next_to_clean > ring->next_to_use)
492		return ring->next_to_clean - ring->next_to_use - 1;
493
494	return ring->count + ring->next_to_clean - ring->next_to_use - 1;
495}
496
497/**
498 * e1000e_systim_to_hwtstamp - convert system time value to hw time stamp
499 * @adapter: board private structure
500 * @hwtstamps: time stamp structure to update
501 * @systim: unsigned 64bit system time value.
502 *
503 * Convert the system time value stored in the RX/TXSTMP registers into a
504 * hwtstamp which can be used by the upper level time stamping functions.
505 *
506 * The 'systim_lock' spinlock is used to protect the consistency of the
507 * system time value. This is needed because reading the 64 bit time
508 * value involves reading two 32 bit registers. The first read latches the
509 * value.
510 **/
511static void e1000e_systim_to_hwtstamp(struct e1000_adapter *adapter,
512				      struct skb_shared_hwtstamps *hwtstamps,
513				      u64 systim)
514{
515	u64 ns;
516	unsigned long flags;
517
518	spin_lock_irqsave(&adapter->systim_lock, flags);
519	ns = timecounter_cyc2time(&adapter->tc, systim);
520	spin_unlock_irqrestore(&adapter->systim_lock, flags);
521
522	memset(hwtstamps, 0, sizeof(*hwtstamps));
523	hwtstamps->hwtstamp = ns_to_ktime(ns);
524}
525
526/**
527 * e1000e_rx_hwtstamp - utility function which checks for Rx time stamp
528 * @adapter: board private structure
529 * @status: descriptor extended error and status field
530 * @skb: particular skb to include time stamp
531 *
532 * If the time stamp is valid, convert it into the timecounter ns value
533 * and store that result into the shhwtstamps structure which is passed
534 * up the network stack.
535 **/
536static void e1000e_rx_hwtstamp(struct e1000_adapter *adapter, u32 status,
537			       struct sk_buff *skb)
538{
539	struct e1000_hw *hw = &adapter->hw;
540	u64 rxstmp;
541
542	if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP) ||
543	    !(status & E1000_RXDEXT_STATERR_TST) ||
544	    !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID))
545		return;
546
547	/* The Rx time stamp registers contain the time stamp.  No other
548	 * received packet will be time stamped until the Rx time stamp
549	 * registers are read.  Because only one packet can be time stamped
550	 * at a time, the register values must belong to this packet and
551	 * therefore none of the other additional attributes need to be
552	 * compared.
553	 */
554	rxstmp = (u64)er32(RXSTMPL);
555	rxstmp |= (u64)er32(RXSTMPH) << 32;
556	e1000e_systim_to_hwtstamp(adapter, skb_hwtstamps(skb), rxstmp);
557
558	adapter->flags2 &= ~FLAG2_CHECK_RX_HWTSTAMP;
559}
560
561/**
562 * e1000_receive_skb - helper function to handle Rx indications
563 * @adapter: board private structure
564 * @staterr: descriptor extended error and status field as written by hardware
565 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
566 * @skb: pointer to sk_buff to be indicated to stack
567 **/
568static void e1000_receive_skb(struct e1000_adapter *adapter,
569			      struct net_device *netdev, struct sk_buff *skb,
570			      u32 staterr, __le16 vlan)
571{
572	u16 tag = le16_to_cpu(vlan);
573
574	e1000e_rx_hwtstamp(adapter, staterr, skb);
575
576	skb->protocol = eth_type_trans(skb, netdev);
577
578	if (staterr & E1000_RXD_STAT_VP)
579		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tag);
580
581	napi_gro_receive(&adapter->napi, skb);
582}
583
584/**
585 * e1000_rx_checksum - Receive Checksum Offload
586 * @adapter: board private structure
587 * @status_err: receive descriptor status and error fields
588 * @csum: receive descriptor csum field
589 * @sk_buff: socket buffer with received data
590 **/
591static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
592			      struct sk_buff *skb)
593{
594	u16 status = (u16)status_err;
595	u8 errors = (u8)(status_err >> 24);
596
597	skb_checksum_none_assert(skb);
598
599	/* Rx checksum disabled */
600	if (!(adapter->netdev->features & NETIF_F_RXCSUM))
601		return;
602
603	/* Ignore Checksum bit is set */
604	if (status & E1000_RXD_STAT_IXSM)
605		return;
606
607	/* TCP/UDP checksum error bit or IP checksum error bit is set */
608	if (errors & (E1000_RXD_ERR_TCPE | E1000_RXD_ERR_IPE)) {
609		/* let the stack verify checksum errors */
610		adapter->hw_csum_err++;
611		return;
612	}
613
614	/* TCP/UDP Checksum has not been calculated */
615	if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
616		return;
617
618	/* It must be a TCP or UDP packet with a valid checksum */
619	skb->ip_summed = CHECKSUM_UNNECESSARY;
620	adapter->hw_csum_good++;
621}
622
623static void e1000e_update_rdt_wa(struct e1000_ring *rx_ring, unsigned int i)
624{
625	struct e1000_adapter *adapter = rx_ring->adapter;
626	struct e1000_hw *hw = &adapter->hw;
627	s32 ret_val = __ew32_prepare(hw);
628
629	writel(i, rx_ring->tail);
630
631	if (unlikely(!ret_val && (i != readl(rx_ring->tail)))) {
632		u32 rctl = er32(RCTL);
633
634		ew32(RCTL, rctl & ~E1000_RCTL_EN);
635		e_err("ME firmware caused invalid RDT - resetting\n");
636		schedule_work(&adapter->reset_task);
637	}
638}
639
640static void e1000e_update_tdt_wa(struct e1000_ring *tx_ring, unsigned int i)
641{
642	struct e1000_adapter *adapter = tx_ring->adapter;
643	struct e1000_hw *hw = &adapter->hw;
644	s32 ret_val = __ew32_prepare(hw);
645
646	writel(i, tx_ring->tail);
647
648	if (unlikely(!ret_val && (i != readl(tx_ring->tail)))) {
649		u32 tctl = er32(TCTL);
650
651		ew32(TCTL, tctl & ~E1000_TCTL_EN);
652		e_err("ME firmware caused invalid TDT - resetting\n");
653		schedule_work(&adapter->reset_task);
654	}
655}
656
657/**
658 * e1000_alloc_rx_buffers - Replace used receive buffers
659 * @rx_ring: Rx descriptor ring
660 **/
661static void e1000_alloc_rx_buffers(struct e1000_ring *rx_ring,
662				   int cleaned_count, gfp_t gfp)
663{
664	struct e1000_adapter *adapter = rx_ring->adapter;
665	struct net_device *netdev = adapter->netdev;
666	struct pci_dev *pdev = adapter->pdev;
667	union e1000_rx_desc_extended *rx_desc;
668	struct e1000_buffer *buffer_info;
669	struct sk_buff *skb;
670	unsigned int i;
671	unsigned int bufsz = adapter->rx_buffer_len;
672
673	i = rx_ring->next_to_use;
674	buffer_info = &rx_ring->buffer_info[i];
675
676	while (cleaned_count--) {
677		skb = buffer_info->skb;
678		if (skb) {
679			skb_trim(skb, 0);
680			goto map_skb;
681		}
682
683		skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
684		if (!skb) {
685			/* Better luck next round */
686			adapter->alloc_rx_buff_failed++;
687			break;
688		}
689
690		buffer_info->skb = skb;
691map_skb:
692		buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
693						  adapter->rx_buffer_len,
694						  DMA_FROM_DEVICE);
695		if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
696			dev_err(&pdev->dev, "Rx DMA map failed\n");
697			adapter->rx_dma_failed++;
698			break;
699		}
700
701		rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
702		rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
703
704		if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
705			/* Force memory writes to complete before letting h/w
706			 * know there are new descriptors to fetch.  (Only
707			 * applicable for weak-ordered memory model archs,
708			 * such as IA-64).
709			 */
710			wmb();
711			if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
712				e1000e_update_rdt_wa(rx_ring, i);
713			else
714				writel(i, rx_ring->tail);
715		}
716		i++;
717		if (i == rx_ring->count)
718			i = 0;
719		buffer_info = &rx_ring->buffer_info[i];
720	}
721
722	rx_ring->next_to_use = i;
723}
724
725/**
726 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
727 * @rx_ring: Rx descriptor ring
728 **/
729static void e1000_alloc_rx_buffers_ps(struct e1000_ring *rx_ring,
730				      int cleaned_count, gfp_t gfp)
731{
732	struct e1000_adapter *adapter = rx_ring->adapter;
733	struct net_device *netdev = adapter->netdev;
734	struct pci_dev *pdev = adapter->pdev;
735	union e1000_rx_desc_packet_split *rx_desc;
736	struct e1000_buffer *buffer_info;
737	struct e1000_ps_page *ps_page;
738	struct sk_buff *skb;
739	unsigned int i, j;
740
741	i = rx_ring->next_to_use;
742	buffer_info = &rx_ring->buffer_info[i];
743
744	while (cleaned_count--) {
745		rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
746
747		for (j = 0; j < PS_PAGE_BUFFERS; j++) {
748			ps_page = &buffer_info->ps_pages[j];
749			if (j >= adapter->rx_ps_pages) {
750				/* all unused desc entries get hw null ptr */
751				rx_desc->read.buffer_addr[j + 1] =
752				    ~cpu_to_le64(0);
753				continue;
754			}
755			if (!ps_page->page) {
756				ps_page->page = alloc_page(gfp);
757				if (!ps_page->page) {
758					adapter->alloc_rx_buff_failed++;
759					goto no_buffers;
760				}
761				ps_page->dma = dma_map_page(&pdev->dev,
762							    ps_page->page,
763							    0, PAGE_SIZE,
764							    DMA_FROM_DEVICE);
765				if (dma_mapping_error(&pdev->dev,
766						      ps_page->dma)) {
767					dev_err(&adapter->pdev->dev,
768						"Rx DMA page map failed\n");
769					adapter->rx_dma_failed++;
770					goto no_buffers;
771				}
772			}
773			/* Refresh the desc even if buffer_addrs
774			 * didn't change because each write-back
775			 * erases this info.
776			 */
777			rx_desc->read.buffer_addr[j + 1] =
778			    cpu_to_le64(ps_page->dma);
779		}
780
781		skb = __netdev_alloc_skb_ip_align(netdev, adapter->rx_ps_bsize0,
782						  gfp);
783
784		if (!skb) {
785			adapter->alloc_rx_buff_failed++;
786			break;
787		}
788
789		buffer_info->skb = skb;
790		buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
791						  adapter->rx_ps_bsize0,
792						  DMA_FROM_DEVICE);
793		if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
794			dev_err(&pdev->dev, "Rx DMA map failed\n");
795			adapter->rx_dma_failed++;
796			/* cleanup skb */
797			dev_kfree_skb_any(skb);
798			buffer_info->skb = NULL;
799			break;
800		}
801
802		rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
803
804		if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
805			/* Force memory writes to complete before letting h/w
806			 * know there are new descriptors to fetch.  (Only
807			 * applicable for weak-ordered memory model archs,
808			 * such as IA-64).
809			 */
810			wmb();
811			if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
812				e1000e_update_rdt_wa(rx_ring, i << 1);
813			else
814				writel(i << 1, rx_ring->tail);
815		}
816
817		i++;
818		if (i == rx_ring->count)
819			i = 0;
820		buffer_info = &rx_ring->buffer_info[i];
821	}
822
823no_buffers:
824	rx_ring->next_to_use = i;
825}
826
827/**
828 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
829 * @rx_ring: Rx descriptor ring
830 * @cleaned_count: number of buffers to allocate this pass
831 **/
832
833static void e1000_alloc_jumbo_rx_buffers(struct e1000_ring *rx_ring,
834					 int cleaned_count, gfp_t gfp)
835{
836	struct e1000_adapter *adapter = rx_ring->adapter;
837	struct net_device *netdev = adapter->netdev;
838	struct pci_dev *pdev = adapter->pdev;
839	union e1000_rx_desc_extended *rx_desc;
840	struct e1000_buffer *buffer_info;
841	struct sk_buff *skb;
842	unsigned int i;
843	unsigned int bufsz = 256 - 16;	/* for skb_reserve */
844
845	i = rx_ring->next_to_use;
846	buffer_info = &rx_ring->buffer_info[i];
847
848	while (cleaned_count--) {
849		skb = buffer_info->skb;
850		if (skb) {
851			skb_trim(skb, 0);
852			goto check_page;
853		}
854
855		skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
856		if (unlikely(!skb)) {
857			/* Better luck next round */
858			adapter->alloc_rx_buff_failed++;
859			break;
860		}
861
862		buffer_info->skb = skb;
863check_page:
864		/* allocate a new page if necessary */
865		if (!buffer_info->page) {
866			buffer_info->page = alloc_page(gfp);
867			if (unlikely(!buffer_info->page)) {
868				adapter->alloc_rx_buff_failed++;
869				break;
870			}
871		}
872
873		if (!buffer_info->dma) {
874			buffer_info->dma = dma_map_page(&pdev->dev,
875							buffer_info->page, 0,
876							PAGE_SIZE,
877							DMA_FROM_DEVICE);
878			if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
879				adapter->alloc_rx_buff_failed++;
880				break;
881			}
882		}
883
884		rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
885		rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
886
887		if (unlikely(++i == rx_ring->count))
888			i = 0;
889		buffer_info = &rx_ring->buffer_info[i];
890	}
891
892	if (likely(rx_ring->next_to_use != i)) {
893		rx_ring->next_to_use = i;
894		if (unlikely(i-- == 0))
895			i = (rx_ring->count - 1);
896
897		/* Force memory writes to complete before letting h/w
898		 * know there are new descriptors to fetch.  (Only
899		 * applicable for weak-ordered memory model archs,
900		 * such as IA-64).
901		 */
902		wmb();
903		if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
904			e1000e_update_rdt_wa(rx_ring, i);
905		else
906			writel(i, rx_ring->tail);
907	}
908}
909
910static inline void e1000_rx_hash(struct net_device *netdev, __le32 rss,
911				 struct sk_buff *skb)
912{
913	if (netdev->features & NETIF_F_RXHASH)
914		skb_set_hash(skb, le32_to_cpu(rss), PKT_HASH_TYPE_L3);
915}
916
917/**
918 * e1000_clean_rx_irq - Send received data up the network stack
919 * @rx_ring: Rx descriptor ring
920 *
921 * the return value indicates whether actual cleaning was done, there
922 * is no guarantee that everything was cleaned
923 **/
924static bool e1000_clean_rx_irq(struct e1000_ring *rx_ring, int *work_done,
925			       int work_to_do)
926{
927	struct e1000_adapter *adapter = rx_ring->adapter;
928	struct net_device *netdev = adapter->netdev;
929	struct pci_dev *pdev = adapter->pdev;
930	struct e1000_hw *hw = &adapter->hw;
931	union e1000_rx_desc_extended *rx_desc, *next_rxd;
932	struct e1000_buffer *buffer_info, *next_buffer;
933	u32 length, staterr;
934	unsigned int i;
935	int cleaned_count = 0;
936	bool cleaned = false;
937	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
938
939	i = rx_ring->next_to_clean;
940	rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
941	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
942	buffer_info = &rx_ring->buffer_info[i];
943
944	while (staterr & E1000_RXD_STAT_DD) {
945		struct sk_buff *skb;
946
947		if (*work_done >= work_to_do)
948			break;
949		(*work_done)++;
950		dma_rmb();	/* read descriptor and rx_buffer_info after status DD */
951
952		skb = buffer_info->skb;
953		buffer_info->skb = NULL;
954
955		prefetch(skb->data - NET_IP_ALIGN);
956
957		i++;
958		if (i == rx_ring->count)
959			i = 0;
960		next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
961		prefetch(next_rxd);
962
963		next_buffer = &rx_ring->buffer_info[i];
964
965		cleaned = true;
966		cleaned_count++;
967		dma_unmap_single(&pdev->dev, buffer_info->dma,
968				 adapter->rx_buffer_len, DMA_FROM_DEVICE);
969		buffer_info->dma = 0;
970
971		length = le16_to_cpu(rx_desc->wb.upper.length);
972
973		/* !EOP means multiple descriptors were used to store a single
974		 * packet, if that's the case we need to toss it.  In fact, we
975		 * need to toss every packet with the EOP bit clear and the
976		 * next frame that _does_ have the EOP bit set, as it is by
977		 * definition only a frame fragment
978		 */
979		if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
980			adapter->flags2 |= FLAG2_IS_DISCARDING;
981
982		if (adapter->flags2 & FLAG2_IS_DISCARDING) {
983			/* All receives must fit into a single buffer */
984			e_dbg("Receive packet consumed multiple buffers\n");
985			/* recycle */
986			buffer_info->skb = skb;
987			if (staterr & E1000_RXD_STAT_EOP)
988				adapter->flags2 &= ~FLAG2_IS_DISCARDING;
989			goto next_desc;
990		}
991
992		if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
993			     !(netdev->features & NETIF_F_RXALL))) {
994			/* recycle */
995			buffer_info->skb = skb;
996			goto next_desc;
997		}
998
999		/* adjust length to remove Ethernet CRC */
1000		if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1001			/* If configured to store CRC, don't subtract FCS,
1002			 * but keep the FCS bytes out of the total_rx_bytes
1003			 * counter
1004			 */
1005			if (netdev->features & NETIF_F_RXFCS)
1006				total_rx_bytes -= 4;
1007			else
1008				length -= 4;
1009		}
1010
1011		total_rx_bytes += length;
1012		total_rx_packets++;
1013
1014		/* code added for copybreak, this should improve
1015		 * performance for small packets with large amounts
1016		 * of reassembly being done in the stack
1017		 */
1018		if (length < copybreak) {
1019			struct sk_buff *new_skb =
1020				napi_alloc_skb(&adapter->napi, length);
1021			if (new_skb) {
1022				skb_copy_to_linear_data_offset(new_skb,
1023							       -NET_IP_ALIGN,
1024							       (skb->data -
1025								NET_IP_ALIGN),
1026							       (length +
1027								NET_IP_ALIGN));
1028				/* save the skb in buffer_info as good */
1029				buffer_info->skb = skb;
1030				skb = new_skb;
1031			}
1032			/* else just continue with the old one */
1033		}
1034		/* end copybreak code */
1035		skb_put(skb, length);
1036
1037		/* Receive Checksum Offload */
1038		e1000_rx_checksum(adapter, staterr, skb);
1039
1040		e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1041
1042		e1000_receive_skb(adapter, netdev, skb, staterr,
1043				  rx_desc->wb.upper.vlan);
1044
1045next_desc:
1046		rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1047
1048		/* return some buffers to hardware, one at a time is too slow */
1049		if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1050			adapter->alloc_rx_buf(rx_ring, cleaned_count,
1051					      GFP_ATOMIC);
1052			cleaned_count = 0;
1053		}
1054
1055		/* use prefetched values */
1056		rx_desc = next_rxd;
1057		buffer_info = next_buffer;
1058
1059		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1060	}
1061	rx_ring->next_to_clean = i;
1062
1063	cleaned_count = e1000_desc_unused(rx_ring);
1064	if (cleaned_count)
1065		adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1066
1067	adapter->total_rx_bytes += total_rx_bytes;
1068	adapter->total_rx_packets += total_rx_packets;
1069	return cleaned;
1070}
1071
1072static void e1000_put_txbuf(struct e1000_ring *tx_ring,
1073			    struct e1000_buffer *buffer_info)
1074{
1075	struct e1000_adapter *adapter = tx_ring->adapter;
1076
1077	if (buffer_info->dma) {
1078		if (buffer_info->mapped_as_page)
1079			dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1080				       buffer_info->length, DMA_TO_DEVICE);
1081		else
1082			dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1083					 buffer_info->length, DMA_TO_DEVICE);
1084		buffer_info->dma = 0;
1085	}
1086	if (buffer_info->skb) {
1087		dev_kfree_skb_any(buffer_info->skb);
1088		buffer_info->skb = NULL;
1089	}
1090	buffer_info->time_stamp = 0;
1091}
1092
1093static void e1000_print_hw_hang(struct work_struct *work)
1094{
1095	struct e1000_adapter *adapter = container_of(work,
1096						     struct e1000_adapter,
1097						     print_hang_task);
1098	struct net_device *netdev = adapter->netdev;
1099	struct e1000_ring *tx_ring = adapter->tx_ring;
1100	unsigned int i = tx_ring->next_to_clean;
1101	unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1102	struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1103	struct e1000_hw *hw = &adapter->hw;
1104	u16 phy_status, phy_1000t_status, phy_ext_status;
1105	u16 pci_status;
1106
1107	if (test_bit(__E1000_DOWN, &adapter->state))
1108		return;
1109
1110	if (!adapter->tx_hang_recheck && (adapter->flags2 & FLAG2_DMA_BURST)) {
1111		/* May be block on write-back, flush and detect again
1112		 * flush pending descriptor writebacks to memory
1113		 */
1114		ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1115		/* execute the writes immediately */
1116		e1e_flush();
1117		/* Due to rare timing issues, write to TIDV again to ensure
1118		 * the write is successful
1119		 */
1120		ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1121		/* execute the writes immediately */
1122		e1e_flush();
1123		adapter->tx_hang_recheck = true;
1124		return;
1125	}
1126	adapter->tx_hang_recheck = false;
1127
1128	if (er32(TDH(0)) == er32(TDT(0))) {
1129		e_dbg("false hang detected, ignoring\n");
1130		return;
1131	}
1132
1133	/* Real hang detected */
1134	netif_stop_queue(netdev);
1135
1136	e1e_rphy(hw, MII_BMSR, &phy_status);
1137	e1e_rphy(hw, MII_STAT1000, &phy_1000t_status);
1138	e1e_rphy(hw, MII_ESTATUS, &phy_ext_status);
1139
1140	pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1141
1142	/* detected Hardware unit hang */
1143	e_err("Detected Hardware Unit Hang:\n"
1144	      "  TDH                  <%x>\n"
1145	      "  TDT                  <%x>\n"
1146	      "  next_to_use          <%x>\n"
1147	      "  next_to_clean        <%x>\n"
1148	      "buffer_info[next_to_clean]:\n"
1149	      "  time_stamp           <%lx>\n"
1150	      "  next_to_watch        <%x>\n"
1151	      "  jiffies              <%lx>\n"
1152	      "  next_to_watch.status <%x>\n"
1153	      "MAC Status             <%x>\n"
1154	      "PHY Status             <%x>\n"
1155	      "PHY 1000BASE-T Status  <%x>\n"
1156	      "PHY Extended Status    <%x>\n"
1157	      "PCI Status             <%x>\n",
1158	      readl(tx_ring->head), readl(tx_ring->tail), tx_ring->next_to_use,
1159	      tx_ring->next_to_clean, tx_ring->buffer_info[eop].time_stamp,
1160	      eop, jiffies, eop_desc->upper.fields.status, er32(STATUS),
1161	      phy_status, phy_1000t_status, phy_ext_status, pci_status);
1162
1163	e1000e_dump(adapter);
1164
1165	/* Suggest workaround for known h/w issue */
1166	if ((hw->mac.type == e1000_pchlan) && (er32(CTRL) & E1000_CTRL_TFCE))
1167		e_err("Try turning off Tx pause (flow control) via ethtool\n");
1168}
1169
1170/**
1171 * e1000e_tx_hwtstamp_work - check for Tx time stamp
1172 * @work: pointer to work struct
1173 *
1174 * This work function polls the TSYNCTXCTL valid bit to determine when a
1175 * timestamp has been taken for the current stored skb.  The timestamp must
1176 * be for this skb because only one such packet is allowed in the queue.
1177 */
1178static void e1000e_tx_hwtstamp_work(struct work_struct *work)
1179{
1180	struct e1000_adapter *adapter = container_of(work, struct e1000_adapter,
1181						     tx_hwtstamp_work);
1182	struct e1000_hw *hw = &adapter->hw;
1183
1184	if (er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID) {
1185		struct skb_shared_hwtstamps shhwtstamps;
1186		u64 txstmp;
1187
1188		txstmp = er32(TXSTMPL);
1189		txstmp |= (u64)er32(TXSTMPH) << 32;
1190
1191		e1000e_systim_to_hwtstamp(adapter, &shhwtstamps, txstmp);
1192
1193		skb_tstamp_tx(adapter->tx_hwtstamp_skb, &shhwtstamps);
1194		dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
1195		adapter->tx_hwtstamp_skb = NULL;
1196	} else if (time_after(jiffies, adapter->tx_hwtstamp_start
1197			      + adapter->tx_timeout_factor * HZ)) {
1198		dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
1199		adapter->tx_hwtstamp_skb = NULL;
1200		adapter->tx_hwtstamp_timeouts++;
1201		e_warn("clearing Tx timestamp hang\n");
1202	} else {
1203		/* reschedule to check later */
1204		schedule_work(&adapter->tx_hwtstamp_work);
1205	}
1206}
1207
1208/**
1209 * e1000_clean_tx_irq - Reclaim resources after transmit completes
1210 * @tx_ring: Tx descriptor ring
1211 *
1212 * the return value indicates whether actual cleaning was done, there
1213 * is no guarantee that everything was cleaned
1214 **/
1215static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
1216{
1217	struct e1000_adapter *adapter = tx_ring->adapter;
1218	struct net_device *netdev = adapter->netdev;
1219	struct e1000_hw *hw = &adapter->hw;
1220	struct e1000_tx_desc *tx_desc, *eop_desc;
1221	struct e1000_buffer *buffer_info;
1222	unsigned int i, eop;
1223	unsigned int count = 0;
1224	unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1225	unsigned int bytes_compl = 0, pkts_compl = 0;
1226
1227	i = tx_ring->next_to_clean;
1228	eop = tx_ring->buffer_info[i].next_to_watch;
1229	eop_desc = E1000_TX_DESC(*tx_ring, eop);
1230
1231	while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1232	       (count < tx_ring->count)) {
1233		bool cleaned = false;
1234
1235		dma_rmb();		/* read buffer_info after eop_desc */
1236		for (; !cleaned; count++) {
1237			tx_desc = E1000_TX_DESC(*tx_ring, i);
1238			buffer_info = &tx_ring->buffer_info[i];
1239			cleaned = (i == eop);
1240
1241			if (cleaned) {
1242				total_tx_packets += buffer_info->segs;
1243				total_tx_bytes += buffer_info->bytecount;
1244				if (buffer_info->skb) {
1245					bytes_compl += buffer_info->skb->len;
1246					pkts_compl++;
1247				}
1248			}
1249
1250			e1000_put_txbuf(tx_ring, buffer_info);
1251			tx_desc->upper.data = 0;
1252
1253			i++;
1254			if (i == tx_ring->count)
1255				i = 0;
1256		}
1257
1258		if (i == tx_ring->next_to_use)
1259			break;
1260		eop = tx_ring->buffer_info[i].next_to_watch;
1261		eop_desc = E1000_TX_DESC(*tx_ring, eop);
1262	}
1263
1264	tx_ring->next_to_clean = i;
1265
1266	netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1267
1268#define TX_WAKE_THRESHOLD 32
1269	if (count && netif_carrier_ok(netdev) &&
1270	    e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1271		/* Make sure that anybody stopping the queue after this
1272		 * sees the new next_to_clean.
1273		 */
1274		smp_mb();
1275
1276		if (netif_queue_stopped(netdev) &&
1277		    !(test_bit(__E1000_DOWN, &adapter->state))) {
1278			netif_wake_queue(netdev);
1279			++adapter->restart_queue;
1280		}
1281	}
1282
1283	if (adapter->detect_tx_hung) {
1284		/* Detect a transmit hang in hardware, this serializes the
1285		 * check with the clearing of time_stamp and movement of i
1286		 */
1287		adapter->detect_tx_hung = false;
1288		if (tx_ring->buffer_info[i].time_stamp &&
1289		    time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1290			       + (adapter->tx_timeout_factor * HZ)) &&
1291		    !(er32(STATUS) & E1000_STATUS_TXOFF))
1292			schedule_work(&adapter->print_hang_task);
1293		else
1294			adapter->tx_hang_recheck = false;
1295	}
1296	adapter->total_tx_bytes += total_tx_bytes;
1297	adapter->total_tx_packets += total_tx_packets;
1298	return count < tx_ring->count;
1299}
1300
1301/**
1302 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1303 * @rx_ring: Rx descriptor ring
1304 *
1305 * the return value indicates whether actual cleaning was done, there
1306 * is no guarantee that everything was cleaned
1307 **/
1308static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
1309				  int work_to_do)
1310{
1311	struct e1000_adapter *adapter = rx_ring->adapter;
1312	struct e1000_hw *hw = &adapter->hw;
1313	union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1314	struct net_device *netdev = adapter->netdev;
1315	struct pci_dev *pdev = adapter->pdev;
1316	struct e1000_buffer *buffer_info, *next_buffer;
1317	struct e1000_ps_page *ps_page;
1318	struct sk_buff *skb;
1319	unsigned int i, j;
1320	u32 length, staterr;
1321	int cleaned_count = 0;
1322	bool cleaned = false;
1323	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1324
1325	i = rx_ring->next_to_clean;
1326	rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1327	staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1328	buffer_info = &rx_ring->buffer_info[i];
1329
1330	while (staterr & E1000_RXD_STAT_DD) {
1331		if (*work_done >= work_to_do)
1332			break;
1333		(*work_done)++;
1334		skb = buffer_info->skb;
1335		dma_rmb();	/* read descriptor and rx_buffer_info after status DD */
1336
1337		/* in the packet split case this is header only */
1338		prefetch(skb->data - NET_IP_ALIGN);
1339
1340		i++;
1341		if (i == rx_ring->count)
1342			i = 0;
1343		next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1344		prefetch(next_rxd);
1345
1346		next_buffer = &rx_ring->buffer_info[i];
1347
1348		cleaned = true;
1349		cleaned_count++;
1350		dma_unmap_single(&pdev->dev, buffer_info->dma,
1351				 adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1352		buffer_info->dma = 0;
1353
1354		/* see !EOP comment in other Rx routine */
1355		if (!(staterr & E1000_RXD_STAT_EOP))
1356			adapter->flags2 |= FLAG2_IS_DISCARDING;
1357
1358		if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1359			e_dbg("Packet Split buffers didn't pick up the full packet\n");
1360			dev_kfree_skb_irq(skb);
1361			if (staterr & E1000_RXD_STAT_EOP)
1362				adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1363			goto next_desc;
1364		}
1365
1366		if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1367			     !(netdev->features & NETIF_F_RXALL))) {
1368			dev_kfree_skb_irq(skb);
1369			goto next_desc;
1370		}
1371
1372		length = le16_to_cpu(rx_desc->wb.middle.length0);
1373
1374		if (!length) {
1375			e_dbg("Last part of the packet spanning multiple descriptors\n");
1376			dev_kfree_skb_irq(skb);
1377			goto next_desc;
1378		}
1379
1380		/* Good Receive */
1381		skb_put(skb, length);
1382
1383		{
1384			/* this looks ugly, but it seems compiler issues make
1385			 * it more efficient than reusing j
1386			 */
1387			int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1388
1389			/* page alloc/put takes too long and effects small
1390			 * packet throughput, so unsplit small packets and
1391			 * save the alloc/put only valid in softirq (napi)
1392			 * context to call kmap_*
1393			 */
1394			if (l1 && (l1 <= copybreak) &&
1395			    ((length + l1) <= adapter->rx_ps_bsize0)) {
1396				u8 *vaddr;
1397
1398				ps_page = &buffer_info->ps_pages[0];
1399
1400				/* there is no documentation about how to call
1401				 * kmap_atomic, so we can't hold the mapping
1402				 * very long
1403				 */
1404				dma_sync_single_for_cpu(&pdev->dev,
1405							ps_page->dma,
1406							PAGE_SIZE,
1407							DMA_FROM_DEVICE);
1408				vaddr = kmap_atomic(ps_page->page);
1409				memcpy(skb_tail_pointer(skb), vaddr, l1);
1410				kunmap_atomic(vaddr);
1411				dma_sync_single_for_device(&pdev->dev,
1412							   ps_page->dma,
1413							   PAGE_SIZE,
1414							   DMA_FROM_DEVICE);
1415
1416				/* remove the CRC */
1417				if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1418					if (!(netdev->features & NETIF_F_RXFCS))
1419						l1 -= 4;
1420				}
1421
1422				skb_put(skb, l1);
1423				goto copydone;
1424			}	/* if */
1425		}
1426
1427		for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1428			length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1429			if (!length)
1430				break;
1431
1432			ps_page = &buffer_info->ps_pages[j];
1433			dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1434				       DMA_FROM_DEVICE);
1435			ps_page->dma = 0;
1436			skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1437			ps_page->page = NULL;
1438			skb->len += length;
1439			skb->data_len += length;
1440			skb->truesize += PAGE_SIZE;
1441		}
1442
1443		/* strip the ethernet crc, problem is we're using pages now so
1444		 * this whole operation can get a little cpu intensive
1445		 */
1446		if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1447			if (!(netdev->features & NETIF_F_RXFCS))
1448				pskb_trim(skb, skb->len - 4);
1449		}
1450
1451copydone:
1452		total_rx_bytes += skb->len;
1453		total_rx_packets++;
1454
1455		e1000_rx_checksum(adapter, staterr, skb);
1456
1457		e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1458
1459		if (rx_desc->wb.upper.header_status &
1460		    cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1461			adapter->rx_hdr_split++;
1462
1463		e1000_receive_skb(adapter, netdev, skb, staterr,
1464				  rx_desc->wb.middle.vlan);
1465
1466next_desc:
1467		rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1468		buffer_info->skb = NULL;
1469
1470		/* return some buffers to hardware, one at a time is too slow */
1471		if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1472			adapter->alloc_rx_buf(rx_ring, cleaned_count,
1473					      GFP_ATOMIC);
1474			cleaned_count = 0;
1475		}
1476
1477		/* use prefetched values */
1478		rx_desc = next_rxd;
1479		buffer_info = next_buffer;
1480
1481		staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1482	}
1483	rx_ring->next_to_clean = i;
1484
1485	cleaned_count = e1000_desc_unused(rx_ring);
1486	if (cleaned_count)
1487		adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1488
1489	adapter->total_rx_bytes += total_rx_bytes;
1490	adapter->total_rx_packets += total_rx_packets;
1491	return cleaned;
1492}
1493
1494/**
1495 * e1000_consume_page - helper function
1496 **/
1497static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1498			       u16 length)
1499{
1500	bi->page = NULL;
1501	skb->len += length;
1502	skb->data_len += length;
1503	skb->truesize += PAGE_SIZE;
1504}
1505
1506/**
1507 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1508 * @adapter: board private structure
1509 *
1510 * the return value indicates whether actual cleaning was done, there
1511 * is no guarantee that everything was cleaned
1512 **/
1513static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1514				     int work_to_do)
1515{
1516	struct e1000_adapter *adapter = rx_ring->adapter;
1517	struct net_device *netdev = adapter->netdev;
1518	struct pci_dev *pdev = adapter->pdev;
1519	union e1000_rx_desc_extended *rx_desc, *next_rxd;
1520	struct e1000_buffer *buffer_info, *next_buffer;
1521	u32 length, staterr;
1522	unsigned int i;
1523	int cleaned_count = 0;
1524	bool cleaned = false;
1525	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1526	struct skb_shared_info *shinfo;
1527
1528	i = rx_ring->next_to_clean;
1529	rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1530	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1531	buffer_info = &rx_ring->buffer_info[i];
1532
1533	while (staterr & E1000_RXD_STAT_DD) {
1534		struct sk_buff *skb;
1535
1536		if (*work_done >= work_to_do)
1537			break;
1538		(*work_done)++;
1539		dma_rmb();	/* read descriptor and rx_buffer_info after status DD */
1540
1541		skb = buffer_info->skb;
1542		buffer_info->skb = NULL;
1543
1544		++i;
1545		if (i == rx_ring->count)
1546			i = 0;
1547		next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1548		prefetch(next_rxd);
1549
1550		next_buffer = &rx_ring->buffer_info[i];
1551
1552		cleaned = true;
1553		cleaned_count++;
1554		dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1555			       DMA_FROM_DEVICE);
1556		buffer_info->dma = 0;
1557
1558		length = le16_to_cpu(rx_desc->wb.upper.length);
1559
1560		/* errors is only valid for DD + EOP descriptors */
1561		if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1562			     ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1563			      !(netdev->features & NETIF_F_RXALL)))) {
1564			/* recycle both page and skb */
1565			buffer_info->skb = skb;
1566			/* an error means any chain goes out the window too */
1567			if (rx_ring->rx_skb_top)
1568				dev_kfree_skb_irq(rx_ring->rx_skb_top);
1569			rx_ring->rx_skb_top = NULL;
1570			goto next_desc;
1571		}
1572#define rxtop (rx_ring->rx_skb_top)
1573		if (!(staterr & E1000_RXD_STAT_EOP)) {
1574			/* this descriptor is only the beginning (or middle) */
1575			if (!rxtop) {
1576				/* this is the beginning of a chain */
1577				rxtop = skb;
1578				skb_fill_page_desc(rxtop, 0, buffer_info->page,
1579						   0, length);
1580			} else {
1581				/* this is the middle of a chain */
1582				shinfo = skb_shinfo(rxtop);
1583				skb_fill_page_desc(rxtop, shinfo->nr_frags,
1584						   buffer_info->page, 0,
1585						   length);
1586				/* re-use the skb, only consumed the page */
1587				buffer_info->skb = skb;
1588			}
1589			e1000_consume_page(buffer_info, rxtop, length);
1590			goto next_desc;
1591		} else {
1592			if (rxtop) {
1593				/* end of the chain */
1594				shinfo = skb_shinfo(rxtop);
1595				skb_fill_page_desc(rxtop, shinfo->nr_frags,
1596						   buffer_info->page, 0,
1597						   length);
1598				/* re-use the current skb, we only consumed the
1599				 * page
1600				 */
1601				buffer_info->skb = skb;
1602				skb = rxtop;
1603				rxtop = NULL;
1604				e1000_consume_page(buffer_info, skb, length);
1605			} else {
1606				/* no chain, got EOP, this buf is the packet
1607				 * copybreak to save the put_page/alloc_page
1608				 */
1609				if (length <= copybreak &&
1610				    skb_tailroom(skb) >= length) {
1611					u8 *vaddr;
1612					vaddr = kmap_atomic(buffer_info->page);
1613					memcpy(skb_tail_pointer(skb), vaddr,
1614					       length);
1615					kunmap_atomic(vaddr);
1616					/* re-use the page, so don't erase
1617					 * buffer_info->page
1618					 */
1619					skb_put(skb, length);
1620				} else {
1621					skb_fill_page_desc(skb, 0,
1622							   buffer_info->page, 0,
1623							   length);
1624					e1000_consume_page(buffer_info, skb,
1625							   length);
1626				}
1627			}
1628		}
1629
1630		/* Receive Checksum Offload */
1631		e1000_rx_checksum(adapter, staterr, skb);
1632
1633		e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1634
1635		/* probably a little skewed due to removing CRC */
1636		total_rx_bytes += skb->len;
1637		total_rx_packets++;
1638
1639		/* eth type trans needs skb->data to point to something */
1640		if (!pskb_may_pull(skb, ETH_HLEN)) {
1641			e_err("pskb_may_pull failed.\n");
1642			dev_kfree_skb_irq(skb);
1643			goto next_desc;
1644		}
1645
1646		e1000_receive_skb(adapter, netdev, skb, staterr,
1647				  rx_desc->wb.upper.vlan);
1648
1649next_desc:
1650		rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1651
1652		/* return some buffers to hardware, one at a time is too slow */
1653		if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1654			adapter->alloc_rx_buf(rx_ring, cleaned_count,
1655					      GFP_ATOMIC);
1656			cleaned_count = 0;
1657		}
1658
1659		/* use prefetched values */
1660		rx_desc = next_rxd;
1661		buffer_info = next_buffer;
1662
1663		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1664	}
1665	rx_ring->next_to_clean = i;
1666
1667	cleaned_count = e1000_desc_unused(rx_ring);
1668	if (cleaned_count)
1669		adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1670
1671	adapter->total_rx_bytes += total_rx_bytes;
1672	adapter->total_rx_packets += total_rx_packets;
1673	return cleaned;
1674}
1675
1676/**
1677 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1678 * @rx_ring: Rx descriptor ring
1679 **/
1680static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1681{
1682	struct e1000_adapter *adapter = rx_ring->adapter;
1683	struct e1000_buffer *buffer_info;
1684	struct e1000_ps_page *ps_page;
1685	struct pci_dev *pdev = adapter->pdev;
1686	unsigned int i, j;
1687
1688	/* Free all the Rx ring sk_buffs */
1689	for (i = 0; i < rx_ring->count; i++) {
1690		buffer_info = &rx_ring->buffer_info[i];
1691		if (buffer_info->dma) {
1692			if (adapter->clean_rx == e1000_clean_rx_irq)
1693				dma_unmap_single(&pdev->dev, buffer_info->dma,
1694						 adapter->rx_buffer_len,
1695						 DMA_FROM_DEVICE);
1696			else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1697				dma_unmap_page(&pdev->dev, buffer_info->dma,
1698					       PAGE_SIZE, DMA_FROM_DEVICE);
1699			else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1700				dma_unmap_single(&pdev->dev, buffer_info->dma,
1701						 adapter->rx_ps_bsize0,
1702						 DMA_FROM_DEVICE);
1703			buffer_info->dma = 0;
1704		}
1705
1706		if (buffer_info->page) {
1707			put_page(buffer_info->page);
1708			buffer_info->page = NULL;
1709		}
1710
1711		if (buffer_info->skb) {
1712			dev_kfree_skb(buffer_info->skb);
1713			buffer_info->skb = NULL;
1714		}
1715
1716		for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1717			ps_page = &buffer_info->ps_pages[j];
1718			if (!ps_page->page)
1719				break;
1720			dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1721				       DMA_FROM_DEVICE);
1722			ps_page->dma = 0;
1723			put_page(ps_page->page);
1724			ps_page->page = NULL;
1725		}
1726	}
1727
1728	/* there also may be some cached data from a chained receive */
1729	if (rx_ring->rx_skb_top) {
1730		dev_kfree_skb(rx_ring->rx_skb_top);
1731		rx_ring->rx_skb_top = NULL;
1732	}
1733
1734	/* Zero out the descriptor ring */
1735	memset(rx_ring->desc, 0, rx_ring->size);
1736
1737	rx_ring->next_to_clean = 0;
1738	rx_ring->next_to_use = 0;
1739	adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1740}
1741
1742static void e1000e_downshift_workaround(struct work_struct *work)
1743{
1744	struct e1000_adapter *adapter = container_of(work,
1745						     struct e1000_adapter,
1746						     downshift_task);
1747
1748	if (test_bit(__E1000_DOWN, &adapter->state))
1749		return;
1750
1751	e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1752}
1753
1754/**
1755 * e1000_intr_msi - Interrupt Handler
1756 * @irq: interrupt number
1757 * @data: pointer to a network interface device structure
1758 **/
1759static irqreturn_t e1000_intr_msi(int __always_unused irq, void *data)
1760{
1761	struct net_device *netdev = data;
1762	struct e1000_adapter *adapter = netdev_priv(netdev);
1763	struct e1000_hw *hw = &adapter->hw;
1764	u32 icr = er32(ICR);
1765
1766	/* read ICR disables interrupts using IAM */
1767	if (icr & E1000_ICR_LSC) {
1768		hw->mac.get_link_status = true;
1769		/* ICH8 workaround-- Call gig speed drop workaround on cable
1770		 * disconnect (LSC) before accessing any PHY registers
1771		 */
1772		if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1773		    (!(er32(STATUS) & E1000_STATUS_LU)))
1774			schedule_work(&adapter->downshift_task);
1775
1776		/* 80003ES2LAN workaround-- For packet buffer work-around on
1777		 * link down event; disable receives here in the ISR and reset
1778		 * adapter in watchdog
1779		 */
1780		if (netif_carrier_ok(netdev) &&
1781		    adapter->flags & FLAG_RX_NEEDS_RESTART) {
1782			/* disable receives */
1783			u32 rctl = er32(RCTL);
1784
1785			ew32(RCTL, rctl & ~E1000_RCTL_EN);
1786			adapter->flags |= FLAG_RESTART_NOW;
1787		}
1788		/* guard against interrupt when we're going down */
1789		if (!test_bit(__E1000_DOWN, &adapter->state))
1790			mod_timer(&adapter->watchdog_timer, jiffies + 1);
1791	}
1792
1793	/* Reset on uncorrectable ECC error */
1794	if ((icr & E1000_ICR_ECCER) && ((hw->mac.type == e1000_pch_lpt) ||
1795					(hw->mac.type == e1000_pch_spt))) {
1796		u32 pbeccsts = er32(PBECCSTS);
1797
1798		adapter->corr_errors +=
1799		    pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1800		adapter->uncorr_errors +=
1801		    (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
1802		    E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
1803
1804		/* Do the reset outside of interrupt context */
1805		schedule_work(&adapter->reset_task);
1806
1807		/* return immediately since reset is imminent */
1808		return IRQ_HANDLED;
1809	}
1810
1811	if (napi_schedule_prep(&adapter->napi)) {
1812		adapter->total_tx_bytes = 0;
1813		adapter->total_tx_packets = 0;
1814		adapter->total_rx_bytes = 0;
1815		adapter->total_rx_packets = 0;
1816		__napi_schedule(&adapter->napi);
1817	}
1818
1819	return IRQ_HANDLED;
1820}
1821
1822/**
1823 * e1000_intr - Interrupt Handler
1824 * @irq: interrupt number
1825 * @data: pointer to a network interface device structure
1826 **/
1827static irqreturn_t e1000_intr(int __always_unused irq, void *data)
1828{
1829	struct net_device *netdev = data;
1830	struct e1000_adapter *adapter = netdev_priv(netdev);
1831	struct e1000_hw *hw = &adapter->hw;
1832	u32 rctl, icr = er32(ICR);
1833
1834	if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1835		return IRQ_NONE;	/* Not our interrupt */
1836
1837	/* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1838	 * not set, then the adapter didn't send an interrupt
1839	 */
1840	if (!(icr & E1000_ICR_INT_ASSERTED))
1841		return IRQ_NONE;
1842
1843	/* Interrupt Auto-Mask...upon reading ICR,
1844	 * interrupts are masked.  No need for the
1845	 * IMC write
1846	 */
1847
1848	if (icr & E1000_ICR_LSC) {
1849		hw->mac.get_link_status = true;
1850		/* ICH8 workaround-- Call gig speed drop workaround on cable
1851		 * disconnect (LSC) before accessing any PHY registers
1852		 */
1853		if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1854		    (!(er32(STATUS) & E1000_STATUS_LU)))
1855			schedule_work(&adapter->downshift_task);
1856
1857		/* 80003ES2LAN workaround--
1858		 * For packet buffer work-around on link down event;
1859		 * disable receives here in the ISR and
1860		 * reset adapter in watchdog
1861		 */
1862		if (netif_carrier_ok(netdev) &&
1863		    (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1864			/* disable receives */
1865			rctl = er32(RCTL);
1866			ew32(RCTL, rctl & ~E1000_RCTL_EN);
1867			adapter->flags |= FLAG_RESTART_NOW;
1868		}
1869		/* guard against interrupt when we're going down */
1870		if (!test_bit(__E1000_DOWN, &adapter->state))
1871			mod_timer(&adapter->watchdog_timer, jiffies + 1);
1872	}
1873
1874	/* Reset on uncorrectable ECC error */
1875	if ((icr & E1000_ICR_ECCER) && ((hw->mac.type == e1000_pch_lpt) ||
1876					(hw->mac.type == e1000_pch_spt))) {
1877		u32 pbeccsts = er32(PBECCSTS);
1878
1879		adapter->corr_errors +=
1880		    pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1881		adapter->uncorr_errors +=
1882		    (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
1883		    E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
1884
1885		/* Do the reset outside of interrupt context */
1886		schedule_work(&adapter->reset_task);
1887
1888		/* return immediately since reset is imminent */
1889		return IRQ_HANDLED;
1890	}
1891
1892	if (napi_schedule_prep(&adapter->napi)) {
1893		adapter->total_tx_bytes = 0;
1894		adapter->total_tx_packets = 0;
1895		adapter->total_rx_bytes = 0;
1896		adapter->total_rx_packets = 0;
1897		__napi_schedule(&adapter->napi);
1898	}
1899
1900	return IRQ_HANDLED;
1901}
1902
1903static irqreturn_t e1000_msix_other(int __always_unused irq, void *data)
1904{
1905	struct net_device *netdev = data;
1906	struct e1000_adapter *adapter = netdev_priv(netdev);
1907	struct e1000_hw *hw = &adapter->hw;
1908	u32 icr = er32(ICR);
1909
1910	if (!(icr & E1000_ICR_INT_ASSERTED)) {
1911		if (!test_bit(__E1000_DOWN, &adapter->state))
1912			ew32(IMS, E1000_IMS_OTHER);
1913		return IRQ_NONE;
1914	}
1915
1916	if (icr & adapter->eiac_mask)
1917		ew32(ICS, (icr & adapter->eiac_mask));
1918
1919	if (icr & E1000_ICR_OTHER) {
1920		if (!(icr & E1000_ICR_LSC))
1921			goto no_link_interrupt;
1922		hw->mac.get_link_status = true;
1923		/* guard against interrupt when we're going down */
1924		if (!test_bit(__E1000_DOWN, &adapter->state))
1925			mod_timer(&adapter->watchdog_timer, jiffies + 1);
1926	}
1927
1928no_link_interrupt:
1929	if (!test_bit(__E1000_DOWN, &adapter->state))
1930		ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1931
1932	return IRQ_HANDLED;
1933}
1934
1935static irqreturn_t e1000_intr_msix_tx(int __always_unused irq, void *data)
1936{
1937	struct net_device *netdev = data;
1938	struct e1000_adapter *adapter = netdev_priv(netdev);
1939	struct e1000_hw *hw = &adapter->hw;
1940	struct e1000_ring *tx_ring = adapter->tx_ring;
1941
1942	adapter->total_tx_bytes = 0;
1943	adapter->total_tx_packets = 0;
1944
1945	if (!e1000_clean_tx_irq(tx_ring))
1946		/* Ring was not completely cleaned, so fire another interrupt */
1947		ew32(ICS, tx_ring->ims_val);
1948
1949	return IRQ_HANDLED;
1950}
1951
1952static irqreturn_t e1000_intr_msix_rx(int __always_unused irq, void *data)
1953{
1954	struct net_device *netdev = data;
1955	struct e1000_adapter *adapter = netdev_priv(netdev);
1956	struct e1000_ring *rx_ring = adapter->rx_ring;
1957
1958	/* Write the ITR value calculated at the end of the
1959	 * previous interrupt.
1960	 */
1961	if (rx_ring->set_itr) {
1962		writel(1000000000 / (rx_ring->itr_val * 256),
1963		       rx_ring->itr_register);
1964		rx_ring->set_itr = 0;
1965	}
1966
1967	if (napi_schedule_prep(&adapter->napi)) {
1968		adapter->total_rx_bytes = 0;
1969		adapter->total_rx_packets = 0;
1970		__napi_schedule(&adapter->napi);
1971	}
1972	return IRQ_HANDLED;
1973}
1974
1975/**
1976 * e1000_configure_msix - Configure MSI-X hardware
1977 *
1978 * e1000_configure_msix sets up the hardware to properly
1979 * generate MSI-X interrupts.
1980 **/
1981static void e1000_configure_msix(struct e1000_adapter *adapter)
1982{
1983	struct e1000_hw *hw = &adapter->hw;
1984	struct e1000_ring *rx_ring = adapter->rx_ring;
1985	struct e1000_ring *tx_ring = adapter->tx_ring;
1986	int vector = 0;
1987	u32 ctrl_ext, ivar = 0;
1988
1989	adapter->eiac_mask = 0;
1990
1991	/* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1992	if (hw->mac.type == e1000_82574) {
1993		u32 rfctl = er32(RFCTL);
1994
1995		rfctl |= E1000_RFCTL_ACK_DIS;
1996		ew32(RFCTL, rfctl);
1997	}
1998
1999	/* Configure Rx vector */
2000	rx_ring->ims_val = E1000_IMS_RXQ0;
2001	adapter->eiac_mask |= rx_ring->ims_val;
2002	if (rx_ring->itr_val)
2003		writel(1000000000 / (rx_ring->itr_val * 256),
2004		       rx_ring->itr_register);
2005	else
2006		writel(1, rx_ring->itr_register);
2007	ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
2008
2009	/* Configure Tx vector */
2010	tx_ring->ims_val = E1000_IMS_TXQ0;
2011	vector++;
2012	if (tx_ring->itr_val)
2013		writel(1000000000 / (tx_ring->itr_val * 256),
2014		       tx_ring->itr_register);
2015	else
2016		writel(1, tx_ring->itr_register);
2017	adapter->eiac_mask |= tx_ring->ims_val;
2018	ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
2019
2020	/* set vector for Other Causes, e.g. link changes */
2021	vector++;
2022	ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
2023	if (rx_ring->itr_val)
2024		writel(1000000000 / (rx_ring->itr_val * 256),
2025		       hw->hw_addr + E1000_EITR_82574(vector));
2026	else
2027		writel(1, hw->hw_addr + E1000_EITR_82574(vector));
2028
2029	/* Cause Tx interrupts on every write back */
2030	ivar |= (1 << 31);
2031
2032	ew32(IVAR, ivar);
2033
2034	/* enable MSI-X PBA support */
2035	ctrl_ext = er32(CTRL_EXT);
2036	ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
2037
2038	/* Auto-Mask Other interrupts upon ICR read */
2039	ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
2040	ctrl_ext |= E1000_CTRL_EXT_EIAME;
2041	ew32(CTRL_EXT, ctrl_ext);
2042	e1e_flush();
2043}
2044
2045void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
2046{
2047	if (adapter->msix_entries) {
2048		pci_disable_msix(adapter->pdev);
2049		kfree(adapter->msix_entries);
2050		adapter->msix_entries = NULL;
2051	} else if (adapter->flags & FLAG_MSI_ENABLED) {
2052		pci_disable_msi(adapter->pdev);
2053		adapter->flags &= ~FLAG_MSI_ENABLED;
2054	}
2055}
2056
2057/**
2058 * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
2059 *
2060 * Attempt to configure interrupts using the best available
2061 * capabilities of the hardware and kernel.
2062 **/
2063void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
2064{
2065	int err;
2066	int i;
2067
2068	switch (adapter->int_mode) {
2069	case E1000E_INT_MODE_MSIX:
2070		if (adapter->flags & FLAG_HAS_MSIX) {
2071			adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
2072			adapter->msix_entries = kcalloc(adapter->num_vectors,
2073							sizeof(struct
2074							       msix_entry),
2075							GFP_KERNEL);
2076			if (adapter->msix_entries) {
2077				struct e1000_adapter *a = adapter;
2078
2079				for (i = 0; i < adapter->num_vectors; i++)
2080					adapter->msix_entries[i].entry = i;
2081
2082				err = pci_enable_msix_range(a->pdev,
2083							    a->msix_entries,
2084							    a->num_vectors,
2085							    a->num_vectors);
2086				if (err > 0)
2087					return;
2088			}
2089			/* MSI-X failed, so fall through and try MSI */
2090			e_err("Failed to initialize MSI-X interrupts.  Falling back to MSI interrupts.\n");
2091			e1000e_reset_interrupt_capability(adapter);
2092		}
2093		adapter->int_mode = E1000E_INT_MODE_MSI;
2094		/* Fall through */
2095	case E1000E_INT_MODE_MSI:
2096		if (!pci_enable_msi(adapter->pdev)) {
2097			adapter->flags |= FLAG_MSI_ENABLED;
2098		} else {
2099			adapter->int_mode = E1000E_INT_MODE_LEGACY;
2100			e_err("Failed to initialize MSI interrupts.  Falling back to legacy interrupts.\n");
2101		}
2102		/* Fall through */
2103	case E1000E_INT_MODE_LEGACY:
2104		/* Don't do anything; this is the system default */
2105		break;
2106	}
2107
2108	/* store the number of vectors being used */
2109	adapter->num_vectors = 1;
2110}
2111
2112/**
2113 * e1000_request_msix - Initialize MSI-X interrupts
2114 *
2115 * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
2116 * kernel.
2117 **/
2118static int e1000_request_msix(struct e1000_adapter *adapter)
2119{
2120	struct net_device *netdev = adapter->netdev;
2121	int err = 0, vector = 0;
2122
2123	if (strlen(netdev->name) < (IFNAMSIZ - 5))
2124		snprintf(adapter->rx_ring->name,
2125			 sizeof(adapter->rx_ring->name) - 1,
2126			 "%s-rx-0", netdev->name);
2127	else
2128		memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
2129	err = request_irq(adapter->msix_entries[vector].vector,
2130			  e1000_intr_msix_rx, 0, adapter->rx_ring->name,
2131			  netdev);
2132	if (err)
2133		return err;
2134	adapter->rx_ring->itr_register = adapter->hw.hw_addr +
2135	    E1000_EITR_82574(vector);
2136	adapter->rx_ring->itr_val = adapter->itr;
2137	vector++;
2138
2139	if (strlen(netdev->name) < (IFNAMSIZ - 5))
2140		snprintf(adapter->tx_ring->name,
2141			 sizeof(adapter->tx_ring->name) - 1,
2142			 "%s-tx-0", netdev->name);
2143	else
2144		memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
2145	err = request_irq(adapter->msix_entries[vector].vector,
2146			  e1000_intr_msix_tx, 0, adapter->tx_ring->name,
2147			  netdev);
2148	if (err)
2149		return err;
2150	adapter->tx_ring->itr_register = adapter->hw.hw_addr +
2151	    E1000_EITR_82574(vector);
2152	adapter->tx_ring->itr_val = adapter->itr;
2153	vector++;
2154
2155	err = request_irq(adapter->msix_entries[vector].vector,
2156			  e1000_msix_other, 0, netdev->name, netdev);
2157	if (err)
2158		return err;
2159
2160	e1000_configure_msix(adapter);
2161
2162	return 0;
2163}
2164
2165/**
2166 * e1000_request_irq - initialize interrupts
2167 *
2168 * Attempts to configure interrupts using the best available
2169 * capabilities of the hardware and kernel.
2170 **/
2171static int e1000_request_irq(struct e1000_adapter *adapter)
2172{
2173	struct net_device *netdev = adapter->netdev;
2174	int err;
2175
2176	if (adapter->msix_entries) {
2177		err = e1000_request_msix(adapter);
2178		if (!err)
2179			return err;
2180		/* fall back to MSI */
2181		e1000e_reset_interrupt_capability(adapter);
2182		adapter->int_mode = E1000E_INT_MODE_MSI;
2183		e1000e_set_interrupt_capability(adapter);
2184	}
2185	if (adapter->flags & FLAG_MSI_ENABLED) {
2186		err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2187				  netdev->name, netdev);
2188		if (!err)
2189			return err;
2190
2191		/* fall back to legacy interrupt */
2192		e1000e_reset_interrupt_capability(adapter);
2193		adapter->int_mode = E1000E_INT_MODE_LEGACY;
2194	}
2195
2196	err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2197			  netdev->name, netdev);
2198	if (err)
2199		e_err("Unable to allocate interrupt, Error: %d\n", err);
2200
2201	return err;
2202}
2203
2204static void e1000_free_irq(struct e1000_adapter *adapter)
2205{
2206	struct net_device *netdev = adapter->netdev;
2207
2208	if (adapter->msix_entries) {
2209		int vector = 0;
2210
2211		free_irq(adapter->msix_entries[vector].vector, netdev);
2212		vector++;
2213
2214		free_irq(adapter->msix_entries[vector].vector, netdev);
2215		vector++;
2216
2217		/* Other Causes interrupt vector */
2218		free_irq(adapter->msix_entries[vector].vector, netdev);
2219		return;
2220	}
2221
2222	free_irq(adapter->pdev->irq, netdev);
2223}
2224
2225/**
2226 * e1000_irq_disable - Mask off interrupt generation on the NIC
2227 **/
2228static void e1000_irq_disable(struct e1000_adapter *adapter)
2229{
2230	struct e1000_hw *hw = &adapter->hw;
2231
2232	ew32(IMC, ~0);
2233	if (adapter->msix_entries)
2234		ew32(EIAC_82574, 0);
2235	e1e_flush();
2236
2237	if (adapter->msix_entries) {
2238		int i;
2239
2240		for (i = 0; i < adapter->num_vectors; i++)
2241			synchronize_irq(adapter->msix_entries[i].vector);
2242	} else {
2243		synchronize_irq(adapter->pdev->irq);
2244	}
2245}
2246
2247/**
2248 * e1000_irq_enable - Enable default interrupt generation settings
2249 **/
2250static void e1000_irq_enable(struct e1000_adapter *adapter)
2251{
2252	struct e1000_hw *hw = &adapter->hw;
2253
2254	if (adapter->msix_entries) {
2255		ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2256		ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
2257	} else if ((hw->mac.type == e1000_pch_lpt) ||
2258		   (hw->mac.type == e1000_pch_spt)) {
2259		ew32(IMS, IMS_ENABLE_MASK | E1000_IMS_ECCER);
2260	} else {
2261		ew32(IMS, IMS_ENABLE_MASK);
2262	}
2263	e1e_flush();
2264}
2265
2266/**
2267 * e1000e_get_hw_control - get control of the h/w from f/w
2268 * @adapter: address of board private structure
2269 *
2270 * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2271 * For ASF and Pass Through versions of f/w this means that
2272 * the driver is loaded. For AMT version (only with 82573)
2273 * of the f/w this means that the network i/f is open.
2274 **/
2275void e1000e_get_hw_control(struct e1000_adapter *adapter)
2276{
2277	struct e1000_hw *hw = &adapter->hw;
2278	u32 ctrl_ext;
2279	u32 swsm;
2280
2281	/* Let firmware know the driver has taken over */
2282	if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2283		swsm = er32(SWSM);
2284		ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2285	} else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2286		ctrl_ext = er32(CTRL_EXT);
2287		ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2288	}
2289}
2290
2291/**
2292 * e1000e_release_hw_control - release control of the h/w to f/w
2293 * @adapter: address of board private structure
2294 *
2295 * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2296 * For ASF and Pass Through versions of f/w this means that the
2297 * driver is no longer loaded. For AMT version (only with 82573) i
2298 * of the f/w this means that the network i/f is closed.
2299 *
2300 **/
2301void e1000e_release_hw_control(struct e1000_adapter *adapter)
2302{
2303	struct e1000_hw *hw = &adapter->hw;
2304	u32 ctrl_ext;
2305	u32 swsm;
2306
2307	/* Let firmware taken over control of h/w */
2308	if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2309		swsm = er32(SWSM);
2310		ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2311	} else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2312		ctrl_ext = er32(CTRL_EXT);
2313		ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2314	}
2315}
2316
2317/**
2318 * e1000_alloc_ring_dma - allocate memory for a ring structure
2319 **/
2320static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2321				struct e1000_ring *ring)
2322{
2323	struct pci_dev *pdev = adapter->pdev;
2324
2325	ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2326					GFP_KERNEL);
2327	if (!ring->desc)
2328		return -ENOMEM;
2329
2330	return 0;
2331}
2332
2333/**
2334 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2335 * @tx_ring: Tx descriptor ring
2336 *
2337 * Return 0 on success, negative on failure
2338 **/
2339int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2340{
2341	struct e1000_adapter *adapter = tx_ring->adapter;
2342	int err = -ENOMEM, size;
2343
2344	size = sizeof(struct e1000_buffer) * tx_ring->count;
2345	tx_ring->buffer_info = vzalloc(size);
2346	if (!tx_ring->buffer_info)
2347		goto err;
2348
2349	/* round up to nearest 4K */
2350	tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2351	tx_ring->size = ALIGN(tx_ring->size, 4096);
2352
2353	err = e1000_alloc_ring_dma(adapter, tx_ring);
2354	if (err)
2355		goto err;
2356
2357	tx_ring->next_to_use = 0;
2358	tx_ring->next_to_clean = 0;
2359
2360	return 0;
2361err:
2362	vfree(tx_ring->buffer_info);
2363	e_err("Unable to allocate memory for the transmit descriptor ring\n");
2364	return err;
2365}
2366
2367/**
2368 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2369 * @rx_ring: Rx descriptor ring
2370 *
2371 * Returns 0 on success, negative on failure
2372 **/
2373int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2374{
2375	struct e1000_adapter *adapter = rx_ring->adapter;
2376	struct e1000_buffer *buffer_info;
2377	int i, size, desc_len, err = -ENOMEM;
2378
2379	size = sizeof(struct e1000_buffer) * rx_ring->count;
2380	rx_ring->buffer_info = vzalloc(size);
2381	if (!rx_ring->buffer_info)
2382		goto err;
2383
2384	for (i = 0; i < rx_ring->count; i++) {
2385		buffer_info = &rx_ring->buffer_info[i];
2386		buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2387						sizeof(struct e1000_ps_page),
2388						GFP_KERNEL);
2389		if (!buffer_info->ps_pages)
2390			goto err_pages;
2391	}
2392
2393	desc_len = sizeof(union e1000_rx_desc_packet_split);
2394
2395	/* Round up to nearest 4K */
2396	rx_ring->size = rx_ring->count * desc_len;
2397	rx_ring->size = ALIGN(rx_ring->size, 4096);
2398
2399	err = e1000_alloc_ring_dma(adapter, rx_ring);
2400	if (err)
2401		goto err_pages;
2402
2403	rx_ring->next_to_clean = 0;
2404	rx_ring->next_to_use = 0;
2405	rx_ring->rx_skb_top = NULL;
2406
2407	return 0;
2408
2409err_pages:
2410	for (i = 0; i < rx_ring->count; i++) {
2411		buffer_info = &rx_ring->buffer_info[i];
2412		kfree(buffer_info->ps_pages);
2413	}
2414err:
2415	vfree(rx_ring->buffer_info);
2416	e_err("Unable to allocate memory for the receive descriptor ring\n");
2417	return err;
2418}
2419
2420/**
2421 * e1000_clean_tx_ring - Free Tx Buffers
2422 * @tx_ring: Tx descriptor ring
2423 **/
2424static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2425{
2426	struct e1000_adapter *adapter = tx_ring->adapter;
2427	struct e1000_buffer *buffer_info;
2428	unsigned long size;
2429	unsigned int i;
2430
2431	for (i = 0; i < tx_ring->count; i++) {
2432		buffer_info = &tx_ring->buffer_info[i];
2433		e1000_put_txbuf(tx_ring, buffer_info);
2434	}
2435
2436	netdev_reset_queue(adapter->netdev);
2437	size = sizeof(struct e1000_buffer) * tx_ring->count;
2438	memset(tx_ring->buffer_info, 0, size);
2439
2440	memset(tx_ring->desc, 0, tx_ring->size);
2441
2442	tx_ring->next_to_use = 0;
2443	tx_ring->next_to_clean = 0;
2444}
2445
2446/**
2447 * e1000e_free_tx_resources - Free Tx Resources per Queue
2448 * @tx_ring: Tx descriptor ring
2449 *
2450 * Free all transmit software resources
2451 **/
2452void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2453{
2454	struct e1000_adapter *adapter = tx_ring->adapter;
2455	struct pci_dev *pdev = adapter->pdev;
2456
2457	e1000_clean_tx_ring(tx_ring);
2458
2459	vfree(tx_ring->buffer_info);
2460	tx_ring->buffer_info = NULL;
2461
2462	dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2463			  tx_ring->dma);
2464	tx_ring->desc = NULL;
2465}
2466
2467/**
2468 * e1000e_free_rx_resources - Free Rx Resources
2469 * @rx_ring: Rx descriptor ring
2470 *
2471 * Free all receive software resources
2472 **/
2473void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2474{
2475	struct e1000_adapter *adapter = rx_ring->adapter;
2476	struct pci_dev *pdev = adapter->pdev;
2477	int i;
2478
2479	e1000_clean_rx_ring(rx_ring);
2480
2481	for (i = 0; i < rx_ring->count; i++)
2482		kfree(rx_ring->buffer_info[i].ps_pages);
2483
2484	vfree(rx_ring->buffer_info);
2485	rx_ring->buffer_info = NULL;
2486
2487	dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2488			  rx_ring->dma);
2489	rx_ring->desc = NULL;
2490}
2491
2492/**
2493 * e1000_update_itr - update the dynamic ITR value based on statistics
2494 * @adapter: pointer to adapter
2495 * @itr_setting: current adapter->itr
2496 * @packets: the number of packets during this measurement interval
2497 * @bytes: the number of bytes during this measurement interval
2498 *
2499 *      Stores a new ITR value based on packets and byte
2500 *      counts during the last interrupt.  The advantage of per interrupt
2501 *      computation is faster updates and more accurate ITR for the current
2502 *      traffic pattern.  Constants in this function were computed
2503 *      based on theoretical maximum wire speed and thresholds were set based
2504 *      on testing data as well as attempting to minimize response time
2505 *      while increasing bulk throughput.  This functionality is controlled
2506 *      by the InterruptThrottleRate module parameter.
2507 **/
2508static unsigned int e1000_update_itr(u16 itr_setting, int packets, int bytes)
2509{
2510	unsigned int retval = itr_setting;
2511
2512	if (packets == 0)
2513		return itr_setting;
2514
2515	switch (itr_setting) {
2516	case lowest_latency:
2517		/* handle TSO and jumbo frames */
2518		if (bytes / packets > 8000)
2519			retval = bulk_latency;
2520		else if ((packets < 5) && (bytes > 512))
2521			retval = low_latency;
2522		break;
2523	case low_latency:	/* 50 usec aka 20000 ints/s */
2524		if (bytes > 10000) {
2525			/* this if handles the TSO accounting */
2526			if (bytes / packets > 8000)
2527				retval = bulk_latency;
2528			else if ((packets < 10) || ((bytes / packets) > 1200))
2529				retval = bulk_latency;
2530			else if ((packets > 35))
2531				retval = lowest_latency;
2532		} else if (bytes / packets > 2000) {
2533			retval = bulk_latency;
2534		} else if (packets <= 2 && bytes < 512) {
2535			retval = lowest_latency;
2536		}
2537		break;
2538	case bulk_latency:	/* 250 usec aka 4000 ints/s */
2539		if (bytes > 25000) {
2540			if (packets > 35)
2541				retval = low_latency;
2542		} else if (bytes < 6000) {
2543			retval = low_latency;
2544		}
2545		break;
2546	}
2547
2548	return retval;
2549}
2550
2551static void e1000_set_itr(struct e1000_adapter *adapter)
2552{
2553	u16 current_itr;
2554	u32 new_itr = adapter->itr;
2555
2556	/* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2557	if (adapter->link_speed != SPEED_1000) {
2558		current_itr = 0;
2559		new_itr = 4000;
2560		goto set_itr_now;
2561	}
2562
2563	if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2564		new_itr = 0;
2565		goto set_itr_now;
2566	}
2567
2568	adapter->tx_itr = e1000_update_itr(adapter->tx_itr,
2569					   adapter->total_tx_packets,
2570					   adapter->total_tx_bytes);
2571	/* conservative mode (itr 3) eliminates the lowest_latency setting */
2572	if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2573		adapter->tx_itr = low_latency;
2574
2575	adapter->rx_itr = e1000_update_itr(adapter->rx_itr,
2576					   adapter->total_rx_packets,
2577					   adapter->total_rx_bytes);
2578	/* conservative mode (itr 3) eliminates the lowest_latency setting */
2579	if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2580		adapter->rx_itr = low_latency;
2581
2582	current_itr = max(adapter->rx_itr, adapter->tx_itr);
2583
2584	/* counts and packets in update_itr are dependent on these numbers */
2585	switch (current_itr) {
2586	case lowest_latency:
2587		new_itr = 70000;
2588		break;
2589	case low_latency:
2590		new_itr = 20000;	/* aka hwitr = ~200 */
2591		break;
2592	case bulk_latency:
2593		new_itr = 4000;
2594		break;
2595	default:
2596		break;
2597	}
2598
2599set_itr_now:
2600	if (new_itr != adapter->itr) {
2601		/* this attempts to bias the interrupt rate towards Bulk
2602		 * by adding intermediate steps when interrupt rate is
2603		 * increasing
2604		 */
2605		new_itr = new_itr > adapter->itr ?
2606		    min(adapter->itr + (new_itr >> 2), new_itr) : new_itr;
2607		adapter->itr = new_itr;
2608		adapter->rx_ring->itr_val = new_itr;
2609		if (adapter->msix_entries)
2610			adapter->rx_ring->set_itr = 1;
2611		else
2612			e1000e_write_itr(adapter, new_itr);
2613	}
2614}
2615
2616/**
2617 * e1000e_write_itr - write the ITR value to the appropriate registers
2618 * @adapter: address of board private structure
2619 * @itr: new ITR value to program
2620 *
2621 * e1000e_write_itr determines if the adapter is in MSI-X mode
2622 * and, if so, writes the EITR registers with the ITR value.
2623 * Otherwise, it writes the ITR value into the ITR register.
2624 **/
2625void e1000e_write_itr(struct e1000_adapter *adapter, u32 itr)
2626{
2627	struct e1000_hw *hw = &adapter->hw;
2628	u32 new_itr = itr ? 1000000000 / (itr * 256) : 0;
2629
2630	if (adapter->msix_entries) {
2631		int vector;
2632
2633		for (vector = 0; vector < adapter->num_vectors; vector++)
2634			writel(new_itr, hw->hw_addr + E1000_EITR_82574(vector));
2635	} else {
2636		ew32(ITR, new_itr);
2637	}
2638}
2639
2640/**
2641 * e1000_alloc_queues - Allocate memory for all rings
2642 * @adapter: board private structure to initialize
2643 **/
2644static int e1000_alloc_queues(struct e1000_adapter *adapter)
2645{
2646	int size = sizeof(struct e1000_ring);
2647
2648	adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2649	if (!adapter->tx_ring)
2650		goto err;
2651	adapter->tx_ring->count = adapter->tx_ring_count;
2652	adapter->tx_ring->adapter = adapter;
2653
2654	adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2655	if (!adapter->rx_ring)
2656		goto err;
2657	adapter->rx_ring->count = adapter->rx_ring_count;
2658	adapter->rx_ring->adapter = adapter;
2659
2660	return 0;
2661err:
2662	e_err("Unable to allocate memory for queues\n");
2663	kfree(adapter->rx_ring);
2664	kfree(adapter->tx_ring);
2665	return -ENOMEM;
2666}
2667
2668/**
2669 * e1000e_poll - NAPI Rx polling callback
2670 * @napi: struct associated with this polling callback
2671 * @weight: number of packets driver is allowed to process this poll
2672 **/
2673static int e1000e_poll(struct napi_struct *napi, int weight)
2674{
2675	struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
2676						     napi);
2677	struct e1000_hw *hw = &adapter->hw;
2678	struct net_device *poll_dev = adapter->netdev;
2679	int tx_cleaned = 1, work_done = 0;
2680
2681	adapter = netdev_priv(poll_dev);
2682
2683	if (!adapter->msix_entries ||
2684	    (adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2685		tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2686
2687	adapter->clean_rx(adapter->rx_ring, &work_done, weight);
2688
2689	if (!tx_cleaned)
2690		work_done = weight;
2691
2692	/* If weight not fully consumed, exit the polling mode */
2693	if (work_done < weight) {
2694		if (adapter->itr_setting & 3)
2695			e1000_set_itr(adapter);
2696		napi_complete_done(napi, work_done);
2697		if (!test_bit(__E1000_DOWN, &adapter->state)) {
2698			if (adapter->msix_entries)
2699				ew32(IMS, adapter->rx_ring->ims_val);
2700			else
2701				e1000_irq_enable(adapter);
2702		}
2703	}
2704
2705	return work_done;
2706}
2707
2708static int e1000_vlan_rx_add_vid(struct net_device *netdev,
2709				 __always_unused __be16 proto, u16 vid)
2710{
2711	struct e1000_adapter *adapter = netdev_priv(netdev);
2712	struct e1000_hw *hw = &adapter->hw;
2713	u32 vfta, index;
2714
2715	/* don't update vlan cookie if already programmed */
2716	if ((adapter->hw.mng_cookie.status &
2717	     E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2718	    (vid == adapter->mng_vlan_id))
2719		return 0;
2720
2721	/* add VID to 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	set_bit(vid, adapter->active_vlans);
2730
2731	return 0;
2732}
2733
2734static int e1000_vlan_rx_kill_vid(struct net_device *netdev,
2735				  __always_unused __be16 proto, u16 vid)
2736{
2737	struct e1000_adapter *adapter = netdev_priv(netdev);
2738	struct e1000_hw *hw = &adapter->hw;
2739	u32 vfta, index;
2740
2741	if ((adapter->hw.mng_cookie.status &
2742	     E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2743	    (vid == adapter->mng_vlan_id)) {
2744		/* release control to f/w */
2745		e1000e_release_hw_control(adapter);
2746		return 0;
2747	}
2748
2749	/* remove VID from filter table */
2750	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2751		index = (vid >> 5) & 0x7F;
2752		vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2753		vfta &= ~(1 << (vid & 0x1F));
2754		hw->mac.ops.write_vfta(hw, index, vfta);
2755	}
2756
2757	clear_bit(vid, adapter->active_vlans);
2758
2759	return 0;
2760}
2761
2762/**
2763 * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2764 * @adapter: board private structure to initialize
2765 **/
2766static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2767{
2768	struct net_device *netdev = adapter->netdev;
2769	struct e1000_hw *hw = &adapter->hw;
2770	u32 rctl;
2771
2772	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2773		/* disable VLAN receive filtering */
2774		rctl = er32(RCTL);
2775		rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2776		ew32(RCTL, rctl);
2777
2778		if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2779			e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
2780					       adapter->mng_vlan_id);
2781			adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2782		}
2783	}
2784}
2785
2786/**
2787 * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2788 * @adapter: board private structure to initialize
2789 **/
2790static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2791{
2792	struct e1000_hw *hw = &adapter->hw;
2793	u32 rctl;
2794
2795	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2796		/* enable VLAN receive filtering */
2797		rctl = er32(RCTL);
2798		rctl |= E1000_RCTL_VFE;
2799		rctl &= ~E1000_RCTL_CFIEN;
2800		ew32(RCTL, rctl);
2801	}
2802}
2803
2804/**
2805 * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2806 * @adapter: board private structure to initialize
2807 **/
2808static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2809{
2810	struct e1000_hw *hw = &adapter->hw;
2811	u32 ctrl;
2812
2813	/* disable VLAN tag insert/strip */
2814	ctrl = er32(CTRL);
2815	ctrl &= ~E1000_CTRL_VME;
2816	ew32(CTRL, ctrl);
2817}
2818
2819/**
2820 * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2821 * @adapter: board private structure to initialize
2822 **/
2823static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2824{
2825	struct e1000_hw *hw = &adapter->hw;
2826	u32 ctrl;
2827
2828	/* enable VLAN tag insert/strip */
2829	ctrl = er32(CTRL);
2830	ctrl |= E1000_CTRL_VME;
2831	ew32(CTRL, ctrl);
2832}
2833
2834static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2835{
2836	struct net_device *netdev = adapter->netdev;
2837	u16 vid = adapter->hw.mng_cookie.vlan_id;
2838	u16 old_vid = adapter->mng_vlan_id;
2839
2840	if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2841		e1000_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
2842		adapter->mng_vlan_id = vid;
2843	}
2844
2845	if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2846		e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q), old_vid);
2847}
2848
2849static void e1000_restore_vlan(struct e1000_adapter *adapter)
2850{
2851	u16 vid;
2852
2853	e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
2854
2855	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2856	    e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
2857}
2858
2859static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2860{
2861	struct e1000_hw *hw = &adapter->hw;
2862	u32 manc, manc2h, mdef, i, j;
2863
2864	if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2865		return;
2866
2867	manc = er32(MANC);
2868
2869	/* enable receiving management packets to the host. this will probably
2870	 * generate destination unreachable messages from the host OS, but
2871	 * the packets will be handled on SMBUS
2872	 */
2873	manc |= E1000_MANC_EN_MNG2HOST;
2874	manc2h = er32(MANC2H);
2875
2876	switch (hw->mac.type) {
2877	default:
2878		manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2879		break;
2880	case e1000_82574:
2881	case e1000_82583:
2882		/* Check if IPMI pass-through decision filter already exists;
2883		 * if so, enable it.
2884		 */
2885		for (i = 0, j = 0; i < 8; i++) {
2886			mdef = er32(MDEF(i));
2887
2888			/* Ignore filters with anything other than IPMI ports */
2889			if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2890				continue;
2891
2892			/* Enable this decision filter in MANC2H */
2893			if (mdef)
2894				manc2h |= (1 << i);
2895
2896			j |= mdef;
2897		}
2898
2899		if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2900			break;
2901
2902		/* Create new decision filter in an empty filter */
2903		for (i = 0, j = 0; i < 8; i++)
2904			if (er32(MDEF(i)) == 0) {
2905				ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2906					       E1000_MDEF_PORT_664));
2907				manc2h |= (1 << 1);
2908				j++;
2909				break;
2910			}
2911
2912		if (!j)
2913			e_warn("Unable to create IPMI pass-through filter\n");
2914		break;
2915	}
2916
2917	ew32(MANC2H, manc2h);
2918	ew32(MANC, manc);
2919}
2920
2921/**
2922 * e1000_configure_tx - Configure Transmit Unit after Reset
2923 * @adapter: board private structure
2924 *
2925 * Configure the Tx unit of the MAC after a reset.
2926 **/
2927static void e1000_configure_tx(struct e1000_adapter *adapter)
2928{
2929	struct e1000_hw *hw = &adapter->hw;
2930	struct e1000_ring *tx_ring = adapter->tx_ring;
2931	u64 tdba;
2932	u32 tdlen, tctl, tarc;
2933
2934	/* Setup the HW Tx Head and Tail descriptor pointers */
2935	tdba = tx_ring->dma;
2936	tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2937	ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
2938	ew32(TDBAH(0), (tdba >> 32));
2939	ew32(TDLEN(0), tdlen);
2940	ew32(TDH(0), 0);
2941	ew32(TDT(0), 0);
2942	tx_ring->head = adapter->hw.hw_addr + E1000_TDH(0);
2943	tx_ring->tail = adapter->hw.hw_addr + E1000_TDT(0);
2944
2945	writel(0, tx_ring->head);
2946	if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
2947		e1000e_update_tdt_wa(tx_ring, 0);
2948	else
2949		writel(0, tx_ring->tail);
2950
2951	/* Set the Tx Interrupt Delay register */
2952	ew32(TIDV, adapter->tx_int_delay);
2953	/* Tx irq moderation */
2954	ew32(TADV, adapter->tx_abs_int_delay);
2955
2956	if (adapter->flags2 & FLAG2_DMA_BURST) {
2957		u32 txdctl = er32(TXDCTL(0));
2958
2959		txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2960			    E1000_TXDCTL_WTHRESH);
2961		/* set up some performance related parameters to encourage the
2962		 * hardware to use the bus more efficiently in bursts, depends
2963		 * on the tx_int_delay to be enabled,
2964		 * wthresh = 1 ==> burst write is disabled to avoid Tx stalls
2965		 * hthresh = 1 ==> prefetch when one or more available
2966		 * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2967		 * BEWARE: this seems to work but should be considered first if
2968		 * there are Tx hangs or other Tx related bugs
2969		 */
2970		txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2971		ew32(TXDCTL(0), txdctl);
2972	}
2973	/* erratum work around: set txdctl the same for both queues */
2974	ew32(TXDCTL(1), er32(TXDCTL(0)));
2975
2976	/* Program the Transmit Control Register */
2977	tctl = er32(TCTL);
2978	tctl &= ~E1000_TCTL_CT;
2979	tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2980		(E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2981
2982	if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2983		tarc = er32(TARC(0));
2984		/* set the speed mode bit, we'll clear it if we're not at
2985		 * gigabit link later
2986		 */
2987#define SPEED_MODE_BIT (1 << 21)
2988		tarc |= SPEED_MODE_BIT;
2989		ew32(TARC(0), tarc);
2990	}
2991
2992	/* errata: program both queues to unweighted RR */
2993	if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2994		tarc = er32(TARC(0));
2995		tarc |= 1;
2996		ew32(TARC(0), tarc);
2997		tarc = er32(TARC(1));
2998		tarc |= 1;
2999		ew32(TARC(1), tarc);
3000	}
3001
3002	/* Setup Transmit Descriptor Settings for eop descriptor */
3003	adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
3004
3005	/* only set IDE if we are delaying interrupts using the timers */
3006	if (adapter->tx_int_delay)
3007		adapter->txd_cmd |= E1000_TXD_CMD_IDE;
3008
3009	/* enable Report Status bit */
3010	adapter->txd_cmd |= E1000_TXD_CMD_RS;
3011
3012	ew32(TCTL, tctl);
3013
3014	hw->mac.ops.config_collision_dist(hw);
3015
3016	/* SPT Si errata workaround to avoid data corruption */
3017	if (hw->mac.type == e1000_pch_spt) {
3018		u32 reg_val;
3019
3020		reg_val = er32(IOSFPC);
3021		reg_val |= E1000_RCTL_RDMTS_HEX;
3022		ew32(IOSFPC, reg_val);
3023
3024		reg_val = er32(TARC(0));
3025		reg_val |= E1000_TARC0_CB_MULTIQ_3_REQ;
3026		ew32(TARC(0), reg_val);
3027	}
3028}
3029
3030/**
3031 * e1000_setup_rctl - configure the receive control registers
3032 * @adapter: Board private structure
3033 **/
3034#define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
3035			   (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
3036static void e1000_setup_rctl(struct e1000_adapter *adapter)
3037{
3038	struct e1000_hw *hw = &adapter->hw;
3039	u32 rctl, rfctl;
3040	u32 pages = 0;
3041
3042	/* Workaround Si errata on PCHx - configure jumbo frame flow.
3043	 * If jumbo frames not set, program related MAC/PHY registers
3044	 * to h/w defaults
3045	 */
3046	if (hw->mac.type >= e1000_pch2lan) {
3047		s32 ret_val;
3048
3049		if (adapter->netdev->mtu > ETH_DATA_LEN)
3050			ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
3051		else
3052			ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
3053
3054		if (ret_val)
3055			e_dbg("failed to enable|disable jumbo frame workaround mode\n");
3056	}
3057
3058	/* Program MC offset vector base */
3059	rctl = er32(RCTL);
3060	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3061	rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
3062	    E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
3063	    (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3064
3065	/* Do not Store bad packets */
3066	rctl &= ~E1000_RCTL_SBP;
3067
3068	/* Enable Long Packet receive */
3069	if (adapter->netdev->mtu <= ETH_DATA_LEN)
3070		rctl &= ~E1000_RCTL_LPE;
3071	else
3072		rctl |= E1000_RCTL_LPE;
3073
3074	/* Some systems expect that the CRC is included in SMBUS traffic. The
3075	 * hardware strips the CRC before sending to both SMBUS (BMC) and to
3076	 * host memory when this is enabled
3077	 */
3078	if (adapter->flags2 & FLAG2_CRC_STRIPPING)
3079		rctl |= E1000_RCTL_SECRC;
3080
3081	/* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
3082	if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
3083		u16 phy_data;
3084
3085		e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
3086		phy_data &= 0xfff8;
3087		phy_data |= (1 << 2);
3088		e1e_wphy(hw, PHY_REG(770, 26), phy_data);
3089
3090		e1e_rphy(hw, 22, &phy_data);
3091		phy_data &= 0x0fff;
3092		phy_data |= (1 << 14);
3093		e1e_wphy(hw, 0x10, 0x2823);
3094		e1e_wphy(hw, 0x11, 0x0003);
3095		e1e_wphy(hw, 22, phy_data);
3096	}
3097
3098	/* Setup buffer sizes */
3099	rctl &= ~E1000_RCTL_SZ_4096;
3100	rctl |= E1000_RCTL_BSEX;
3101	switch (adapter->rx_buffer_len) {
3102	case 2048:
3103	default:
3104		rctl |= E1000_RCTL_SZ_2048;
3105		rctl &= ~E1000_RCTL_BSEX;
3106		break;
3107	case 4096:
3108		rctl |= E1000_RCTL_SZ_4096;
3109		break;
3110	case 8192:
3111		rctl |= E1000_RCTL_SZ_8192;
3112		break;
3113	case 16384:
3114		rctl |= E1000_RCTL_SZ_16384;
3115		break;
3116	}
3117
3118	/* Enable Extended Status in all Receive Descriptors */
3119	rfctl = er32(RFCTL);
3120	rfctl |= E1000_RFCTL_EXTEN;
3121	ew32(RFCTL, rfctl);
3122
3123	/* 82571 and greater support packet-split where the protocol
3124	 * header is placed in skb->data and the packet data is
3125	 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
3126	 * In the case of a non-split, skb->data is linearly filled,
3127	 * followed by the page buffers.  Therefore, skb->data is
3128	 * sized to hold the largest protocol header.
3129	 *
3130	 * allocations using alloc_page take too long for regular MTU
3131	 * so only enable packet split for jumbo frames
3132	 *
3133	 * Using pages when the page size is greater than 16k wastes
3134	 * a lot of memory, since we allocate 3 pages at all times
3135	 * per packet.
3136	 */
3137	pages = PAGE_USE_COUNT(adapter->netdev->mtu);
3138	if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
3139		adapter->rx_ps_pages = pages;
3140	else
3141		adapter->rx_ps_pages = 0;
3142
3143	if (adapter->rx_ps_pages) {
3144		u32 psrctl = 0;
3145
3146		/* Enable Packet split descriptors */
3147		rctl |= E1000_RCTL_DTYP_PS;
3148
3149		psrctl |= adapter->rx_ps_bsize0 >> E1000_PSRCTL_BSIZE0_SHIFT;
3150
3151		switch (adapter->rx_ps_pages) {
3152		case 3:
3153			psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE3_SHIFT;
3154			/* fall-through */
3155		case 2:
3156			psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE2_SHIFT;
3157			/* fall-through */
3158		case 1:
3159			psrctl |= PAGE_SIZE >> E1000_PSRCTL_BSIZE1_SHIFT;
3160			break;
3161		}
3162
3163		ew32(PSRCTL, psrctl);
3164	}
3165
3166	/* This is useful for sniffing bad packets. */
3167	if (adapter->netdev->features & NETIF_F_RXALL) {
3168		/* UPE and MPE will be handled by normal PROMISC logic
3169		 * in e1000e_set_rx_mode
3170		 */
3171		rctl |= (E1000_RCTL_SBP |	/* Receive bad packets */
3172			 E1000_RCTL_BAM |	/* RX All Bcast Pkts */
3173			 E1000_RCTL_PMCF);	/* RX All MAC Ctrl Pkts */
3174
3175		rctl &= ~(E1000_RCTL_VFE |	/* Disable VLAN filter */
3176			  E1000_RCTL_DPF |	/* Allow filtered pause */
3177			  E1000_RCTL_CFIEN);	/* Dis VLAN CFIEN Filter */
3178		/* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3179		 * and that breaks VLANs.
3180		 */
3181	}
3182
3183	ew32(RCTL, rctl);
3184	/* just started the receive unit, no need to restart */
3185	adapter->flags &= ~FLAG_RESTART_NOW;
3186}
3187
3188/**
3189 * e1000_configure_rx - Configure Receive Unit after Reset
3190 * @adapter: board private structure
3191 *
3192 * Configure the Rx unit of the MAC after a reset.
3193 **/
3194static void e1000_configure_rx(struct e1000_adapter *adapter)
3195{
3196	struct e1000_hw *hw = &adapter->hw;
3197	struct e1000_ring *rx_ring = adapter->rx_ring;
3198	u64 rdba;
3199	u32 rdlen, rctl, rxcsum, ctrl_ext;
3200
3201	if (adapter->rx_ps_pages) {
3202		/* this is a 32 byte descriptor */
3203		rdlen = rx_ring->count *
3204		    sizeof(union e1000_rx_desc_packet_split);
3205		adapter->clean_rx = e1000_clean_rx_irq_ps;
3206		adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3207	} else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3208		rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3209		adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3210		adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3211	} else {
3212		rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3213		adapter->clean_rx = e1000_clean_rx_irq;
3214		adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3215	}
3216
3217	/* disable receives while setting up the descriptors */
3218	rctl = er32(RCTL);
3219	if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3220		ew32(RCTL, rctl & ~E1000_RCTL_EN);
3221	e1e_flush();
3222	usleep_range(10000, 20000);
3223
3224	if (adapter->flags2 & FLAG2_DMA_BURST) {
3225		/* set the writeback threshold (only takes effect if the RDTR
3226		 * is set). set GRAN=1 and write back up to 0x4 worth, and
3227		 * enable prefetching of 0x20 Rx descriptors
3228		 * granularity = 01
3229		 * wthresh = 04,
3230		 * hthresh = 04,
3231		 * pthresh = 0x20
3232		 */
3233		ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3234		ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3235
3236		/* override the delay timers for enabling bursting, only if
3237		 * the value was not set by the user via module options
3238		 */
3239		if (adapter->rx_int_delay == DEFAULT_RDTR)
3240			adapter->rx_int_delay = BURST_RDTR;
3241		if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3242			adapter->rx_abs_int_delay = BURST_RADV;
3243	}
3244
3245	/* set the Receive Delay Timer Register */
3246	ew32(RDTR, adapter->rx_int_delay);
3247
3248	/* irq moderation */
3249	ew32(RADV, adapter->rx_abs_int_delay);
3250	if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3251		e1000e_write_itr(adapter, adapter->itr);
3252
3253	ctrl_ext = er32(CTRL_EXT);
3254	/* Auto-Mask interrupts upon ICR access */
3255	ctrl_ext |= E1000_CTRL_EXT_IAME;
3256	ew32(IAM, 0xffffffff);
3257	ew32(CTRL_EXT, ctrl_ext);
3258	e1e_flush();
3259
3260	/* Setup the HW Rx Head and Tail Descriptor Pointers and
3261	 * the Base and Length of the Rx Descriptor Ring
3262	 */
3263	rdba = rx_ring->dma;
3264	ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
3265	ew32(RDBAH(0), (rdba >> 32));
3266	ew32(RDLEN(0), rdlen);
3267	ew32(RDH(0), 0);
3268	ew32(RDT(0), 0);
3269	rx_ring->head = adapter->hw.hw_addr + E1000_RDH(0);
3270	rx_ring->tail = adapter->hw.hw_addr + E1000_RDT(0);
3271
3272	writel(0, rx_ring->head);
3273	if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
3274		e1000e_update_rdt_wa(rx_ring, 0);
3275	else
3276		writel(0, rx_ring->tail);
3277
3278	/* Enable Receive Checksum Offload for TCP and UDP */
3279	rxcsum = er32(RXCSUM);
3280	if (adapter->netdev->features & NETIF_F_RXCSUM)
3281		rxcsum |= E1000_RXCSUM_TUOFL;
3282	else
3283		rxcsum &= ~E1000_RXCSUM_TUOFL;
3284	ew32(RXCSUM, rxcsum);
3285
3286	/* With jumbo frames, excessive C-state transition latencies result
3287	 * in dropped transactions.
3288	 */
3289	if (adapter->netdev->mtu > ETH_DATA_LEN) {
3290		u32 lat =
3291		    ((er32(PBA) & E1000_PBA_RXA_MASK) * 1024 -
3292		     adapter->max_frame_size) * 8 / 1000;
3293
3294		if (adapter->flags & FLAG_IS_ICH) {
3295			u32 rxdctl = er32(RXDCTL(0));
3296
3297			ew32(RXDCTL(0), rxdctl | 0x3);
3298		}
3299
3300		pm_qos_update_request(&adapter->pm_qos_req, lat);
3301	} else {
3302		pm_qos_update_request(&adapter->pm_qos_req,
3303				      PM_QOS_DEFAULT_VALUE);
3304	}
3305
3306	/* Enable Receives */
3307	ew32(RCTL, rctl);
3308}
3309
3310/**
3311 * e1000e_write_mc_addr_list - write multicast addresses to MTA
3312 * @netdev: network interface device structure
3313 *
3314 * Writes multicast address list to the MTA hash table.
3315 * Returns: -ENOMEM on failure
3316 *                0 on no addresses written
3317 *                X on writing X addresses to MTA
3318 */
3319static int e1000e_write_mc_addr_list(struct net_device *netdev)
3320{
3321	struct e1000_adapter *adapter = netdev_priv(netdev);
3322	struct e1000_hw *hw = &adapter->hw;
3323	struct netdev_hw_addr *ha;
3324	u8 *mta_list;
3325	int i;
3326
3327	if (netdev_mc_empty(netdev)) {
3328		/* nothing to program, so clear mc list */
3329		hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3330		return 0;
3331	}
3332
3333	mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3334	if (!mta_list)
3335		return -ENOMEM;
3336
3337	/* update_mc_addr_list expects a packed array of only addresses. */
3338	i = 0;
3339	netdev_for_each_mc_addr(ha, netdev)
3340	    memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3341
3342	hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3343	kfree(mta_list);
3344
3345	return netdev_mc_count(netdev);
3346}
3347
3348/**
3349 * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3350 * @netdev: network interface device structure
3351 *
3352 * Writes unicast address list to the RAR table.
3353 * Returns: -ENOMEM on failure/insufficient address space
3354 *                0 on no addresses written
3355 *                X on writing X addresses to the RAR table
3356 **/
3357static int e1000e_write_uc_addr_list(struct net_device *netdev)
3358{
3359	struct e1000_adapter *adapter = netdev_priv(netdev);
3360	struct e1000_hw *hw = &adapter->hw;
3361	unsigned int rar_entries;
3362	int count = 0;
3363
3364	rar_entries = hw->mac.ops.rar_get_count(hw);
3365
3366	/* save a rar entry for our hardware address */
3367	rar_entries--;
3368
3369	/* save a rar entry for the LAA workaround */
3370	if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3371		rar_entries--;
3372
3373	/* return ENOMEM indicating insufficient memory for addresses */
3374	if (netdev_uc_count(netdev) > rar_entries)
3375		return -ENOMEM;
3376
3377	if (!netdev_uc_empty(netdev) && rar_entries) {
3378		struct netdev_hw_addr *ha;
3379
3380		/* write the addresses in reverse order to avoid write
3381		 * combining
3382		 */
3383		netdev_for_each_uc_addr(ha, netdev) {
3384			int rval;
3385
3386			if (!rar_entries)
3387				break;
3388			rval = hw->mac.ops.rar_set(hw, ha->addr, rar_entries--);
3389			if (rval < 0)
3390				return -ENOMEM;
3391			count++;
3392		}
3393	}
3394
3395	/* zero out the remaining RAR entries not used above */
3396	for (; rar_entries > 0; rar_entries--) {
3397		ew32(RAH(rar_entries), 0);
3398		ew32(RAL(rar_entries), 0);
3399	}
3400	e1e_flush();
3401
3402	return count;
3403}
3404
3405/**
3406 * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3407 * @netdev: network interface device structure
3408 *
3409 * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3410 * address list or the network interface flags are updated.  This routine is
3411 * responsible for configuring the hardware for proper unicast, multicast,
3412 * promiscuous mode, and all-multi behavior.
3413 **/
3414static void e1000e_set_rx_mode(struct net_device *netdev)
3415{
3416	struct e1000_adapter *adapter = netdev_priv(netdev);
3417	struct e1000_hw *hw = &adapter->hw;
3418	u32 rctl;
3419
3420	if (pm_runtime_suspended(netdev->dev.parent))
3421		return;
3422
3423	/* Check for Promiscuous and All Multicast modes */
3424	rctl = er32(RCTL);
3425
3426	/* clear the affected bits */
3427	rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3428
3429	if (netdev->flags & IFF_PROMISC) {
3430		rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3431		/* Do not hardware filter VLANs in promisc mode */
3432		e1000e_vlan_filter_disable(adapter);
3433	} else {
3434		int count;
3435
3436		if (netdev->flags & IFF_ALLMULTI) {
3437			rctl |= E1000_RCTL_MPE;
3438		} else {
3439			/* Write addresses to the MTA, if the attempt fails
3440			 * then we should just turn on promiscuous mode so
3441			 * that we can at least receive multicast traffic
3442			 */
3443			count = e1000e_write_mc_addr_list(netdev);
3444			if (count < 0)
3445				rctl |= E1000_RCTL_MPE;
3446		}
3447		e1000e_vlan_filter_enable(adapter);
3448		/* Write addresses to available RAR registers, if there is not
3449		 * sufficient space to store all the addresses then enable
3450		 * unicast promiscuous mode
3451		 */
3452		count = e1000e_write_uc_addr_list(netdev);
3453		if (count < 0)
3454			rctl |= E1000_RCTL_UPE;
3455	}
3456
3457	ew32(RCTL, rctl);
3458
3459	if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
3460		e1000e_vlan_strip_enable(adapter);
3461	else
3462		e1000e_vlan_strip_disable(adapter);
3463}
3464
3465static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3466{
3467	struct e1000_hw *hw = &adapter->hw;
3468	u32 mrqc, rxcsum;
3469	u32 rss_key[10];
3470	int i;
3471
3472	netdev_rss_key_fill(rss_key, sizeof(rss_key));
3473	for (i = 0; i < 10; i++)
3474		ew32(RSSRK(i), rss_key[i]);
3475
3476	/* Direct all traffic to queue 0 */
3477	for (i = 0; i < 32; i++)
3478		ew32(RETA(i), 0);
3479
3480	/* Disable raw packet checksumming so that RSS hash is placed in
3481	 * descriptor on writeback.
3482	 */
3483	rxcsum = er32(RXCSUM);
3484	rxcsum |= E1000_RXCSUM_PCSD;
3485
3486	ew32(RXCSUM, rxcsum);
3487
3488	mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3489		E1000_MRQC_RSS_FIELD_IPV4_TCP |
3490		E1000_MRQC_RSS_FIELD_IPV6 |
3491		E1000_MRQC_RSS_FIELD_IPV6_TCP |
3492		E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3493
3494	ew32(MRQC, mrqc);
3495}
3496
3497/**
3498 * e1000e_get_base_timinca - get default SYSTIM time increment attributes
3499 * @adapter: board private structure
3500 * @timinca: pointer to returned time increment attributes
3501 *
3502 * Get attributes for incrementing the System Time Register SYSTIML/H at
3503 * the default base frequency, and set the cyclecounter shift value.
3504 **/
3505s32 e1000e_get_base_timinca(struct e1000_adapter *adapter, u32 *timinca)
3506{
3507	struct e1000_hw *hw = &adapter->hw;
3508	u32 incvalue, incperiod, shift;
3509
3510	/* Make sure clock is enabled on I217/I218/I219  before checking
3511	 * the frequency
3512	 */
3513	if (((hw->mac.type == e1000_pch_lpt) ||
3514	     (hw->mac.type == e1000_pch_spt)) &&
3515	    !(er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) &&
3516	    !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_ENABLED)) {
3517		u32 fextnvm7 = er32(FEXTNVM7);
3518
3519		if (!(fextnvm7 & (1 << 0))) {
3520			ew32(FEXTNVM7, fextnvm7 | (1 << 0));
3521			e1e_flush();
3522		}
3523	}
3524
3525	switch (hw->mac.type) {
3526	case e1000_pch2lan:
3527	case e1000_pch_lpt:
3528		if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3529			/* Stable 96MHz frequency */
3530			incperiod = INCPERIOD_96MHz;
3531			incvalue = INCVALUE_96MHz;
3532			shift = INCVALUE_SHIFT_96MHz;
3533			adapter->cc.shift = shift + INCPERIOD_SHIFT_96MHz;
3534		} else {
3535			/* Stable 25MHz frequency */
3536			incperiod = INCPERIOD_25MHz;
3537			incvalue = INCVALUE_25MHz;
3538			shift = INCVALUE_SHIFT_25MHz;
3539			adapter->cc.shift = shift;
3540		}
3541		break;
3542	case e1000_pch_spt:
3543		if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3544			/* Stable 24MHz frequency */
3545			incperiod = INCPERIOD_24MHz;
3546			incvalue = INCVALUE_24MHz;
3547			shift = INCVALUE_SHIFT_24MHz;
3548			adapter->cc.shift = shift;
3549			break;
3550		}
3551		return -EINVAL;
3552	case e1000_82574:
3553	case e1000_82583:
3554		/* Stable 25MHz frequency */
3555		incperiod = INCPERIOD_25MHz;
3556		incvalue = INCVALUE_25MHz;
3557		shift = INCVALUE_SHIFT_25MHz;
3558		adapter->cc.shift = shift;
3559		break;
3560	default:
3561		return -EINVAL;
3562	}
3563
3564	*timinca = ((incperiod << E1000_TIMINCA_INCPERIOD_SHIFT) |
3565		    ((incvalue << shift) & E1000_TIMINCA_INCVALUE_MASK));
3566
3567	return 0;
3568}
3569
3570/**
3571 * e1000e_config_hwtstamp - configure the hwtstamp registers and enable/disable
3572 * @adapter: board private structure
3573 *
3574 * Outgoing time stamping can be enabled and disabled. Play nice and
3575 * disable it when requested, although it shouldn't cause any overhead
3576 * when no packet needs it. At most one packet in the queue may be
3577 * marked for time stamping, otherwise it would be impossible to tell
3578 * for sure to which packet the hardware time stamp belongs.
3579 *
3580 * Incoming time stamping has to be configured via the hardware filters.
3581 * Not all combinations are supported, in particular event type has to be
3582 * specified. Matching the kind of event packet is not supported, with the
3583 * exception of "all V2 events regardless of level 2 or 4".
3584 **/
3585static int e1000e_config_hwtstamp(struct e1000_adapter *adapter,
3586				  struct hwtstamp_config *config)
3587{
3588	struct e1000_hw *hw = &adapter->hw;
3589	u32 tsync_tx_ctl = E1000_TSYNCTXCTL_ENABLED;
3590	u32 tsync_rx_ctl = E1000_TSYNCRXCTL_ENABLED;
3591	u32 rxmtrl = 0;
3592	u16 rxudp = 0;
3593	bool is_l4 = false;
3594	bool is_l2 = false;
3595	u32 regval;
3596	s32 ret_val;
3597
3598	if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP))
3599		return -EINVAL;
3600
3601	/* flags reserved for future extensions - must be zero */
3602	if (config->flags)
3603		return -EINVAL;
3604
3605	switch (config->tx_type) {
3606	case HWTSTAMP_TX_OFF:
3607		tsync_tx_ctl = 0;
3608		break;
3609	case HWTSTAMP_TX_ON:
3610		break;
3611	default:
3612		return -ERANGE;
3613	}
3614
3615	switch (config->rx_filter) {
3616	case HWTSTAMP_FILTER_NONE:
3617		tsync_rx_ctl = 0;
3618		break;
3619	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3620		tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3621		rxmtrl = E1000_RXMTRL_PTP_V1_SYNC_MESSAGE;
3622		is_l4 = true;
3623		break;
3624	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3625		tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3626		rxmtrl = E1000_RXMTRL_PTP_V1_DELAY_REQ_MESSAGE;
3627		is_l4 = true;
3628		break;
3629	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3630		/* Also time stamps V2 L2 Path Delay Request/Response */
3631		tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3632		rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3633		is_l2 = true;
3634		break;
3635	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3636		/* Also time stamps V2 L2 Path Delay Request/Response. */
3637		tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3638		rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3639		is_l2 = true;
3640		break;
3641	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3642		/* Hardware cannot filter just V2 L4 Sync messages;
3643		 * fall-through to V2 (both L2 and L4) Sync.
3644		 */
3645	case HWTSTAMP_FILTER_PTP_V2_SYNC:
3646		/* Also time stamps V2 Path Delay Request/Response. */
3647		tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3648		rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3649		is_l2 = true;
3650		is_l4 = true;
3651		break;
3652	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3653		/* Hardware cannot filter just V2 L4 Delay Request messages;
3654		 * fall-through to V2 (both L2 and L4) Delay Request.
3655		 */
3656	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3657		/* Also time stamps V2 Path Delay Request/Response. */
3658		tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3659		rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3660		is_l2 = true;
3661		is_l4 = true;
3662		break;
3663	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3664	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3665		/* Hardware cannot filter just V2 L4 or L2 Event messages;
3666		 * fall-through to all V2 (both L2 and L4) Events.
3667		 */
3668	case HWTSTAMP_FILTER_PTP_V2_EVENT:
3669		tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_EVENT_V2;
3670		config->rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
3671		is_l2 = true;
3672		is_l4 = true;
3673		break;
3674	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3675		/* For V1, the hardware can only filter Sync messages or
3676		 * Delay Request messages but not both so fall-through to
3677		 * time stamp all packets.
3678		 */
3679	case HWTSTAMP_FILTER_ALL:
3680		is_l2 = true;
3681		is_l4 = true;
3682		tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_ALL;
3683		config->rx_filter = HWTSTAMP_FILTER_ALL;
3684		break;
3685	default:
3686		return -ERANGE;
3687	}
3688
3689	adapter->hwtstamp_config = *config;
3690
3691	/* enable/disable Tx h/w time stamping */
3692	regval = er32(TSYNCTXCTL);
3693	regval &= ~E1000_TSYNCTXCTL_ENABLED;
3694	regval |= tsync_tx_ctl;
3695	ew32(TSYNCTXCTL, regval);
3696	if ((er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) !=
3697	    (regval & E1000_TSYNCTXCTL_ENABLED)) {
3698		e_err("Timesync Tx Control register not set as expected\n");
3699		return -EAGAIN;
3700	}
3701
3702	/* enable/disable Rx h/w time stamping */
3703	regval = er32(TSYNCRXCTL);
3704	regval &= ~(E1000_TSYNCRXCTL_ENABLED | E1000_TSYNCRXCTL_TYPE_MASK);
3705	regval |= tsync_rx_ctl;
3706	ew32(TSYNCRXCTL, regval);
3707	if ((er32(TSYNCRXCTL) & (E1000_TSYNCRXCTL_ENABLED |
3708				 E1000_TSYNCRXCTL_TYPE_MASK)) !=
3709	    (regval & (E1000_TSYNCRXCTL_ENABLED |
3710		       E1000_TSYNCRXCTL_TYPE_MASK))) {
3711		e_err("Timesync Rx Control register not set as expected\n");
3712		return -EAGAIN;
3713	}
3714
3715	/* L2: define ethertype filter for time stamped packets */
3716	if (is_l2)
3717		rxmtrl |= ETH_P_1588;
3718
3719	/* define which PTP packets get time stamped */
3720	ew32(RXMTRL, rxmtrl);
3721
3722	/* Filter by destination port */
3723	if (is_l4) {
3724		rxudp = PTP_EV_PORT;
3725		cpu_to_be16s(&rxudp);
3726	}
3727	ew32(RXUDP, rxudp);
3728
3729	e1e_flush();
3730
3731	/* Clear TSYNCRXCTL_VALID & TSYNCTXCTL_VALID bit */
3732	er32(RXSTMPH);
3733	er32(TXSTMPH);
3734
3735	/* Get and set the System Time Register SYSTIM base frequency */
3736	ret_val = e1000e_get_base_timinca(adapter, &regval);
3737	if (ret_val)
3738		return ret_val;
3739	ew32(TIMINCA, regval);
3740
3741	/* reset the ns time counter */
3742	timecounter_init(&adapter->tc, &adapter->cc,
3743			 ktime_to_ns(ktime_get_real()));
3744
3745	return 0;
3746}
3747
3748/**
3749 * e1000_configure - configure the hardware for Rx and Tx
3750 * @adapter: private board structure
3751 **/
3752static void e1000_configure(struct e1000_adapter *adapter)
3753{
3754	struct e1000_ring *rx_ring = adapter->rx_ring;
3755
3756	e1000e_set_rx_mode(adapter->netdev);
3757
3758	e1000_restore_vlan(adapter);
3759	e1000_init_manageability_pt(adapter);
3760
3761	e1000_configure_tx(adapter);
3762
3763	if (adapter->netdev->features & NETIF_F_RXHASH)
3764		e1000e_setup_rss_hash(adapter);
3765	e1000_setup_rctl(adapter);
3766	e1000_configure_rx(adapter);
3767	adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3768}
3769
3770/**
3771 * e1000e_power_up_phy - restore link in case the phy was powered down
3772 * @adapter: address of board private structure
3773 *
3774 * The phy may be powered down to save power and turn off link when the
3775 * driver is unloaded and wake on lan is not enabled (among others)
3776 * *** this routine MUST be followed by a call to e1000e_reset ***
3777 **/
3778void e1000e_power_up_phy(struct e1000_adapter *adapter)
3779{
3780	if (adapter->hw.phy.ops.power_up)
3781		adapter->hw.phy.ops.power_up(&adapter->hw);
3782
3783	adapter->hw.mac.ops.setup_link(&adapter->hw);
3784}
3785
3786/**
3787 * e1000_power_down_phy - Power down the PHY
3788 *
3789 * Power down the PHY so no link is implied when interface is down.
3790 * The PHY cannot be powered down if management or WoL is active.
3791 */
3792static void e1000_power_down_phy(struct e1000_adapter *adapter)
3793{
3794	if (adapter->hw.phy.ops.power_down)
3795		adapter->hw.phy.ops.power_down(&adapter->hw);
3796}
3797
3798/**
3799 * e1000_flush_tx_ring - remove all descriptors from the tx_ring
3800 *
3801 * We want to clear all pending descriptors from the TX ring.
3802 * zeroing happens when the HW reads the regs. We  assign the ring itself as
3803 * the data of the next descriptor. We don't care about the data we are about
3804 * to reset the HW.
3805 */
3806static void e1000_flush_tx_ring(struct e1000_adapter *adapter)
3807{
3808	struct e1000_hw *hw = &adapter->hw;
3809	struct e1000_ring *tx_ring = adapter->tx_ring;
3810	struct e1000_tx_desc *tx_desc = NULL;
3811	u32 tdt, tctl, txd_lower = E1000_TXD_CMD_IFCS;
3812	u16 size = 512;
3813
3814	tctl = er32(TCTL);
3815	ew32(TCTL, tctl | E1000_TCTL_EN);
3816	tdt = er32(TDT(0));
3817	BUG_ON(tdt != tx_ring->next_to_use);
3818	tx_desc =  E1000_TX_DESC(*tx_ring, tx_ring->next_to_use);
3819	tx_desc->buffer_addr = tx_ring->dma;
3820
3821	tx_desc->lower.data = cpu_to_le32(txd_lower | size);
3822	tx_desc->upper.data = 0;
3823	/* flush descriptors to memory before notifying the HW */
3824	wmb();
3825	tx_ring->next_to_use++;
3826	if (tx_ring->next_to_use == tx_ring->count)
3827		tx_ring->next_to_use = 0;
3828	ew32(TDT(0), tx_ring->next_to_use);
3829	mmiowb();
3830	usleep_range(200, 250);
3831}
3832
3833/**
3834 * e1000_flush_rx_ring - remove all descriptors from the rx_ring
3835 *
3836 * Mark all descriptors in the RX ring as consumed and disable the rx ring
3837 */
3838static void e1000_flush_rx_ring(struct e1000_adapter *adapter)
3839{
3840	u32 rctl, rxdctl;
3841	struct e1000_hw *hw = &adapter->hw;
3842
3843	rctl = er32(RCTL);
3844	ew32(RCTL, rctl & ~E1000_RCTL_EN);
3845	e1e_flush();
3846	usleep_range(100, 150);
3847
3848	rxdctl = er32(RXDCTL(0));
3849	/* zero the lower 14 bits (prefetch and host thresholds) */
3850	rxdctl &= 0xffffc000;
3851
3852	/* update thresholds: prefetch threshold to 31, host threshold to 1
3853	 * and make sure the granularity is "descriptors" and not "cache lines"
3854	 */
3855	rxdctl |= (0x1F | (1 << 8) | E1000_RXDCTL_THRESH_UNIT_DESC);
3856
3857	ew32(RXDCTL(0), rxdctl);
3858	/* momentarily enable the RX ring for the changes to take effect */
3859	ew32(RCTL, rctl | E1000_RCTL_EN);
3860	e1e_flush();
3861	usleep_range(100, 150);
3862	ew32(RCTL, rctl & ~E1000_RCTL_EN);
3863}
3864
3865/**
3866 * e1000_flush_desc_rings - remove all descriptors from the descriptor rings
3867 *
3868 * In i219, the descriptor rings must be emptied before resetting the HW
3869 * or before changing the device state to D3 during runtime (runtime PM).
3870 *
3871 * Failure to do this will cause the HW to enter a unit hang state which can
3872 * only be released by PCI reset on the device
3873 *
3874 */
3875
3876static void e1000_flush_desc_rings(struct e1000_adapter *adapter)
3877{
3878	u16 hang_state;
3879	u32 fext_nvm11, tdlen;
3880	struct e1000_hw *hw = &adapter->hw;
3881
3882	/* First, disable MULR fix in FEXTNVM11 */
3883	fext_nvm11 = er32(FEXTNVM11);
3884	fext_nvm11 |= E1000_FEXTNVM11_DISABLE_MULR_FIX;
3885	ew32(FEXTNVM11, fext_nvm11);
3886	/* do nothing if we're not in faulty state, or if the queue is empty */
3887	tdlen = er32(TDLEN(0));
3888	pci_read_config_word(adapter->pdev, PCICFG_DESC_RING_STATUS,
3889			     &hang_state);
3890	if (!(hang_state & FLUSH_DESC_REQUIRED) || !tdlen)
3891		return;
3892	e1000_flush_tx_ring(adapter);
3893	/* recheck, maybe the fault is caused by the rx ring */
3894	pci_read_config_word(adapter->pdev, PCICFG_DESC_RING_STATUS,
3895			     &hang_state);
3896	if (hang_state & FLUSH_DESC_REQUIRED)
3897		e1000_flush_rx_ring(adapter);
3898}
3899
3900/**
3901 * e1000e_reset - bring the hardware into a known good state
3902 *
3903 * This function boots the hardware and enables some settings that
3904 * require a configuration cycle of the hardware - those cannot be
3905 * set/changed during runtime. After reset the device needs to be
3906 * properly configured for Rx, Tx etc.
3907 */
3908void e1000e_reset(struct e1000_adapter *adapter)
3909{
3910	struct e1000_mac_info *mac = &adapter->hw.mac;
3911	struct e1000_fc_info *fc = &adapter->hw.fc;
3912	struct e1000_hw *hw = &adapter->hw;
3913	u32 tx_space, min_tx_space, min_rx_space;
3914	u32 pba = adapter->pba;
3915	u16 hwm;
3916
3917	/* reset Packet Buffer Allocation to default */
3918	ew32(PBA, pba);
3919
3920	if (adapter->max_frame_size > (VLAN_ETH_FRAME_LEN + ETH_FCS_LEN)) {
3921		/* To maintain wire speed transmits, the Tx FIFO should be
3922		 * large enough to accommodate two full transmit packets,
3923		 * rounded up to the next 1KB and expressed in KB.  Likewise,
3924		 * the Rx FIFO should be large enough to accommodate at least
3925		 * one full receive packet and is similarly rounded up and
3926		 * expressed in KB.
3927		 */
3928		pba = er32(PBA);
3929		/* upper 16 bits has Tx packet buffer allocation size in KB */
3930		tx_space = pba >> 16;
3931		/* lower 16 bits has Rx packet buffer allocation size in KB */
3932		pba &= 0xffff;
3933		/* the Tx fifo also stores 16 bytes of information about the Tx
3934		 * but don't include ethernet FCS because hardware appends it
3935		 */
3936		min_tx_space = (adapter->max_frame_size +
3937				sizeof(struct e1000_tx_desc) - ETH_FCS_LEN) * 2;
3938		min_tx_space = ALIGN(min_tx_space, 1024);
3939		min_tx_space >>= 10;
3940		/* software strips receive CRC, so leave room for it */
3941		min_rx_space = adapter->max_frame_size;
3942		min_rx_space = ALIGN(min_rx_space, 1024);
3943		min_rx_space >>= 10;
3944
3945		/* If current Tx allocation is less than the min Tx FIFO size,
3946		 * and the min Tx FIFO size is less than the current Rx FIFO
3947		 * allocation, take space away from current Rx allocation
3948		 */
3949		if ((tx_space < min_tx_space) &&
3950		    ((min_tx_space - tx_space) < pba)) {
3951			pba -= min_tx_space - tx_space;
3952
3953			/* if short on Rx space, Rx wins and must trump Tx
3954			 * adjustment
3955			 */
3956			if (pba < min_rx_space)
3957				pba = min_rx_space;
3958		}
3959
3960		ew32(PBA, pba);
3961	}
3962
3963	/* flow control settings
3964	 *
3965	 * The high water mark must be low enough to fit one full frame
3966	 * (or the size used for early receive) above it in the Rx FIFO.
3967	 * Set it to the lower of:
3968	 * - 90% of the Rx FIFO size, and
3969	 * - the full Rx FIFO size minus one full frame
3970	 */
3971	if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3972		fc->pause_time = 0xFFFF;
3973	else
3974		fc->pause_time = E1000_FC_PAUSE_TIME;
3975	fc->send_xon = true;
3976	fc->current_mode = fc->requested_mode;
3977
3978	switch (hw->mac.type) {
3979	case e1000_ich9lan:
3980	case e1000_ich10lan:
3981		if (adapter->netdev->mtu > ETH_DATA_LEN) {
3982			pba = 14;
3983			ew32(PBA, pba);
3984			fc->high_water = 0x2800;
3985			fc->low_water = fc->high_water - 8;
3986			break;
3987		}
3988		/* fall-through */
3989	default:
3990		hwm = min(((pba << 10) * 9 / 10),
3991			  ((pba << 10) - adapter->max_frame_size));
3992
3993		fc->high_water = hwm & E1000_FCRTH_RTH;	/* 8-byte granularity */
3994		fc->low_water = fc->high_water - 8;
3995		break;
3996	case e1000_pchlan:
3997		/* Workaround PCH LOM adapter hangs with certain network
3998		 * loads.  If hangs persist, try disabling Tx flow control.
3999		 */
4000		if (adapter->netdev->mtu > ETH_DATA_LEN) {
4001			fc->high_water = 0x3500;
4002			fc->low_water = 0x1500;
4003		} else {
4004			fc->high_water = 0x5000;
4005			fc->low_water = 0x3000;
4006		}
4007		fc->refresh_time = 0x1000;
4008		break;
4009	case e1000_pch2lan:
4010	case e1000_pch_lpt:
4011	case e1000_pch_spt:
4012		fc->refresh_time = 0x0400;
4013
4014		if (adapter->netdev->mtu <= ETH_DATA_LEN) {
4015			fc->high_water = 0x05C20;
4016			fc->low_water = 0x05048;
4017			fc->pause_time = 0x0650;
4018			break;
4019		}
4020
4021		pba = 14;
4022		ew32(PBA, pba);
4023		fc->high_water = ((pba << 10) * 9 / 10) & E1000_FCRTH_RTH;
4024		fc->low_water = ((pba << 10) * 8 / 10) & E1000_FCRTL_RTL;
4025		break;
4026	}
4027
4028	/* Alignment of Tx data is on an arbitrary byte boundary with the
4029	 * maximum size per Tx descriptor limited only to the transmit
4030	 * allocation of the packet buffer minus 96 bytes with an upper
4031	 * limit of 24KB due to receive synchronization limitations.
4032	 */
4033	adapter->tx_fifo_limit = min_t(u32, ((er32(PBA) >> 16) << 10) - 96,
4034				       24 << 10);
4035
4036	/* Disable Adaptive Interrupt Moderation if 2 full packets cannot
4037	 * fit in receive buffer.
4038	 */
4039	if (adapter->itr_setting & 0x3) {
4040		if ((adapter->max_frame_size * 2) > (pba << 10)) {
4041			if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
4042				dev_info(&adapter->pdev->dev,
4043					 "Interrupt Throttle Rate off\n");
4044				adapter->flags2 |= FLAG2_DISABLE_AIM;
4045				e1000e_write_itr(adapter, 0);
4046			}
4047		} else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
4048			dev_info(&adapter->pdev->dev,
4049				 "Interrupt Throttle Rate on\n");
4050			adapter->flags2 &= ~FLAG2_DISABLE_AIM;
4051			adapter->itr = 20000;
4052			e1000e_write_itr(adapter, adapter->itr);
4053		}
4054	}
4055
4056	if (hw->mac.type == e1000_pch_spt)
4057		e1000_flush_desc_rings(adapter);
4058	/* Allow time for pending master requests to run */
4059	mac->ops.reset_hw(hw);
4060
4061	/* For parts with AMT enabled, let the firmware know
4062	 * that the network interface is in control
4063	 */
4064	if (adapter->flags & FLAG_HAS_AMT)
4065		e1000e_get_hw_control(adapter);
4066
4067	ew32(WUC, 0);
4068
4069	if (mac->ops.init_hw(hw))
4070		e_err("Hardware Error\n");
4071
4072	e1000_update_mng_vlan(adapter);
4073
4074	/* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
4075	ew32(VET, ETH_P_8021Q);
4076
4077	e1000e_reset_adaptive(hw);
4078
4079	/* initialize systim and reset the ns time counter */
4080	e1000e_config_hwtstamp(adapter, &adapter->hwtstamp_config);
4081
4082	/* Set EEE advertisement as appropriate */
4083	if (adapter->flags2 & FLAG2_HAS_EEE) {
4084		s32 ret_val;
4085		u16 adv_addr;
4086
4087		switch (hw->phy.type) {
4088		case e1000_phy_82579:
4089			adv_addr = I82579_EEE_ADVERTISEMENT;
4090			break;
4091		case e1000_phy_i217:
4092			adv_addr = I217_EEE_ADVERTISEMENT;
4093			break;
4094		default:
4095			dev_err(&adapter->pdev->dev,
4096				"Invalid PHY type setting EEE advertisement\n");
4097			return;
4098		}
4099
4100		ret_val = hw->phy.ops.acquire(hw);
4101		if (ret_val) {
4102			dev_err(&adapter->pdev->dev,
4103				"EEE advertisement - unable to acquire PHY\n");
4104			return;
4105		}
4106
4107		e1000_write_emi_reg_locked(hw, adv_addr,
4108					   hw->dev_spec.ich8lan.eee_disable ?
4109					   0 : adapter->eee_advert);
4110
4111		hw->phy.ops.release(hw);
4112	}
4113
4114	if (!netif_running(adapter->netdev) &&
4115	    !test_bit(__E1000_TESTING, &adapter->state))
4116		e1000_power_down_phy(adapter);
4117
4118	e1000_get_phy_info(hw);
4119
4120	if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
4121	    !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
4122		u16 phy_data = 0;
4123		/* speed up time to link by disabling smart power down, ignore
4124		 * the return value of this function because there is nothing
4125		 * different we would do if it failed
4126		 */
4127		e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
4128		phy_data &= ~IGP02E1000_PM_SPD;
4129		e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
4130	}
4131	if (hw->mac.type == e1000_pch_spt && adapter->int_mode == 0) {
4132		u32 reg;
4133
4134		/* Fextnvm7 @ 0xe4[2] = 1 */
4135		reg = er32(FEXTNVM7);
4136		reg |= E1000_FEXTNVM7_SIDE_CLK_UNGATE;
4137		ew32(FEXTNVM7, reg);
4138		/* Fextnvm9 @ 0x5bb4[13:12] = 11 */
4139		reg = er32(FEXTNVM9);
4140		reg |= E1000_FEXTNVM9_IOSFSB_CLKGATE_DIS |
4141		       E1000_FEXTNVM9_IOSFSB_CLKREQ_DIS;
4142		ew32(FEXTNVM9, reg);
4143	}
4144
4145}
4146
4147int e1000e_up(struct e1000_adapter *adapter)
4148{
4149	struct e1000_hw *hw = &adapter->hw;
4150
4151	/* hardware has been reset, we need to reload some things */
4152	e1000_configure(adapter);
4153
4154	clear_bit(__E1000_DOWN, &adapter->state);
4155
4156	if (adapter->msix_entries)
4157		e1000_configure_msix(adapter);
4158	e1000_irq_enable(adapter);
4159
4160	netif_start_queue(adapter->netdev);
4161
4162	/* fire a link change interrupt to start the watchdog */
4163	if (adapter->msix_entries)
4164		ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
4165	else
4166		ew32(ICS, E1000_ICS_LSC);
4167
4168	return 0;
4169}
4170
4171static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
4172{
4173	struct e1000_hw *hw = &adapter->hw;
4174
4175	if (!(adapter->flags2 & FLAG2_DMA_BURST))
4176		return;
4177
4178	/* flush pending descriptor writebacks to memory */
4179	ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4180	ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4181
4182	/* execute the writes immediately */
4183	e1e_flush();
4184
4185	/* due to rare timing issues, write to TIDV/RDTR again to ensure the
4186	 * write is successful
4187	 */
4188	ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4189	ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4190
4191	/* execute the writes immediately */
4192	e1e_flush();
4193}
4194
4195static void e1000e_update_stats(struct e1000_adapter *adapter);
4196
4197/**
4198 * e1000e_down - quiesce the device and optionally reset the hardware
4199 * @adapter: board private structure
4200 * @reset: boolean flag to reset the hardware or not
4201 */
4202void e1000e_down(struct e1000_adapter *adapter, bool reset)
4203{
4204	struct net_device *netdev = adapter->netdev;
4205	struct e1000_hw *hw = &adapter->hw;
4206	u32 tctl, rctl;
4207
4208	/* signal that we're down so the interrupt handler does not
4209	 * reschedule our watchdog timer
4210	 */
4211	set_bit(__E1000_DOWN, &adapter->state);
4212
4213	netif_carrier_off(netdev);
4214
4215	/* disable receives in the hardware */
4216	rctl = er32(RCTL);
4217	if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
4218		ew32(RCTL, rctl & ~E1000_RCTL_EN);
4219	/* flush and sleep below */
4220
4221	netif_stop_queue(netdev);
4222
4223	/* disable transmits in the hardware */
4224	tctl = er32(TCTL);
4225	tctl &= ~E1000_TCTL_EN;
4226	ew32(TCTL, tctl);
4227
4228	/* flush both disables and wait for them to finish */
4229	e1e_flush();
4230	usleep_range(10000, 20000);
4231
4232	e1000_irq_disable(adapter);
4233
4234	napi_synchronize(&adapter->napi);
4235
4236	del_timer_sync(&adapter->watchdog_timer);
4237	del_timer_sync(&adapter->phy_info_timer);
4238
4239	spin_lock(&adapter->stats64_lock);
4240	e1000e_update_stats(adapter);
4241	spin_unlock(&adapter->stats64_lock);
4242
4243	e1000e_flush_descriptors(adapter);
4244
4245	adapter->link_speed = 0;
4246	adapter->link_duplex = 0;
4247
4248	/* Disable Si errata workaround on PCHx for jumbo frame flow */
4249	if ((hw->mac.type >= e1000_pch2lan) &&
4250	    (adapter->netdev->mtu > ETH_DATA_LEN) &&
4251	    e1000_lv_jumbo_workaround_ich8lan(hw, false))
4252		e_dbg("failed to disable jumbo frame workaround mode\n");
4253
4254	if (!pci_channel_offline(adapter->pdev)) {
4255		if (reset)
4256			e1000e_reset(adapter);
4257		else if (hw->mac.type == e1000_pch_spt)
4258			e1000_flush_desc_rings(adapter);
4259	}
4260	e1000_clean_tx_ring(adapter->tx_ring);
4261	e1000_clean_rx_ring(adapter->rx_ring);
4262}
4263
4264void e1000e_reinit_locked(struct e1000_adapter *adapter)
4265{
4266	might_sleep();
4267	while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4268		usleep_range(1000, 2000);
4269	e1000e_down(adapter, true);
4270	e1000e_up(adapter);
4271	clear_bit(__E1000_RESETTING, &adapter->state);
4272}
4273
4274/**
4275 * e1000e_cyclecounter_read - read raw cycle counter (used by time counter)
4276 * @cc: cyclecounter structure
4277 **/
4278static cycle_t e1000e_cyclecounter_read(const struct cyclecounter *cc)
4279{
4280	struct e1000_adapter *adapter = container_of(cc, struct e1000_adapter,
4281						     cc);
4282	struct e1000_hw *hw = &adapter->hw;
4283	u32 systimel_1, systimel_2, systimeh;
4284	cycle_t systim, systim_next;
4285	/* SYSTIMH latching upon SYSTIML read does not work well.
4286	 * This means that if SYSTIML overflows after we read it but before
4287	 * we read SYSTIMH, the value of SYSTIMH has been incremented and we
4288	 * will experience a huge non linear increment in the systime value
4289	 * to fix that we test for overflow and if true, we re-read systime.
4290	 */
4291	systimel_1 = er32(SYSTIML);
4292	systimeh = er32(SYSTIMH);
4293	systimel_2 = er32(SYSTIML);
4294	/* Check for overflow. If there was no overflow, use the values */
4295	if (systimel_1 < systimel_2) {
4296		systim = (cycle_t)systimel_1;
4297		systim |= (cycle_t)systimeh << 32;
4298	} else {
4299		/* There was an overflow, read again SYSTIMH, and use
4300		 * systimel_2
4301		 */
4302		systimeh = er32(SYSTIMH);
4303		systim = (cycle_t)systimel_2;
4304		systim |= (cycle_t)systimeh << 32;
4305	}
4306
4307	if ((hw->mac.type == e1000_82574) || (hw->mac.type == e1000_82583)) {
4308		u64 incvalue, time_delta, rem, temp;
4309		int i;
4310
4311		/* errata for 82574/82583 possible bad bits read from SYSTIMH/L
4312		 * check to see that the time is incrementing at a reasonable
4313		 * rate and is a multiple of incvalue
4314		 */
4315		incvalue = er32(TIMINCA) & E1000_TIMINCA_INCVALUE_MASK;
4316		for (i = 0; i < E1000_MAX_82574_SYSTIM_REREADS; i++) {
4317			/* latch SYSTIMH on read of SYSTIML */
4318			systim_next = (cycle_t)er32(SYSTIML);
4319			systim_next |= (cycle_t)er32(SYSTIMH) << 32;
4320
4321			time_delta = systim_next - systim;
4322			temp = time_delta;
4323			rem = do_div(temp, incvalue);
4324
4325			systim = systim_next;
4326
4327			if ((time_delta < E1000_82574_SYSTIM_EPSILON) &&
4328			    (rem == 0))
4329				break;
4330		}
4331	}
4332	return systim;
4333}
4334
4335/**
4336 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
4337 * @adapter: board private structure to initialize
4338 *
4339 * e1000_sw_init initializes the Adapter private data structure.
4340 * Fields are initialized based on PCI device information and
4341 * OS network device settings (MTU size).
4342 **/
4343static int e1000_sw_init(struct e1000_adapter *adapter)
4344{
4345	struct net_device *netdev = adapter->netdev;
4346
4347	adapter->rx_buffer_len = VLAN_ETH_FRAME_LEN + ETH_FCS_LEN;
4348	adapter->rx_ps_bsize0 = 128;
4349	adapter->max_frame_size = netdev->mtu + VLAN_ETH_HLEN + ETH_FCS_LEN;
4350	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
4351	adapter->tx_ring_count = E1000_DEFAULT_TXD;
4352	adapter->rx_ring_count = E1000_DEFAULT_RXD;
4353
4354	spin_lock_init(&adapter->stats64_lock);
4355
4356	e1000e_set_interrupt_capability(adapter);
4357
4358	if (e1000_alloc_queues(adapter))
4359		return -ENOMEM;
4360
4361	/* Setup hardware time stamping cyclecounter */
4362	if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
4363		adapter->cc.read = e1000e_cyclecounter_read;
4364		adapter->cc.mask = CYCLECOUNTER_MASK(64);
4365		adapter->cc.mult = 1;
4366		/* cc.shift set in e1000e_get_base_tininca() */
4367
4368		spin_lock_init(&adapter->systim_lock);
4369		INIT_WORK(&adapter->tx_hwtstamp_work, e1000e_tx_hwtstamp_work);
4370	}
4371
4372	/* Explicitly disable IRQ since the NIC can be in any state. */
4373	e1000_irq_disable(adapter);
4374
4375	set_bit(__E1000_DOWN, &adapter->state);
4376	return 0;
4377}
4378
4379/**
4380 * e1000_intr_msi_test - Interrupt Handler
4381 * @irq: interrupt number
4382 * @data: pointer to a network interface device structure
4383 **/
4384static irqreturn_t e1000_intr_msi_test(int __always_unused irq, void *data)
4385{
4386	struct net_device *netdev = data;
4387	struct e1000_adapter *adapter = netdev_priv(netdev);
4388	struct e1000_hw *hw = &adapter->hw;
4389	u32 icr = er32(ICR);
4390
4391	e_dbg("icr is %08X\n", icr);
4392	if (icr & E1000_ICR_RXSEQ) {
4393		adapter->flags &= ~FLAG_MSI_TEST_FAILED;
4394		/* Force memory writes to complete before acknowledging the
4395		 * interrupt is handled.
4396		 */
4397		wmb();
4398	}
4399
4400	return IRQ_HANDLED;
4401}
4402
4403/**
4404 * e1000_test_msi_interrupt - Returns 0 for successful test
4405 * @adapter: board private struct
4406 *
4407 * code flow taken from tg3.c
4408 **/
4409static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
4410{
4411	struct net_device *netdev = adapter->netdev;
4412	struct e1000_hw *hw = &adapter->hw;
4413	int err;
4414
4415	/* poll_enable hasn't been called yet, so don't need disable */
4416	/* clear any pending events */
4417	er32(ICR);
4418
4419	/* free the real vector and request a test handler */
4420	e1000_free_irq(adapter);
4421	e1000e_reset_interrupt_capability(adapter);
4422
4423	/* Assume that the test fails, if it succeeds then the test
4424	 * MSI irq handler will unset this flag
4425	 */
4426	adapter->flags |= FLAG_MSI_TEST_FAILED;
4427
4428	err = pci_enable_msi(adapter->pdev);
4429	if (err)
4430		goto msi_test_failed;
4431
4432	err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
4433			  netdev->name, netdev);
4434	if (err) {
4435		pci_disable_msi(adapter->pdev);
4436		goto msi_test_failed;
4437	}
4438
4439	/* Force memory writes to complete before enabling and firing an
4440	 * interrupt.
4441	 */
4442	wmb();
4443
4444	e1000_irq_enable(adapter);
4445
4446	/* fire an unusual interrupt on the test handler */
4447	ew32(ICS, E1000_ICS_RXSEQ);
4448	e1e_flush();
4449	msleep(100);
4450
4451	e1000_irq_disable(adapter);
4452
4453	rmb();			/* read flags after interrupt has been fired */
4454
4455	if (adapter->flags & FLAG_MSI_TEST_FAILED) {
4456		adapter->int_mode = E1000E_INT_MODE_LEGACY;
4457		e_info("MSI interrupt test failed, using legacy interrupt.\n");
4458	} else {
4459		e_dbg("MSI interrupt test succeeded!\n");
4460	}
4461
4462	free_irq(adapter->pdev->irq, netdev);
4463	pci_disable_msi(adapter->pdev);
4464
4465msi_test_failed:
4466	e1000e_set_interrupt_capability(adapter);
4467	return e1000_request_irq(adapter);
4468}
4469
4470/**
4471 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
4472 * @adapter: board private struct
4473 *
4474 * code flow taken from tg3.c, called with e1000 interrupts disabled.
4475 **/
4476static int e1000_test_msi(struct e1000_adapter *adapter)
4477{
4478	int err;
4479	u16 pci_cmd;
4480
4481	if (!(adapter->flags & FLAG_MSI_ENABLED))
4482		return 0;
4483
4484	/* disable SERR in case the MSI write causes a master abort */
4485	pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4486	if (pci_cmd & PCI_COMMAND_SERR)
4487		pci_write_config_word(adapter->pdev, PCI_COMMAND,
4488				      pci_cmd & ~PCI_COMMAND_SERR);
4489
4490	err = e1000_test_msi_interrupt(adapter);
4491
4492	/* re-enable SERR */
4493	if (pci_cmd & PCI_COMMAND_SERR) {
4494		pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4495		pci_cmd |= PCI_COMMAND_SERR;
4496		pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
4497	}
4498
4499	return err;
4500}
4501
4502/**
4503 * e1000_open - Called when a network interface is made active
4504 * @netdev: network interface device structure
4505 *
4506 * Returns 0 on success, negative value on failure
4507 *
4508 * The open entry point is called when a network interface is made
4509 * active by the system (IFF_UP).  At this point all resources needed
4510 * for transmit and receive operations are allocated, the interrupt
4511 * handler is registered with the OS, the watchdog timer is started,
4512 * and the stack is notified that the interface is ready.
4513 **/
4514static int e1000_open(struct net_device *netdev)
4515{
4516	struct e1000_adapter *adapter = netdev_priv(netdev);
4517	struct e1000_hw *hw = &adapter->hw;
4518	struct pci_dev *pdev = adapter->pdev;
4519	int err;
4520
4521	/* disallow open during test */
4522	if (test_bit(__E1000_TESTING, &adapter->state))
4523		return -EBUSY;
4524
4525	pm_runtime_get_sync(&pdev->dev);
4526
4527	netif_carrier_off(netdev);
4528
4529	/* allocate transmit descriptors */
4530	err = e1000e_setup_tx_resources(adapter->tx_ring);
4531	if (err)
4532		goto err_setup_tx;
4533
4534	/* allocate receive descriptors */
4535	err = e1000e_setup_rx_resources(adapter->rx_ring);
4536	if (err)
4537		goto err_setup_rx;
4538
4539	/* If AMT is enabled, let the firmware know that the network
4540	 * interface is now open and reset the part to a known state.
4541	 */
4542	if (adapter->flags & FLAG_HAS_AMT) {
4543		e1000e_get_hw_control(adapter);
4544		e1000e_reset(adapter);
4545	}
4546
4547	e1000e_power_up_phy(adapter);
4548
4549	adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4550	if ((adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
4551		e1000_update_mng_vlan(adapter);
4552
4553	/* DMA latency requirement to workaround jumbo issue */
4554	pm_qos_add_request(&adapter->pm_qos_req, PM_QOS_CPU_DMA_LATENCY,
4555			   PM_QOS_DEFAULT_VALUE);
4556
4557	/* before we allocate an interrupt, we must be ready to handle it.
4558	 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
4559	 * as soon as we call pci_request_irq, so we have to setup our
4560	 * clean_rx handler before we do so.
4561	 */
4562	e1000_configure(adapter);
4563
4564	err = e1000_request_irq(adapter);
4565	if (err)
4566		goto err_req_irq;
4567
4568	/* Work around PCIe errata with MSI interrupts causing some chipsets to
4569	 * ignore e1000e MSI messages, which means we need to test our MSI
4570	 * interrupt now
4571	 */
4572	if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
4573		err = e1000_test_msi(adapter);
4574		if (err) {
4575			e_err("Interrupt allocation failed\n");
4576			goto err_req_irq;
4577		}
4578	}
4579
4580	/* From here on the code is the same as e1000e_up() */
4581	clear_bit(__E1000_DOWN, &adapter->state);
4582
4583	napi_enable(&adapter->napi);
4584
4585	e1000_irq_enable(adapter);
4586
4587	adapter->tx_hang_recheck = false;
4588	netif_start_queue(netdev);
4589
4590	hw->mac.get_link_status = true;
4591	pm_runtime_put(&pdev->dev);
4592
4593	/* fire a link status change interrupt to start the watchdog */
4594	if (adapter->msix_entries)
4595		ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
4596	else
4597		ew32(ICS, E1000_ICS_LSC);
4598
4599	return 0;
4600
4601err_req_irq:
4602	pm_qos_remove_request(&adapter->pm_qos_req);
4603	e1000e_release_hw_control(adapter);
4604	e1000_power_down_phy(adapter);
4605	e1000e_free_rx_resources(adapter->rx_ring);
4606err_setup_rx:
4607	e1000e_free_tx_resources(adapter->tx_ring);
4608err_setup_tx:
4609	e1000e_reset(adapter);
4610	pm_runtime_put_sync(&pdev->dev);
4611
4612	return err;
4613}
4614
4615/**
4616 * e1000_close - Disables a network interface
4617 * @netdev: network interface device structure
4618 *
4619 * Returns 0, this is not allowed to fail
4620 *
4621 * The close entry point is called when an interface is de-activated
4622 * by the OS.  The hardware is still under the drivers control, but
4623 * needs to be disabled.  A global MAC reset is issued to stop the
4624 * hardware, and all transmit and receive resources are freed.
4625 **/
4626static int e1000_close(struct net_device *netdev)
4627{
4628	struct e1000_adapter *adapter = netdev_priv(netdev);
4629	struct pci_dev *pdev = adapter->pdev;
4630	int count = E1000_CHECK_RESET_COUNT;
4631
4632	while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
4633		usleep_range(10000, 20000);
4634
4635	WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4636
4637	pm_runtime_get_sync(&pdev->dev);
4638
4639	if (!test_bit(__E1000_DOWN, &adapter->state)) {
4640		e1000e_down(adapter, true);
4641		e1000_free_irq(adapter);
4642
4643		/* Link status message must follow this format */
4644		pr_info("%s NIC Link is Down\n", adapter->netdev->name);
4645	}
4646
4647	napi_disable(&adapter->napi);
4648
4649	e1000e_free_tx_resources(adapter->tx_ring);
4650	e1000e_free_rx_resources(adapter->rx_ring);
4651
4652	/* kill manageability vlan ID if supported, but not if a vlan with
4653	 * the same ID is registered on the host OS (let 8021q kill it)
4654	 */
4655	if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
4656		e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
4657				       adapter->mng_vlan_id);
4658
4659	/* If AMT is enabled, let the firmware know that the network
4660	 * interface is now closed
4661	 */
4662	if ((adapter->flags & FLAG_HAS_AMT) &&
4663	    !test_bit(__E1000_TESTING, &adapter->state))
4664		e1000e_release_hw_control(adapter);
4665
4666	pm_qos_remove_request(&adapter->pm_qos_req);
4667
4668	pm_runtime_put_sync(&pdev->dev);
4669
4670	return 0;
4671}
4672
4673/**
4674 * e1000_set_mac - Change the Ethernet Address of the NIC
4675 * @netdev: network interface device structure
4676 * @p: pointer to an address structure
4677 *
4678 * Returns 0 on success, negative on failure
4679 **/
4680static int e1000_set_mac(struct net_device *netdev, void *p)
4681{
4682	struct e1000_adapter *adapter = netdev_priv(netdev);
4683	struct e1000_hw *hw = &adapter->hw;
4684	struct sockaddr *addr = p;
4685
4686	if (!is_valid_ether_addr(addr->sa_data))
4687		return -EADDRNOTAVAIL;
4688
4689	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4690	memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
4691
4692	hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
4693
4694	if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
4695		/* activate the work around */
4696		e1000e_set_laa_state_82571(&adapter->hw, 1);
4697
4698		/* Hold a copy of the LAA in RAR[14] This is done so that
4699		 * between the time RAR[0] gets clobbered  and the time it
4700		 * gets fixed (in e1000_watchdog), the actual LAA is in one
4701		 * of the RARs and no incoming packets directed to this port
4702		 * are dropped. Eventually the LAA will be in RAR[0] and
4703		 * RAR[14]
4704		 */
4705		hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr,
4706				    adapter->hw.mac.rar_entry_count - 1);
4707	}
4708
4709	return 0;
4710}
4711
4712/**
4713 * e1000e_update_phy_task - work thread to update phy
4714 * @work: pointer to our work struct
4715 *
4716 * this worker thread exists because we must acquire a
4717 * semaphore to read the phy, which we could msleep while
4718 * waiting for it, and we can't msleep in a timer.
4719 **/
4720static void e1000e_update_phy_task(struct work_struct *work)
4721{
4722	struct e1000_adapter *adapter = container_of(work,
4723						     struct e1000_adapter,
4724						     update_phy_task);
4725	struct e1000_hw *hw = &adapter->hw;
4726
4727	if (test_bit(__E1000_DOWN, &adapter->state))
4728		return;
4729
4730	e1000_get_phy_info(hw);
4731
4732	/* Enable EEE on 82579 after link up */
4733	if (hw->phy.type >= e1000_phy_82579)
4734		e1000_set_eee_pchlan(hw);
4735}
4736
4737/**
4738 * e1000_update_phy_info - timre call-back to update PHY info
4739 * @data: pointer to adapter cast into an unsigned long
4740 *
4741 * Need to wait a few seconds after link up to get diagnostic information from
4742 * the phy
4743 **/
4744static void e1000_update_phy_info(unsigned long data)
4745{
4746	struct e1000_adapter *adapter = (struct e1000_adapter *)data;
4747
4748	if (test_bit(__E1000_DOWN, &adapter->state))
4749		return;
4750
4751	schedule_work(&adapter->update_phy_task);
4752}
4753
4754/**
4755 * e1000e_update_phy_stats - Update the PHY statistics counters
4756 * @adapter: board private structure
4757 *
4758 * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4759 **/
4760static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4761{
4762	struct e1000_hw *hw = &adapter->hw;
4763	s32 ret_val;
4764	u16 phy_data;
4765
4766	ret_val = hw->phy.ops.acquire(hw);
4767	if (ret_val)
4768		return;
4769
4770	/* A page set is expensive so check if already on desired page.
4771	 * If not, set to the page with the PHY status registers.
4772	 */
4773	hw->phy.addr = 1;
4774	ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4775					   &phy_data);
4776	if (ret_val)
4777		goto release;
4778	if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4779		ret_val = hw->phy.ops.set_page(hw,
4780					       HV_STATS_PAGE << IGP_PAGE_SHIFT);
4781		if (ret_val)
4782			goto release;
4783	}
4784
4785	/* Single Collision Count */
4786	hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4787	ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4788	if (!ret_val)
4789		adapter->stats.scc += phy_data;
4790
4791	/* Excessive Collision Count */
4792	hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4793	ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4794	if (!ret_val)
4795		adapter->stats.ecol += phy_data;
4796
4797	/* Multiple Collision Count */
4798	hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4799	ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4800	if (!ret_val)
4801		adapter->stats.mcc += phy_data;
4802
4803	/* Late Collision Count */
4804	hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4805	ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4806	if (!ret_val)
4807		adapter->stats.latecol += phy_data;
4808
4809	/* Collision Count - also used for adaptive IFS */
4810	hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4811	ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4812	if (!ret_val)
4813		hw->mac.collision_delta = phy_data;
4814
4815	/* Defer Count */
4816	hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4817	ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4818	if (!ret_val)
4819		adapter->stats.dc += phy_data;
4820
4821	/* Transmit with no CRS */
4822	hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4823	ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4824	if (!ret_val)
4825		adapter->stats.tncrs += phy_data;
4826
4827release:
4828	hw->phy.ops.release(hw);
4829}
4830
4831/**
4832 * e1000e_update_stats - Update the board statistics counters
4833 * @adapter: board private structure
4834 **/
4835static void e1000e_update_stats(struct e1000_adapter *adapter)
4836{
4837	struct net_device *netdev = adapter->netdev;
4838	struct e1000_hw *hw = &adapter->hw;
4839	struct pci_dev *pdev = adapter->pdev;
4840
4841	/* Prevent stats update while adapter is being reset, or if the pci
4842	 * connection is down.
4843	 */
4844	if (adapter->link_speed == 0)
4845		return;
4846	if (pci_channel_offline(pdev))
4847		return;
4848
4849	adapter->stats.crcerrs += er32(CRCERRS);
4850	adapter->stats.gprc += er32(GPRC);
4851	adapter->stats.gorc += er32(GORCL);
4852	er32(GORCH);		/* Clear gorc */
4853	adapter->stats.bprc += er32(BPRC);
4854	adapter->stats.mprc += er32(MPRC);
4855	adapter->stats.roc += er32(ROC);
4856
4857	adapter->stats.mpc += er32(MPC);
4858
4859	/* Half-duplex statistics */
4860	if (adapter->link_duplex == HALF_DUPLEX) {
4861		if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4862			e1000e_update_phy_stats(adapter);
4863		} else {
4864			adapter->stats.scc += er32(SCC);
4865			adapter->stats.ecol += er32(ECOL);
4866			adapter->stats.mcc += er32(MCC);
4867			adapter->stats.latecol += er32(LATECOL);
4868			adapter->stats.dc += er32(DC);
4869
4870			hw->mac.collision_delta = er32(COLC);
4871
4872			if ((hw->mac.type != e1000_82574) &&
4873			    (hw->mac.type != e1000_82583))
4874				adapter->stats.tncrs += er32(TNCRS);
4875		}
4876		adapter->stats.colc += hw->mac.collision_delta;
4877	}
4878
4879	adapter->stats.xonrxc += er32(XONRXC);
4880	adapter->stats.xontxc += er32(XONTXC);
4881	adapter->stats.xoffrxc += er32(XOFFRXC);
4882	adapter->stats.xofftxc += er32(XOFFTXC);
4883	adapter->stats.gptc += er32(GPTC);
4884	adapter->stats.gotc += er32(GOTCL);
4885	er32(GOTCH);		/* Clear gotc */
4886	adapter->stats.rnbc += er32(RNBC);
4887	adapter->stats.ruc += er32(RUC);
4888
4889	adapter->stats.mptc += er32(MPTC);
4890	adapter->stats.bptc += er32(BPTC);
4891
4892	/* used for adaptive IFS */
4893
4894	hw->mac.tx_packet_delta = er32(TPT);
4895	adapter->stats.tpt += hw->mac.tx_packet_delta;
4896
4897	adapter->stats.algnerrc += er32(ALGNERRC);
4898	adapter->stats.rxerrc += er32(RXERRC);
4899	adapter->stats.cexterr += er32(CEXTERR);
4900	adapter->stats.tsctc += er32(TSCTC);
4901	adapter->stats.tsctfc += er32(TSCTFC);
4902
4903	/* Fill out the OS statistics structure */
4904	netdev->stats.multicast = adapter->stats.mprc;
4905	netdev->stats.collisions = adapter->stats.colc;
4906
4907	/* Rx Errors */
4908
4909	/* RLEC on some newer hardware can be incorrect so build
4910	 * our own version based on RUC and ROC
4911	 */
4912	netdev->stats.rx_errors = adapter->stats.rxerrc +
4913	    adapter->stats.crcerrs + adapter->stats.algnerrc +
4914	    adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
4915	netdev->stats.rx_length_errors = adapter->stats.ruc +
4916	    adapter->stats.roc;
4917	netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4918	netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4919	netdev->stats.rx_missed_errors = adapter->stats.mpc;
4920
4921	/* Tx Errors */
4922	netdev->stats.tx_errors = adapter->stats.ecol + adapter->stats.latecol;
4923	netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4924	netdev->stats.tx_window_errors = adapter->stats.latecol;
4925	netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4926
4927	/* Tx Dropped needs to be maintained elsewhere */
4928
4929	/* Management Stats */
4930	adapter->stats.mgptc += er32(MGTPTC);
4931	adapter->stats.mgprc += er32(MGTPRC);
4932	adapter->stats.mgpdc += er32(MGTPDC);
4933
4934	/* Correctable ECC Errors */
4935	if ((hw->mac.type == e1000_pch_lpt) ||
4936	    (hw->mac.type == e1000_pch_spt)) {
4937		u32 pbeccsts = er32(PBECCSTS);
4938
4939		adapter->corr_errors +=
4940		    pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
4941		adapter->uncorr_errors +=
4942		    (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
4943		    E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
4944	}
4945}
4946
4947/**
4948 * e1000_phy_read_status - Update the PHY register status snapshot
4949 * @adapter: board private structure
4950 **/
4951static void e1000_phy_read_status(struct e1000_adapter *adapter)
4952{
4953	struct e1000_hw *hw = &adapter->hw;
4954	struct e1000_phy_regs *phy = &adapter->phy_regs;
4955
4956	if (!pm_runtime_suspended((&adapter->pdev->dev)->parent) &&
4957	    (er32(STATUS) & E1000_STATUS_LU) &&
4958	    (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4959		int ret_val;
4960
4961		ret_val = e1e_rphy(hw, MII_BMCR, &phy->bmcr);
4962		ret_val |= e1e_rphy(hw, MII_BMSR, &phy->bmsr);
4963		ret_val |= e1e_rphy(hw, MII_ADVERTISE, &phy->advertise);
4964		ret_val |= e1e_rphy(hw, MII_LPA, &phy->lpa);
4965		ret_val |= e1e_rphy(hw, MII_EXPANSION, &phy->expansion);
4966		ret_val |= e1e_rphy(hw, MII_CTRL1000, &phy->ctrl1000);
4967		ret_val |= e1e_rphy(hw, MII_STAT1000, &phy->stat1000);
4968		ret_val |= e1e_rphy(hw, MII_ESTATUS, &phy->estatus);
4969		if (ret_val)
4970			e_warn("Error reading PHY register\n");
4971	} else {
4972		/* Do not read PHY registers if link is not up
4973		 * Set values to typical power-on defaults
4974		 */
4975		phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4976		phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4977			     BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4978			     BMSR_ERCAP);
4979		phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4980				  ADVERTISE_ALL | ADVERTISE_CSMA);
4981		phy->lpa = 0;
4982		phy->expansion = EXPANSION_ENABLENPAGE;
4983		phy->ctrl1000 = ADVERTISE_1000FULL;
4984		phy->stat1000 = 0;
4985		phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4986	}
4987}
4988
4989static void e1000_print_link_info(struct e1000_adapter *adapter)
4990{
4991	struct e1000_hw *hw = &adapter->hw;
4992	u32 ctrl = er32(CTRL);
4993
4994	/* Link status message must follow this format for user tools */
4995	pr_info("%s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4996		adapter->netdev->name, adapter->link_speed,
4997		adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
4998		(ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
4999		(ctrl & E1000_CTRL_RFCE) ? "Rx" :
5000		(ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
5001}
5002
5003static bool e1000e_has_link(struct e1000_adapter *adapter)
5004{
5005	struct e1000_hw *hw = &adapter->hw;
5006	bool link_active = false;
5007	s32 ret_val = 0;
5008
5009	/* get_link_status is set on LSC (link status) interrupt or
5010	 * Rx sequence error interrupt.  get_link_status will stay
5011	 * false until the check_for_link establishes link
5012	 * for copper adapters ONLY
5013	 */
5014	switch (hw->phy.media_type) {
5015	case e1000_media_type_copper:
5016		if (hw->mac.get_link_status) {
5017			ret_val = hw->mac.ops.check_for_link(hw);
5018			link_active = !hw->mac.get_link_status;
5019		} else {
5020			link_active = true;
5021		}
5022		break;
5023	case e1000_media_type_fiber:
5024		ret_val = hw->mac.ops.check_for_link(hw);
5025		link_active = !!(er32(STATUS) & E1000_STATUS_LU);
5026		break;
5027	case e1000_media_type_internal_serdes:
5028		ret_val = hw->mac.ops.check_for_link(hw);
5029		link_active = adapter->hw.mac.serdes_has_link;
5030		break;
5031	default:
5032	case e1000_media_type_unknown:
5033		break;
5034	}
5035
5036	if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
5037	    (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
5038		/* See e1000_kmrn_lock_loss_workaround_ich8lan() */
5039		e_info("Gigabit has been disabled, downgrading speed\n");
5040	}
5041
5042	return link_active;
5043}
5044
5045static void e1000e_enable_receives(struct e1000_adapter *adapter)
5046{
5047	/* make sure the receive unit is started */
5048	if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5049	    (adapter->flags & FLAG_RESTART_NOW)) {
5050		struct e1000_hw *hw = &adapter->hw;
5051		u32 rctl = er32(RCTL);
5052
5053		ew32(RCTL, rctl | E1000_RCTL_EN);
5054		adapter->flags &= ~FLAG_RESTART_NOW;
5055	}
5056}
5057
5058static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
5059{
5060	struct e1000_hw *hw = &adapter->hw;
5061
5062	/* With 82574 controllers, PHY needs to be checked periodically
5063	 * for hung state and reset, if two calls return true
5064	 */
5065	if (e1000_check_phy_82574(hw))
5066		adapter->phy_hang_count++;
5067	else
5068		adapter->phy_hang_count = 0;
5069
5070	if (adapter->phy_hang_count > 1) {
5071		adapter->phy_hang_count = 0;
5072		e_dbg("PHY appears hung - resetting\n");
5073		schedule_work(&adapter->reset_task);
5074	}
5075}
5076
5077/**
5078 * e1000_watchdog - Timer Call-back
5079 * @data: pointer to adapter cast into an unsigned long
5080 **/
5081static void e1000_watchdog(unsigned long data)
5082{
5083	struct e1000_adapter *adapter = (struct e1000_adapter *)data;
5084
5085	/* Do the rest outside of interrupt context */
5086	schedule_work(&adapter->watchdog_task);
5087
5088	/* TODO: make this use queue_delayed_work() */
5089}
5090
5091static void e1000_watchdog_task(struct work_struct *work)
5092{
5093	struct e1000_adapter *adapter = container_of(work,
5094						     struct e1000_adapter,
5095						     watchdog_task);
5096	struct net_device *netdev = adapter->netdev;
5097	struct e1000_mac_info *mac = &adapter->hw.mac;
5098	struct e1000_phy_info *phy = &adapter->hw.phy;
5099	struct e1000_ring *tx_ring = adapter->tx_ring;
5100	struct e1000_hw *hw = &adapter->hw;
5101	u32 link, tctl;
5102
5103	if (test_bit(__E1000_DOWN, &adapter->state))
5104		return;
5105
5106	link = e1000e_has_link(adapter);
5107	if ((netif_carrier_ok(netdev)) && link) {
5108		/* Cancel scheduled suspend requests. */
5109		pm_runtime_resume(netdev->dev.parent);
5110
5111		e1000e_enable_receives(adapter);
5112		goto link_up;
5113	}
5114
5115	if ((e1000e_enable_tx_pkt_filtering(hw)) &&
5116	    (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
5117		e1000_update_mng_vlan(adapter);
5118
5119	if (link) {
5120		if (!netif_carrier_ok(netdev)) {
5121			bool txb2b = true;
5122
5123			/* Cancel scheduled suspend requests. */
5124			pm_runtime_resume(netdev->dev.parent);
5125
5126			/* update snapshot of PHY registers on LSC */
5127			e1000_phy_read_status(adapter);
5128			mac->ops.get_link_up_info(&adapter->hw,
5129						  &adapter->link_speed,
5130						  &adapter->link_duplex);
5131			e1000_print_link_info(adapter);
5132
5133			/* check if SmartSpeed worked */
5134			e1000e_check_downshift(hw);
5135			if (phy->speed_downgraded)
5136				netdev_warn(netdev,
5137					    "Link Speed was downgraded by SmartSpeed\n");
5138
5139			/* On supported PHYs, check for duplex mismatch only
5140			 * if link has autonegotiated at 10/100 half
5141			 */
5142			if ((hw->phy.type == e1000_phy_igp_3 ||
5143			     hw->phy.type == e1000_phy_bm) &&
5144			    hw->mac.autoneg &&
5145			    (adapter->link_speed == SPEED_10 ||
5146			     adapter->link_speed == SPEED_100) &&
5147			    (adapter->link_duplex == HALF_DUPLEX)) {
5148				u16 autoneg_exp;
5149
5150				e1e_rphy(hw, MII_EXPANSION, &autoneg_exp);
5151
5152				if (!(autoneg_exp & EXPANSION_NWAY))
5153					e_info("Autonegotiated half duplex but link partner cannot autoneg.  Try forcing full duplex if link gets many collisions.\n");
5154			}
5155
5156			/* adjust timeout factor according to speed/duplex */
5157			adapter->tx_timeout_factor = 1;
5158			switch (adapter->link_speed) {
5159			case SPEED_10:
5160				txb2b = false;
5161				adapter->tx_timeout_factor = 16;
5162				break;
5163			case SPEED_100:
5164				txb2b = false;
5165				adapter->tx_timeout_factor = 10;
5166				break;
5167			}
5168
5169			/* workaround: re-program speed mode bit after
5170			 * link-up event
5171			 */
5172			if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
5173			    !txb2b) {
5174				u32 tarc0;
5175
5176				tarc0 = er32(TARC(0));
5177				tarc0 &= ~SPEED_MODE_BIT;
5178				ew32(TARC(0), tarc0);
5179			}
5180
5181			/* disable TSO for pcie and 10/100 speeds, to avoid
5182			 * some hardware issues
5183			 */
5184			if (!(adapter->flags & FLAG_TSO_FORCE)) {
5185				switch (adapter->link_speed) {
5186				case SPEED_10:
5187				case SPEED_100:
5188					e_info("10/100 speed: disabling TSO\n");
5189					netdev->features &= ~NETIF_F_TSO;
5190					netdev->features &= ~NETIF_F_TSO6;
5191					break;
5192				case SPEED_1000:
5193					netdev->features |= NETIF_F_TSO;
5194					netdev->features |= NETIF_F_TSO6;
5195					break;
5196				default:
5197					/* oops */
5198					break;
5199				}
5200			}
5201
5202			/* enable transmits in the hardware, need to do this
5203			 * after setting TARC(0)
5204			 */
5205			tctl = er32(TCTL);
5206			tctl |= E1000_TCTL_EN;
5207			ew32(TCTL, tctl);
5208
5209			/* Perform any post-link-up configuration before
5210			 * reporting link up.
5211			 */
5212			if (phy->ops.cfg_on_link_up)
5213				phy->ops.cfg_on_link_up(hw);
5214
5215			netif_carrier_on(netdev);
5216
5217			if (!test_bit(__E1000_DOWN, &adapter->state))
5218				mod_timer(&adapter->phy_info_timer,
5219					  round_jiffies(jiffies + 2 * HZ));
5220		}
5221	} else {
5222		if (netif_carrier_ok(netdev)) {
5223			adapter->link_speed = 0;
5224			adapter->link_duplex = 0;
5225			/* Link status message must follow this format */
5226			pr_info("%s NIC Link is Down\n", adapter->netdev->name);
5227			netif_carrier_off(netdev);
5228			if (!test_bit(__E1000_DOWN, &adapter->state))
5229				mod_timer(&adapter->phy_info_timer,
5230					  round_jiffies(jiffies + 2 * HZ));
5231
5232			/* 8000ES2LAN requires a Rx packet buffer work-around
5233			 * on link down event; reset the controller to flush
5234			 * the Rx packet buffer.
5235			 */
5236			if (adapter->flags & FLAG_RX_NEEDS_RESTART)
5237				adapter->flags |= FLAG_RESTART_NOW;
5238			else
5239				pm_schedule_suspend(netdev->dev.parent,
5240						    LINK_TIMEOUT);
5241		}
5242	}
5243
5244link_up:
5245	spin_lock(&adapter->stats64_lock);
5246	e1000e_update_stats(adapter);
5247
5248	mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
5249	adapter->tpt_old = adapter->stats.tpt;
5250	mac->collision_delta = adapter->stats.colc - adapter->colc_old;
5251	adapter->colc_old = adapter->stats.colc;
5252
5253	adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
5254	adapter->gorc_old = adapter->stats.gorc;
5255	adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
5256	adapter->gotc_old = adapter->stats.gotc;
5257	spin_unlock(&adapter->stats64_lock);
5258
5259	/* If the link is lost the controller stops DMA, but
5260	 * if there is queued Tx work it cannot be done.  So
5261	 * reset the controller to flush the Tx packet buffers.
5262	 */
5263	if (!netif_carrier_ok(netdev) &&
5264	    (e1000_desc_unused(tx_ring) + 1 < tx_ring->count))
5265		adapter->flags |= FLAG_RESTART_NOW;
5266
5267	/* If reset is necessary, do it outside of interrupt context. */
5268	if (adapter->flags & FLAG_RESTART_NOW) {
5269		schedule_work(&adapter->reset_task);
5270		/* return immediately since reset is imminent */
5271		return;
5272	}
5273
5274	e1000e_update_adaptive(&adapter->hw);
5275
5276	/* Simple mode for Interrupt Throttle Rate (ITR) */
5277	if (adapter->itr_setting == 4) {
5278		/* Symmetric Tx/Rx gets a reduced ITR=2000;
5279		 * Total asymmetrical Tx or Rx gets ITR=8000;
5280		 * everyone else is between 2000-8000.
5281		 */
5282		u32 goc = (adapter->gotc + adapter->gorc) / 10000;
5283		u32 dif = (adapter->gotc > adapter->gorc ?
5284			   adapter->gotc - adapter->gorc :
5285			   adapter->gorc - adapter->gotc) / 10000;
5286		u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
5287
5288		e1000e_write_itr(adapter, itr);
5289	}
5290
5291	/* Cause software interrupt to ensure Rx ring is cleaned */
5292	if (adapter->msix_entries)
5293		ew32(ICS, adapter->rx_ring->ims_val);
5294	else
5295		ew32(ICS, E1000_ICS_RXDMT0);
5296
5297	/* flush pending descriptors to memory before detecting Tx hang */
5298	e1000e_flush_descriptors(adapter);
5299
5300	/* Force detection of hung controller every watchdog period */
5301	adapter->detect_tx_hung = true;
5302
5303	/* With 82571 controllers, LAA may be overwritten due to controller
5304	 * reset from the other port. Set the appropriate LAA in RAR[0]
5305	 */
5306	if (e1000e_get_laa_state_82571(hw))
5307		hw->mac.ops.rar_set(hw, adapter->hw.mac.addr, 0);
5308
5309	if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
5310		e1000e_check_82574_phy_workaround(adapter);
5311
5312	/* Clear valid timestamp stuck in RXSTMPL/H due to a Rx error */
5313	if (adapter->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE) {
5314		if ((adapter->flags2 & FLAG2_CHECK_RX_HWTSTAMP) &&
5315		    (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID)) {
5316			er32(RXSTMPH);
5317			adapter->rx_hwtstamp_cleared++;
5318		} else {
5319			adapter->flags2 |= FLAG2_CHECK_RX_HWTSTAMP;
5320		}
5321	}
5322
5323	/* Reset the timer */
5324	if (!test_bit(__E1000_DOWN, &adapter->state))
5325		mod_timer(&adapter->watchdog_timer,
5326			  round_jiffies(jiffies + 2 * HZ));
5327}
5328
5329#define E1000_TX_FLAGS_CSUM		0x00000001
5330#define E1000_TX_FLAGS_VLAN		0x00000002
5331#define E1000_TX_FLAGS_TSO		0x00000004
5332#define E1000_TX_FLAGS_IPV4		0x00000008
5333#define E1000_TX_FLAGS_NO_FCS		0x00000010
5334#define E1000_TX_FLAGS_HWTSTAMP		0x00000020
5335#define E1000_TX_FLAGS_VLAN_MASK	0xffff0000
5336#define E1000_TX_FLAGS_VLAN_SHIFT	16
5337
5338static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb,
5339		     __be16 protocol)
5340{
5341	struct e1000_context_desc *context_desc;
5342	struct e1000_buffer *buffer_info;
5343	unsigned int i;
5344	u32 cmd_length = 0;
5345	u16 ipcse = 0, mss;
5346	u8 ipcss, ipcso, tucss, tucso, hdr_len;
5347	int err;
5348
5349	if (!skb_is_gso(skb))
5350		return 0;
5351
5352	err = skb_cow_head(skb, 0);
5353	if (err < 0)
5354		return err;
5355
5356	hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5357	mss = skb_shinfo(skb)->gso_size;
5358	if (protocol == htons(ETH_P_IP)) {
5359		struct iphdr *iph = ip_hdr(skb);
5360		iph->tot_len = 0;
5361		iph->check = 0;
5362		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
5363							 0, IPPROTO_TCP, 0);
5364		cmd_length = E1000_TXD_CMD_IP;
5365		ipcse = skb_transport_offset(skb) - 1;
5366	} else if (skb_is_gso_v6(skb)) {
5367		ipv6_hdr(skb)->payload_len = 0;
5368		tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5369						       &ipv6_hdr(skb)->daddr,
5370						       0, IPPROTO_TCP, 0);
5371		ipcse = 0;
5372	}
5373	ipcss = skb_network_offset(skb);
5374	ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
5375	tucss = skb_transport_offset(skb);
5376	tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
5377
5378	cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
5379		       E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
5380
5381	i = tx_ring->next_to_use;
5382	context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5383	buffer_info = &tx_ring->buffer_info[i];
5384
5385	context_desc->lower_setup.ip_fields.ipcss = ipcss;
5386	context_desc->lower_setup.ip_fields.ipcso = ipcso;
5387	context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
5388	context_desc->upper_setup.tcp_fields.tucss = tucss;
5389	context_desc->upper_setup.tcp_fields.tucso = tucso;
5390	context_desc->upper_setup.tcp_fields.tucse = 0;
5391	context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
5392	context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
5393	context_desc->cmd_and_length = cpu_to_le32(cmd_length);
5394
5395	buffer_info->time_stamp = jiffies;
5396	buffer_info->next_to_watch = i;
5397
5398	i++;
5399	if (i == tx_ring->count)
5400		i = 0;
5401	tx_ring->next_to_use = i;
5402
5403	return 1;
5404}
5405
5406static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb,
5407			  __be16 protocol)
5408{
5409	struct e1000_adapter *adapter = tx_ring->adapter;
5410	struct e1000_context_desc *context_desc;
5411	struct e1000_buffer *buffer_info;
5412	unsigned int i;
5413	u8 css;
5414	u32 cmd_len = E1000_TXD_CMD_DEXT;
5415
5416	if (skb->ip_summed != CHECKSUM_PARTIAL)
5417		return false;
5418
5419	switch (protocol) {
5420	case cpu_to_be16(ETH_P_IP):
5421		if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5422			cmd_len |= E1000_TXD_CMD_TCP;
5423		break;
5424	case cpu_to_be16(ETH_P_IPV6):
5425		/* XXX not handling all IPV6 headers */
5426		if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5427			cmd_len |= E1000_TXD_CMD_TCP;
5428		break;
5429	default:
5430		if (unlikely(net_ratelimit()))
5431			e_warn("checksum_partial proto=%x!\n",
5432			       be16_to_cpu(protocol));
5433		break;
5434	}
5435
5436	css = skb_checksum_start_offset(skb);
5437
5438	i = tx_ring->next_to_use;
5439	buffer_info = &tx_ring->buffer_info[i];
5440	context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5441
5442	context_desc->lower_setup.ip_config = 0;
5443	context_desc->upper_setup.tcp_fields.tucss = css;
5444	context_desc->upper_setup.tcp_fields.tucso = css + skb->csum_offset;
5445	context_desc->upper_setup.tcp_fields.tucse = 0;
5446	context_desc->tcp_seg_setup.data = 0;
5447	context_desc->cmd_and_length = cpu_to_le32(cmd_len);
5448
5449	buffer_info->time_stamp = jiffies;
5450	buffer_info->next_to_watch = i;
5451
5452	i++;
5453	if (i == tx_ring->count)
5454		i = 0;
5455	tx_ring->next_to_use = i;
5456
5457	return true;
5458}
5459
5460static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
5461			unsigned int first, unsigned int max_per_txd,
5462			unsigned int nr_frags)
5463{
5464	struct e1000_adapter *adapter = tx_ring->adapter;
5465	struct pci_dev *pdev = adapter->pdev;
5466	struct e1000_buffer *buffer_info;
5467	unsigned int len = skb_headlen(skb);
5468	unsigned int offset = 0, size, count = 0, i;
5469	unsigned int f, bytecount, segs;
5470
5471	i = tx_ring->next_to_use;
5472
5473	while (len) {
5474		buffer_info = &tx_ring->buffer_info[i];
5475		size = min(len, max_per_txd);
5476
5477		buffer_info->length = size;
5478		buffer_info->time_stamp = jiffies;
5479		buffer_info->next_to_watch = i;
5480		buffer_info->dma = dma_map_single(&pdev->dev,
5481						  skb->data + offset,
5482						  size, DMA_TO_DEVICE);
5483		buffer_info->mapped_as_page = false;
5484		if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5485			goto dma_error;
5486
5487		len -= size;
5488		offset += size;
5489		count++;
5490
5491		if (len) {
5492			i++;
5493			if (i == tx_ring->count)
5494				i = 0;
5495		}
5496	}
5497
5498	for (f = 0; f < nr_frags; f++) {
5499		const struct skb_frag_struct *frag;
5500
5501		frag = &skb_shinfo(skb)->frags[f];
5502		len = skb_frag_size(frag);
5503		offset = 0;
5504
5505		while (len) {
5506			i++;
5507			if (i == tx_ring->count)
5508				i = 0;
5509
5510			buffer_info = &tx_ring->buffer_info[i];
5511			size = min(len, max_per_txd);
5512
5513			buffer_info->length = size;
5514			buffer_info->time_stamp = jiffies;
5515			buffer_info->next_to_watch = i;
5516			buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
5517							    offset, size,
5518							    DMA_TO_DEVICE);
5519			buffer_info->mapped_as_page = true;
5520			if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5521				goto dma_error;
5522
5523			len -= size;
5524			offset += size;
5525			count++;
5526		}
5527	}
5528
5529	segs = skb_shinfo(skb)->gso_segs ? : 1;
5530	/* multiply data chunks by size of headers */
5531	bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
5532
5533	tx_ring->buffer_info[i].skb = skb;
5534	tx_ring->buffer_info[i].segs = segs;
5535	tx_ring->buffer_info[i].bytecount = bytecount;
5536	tx_ring->buffer_info[first].next_to_watch = i;
5537
5538	return count;
5539
5540dma_error:
5541	dev_err(&pdev->dev, "Tx DMA map failed\n");
5542	buffer_info->dma = 0;
5543	if (count)
5544		count--;
5545
5546	while (count--) {
5547		if (i == 0)
5548			i += tx_ring->count;
5549		i--;
5550		buffer_info = &tx_ring->buffer_info[i];
5551		e1000_put_txbuf(tx_ring, buffer_info);
5552	}
5553
5554	return 0;
5555}
5556
5557static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
5558{
5559	struct e1000_adapter *adapter = tx_ring->adapter;
5560	struct e1000_tx_desc *tx_desc = NULL;
5561	struct e1000_buffer *buffer_info;
5562	u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
5563	unsigned int i;
5564
5565	if (tx_flags & E1000_TX_FLAGS_TSO) {
5566		txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
5567		    E1000_TXD_CMD_TSE;
5568		txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5569
5570		if (tx_flags & E1000_TX_FLAGS_IPV4)
5571			txd_upper |= E1000_TXD_POPTS_IXSM << 8;
5572	}
5573
5574	if (tx_flags & E1000_TX_FLAGS_CSUM) {
5575		txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5576		txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5577	}
5578
5579	if (tx_flags & E1000_TX_FLAGS_VLAN) {
5580		txd_lower |= E1000_TXD_CMD_VLE;
5581		txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
5582	}
5583
5584	if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5585		txd_lower &= ~(E1000_TXD_CMD_IFCS);
5586
5587	if (unlikely(tx_flags & E1000_TX_FLAGS_HWTSTAMP)) {
5588		txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5589		txd_upper |= E1000_TXD_EXTCMD_TSTAMP;
5590	}
5591
5592	i = tx_ring->next_to_use;
5593
5594	do {
5595		buffer_info = &tx_ring->buffer_info[i];
5596		tx_desc = E1000_TX_DESC(*tx_ring, i);
5597		tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
5598		tx_desc->lower.data = cpu_to_le32(txd_lower |
5599						  buffer_info->length);
5600		tx_desc->upper.data = cpu_to_le32(txd_upper);
5601
5602		i++;
5603		if (i == tx_ring->count)
5604			i = 0;
5605	} while (--count > 0);
5606
5607	tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
5608
5609	/* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
5610	if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5611		tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
5612
5613	/* Force memory writes to complete before letting h/w
5614	 * know there are new descriptors to fetch.  (Only
5615	 * applicable for weak-ordered memory model archs,
5616	 * such as IA-64).
5617	 */
5618	wmb();
5619
5620	tx_ring->next_to_use = i;
5621}
5622
5623#define MINIMUM_DHCP_PACKET_SIZE 282
5624static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
5625				    struct sk_buff *skb)
5626{
5627	struct e1000_hw *hw = &adapter->hw;
5628	u16 length, offset;
5629
5630	if (skb_vlan_tag_present(skb) &&
5631	    !((skb_vlan_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
5632	      (adapter->hw.mng_cookie.status &
5633	       E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
5634		return 0;
5635
5636	if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
5637		return 0;
5638
5639	if (((struct ethhdr *)skb->data)->h_proto != htons(ETH_P_IP))
5640		return 0;
5641
5642	{
5643		const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data + 14);
5644		struct udphdr *udp;
5645
5646		if (ip->protocol != IPPROTO_UDP)
5647			return 0;
5648
5649		udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
5650		if (ntohs(udp->dest) != 67)
5651			return 0;
5652
5653		offset = (u8 *)udp + 8 - skb->data;
5654		length = skb->len - offset;
5655		return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
5656	}
5657
5658	return 0;
5659}
5660
5661static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5662{
5663	struct e1000_adapter *adapter = tx_ring->adapter;
5664
5665	netif_stop_queue(adapter->netdev);
5666	/* Herbert's original patch had:
5667	 *  smp_mb__after_netif_stop_queue();
5668	 * but since that doesn't exist yet, just open code it.
5669	 */
5670	smp_mb();
5671
5672	/* We need to check again in a case another CPU has just
5673	 * made room available.
5674	 */
5675	if (e1000_desc_unused(tx_ring) < size)
5676		return -EBUSY;
5677
5678	/* A reprieve! */
5679	netif_start_queue(adapter->netdev);
5680	++adapter->restart_queue;
5681	return 0;
5682}
5683
5684static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5685{
5686	BUG_ON(size > tx_ring->count);
5687
5688	if (e1000_desc_unused(tx_ring) >= size)
5689		return 0;
5690	return __e1000_maybe_stop_tx(tx_ring, size);
5691}
5692
5693static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
5694				    struct net_device *netdev)
5695{
5696	struct e1000_adapter *adapter = netdev_priv(netdev);
5697	struct e1000_ring *tx_ring = adapter->tx_ring;
5698	unsigned int first;
5699	unsigned int tx_flags = 0;
5700	unsigned int len = skb_headlen(skb);
5701	unsigned int nr_frags;
5702	unsigned int mss;
5703	int count = 0;
5704	int tso;
5705	unsigned int f;
5706	__be16 protocol = vlan_get_protocol(skb);
5707
5708	if (test_bit(__E1000_DOWN, &adapter->state)) {
5709		dev_kfree_skb_any(skb);
5710		return NETDEV_TX_OK;
5711	}
5712
5713	if (skb->len <= 0) {
5714		dev_kfree_skb_any(skb);
5715		return NETDEV_TX_OK;
5716	}
5717
5718	/* The minimum packet size with TCTL.PSP set is 17 bytes so
5719	 * pad skb in order to meet this minimum size requirement
5720	 */
5721	if (skb_put_padto(skb, 17))
5722		return NETDEV_TX_OK;
5723
5724	mss = skb_shinfo(skb)->gso_size;
5725	if (mss) {
5726		u8 hdr_len;
5727
5728		/* TSO Workaround for 82571/2/3 Controllers -- if skb->data
5729		 * points to just header, pull a few bytes of payload from
5730		 * frags into skb->data
5731		 */
5732		hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5733		/* we do this workaround for ES2LAN, but it is un-necessary,
5734		 * avoiding it could save a lot of cycles
5735		 */
5736		if (skb->data_len && (hdr_len == len)) {
5737			unsigned int pull_size;
5738
5739			pull_size = min_t(unsigned int, 4, skb->data_len);
5740			if (!__pskb_pull_tail(skb, pull_size)) {
5741				e_err("__pskb_pull_tail failed.\n");
5742				dev_kfree_skb_any(skb);
5743				return NETDEV_TX_OK;
5744			}
5745			len = skb_headlen(skb);
5746		}
5747	}
5748
5749	/* reserve a descriptor for the offload context */
5750	if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5751		count++;
5752	count++;
5753
5754	count += DIV_ROUND_UP(len, adapter->tx_fifo_limit);
5755
5756	nr_frags = skb_shinfo(skb)->nr_frags;
5757	for (f = 0; f < nr_frags; f++)
5758		count += DIV_ROUND_UP(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5759				      adapter->tx_fifo_limit);
5760
5761	if (adapter->hw.mac.tx_pkt_filtering)
5762		e1000_transfer_dhcp_info(adapter, skb);
5763
5764	/* need: count + 2 desc gap to keep tail from touching
5765	 * head, otherwise try next time
5766	 */
5767	if (e1000_maybe_stop_tx(tx_ring, count + 2))
5768		return NETDEV_TX_BUSY;
5769
5770	if (skb_vlan_tag_present(skb)) {
5771		tx_flags |= E1000_TX_FLAGS_VLAN;
5772		tx_flags |= (skb_vlan_tag_get(skb) <<
5773			     E1000_TX_FLAGS_VLAN_SHIFT);
5774	}
5775
5776	first = tx_ring->next_to_use;
5777
5778	tso = e1000_tso(tx_ring, skb, protocol);
5779	if (tso < 0) {
5780		dev_kfree_skb_any(skb);
5781		return NETDEV_TX_OK;
5782	}
5783
5784	if (tso)
5785		tx_flags |= E1000_TX_FLAGS_TSO;
5786	else if (e1000_tx_csum(tx_ring, skb, protocol))
5787		tx_flags |= E1000_TX_FLAGS_CSUM;
5788
5789	/* Old method was to assume IPv4 packet by default if TSO was enabled.
5790	 * 82571 hardware supports TSO capabilities for IPv6 as well...
5791	 * no longer assume, we must.
5792	 */
5793	if (protocol == htons(ETH_P_IP))
5794		tx_flags |= E1000_TX_FLAGS_IPV4;
5795
5796	if (unlikely(skb->no_fcs))
5797		tx_flags |= E1000_TX_FLAGS_NO_FCS;
5798
5799	/* if count is 0 then mapping error has occurred */
5800	count = e1000_tx_map(tx_ring, skb, first, adapter->tx_fifo_limit,
5801			     nr_frags);
5802	if (count) {
5803		if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
5804		    (adapter->flags & FLAG_HAS_HW_TIMESTAMP) &&
5805		    !adapter->tx_hwtstamp_skb) {
5806			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
5807			tx_flags |= E1000_TX_FLAGS_HWTSTAMP;
5808			adapter->tx_hwtstamp_skb = skb_get(skb);
5809			adapter->tx_hwtstamp_start = jiffies;
5810			schedule_work(&adapter->tx_hwtstamp_work);
5811		} else {
5812			skb_tx_timestamp(skb);
5813		}
5814
5815		netdev_sent_queue(netdev, skb->len);
5816		e1000_tx_queue(tx_ring, tx_flags, count);
5817		/* Make sure there is space in the ring for the next send. */
5818		e1000_maybe_stop_tx(tx_ring,
5819				    (MAX_SKB_FRAGS *
5820				     DIV_ROUND_UP(PAGE_SIZE,
5821						  adapter->tx_fifo_limit) + 2));
5822
5823		if (!skb->xmit_more ||
5824		    netif_xmit_stopped(netdev_get_tx_queue(netdev, 0))) {
5825			if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
5826				e1000e_update_tdt_wa(tx_ring,
5827						     tx_ring->next_to_use);
5828			else
5829				writel(tx_ring->next_to_use, tx_ring->tail);
5830
5831			/* we need this if more than one processor can write
5832			 * to our tail at a time, it synchronizes IO on
5833			 *IA64/Altix systems
5834			 */
5835			mmiowb();
5836		}
5837	} else {
5838		dev_kfree_skb_any(skb);
5839		tx_ring->buffer_info[first].time_stamp = 0;
5840		tx_ring->next_to_use = first;
5841	}
5842
5843	return NETDEV_TX_OK;
5844}
5845
5846/**
5847 * e1000_tx_timeout - Respond to a Tx Hang
5848 * @netdev: network interface device structure
5849 **/
5850static void e1000_tx_timeout(struct net_device *netdev)
5851{
5852	struct e1000_adapter *adapter = netdev_priv(netdev);
5853
5854	/* Do the reset outside of interrupt context */
5855	adapter->tx_timeout_count++;
5856	schedule_work(&adapter->reset_task);
5857}
5858
5859static void e1000_reset_task(struct work_struct *work)
5860{
5861	struct e1000_adapter *adapter;
5862	adapter = container_of(work, struct e1000_adapter, reset_task);
5863
5864	/* don't run the task if already down */
5865	if (test_bit(__E1000_DOWN, &adapter->state))
5866		return;
5867
5868	if (!(adapter->flags & FLAG_RESTART_NOW)) {
5869		e1000e_dump(adapter);
5870		e_err("Reset adapter unexpectedly\n");
5871	}
5872	e1000e_reinit_locked(adapter);
5873}
5874
5875/**
5876 * e1000_get_stats64 - Get System Network Statistics
5877 * @netdev: network interface device structure
5878 * @stats: rtnl_link_stats64 pointer
5879 *
5880 * Returns the address of the device statistics structure.
5881 **/
5882struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
5883					     struct rtnl_link_stats64 *stats)
5884{
5885	struct e1000_adapter *adapter = netdev_priv(netdev);
5886
5887	memset(stats, 0, sizeof(struct rtnl_link_stats64));
5888	spin_lock(&adapter->stats64_lock);
5889	e1000e_update_stats(adapter);
5890	/* Fill out the OS statistics structure */
5891	stats->rx_bytes = adapter->stats.gorc;
5892	stats->rx_packets = adapter->stats.gprc;
5893	stats->tx_bytes = adapter->stats.gotc;
5894	stats->tx_packets = adapter->stats.gptc;
5895	stats->multicast = adapter->stats.mprc;
5896	stats->collisions = adapter->stats.colc;
5897
5898	/* Rx Errors */
5899
5900	/* RLEC on some newer hardware can be incorrect so build
5901	 * our own version based on RUC and ROC
5902	 */
5903	stats->rx_errors = adapter->stats.rxerrc +
5904	    adapter->stats.crcerrs + adapter->stats.algnerrc +
5905	    adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
5906	stats->rx_length_errors = adapter->stats.ruc + adapter->stats.roc;
5907	stats->rx_crc_errors = adapter->stats.crcerrs;
5908	stats->rx_frame_errors = adapter->stats.algnerrc;
5909	stats->rx_missed_errors = adapter->stats.mpc;
5910
5911	/* Tx Errors */
5912	stats->tx_errors = adapter->stats.ecol + adapter->stats.latecol;
5913	stats->tx_aborted_errors = adapter->stats.ecol;
5914	stats->tx_window_errors = adapter->stats.latecol;
5915	stats->tx_carrier_errors = adapter->stats.tncrs;
5916
5917	/* Tx Dropped needs to be maintained elsewhere */
5918
5919	spin_unlock(&adapter->stats64_lock);
5920	return stats;
5921}
5922
5923/**
5924 * e1000_change_mtu - Change the Maximum Transfer Unit
5925 * @netdev: network interface device structure
5926 * @new_mtu: new value for maximum frame size
5927 *
5928 * Returns 0 on success, negative on failure
5929 **/
5930static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5931{
5932	struct e1000_adapter *adapter = netdev_priv(netdev);
5933	int max_frame = new_mtu + VLAN_ETH_HLEN + ETH_FCS_LEN;
5934
5935	/* Jumbo frame support */
5936	if ((max_frame > (VLAN_ETH_FRAME_LEN + ETH_FCS_LEN)) &&
5937	    !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5938		e_err("Jumbo Frames not supported.\n");
5939		return -EINVAL;
5940	}
5941
5942	/* Supported frame sizes */
5943	if ((new_mtu < (VLAN_ETH_ZLEN + ETH_FCS_LEN)) ||
5944	    (max_frame > adapter->max_hw_frame_size)) {
5945		e_err("Unsupported MTU setting\n");
5946		return -EINVAL;
5947	}
5948
5949	/* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
5950	if ((adapter->hw.mac.type >= e1000_pch2lan) &&
5951	    !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5952	    (new_mtu > ETH_DATA_LEN)) {
5953		e_err("Jumbo Frames not supported on this device when CRC stripping is disabled.\n");
5954		return -EINVAL;
5955	}
5956
5957	while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5958		usleep_range(1000, 2000);
5959	/* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5960	adapter->max_frame_size = max_frame;
5961	e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5962	netdev->mtu = new_mtu;
5963
5964	pm_runtime_get_sync(netdev->dev.parent);
5965
5966	if (netif_running(netdev))
5967		e1000e_down(adapter, true);
5968
5969	/* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5970	 * means we reserve 2 more, this pushes us to allocate from the next
5971	 * larger slab size.
5972	 * i.e. RXBUFFER_2048 --> size-4096 slab
5973	 * However with the new *_jumbo_rx* routines, jumbo receives will use
5974	 * fragmented skbs
5975	 */
5976
5977	if (max_frame <= 2048)
5978		adapter->rx_buffer_len = 2048;
5979	else
5980		adapter->rx_buffer_len = 4096;
5981
5982	/* adjust allocation if LPE protects us, and we aren't using SBP */
5983	if (max_frame <= (VLAN_ETH_FRAME_LEN + ETH_FCS_LEN))
5984		adapter->rx_buffer_len = VLAN_ETH_FRAME_LEN + ETH_FCS_LEN;
5985
5986	if (netif_running(netdev))
5987		e1000e_up(adapter);
5988	else
5989		e1000e_reset(adapter);
5990
5991	pm_runtime_put_sync(netdev->dev.parent);
5992
5993	clear_bit(__E1000_RESETTING, &adapter->state);
5994
5995	return 0;
5996}
5997
5998static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5999			   int cmd)
6000{
6001	struct e1000_adapter *adapter = netdev_priv(netdev);
6002	struct mii_ioctl_data *data = if_mii(ifr);
6003
6004	if (adapter->hw.phy.media_type != e1000_media_type_copper)
6005		return -EOPNOTSUPP;
6006
6007	switch (cmd) {
6008	case SIOCGMIIPHY:
6009		data->phy_id = adapter->hw.phy.addr;
6010		break;
6011	case SIOCGMIIREG:
6012		e1000_phy_read_status(adapter);
6013
6014		switch (data->reg_num & 0x1F) {
6015		case MII_BMCR:
6016			data->val_out = adapter->phy_regs.bmcr;
6017			break;
6018		case MII_BMSR:
6019			data->val_out = adapter->phy_regs.bmsr;
6020			break;
6021		case MII_PHYSID1:
6022			data->val_out = (adapter->hw.phy.id >> 16);
6023			break;
6024		case MII_PHYSID2:
6025			data->val_out = (adapter->hw.phy.id & 0xFFFF);
6026			break;
6027		case MII_ADVERTISE:
6028			data->val_out = adapter->phy_regs.advertise;
6029			break;
6030		case MII_LPA:
6031			data->val_out = adapter->phy_regs.lpa;
6032			break;
6033		case MII_EXPANSION:
6034			data->val_out = adapter->phy_regs.expansion;
6035			break;
6036		case MII_CTRL1000:
6037			data->val_out = adapter->phy_regs.ctrl1000;
6038			break;
6039		case MII_STAT1000:
6040			data->val_out = adapter->phy_regs.stat1000;
6041			break;
6042		case MII_ESTATUS:
6043			data->val_out = adapter->phy_regs.estatus;
6044			break;
6045		default:
6046			return -EIO;
6047		}
6048		break;
6049	case SIOCSMIIREG:
6050	default:
6051		return -EOPNOTSUPP;
6052	}
6053	return 0;
6054}
6055
6056/**
6057 * e1000e_hwtstamp_ioctl - control hardware time stamping
6058 * @netdev: network interface device structure
6059 * @ifreq: interface request
6060 *
6061 * Outgoing time stamping can be enabled and disabled. Play nice and
6062 * disable it when requested, although it shouldn't cause any overhead
6063 * when no packet needs it. At most one packet in the queue may be
6064 * marked for time stamping, otherwise it would be impossible to tell
6065 * for sure to which packet the hardware time stamp belongs.
6066 *
6067 * Incoming time stamping has to be configured via the hardware filters.
6068 * Not all combinations are supported, in particular event type has to be
6069 * specified. Matching the kind of event packet is not supported, with the
6070 * exception of "all V2 events regardless of level 2 or 4".
6071 **/
6072static int e1000e_hwtstamp_set(struct net_device *netdev, struct ifreq *ifr)
6073{
6074	struct e1000_adapter *adapter = netdev_priv(netdev);
6075	struct hwtstamp_config config;
6076	int ret_val;
6077
6078	if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
6079		return -EFAULT;
6080
6081	ret_val = e1000e_config_hwtstamp(adapter, &config);
6082	if (ret_val)
6083		return ret_val;
6084
6085	switch (config.rx_filter) {
6086	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
6087	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
6088	case HWTSTAMP_FILTER_PTP_V2_SYNC:
6089	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
6090	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
6091	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
6092		/* With V2 type filters which specify a Sync or Delay Request,
6093		 * Path Delay Request/Response messages are also time stamped
6094		 * by hardware so notify the caller the requested packets plus
6095		 * some others are time stamped.
6096		 */
6097		config.rx_filter = HWTSTAMP_FILTER_SOME;
6098		break;
6099	default:
6100		break;
6101	}
6102
6103	return copy_to_user(ifr->ifr_data, &config,
6104			    sizeof(config)) ? -EFAULT : 0;
6105}
6106
6107static int e1000e_hwtstamp_get(struct net_device *netdev, struct ifreq *ifr)
6108{
6109	struct e1000_adapter *adapter = netdev_priv(netdev);
6110
6111	return copy_to_user(ifr->ifr_data, &adapter->hwtstamp_config,
6112			    sizeof(adapter->hwtstamp_config)) ? -EFAULT : 0;
6113}
6114
6115static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6116{
6117	switch (cmd) {
6118	case SIOCGMIIPHY:
6119	case SIOCGMIIREG:
6120	case SIOCSMIIREG:
6121		return e1000_mii_ioctl(netdev, ifr, cmd);
6122	case SIOCSHWTSTAMP:
6123		return e1000e_hwtstamp_set(netdev, ifr);
6124	case SIOCGHWTSTAMP:
6125		return e1000e_hwtstamp_get(netdev, ifr);
6126	default:
6127		return -EOPNOTSUPP;
6128	}
6129}
6130
6131static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
6132{
6133	struct e1000_hw *hw = &adapter->hw;
6134	u32 i, mac_reg, wuc;
6135	u16 phy_reg, wuc_enable;
6136	int retval;
6137
6138	/* copy MAC RARs to PHY RARs */
6139	e1000_copy_rx_addrs_to_phy_ich8lan(hw);
6140
6141	retval = hw->phy.ops.acquire(hw);
6142	if (retval) {
6143		e_err("Could not acquire PHY\n");
6144		return retval;
6145	}
6146
6147	/* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
6148	retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
6149	if (retval)
6150		goto release;
6151
6152	/* copy MAC MTA to PHY MTA - only needed for pchlan */
6153	for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
6154		mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
6155		hw->phy.ops.write_reg_page(hw, BM_MTA(i),
6156					   (u16)(mac_reg & 0xFFFF));
6157		hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
6158					   (u16)((mac_reg >> 16) & 0xFFFF));
6159	}
6160
6161	/* configure PHY Rx Control register */
6162	hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
6163	mac_reg = er32(RCTL);
6164	if (mac_reg & E1000_RCTL_UPE)
6165		phy_reg |= BM_RCTL_UPE;
6166	if (mac_reg & E1000_RCTL_MPE)
6167		phy_reg |= BM_RCTL_MPE;
6168	phy_reg &= ~(BM_RCTL_MO_MASK);
6169	if (mac_reg & E1000_RCTL_MO_3)
6170		phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
6171			    << BM_RCTL_MO_SHIFT);
6172	if (mac_reg & E1000_RCTL_BAM)
6173		phy_reg |= BM_RCTL_BAM;
6174	if (mac_reg & E1000_RCTL_PMCF)
6175		phy_reg |= BM_RCTL_PMCF;
6176	mac_reg = er32(CTRL);
6177	if (mac_reg & E1000_CTRL_RFCE)
6178		phy_reg |= BM_RCTL_RFCE;
6179	hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
6180
6181	wuc = E1000_WUC_PME_EN;
6182	if (wufc & (E1000_WUFC_MAG | E1000_WUFC_LNKC))
6183		wuc |= E1000_WUC_APME;
6184
6185	/* enable PHY wakeup in MAC register */
6186	ew32(WUFC, wufc);
6187	ew32(WUC, (E1000_WUC_PHY_WAKE | E1000_WUC_APMPME |
6188		   E1000_WUC_PME_STATUS | wuc));
6189
6190	/* configure and enable PHY wakeup in PHY registers */
6191	hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
6192	hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, wuc);
6193
6194	/* activate PHY wakeup */
6195	wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
6196	retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
6197	if (retval)
6198		e_err("Could not set PHY Host Wakeup bit\n");
6199release:
6200	hw->phy.ops.release(hw);
6201
6202	return retval;
6203}
6204
6205static void e1000e_flush_lpic(struct pci_dev *pdev)
6206{
6207	struct net_device *netdev = pci_get_drvdata(pdev);
6208	struct e1000_adapter *adapter = netdev_priv(netdev);
6209	struct e1000_hw *hw = &adapter->hw;
6210	u32 ret_val;
6211
6212	pm_runtime_get_sync(netdev->dev.parent);
6213
6214	ret_val = hw->phy.ops.acquire(hw);
6215	if (ret_val)
6216		goto fl_out;
6217
6218	pr_info("EEE TX LPI TIMER: %08X\n",
6219		er32(LPIC) >> E1000_LPIC_LPIET_SHIFT);
6220
6221	hw->phy.ops.release(hw);
6222
6223fl_out:
6224	pm_runtime_put_sync(netdev->dev.parent);
6225}
6226
6227static int e1000e_pm_freeze(struct device *dev)
6228{
6229	struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
6230	struct e1000_adapter *adapter = netdev_priv(netdev);
6231
6232	netif_device_detach(netdev);
6233
6234	if (netif_running(netdev)) {
6235		int count = E1000_CHECK_RESET_COUNT;
6236
6237		while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
6238			usleep_range(10000, 20000);
6239
6240		WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
6241
6242		/* Quiesce the device without resetting the hardware */
6243		e1000e_down(adapter, false);
6244		e1000_free_irq(adapter);
6245	}
6246	e1000e_reset_interrupt_capability(adapter);
6247
6248	/* Allow time for pending master requests to run */
6249	e1000e_disable_pcie_master(&adapter->hw);
6250
6251	return 0;
6252}
6253
6254static int __e1000_shutdown(struct pci_dev *pdev, bool runtime)
6255{
6256	struct net_device *netdev = pci_get_drvdata(pdev);
6257	struct e1000_adapter *adapter = netdev_priv(netdev);
6258	struct e1000_hw *hw = &adapter->hw;
6259	u32 ctrl, ctrl_ext, rctl, status;
6260	/* Runtime suspend should only enable wakeup for link changes */
6261	u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
6262	int retval = 0;
6263
6264	status = er32(STATUS);
6265	if (status & E1000_STATUS_LU)
6266		wufc &= ~E1000_WUFC_LNKC;
6267
6268	if (wufc) {
6269		e1000_setup_rctl(adapter);
6270		e1000e_set_rx_mode(netdev);
6271
6272		/* turn on all-multi mode if wake on multicast is enabled */
6273		if (wufc & E1000_WUFC_MC) {
6274			rctl = er32(RCTL);
6275			rctl |= E1000_RCTL_MPE;
6276			ew32(RCTL, rctl);
6277		}
6278
6279		ctrl = er32(CTRL);
6280		ctrl |= E1000_CTRL_ADVD3WUC;
6281		if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
6282			ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
6283		ew32(CTRL, ctrl);
6284
6285		if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
6286		    adapter->hw.phy.media_type ==
6287		    e1000_media_type_internal_serdes) {
6288			/* keep the laser running in D3 */
6289			ctrl_ext = er32(CTRL_EXT);
6290			ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
6291			ew32(CTRL_EXT, ctrl_ext);
6292		}
6293
6294		if (!runtime)
6295			e1000e_power_up_phy(adapter);
6296
6297		if (adapter->flags & FLAG_IS_ICH)
6298			e1000_suspend_workarounds_ich8lan(&adapter->hw);
6299
6300		if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6301			/* enable wakeup by the PHY */
6302			retval = e1000_init_phy_wakeup(adapter, wufc);
6303			if (retval)
6304				return retval;
6305		} else {
6306			/* enable wakeup by the MAC */
6307			ew32(WUFC, wufc);
6308			ew32(WUC, E1000_WUC_PME_EN);
6309		}
6310	} else {
6311		ew32(WUC, 0);
6312		ew32(WUFC, 0);
6313
6314		e1000_power_down_phy(adapter);
6315	}
6316
6317	if (adapter->hw.phy.type == e1000_phy_igp_3) {
6318		e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
6319	} else if ((hw->mac.type == e1000_pch_lpt) ||
6320		   (hw->mac.type == e1000_pch_spt)) {
6321		if (!(wufc & (E1000_WUFC_EX | E1000_WUFC_MC | E1000_WUFC_BC)))
6322			/* ULP does not support wake from unicast, multicast
6323			 * or broadcast.
6324			 */
6325			retval = e1000_enable_ulp_lpt_lp(hw, !runtime);
6326
6327		if (retval)
6328			return retval;
6329	}
6330
6331	/* Ensure that the appropriate bits are set in LPI_CTRL
6332	 * for EEE in Sx
6333	 */
6334	if ((hw->phy.type >= e1000_phy_i217) &&
6335	    adapter->eee_advert && hw->dev_spec.ich8lan.eee_lp_ability) {
6336		u16 lpi_ctrl = 0;
6337
6338		retval = hw->phy.ops.acquire(hw);
6339		if (!retval) {
6340			retval = e1e_rphy_locked(hw, I82579_LPI_CTRL,
6341						 &lpi_ctrl);
6342			if (!retval) {
6343				if (adapter->eee_advert &
6344				    hw->dev_spec.ich8lan.eee_lp_ability &
6345				    I82579_EEE_100_SUPPORTED)
6346					lpi_ctrl |= I82579_LPI_CTRL_100_ENABLE;
6347				if (adapter->eee_advert &
6348				    hw->dev_spec.ich8lan.eee_lp_ability &
6349				    I82579_EEE_1000_SUPPORTED)
6350					lpi_ctrl |= I82579_LPI_CTRL_1000_ENABLE;
6351
6352				retval = e1e_wphy_locked(hw, I82579_LPI_CTRL,
6353							 lpi_ctrl);
6354			}
6355		}
6356		hw->phy.ops.release(hw);
6357	}
6358
6359	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
6360	 * would have already happened in close and is redundant.
6361	 */
6362	e1000e_release_hw_control(adapter);
6363
6364	pci_clear_master(pdev);
6365
6366	/* The pci-e switch on some quad port adapters will report a
6367	 * correctable error when the MAC transitions from D0 to D3.  To
6368	 * prevent this we need to mask off the correctable errors on the
6369	 * downstream port of the pci-e switch.
6370	 *
6371	 * We don't have the associated upstream bridge while assigning
6372	 * the PCI device into guest. For example, the KVM on power is
6373	 * one of the cases.
6374	 */
6375	if (adapter->flags & FLAG_IS_QUAD_PORT) {
6376		struct pci_dev *us_dev = pdev->bus->self;
6377		u16 devctl;
6378
6379		if (!us_dev)
6380			return 0;
6381
6382		pcie_capability_read_word(us_dev, PCI_EXP_DEVCTL, &devctl);
6383		pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL,
6384					   (devctl & ~PCI_EXP_DEVCTL_CERE));
6385
6386		pci_save_state(pdev);
6387		pci_prepare_to_sleep(pdev);
6388
6389		pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL, devctl);
6390	}
6391
6392	return 0;
6393}
6394
6395/**
6396 * __e1000e_disable_aspm - Disable ASPM states
6397 * @pdev: pointer to PCI device struct
6398 * @state: bit-mask of ASPM states to disable
6399 * @locked: indication if this context holds pci_bus_sem locked.
6400 *
6401 * Some devices *must* have certain ASPM states disabled per hardware errata.
6402 **/
6403static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state, int locked)
6404{
6405	struct pci_dev *parent = pdev->bus->self;
6406	u16 aspm_dis_mask = 0;
6407	u16 pdev_aspmc, parent_aspmc;
6408
6409	switch (state) {
6410	case PCIE_LINK_STATE_L0S:
6411	case PCIE_LINK_STATE_L0S | PCIE_LINK_STATE_L1:
6412		aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L0S;
6413		/* fall-through - can't have L1 without L0s */
6414	case PCIE_LINK_STATE_L1:
6415		aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L1;
6416		break;
6417	default:
6418		return;
6419	}
6420
6421	pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6422	pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6423
6424	if (parent) {
6425		pcie_capability_read_word(parent, PCI_EXP_LNKCTL,
6426					  &parent_aspmc);
6427		parent_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6428	}
6429
6430	/* Nothing to do if the ASPM states to be disabled already are */
6431	if (!(pdev_aspmc & aspm_dis_mask) &&
6432	    (!parent || !(parent_aspmc & aspm_dis_mask)))
6433		return;
6434
6435	dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
6436		 (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L0S) ?
6437		 "L0s" : "",
6438		 (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L1) ?
6439		 "L1" : "");
6440
6441#ifdef CONFIG_PCIEASPM
6442	if (locked)
6443		pci_disable_link_state_locked(pdev, state);
6444	else
6445		pci_disable_link_state(pdev, state);
6446
6447	/* Double-check ASPM control.  If not disabled by the above, the
6448	 * BIOS is preventing that from happening (or CONFIG_PCIEASPM is
6449	 * not enabled); override by writing PCI config space directly.
6450	 */
6451	pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6452	pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6453
6454	if (!(aspm_dis_mask & pdev_aspmc))
6455		return;
6456#endif
6457
6458	/* Both device and parent should have the same ASPM setting.
6459	 * Disable ASPM in downstream component first and then upstream.
6460	 */
6461	pcie_capability_clear_word(pdev, PCI_EXP_LNKCTL, aspm_dis_mask);
6462
6463	if (parent)
6464		pcie_capability_clear_word(parent, PCI_EXP_LNKCTL,
6465					   aspm_dis_mask);
6466}
6467
6468/**
6469 * e1000e_disable_aspm - Disable ASPM states.
6470 * @pdev: pointer to PCI device struct
6471 * @state: bit-mask of ASPM states to disable
6472 *
6473 * This function acquires the pci_bus_sem!
6474 * Some devices *must* have certain ASPM states disabled per hardware errata.
6475 **/
6476static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
6477{
6478	__e1000e_disable_aspm(pdev, state, 0);
6479}
6480
6481/**
6482 * e1000e_disable_aspm_locked   Disable ASPM states.
6483 * @pdev: pointer to PCI device struct
6484 * @state: bit-mask of ASPM states to disable
6485 *
6486 * This function must be called with pci_bus_sem acquired!
6487 * Some devices *must* have certain ASPM states disabled per hardware errata.
6488 **/
6489static void e1000e_disable_aspm_locked(struct pci_dev *pdev, u16 state)
6490{
6491	__e1000e_disable_aspm(pdev, state, 1);
6492}
6493
6494#ifdef CONFIG_PM
6495static int __e1000_resume(struct pci_dev *pdev)
6496{
6497	struct net_device *netdev = pci_get_drvdata(pdev);
6498	struct e1000_adapter *adapter = netdev_priv(netdev);
6499	struct e1000_hw *hw = &adapter->hw;
6500	u16 aspm_disable_flag = 0;
6501
6502	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
6503		aspm_disable_flag = PCIE_LINK_STATE_L0S;
6504	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
6505		aspm_disable_flag |= PCIE_LINK_STATE_L1;
6506	if (aspm_disable_flag)
6507		e1000e_disable_aspm(pdev, aspm_disable_flag);
6508
6509	pci_set_master(pdev);
6510
6511	if (hw->mac.type >= e1000_pch2lan)
6512		e1000_resume_workarounds_pchlan(&adapter->hw);
6513
6514	e1000e_power_up_phy(adapter);
6515
6516	/* report the system wakeup cause from S3/S4 */
6517	if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6518		u16 phy_data;
6519
6520		e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
6521		if (phy_data) {
6522			e_info("PHY Wakeup cause - %s\n",
6523			       phy_data & E1000_WUS_EX ? "Unicast Packet" :
6524			       phy_data & E1000_WUS_MC ? "Multicast Packet" :
6525			       phy_data & E1000_WUS_BC ? "Broadcast Packet" :
6526			       phy_data & E1000_WUS_MAG ? "Magic Packet" :
6527			       phy_data & E1000_WUS_LNKC ?
6528			       "Link Status Change" : "other");
6529		}
6530		e1e_wphy(&adapter->hw, BM_WUS, ~0);
6531	} else {
6532		u32 wus = er32(WUS);
6533
6534		if (wus) {
6535			e_info("MAC Wakeup cause - %s\n",
6536			       wus & E1000_WUS_EX ? "Unicast Packet" :
6537			       wus & E1000_WUS_MC ? "Multicast Packet" :
6538			       wus & E1000_WUS_BC ? "Broadcast Packet" :
6539			       wus & E1000_WUS_MAG ? "Magic Packet" :
6540			       wus & E1000_WUS_LNKC ? "Link Status Change" :
6541			       "other");
6542		}
6543		ew32(WUS, ~0);
6544	}
6545
6546	e1000e_reset(adapter);
6547
6548	e1000_init_manageability_pt(adapter);
6549
6550	/* If the controller has AMT, do not set DRV_LOAD until the interface
6551	 * is up.  For all other cases, let the f/w know that the h/w is now
6552	 * under the control of the driver.
6553	 */
6554	if (!(adapter->flags & FLAG_HAS_AMT))
6555		e1000e_get_hw_control(adapter);
6556
6557	return 0;
6558}
6559
6560#ifdef CONFIG_PM_SLEEP
6561static int e1000e_pm_thaw(struct device *dev)
6562{
6563	struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
6564	struct e1000_adapter *adapter = netdev_priv(netdev);
6565
6566	e1000e_set_interrupt_capability(adapter);
6567	if (netif_running(netdev)) {
6568		u32 err = e1000_request_irq(adapter);
6569
6570		if (err)
6571			return err;
6572
6573		e1000e_up(adapter);
6574	}
6575
6576	netif_device_attach(netdev);
6577
6578	return 0;
6579}
6580
6581static int e1000e_pm_suspend(struct device *dev)
6582{
6583	struct pci_dev *pdev = to_pci_dev(dev);
6584
6585	e1000e_flush_lpic(pdev);
6586
6587	e1000e_pm_freeze(dev);
6588
6589	return __e1000_shutdown(pdev, false);
6590}
6591
6592static int e1000e_pm_resume(struct device *dev)
6593{
6594	struct pci_dev *pdev = to_pci_dev(dev);
6595	int rc;
6596
6597	rc = __e1000_resume(pdev);
6598	if (rc)
6599		return rc;
6600
6601	return e1000e_pm_thaw(dev);
6602}
6603#endif /* CONFIG_PM_SLEEP */
6604
6605static int e1000e_pm_runtime_idle(struct device *dev)
6606{
6607	struct pci_dev *pdev = to_pci_dev(dev);
6608	struct net_device *netdev = pci_get_drvdata(pdev);
6609	struct e1000_adapter *adapter = netdev_priv(netdev);
6610	u16 eee_lp;
6611
6612	eee_lp = adapter->hw.dev_spec.ich8lan.eee_lp_ability;
6613
6614	if (!e1000e_has_link(adapter)) {
6615		adapter->hw.dev_spec.ich8lan.eee_lp_ability = eee_lp;
6616		pm_schedule_suspend(dev, 5 * MSEC_PER_SEC);
6617	}
6618
6619	return -EBUSY;
6620}
6621
6622static int e1000e_pm_runtime_resume(struct device *dev)
6623{
6624	struct pci_dev *pdev = to_pci_dev(dev);
6625	struct net_device *netdev = pci_get_drvdata(pdev);
6626	struct e1000_adapter *adapter = netdev_priv(netdev);
6627	int rc;
6628
6629	rc = __e1000_resume(pdev);
6630	if (rc)
6631		return rc;
6632
6633	if (netdev->flags & IFF_UP)
6634		rc = e1000e_up(adapter);
6635
6636	return rc;
6637}
6638
6639static int e1000e_pm_runtime_suspend(struct device *dev)
6640{
6641	struct pci_dev *pdev = to_pci_dev(dev);
6642	struct net_device *netdev = pci_get_drvdata(pdev);
6643	struct e1000_adapter *adapter = netdev_priv(netdev);
6644
6645	if (netdev->flags & IFF_UP) {
6646		int count = E1000_CHECK_RESET_COUNT;
6647
6648		while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
6649			usleep_range(10000, 20000);
6650
6651		WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
6652
6653		/* Down the device without resetting the hardware */
6654		e1000e_down(adapter, false);
6655	}
6656
6657	if (__e1000_shutdown(pdev, true)) {
6658		e1000e_pm_runtime_resume(dev);
6659		return -EBUSY;
6660	}
6661
6662	return 0;
6663}
6664#endif /* CONFIG_PM */
6665
6666static void e1000_shutdown(struct pci_dev *pdev)
6667{
6668	e1000e_flush_lpic(pdev);
6669
6670	e1000e_pm_freeze(&pdev->dev);
6671
6672	__e1000_shutdown(pdev, false);
6673}
6674
6675#ifdef CONFIG_NET_POLL_CONTROLLER
6676
6677static irqreturn_t e1000_intr_msix(int __always_unused irq, void *data)
6678{
6679	struct net_device *netdev = data;
6680	struct e1000_adapter *adapter = netdev_priv(netdev);
6681
6682	if (adapter->msix_entries) {
6683		int vector, msix_irq;
6684
6685		vector = 0;
6686		msix_irq = adapter->msix_entries[vector].vector;
6687		disable_irq(msix_irq);
6688		e1000_intr_msix_rx(msix_irq, netdev);
6689		enable_irq(msix_irq);
6690
6691		vector++;
6692		msix_irq = adapter->msix_entries[vector].vector;
6693		disable_irq(msix_irq);
6694		e1000_intr_msix_tx(msix_irq, netdev);
6695		enable_irq(msix_irq);
6696
6697		vector++;
6698		msix_irq = adapter->msix_entries[vector].vector;
6699		disable_irq(msix_irq);
6700		e1000_msix_other(msix_irq, netdev);
6701		enable_irq(msix_irq);
6702	}
6703
6704	return IRQ_HANDLED;
6705}
6706
6707/**
6708 * e1000_netpoll
6709 * @netdev: network interface device structure
6710 *
6711 * Polling 'interrupt' - used by things like netconsole to send skbs
6712 * without having to re-enable interrupts. It's not called while
6713 * the interrupt routine is executing.
6714 */
6715static void e1000_netpoll(struct net_device *netdev)
6716{
6717	struct e1000_adapter *adapter = netdev_priv(netdev);
6718
6719	switch (adapter->int_mode) {
6720	case E1000E_INT_MODE_MSIX:
6721		e1000_intr_msix(adapter->pdev->irq, netdev);
6722		break;
6723	case E1000E_INT_MODE_MSI:
6724		disable_irq(adapter->pdev->irq);
6725		e1000_intr_msi(adapter->pdev->irq, netdev);
6726		enable_irq(adapter->pdev->irq);
6727		break;
6728	default:		/* E1000E_INT_MODE_LEGACY */
6729		disable_irq(adapter->pdev->irq);
6730		e1000_intr(adapter->pdev->irq, netdev);
6731		enable_irq(adapter->pdev->irq);
6732		break;
6733	}
6734}
6735#endif
6736
6737/**
6738 * e1000_io_error_detected - called when PCI error is detected
6739 * @pdev: Pointer to PCI device
6740 * @state: The current pci connection state
6741 *
6742 * This function is called after a PCI bus error affecting
6743 * this device has been detected.
6744 */
6745static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
6746						pci_channel_state_t state)
6747{
6748	struct net_device *netdev = pci_get_drvdata(pdev);
6749	struct e1000_adapter *adapter = netdev_priv(netdev);
6750
6751	netif_device_detach(netdev);
6752
6753	if (state == pci_channel_io_perm_failure)
6754		return PCI_ERS_RESULT_DISCONNECT;
6755
6756	if (netif_running(netdev))
6757		e1000e_down(adapter, true);
6758	pci_disable_device(pdev);
6759
6760	/* Request a slot slot reset. */
6761	return PCI_ERS_RESULT_NEED_RESET;
6762}
6763
6764/**
6765 * e1000_io_slot_reset - called after the pci bus has been reset.
6766 * @pdev: Pointer to PCI device
6767 *
6768 * Restart the card from scratch, as if from a cold-boot. Implementation
6769 * resembles the first-half of the e1000e_pm_resume routine.
6770 */
6771static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
6772{
6773	struct net_device *netdev = pci_get_drvdata(pdev);
6774	struct e1000_adapter *adapter = netdev_priv(netdev);
6775	struct e1000_hw *hw = &adapter->hw;
6776	u16 aspm_disable_flag = 0;
6777	int err;
6778	pci_ers_result_t result;
6779
6780	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
6781		aspm_disable_flag = PCIE_LINK_STATE_L0S;
6782	if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
6783		aspm_disable_flag |= PCIE_LINK_STATE_L1;
6784	if (aspm_disable_flag)
6785		e1000e_disable_aspm_locked(pdev, aspm_disable_flag);
6786
6787	err = pci_enable_device_mem(pdev);
6788	if (err) {
6789		dev_err(&pdev->dev,
6790			"Cannot re-enable PCI device after reset.\n");
6791		result = PCI_ERS_RESULT_DISCONNECT;
6792	} else {
6793		pdev->state_saved = true;
6794		pci_restore_state(pdev);
6795		pci_set_master(pdev);
6796
6797		pci_enable_wake(pdev, PCI_D3hot, 0);
6798		pci_enable_wake(pdev, PCI_D3cold, 0);
6799
6800		e1000e_reset(adapter);
6801		ew32(WUS, ~0);
6802		result = PCI_ERS_RESULT_RECOVERED;
6803	}
6804
6805	pci_cleanup_aer_uncorrect_error_status(pdev);
6806
6807	return result;
6808}
6809
6810/**
6811 * e1000_io_resume - called when traffic can start flowing again.
6812 * @pdev: Pointer to PCI device
6813 *
6814 * This callback is called when the error recovery driver tells us that
6815 * its OK to resume normal operation. Implementation resembles the
6816 * second-half of the e1000e_pm_resume routine.
6817 */
6818static void e1000_io_resume(struct pci_dev *pdev)
6819{
6820	struct net_device *netdev = pci_get_drvdata(pdev);
6821	struct e1000_adapter *adapter = netdev_priv(netdev);
6822
6823	e1000_init_manageability_pt(adapter);
6824
6825	if (netif_running(netdev)) {
6826		if (e1000e_up(adapter)) {
6827			dev_err(&pdev->dev,
6828				"can't bring device back up after reset\n");
6829			return;
6830		}
6831	}
6832
6833	netif_device_attach(netdev);
6834
6835	/* If the controller has AMT, do not set DRV_LOAD until the interface
6836	 * is up.  For all other cases, let the f/w know that the h/w is now
6837	 * under the control of the driver.
6838	 */
6839	if (!(adapter->flags & FLAG_HAS_AMT))
6840		e1000e_get_hw_control(adapter);
6841}
6842
6843static void e1000_print_device_info(struct e1000_adapter *adapter)
6844{
6845	struct e1000_hw *hw = &adapter->hw;
6846	struct net_device *netdev = adapter->netdev;
6847	u32 ret_val;
6848	u8 pba_str[E1000_PBANUM_LENGTH];
6849
6850	/* print bus type/speed/width info */
6851	e_info("(PCI Express:2.5GT/s:%s) %pM\n",
6852	       /* bus width */
6853	       ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
6854		"Width x1"),
6855	       /* MAC address */
6856	       netdev->dev_addr);
6857	e_info("Intel(R) PRO/%s Network Connection\n",
6858	       (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
6859	ret_val = e1000_read_pba_string_generic(hw, pba_str,
6860						E1000_PBANUM_LENGTH);
6861	if (ret_val)
6862		strlcpy((char *)pba_str, "Unknown", sizeof(pba_str));
6863	e_info("MAC: %d, PHY: %d, PBA No: %s\n",
6864	       hw->mac.type, hw->phy.type, pba_str);
6865}
6866
6867static void e1000_eeprom_checks(struct e1000_adapter *adapter)
6868{
6869	struct e1000_hw *hw = &adapter->hw;
6870	int ret_val;
6871	u16 buf = 0;
6872
6873	if (hw->mac.type != e1000_82573)
6874		return;
6875
6876	ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
6877	le16_to_cpus(&buf);
6878	if (!ret_val && (!(buf & (1 << 0)))) {
6879		/* Deep Smart Power Down (DSPD) */
6880		dev_warn(&adapter->pdev->dev,
6881			 "Warning: detected DSPD enabled in EEPROM\n");
6882	}
6883}
6884
6885static netdev_features_t e1000_fix_features(struct net_device *netdev,
6886					    netdev_features_t features)
6887{
6888	struct e1000_adapter *adapter = netdev_priv(netdev);
6889	struct e1000_hw *hw = &adapter->hw;
6890
6891	/* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
6892	if ((hw->mac.type >= e1000_pch2lan) && (netdev->mtu > ETH_DATA_LEN))
6893		features &= ~NETIF_F_RXFCS;
6894
6895	return features;
6896}
6897
6898static int e1000_set_features(struct net_device *netdev,
6899			      netdev_features_t features)
6900{
6901	struct e1000_adapter *adapter = netdev_priv(netdev);
6902	netdev_features_t changed = features ^ netdev->features;
6903
6904	if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
6905		adapter->flags |= FLAG_TSO_FORCE;
6906
6907	if (!(changed & (NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_TX |
6908			 NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_RXFCS |
6909			 NETIF_F_RXALL)))
6910		return 0;
6911
6912	if (changed & NETIF_F_RXFCS) {
6913		if (features & NETIF_F_RXFCS) {
6914			adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6915		} else {
6916			/* We need to take it back to defaults, which might mean
6917			 * stripping is still disabled at the adapter level.
6918			 */
6919			if (adapter->flags2 & FLAG2_DFLT_CRC_STRIPPING)
6920				adapter->flags2 |= FLAG2_CRC_STRIPPING;
6921			else
6922				adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6923		}
6924	}
6925
6926	netdev->features = features;
6927
6928	if (netif_running(netdev))
6929		e1000e_reinit_locked(adapter);
6930	else
6931		e1000e_reset(adapter);
6932
6933	return 0;
6934}
6935
6936static const struct net_device_ops e1000e_netdev_ops = {
6937	.ndo_open		= e1000_open,
6938	.ndo_stop		= e1000_close,
6939	.ndo_start_xmit		= e1000_xmit_frame,
6940	.ndo_get_stats64	= e1000e_get_stats64,
6941	.ndo_set_rx_mode	= e1000e_set_rx_mode,
6942	.ndo_set_mac_address	= e1000_set_mac,
6943	.ndo_change_mtu		= e1000_change_mtu,
6944	.ndo_do_ioctl		= e1000_ioctl,
6945	.ndo_tx_timeout		= e1000_tx_timeout,
6946	.ndo_validate_addr	= eth_validate_addr,
6947
6948	.ndo_vlan_rx_add_vid	= e1000_vlan_rx_add_vid,
6949	.ndo_vlan_rx_kill_vid	= e1000_vlan_rx_kill_vid,
6950#ifdef CONFIG_NET_POLL_CONTROLLER
6951	.ndo_poll_controller	= e1000_netpoll,
6952#endif
6953	.ndo_set_features = e1000_set_features,
6954	.ndo_fix_features = e1000_fix_features,
6955	.ndo_features_check	= passthru_features_check,
6956};
6957
6958/**
6959 * e1000_probe - Device Initialization Routine
6960 * @pdev: PCI device information struct
6961 * @ent: entry in e1000_pci_tbl
6962 *
6963 * Returns 0 on success, negative on failure
6964 *
6965 * e1000_probe initializes an adapter identified by a pci_dev structure.
6966 * The OS initialization, configuring of the adapter private structure,
6967 * and a hardware reset occur.
6968 **/
6969static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
6970{
6971	struct net_device *netdev;
6972	struct e1000_adapter *adapter;
6973	struct e1000_hw *hw;
6974	const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
6975	resource_size_t mmio_start, mmio_len;
6976	resource_size_t flash_start, flash_len;
6977	static int cards_found;
6978	u16 aspm_disable_flag = 0;
6979	int bars, i, err, pci_using_dac;
6980	u16 eeprom_data = 0;
6981	u16 eeprom_apme_mask = E1000_EEPROM_APME;
6982	s32 rval = 0;
6983
6984	if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
6985		aspm_disable_flag = PCIE_LINK_STATE_L0S;
6986	if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
6987		aspm_disable_flag |= PCIE_LINK_STATE_L1;
6988	if (aspm_disable_flag)
6989		e1000e_disable_aspm(pdev, aspm_disable_flag);
6990
6991	err = pci_enable_device_mem(pdev);
6992	if (err)
6993		return err;
6994
6995	pci_using_dac = 0;
6996	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
6997	if (!err) {
6998		pci_using_dac = 1;
6999	} else {
7000		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
7001		if (err) {
7002			dev_err(&pdev->dev,
7003				"No usable DMA configuration, aborting\n");
7004			goto err_dma;
7005		}
7006	}
7007
7008	bars = pci_select_bars(pdev, IORESOURCE_MEM);
7009	err = pci_request_selected_regions_exclusive(pdev, bars,
7010						     e1000e_driver_name);
7011	if (err)
7012		goto err_pci_reg;
7013
7014	/* AER (Advanced Error Reporting) hooks */
7015	pci_enable_pcie_error_reporting(pdev);
7016
7017	pci_set_master(pdev);
7018	/* PCI config space info */
7019	err = pci_save_state(pdev);
7020	if (err)
7021		goto err_alloc_etherdev;
7022
7023	err = -ENOMEM;
7024	netdev = alloc_etherdev(sizeof(struct e1000_adapter));
7025	if (!netdev)
7026		goto err_alloc_etherdev;
7027
7028	SET_NETDEV_DEV(netdev, &pdev->dev);
7029
7030	netdev->irq = pdev->irq;
7031
7032	pci_set_drvdata(pdev, netdev);
7033	adapter = netdev_priv(netdev);
7034	hw = &adapter->hw;
7035	adapter->netdev = netdev;
7036	adapter->pdev = pdev;
7037	adapter->ei = ei;
7038	adapter->pba = ei->pba;
7039	adapter->flags = ei->flags;
7040	adapter->flags2 = ei->flags2;
7041	adapter->hw.adapter = adapter;
7042	adapter->hw.mac.type = ei->mac;
7043	adapter->max_hw_frame_size = ei->max_hw_frame_size;
7044	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
7045
7046	mmio_start = pci_resource_start(pdev, 0);
7047	mmio_len = pci_resource_len(pdev, 0);
7048
7049	err = -EIO;
7050	adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
7051	if (!adapter->hw.hw_addr)
7052		goto err_ioremap;
7053
7054	if ((adapter->flags & FLAG_HAS_FLASH) &&
7055	    (pci_resource_flags(pdev, 1) & IORESOURCE_MEM) &&
7056	    (hw->mac.type < e1000_pch_spt)) {
7057		flash_start = pci_resource_start(pdev, 1);
7058		flash_len = pci_resource_len(pdev, 1);
7059		adapter->hw.flash_address = ioremap(flash_start, flash_len);
7060		if (!adapter->hw.flash_address)
7061			goto err_flashmap;
7062	}
7063
7064	/* Set default EEE advertisement */
7065	if (adapter->flags2 & FLAG2_HAS_EEE)
7066		adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
7067
7068	/* construct the net_device struct */
7069	netdev->netdev_ops = &e1000e_netdev_ops;
7070	e1000e_set_ethtool_ops(netdev);
7071	netdev->watchdog_timeo = 5 * HZ;
7072	netif_napi_add(netdev, &adapter->napi, e1000e_poll, 64);
7073	strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
7074
7075	netdev->mem_start = mmio_start;
7076	netdev->mem_end = mmio_start + mmio_len;
7077
7078	adapter->bd_number = cards_found++;
7079
7080	e1000e_check_options(adapter);
7081
7082	/* setup adapter struct */
7083	err = e1000_sw_init(adapter);
7084	if (err)
7085		goto err_sw_init;
7086
7087	memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
7088	memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
7089	memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
7090
7091	err = ei->get_variants(adapter);
7092	if (err)
7093		goto err_hw_init;
7094
7095	if ((adapter->flags & FLAG_IS_ICH) &&
7096	    (adapter->flags & FLAG_READ_ONLY_NVM) &&
7097	    (hw->mac.type < e1000_pch_spt))
7098		e1000e_write_protect_nvm_ich8lan(&adapter->hw);
7099
7100	hw->mac.ops.get_bus_info(&adapter->hw);
7101
7102	adapter->hw.phy.autoneg_wait_to_complete = 0;
7103
7104	/* Copper options */
7105	if (adapter->hw.phy.media_type == e1000_media_type_copper) {
7106		adapter->hw.phy.mdix = AUTO_ALL_MODES;
7107		adapter->hw.phy.disable_polarity_correction = 0;
7108		adapter->hw.phy.ms_type = e1000_ms_hw_default;
7109	}
7110
7111	if (hw->phy.ops.check_reset_block && hw->phy.ops.check_reset_block(hw))
7112		dev_info(&pdev->dev,
7113			 "PHY reset is blocked due to SOL/IDER session.\n");
7114
7115	/* Set initial default active device features */
7116	netdev->features = (NETIF_F_SG |
7117			    NETIF_F_HW_VLAN_CTAG_RX |
7118			    NETIF_F_HW_VLAN_CTAG_TX |
7119			    NETIF_F_TSO |
7120			    NETIF_F_TSO6 |
7121			    NETIF_F_RXHASH |
7122			    NETIF_F_RXCSUM |
7123			    NETIF_F_HW_CSUM);
7124
7125	/* Set user-changeable features (subset of all device features) */
7126	netdev->hw_features = netdev->features;
7127	netdev->hw_features |= NETIF_F_RXFCS;
7128	netdev->priv_flags |= IFF_SUPP_NOFCS;
7129	netdev->hw_features |= NETIF_F_RXALL;
7130
7131	if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
7132		netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
7133
7134	netdev->vlan_features |= (NETIF_F_SG |
7135				  NETIF_F_TSO |
7136				  NETIF_F_TSO6 |
7137				  NETIF_F_HW_CSUM);
7138
7139	netdev->priv_flags |= IFF_UNICAST_FLT;
7140
7141	if (pci_using_dac) {
7142		netdev->features |= NETIF_F_HIGHDMA;
7143		netdev->vlan_features |= NETIF_F_HIGHDMA;
7144	}
7145
7146	if (e1000e_enable_mng_pass_thru(&adapter->hw))
7147		adapter->flags |= FLAG_MNG_PT_ENABLED;
7148
7149	/* before reading the NVM, reset the controller to
7150	 * put the device in a known good starting state
7151	 */
7152	adapter->hw.mac.ops.reset_hw(&adapter->hw);
7153
7154	/* systems with ASPM and others may see the checksum fail on the first
7155	 * attempt. Let's give it a few tries
7156	 */
7157	for (i = 0;; i++) {
7158		if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
7159			break;
7160		if (i == 2) {
7161			dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
7162			err = -EIO;
7163			goto err_eeprom;
7164		}
7165	}
7166
7167	e1000_eeprom_checks(adapter);
7168
7169	/* copy the MAC address */
7170	if (e1000e_read_mac_addr(&adapter->hw))
7171		dev_err(&pdev->dev,
7172			"NVM Read Error while reading MAC address\n");
7173
7174	memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
7175
7176	if (!is_valid_ether_addr(netdev->dev_addr)) {
7177		dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
7178			netdev->dev_addr);
7179		err = -EIO;
7180		goto err_eeprom;
7181	}
7182
7183	init_timer(&adapter->watchdog_timer);
7184	adapter->watchdog_timer.function = e1000_watchdog;
7185	adapter->watchdog_timer.data = (unsigned long)adapter;
7186
7187	init_timer(&adapter->phy_info_timer);
7188	adapter->phy_info_timer.function = e1000_update_phy_info;
7189	adapter->phy_info_timer.data = (unsigned long)adapter;
7190
7191	INIT_WORK(&adapter->reset_task, e1000_reset_task);
7192	INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
7193	INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
7194	INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
7195	INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
7196
7197	/* Initialize link parameters. User can change them with ethtool */
7198	adapter->hw.mac.autoneg = 1;
7199	adapter->fc_autoneg = true;
7200	adapter->hw.fc.requested_mode = e1000_fc_default;
7201	adapter->hw.fc.current_mode = e1000_fc_default;
7202	adapter->hw.phy.autoneg_advertised = 0x2f;
7203
7204	/* Initial Wake on LAN setting - If APM wake is enabled in
7205	 * the EEPROM, enable the ACPI Magic Packet filter
7206	 */
7207	if (adapter->flags & FLAG_APME_IN_WUC) {
7208		/* APME bit in EEPROM is mapped to WUC.APME */
7209		eeprom_data = er32(WUC);
7210		eeprom_apme_mask = E1000_WUC_APME;
7211		if ((hw->mac.type > e1000_ich10lan) &&
7212		    (eeprom_data & E1000_WUC_PHY_WAKE))
7213			adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
7214	} else if (adapter->flags & FLAG_APME_IN_CTRL3) {
7215		if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
7216		    (adapter->hw.bus.func == 1))
7217			rval = e1000_read_nvm(&adapter->hw,
7218					      NVM_INIT_CONTROL3_PORT_B,
7219					      1, &eeprom_data);
7220		else
7221			rval = e1000_read_nvm(&adapter->hw,
7222					      NVM_INIT_CONTROL3_PORT_A,
7223					      1, &eeprom_data);
7224	}
7225
7226	/* fetch WoL from EEPROM */
7227	if (rval)
7228		e_dbg("NVM read error getting WoL initial values: %d\n", rval);
7229	else if (eeprom_data & eeprom_apme_mask)
7230		adapter->eeprom_wol |= E1000_WUFC_MAG;
7231
7232	/* now that we have the eeprom settings, apply the special cases
7233	 * where the eeprom may be wrong or the board simply won't support
7234	 * wake on lan on a particular port
7235	 */
7236	if (!(adapter->flags & FLAG_HAS_WOL))
7237		adapter->eeprom_wol = 0;
7238
7239	/* initialize the wol settings based on the eeprom settings */
7240	adapter->wol = adapter->eeprom_wol;
7241
7242	/* make sure adapter isn't asleep if manageability is enabled */
7243	if (adapter->wol || (adapter->flags & FLAG_MNG_PT_ENABLED) ||
7244	    (hw->mac.ops.check_mng_mode(hw)))
7245		device_wakeup_enable(&pdev->dev);
7246
7247	/* save off EEPROM version number */
7248	rval = e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
7249
7250	if (rval) {
7251		e_dbg("NVM read error getting EEPROM version: %d\n", rval);
7252		adapter->eeprom_vers = 0;
7253	}
7254
7255	/* reset the hardware with the new settings */
7256	e1000e_reset(adapter);
7257
7258	/* If the controller has AMT, do not set DRV_LOAD until the interface
7259	 * is up.  For all other cases, let the f/w know that the h/w is now
7260	 * under the control of the driver.
7261	 */
7262	if (!(adapter->flags & FLAG_HAS_AMT))
7263		e1000e_get_hw_control(adapter);
7264
7265	strlcpy(netdev->name, "eth%d", sizeof(netdev->name));
7266	err = register_netdev(netdev);
7267	if (err)
7268		goto err_register;
7269
7270	/* carrier off reporting is important to ethtool even BEFORE open */
7271	netif_carrier_off(netdev);
7272
7273	/* init PTP hardware clock */
7274	e1000e_ptp_init(adapter);
7275
7276	e1000_print_device_info(adapter);
7277
7278	if (pci_dev_run_wake(pdev))
7279		pm_runtime_put_noidle(&pdev->dev);
7280
7281	return 0;
7282
7283err_register:
7284	if (!(adapter->flags & FLAG_HAS_AMT))
7285		e1000e_release_hw_control(adapter);
7286err_eeprom:
7287	if (hw->phy.ops.check_reset_block && !hw->phy.ops.check_reset_block(hw))
7288		e1000_phy_hw_reset(&adapter->hw);
7289err_hw_init:
7290	kfree(adapter->tx_ring);
7291	kfree(adapter->rx_ring);
7292err_sw_init:
7293	if ((adapter->hw.flash_address) && (hw->mac.type < e1000_pch_spt))
7294		iounmap(adapter->hw.flash_address);
7295	e1000e_reset_interrupt_capability(adapter);
7296err_flashmap:
7297	iounmap(adapter->hw.hw_addr);
7298err_ioremap:
7299	free_netdev(netdev);
7300err_alloc_etherdev:
7301	pci_release_selected_regions(pdev,
7302				     pci_select_bars(pdev, IORESOURCE_MEM));
7303err_pci_reg:
7304err_dma:
7305	pci_disable_device(pdev);
7306	return err;
7307}
7308
7309/**
7310 * e1000_remove - Device Removal Routine
7311 * @pdev: PCI device information struct
7312 *
7313 * e1000_remove is called by the PCI subsystem to alert the driver
7314 * that it should release a PCI device.  The could be caused by a
7315 * Hot-Plug event, or because the driver is going to be removed from
7316 * memory.
7317 **/
7318static void e1000_remove(struct pci_dev *pdev)
7319{
7320	struct net_device *netdev = pci_get_drvdata(pdev);
7321	struct e1000_adapter *adapter = netdev_priv(netdev);
7322	bool down = test_bit(__E1000_DOWN, &adapter->state);
7323
7324	e1000e_ptp_remove(adapter);
7325
7326	/* The timers may be rescheduled, so explicitly disable them
7327	 * from being rescheduled.
7328	 */
7329	if (!down)
7330		set_bit(__E1000_DOWN, &adapter->state);
7331	del_timer_sync(&adapter->watchdog_timer);
7332	del_timer_sync(&adapter->phy_info_timer);
7333
7334	cancel_work_sync(&adapter->reset_task);
7335	cancel_work_sync(&adapter->watchdog_task);
7336	cancel_work_sync(&adapter->downshift_task);
7337	cancel_work_sync(&adapter->update_phy_task);
7338	cancel_work_sync(&adapter->print_hang_task);
7339
7340	if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
7341		cancel_work_sync(&adapter->tx_hwtstamp_work);
7342		if (adapter->tx_hwtstamp_skb) {
7343			dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
7344			adapter->tx_hwtstamp_skb = NULL;
7345		}
7346	}
7347
7348	/* Don't lie to e1000_close() down the road. */
7349	if (!down)
7350		clear_bit(__E1000_DOWN, &adapter->state);
7351	unregister_netdev(netdev);
7352
7353	if (pci_dev_run_wake(pdev))
7354		pm_runtime_get_noresume(&pdev->dev);
7355
7356	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
7357	 * would have already happened in close and is redundant.
7358	 */
7359	e1000e_release_hw_control(adapter);
7360
7361	e1000e_reset_interrupt_capability(adapter);
7362	kfree(adapter->tx_ring);
7363	kfree(adapter->rx_ring);
7364
7365	iounmap(adapter->hw.hw_addr);
7366	if ((adapter->hw.flash_address) &&
7367	    (adapter->hw.mac.type < e1000_pch_spt))
7368		iounmap(adapter->hw.flash_address);
7369	pci_release_selected_regions(pdev,
7370				     pci_select_bars(pdev, IORESOURCE_MEM));
7371
7372	free_netdev(netdev);
7373
7374	/* AER disable */
7375	pci_disable_pcie_error_reporting(pdev);
7376
7377	pci_disable_device(pdev);
7378}
7379
7380/* PCI Error Recovery (ERS) */
7381static const struct pci_error_handlers e1000_err_handler = {
7382	.error_detected = e1000_io_error_detected,
7383	.slot_reset = e1000_io_slot_reset,
7384	.resume = e1000_io_resume,
7385};
7386
7387static const struct pci_device_id e1000_pci_tbl[] = {
7388	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
7389	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
7390	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
7391	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP),
7392	  board_82571 },
7393	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
7394	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
7395	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
7396	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
7397	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
7398
7399	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
7400	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
7401	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
7402	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
7403
7404	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
7405	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
7406	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
7407
7408	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
7409	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
7410	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
7411
7412	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
7413	  board_80003es2lan },
7414	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
7415	  board_80003es2lan },
7416	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
7417	  board_80003es2lan },
7418	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
7419	  board_80003es2lan },
7420
7421	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
7422	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
7423	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
7424	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
7425	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
7426	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
7427	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
7428	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
7429
7430	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
7431	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
7432	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
7433	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
7434	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
7435	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
7436	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
7437	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
7438	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
7439
7440	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
7441	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
7442	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
7443
7444	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
7445	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
7446	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
7447
7448	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
7449	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
7450	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
7451	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
7452
7453	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
7454	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
7455
7456	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_LM), board_pch_lpt },
7457	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_V), board_pch_lpt },
7458	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_LM), board_pch_lpt },
7459	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_V), board_pch_lpt },
7460	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM2), board_pch_lpt },
7461	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V2), board_pch_lpt },
7462	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM3), board_pch_lpt },
7463	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V3), board_pch_lpt },
7464	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM), board_pch_spt },
7465	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V), board_pch_spt },
7466	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM2), board_pch_spt },
7467	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V2), board_pch_spt },
7468
7469	{ 0, 0, 0, 0, 0, 0, 0 }	/* terminate list */
7470};
7471MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
7472
7473static const struct dev_pm_ops e1000_pm_ops = {
7474#ifdef CONFIG_PM_SLEEP
7475	.suspend	= e1000e_pm_suspend,
7476	.resume		= e1000e_pm_resume,
7477	.freeze		= e1000e_pm_freeze,
7478	.thaw		= e1000e_pm_thaw,
7479	.poweroff	= e1000e_pm_suspend,
7480	.restore	= e1000e_pm_resume,
7481#endif
7482	SET_RUNTIME_PM_OPS(e1000e_pm_runtime_suspend, e1000e_pm_runtime_resume,
7483			   e1000e_pm_runtime_idle)
7484};
7485
7486/* PCI Device API Driver */
7487static struct pci_driver e1000_driver = {
7488	.name     = e1000e_driver_name,
7489	.id_table = e1000_pci_tbl,
7490	.probe    = e1000_probe,
7491	.remove   = e1000_remove,
7492	.driver   = {
7493		.pm = &e1000_pm_ops,
7494	},
7495	.shutdown = e1000_shutdown,
7496	.err_handler = &e1000_err_handler
7497};
7498
7499/**
7500 * e1000_init_module - Driver Registration Routine
7501 *
7502 * e1000_init_module is the first routine called when the driver is
7503 * loaded. All it does is register with the PCI subsystem.
7504 **/
7505static int __init e1000_init_module(void)
7506{
7507	int ret;
7508
7509	pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
7510		e1000e_driver_version);
7511	pr_info("Copyright(c) 1999 - 2015 Intel Corporation.\n");
7512	ret = pci_register_driver(&e1000_driver);
7513
7514	return ret;
7515}
7516module_init(e1000_init_module);
7517
7518/**
7519 * e1000_exit_module - Driver Exit Cleanup Routine
7520 *
7521 * e1000_exit_module is called just before the driver is removed
7522 * from memory.
7523 **/
7524static void __exit e1000_exit_module(void)
7525{
7526	pci_unregister_driver(&e1000_driver);
7527}
7528module_exit(e1000_exit_module);
7529
7530MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
7531MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
7532MODULE_LICENSE("GPL");
7533MODULE_VERSION(DRV_VERSION);
7534
7535/* netdev.c */
7536