1 /*******************************************************************************
2 
3   Intel 10 Gigabit PCI Express Linux driver
4   Copyright(c) 1999 - 2012 Intel Corporation.
5 
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9 
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14 
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18 
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21 
22   Contact Information:
23   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25 
26 *******************************************************************************/
27 
28 #include <linux/types.h>
29 #include <linux/module.h>
30 #include <linux/pci.h>
31 #include <linux/netdevice.h>
32 #include <linux/vmalloc.h>
33 #include <linux/string.h>
34 #include <linux/in.h>
35 #include <linux/interrupt.h>
36 #include <linux/ip.h>
37 #include <linux/tcp.h>
38 #include <linux/sctp.h>
39 #include <linux/pkt_sched.h>
40 #include <linux/ipv6.h>
41 #include <linux/slab.h>
42 #include <net/checksum.h>
43 #include <net/ip6_checksum.h>
44 #include <linux/ethtool.h>
45 #include <linux/if.h>
46 #include <linux/if_vlan.h>
47 #include <linux/prefetch.h>
48 #include <scsi/fc/fc_fcoe.h>
49 
50 #include "ixgbe.h"
51 #include "ixgbe_common.h"
52 #include "ixgbe_dcb_82599.h"
53 #include "ixgbe_sriov.h"
54 
55 char ixgbe_driver_name[] = "ixgbe";
56 static const char ixgbe_driver_string[] =
57 			      "Intel(R) 10 Gigabit PCI Express Network Driver";
58 #ifdef IXGBE_FCOE
59 char ixgbe_default_device_descr[] =
60 			      "Intel(R) 10 Gigabit Network Connection";
61 #else
62 static char ixgbe_default_device_descr[] =
63 			      "Intel(R) 10 Gigabit Network Connection";
64 #endif
65 #define MAJ 3
66 #define MIN 9
67 #define BUILD 15
68 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
69 	__stringify(BUILD) "-k"
70 const char ixgbe_driver_version[] = DRV_VERSION;
71 static const char ixgbe_copyright[] =
72 				"Copyright (c) 1999-2012 Intel Corporation.";
73 
74 static const struct ixgbe_info *ixgbe_info_tbl[] = {
75 	[board_82598] = &ixgbe_82598_info,
76 	[board_82599] = &ixgbe_82599_info,
77 	[board_X540] = &ixgbe_X540_info,
78 };
79 
80 /* ixgbe_pci_tbl - PCI Device ID Table
81  *
82  * Wildcard entries (PCI_ANY_ID) should come last
83  * Last entry must be all 0s
84  *
85  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
86  *   Class, Class Mask, private data (not used) }
87  */
88 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
89 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
90 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
91 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
92 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
93 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
94 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
95 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
96 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
97 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
98 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
99 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
100 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
101 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
102 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
103 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
104 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
105 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
106 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
107 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
108 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
109 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
110 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
111 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
112 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
113 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
114 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
115 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
116 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
117 	/* required last entry */
118 	{0, }
119 };
120 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
121 
122 #ifdef CONFIG_IXGBE_DCA
123 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
124 			    void *p);
125 static struct notifier_block dca_notifier = {
126 	.notifier_call = ixgbe_notify_dca,
127 	.next          = NULL,
128 	.priority      = 0
129 };
130 #endif
131 
132 #ifdef CONFIG_PCI_IOV
133 static unsigned int max_vfs;
134 module_param(max_vfs, uint, 0);
135 MODULE_PARM_DESC(max_vfs,
136 		 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63");
137 #endif /* CONFIG_PCI_IOV */
138 
139 static unsigned int allow_unsupported_sfp;
140 module_param(allow_unsupported_sfp, uint, 0);
141 MODULE_PARM_DESC(allow_unsupported_sfp,
142 		 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
143 
144 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
145 static int debug = -1;
146 module_param(debug, int, 0);
147 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
148 
149 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
150 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
151 MODULE_LICENSE("GPL");
152 MODULE_VERSION(DRV_VERSION);
153 
154 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
155 {
156 	if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
157 	    !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
158 		schedule_work(&adapter->service_task);
159 }
160 
161 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
162 {
163 	BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
164 
165 	/* flush memory to make sure state is correct before next watchdog */
166 	smp_mb__before_clear_bit();
167 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
168 }
169 
170 struct ixgbe_reg_info {
171 	u32 ofs;
172 	char *name;
173 };
174 
175 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
176 
177 	/* General Registers */
178 	{IXGBE_CTRL, "CTRL"},
179 	{IXGBE_STATUS, "STATUS"},
180 	{IXGBE_CTRL_EXT, "CTRL_EXT"},
181 
182 	/* Interrupt Registers */
183 	{IXGBE_EICR, "EICR"},
184 
185 	/* RX Registers */
186 	{IXGBE_SRRCTL(0), "SRRCTL"},
187 	{IXGBE_DCA_RXCTRL(0), "DRXCTL"},
188 	{IXGBE_RDLEN(0), "RDLEN"},
189 	{IXGBE_RDH(0), "RDH"},
190 	{IXGBE_RDT(0), "RDT"},
191 	{IXGBE_RXDCTL(0), "RXDCTL"},
192 	{IXGBE_RDBAL(0), "RDBAL"},
193 	{IXGBE_RDBAH(0), "RDBAH"},
194 
195 	/* TX Registers */
196 	{IXGBE_TDBAL(0), "TDBAL"},
197 	{IXGBE_TDBAH(0), "TDBAH"},
198 	{IXGBE_TDLEN(0), "TDLEN"},
199 	{IXGBE_TDH(0), "TDH"},
200 	{IXGBE_TDT(0), "TDT"},
201 	{IXGBE_TXDCTL(0), "TXDCTL"},
202 
203 	/* List Terminator */
204 	{}
205 };
206 
207 
208 /*
209  * ixgbe_regdump - register printout routine
210  */
211 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
212 {
213 	int i = 0, j = 0;
214 	char rname[16];
215 	u32 regs[64];
216 
217 	switch (reginfo->ofs) {
218 	case IXGBE_SRRCTL(0):
219 		for (i = 0; i < 64; i++)
220 			regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
221 		break;
222 	case IXGBE_DCA_RXCTRL(0):
223 		for (i = 0; i < 64; i++)
224 			regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
225 		break;
226 	case IXGBE_RDLEN(0):
227 		for (i = 0; i < 64; i++)
228 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
229 		break;
230 	case IXGBE_RDH(0):
231 		for (i = 0; i < 64; i++)
232 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
233 		break;
234 	case IXGBE_RDT(0):
235 		for (i = 0; i < 64; i++)
236 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
237 		break;
238 	case IXGBE_RXDCTL(0):
239 		for (i = 0; i < 64; i++)
240 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
241 		break;
242 	case IXGBE_RDBAL(0):
243 		for (i = 0; i < 64; i++)
244 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
245 		break;
246 	case IXGBE_RDBAH(0):
247 		for (i = 0; i < 64; i++)
248 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
249 		break;
250 	case IXGBE_TDBAL(0):
251 		for (i = 0; i < 64; i++)
252 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
253 		break;
254 	case IXGBE_TDBAH(0):
255 		for (i = 0; i < 64; i++)
256 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
257 		break;
258 	case IXGBE_TDLEN(0):
259 		for (i = 0; i < 64; i++)
260 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
261 		break;
262 	case IXGBE_TDH(0):
263 		for (i = 0; i < 64; i++)
264 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
265 		break;
266 	case IXGBE_TDT(0):
267 		for (i = 0; i < 64; i++)
268 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
269 		break;
270 	case IXGBE_TXDCTL(0):
271 		for (i = 0; i < 64; i++)
272 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
273 		break;
274 	default:
275 		pr_info("%-15s %08x\n", reginfo->name,
276 			IXGBE_READ_REG(hw, reginfo->ofs));
277 		return;
278 	}
279 
280 	for (i = 0; i < 8; i++) {
281 		snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
282 		pr_err("%-15s", rname);
283 		for (j = 0; j < 8; j++)
284 			pr_cont(" %08x", regs[i*8+j]);
285 		pr_cont("\n");
286 	}
287 
288 }
289 
290 /*
291  * ixgbe_dump - Print registers, tx-rings and rx-rings
292  */
293 static void ixgbe_dump(struct ixgbe_adapter *adapter)
294 {
295 	struct net_device *netdev = adapter->netdev;
296 	struct ixgbe_hw *hw = &adapter->hw;
297 	struct ixgbe_reg_info *reginfo;
298 	int n = 0;
299 	struct ixgbe_ring *tx_ring;
300 	struct ixgbe_tx_buffer *tx_buffer;
301 	union ixgbe_adv_tx_desc *tx_desc;
302 	struct my_u0 { u64 a; u64 b; } *u0;
303 	struct ixgbe_ring *rx_ring;
304 	union ixgbe_adv_rx_desc *rx_desc;
305 	struct ixgbe_rx_buffer *rx_buffer_info;
306 	u32 staterr;
307 	int i = 0;
308 
309 	if (!netif_msg_hw(adapter))
310 		return;
311 
312 	/* Print netdevice Info */
313 	if (netdev) {
314 		dev_info(&adapter->pdev->dev, "Net device Info\n");
315 		pr_info("Device Name     state            "
316 			"trans_start      last_rx\n");
317 		pr_info("%-15s %016lX %016lX %016lX\n",
318 			netdev->name,
319 			netdev->state,
320 			netdev->trans_start,
321 			netdev->last_rx);
322 	}
323 
324 	/* Print Registers */
325 	dev_info(&adapter->pdev->dev, "Register Dump\n");
326 	pr_info(" Register Name   Value\n");
327 	for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
328 	     reginfo->name; reginfo++) {
329 		ixgbe_regdump(hw, reginfo);
330 	}
331 
332 	/* Print TX Ring Summary */
333 	if (!netdev || !netif_running(netdev))
334 		goto exit;
335 
336 	dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
337 	pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
338 	for (n = 0; n < adapter->num_tx_queues; n++) {
339 		tx_ring = adapter->tx_ring[n];
340 		tx_buffer = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
341 		pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
342 			   n, tx_ring->next_to_use, tx_ring->next_to_clean,
343 			   (u64)dma_unmap_addr(tx_buffer, dma),
344 			   dma_unmap_len(tx_buffer, len),
345 			   tx_buffer->next_to_watch,
346 			   (u64)tx_buffer->time_stamp);
347 	}
348 
349 	/* Print TX Rings */
350 	if (!netif_msg_tx_done(adapter))
351 		goto rx_ring_summary;
352 
353 	dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
354 
355 	/* Transmit Descriptor Formats
356 	 *
357 	 * Advanced Transmit Descriptor
358 	 *   +--------------------------------------------------------------+
359 	 * 0 |         Buffer Address [63:0]                                |
360 	 *   +--------------------------------------------------------------+
361 	 * 8 |  PAYLEN  | PORTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
362 	 *   +--------------------------------------------------------------+
363 	 *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
364 	 */
365 
366 	for (n = 0; n < adapter->num_tx_queues; n++) {
367 		tx_ring = adapter->tx_ring[n];
368 		pr_info("------------------------------------\n");
369 		pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
370 		pr_info("------------------------------------\n");
371 		pr_info("T [desc]     [address 63:0  ] "
372 			"[PlPOIdStDDt Ln] [bi->dma       ] "
373 			"leng  ntw timestamp        bi->skb\n");
374 
375 		for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
376 			tx_desc = IXGBE_TX_DESC(tx_ring, i);
377 			tx_buffer = &tx_ring->tx_buffer_info[i];
378 			u0 = (struct my_u0 *)tx_desc;
379 			pr_info("T [0x%03X]    %016llX %016llX %016llX"
380 				" %04X  %p %016llX %p", i,
381 				le64_to_cpu(u0->a),
382 				le64_to_cpu(u0->b),
383 				(u64)dma_unmap_addr(tx_buffer, dma),
384 				dma_unmap_len(tx_buffer, len),
385 				tx_buffer->next_to_watch,
386 				(u64)tx_buffer->time_stamp,
387 				tx_buffer->skb);
388 			if (i == tx_ring->next_to_use &&
389 				i == tx_ring->next_to_clean)
390 				pr_cont(" NTC/U\n");
391 			else if (i == tx_ring->next_to_use)
392 				pr_cont(" NTU\n");
393 			else if (i == tx_ring->next_to_clean)
394 				pr_cont(" NTC\n");
395 			else
396 				pr_cont("\n");
397 
398 			if (netif_msg_pktdata(adapter) &&
399 			    tx_buffer->skb)
400 				print_hex_dump(KERN_INFO, "",
401 					DUMP_PREFIX_ADDRESS, 16, 1,
402 					tx_buffer->skb->data,
403 					dma_unmap_len(tx_buffer, len),
404 					true);
405 		}
406 	}
407 
408 	/* Print RX Rings Summary */
409 rx_ring_summary:
410 	dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
411 	pr_info("Queue [NTU] [NTC]\n");
412 	for (n = 0; n < adapter->num_rx_queues; n++) {
413 		rx_ring = adapter->rx_ring[n];
414 		pr_info("%5d %5X %5X\n",
415 			n, rx_ring->next_to_use, rx_ring->next_to_clean);
416 	}
417 
418 	/* Print RX Rings */
419 	if (!netif_msg_rx_status(adapter))
420 		goto exit;
421 
422 	dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
423 
424 	/* Advanced Receive Descriptor (Read) Format
425 	 *    63                                           1        0
426 	 *    +-----------------------------------------------------+
427 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
428 	 *    +----------------------------------------------+------+
429 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
430 	 *    +-----------------------------------------------------+
431 	 *
432 	 *
433 	 * Advanced Receive Descriptor (Write-Back) Format
434 	 *
435 	 *   63       48 47    32 31  30      21 20 16 15   4 3     0
436 	 *   +------------------------------------------------------+
437 	 * 0 | Packet     IP     |SPH| HDR_LEN   | RSV|Packet|  RSS |
438 	 *   | Checksum   Ident  |   |           |    | Type | Type |
439 	 *   +------------------------------------------------------+
440 	 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
441 	 *   +------------------------------------------------------+
442 	 *   63       48 47    32 31            20 19               0
443 	 */
444 	for (n = 0; n < adapter->num_rx_queues; n++) {
445 		rx_ring = adapter->rx_ring[n];
446 		pr_info("------------------------------------\n");
447 		pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
448 		pr_info("------------------------------------\n");
449 		pr_info("R  [desc]      [ PktBuf     A0] "
450 			"[  HeadBuf   DD] [bi->dma       ] [bi->skb] "
451 			"<-- Adv Rx Read format\n");
452 		pr_info("RWB[desc]      [PcsmIpSHl PtRs] "
453 			"[vl er S cks ln] ---------------- [bi->skb] "
454 			"<-- Adv Rx Write-Back format\n");
455 
456 		for (i = 0; i < rx_ring->count; i++) {
457 			rx_buffer_info = &rx_ring->rx_buffer_info[i];
458 			rx_desc = IXGBE_RX_DESC(rx_ring, i);
459 			u0 = (struct my_u0 *)rx_desc;
460 			staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
461 			if (staterr & IXGBE_RXD_STAT_DD) {
462 				/* Descriptor Done */
463 				pr_info("RWB[0x%03X]     %016llX "
464 					"%016llX ---------------- %p", i,
465 					le64_to_cpu(u0->a),
466 					le64_to_cpu(u0->b),
467 					rx_buffer_info->skb);
468 			} else {
469 				pr_info("R  [0x%03X]     %016llX "
470 					"%016llX %016llX %p", i,
471 					le64_to_cpu(u0->a),
472 					le64_to_cpu(u0->b),
473 					(u64)rx_buffer_info->dma,
474 					rx_buffer_info->skb);
475 
476 				if (netif_msg_pktdata(adapter) &&
477 				    rx_buffer_info->dma) {
478 					print_hex_dump(KERN_INFO, "",
479 					   DUMP_PREFIX_ADDRESS, 16, 1,
480 					   page_address(rx_buffer_info->page) +
481 						    rx_buffer_info->page_offset,
482 					   ixgbe_rx_bufsz(rx_ring), true);
483 				}
484 			}
485 
486 			if (i == rx_ring->next_to_use)
487 				pr_cont(" NTU\n");
488 			else if (i == rx_ring->next_to_clean)
489 				pr_cont(" NTC\n");
490 			else
491 				pr_cont("\n");
492 
493 		}
494 	}
495 
496 exit:
497 	return;
498 }
499 
500 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
501 {
502 	u32 ctrl_ext;
503 
504 	/* Let firmware take over control of h/w */
505 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
506 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
507 			ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
508 }
509 
510 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
511 {
512 	u32 ctrl_ext;
513 
514 	/* Let firmware know the driver has taken over */
515 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
516 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
517 			ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
518 }
519 
520 /**
521  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
522  * @adapter: pointer to adapter struct
523  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
524  * @queue: queue to map the corresponding interrupt to
525  * @msix_vector: the vector to map to the corresponding queue
526  *
527  */
528 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
529 			   u8 queue, u8 msix_vector)
530 {
531 	u32 ivar, index;
532 	struct ixgbe_hw *hw = &adapter->hw;
533 	switch (hw->mac.type) {
534 	case ixgbe_mac_82598EB:
535 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
536 		if (direction == -1)
537 			direction = 0;
538 		index = (((direction * 64) + queue) >> 2) & 0x1F;
539 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
540 		ivar &= ~(0xFF << (8 * (queue & 0x3)));
541 		ivar |= (msix_vector << (8 * (queue & 0x3)));
542 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
543 		break;
544 	case ixgbe_mac_82599EB:
545 	case ixgbe_mac_X540:
546 		if (direction == -1) {
547 			/* other causes */
548 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
549 			index = ((queue & 1) * 8);
550 			ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
551 			ivar &= ~(0xFF << index);
552 			ivar |= (msix_vector << index);
553 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
554 			break;
555 		} else {
556 			/* tx or rx causes */
557 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
558 			index = ((16 * (queue & 1)) + (8 * direction));
559 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
560 			ivar &= ~(0xFF << index);
561 			ivar |= (msix_vector << index);
562 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
563 			break;
564 		}
565 	default:
566 		break;
567 	}
568 }
569 
570 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
571 					  u64 qmask)
572 {
573 	u32 mask;
574 
575 	switch (adapter->hw.mac.type) {
576 	case ixgbe_mac_82598EB:
577 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
578 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
579 		break;
580 	case ixgbe_mac_82599EB:
581 	case ixgbe_mac_X540:
582 		mask = (qmask & 0xFFFFFFFF);
583 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
584 		mask = (qmask >> 32);
585 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
586 		break;
587 	default:
588 		break;
589 	}
590 }
591 
592 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *ring,
593 				      struct ixgbe_tx_buffer *tx_buffer)
594 {
595 	if (tx_buffer->skb) {
596 		dev_kfree_skb_any(tx_buffer->skb);
597 		if (dma_unmap_len(tx_buffer, len))
598 			dma_unmap_single(ring->dev,
599 					 dma_unmap_addr(tx_buffer, dma),
600 					 dma_unmap_len(tx_buffer, len),
601 					 DMA_TO_DEVICE);
602 	} else if (dma_unmap_len(tx_buffer, len)) {
603 		dma_unmap_page(ring->dev,
604 			       dma_unmap_addr(tx_buffer, dma),
605 			       dma_unmap_len(tx_buffer, len),
606 			       DMA_TO_DEVICE);
607 	}
608 	tx_buffer->next_to_watch = NULL;
609 	tx_buffer->skb = NULL;
610 	dma_unmap_len_set(tx_buffer, len, 0);
611 	/* tx_buffer must be completely set up in the transmit path */
612 }
613 
614 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
615 {
616 	struct ixgbe_hw *hw = &adapter->hw;
617 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
618 	int i;
619 	u32 data;
620 
621 	if ((hw->fc.current_mode != ixgbe_fc_full) &&
622 	    (hw->fc.current_mode != ixgbe_fc_rx_pause))
623 		return;
624 
625 	switch (hw->mac.type) {
626 	case ixgbe_mac_82598EB:
627 		data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
628 		break;
629 	default:
630 		data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
631 	}
632 	hwstats->lxoffrxc += data;
633 
634 	/* refill credits (no tx hang) if we received xoff */
635 	if (!data)
636 		return;
637 
638 	for (i = 0; i < adapter->num_tx_queues; i++)
639 		clear_bit(__IXGBE_HANG_CHECK_ARMED,
640 			  &adapter->tx_ring[i]->state);
641 }
642 
643 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
644 {
645 	struct ixgbe_hw *hw = &adapter->hw;
646 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
647 	u32 xoff[8] = {0};
648 	int i;
649 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
650 
651 	if (adapter->ixgbe_ieee_pfc)
652 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
653 
654 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
655 		ixgbe_update_xoff_rx_lfc(adapter);
656 		return;
657 	}
658 
659 	/* update stats for each tc, only valid with PFC enabled */
660 	for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
661 		switch (hw->mac.type) {
662 		case ixgbe_mac_82598EB:
663 			xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
664 			break;
665 		default:
666 			xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
667 		}
668 		hwstats->pxoffrxc[i] += xoff[i];
669 	}
670 
671 	/* disarm tx queues that have received xoff frames */
672 	for (i = 0; i < adapter->num_tx_queues; i++) {
673 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
674 		u8 tc = tx_ring->dcb_tc;
675 
676 		if (xoff[tc])
677 			clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
678 	}
679 }
680 
681 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
682 {
683 	return ring->stats.packets;
684 }
685 
686 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
687 {
688 	struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
689 	struct ixgbe_hw *hw = &adapter->hw;
690 
691 	u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
692 	u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
693 
694 	if (head != tail)
695 		return (head < tail) ?
696 			tail - head : (tail + ring->count - head);
697 
698 	return 0;
699 }
700 
701 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
702 {
703 	u32 tx_done = ixgbe_get_tx_completed(tx_ring);
704 	u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
705 	u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
706 	bool ret = false;
707 
708 	clear_check_for_tx_hang(tx_ring);
709 
710 	/*
711 	 * Check for a hung queue, but be thorough. This verifies
712 	 * that a transmit has been completed since the previous
713 	 * check AND there is at least one packet pending. The
714 	 * ARMED bit is set to indicate a potential hang. The
715 	 * bit is cleared if a pause frame is received to remove
716 	 * false hang detection due to PFC or 802.3x frames. By
717 	 * requiring this to fail twice we avoid races with
718 	 * pfc clearing the ARMED bit and conditions where we
719 	 * run the check_tx_hang logic with a transmit completion
720 	 * pending but without time to complete it yet.
721 	 */
722 	if ((tx_done_old == tx_done) && tx_pending) {
723 		/* make sure it is true for two checks in a row */
724 		ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
725 				       &tx_ring->state);
726 	} else {
727 		/* update completed stats and continue */
728 		tx_ring->tx_stats.tx_done_old = tx_done;
729 		/* reset the countdown */
730 		clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
731 	}
732 
733 	return ret;
734 }
735 
736 /**
737  * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
738  * @adapter: driver private struct
739  **/
740 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
741 {
742 
743 	/* Do the reset outside of interrupt context */
744 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
745 		adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
746 		ixgbe_service_event_schedule(adapter);
747 	}
748 }
749 
750 /**
751  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
752  * @q_vector: structure containing interrupt and ring information
753  * @tx_ring: tx ring to clean
754  **/
755 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
756 			       struct ixgbe_ring *tx_ring)
757 {
758 	struct ixgbe_adapter *adapter = q_vector->adapter;
759 	struct ixgbe_tx_buffer *tx_buffer;
760 	union ixgbe_adv_tx_desc *tx_desc;
761 	unsigned int total_bytes = 0, total_packets = 0;
762 	unsigned int budget = q_vector->tx.work_limit;
763 	unsigned int i = tx_ring->next_to_clean;
764 
765 	if (test_bit(__IXGBE_DOWN, &adapter->state))
766 		return true;
767 
768 	tx_buffer = &tx_ring->tx_buffer_info[i];
769 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
770 	i -= tx_ring->count;
771 
772 	do {
773 		union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
774 
775 		/* if next_to_watch is not set then there is no work pending */
776 		if (!eop_desc)
777 			break;
778 
779 		/* prevent any other reads prior to eop_desc */
780 		rmb();
781 
782 		/* if DD is not set pending work has not been completed */
783 		if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
784 			break;
785 
786 		/* clear next_to_watch to prevent false hangs */
787 		tx_buffer->next_to_watch = NULL;
788 
789 		/* update the statistics for this packet */
790 		total_bytes += tx_buffer->bytecount;
791 		total_packets += tx_buffer->gso_segs;
792 
793 #ifdef CONFIG_IXGBE_PTP
794 		if (unlikely(tx_buffer->tx_flags & IXGBE_TX_FLAGS_TSTAMP))
795 			ixgbe_ptp_tx_hwtstamp(q_vector, tx_buffer->skb);
796 #endif
797 
798 		/* free the skb */
799 		dev_kfree_skb_any(tx_buffer->skb);
800 
801 		/* unmap skb header data */
802 		dma_unmap_single(tx_ring->dev,
803 				 dma_unmap_addr(tx_buffer, dma),
804 				 dma_unmap_len(tx_buffer, len),
805 				 DMA_TO_DEVICE);
806 
807 		/* clear tx_buffer data */
808 		tx_buffer->skb = NULL;
809 		dma_unmap_len_set(tx_buffer, len, 0);
810 
811 		/* unmap remaining buffers */
812 		while (tx_desc != eop_desc) {
813 			tx_buffer++;
814 			tx_desc++;
815 			i++;
816 			if (unlikely(!i)) {
817 				i -= tx_ring->count;
818 				tx_buffer = tx_ring->tx_buffer_info;
819 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
820 			}
821 
822 			/* unmap any remaining paged data */
823 			if (dma_unmap_len(tx_buffer, len)) {
824 				dma_unmap_page(tx_ring->dev,
825 					       dma_unmap_addr(tx_buffer, dma),
826 					       dma_unmap_len(tx_buffer, len),
827 					       DMA_TO_DEVICE);
828 				dma_unmap_len_set(tx_buffer, len, 0);
829 			}
830 		}
831 
832 		/* move us one more past the eop_desc for start of next pkt */
833 		tx_buffer++;
834 		tx_desc++;
835 		i++;
836 		if (unlikely(!i)) {
837 			i -= tx_ring->count;
838 			tx_buffer = tx_ring->tx_buffer_info;
839 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
840 		}
841 
842 		/* issue prefetch for next Tx descriptor */
843 		prefetch(tx_desc);
844 
845 		/* update budget accounting */
846 		budget--;
847 	} while (likely(budget));
848 
849 	i += tx_ring->count;
850 	tx_ring->next_to_clean = i;
851 	u64_stats_update_begin(&tx_ring->syncp);
852 	tx_ring->stats.bytes += total_bytes;
853 	tx_ring->stats.packets += total_packets;
854 	u64_stats_update_end(&tx_ring->syncp);
855 	q_vector->tx.total_bytes += total_bytes;
856 	q_vector->tx.total_packets += total_packets;
857 
858 	if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
859 		/* schedule immediate reset if we believe we hung */
860 		struct ixgbe_hw *hw = &adapter->hw;
861 		e_err(drv, "Detected Tx Unit Hang\n"
862 			"  Tx Queue             <%d>\n"
863 			"  TDH, TDT             <%x>, <%x>\n"
864 			"  next_to_use          <%x>\n"
865 			"  next_to_clean        <%x>\n"
866 			"tx_buffer_info[next_to_clean]\n"
867 			"  time_stamp           <%lx>\n"
868 			"  jiffies              <%lx>\n",
869 			tx_ring->queue_index,
870 			IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
871 			IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
872 			tx_ring->next_to_use, i,
873 			tx_ring->tx_buffer_info[i].time_stamp, jiffies);
874 
875 		netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
876 
877 		e_info(probe,
878 		       "tx hang %d detected on queue %d, resetting adapter\n",
879 			adapter->tx_timeout_count + 1, tx_ring->queue_index);
880 
881 		/* schedule immediate reset if we believe we hung */
882 		ixgbe_tx_timeout_reset(adapter);
883 
884 		/* the adapter is about to reset, no point in enabling stuff */
885 		return true;
886 	}
887 
888 	netdev_tx_completed_queue(txring_txq(tx_ring),
889 				  total_packets, total_bytes);
890 
891 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
892 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
893 		     (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
894 		/* Make sure that anybody stopping the queue after this
895 		 * sees the new next_to_clean.
896 		 */
897 		smp_mb();
898 		if (__netif_subqueue_stopped(tx_ring->netdev,
899 					     tx_ring->queue_index)
900 		    && !test_bit(__IXGBE_DOWN, &adapter->state)) {
901 			netif_wake_subqueue(tx_ring->netdev,
902 					    tx_ring->queue_index);
903 			++tx_ring->tx_stats.restart_queue;
904 		}
905 	}
906 
907 	return !!budget;
908 }
909 
910 #ifdef CONFIG_IXGBE_DCA
911 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
912 				struct ixgbe_ring *tx_ring,
913 				int cpu)
914 {
915 	struct ixgbe_hw *hw = &adapter->hw;
916 	u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
917 	u16 reg_offset;
918 
919 	switch (hw->mac.type) {
920 	case ixgbe_mac_82598EB:
921 		reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
922 		break;
923 	case ixgbe_mac_82599EB:
924 	case ixgbe_mac_X540:
925 		reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
926 		txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
927 		break;
928 	default:
929 		/* for unknown hardware do not write register */
930 		return;
931 	}
932 
933 	/*
934 	 * We can enable relaxed ordering for reads, but not writes when
935 	 * DCA is enabled.  This is due to a known issue in some chipsets
936 	 * which will cause the DCA tag to be cleared.
937 	 */
938 	txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
939 		  IXGBE_DCA_TXCTRL_DATA_RRO_EN |
940 		  IXGBE_DCA_TXCTRL_DESC_DCA_EN;
941 
942 	IXGBE_WRITE_REG(hw, reg_offset, txctrl);
943 }
944 
945 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
946 				struct ixgbe_ring *rx_ring,
947 				int cpu)
948 {
949 	struct ixgbe_hw *hw = &adapter->hw;
950 	u32 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
951 	u8 reg_idx = rx_ring->reg_idx;
952 
953 
954 	switch (hw->mac.type) {
955 	case ixgbe_mac_82599EB:
956 	case ixgbe_mac_X540:
957 		rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
958 		break;
959 	default:
960 		break;
961 	}
962 
963 	/*
964 	 * We can enable relaxed ordering for reads, but not writes when
965 	 * DCA is enabled.  This is due to a known issue in some chipsets
966 	 * which will cause the DCA tag to be cleared.
967 	 */
968 	rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
969 		  IXGBE_DCA_RXCTRL_DATA_DCA_EN |
970 		  IXGBE_DCA_RXCTRL_DESC_DCA_EN;
971 
972 	IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
973 }
974 
975 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
976 {
977 	struct ixgbe_adapter *adapter = q_vector->adapter;
978 	struct ixgbe_ring *ring;
979 	int cpu = get_cpu();
980 
981 	if (q_vector->cpu == cpu)
982 		goto out_no_update;
983 
984 	ixgbe_for_each_ring(ring, q_vector->tx)
985 		ixgbe_update_tx_dca(adapter, ring, cpu);
986 
987 	ixgbe_for_each_ring(ring, q_vector->rx)
988 		ixgbe_update_rx_dca(adapter, ring, cpu);
989 
990 	q_vector->cpu = cpu;
991 out_no_update:
992 	put_cpu();
993 }
994 
995 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
996 {
997 	int i;
998 
999 	if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1000 		return;
1001 
1002 	/* always use CB2 mode, difference is masked in the CB driver */
1003 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1004 
1005 	for (i = 0; i < adapter->num_q_vectors; i++) {
1006 		adapter->q_vector[i]->cpu = -1;
1007 		ixgbe_update_dca(adapter->q_vector[i]);
1008 	}
1009 }
1010 
1011 static int __ixgbe_notify_dca(struct device *dev, void *data)
1012 {
1013 	struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1014 	unsigned long event = *(unsigned long *)data;
1015 
1016 	if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1017 		return 0;
1018 
1019 	switch (event) {
1020 	case DCA_PROVIDER_ADD:
1021 		/* if we're already enabled, don't do it again */
1022 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1023 			break;
1024 		if (dca_add_requester(dev) == 0) {
1025 			adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1026 			ixgbe_setup_dca(adapter);
1027 			break;
1028 		}
1029 		/* Fall Through since DCA is disabled. */
1030 	case DCA_PROVIDER_REMOVE:
1031 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1032 			dca_remove_requester(dev);
1033 			adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1034 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1035 		}
1036 		break;
1037 	}
1038 
1039 	return 0;
1040 }
1041 
1042 #endif /* CONFIG_IXGBE_DCA */
1043 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1044 				 union ixgbe_adv_rx_desc *rx_desc,
1045 				 struct sk_buff *skb)
1046 {
1047 	if (ring->netdev->features & NETIF_F_RXHASH)
1048 		skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
1049 }
1050 
1051 #ifdef IXGBE_FCOE
1052 /**
1053  * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1054  * @ring: structure containing ring specific data
1055  * @rx_desc: advanced rx descriptor
1056  *
1057  * Returns : true if it is FCoE pkt
1058  */
1059 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1060 				    union ixgbe_adv_rx_desc *rx_desc)
1061 {
1062 	__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1063 
1064 	return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1065 	       ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1066 		(cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1067 			     IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1068 }
1069 
1070 #endif /* IXGBE_FCOE */
1071 /**
1072  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1073  * @ring: structure containing ring specific data
1074  * @rx_desc: current Rx descriptor being processed
1075  * @skb: skb currently being received and modified
1076  **/
1077 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1078 				     union ixgbe_adv_rx_desc *rx_desc,
1079 				     struct sk_buff *skb)
1080 {
1081 	skb_checksum_none_assert(skb);
1082 
1083 	/* Rx csum disabled */
1084 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
1085 		return;
1086 
1087 	/* if IP and error */
1088 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1089 	    ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1090 		ring->rx_stats.csum_err++;
1091 		return;
1092 	}
1093 
1094 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1095 		return;
1096 
1097 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1098 		__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1099 
1100 		/*
1101 		 * 82599 errata, UDP frames with a 0 checksum can be marked as
1102 		 * checksum errors.
1103 		 */
1104 		if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1105 		    test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1106 			return;
1107 
1108 		ring->rx_stats.csum_err++;
1109 		return;
1110 	}
1111 
1112 	/* It must be a TCP or UDP packet with a valid checksum */
1113 	skb->ip_summed = CHECKSUM_UNNECESSARY;
1114 }
1115 
1116 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1117 {
1118 	rx_ring->next_to_use = val;
1119 
1120 	/* update next to alloc since we have filled the ring */
1121 	rx_ring->next_to_alloc = val;
1122 	/*
1123 	 * Force memory writes to complete before letting h/w
1124 	 * know there are new descriptors to fetch.  (Only
1125 	 * applicable for weak-ordered memory model archs,
1126 	 * such as IA-64).
1127 	 */
1128 	wmb();
1129 	writel(val, rx_ring->tail);
1130 }
1131 
1132 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1133 				    struct ixgbe_rx_buffer *bi)
1134 {
1135 	struct page *page = bi->page;
1136 	dma_addr_t dma = bi->dma;
1137 
1138 	/* since we are recycling buffers we should seldom need to alloc */
1139 	if (likely(dma))
1140 		return true;
1141 
1142 	/* alloc new page for storage */
1143 	if (likely(!page)) {
1144 		page = __skb_alloc_pages(GFP_ATOMIC | __GFP_COLD | __GFP_COMP,
1145 					 bi->skb, ixgbe_rx_pg_order(rx_ring));
1146 		if (unlikely(!page)) {
1147 			rx_ring->rx_stats.alloc_rx_page_failed++;
1148 			return false;
1149 		}
1150 		bi->page = page;
1151 	}
1152 
1153 	/* map page for use */
1154 	dma = dma_map_page(rx_ring->dev, page, 0,
1155 			   ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1156 
1157 	/*
1158 	 * if mapping failed free memory back to system since
1159 	 * there isn't much point in holding memory we can't use
1160 	 */
1161 	if (dma_mapping_error(rx_ring->dev, dma)) {
1162 		__free_pages(page, ixgbe_rx_pg_order(rx_ring));
1163 		bi->page = NULL;
1164 
1165 		rx_ring->rx_stats.alloc_rx_page_failed++;
1166 		return false;
1167 	}
1168 
1169 	bi->dma = dma;
1170 	bi->page_offset ^= ixgbe_rx_bufsz(rx_ring);
1171 
1172 	return true;
1173 }
1174 
1175 /**
1176  * ixgbe_alloc_rx_buffers - Replace used receive buffers
1177  * @rx_ring: ring to place buffers on
1178  * @cleaned_count: number of buffers to replace
1179  **/
1180 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1181 {
1182 	union ixgbe_adv_rx_desc *rx_desc;
1183 	struct ixgbe_rx_buffer *bi;
1184 	u16 i = rx_ring->next_to_use;
1185 
1186 	/* nothing to do */
1187 	if (!cleaned_count)
1188 		return;
1189 
1190 	rx_desc = IXGBE_RX_DESC(rx_ring, i);
1191 	bi = &rx_ring->rx_buffer_info[i];
1192 	i -= rx_ring->count;
1193 
1194 	do {
1195 		if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1196 			break;
1197 
1198 		/*
1199 		 * Refresh the desc even if buffer_addrs didn't change
1200 		 * because each write-back erases this info.
1201 		 */
1202 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1203 
1204 		rx_desc++;
1205 		bi++;
1206 		i++;
1207 		if (unlikely(!i)) {
1208 			rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1209 			bi = rx_ring->rx_buffer_info;
1210 			i -= rx_ring->count;
1211 		}
1212 
1213 		/* clear the hdr_addr for the next_to_use descriptor */
1214 		rx_desc->read.hdr_addr = 0;
1215 
1216 		cleaned_count--;
1217 	} while (cleaned_count);
1218 
1219 	i += rx_ring->count;
1220 
1221 	if (rx_ring->next_to_use != i)
1222 		ixgbe_release_rx_desc(rx_ring, i);
1223 }
1224 
1225 /**
1226  * ixgbe_get_headlen - determine size of header for RSC/LRO/GRO/FCOE
1227  * @data: pointer to the start of the headers
1228  * @max_len: total length of section to find headers in
1229  *
1230  * This function is meant to determine the length of headers that will
1231  * be recognized by hardware for LRO, GRO, and RSC offloads.  The main
1232  * motivation of doing this is to only perform one pull for IPv4 TCP
1233  * packets so that we can do basic things like calculating the gso_size
1234  * based on the average data per packet.
1235  **/
1236 static unsigned int ixgbe_get_headlen(unsigned char *data,
1237 				      unsigned int max_len)
1238 {
1239 	union {
1240 		unsigned char *network;
1241 		/* l2 headers */
1242 		struct ethhdr *eth;
1243 		struct vlan_hdr *vlan;
1244 		/* l3 headers */
1245 		struct iphdr *ipv4;
1246 	} hdr;
1247 	__be16 protocol;
1248 	u8 nexthdr = 0;	/* default to not TCP */
1249 	u8 hlen;
1250 
1251 	/* this should never happen, but better safe than sorry */
1252 	if (max_len < ETH_HLEN)
1253 		return max_len;
1254 
1255 	/* initialize network frame pointer */
1256 	hdr.network = data;
1257 
1258 	/* set first protocol and move network header forward */
1259 	protocol = hdr.eth->h_proto;
1260 	hdr.network += ETH_HLEN;
1261 
1262 	/* handle any vlan tag if present */
1263 	if (protocol == __constant_htons(ETH_P_8021Q)) {
1264 		if ((hdr.network - data) > (max_len - VLAN_HLEN))
1265 			return max_len;
1266 
1267 		protocol = hdr.vlan->h_vlan_encapsulated_proto;
1268 		hdr.network += VLAN_HLEN;
1269 	}
1270 
1271 	/* handle L3 protocols */
1272 	if (protocol == __constant_htons(ETH_P_IP)) {
1273 		if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
1274 			return max_len;
1275 
1276 		/* access ihl as a u8 to avoid unaligned access on ia64 */
1277 		hlen = (hdr.network[0] & 0x0F) << 2;
1278 
1279 		/* verify hlen meets minimum size requirements */
1280 		if (hlen < sizeof(struct iphdr))
1281 			return hdr.network - data;
1282 
1283 		/* record next protocol */
1284 		nexthdr = hdr.ipv4->protocol;
1285 		hdr.network += hlen;
1286 #ifdef IXGBE_FCOE
1287 	} else if (protocol == __constant_htons(ETH_P_FCOE)) {
1288 		if ((hdr.network - data) > (max_len - FCOE_HEADER_LEN))
1289 			return max_len;
1290 		hdr.network += FCOE_HEADER_LEN;
1291 #endif
1292 	} else {
1293 		return hdr.network - data;
1294 	}
1295 
1296 	/* finally sort out TCP */
1297 	if (nexthdr == IPPROTO_TCP) {
1298 		if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
1299 			return max_len;
1300 
1301 		/* access doff as a u8 to avoid unaligned access on ia64 */
1302 		hlen = (hdr.network[12] & 0xF0) >> 2;
1303 
1304 		/* verify hlen meets minimum size requirements */
1305 		if (hlen < sizeof(struct tcphdr))
1306 			return hdr.network - data;
1307 
1308 		hdr.network += hlen;
1309 	}
1310 
1311 	/*
1312 	 * If everything has gone correctly hdr.network should be the
1313 	 * data section of the packet and will be the end of the header.
1314 	 * If not then it probably represents the end of the last recognized
1315 	 * header.
1316 	 */
1317 	if ((hdr.network - data) < max_len)
1318 		return hdr.network - data;
1319 	else
1320 		return max_len;
1321 }
1322 
1323 static void ixgbe_get_rsc_cnt(struct ixgbe_ring *rx_ring,
1324 			      union ixgbe_adv_rx_desc *rx_desc,
1325 			      struct sk_buff *skb)
1326 {
1327 	__le32 rsc_enabled;
1328 	u32 rsc_cnt;
1329 
1330 	if (!ring_is_rsc_enabled(rx_ring))
1331 		return;
1332 
1333 	rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1334 		      cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1335 
1336 	/* If this is an RSC frame rsc_cnt should be non-zero */
1337 	if (!rsc_enabled)
1338 		return;
1339 
1340 	rsc_cnt = le32_to_cpu(rsc_enabled);
1341 	rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1342 
1343 	IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1344 }
1345 
1346 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1347 				   struct sk_buff *skb)
1348 {
1349 	u16 hdr_len = skb_headlen(skb);
1350 
1351 	/* set gso_size to avoid messing up TCP MSS */
1352 	skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1353 						 IXGBE_CB(skb)->append_cnt);
1354 }
1355 
1356 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1357 				   struct sk_buff *skb)
1358 {
1359 	/* if append_cnt is 0 then frame is not RSC */
1360 	if (!IXGBE_CB(skb)->append_cnt)
1361 		return;
1362 
1363 	rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1364 	rx_ring->rx_stats.rsc_flush++;
1365 
1366 	ixgbe_set_rsc_gso_size(rx_ring, skb);
1367 
1368 	/* gso_size is computed using append_cnt so always clear it last */
1369 	IXGBE_CB(skb)->append_cnt = 0;
1370 }
1371 
1372 /**
1373  * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1374  * @rx_ring: rx descriptor ring packet is being transacted on
1375  * @rx_desc: pointer to the EOP Rx descriptor
1376  * @skb: pointer to current skb being populated
1377  *
1378  * This function checks the ring, descriptor, and packet information in
1379  * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1380  * other fields within the skb.
1381  **/
1382 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1383 				     union ixgbe_adv_rx_desc *rx_desc,
1384 				     struct sk_buff *skb)
1385 {
1386 	struct net_device *dev = rx_ring->netdev;
1387 
1388 	ixgbe_update_rsc_stats(rx_ring, skb);
1389 
1390 	ixgbe_rx_hash(rx_ring, rx_desc, skb);
1391 
1392 	ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1393 
1394 #ifdef CONFIG_IXGBE_PTP
1395 	ixgbe_ptp_rx_hwtstamp(rx_ring->q_vector, rx_desc, skb);
1396 #endif
1397 
1398 	if ((dev->features & NETIF_F_HW_VLAN_RX) &&
1399 	    ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1400 		u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1401 		__vlan_hwaccel_put_tag(skb, vid);
1402 	}
1403 
1404 	skb_record_rx_queue(skb, rx_ring->queue_index);
1405 
1406 	skb->protocol = eth_type_trans(skb, dev);
1407 }
1408 
1409 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1410 			 struct sk_buff *skb)
1411 {
1412 	struct ixgbe_adapter *adapter = q_vector->adapter;
1413 
1414 	if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1415 		napi_gro_receive(&q_vector->napi, skb);
1416 	else
1417 		netif_rx(skb);
1418 }
1419 
1420 /**
1421  * ixgbe_is_non_eop - process handling of non-EOP buffers
1422  * @rx_ring: Rx ring being processed
1423  * @rx_desc: Rx descriptor for current buffer
1424  * @skb: Current socket buffer containing buffer in progress
1425  *
1426  * This function updates next to clean.  If the buffer is an EOP buffer
1427  * this function exits returning false, otherwise it will place the
1428  * sk_buff in the next buffer to be chained and return true indicating
1429  * that this is in fact a non-EOP buffer.
1430  **/
1431 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1432 			     union ixgbe_adv_rx_desc *rx_desc,
1433 			     struct sk_buff *skb)
1434 {
1435 	u32 ntc = rx_ring->next_to_clean + 1;
1436 
1437 	/* fetch, update, and store next to clean */
1438 	ntc = (ntc < rx_ring->count) ? ntc : 0;
1439 	rx_ring->next_to_clean = ntc;
1440 
1441 	prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1442 
1443 	if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1444 		return false;
1445 
1446 	/* append_cnt indicates packet is RSC, if so fetch nextp */
1447 	if (IXGBE_CB(skb)->append_cnt) {
1448 		ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1449 		ntc &= IXGBE_RXDADV_NEXTP_MASK;
1450 		ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1451 	}
1452 
1453 	/* place skb in next buffer to be received */
1454 	rx_ring->rx_buffer_info[ntc].skb = skb;
1455 	rx_ring->rx_stats.non_eop_descs++;
1456 
1457 	return true;
1458 }
1459 
1460 /**
1461  * ixgbe_cleanup_headers - Correct corrupted or empty headers
1462  * @rx_ring: rx descriptor ring packet is being transacted on
1463  * @rx_desc: pointer to the EOP Rx descriptor
1464  * @skb: pointer to current skb being fixed
1465  *
1466  * Check for corrupted packet headers caused by senders on the local L2
1467  * embedded NIC switch not setting up their Tx Descriptors right.  These
1468  * should be very rare.
1469  *
1470  * Also address the case where we are pulling data in on pages only
1471  * and as such no data is present in the skb header.
1472  *
1473  * In addition if skb is not at least 60 bytes we need to pad it so that
1474  * it is large enough to qualify as a valid Ethernet frame.
1475  *
1476  * Returns true if an error was encountered and skb was freed.
1477  **/
1478 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1479 				  union ixgbe_adv_rx_desc *rx_desc,
1480 				  struct sk_buff *skb)
1481 {
1482 	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1483 	struct net_device *netdev = rx_ring->netdev;
1484 	unsigned char *va;
1485 	unsigned int pull_len;
1486 
1487 	/* if the page was released unmap it, else just sync our portion */
1488 	if (unlikely(IXGBE_CB(skb)->page_released)) {
1489 		dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1490 			       ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1491 		IXGBE_CB(skb)->page_released = false;
1492 	} else {
1493 		dma_sync_single_range_for_cpu(rx_ring->dev,
1494 					      IXGBE_CB(skb)->dma,
1495 					      frag->page_offset,
1496 					      ixgbe_rx_bufsz(rx_ring),
1497 					      DMA_FROM_DEVICE);
1498 	}
1499 	IXGBE_CB(skb)->dma = 0;
1500 
1501 	/* verify that the packet does not have any known errors */
1502 	if (unlikely(ixgbe_test_staterr(rx_desc,
1503 					IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1504 	    !(netdev->features & NETIF_F_RXALL))) {
1505 		dev_kfree_skb_any(skb);
1506 		return true;
1507 	}
1508 
1509 	/*
1510 	 * it is valid to use page_address instead of kmap since we are
1511 	 * working with pages allocated out of the lomem pool per
1512 	 * alloc_page(GFP_ATOMIC)
1513 	 */
1514 	va = skb_frag_address(frag);
1515 
1516 	/*
1517 	 * we need the header to contain the greater of either ETH_HLEN or
1518 	 * 60 bytes if the skb->len is less than 60 for skb_pad.
1519 	 */
1520 	pull_len = skb_frag_size(frag);
1521 	if (pull_len > IXGBE_RX_HDR_SIZE)
1522 		pull_len = ixgbe_get_headlen(va, IXGBE_RX_HDR_SIZE);
1523 
1524 	/* align pull length to size of long to optimize memcpy performance */
1525 	skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1526 
1527 	/* update all of the pointers */
1528 	skb_frag_size_sub(frag, pull_len);
1529 	frag->page_offset += pull_len;
1530 	skb->data_len -= pull_len;
1531 	skb->tail += pull_len;
1532 
1533 	/*
1534 	 * if we sucked the frag empty then we should free it,
1535 	 * if there are other frags here something is screwed up in hardware
1536 	 */
1537 	if (skb_frag_size(frag) == 0) {
1538 		BUG_ON(skb_shinfo(skb)->nr_frags != 1);
1539 		skb_shinfo(skb)->nr_frags = 0;
1540 		__skb_frag_unref(frag);
1541 		skb->truesize -= ixgbe_rx_bufsz(rx_ring);
1542 	}
1543 
1544 #ifdef IXGBE_FCOE
1545 	/* do not attempt to pad FCoE Frames as this will disrupt DDP */
1546 	if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1547 		return false;
1548 
1549 #endif
1550 	/* if skb_pad returns an error the skb was freed */
1551 	if (unlikely(skb->len < 60)) {
1552 		int pad_len = 60 - skb->len;
1553 
1554 		if (skb_pad(skb, pad_len))
1555 			return true;
1556 		__skb_put(skb, pad_len);
1557 	}
1558 
1559 	return false;
1560 }
1561 
1562 /**
1563  * ixgbe_can_reuse_page - determine if we can reuse a page
1564  * @rx_buffer: pointer to rx_buffer containing the page we want to reuse
1565  *
1566  * Returns true if page can be reused in another Rx buffer
1567  **/
1568 static inline bool ixgbe_can_reuse_page(struct ixgbe_rx_buffer *rx_buffer)
1569 {
1570 	struct page *page = rx_buffer->page;
1571 
1572 	/* if we are only owner of page and it is local we can reuse it */
1573 	return likely(page_count(page) == 1) &&
1574 	       likely(page_to_nid(page) == numa_node_id());
1575 }
1576 
1577 /**
1578  * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1579  * @rx_ring: rx descriptor ring to store buffers on
1580  * @old_buff: donor buffer to have page reused
1581  *
1582  * Syncronizes page for reuse by the adapter
1583  **/
1584 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1585 				struct ixgbe_rx_buffer *old_buff)
1586 {
1587 	struct ixgbe_rx_buffer *new_buff;
1588 	u16 nta = rx_ring->next_to_alloc;
1589 	u16 bufsz = ixgbe_rx_bufsz(rx_ring);
1590 
1591 	new_buff = &rx_ring->rx_buffer_info[nta];
1592 
1593 	/* update, and store next to alloc */
1594 	nta++;
1595 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1596 
1597 	/* transfer page from old buffer to new buffer */
1598 	new_buff->page = old_buff->page;
1599 	new_buff->dma = old_buff->dma;
1600 
1601 	/* flip page offset to other buffer and store to new_buff */
1602 	new_buff->page_offset = old_buff->page_offset ^ bufsz;
1603 
1604 	/* sync the buffer for use by the device */
1605 	dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1606 					 new_buff->page_offset, bufsz,
1607 					 DMA_FROM_DEVICE);
1608 
1609 	/* bump ref count on page before it is given to the stack */
1610 	get_page(new_buff->page);
1611 }
1612 
1613 /**
1614  * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1615  * @rx_ring: rx descriptor ring to transact packets on
1616  * @rx_buffer: buffer containing page to add
1617  * @rx_desc: descriptor containing length of buffer written by hardware
1618  * @skb: sk_buff to place the data into
1619  *
1620  * This function is based on skb_add_rx_frag.  I would have used that
1621  * function however it doesn't handle the truesize case correctly since we
1622  * are allocating more memory than might be used for a single receive.
1623  **/
1624 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1625 			      struct ixgbe_rx_buffer *rx_buffer,
1626 			      struct sk_buff *skb, int size)
1627 {
1628 	skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1629 			   rx_buffer->page, rx_buffer->page_offset,
1630 			   size);
1631 	skb->len += size;
1632 	skb->data_len += size;
1633 	skb->truesize += ixgbe_rx_bufsz(rx_ring);
1634 }
1635 
1636 /**
1637  * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1638  * @q_vector: structure containing interrupt and ring information
1639  * @rx_ring: rx descriptor ring to transact packets on
1640  * @budget: Total limit on number of packets to process
1641  *
1642  * This function provides a "bounce buffer" approach to Rx interrupt
1643  * processing.  The advantage to this is that on systems that have
1644  * expensive overhead for IOMMU access this provides a means of avoiding
1645  * it by maintaining the mapping of the page to the syste.
1646  *
1647  * Returns true if all work is completed without reaching budget
1648  **/
1649 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1650 			       struct ixgbe_ring *rx_ring,
1651 			       int budget)
1652 {
1653 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1654 #ifdef IXGBE_FCOE
1655 	struct ixgbe_adapter *adapter = q_vector->adapter;
1656 	int ddp_bytes = 0;
1657 #endif /* IXGBE_FCOE */
1658 	u16 cleaned_count = ixgbe_desc_unused(rx_ring);
1659 
1660 	do {
1661 		struct ixgbe_rx_buffer *rx_buffer;
1662 		union ixgbe_adv_rx_desc *rx_desc;
1663 		struct sk_buff *skb;
1664 		struct page *page;
1665 		u16 ntc;
1666 
1667 		/* return some buffers to hardware, one at a time is too slow */
1668 		if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1669 			ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1670 			cleaned_count = 0;
1671 		}
1672 
1673 		ntc = rx_ring->next_to_clean;
1674 		rx_desc = IXGBE_RX_DESC(rx_ring, ntc);
1675 		rx_buffer = &rx_ring->rx_buffer_info[ntc];
1676 
1677 		if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
1678 			break;
1679 
1680 		/*
1681 		 * This memory barrier is needed to keep us from reading
1682 		 * any other fields out of the rx_desc until we know the
1683 		 * RXD_STAT_DD bit is set
1684 		 */
1685 		rmb();
1686 
1687 		page = rx_buffer->page;
1688 		prefetchw(page);
1689 
1690 		skb = rx_buffer->skb;
1691 
1692 		if (likely(!skb)) {
1693 			void *page_addr = page_address(page) +
1694 					  rx_buffer->page_offset;
1695 
1696 			/* prefetch first cache line of first page */
1697 			prefetch(page_addr);
1698 #if L1_CACHE_BYTES < 128
1699 			prefetch(page_addr + L1_CACHE_BYTES);
1700 #endif
1701 
1702 			/* allocate a skb to store the frags */
1703 			skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1704 							IXGBE_RX_HDR_SIZE);
1705 			if (unlikely(!skb)) {
1706 				rx_ring->rx_stats.alloc_rx_buff_failed++;
1707 				break;
1708 			}
1709 
1710 			/*
1711 			 * we will be copying header into skb->data in
1712 			 * pskb_may_pull so it is in our interest to prefetch
1713 			 * it now to avoid a possible cache miss
1714 			 */
1715 			prefetchw(skb->data);
1716 
1717 			/*
1718 			 * Delay unmapping of the first packet. It carries the
1719 			 * header information, HW may still access the header
1720 			 * after the writeback.  Only unmap it when EOP is
1721 			 * reached
1722 			 */
1723 			IXGBE_CB(skb)->dma = rx_buffer->dma;
1724 		} else {
1725 			/* we are reusing so sync this buffer for CPU use */
1726 			dma_sync_single_range_for_cpu(rx_ring->dev,
1727 						      rx_buffer->dma,
1728 						      rx_buffer->page_offset,
1729 						      ixgbe_rx_bufsz(rx_ring),
1730 						      DMA_FROM_DEVICE);
1731 		}
1732 
1733 		/* pull page into skb */
1734 		ixgbe_add_rx_frag(rx_ring, rx_buffer, skb,
1735 				  le16_to_cpu(rx_desc->wb.upper.length));
1736 
1737 		if (ixgbe_can_reuse_page(rx_buffer)) {
1738 			/* hand second half of page back to the ring */
1739 			ixgbe_reuse_rx_page(rx_ring, rx_buffer);
1740 		} else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
1741 			/* the page has been released from the ring */
1742 			IXGBE_CB(skb)->page_released = true;
1743 		} else {
1744 			/* we are not reusing the buffer so unmap it */
1745 			dma_unmap_page(rx_ring->dev, rx_buffer->dma,
1746 				       ixgbe_rx_pg_size(rx_ring),
1747 				       DMA_FROM_DEVICE);
1748 		}
1749 
1750 		/* clear contents of buffer_info */
1751 		rx_buffer->skb = NULL;
1752 		rx_buffer->dma = 0;
1753 		rx_buffer->page = NULL;
1754 
1755 		ixgbe_get_rsc_cnt(rx_ring, rx_desc, skb);
1756 
1757 		cleaned_count++;
1758 
1759 		/* place incomplete frames back on ring for completion */
1760 		if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
1761 			continue;
1762 
1763 		/* verify the packet layout is correct */
1764 		if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
1765 			continue;
1766 
1767 		/* probably a little skewed due to removing CRC */
1768 		total_rx_bytes += skb->len;
1769 		total_rx_packets++;
1770 
1771 		/* populate checksum, timestamp, VLAN, and protocol */
1772 		ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
1773 
1774 #ifdef IXGBE_FCOE
1775 		/* if ddp, not passing to ULD unless for FCP_RSP or error */
1776 		if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
1777 			ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1778 			if (!ddp_bytes) {
1779 				dev_kfree_skb_any(skb);
1780 				continue;
1781 			}
1782 		}
1783 
1784 #endif /* IXGBE_FCOE */
1785 		ixgbe_rx_skb(q_vector, skb);
1786 
1787 		/* update budget accounting */
1788 		budget--;
1789 	} while (likely(budget));
1790 
1791 #ifdef IXGBE_FCOE
1792 	/* include DDPed FCoE data */
1793 	if (ddp_bytes > 0) {
1794 		unsigned int mss;
1795 
1796 		mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1797 			sizeof(struct fc_frame_header) -
1798 			sizeof(struct fcoe_crc_eof);
1799 		if (mss > 512)
1800 			mss &= ~511;
1801 		total_rx_bytes += ddp_bytes;
1802 		total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1803 	}
1804 
1805 #endif /* IXGBE_FCOE */
1806 	u64_stats_update_begin(&rx_ring->syncp);
1807 	rx_ring->stats.packets += total_rx_packets;
1808 	rx_ring->stats.bytes += total_rx_bytes;
1809 	u64_stats_update_end(&rx_ring->syncp);
1810 	q_vector->rx.total_packets += total_rx_packets;
1811 	q_vector->rx.total_bytes += total_rx_bytes;
1812 
1813 	if (cleaned_count)
1814 		ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1815 
1816 	return !!budget;
1817 }
1818 
1819 /**
1820  * ixgbe_configure_msix - Configure MSI-X hardware
1821  * @adapter: board private structure
1822  *
1823  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1824  * interrupts.
1825  **/
1826 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1827 {
1828 	struct ixgbe_q_vector *q_vector;
1829 	int v_idx;
1830 	u32 mask;
1831 
1832 	/* Populate MSIX to EITR Select */
1833 	if (adapter->num_vfs > 32) {
1834 		u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
1835 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
1836 	}
1837 
1838 	/*
1839 	 * Populate the IVAR table and set the ITR values to the
1840 	 * corresponding register.
1841 	 */
1842 	for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
1843 		struct ixgbe_ring *ring;
1844 		q_vector = adapter->q_vector[v_idx];
1845 
1846 		ixgbe_for_each_ring(ring, q_vector->rx)
1847 			ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1848 
1849 		ixgbe_for_each_ring(ring, q_vector->tx)
1850 			ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1851 
1852 		if (q_vector->tx.ring && !q_vector->rx.ring) {
1853 			/* tx only vector */
1854 			if (adapter->tx_itr_setting == 1)
1855 				q_vector->itr = IXGBE_10K_ITR;
1856 			else
1857 				q_vector->itr = adapter->tx_itr_setting;
1858 		} else {
1859 			/* rx or rx/tx vector */
1860 			if (adapter->rx_itr_setting == 1)
1861 				q_vector->itr = IXGBE_20K_ITR;
1862 			else
1863 				q_vector->itr = adapter->rx_itr_setting;
1864 		}
1865 
1866 		ixgbe_write_eitr(q_vector);
1867 	}
1868 
1869 	switch (adapter->hw.mac.type) {
1870 	case ixgbe_mac_82598EB:
1871 		ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1872 			       v_idx);
1873 		break;
1874 	case ixgbe_mac_82599EB:
1875 	case ixgbe_mac_X540:
1876 		ixgbe_set_ivar(adapter, -1, 1, v_idx);
1877 		break;
1878 	default:
1879 		break;
1880 	}
1881 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1882 
1883 	/* set up to autoclear timer, and the vectors */
1884 	mask = IXGBE_EIMS_ENABLE_MASK;
1885 	mask &= ~(IXGBE_EIMS_OTHER |
1886 		  IXGBE_EIMS_MAILBOX |
1887 		  IXGBE_EIMS_LSC);
1888 
1889 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1890 }
1891 
1892 enum latency_range {
1893 	lowest_latency = 0,
1894 	low_latency = 1,
1895 	bulk_latency = 2,
1896 	latency_invalid = 255
1897 };
1898 
1899 /**
1900  * ixgbe_update_itr - update the dynamic ITR value based on statistics
1901  * @q_vector: structure containing interrupt and ring information
1902  * @ring_container: structure containing ring performance data
1903  *
1904  *      Stores a new ITR value based on packets and byte
1905  *      counts during the last interrupt.  The advantage of per interrupt
1906  *      computation is faster updates and more accurate ITR for the current
1907  *      traffic pattern.  Constants in this function were computed
1908  *      based on theoretical maximum wire speed and thresholds were set based
1909  *      on testing data as well as attempting to minimize response time
1910  *      while increasing bulk throughput.
1911  *      this functionality is controlled by the InterruptThrottleRate module
1912  *      parameter (see ixgbe_param.c)
1913  **/
1914 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
1915 			     struct ixgbe_ring_container *ring_container)
1916 {
1917 	int bytes = ring_container->total_bytes;
1918 	int packets = ring_container->total_packets;
1919 	u32 timepassed_us;
1920 	u64 bytes_perint;
1921 	u8 itr_setting = ring_container->itr;
1922 
1923 	if (packets == 0)
1924 		return;
1925 
1926 	/* simple throttlerate management
1927 	 *   0-10MB/s   lowest (100000 ints/s)
1928 	 *  10-20MB/s   low    (20000 ints/s)
1929 	 *  20-1249MB/s bulk   (8000 ints/s)
1930 	 */
1931 	/* what was last interrupt timeslice? */
1932 	timepassed_us = q_vector->itr >> 2;
1933 	bytes_perint = bytes / timepassed_us; /* bytes/usec */
1934 
1935 	switch (itr_setting) {
1936 	case lowest_latency:
1937 		if (bytes_perint > 10)
1938 			itr_setting = low_latency;
1939 		break;
1940 	case low_latency:
1941 		if (bytes_perint > 20)
1942 			itr_setting = bulk_latency;
1943 		else if (bytes_perint <= 10)
1944 			itr_setting = lowest_latency;
1945 		break;
1946 	case bulk_latency:
1947 		if (bytes_perint <= 20)
1948 			itr_setting = low_latency;
1949 		break;
1950 	}
1951 
1952 	/* clear work counters since we have the values we need */
1953 	ring_container->total_bytes = 0;
1954 	ring_container->total_packets = 0;
1955 
1956 	/* write updated itr to ring container */
1957 	ring_container->itr = itr_setting;
1958 }
1959 
1960 /**
1961  * ixgbe_write_eitr - write EITR register in hardware specific way
1962  * @q_vector: structure containing interrupt and ring information
1963  *
1964  * This function is made to be called by ethtool and by the driver
1965  * when it needs to update EITR registers at runtime.  Hardware
1966  * specific quirks/differences are taken care of here.
1967  */
1968 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1969 {
1970 	struct ixgbe_adapter *adapter = q_vector->adapter;
1971 	struct ixgbe_hw *hw = &adapter->hw;
1972 	int v_idx = q_vector->v_idx;
1973 	u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1974 
1975 	switch (adapter->hw.mac.type) {
1976 	case ixgbe_mac_82598EB:
1977 		/* must write high and low 16 bits to reset counter */
1978 		itr_reg |= (itr_reg << 16);
1979 		break;
1980 	case ixgbe_mac_82599EB:
1981 	case ixgbe_mac_X540:
1982 		/*
1983 		 * set the WDIS bit to not clear the timer bits and cause an
1984 		 * immediate assertion of the interrupt
1985 		 */
1986 		itr_reg |= IXGBE_EITR_CNT_WDIS;
1987 		break;
1988 	default:
1989 		break;
1990 	}
1991 	IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1992 }
1993 
1994 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
1995 {
1996 	u32 new_itr = q_vector->itr;
1997 	u8 current_itr;
1998 
1999 	ixgbe_update_itr(q_vector, &q_vector->tx);
2000 	ixgbe_update_itr(q_vector, &q_vector->rx);
2001 
2002 	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
2003 
2004 	switch (current_itr) {
2005 	/* counts and packets in update_itr are dependent on these numbers */
2006 	case lowest_latency:
2007 		new_itr = IXGBE_100K_ITR;
2008 		break;
2009 	case low_latency:
2010 		new_itr = IXGBE_20K_ITR;
2011 		break;
2012 	case bulk_latency:
2013 		new_itr = IXGBE_8K_ITR;
2014 		break;
2015 	default:
2016 		break;
2017 	}
2018 
2019 	if (new_itr != q_vector->itr) {
2020 		/* do an exponential smoothing */
2021 		new_itr = (10 * new_itr * q_vector->itr) /
2022 			  ((9 * new_itr) + q_vector->itr);
2023 
2024 		/* save the algorithm value here */
2025 		q_vector->itr = new_itr;
2026 
2027 		ixgbe_write_eitr(q_vector);
2028 	}
2029 }
2030 
2031 /**
2032  * ixgbe_check_overtemp_subtask - check for over temperature
2033  * @adapter: pointer to adapter
2034  **/
2035 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2036 {
2037 	struct ixgbe_hw *hw = &adapter->hw;
2038 	u32 eicr = adapter->interrupt_event;
2039 
2040 	if (test_bit(__IXGBE_DOWN, &adapter->state))
2041 		return;
2042 
2043 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2044 	    !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2045 		return;
2046 
2047 	adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2048 
2049 	switch (hw->device_id) {
2050 	case IXGBE_DEV_ID_82599_T3_LOM:
2051 		/*
2052 		 * Since the warning interrupt is for both ports
2053 		 * we don't have to check if:
2054 		 *  - This interrupt wasn't for our port.
2055 		 *  - We may have missed the interrupt so always have to
2056 		 *    check if we  got a LSC
2057 		 */
2058 		if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
2059 		    !(eicr & IXGBE_EICR_LSC))
2060 			return;
2061 
2062 		if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2063 			u32 autoneg;
2064 			bool link_up = false;
2065 
2066 			hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2067 
2068 			if (link_up)
2069 				return;
2070 		}
2071 
2072 		/* Check if this is not due to overtemp */
2073 		if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2074 			return;
2075 
2076 		break;
2077 	default:
2078 		if (!(eicr & IXGBE_EICR_GPI_SDP0))
2079 			return;
2080 		break;
2081 	}
2082 	e_crit(drv,
2083 	       "Network adapter has been stopped because it has over heated. "
2084 	       "Restart the computer. If the problem persists, "
2085 	       "power off the system and replace the adapter\n");
2086 
2087 	adapter->interrupt_event = 0;
2088 }
2089 
2090 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2091 {
2092 	struct ixgbe_hw *hw = &adapter->hw;
2093 
2094 	if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2095 	    (eicr & IXGBE_EICR_GPI_SDP1)) {
2096 		e_crit(probe, "Fan has stopped, replace the adapter\n");
2097 		/* write to clear the interrupt */
2098 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2099 	}
2100 }
2101 
2102 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2103 {
2104 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2105 		return;
2106 
2107 	switch (adapter->hw.mac.type) {
2108 	case ixgbe_mac_82599EB:
2109 		/*
2110 		 * Need to check link state so complete overtemp check
2111 		 * on service task
2112 		 */
2113 		if (((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)) &&
2114 		    (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2115 			adapter->interrupt_event = eicr;
2116 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2117 			ixgbe_service_event_schedule(adapter);
2118 			return;
2119 		}
2120 		return;
2121 	case ixgbe_mac_X540:
2122 		if (!(eicr & IXGBE_EICR_TS))
2123 			return;
2124 		break;
2125 	default:
2126 		return;
2127 	}
2128 
2129 	e_crit(drv,
2130 	       "Network adapter has been stopped because it has over heated. "
2131 	       "Restart the computer. If the problem persists, "
2132 	       "power off the system and replace the adapter\n");
2133 }
2134 
2135 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2136 {
2137 	struct ixgbe_hw *hw = &adapter->hw;
2138 
2139 	if (eicr & IXGBE_EICR_GPI_SDP2) {
2140 		/* Clear the interrupt */
2141 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
2142 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2143 			adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2144 			ixgbe_service_event_schedule(adapter);
2145 		}
2146 	}
2147 
2148 	if (eicr & IXGBE_EICR_GPI_SDP1) {
2149 		/* Clear the interrupt */
2150 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2151 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2152 			adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2153 			ixgbe_service_event_schedule(adapter);
2154 		}
2155 	}
2156 }
2157 
2158 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2159 {
2160 	struct ixgbe_hw *hw = &adapter->hw;
2161 
2162 	adapter->lsc_int++;
2163 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2164 	adapter->link_check_timeout = jiffies;
2165 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2166 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2167 		IXGBE_WRITE_FLUSH(hw);
2168 		ixgbe_service_event_schedule(adapter);
2169 	}
2170 }
2171 
2172 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2173 					   u64 qmask)
2174 {
2175 	u32 mask;
2176 	struct ixgbe_hw *hw = &adapter->hw;
2177 
2178 	switch (hw->mac.type) {
2179 	case ixgbe_mac_82598EB:
2180 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2181 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2182 		break;
2183 	case ixgbe_mac_82599EB:
2184 	case ixgbe_mac_X540:
2185 		mask = (qmask & 0xFFFFFFFF);
2186 		if (mask)
2187 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2188 		mask = (qmask >> 32);
2189 		if (mask)
2190 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2191 		break;
2192 	default:
2193 		break;
2194 	}
2195 	/* skip the flush */
2196 }
2197 
2198 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2199 					    u64 qmask)
2200 {
2201 	u32 mask;
2202 	struct ixgbe_hw *hw = &adapter->hw;
2203 
2204 	switch (hw->mac.type) {
2205 	case ixgbe_mac_82598EB:
2206 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2207 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2208 		break;
2209 	case ixgbe_mac_82599EB:
2210 	case ixgbe_mac_X540:
2211 		mask = (qmask & 0xFFFFFFFF);
2212 		if (mask)
2213 			IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2214 		mask = (qmask >> 32);
2215 		if (mask)
2216 			IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2217 		break;
2218 	default:
2219 		break;
2220 	}
2221 	/* skip the flush */
2222 }
2223 
2224 /**
2225  * ixgbe_irq_enable - Enable default interrupt generation settings
2226  * @adapter: board private structure
2227  **/
2228 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2229 				    bool flush)
2230 {
2231 	u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2232 
2233 	/* don't reenable LSC while waiting for link */
2234 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2235 		mask &= ~IXGBE_EIMS_LSC;
2236 
2237 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2238 		switch (adapter->hw.mac.type) {
2239 		case ixgbe_mac_82599EB:
2240 			mask |= IXGBE_EIMS_GPI_SDP0;
2241 			break;
2242 		case ixgbe_mac_X540:
2243 			mask |= IXGBE_EIMS_TS;
2244 			break;
2245 		default:
2246 			break;
2247 		}
2248 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2249 		mask |= IXGBE_EIMS_GPI_SDP1;
2250 	switch (adapter->hw.mac.type) {
2251 	case ixgbe_mac_82599EB:
2252 		mask |= IXGBE_EIMS_GPI_SDP1;
2253 		mask |= IXGBE_EIMS_GPI_SDP2;
2254 	case ixgbe_mac_X540:
2255 		mask |= IXGBE_EIMS_ECC;
2256 		mask |= IXGBE_EIMS_MAILBOX;
2257 		break;
2258 	default:
2259 		break;
2260 	}
2261 	if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2262 	    !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2263 		mask |= IXGBE_EIMS_FLOW_DIR;
2264 
2265 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2266 	if (queues)
2267 		ixgbe_irq_enable_queues(adapter, ~0);
2268 	if (flush)
2269 		IXGBE_WRITE_FLUSH(&adapter->hw);
2270 }
2271 
2272 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2273 {
2274 	struct ixgbe_adapter *adapter = data;
2275 	struct ixgbe_hw *hw = &adapter->hw;
2276 	u32 eicr;
2277 
2278 	/*
2279 	 * Workaround for Silicon errata.  Use clear-by-write instead
2280 	 * of clear-by-read.  Reading with EICS will return the
2281 	 * interrupt causes without clearing, which later be done
2282 	 * with the write to EICR.
2283 	 */
2284 	eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2285 	IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2286 
2287 	if (eicr & IXGBE_EICR_LSC)
2288 		ixgbe_check_lsc(adapter);
2289 
2290 	if (eicr & IXGBE_EICR_MAILBOX)
2291 		ixgbe_msg_task(adapter);
2292 
2293 	switch (hw->mac.type) {
2294 	case ixgbe_mac_82599EB:
2295 	case ixgbe_mac_X540:
2296 		if (eicr & IXGBE_EICR_ECC)
2297 			e_info(link, "Received unrecoverable ECC Err, please "
2298 			       "reboot\n");
2299 		/* Handle Flow Director Full threshold interrupt */
2300 		if (eicr & IXGBE_EICR_FLOW_DIR) {
2301 			int reinit_count = 0;
2302 			int i;
2303 			for (i = 0; i < adapter->num_tx_queues; i++) {
2304 				struct ixgbe_ring *ring = adapter->tx_ring[i];
2305 				if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2306 						       &ring->state))
2307 					reinit_count++;
2308 			}
2309 			if (reinit_count) {
2310 				/* no more flow director interrupts until after init */
2311 				IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2312 				adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2313 				ixgbe_service_event_schedule(adapter);
2314 			}
2315 		}
2316 		ixgbe_check_sfp_event(adapter, eicr);
2317 		ixgbe_check_overtemp_event(adapter, eicr);
2318 		break;
2319 	default:
2320 		break;
2321 	}
2322 
2323 	ixgbe_check_fan_failure(adapter, eicr);
2324 #ifdef CONFIG_IXGBE_PTP
2325 	ixgbe_ptp_check_pps_event(adapter, eicr);
2326 #endif
2327 
2328 	/* re-enable the original interrupt state, no lsc, no queues */
2329 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
2330 		ixgbe_irq_enable(adapter, false, false);
2331 
2332 	return IRQ_HANDLED;
2333 }
2334 
2335 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2336 {
2337 	struct ixgbe_q_vector *q_vector = data;
2338 
2339 	/* EIAM disabled interrupts (on this vector) for us */
2340 
2341 	if (q_vector->rx.ring || q_vector->tx.ring)
2342 		napi_schedule(&q_vector->napi);
2343 
2344 	return IRQ_HANDLED;
2345 }
2346 
2347 /**
2348  * ixgbe_poll - NAPI Rx polling callback
2349  * @napi: structure for representing this polling device
2350  * @budget: how many packets driver is allowed to clean
2351  *
2352  * This function is used for legacy and MSI, NAPI mode
2353  **/
2354 int ixgbe_poll(struct napi_struct *napi, int budget)
2355 {
2356 	struct ixgbe_q_vector *q_vector =
2357 				container_of(napi, struct ixgbe_q_vector, napi);
2358 	struct ixgbe_adapter *adapter = q_vector->adapter;
2359 	struct ixgbe_ring *ring;
2360 	int per_ring_budget;
2361 	bool clean_complete = true;
2362 
2363 #ifdef CONFIG_IXGBE_DCA
2364 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2365 		ixgbe_update_dca(q_vector);
2366 #endif
2367 
2368 	ixgbe_for_each_ring(ring, q_vector->tx)
2369 		clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
2370 
2371 	/* attempt to distribute budget to each queue fairly, but don't allow
2372 	 * the budget to go below 1 because we'll exit polling */
2373 	if (q_vector->rx.count > 1)
2374 		per_ring_budget = max(budget/q_vector->rx.count, 1);
2375 	else
2376 		per_ring_budget = budget;
2377 
2378 	ixgbe_for_each_ring(ring, q_vector->rx)
2379 		clean_complete &= ixgbe_clean_rx_irq(q_vector, ring,
2380 						     per_ring_budget);
2381 
2382 	/* If all work not completed, return budget and keep polling */
2383 	if (!clean_complete)
2384 		return budget;
2385 
2386 	/* all work done, exit the polling mode */
2387 	napi_complete(napi);
2388 	if (adapter->rx_itr_setting & 1)
2389 		ixgbe_set_itr(q_vector);
2390 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
2391 		ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2392 
2393 	return 0;
2394 }
2395 
2396 /**
2397  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2398  * @adapter: board private structure
2399  *
2400  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2401  * interrupts from the kernel.
2402  **/
2403 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2404 {
2405 	struct net_device *netdev = adapter->netdev;
2406 	int vector, err;
2407 	int ri = 0, ti = 0;
2408 
2409 	for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2410 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2411 		struct msix_entry *entry = &adapter->msix_entries[vector];
2412 
2413 		if (q_vector->tx.ring && q_vector->rx.ring) {
2414 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2415 				 "%s-%s-%d", netdev->name, "TxRx", ri++);
2416 			ti++;
2417 		} else if (q_vector->rx.ring) {
2418 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2419 				 "%s-%s-%d", netdev->name, "rx", ri++);
2420 		} else if (q_vector->tx.ring) {
2421 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2422 				 "%s-%s-%d", netdev->name, "tx", ti++);
2423 		} else {
2424 			/* skip this unused q_vector */
2425 			continue;
2426 		}
2427 		err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2428 				  q_vector->name, q_vector);
2429 		if (err) {
2430 			e_err(probe, "request_irq failed for MSIX interrupt "
2431 			      "Error: %d\n", err);
2432 			goto free_queue_irqs;
2433 		}
2434 		/* If Flow Director is enabled, set interrupt affinity */
2435 		if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2436 			/* assign the mask for this irq */
2437 			irq_set_affinity_hint(entry->vector,
2438 					      &q_vector->affinity_mask);
2439 		}
2440 	}
2441 
2442 	err = request_irq(adapter->msix_entries[vector].vector,
2443 			  ixgbe_msix_other, 0, netdev->name, adapter);
2444 	if (err) {
2445 		e_err(probe, "request_irq for msix_other failed: %d\n", err);
2446 		goto free_queue_irqs;
2447 	}
2448 
2449 	return 0;
2450 
2451 free_queue_irqs:
2452 	while (vector) {
2453 		vector--;
2454 		irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2455 				      NULL);
2456 		free_irq(adapter->msix_entries[vector].vector,
2457 			 adapter->q_vector[vector]);
2458 	}
2459 	adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2460 	pci_disable_msix(adapter->pdev);
2461 	kfree(adapter->msix_entries);
2462 	adapter->msix_entries = NULL;
2463 	return err;
2464 }
2465 
2466 /**
2467  * ixgbe_intr - legacy mode Interrupt Handler
2468  * @irq: interrupt number
2469  * @data: pointer to a network interface device structure
2470  **/
2471 static irqreturn_t ixgbe_intr(int irq, void *data)
2472 {
2473 	struct ixgbe_adapter *adapter = data;
2474 	struct ixgbe_hw *hw = &adapter->hw;
2475 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2476 	u32 eicr;
2477 
2478 	/*
2479 	 * Workaround for silicon errata #26 on 82598.  Mask the interrupt
2480 	 * before the read of EICR.
2481 	 */
2482 	IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2483 
2484 	/* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2485 	 * therefore no explicit interrupt disable is necessary */
2486 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2487 	if (!eicr) {
2488 		/*
2489 		 * shared interrupt alert!
2490 		 * make sure interrupts are enabled because the read will
2491 		 * have disabled interrupts due to EIAM
2492 		 * finish the workaround of silicon errata on 82598.  Unmask
2493 		 * the interrupt that we masked before the EICR read.
2494 		 */
2495 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
2496 			ixgbe_irq_enable(adapter, true, true);
2497 		return IRQ_NONE;	/* Not our interrupt */
2498 	}
2499 
2500 	if (eicr & IXGBE_EICR_LSC)
2501 		ixgbe_check_lsc(adapter);
2502 
2503 	switch (hw->mac.type) {
2504 	case ixgbe_mac_82599EB:
2505 		ixgbe_check_sfp_event(adapter, eicr);
2506 		/* Fall through */
2507 	case ixgbe_mac_X540:
2508 		if (eicr & IXGBE_EICR_ECC)
2509 			e_info(link, "Received unrecoverable ECC err, please "
2510 				     "reboot\n");
2511 		ixgbe_check_overtemp_event(adapter, eicr);
2512 		break;
2513 	default:
2514 		break;
2515 	}
2516 
2517 	ixgbe_check_fan_failure(adapter, eicr);
2518 #ifdef CONFIG_IXGBE_PTP
2519 	ixgbe_ptp_check_pps_event(adapter, eicr);
2520 #endif
2521 
2522 	/* would disable interrupts here but EIAM disabled it */
2523 	napi_schedule(&q_vector->napi);
2524 
2525 	/*
2526 	 * re-enable link(maybe) and non-queue interrupts, no flush.
2527 	 * ixgbe_poll will re-enable the queue interrupts
2528 	 */
2529 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
2530 		ixgbe_irq_enable(adapter, false, false);
2531 
2532 	return IRQ_HANDLED;
2533 }
2534 
2535 /**
2536  * ixgbe_request_irq - initialize interrupts
2537  * @adapter: board private structure
2538  *
2539  * Attempts to configure interrupts using the best available
2540  * capabilities of the hardware and kernel.
2541  **/
2542 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2543 {
2544 	struct net_device *netdev = adapter->netdev;
2545 	int err;
2546 
2547 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2548 		err = ixgbe_request_msix_irqs(adapter);
2549 	else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2550 		err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2551 				  netdev->name, adapter);
2552 	else
2553 		err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2554 				  netdev->name, adapter);
2555 
2556 	if (err)
2557 		e_err(probe, "request_irq failed, Error %d\n", err);
2558 
2559 	return err;
2560 }
2561 
2562 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2563 {
2564 	int vector;
2565 
2566 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2567 		free_irq(adapter->pdev->irq, adapter);
2568 		return;
2569 	}
2570 
2571 	for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2572 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2573 		struct msix_entry *entry = &adapter->msix_entries[vector];
2574 
2575 		/* free only the irqs that were actually requested */
2576 		if (!q_vector->rx.ring && !q_vector->tx.ring)
2577 			continue;
2578 
2579 		/* clear the affinity_mask in the IRQ descriptor */
2580 		irq_set_affinity_hint(entry->vector, NULL);
2581 
2582 		free_irq(entry->vector, q_vector);
2583 	}
2584 
2585 	free_irq(adapter->msix_entries[vector++].vector, adapter);
2586 }
2587 
2588 /**
2589  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2590  * @adapter: board private structure
2591  **/
2592 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2593 {
2594 	switch (adapter->hw.mac.type) {
2595 	case ixgbe_mac_82598EB:
2596 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2597 		break;
2598 	case ixgbe_mac_82599EB:
2599 	case ixgbe_mac_X540:
2600 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2601 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2602 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2603 		break;
2604 	default:
2605 		break;
2606 	}
2607 	IXGBE_WRITE_FLUSH(&adapter->hw);
2608 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2609 		int vector;
2610 
2611 		for (vector = 0; vector < adapter->num_q_vectors; vector++)
2612 			synchronize_irq(adapter->msix_entries[vector].vector);
2613 
2614 		synchronize_irq(adapter->msix_entries[vector++].vector);
2615 	} else {
2616 		synchronize_irq(adapter->pdev->irq);
2617 	}
2618 }
2619 
2620 /**
2621  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2622  *
2623  **/
2624 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2625 {
2626 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2627 
2628 	/* rx/tx vector */
2629 	if (adapter->rx_itr_setting == 1)
2630 		q_vector->itr = IXGBE_20K_ITR;
2631 	else
2632 		q_vector->itr = adapter->rx_itr_setting;
2633 
2634 	ixgbe_write_eitr(q_vector);
2635 
2636 	ixgbe_set_ivar(adapter, 0, 0, 0);
2637 	ixgbe_set_ivar(adapter, 1, 0, 0);
2638 
2639 	e_info(hw, "Legacy interrupt IVAR setup done\n");
2640 }
2641 
2642 /**
2643  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2644  * @adapter: board private structure
2645  * @ring: structure containing ring specific data
2646  *
2647  * Configure the Tx descriptor ring after a reset.
2648  **/
2649 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2650 			     struct ixgbe_ring *ring)
2651 {
2652 	struct ixgbe_hw *hw = &adapter->hw;
2653 	u64 tdba = ring->dma;
2654 	int wait_loop = 10;
2655 	u32 txdctl = IXGBE_TXDCTL_ENABLE;
2656 	u8 reg_idx = ring->reg_idx;
2657 
2658 	/* disable queue to avoid issues while updating state */
2659 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
2660 	IXGBE_WRITE_FLUSH(hw);
2661 
2662 	IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2663 			(tdba & DMA_BIT_MASK(32)));
2664 	IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2665 	IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2666 			ring->count * sizeof(union ixgbe_adv_tx_desc));
2667 	IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2668 	IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2669 	ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2670 
2671 	/*
2672 	 * set WTHRESH to encourage burst writeback, it should not be set
2673 	 * higher than 1 when ITR is 0 as it could cause false TX hangs
2674 	 *
2675 	 * In order to avoid issues WTHRESH + PTHRESH should always be equal
2676 	 * to or less than the number of on chip descriptors, which is
2677 	 * currently 40.
2678 	 */
2679 	if (!ring->q_vector || (ring->q_vector->itr < 8))
2680 		txdctl |= (1 << 16);	/* WTHRESH = 1 */
2681 	else
2682 		txdctl |= (8 << 16);	/* WTHRESH = 8 */
2683 
2684 	/*
2685 	 * Setting PTHRESH to 32 both improves performance
2686 	 * and avoids a TX hang with DFP enabled
2687 	 */
2688 	txdctl |= (1 << 8) |	/* HTHRESH = 1 */
2689 		   32;		/* PTHRESH = 32 */
2690 
2691 	/* reinitialize flowdirector state */
2692 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2693 		ring->atr_sample_rate = adapter->atr_sample_rate;
2694 		ring->atr_count = 0;
2695 		set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2696 	} else {
2697 		ring->atr_sample_rate = 0;
2698 	}
2699 
2700 	clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2701 
2702 	/* enable queue */
2703 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2704 
2705 	/* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2706 	if (hw->mac.type == ixgbe_mac_82598EB &&
2707 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2708 		return;
2709 
2710 	/* poll to verify queue is enabled */
2711 	do {
2712 		usleep_range(1000, 2000);
2713 		txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2714 	} while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2715 	if (!wait_loop)
2716 		e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2717 }
2718 
2719 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2720 {
2721 	struct ixgbe_hw *hw = &adapter->hw;
2722 	u32 rttdcs, mtqc;
2723 	u8 tcs = netdev_get_num_tc(adapter->netdev);
2724 
2725 	if (hw->mac.type == ixgbe_mac_82598EB)
2726 		return;
2727 
2728 	/* disable the arbiter while setting MTQC */
2729 	rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2730 	rttdcs |= IXGBE_RTTDCS_ARBDIS;
2731 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2732 
2733 	/* set transmit pool layout */
2734 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2735 		mtqc = IXGBE_MTQC_VT_ENA;
2736 		if (tcs > 4)
2737 			mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2738 		else if (tcs > 1)
2739 			mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2740 		else if (adapter->ring_feature[RING_F_RSS].indices == 4)
2741 			mtqc |= IXGBE_MTQC_32VF;
2742 		else
2743 			mtqc |= IXGBE_MTQC_64VF;
2744 	} else {
2745 		if (tcs > 4)
2746 			mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2747 		else if (tcs > 1)
2748 			mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2749 		else
2750 			mtqc = IXGBE_MTQC_64Q_1PB;
2751 	}
2752 
2753 	IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
2754 
2755 	/* Enable Security TX Buffer IFG for multiple pb */
2756 	if (tcs) {
2757 		u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
2758 		sectx |= IXGBE_SECTX_DCB;
2759 		IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
2760 	}
2761 
2762 	/* re-enable the arbiter */
2763 	rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2764 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2765 }
2766 
2767 /**
2768  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2769  * @adapter: board private structure
2770  *
2771  * Configure the Tx unit of the MAC after a reset.
2772  **/
2773 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2774 {
2775 	struct ixgbe_hw *hw = &adapter->hw;
2776 	u32 dmatxctl;
2777 	u32 i;
2778 
2779 	ixgbe_setup_mtqc(adapter);
2780 
2781 	if (hw->mac.type != ixgbe_mac_82598EB) {
2782 		/* DMATXCTL.EN must be before Tx queues are enabled */
2783 		dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2784 		dmatxctl |= IXGBE_DMATXCTL_TE;
2785 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2786 	}
2787 
2788 	/* Setup the HW Tx Head and Tail descriptor pointers */
2789 	for (i = 0; i < adapter->num_tx_queues; i++)
2790 		ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2791 }
2792 
2793 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
2794 				 struct ixgbe_ring *ring)
2795 {
2796 	struct ixgbe_hw *hw = &adapter->hw;
2797 	u8 reg_idx = ring->reg_idx;
2798 	u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
2799 
2800 	srrctl |= IXGBE_SRRCTL_DROP_EN;
2801 
2802 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
2803 }
2804 
2805 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
2806 				  struct ixgbe_ring *ring)
2807 {
2808 	struct ixgbe_hw *hw = &adapter->hw;
2809 	u8 reg_idx = ring->reg_idx;
2810 	u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
2811 
2812 	srrctl &= ~IXGBE_SRRCTL_DROP_EN;
2813 
2814 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
2815 }
2816 
2817 #ifdef CONFIG_IXGBE_DCB
2818 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
2819 #else
2820 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
2821 #endif
2822 {
2823 	int i;
2824 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
2825 
2826 	if (adapter->ixgbe_ieee_pfc)
2827 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
2828 
2829 	/*
2830 	 * We should set the drop enable bit if:
2831 	 *  SR-IOV is enabled
2832 	 *   or
2833 	 *  Number of Rx queues > 1 and flow control is disabled
2834 	 *
2835 	 *  This allows us to avoid head of line blocking for security
2836 	 *  and performance reasons.
2837 	 */
2838 	if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
2839 	    !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
2840 		for (i = 0; i < adapter->num_rx_queues; i++)
2841 			ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
2842 	} else {
2843 		for (i = 0; i < adapter->num_rx_queues; i++)
2844 			ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
2845 	}
2846 }
2847 
2848 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2849 
2850 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2851 				   struct ixgbe_ring *rx_ring)
2852 {
2853 	struct ixgbe_hw *hw = &adapter->hw;
2854 	u32 srrctl;
2855 	u8 reg_idx = rx_ring->reg_idx;
2856 
2857 	if (hw->mac.type == ixgbe_mac_82598EB) {
2858 		u16 mask = adapter->ring_feature[RING_F_RSS].mask;
2859 
2860 		/*
2861 		 * if VMDq is not active we must program one srrctl register
2862 		 * per RSS queue since we have enabled RDRXCTL.MVMEN
2863 		 */
2864 		reg_idx &= mask;
2865 	}
2866 
2867 	/* configure header buffer length, needed for RSC */
2868 	srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
2869 
2870 	/* configure the packet buffer length */
2871 #if PAGE_SIZE > IXGBE_MAX_RXBUFFER
2872 	srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2873 #else
2874 	srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2875 #endif
2876 
2877 	/* configure descriptor type */
2878 	srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2879 
2880 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
2881 }
2882 
2883 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2884 {
2885 	struct ixgbe_hw *hw = &adapter->hw;
2886 	static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2887 			  0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2888 			  0x6A3E67EA, 0x14364D17, 0x3BED200D};
2889 	u32 mrqc = 0, reta = 0;
2890 	u32 rxcsum;
2891 	int i, j;
2892 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
2893 
2894 	/*
2895 	 * Program table for at least 2 queues w/ SR-IOV so that VFs can
2896 	 * make full use of any rings they may have.  We will use the
2897 	 * PSRTYPE register to control how many rings we use within the PF.
2898 	 */
2899 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 2))
2900 		rss_i = 2;
2901 
2902 	/* Fill out hash function seeds */
2903 	for (i = 0; i < 10; i++)
2904 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2905 
2906 	/* Fill out redirection table */
2907 	for (i = 0, j = 0; i < 128; i++, j++) {
2908 		if (j == rss_i)
2909 			j = 0;
2910 		/* reta = 4-byte sliding window of
2911 		 * 0x00..(indices-1)(indices-1)00..etc. */
2912 		reta = (reta << 8) | (j * 0x11);
2913 		if ((i & 3) == 3)
2914 			IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2915 	}
2916 
2917 	/* Disable indicating checksum in descriptor, enables RSS hash */
2918 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2919 	rxcsum |= IXGBE_RXCSUM_PCSD;
2920 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2921 
2922 	if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2923 		if (adapter->ring_feature[RING_F_RSS].mask)
2924 			mrqc = IXGBE_MRQC_RSSEN;
2925 	} else {
2926 		u8 tcs = netdev_get_num_tc(adapter->netdev);
2927 
2928 		if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2929 			if (tcs > 4)
2930 				mrqc = IXGBE_MRQC_VMDQRT8TCEN;	/* 8 TCs */
2931 			else if (tcs > 1)
2932 				mrqc = IXGBE_MRQC_VMDQRT4TCEN;	/* 4 TCs */
2933 			else if (adapter->ring_feature[RING_F_RSS].indices == 4)
2934 				mrqc = IXGBE_MRQC_VMDQRSS32EN;
2935 			else
2936 				mrqc = IXGBE_MRQC_VMDQRSS64EN;
2937 		} else {
2938 			if (tcs > 4)
2939 				mrqc = IXGBE_MRQC_RTRSS8TCEN;
2940 			else if (tcs > 1)
2941 				mrqc = IXGBE_MRQC_RTRSS4TCEN;
2942 			else
2943 				mrqc = IXGBE_MRQC_RSSEN;
2944 		}
2945 	}
2946 
2947 	/* Perform hash on these packet types */
2948 	mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4 |
2949 		IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
2950 		IXGBE_MRQC_RSS_FIELD_IPV6 |
2951 		IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2952 
2953 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
2954 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
2955 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
2956 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2957 
2958 	IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2959 }
2960 
2961 /**
2962  * ixgbe_configure_rscctl - enable RSC for the indicated ring
2963  * @adapter:    address of board private structure
2964  * @index:      index of ring to set
2965  **/
2966 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2967 				   struct ixgbe_ring *ring)
2968 {
2969 	struct ixgbe_hw *hw = &adapter->hw;
2970 	u32 rscctrl;
2971 	u8 reg_idx = ring->reg_idx;
2972 
2973 	if (!ring_is_rsc_enabled(ring))
2974 		return;
2975 
2976 	rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2977 	rscctrl |= IXGBE_RSCCTL_RSCEN;
2978 	/*
2979 	 * we must limit the number of descriptors so that the
2980 	 * total size of max desc * buf_len is not greater
2981 	 * than 65536
2982 	 */
2983 #if (PAGE_SIZE <= 8192)
2984 	rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2985 #elif (PAGE_SIZE <= 16384)
2986 	rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2987 #else
2988 	rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2989 #endif
2990 	IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2991 }
2992 
2993 #define IXGBE_MAX_RX_DESC_POLL 10
2994 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2995 				       struct ixgbe_ring *ring)
2996 {
2997 	struct ixgbe_hw *hw = &adapter->hw;
2998 	int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2999 	u32 rxdctl;
3000 	u8 reg_idx = ring->reg_idx;
3001 
3002 	/* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3003 	if (hw->mac.type == ixgbe_mac_82598EB &&
3004 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3005 		return;
3006 
3007 	do {
3008 		usleep_range(1000, 2000);
3009 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3010 	} while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3011 
3012 	if (!wait_loop) {
3013 		e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3014 		      "the polling period\n", reg_idx);
3015 	}
3016 }
3017 
3018 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3019 			    struct ixgbe_ring *ring)
3020 {
3021 	struct ixgbe_hw *hw = &adapter->hw;
3022 	int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3023 	u32 rxdctl;
3024 	u8 reg_idx = ring->reg_idx;
3025 
3026 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3027 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3028 
3029 	/* write value back with RXDCTL.ENABLE bit cleared */
3030 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3031 
3032 	if (hw->mac.type == ixgbe_mac_82598EB &&
3033 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3034 		return;
3035 
3036 	/* the hardware may take up to 100us to really disable the rx queue */
3037 	do {
3038 		udelay(10);
3039 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3040 	} while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3041 
3042 	if (!wait_loop) {
3043 		e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3044 		      "the polling period\n", reg_idx);
3045 	}
3046 }
3047 
3048 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3049 			     struct ixgbe_ring *ring)
3050 {
3051 	struct ixgbe_hw *hw = &adapter->hw;
3052 	u64 rdba = ring->dma;
3053 	u32 rxdctl;
3054 	u8 reg_idx = ring->reg_idx;
3055 
3056 	/* disable queue to avoid issues while updating state */
3057 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3058 	ixgbe_disable_rx_queue(adapter, ring);
3059 
3060 	IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3061 	IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3062 	IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3063 			ring->count * sizeof(union ixgbe_adv_rx_desc));
3064 	IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3065 	IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3066 	ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
3067 
3068 	ixgbe_configure_srrctl(adapter, ring);
3069 	ixgbe_configure_rscctl(adapter, ring);
3070 
3071 	/* If operating in IOV mode set RLPML for X540 */
3072 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
3073 	    hw->mac.type == ixgbe_mac_X540) {
3074 		rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
3075 		rxdctl |= ((ring->netdev->mtu + ETH_HLEN +
3076 			    ETH_FCS_LEN + VLAN_HLEN) | IXGBE_RXDCTL_RLPML_EN);
3077 	}
3078 
3079 	if (hw->mac.type == ixgbe_mac_82598EB) {
3080 		/*
3081 		 * enable cache line friendly hardware writes:
3082 		 * PTHRESH=32 descriptors (half the internal cache),
3083 		 * this also removes ugly rx_no_buffer_count increment
3084 		 * HTHRESH=4 descriptors (to minimize latency on fetch)
3085 		 * WTHRESH=8 burst writeback up to two cache lines
3086 		 */
3087 		rxdctl &= ~0x3FFFFF;
3088 		rxdctl |=  0x080420;
3089 	}
3090 
3091 	/* enable receive descriptor ring */
3092 	rxdctl |= IXGBE_RXDCTL_ENABLE;
3093 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3094 
3095 	ixgbe_rx_desc_queue_enable(adapter, ring);
3096 	ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3097 }
3098 
3099 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3100 {
3101 	struct ixgbe_hw *hw = &adapter->hw;
3102 	int rss_i = adapter->ring_feature[RING_F_RSS].indices;
3103 	int p;
3104 
3105 	/* PSRTYPE must be initialized in non 82598 adapters */
3106 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3107 		      IXGBE_PSRTYPE_UDPHDR |
3108 		      IXGBE_PSRTYPE_IPV4HDR |
3109 		      IXGBE_PSRTYPE_L2HDR |
3110 		      IXGBE_PSRTYPE_IPV6HDR;
3111 
3112 	if (hw->mac.type == ixgbe_mac_82598EB)
3113 		return;
3114 
3115 	if (rss_i > 3)
3116 		psrtype |= 2 << 29;
3117 	else if (rss_i > 1)
3118 		psrtype |= 1 << 29;
3119 
3120 	for (p = 0; p < adapter->num_rx_pools; p++)
3121 		IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(p)),
3122 				psrtype);
3123 }
3124 
3125 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3126 {
3127 	struct ixgbe_hw *hw = &adapter->hw;
3128 	u32 reg_offset, vf_shift;
3129 	u32 gcr_ext, vmdctl;
3130 	int i;
3131 
3132 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3133 		return;
3134 
3135 	vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3136 	vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
3137 	vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
3138 	vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
3139 	vmdctl |= IXGBE_VT_CTL_REPLEN;
3140 	IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
3141 
3142 	vf_shift = VMDQ_P(0) % 32;
3143 	reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
3144 
3145 	/* Enable only the PF's pool for Tx/Rx */
3146 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (~0) << vf_shift);
3147 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
3148 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (~0) << vf_shift);
3149 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
3150 	IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3151 
3152 	/* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3153 	hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
3154 
3155 	/*
3156 	 * Set up VF register offsets for selected VT Mode,
3157 	 * i.e. 32 or 64 VFs for SR-IOV
3158 	 */
3159 	switch (adapter->ring_feature[RING_F_VMDQ].mask) {
3160 	case IXGBE_82599_VMDQ_8Q_MASK:
3161 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
3162 		break;
3163 	case IXGBE_82599_VMDQ_4Q_MASK:
3164 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
3165 		break;
3166 	default:
3167 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
3168 		break;
3169 	}
3170 
3171 	IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3172 
3173 	/* enable Tx loopback for VF/PF communication */
3174 	IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3175 
3176 	/* Enable MAC Anti-Spoofing */
3177 	hw->mac.ops.set_mac_anti_spoofing(hw, (adapter->num_vfs != 0),
3178 					  adapter->num_vfs);
3179 	/* For VFs that have spoof checking turned off */
3180 	for (i = 0; i < adapter->num_vfs; i++) {
3181 		if (!adapter->vfinfo[i].spoofchk_enabled)
3182 			ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
3183 	}
3184 }
3185 
3186 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3187 {
3188 	struct ixgbe_hw *hw = &adapter->hw;
3189 	struct net_device *netdev = adapter->netdev;
3190 	int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3191 	struct ixgbe_ring *rx_ring;
3192 	int i;
3193 	u32 mhadd, hlreg0;
3194 
3195 #ifdef IXGBE_FCOE
3196 	/* adjust max frame to be able to do baby jumbo for FCoE */
3197 	if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3198 	    (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3199 		max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3200 
3201 #endif /* IXGBE_FCOE */
3202 	mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3203 	if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3204 		mhadd &= ~IXGBE_MHADD_MFS_MASK;
3205 		mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3206 
3207 		IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3208 	}
3209 
3210 	/* MHADD will allow an extra 4 bytes past for vlan tagged frames */
3211 	max_frame += VLAN_HLEN;
3212 
3213 	hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3214 	/* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3215 	hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3216 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3217 
3218 	/*
3219 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
3220 	 * the Base and Length of the Rx Descriptor Ring
3221 	 */
3222 	for (i = 0; i < adapter->num_rx_queues; i++) {
3223 		rx_ring = adapter->rx_ring[i];
3224 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3225 			set_ring_rsc_enabled(rx_ring);
3226 		else
3227 			clear_ring_rsc_enabled(rx_ring);
3228 	}
3229 }
3230 
3231 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3232 {
3233 	struct ixgbe_hw *hw = &adapter->hw;
3234 	u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3235 
3236 	switch (hw->mac.type) {
3237 	case ixgbe_mac_82598EB:
3238 		/*
3239 		 * For VMDq support of different descriptor types or
3240 		 * buffer sizes through the use of multiple SRRCTL
3241 		 * registers, RDRXCTL.MVMEN must be set to 1
3242 		 *
3243 		 * also, the manual doesn't mention it clearly but DCA hints
3244 		 * will only use queue 0's tags unless this bit is set.  Side
3245 		 * effects of setting this bit are only that SRRCTL must be
3246 		 * fully programmed [0..15]
3247 		 */
3248 		rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3249 		break;
3250 	case ixgbe_mac_82599EB:
3251 	case ixgbe_mac_X540:
3252 		/* Disable RSC for ACK packets */
3253 		IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3254 		   (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3255 		rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3256 		/* hardware requires some bits to be set by default */
3257 		rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3258 		rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3259 		break;
3260 	default:
3261 		/* We should do nothing since we don't know this hardware */
3262 		return;
3263 	}
3264 
3265 	IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3266 }
3267 
3268 /**
3269  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3270  * @adapter: board private structure
3271  *
3272  * Configure the Rx unit of the MAC after a reset.
3273  **/
3274 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3275 {
3276 	struct ixgbe_hw *hw = &adapter->hw;
3277 	int i;
3278 	u32 rxctrl;
3279 
3280 	/* disable receives while setting up the descriptors */
3281 	rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3282 	IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3283 
3284 	ixgbe_setup_psrtype(adapter);
3285 	ixgbe_setup_rdrxctl(adapter);
3286 
3287 	/* Program registers for the distribution of queues */
3288 	ixgbe_setup_mrqc(adapter);
3289 
3290 	/* set_rx_buffer_len must be called before ring initialization */
3291 	ixgbe_set_rx_buffer_len(adapter);
3292 
3293 	/*
3294 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
3295 	 * the Base and Length of the Rx Descriptor Ring
3296 	 */
3297 	for (i = 0; i < adapter->num_rx_queues; i++)
3298 		ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3299 
3300 	/* disable drop enable for 82598 parts */
3301 	if (hw->mac.type == ixgbe_mac_82598EB)
3302 		rxctrl |= IXGBE_RXCTRL_DMBYPS;
3303 
3304 	/* enable all receives */
3305 	rxctrl |= IXGBE_RXCTRL_RXEN;
3306 	hw->mac.ops.enable_rx_dma(hw, rxctrl);
3307 }
3308 
3309 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3310 {
3311 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3312 	struct ixgbe_hw *hw = &adapter->hw;
3313 
3314 	/* add VID to filter table */
3315 	hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true);
3316 	set_bit(vid, adapter->active_vlans);
3317 
3318 	return 0;
3319 }
3320 
3321 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3322 {
3323 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3324 	struct ixgbe_hw *hw = &adapter->hw;
3325 
3326 	/* remove VID from filter table */
3327 	hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), false);
3328 	clear_bit(vid, adapter->active_vlans);
3329 
3330 	return 0;
3331 }
3332 
3333 /**
3334  * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3335  * @adapter: driver data
3336  */
3337 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3338 {
3339 	struct ixgbe_hw *hw = &adapter->hw;
3340 	u32 vlnctrl;
3341 
3342 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3343 	vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3344 	IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3345 }
3346 
3347 /**
3348  * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3349  * @adapter: driver data
3350  */
3351 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3352 {
3353 	struct ixgbe_hw *hw = &adapter->hw;
3354 	u32 vlnctrl;
3355 
3356 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3357 	vlnctrl |= IXGBE_VLNCTRL_VFE;
3358 	vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3359 	IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3360 }
3361 
3362 /**
3363  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3364  * @adapter: driver data
3365  */
3366 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3367 {
3368 	struct ixgbe_hw *hw = &adapter->hw;
3369 	u32 vlnctrl;
3370 	int i, j;
3371 
3372 	switch (hw->mac.type) {
3373 	case ixgbe_mac_82598EB:
3374 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3375 		vlnctrl &= ~IXGBE_VLNCTRL_VME;
3376 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3377 		break;
3378 	case ixgbe_mac_82599EB:
3379 	case ixgbe_mac_X540:
3380 		for (i = 0; i < adapter->num_rx_queues; i++) {
3381 			j = adapter->rx_ring[i]->reg_idx;
3382 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3383 			vlnctrl &= ~IXGBE_RXDCTL_VME;
3384 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3385 		}
3386 		break;
3387 	default:
3388 		break;
3389 	}
3390 }
3391 
3392 /**
3393  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3394  * @adapter: driver data
3395  */
3396 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3397 {
3398 	struct ixgbe_hw *hw = &adapter->hw;
3399 	u32 vlnctrl;
3400 	int i, j;
3401 
3402 	switch (hw->mac.type) {
3403 	case ixgbe_mac_82598EB:
3404 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3405 		vlnctrl |= IXGBE_VLNCTRL_VME;
3406 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3407 		break;
3408 	case ixgbe_mac_82599EB:
3409 	case ixgbe_mac_X540:
3410 		for (i = 0; i < adapter->num_rx_queues; i++) {
3411 			j = adapter->rx_ring[i]->reg_idx;
3412 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3413 			vlnctrl |= IXGBE_RXDCTL_VME;
3414 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3415 		}
3416 		break;
3417 	default:
3418 		break;
3419 	}
3420 }
3421 
3422 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3423 {
3424 	u16 vid;
3425 
3426 	ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3427 
3428 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3429 		ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3430 }
3431 
3432 /**
3433  * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3434  * @netdev: network interface device structure
3435  *
3436  * Writes unicast address list to the RAR table.
3437  * Returns: -ENOMEM on failure/insufficient address space
3438  *                0 on no addresses written
3439  *                X on writing X addresses to the RAR table
3440  **/
3441 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3442 {
3443 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3444 	struct ixgbe_hw *hw = &adapter->hw;
3445 	unsigned int rar_entries = hw->mac.num_rar_entries - 1;
3446 	int count = 0;
3447 
3448 	/* In SR-IOV mode significantly less RAR entries are available */
3449 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3450 		rar_entries = IXGBE_MAX_PF_MACVLANS - 1;
3451 
3452 	/* return ENOMEM indicating insufficient memory for addresses */
3453 	if (netdev_uc_count(netdev) > rar_entries)
3454 		return -ENOMEM;
3455 
3456 	if (!netdev_uc_empty(netdev)) {
3457 		struct netdev_hw_addr *ha;
3458 		/* return error if we do not support writing to RAR table */
3459 		if (!hw->mac.ops.set_rar)
3460 			return -ENOMEM;
3461 
3462 		netdev_for_each_uc_addr(ha, netdev) {
3463 			if (!rar_entries)
3464 				break;
3465 			hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3466 					    VMDQ_P(0), IXGBE_RAH_AV);
3467 			count++;
3468 		}
3469 	}
3470 	/* write the addresses in reverse order to avoid write combining */
3471 	for (; rar_entries > 0 ; rar_entries--)
3472 		hw->mac.ops.clear_rar(hw, rar_entries);
3473 
3474 	return count;
3475 }
3476 
3477 /**
3478  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3479  * @netdev: network interface device structure
3480  *
3481  * The set_rx_method entry point is called whenever the unicast/multicast
3482  * address list or the network interface flags are updated.  This routine is
3483  * responsible for configuring the hardware for proper unicast, multicast and
3484  * promiscuous mode.
3485  **/
3486 void ixgbe_set_rx_mode(struct net_device *netdev)
3487 {
3488 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3489 	struct ixgbe_hw *hw = &adapter->hw;
3490 	u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3491 	int count;
3492 
3493 	/* Check for Promiscuous and All Multicast modes */
3494 
3495 	fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3496 
3497 	/* set all bits that we expect to always be set */
3498 	fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
3499 	fctrl |= IXGBE_FCTRL_BAM;
3500 	fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3501 	fctrl |= IXGBE_FCTRL_PMCF;
3502 
3503 	/* clear the bits we are changing the status of */
3504 	fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3505 
3506 	if (netdev->flags & IFF_PROMISC) {
3507 		hw->addr_ctrl.user_set_promisc = true;
3508 		fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3509 		vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3510 		/* don't hardware filter vlans in promisc mode */
3511 		ixgbe_vlan_filter_disable(adapter);
3512 	} else {
3513 		if (netdev->flags & IFF_ALLMULTI) {
3514 			fctrl |= IXGBE_FCTRL_MPE;
3515 			vmolr |= IXGBE_VMOLR_MPE;
3516 		} else {
3517 			/*
3518 			 * Write addresses to the MTA, if the attempt fails
3519 			 * then we should just turn on promiscuous mode so
3520 			 * that we can at least receive multicast traffic
3521 			 */
3522 			hw->mac.ops.update_mc_addr_list(hw, netdev);
3523 			vmolr |= IXGBE_VMOLR_ROMPE;
3524 		}
3525 		ixgbe_vlan_filter_enable(adapter);
3526 		hw->addr_ctrl.user_set_promisc = false;
3527 	}
3528 
3529 	/*
3530 	 * Write addresses to available RAR registers, if there is not
3531 	 * sufficient space to store all the addresses then enable
3532 	 * unicast promiscuous mode
3533 	 */
3534 	count = ixgbe_write_uc_addr_list(netdev);
3535 	if (count < 0) {
3536 		fctrl |= IXGBE_FCTRL_UPE;
3537 		vmolr |= IXGBE_VMOLR_ROPE;
3538 	}
3539 
3540 	if (adapter->num_vfs)
3541 		ixgbe_restore_vf_multicasts(adapter);
3542 
3543 	if (hw->mac.type != ixgbe_mac_82598EB) {
3544 		vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
3545 			 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3546 			   IXGBE_VMOLR_ROPE);
3547 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
3548 	}
3549 
3550 	/* This is useful for sniffing bad packets. */
3551 	if (adapter->netdev->features & NETIF_F_RXALL) {
3552 		/* UPE and MPE will be handled by normal PROMISC logic
3553 		 * in e1000e_set_rx_mode */
3554 		fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
3555 			  IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
3556 			  IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
3557 
3558 		fctrl &= ~(IXGBE_FCTRL_DPF);
3559 		/* NOTE:  VLAN filtering is disabled by setting PROMISC */
3560 	}
3561 
3562 	IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3563 
3564 	if (netdev->features & NETIF_F_HW_VLAN_RX)
3565 		ixgbe_vlan_strip_enable(adapter);
3566 	else
3567 		ixgbe_vlan_strip_disable(adapter);
3568 }
3569 
3570 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3571 {
3572 	int q_idx;
3573 
3574 	for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
3575 		napi_enable(&adapter->q_vector[q_idx]->napi);
3576 }
3577 
3578 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3579 {
3580 	int q_idx;
3581 
3582 	for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
3583 		napi_disable(&adapter->q_vector[q_idx]->napi);
3584 }
3585 
3586 #ifdef CONFIG_IXGBE_DCB
3587 /**
3588  * ixgbe_configure_dcb - Configure DCB hardware
3589  * @adapter: ixgbe adapter struct
3590  *
3591  * This is called by the driver on open to configure the DCB hardware.
3592  * This is also called by the gennetlink interface when reconfiguring
3593  * the DCB state.
3594  */
3595 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3596 {
3597 	struct ixgbe_hw *hw = &adapter->hw;
3598 	int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3599 
3600 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3601 		if (hw->mac.type == ixgbe_mac_82598EB)
3602 			netif_set_gso_max_size(adapter->netdev, 65536);
3603 		return;
3604 	}
3605 
3606 	if (hw->mac.type == ixgbe_mac_82598EB)
3607 		netif_set_gso_max_size(adapter->netdev, 32768);
3608 
3609 	hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3610 
3611 #ifdef IXGBE_FCOE
3612 	if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3613 		max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3614 #endif
3615 
3616 	/* reconfigure the hardware */
3617 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
3618 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3619 						DCB_TX_CONFIG);
3620 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3621 						DCB_RX_CONFIG);
3622 		ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3623 	} else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
3624 		ixgbe_dcb_hw_ets(&adapter->hw,
3625 				 adapter->ixgbe_ieee_ets,
3626 				 max_frame);
3627 		ixgbe_dcb_hw_pfc_config(&adapter->hw,
3628 					adapter->ixgbe_ieee_pfc->pfc_en,
3629 					adapter->ixgbe_ieee_ets->prio_tc);
3630 	}
3631 
3632 	/* Enable RSS Hash per TC */
3633 	if (hw->mac.type != ixgbe_mac_82598EB) {
3634 		u32 msb = 0;
3635 		u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
3636 
3637 		while (rss_i) {
3638 			msb++;
3639 			rss_i >>= 1;
3640 		}
3641 
3642 		/* write msb to all 8 TCs in one write */
3643 		IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
3644 	}
3645 }
3646 #endif
3647 
3648 /* Additional bittime to account for IXGBE framing */
3649 #define IXGBE_ETH_FRAMING 20
3650 
3651 /**
3652  * ixgbe_hpbthresh - calculate high water mark for flow control
3653  *
3654  * @adapter: board private structure to calculate for
3655  * @pb: packet buffer to calculate
3656  */
3657 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
3658 {
3659 	struct ixgbe_hw *hw = &adapter->hw;
3660 	struct net_device *dev = adapter->netdev;
3661 	int link, tc, kb, marker;
3662 	u32 dv_id, rx_pba;
3663 
3664 	/* Calculate max LAN frame size */
3665 	tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
3666 
3667 #ifdef IXGBE_FCOE
3668 	/* FCoE traffic class uses FCOE jumbo frames */
3669 	if ((dev->features & NETIF_F_FCOE_MTU) &&
3670 	    (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
3671 	    (pb == ixgbe_fcoe_get_tc(adapter)))
3672 		tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3673 
3674 #endif
3675 	/* Calculate delay value for device */
3676 	switch (hw->mac.type) {
3677 	case ixgbe_mac_X540:
3678 		dv_id = IXGBE_DV_X540(link, tc);
3679 		break;
3680 	default:
3681 		dv_id = IXGBE_DV(link, tc);
3682 		break;
3683 	}
3684 
3685 	/* Loopback switch introduces additional latency */
3686 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3687 		dv_id += IXGBE_B2BT(tc);
3688 
3689 	/* Delay value is calculated in bit times convert to KB */
3690 	kb = IXGBE_BT2KB(dv_id);
3691 	rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
3692 
3693 	marker = rx_pba - kb;
3694 
3695 	/* It is possible that the packet buffer is not large enough
3696 	 * to provide required headroom. In this case throw an error
3697 	 * to user and a do the best we can.
3698 	 */
3699 	if (marker < 0) {
3700 		e_warn(drv, "Packet Buffer(%i) can not provide enough"
3701 			    "headroom to support flow control."
3702 			    "Decrease MTU or number of traffic classes\n", pb);
3703 		marker = tc + 1;
3704 	}
3705 
3706 	return marker;
3707 }
3708 
3709 /**
3710  * ixgbe_lpbthresh - calculate low water mark for for flow control
3711  *
3712  * @adapter: board private structure to calculate for
3713  * @pb: packet buffer to calculate
3714  */
3715 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter)
3716 {
3717 	struct ixgbe_hw *hw = &adapter->hw;
3718 	struct net_device *dev = adapter->netdev;
3719 	int tc;
3720 	u32 dv_id;
3721 
3722 	/* Calculate max LAN frame size */
3723 	tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
3724 
3725 	/* Calculate delay value for device */
3726 	switch (hw->mac.type) {
3727 	case ixgbe_mac_X540:
3728 		dv_id = IXGBE_LOW_DV_X540(tc);
3729 		break;
3730 	default:
3731 		dv_id = IXGBE_LOW_DV(tc);
3732 		break;
3733 	}
3734 
3735 	/* Delay value is calculated in bit times convert to KB */
3736 	return IXGBE_BT2KB(dv_id);
3737 }
3738 
3739 /*
3740  * ixgbe_pbthresh_setup - calculate and setup high low water marks
3741  */
3742 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
3743 {
3744 	struct ixgbe_hw *hw = &adapter->hw;
3745 	int num_tc = netdev_get_num_tc(adapter->netdev);
3746 	int i;
3747 
3748 	if (!num_tc)
3749 		num_tc = 1;
3750 
3751 	hw->fc.low_water = ixgbe_lpbthresh(adapter);
3752 
3753 	for (i = 0; i < num_tc; i++) {
3754 		hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
3755 
3756 		/* Low water marks must not be larger than high water marks */
3757 		if (hw->fc.low_water > hw->fc.high_water[i])
3758 			hw->fc.low_water = 0;
3759 	}
3760 }
3761 
3762 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
3763 {
3764 	struct ixgbe_hw *hw = &adapter->hw;
3765 	int hdrm;
3766 	u8 tc = netdev_get_num_tc(adapter->netdev);
3767 
3768 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3769 	    adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3770 		hdrm = 32 << adapter->fdir_pballoc;
3771 	else
3772 		hdrm = 0;
3773 
3774 	hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
3775 	ixgbe_pbthresh_setup(adapter);
3776 }
3777 
3778 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
3779 {
3780 	struct ixgbe_hw *hw = &adapter->hw;
3781 	struct hlist_node *node, *node2;
3782 	struct ixgbe_fdir_filter *filter;
3783 
3784 	spin_lock(&adapter->fdir_perfect_lock);
3785 
3786 	if (!hlist_empty(&adapter->fdir_filter_list))
3787 		ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
3788 
3789 	hlist_for_each_entry_safe(filter, node, node2,
3790 				  &adapter->fdir_filter_list, fdir_node) {
3791 		ixgbe_fdir_write_perfect_filter_82599(hw,
3792 				&filter->filter,
3793 				filter->sw_idx,
3794 				(filter->action == IXGBE_FDIR_DROP_QUEUE) ?
3795 				IXGBE_FDIR_DROP_QUEUE :
3796 				adapter->rx_ring[filter->action]->reg_idx);
3797 	}
3798 
3799 	spin_unlock(&adapter->fdir_perfect_lock);
3800 }
3801 
3802 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3803 {
3804 	struct ixgbe_hw *hw = &adapter->hw;
3805 
3806 	ixgbe_configure_pb(adapter);
3807 #ifdef CONFIG_IXGBE_DCB
3808 	ixgbe_configure_dcb(adapter);
3809 #endif
3810 
3811 	ixgbe_set_rx_mode(adapter->netdev);
3812 	ixgbe_restore_vlan(adapter);
3813 
3814 	switch (hw->mac.type) {
3815 	case ixgbe_mac_82599EB:
3816 	case ixgbe_mac_X540:
3817 		hw->mac.ops.disable_rx_buff(hw);
3818 		break;
3819 	default:
3820 		break;
3821 	}
3822 
3823 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3824 		ixgbe_init_fdir_signature_82599(&adapter->hw,
3825 						adapter->fdir_pballoc);
3826 	} else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3827 		ixgbe_init_fdir_perfect_82599(&adapter->hw,
3828 					      adapter->fdir_pballoc);
3829 		ixgbe_fdir_filter_restore(adapter);
3830 	}
3831 
3832 	switch (hw->mac.type) {
3833 	case ixgbe_mac_82599EB:
3834 	case ixgbe_mac_X540:
3835 		hw->mac.ops.enable_rx_buff(hw);
3836 		break;
3837 	default:
3838 		break;
3839 	}
3840 
3841 	ixgbe_configure_virtualization(adapter);
3842 
3843 #ifdef IXGBE_FCOE
3844 	/* configure FCoE L2 filters, redirection table, and Rx control */
3845 	ixgbe_configure_fcoe(adapter);
3846 
3847 #endif /* IXGBE_FCOE */
3848 	ixgbe_configure_tx(adapter);
3849 	ixgbe_configure_rx(adapter);
3850 }
3851 
3852 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3853 {
3854 	switch (hw->phy.type) {
3855 	case ixgbe_phy_sfp_avago:
3856 	case ixgbe_phy_sfp_ftl:
3857 	case ixgbe_phy_sfp_intel:
3858 	case ixgbe_phy_sfp_unknown:
3859 	case ixgbe_phy_sfp_passive_tyco:
3860 	case ixgbe_phy_sfp_passive_unknown:
3861 	case ixgbe_phy_sfp_active_unknown:
3862 	case ixgbe_phy_sfp_ftl_active:
3863 		return true;
3864 	case ixgbe_phy_nl:
3865 		if (hw->mac.type == ixgbe_mac_82598EB)
3866 			return true;
3867 	default:
3868 		return false;
3869 	}
3870 }
3871 
3872 /**
3873  * ixgbe_sfp_link_config - set up SFP+ link
3874  * @adapter: pointer to private adapter struct
3875  **/
3876 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3877 {
3878 	/*
3879 	 * We are assuming the worst case scenario here, and that
3880 	 * is that an SFP was inserted/removed after the reset
3881 	 * but before SFP detection was enabled.  As such the best
3882 	 * solution is to just start searching as soon as we start
3883 	 */
3884 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3885 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
3886 
3887 	adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
3888 }
3889 
3890 /**
3891  * ixgbe_non_sfp_link_config - set up non-SFP+ link
3892  * @hw: pointer to private hardware struct
3893  *
3894  * Returns 0 on success, negative on failure
3895  **/
3896 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3897 {
3898 	u32 autoneg;
3899 	bool negotiation, link_up = false;
3900 	u32 ret = IXGBE_ERR_LINK_SETUP;
3901 
3902 	if (hw->mac.ops.check_link)
3903 		ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3904 
3905 	if (ret)
3906 		goto link_cfg_out;
3907 
3908 	autoneg = hw->phy.autoneg_advertised;
3909 	if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
3910 		ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3911 							&negotiation);
3912 	if (ret)
3913 		goto link_cfg_out;
3914 
3915 	if (hw->mac.ops.setup_link)
3916 		ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3917 link_cfg_out:
3918 	return ret;
3919 }
3920 
3921 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3922 {
3923 	struct ixgbe_hw *hw = &adapter->hw;
3924 	u32 gpie = 0;
3925 
3926 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3927 		gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3928 		       IXGBE_GPIE_OCD;
3929 		gpie |= IXGBE_GPIE_EIAME;
3930 		/*
3931 		 * use EIAM to auto-mask when MSI-X interrupt is asserted
3932 		 * this saves a register write for every interrupt
3933 		 */
3934 		switch (hw->mac.type) {
3935 		case ixgbe_mac_82598EB:
3936 			IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3937 			break;
3938 		case ixgbe_mac_82599EB:
3939 		case ixgbe_mac_X540:
3940 		default:
3941 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3942 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3943 			break;
3944 		}
3945 	} else {
3946 		/* legacy interrupts, use EIAM to auto-mask when reading EICR,
3947 		 * specifically only auto mask tx and rx interrupts */
3948 		IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3949 	}
3950 
3951 	/* XXX: to interrupt immediately for EICS writes, enable this */
3952 	/* gpie |= IXGBE_GPIE_EIMEN; */
3953 
3954 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3955 		gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3956 
3957 		switch (adapter->ring_feature[RING_F_VMDQ].mask) {
3958 		case IXGBE_82599_VMDQ_8Q_MASK:
3959 			gpie |= IXGBE_GPIE_VTMODE_16;
3960 			break;
3961 		case IXGBE_82599_VMDQ_4Q_MASK:
3962 			gpie |= IXGBE_GPIE_VTMODE_32;
3963 			break;
3964 		default:
3965 			gpie |= IXGBE_GPIE_VTMODE_64;
3966 			break;
3967 		}
3968 	}
3969 
3970 	/* Enable Thermal over heat sensor interrupt */
3971 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
3972 		switch (adapter->hw.mac.type) {
3973 		case ixgbe_mac_82599EB:
3974 			gpie |= IXGBE_SDP0_GPIEN;
3975 			break;
3976 		case ixgbe_mac_X540:
3977 			gpie |= IXGBE_EIMS_TS;
3978 			break;
3979 		default:
3980 			break;
3981 		}
3982 	}
3983 
3984 	/* Enable fan failure interrupt */
3985 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3986 		gpie |= IXGBE_SDP1_GPIEN;
3987 
3988 	if (hw->mac.type == ixgbe_mac_82599EB) {
3989 		gpie |= IXGBE_SDP1_GPIEN;
3990 		gpie |= IXGBE_SDP2_GPIEN;
3991 	}
3992 
3993 	IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3994 }
3995 
3996 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
3997 {
3998 	struct ixgbe_hw *hw = &adapter->hw;
3999 	int err;
4000 	u32 ctrl_ext;
4001 
4002 	ixgbe_get_hw_control(adapter);
4003 	ixgbe_setup_gpie(adapter);
4004 
4005 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4006 		ixgbe_configure_msix(adapter);
4007 	else
4008 		ixgbe_configure_msi_and_legacy(adapter);
4009 
4010 	/* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
4011 	if (hw->mac.ops.enable_tx_laser &&
4012 	    ((hw->phy.multispeed_fiber) ||
4013 	     ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4014 	      (hw->mac.type == ixgbe_mac_82599EB))))
4015 		hw->mac.ops.enable_tx_laser(hw);
4016 
4017 	clear_bit(__IXGBE_DOWN, &adapter->state);
4018 	ixgbe_napi_enable_all(adapter);
4019 
4020 	if (ixgbe_is_sfp(hw)) {
4021 		ixgbe_sfp_link_config(adapter);
4022 	} else {
4023 		err = ixgbe_non_sfp_link_config(hw);
4024 		if (err)
4025 			e_err(probe, "link_config FAILED %d\n", err);
4026 	}
4027 
4028 	/* clear any pending interrupts, may auto mask */
4029 	IXGBE_READ_REG(hw, IXGBE_EICR);
4030 	ixgbe_irq_enable(adapter, true, true);
4031 
4032 	/*
4033 	 * If this adapter has a fan, check to see if we had a failure
4034 	 * before we enabled the interrupt.
4035 	 */
4036 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
4037 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
4038 		if (esdp & IXGBE_ESDP_SDP1)
4039 			e_crit(drv, "Fan has stopped, replace the adapter\n");
4040 	}
4041 
4042 	/* enable transmits */
4043 	netif_tx_start_all_queues(adapter->netdev);
4044 
4045 	/* bring the link up in the watchdog, this could race with our first
4046 	 * link up interrupt but shouldn't be a problem */
4047 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4048 	adapter->link_check_timeout = jiffies;
4049 	mod_timer(&adapter->service_timer, jiffies);
4050 
4051 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
4052 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
4053 	ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
4054 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
4055 }
4056 
4057 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
4058 {
4059 	WARN_ON(in_interrupt());
4060 	/* put off any impending NetWatchDogTimeout */
4061 	adapter->netdev->trans_start = jiffies;
4062 
4063 	while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
4064 		usleep_range(1000, 2000);
4065 	ixgbe_down(adapter);
4066 	/*
4067 	 * If SR-IOV enabled then wait a bit before bringing the adapter
4068 	 * back up to give the VFs time to respond to the reset.  The
4069 	 * two second wait is based upon the watchdog timer cycle in
4070 	 * the VF driver.
4071 	 */
4072 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4073 		msleep(2000);
4074 	ixgbe_up(adapter);
4075 	clear_bit(__IXGBE_RESETTING, &adapter->state);
4076 }
4077 
4078 void ixgbe_up(struct ixgbe_adapter *adapter)
4079 {
4080 	/* hardware has been reset, we need to reload some things */
4081 	ixgbe_configure(adapter);
4082 
4083 	ixgbe_up_complete(adapter);
4084 }
4085 
4086 void ixgbe_reset(struct ixgbe_adapter *adapter)
4087 {
4088 	struct ixgbe_hw *hw = &adapter->hw;
4089 	int err;
4090 
4091 	/* lock SFP init bit to prevent race conditions with the watchdog */
4092 	while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
4093 		usleep_range(1000, 2000);
4094 
4095 	/* clear all SFP and link config related flags while holding SFP_INIT */
4096 	adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
4097 			     IXGBE_FLAG2_SFP_NEEDS_RESET);
4098 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
4099 
4100 	err = hw->mac.ops.init_hw(hw);
4101 	switch (err) {
4102 	case 0:
4103 	case IXGBE_ERR_SFP_NOT_PRESENT:
4104 	case IXGBE_ERR_SFP_NOT_SUPPORTED:
4105 		break;
4106 	case IXGBE_ERR_MASTER_REQUESTS_PENDING:
4107 		e_dev_err("master disable timed out\n");
4108 		break;
4109 	case IXGBE_ERR_EEPROM_VERSION:
4110 		/* We are running on a pre-production device, log a warning */
4111 		e_dev_warn("This device is a pre-production adapter/LOM. "
4112 			   "Please be aware there may be issues associated with "
4113 			   "your hardware.  If you are experiencing problems "
4114 			   "please contact your Intel or hardware "
4115 			   "representative who provided you with this "
4116 			   "hardware.\n");
4117 		break;
4118 	default:
4119 		e_dev_err("Hardware Error: %d\n", err);
4120 	}
4121 
4122 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
4123 
4124 	/* reprogram the RAR[0] in case user changed it. */
4125 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, VMDQ_P(0), IXGBE_RAH_AV);
4126 
4127 	/* update SAN MAC vmdq pool selection */
4128 	if (hw->mac.san_mac_rar_index)
4129 		hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
4130 }
4131 
4132 /**
4133  * ixgbe_init_rx_page_offset - initialize page offset values for Rx buffers
4134  * @rx_ring: ring to setup
4135  *
4136  * On many IA platforms the L1 cache has a critical stride of 4K, this
4137  * results in each receive buffer starting in the same cache set.  To help
4138  * reduce the pressure on this cache set we can interleave the offsets so
4139  * that only every other buffer will be in the same cache set.
4140  **/
4141 static void ixgbe_init_rx_page_offset(struct ixgbe_ring *rx_ring)
4142 {
4143 	struct ixgbe_rx_buffer *rx_buffer = rx_ring->rx_buffer_info;
4144 	u16 i;
4145 
4146 	for (i = 0; i < rx_ring->count; i += 2) {
4147 		rx_buffer[0].page_offset = 0;
4148 		rx_buffer[1].page_offset = ixgbe_rx_bufsz(rx_ring);
4149 		rx_buffer = &rx_buffer[2];
4150 	}
4151 }
4152 
4153 /**
4154  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4155  * @rx_ring: ring to free buffers from
4156  **/
4157 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4158 {
4159 	struct device *dev = rx_ring->dev;
4160 	unsigned long size;
4161 	u16 i;
4162 
4163 	/* ring already cleared, nothing to do */
4164 	if (!rx_ring->rx_buffer_info)
4165 		return;
4166 
4167 	/* Free all the Rx ring sk_buffs */
4168 	for (i = 0; i < rx_ring->count; i++) {
4169 		struct ixgbe_rx_buffer *rx_buffer;
4170 
4171 		rx_buffer = &rx_ring->rx_buffer_info[i];
4172 		if (rx_buffer->skb) {
4173 			struct sk_buff *skb = rx_buffer->skb;
4174 			if (IXGBE_CB(skb)->page_released) {
4175 				dma_unmap_page(dev,
4176 					       IXGBE_CB(skb)->dma,
4177 					       ixgbe_rx_bufsz(rx_ring),
4178 					       DMA_FROM_DEVICE);
4179 				IXGBE_CB(skb)->page_released = false;
4180 			}
4181 			dev_kfree_skb(skb);
4182 		}
4183 		rx_buffer->skb = NULL;
4184 		if (rx_buffer->dma)
4185 			dma_unmap_page(dev, rx_buffer->dma,
4186 				       ixgbe_rx_pg_size(rx_ring),
4187 				       DMA_FROM_DEVICE);
4188 		rx_buffer->dma = 0;
4189 		if (rx_buffer->page)
4190 			__free_pages(rx_buffer->page,
4191 				     ixgbe_rx_pg_order(rx_ring));
4192 		rx_buffer->page = NULL;
4193 	}
4194 
4195 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4196 	memset(rx_ring->rx_buffer_info, 0, size);
4197 
4198 	ixgbe_init_rx_page_offset(rx_ring);
4199 
4200 	/* Zero out the descriptor ring */
4201 	memset(rx_ring->desc, 0, rx_ring->size);
4202 
4203 	rx_ring->next_to_alloc = 0;
4204 	rx_ring->next_to_clean = 0;
4205 	rx_ring->next_to_use = 0;
4206 }
4207 
4208 /**
4209  * ixgbe_clean_tx_ring - Free Tx Buffers
4210  * @tx_ring: ring to be cleaned
4211  **/
4212 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4213 {
4214 	struct ixgbe_tx_buffer *tx_buffer_info;
4215 	unsigned long size;
4216 	u16 i;
4217 
4218 	/* ring already cleared, nothing to do */
4219 	if (!tx_ring->tx_buffer_info)
4220 		return;
4221 
4222 	/* Free all the Tx ring sk_buffs */
4223 	for (i = 0; i < tx_ring->count; i++) {
4224 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
4225 		ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4226 	}
4227 
4228 	netdev_tx_reset_queue(txring_txq(tx_ring));
4229 
4230 	size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4231 	memset(tx_ring->tx_buffer_info, 0, size);
4232 
4233 	/* Zero out the descriptor ring */
4234 	memset(tx_ring->desc, 0, tx_ring->size);
4235 
4236 	tx_ring->next_to_use = 0;
4237 	tx_ring->next_to_clean = 0;
4238 }
4239 
4240 /**
4241  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4242  * @adapter: board private structure
4243  **/
4244 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4245 {
4246 	int i;
4247 
4248 	for (i = 0; i < adapter->num_rx_queues; i++)
4249 		ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4250 }
4251 
4252 /**
4253  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4254  * @adapter: board private structure
4255  **/
4256 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4257 {
4258 	int i;
4259 
4260 	for (i = 0; i < adapter->num_tx_queues; i++)
4261 		ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4262 }
4263 
4264 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
4265 {
4266 	struct hlist_node *node, *node2;
4267 	struct ixgbe_fdir_filter *filter;
4268 
4269 	spin_lock(&adapter->fdir_perfect_lock);
4270 
4271 	hlist_for_each_entry_safe(filter, node, node2,
4272 				  &adapter->fdir_filter_list, fdir_node) {
4273 		hlist_del(&filter->fdir_node);
4274 		kfree(filter);
4275 	}
4276 	adapter->fdir_filter_count = 0;
4277 
4278 	spin_unlock(&adapter->fdir_perfect_lock);
4279 }
4280 
4281 void ixgbe_down(struct ixgbe_adapter *adapter)
4282 {
4283 	struct net_device *netdev = adapter->netdev;
4284 	struct ixgbe_hw *hw = &adapter->hw;
4285 	u32 rxctrl;
4286 	int i;
4287 
4288 	/* signal that we are down to the interrupt handler */
4289 	set_bit(__IXGBE_DOWN, &adapter->state);
4290 
4291 	/* disable receives */
4292 	rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4293 	IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4294 
4295 	/* disable all enabled rx queues */
4296 	for (i = 0; i < adapter->num_rx_queues; i++)
4297 		/* this call also flushes the previous write */
4298 		ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4299 
4300 	usleep_range(10000, 20000);
4301 
4302 	netif_tx_stop_all_queues(netdev);
4303 
4304 	/* call carrier off first to avoid false dev_watchdog timeouts */
4305 	netif_carrier_off(netdev);
4306 	netif_tx_disable(netdev);
4307 
4308 	ixgbe_irq_disable(adapter);
4309 
4310 	ixgbe_napi_disable_all(adapter);
4311 
4312 	adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
4313 			     IXGBE_FLAG2_RESET_REQUESTED);
4314 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4315 
4316 	del_timer_sync(&adapter->service_timer);
4317 
4318 	if (adapter->num_vfs) {
4319 		/* Clear EITR Select mapping */
4320 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
4321 
4322 		/* Mark all the VFs as inactive */
4323 		for (i = 0 ; i < adapter->num_vfs; i++)
4324 			adapter->vfinfo[i].clear_to_send = false;
4325 
4326 		/* ping all the active vfs to let them know we are going down */
4327 		ixgbe_ping_all_vfs(adapter);
4328 
4329 		/* Disable all VFTE/VFRE TX/RX */
4330 		ixgbe_disable_tx_rx(adapter);
4331 	}
4332 
4333 	/* disable transmits in the hardware now that interrupts are off */
4334 	for (i = 0; i < adapter->num_tx_queues; i++) {
4335 		u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4336 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
4337 	}
4338 
4339 	/* Disable the Tx DMA engine on 82599 and X540 */
4340 	switch (hw->mac.type) {
4341 	case ixgbe_mac_82599EB:
4342 	case ixgbe_mac_X540:
4343 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4344 				(IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4345 				 ~IXGBE_DMATXCTL_TE));
4346 		break;
4347 	default:
4348 		break;
4349 	}
4350 
4351 	if (!pci_channel_offline(adapter->pdev))
4352 		ixgbe_reset(adapter);
4353 
4354 	/* power down the optics for multispeed fiber and 82599 SFP+ fiber */
4355 	if (hw->mac.ops.disable_tx_laser &&
4356 	    ((hw->phy.multispeed_fiber) ||
4357 	     ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4358 	      (hw->mac.type == ixgbe_mac_82599EB))))
4359 		hw->mac.ops.disable_tx_laser(hw);
4360 
4361 	ixgbe_clean_all_tx_rings(adapter);
4362 	ixgbe_clean_all_rx_rings(adapter);
4363 
4364 #ifdef CONFIG_IXGBE_DCA
4365 	/* since we reset the hardware DCA settings were cleared */
4366 	ixgbe_setup_dca(adapter);
4367 #endif
4368 }
4369 
4370 /**
4371  * ixgbe_tx_timeout - Respond to a Tx Hang
4372  * @netdev: network interface device structure
4373  **/
4374 static void ixgbe_tx_timeout(struct net_device *netdev)
4375 {
4376 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4377 
4378 	/* Do the reset outside of interrupt context */
4379 	ixgbe_tx_timeout_reset(adapter);
4380 }
4381 
4382 /**
4383  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4384  * @adapter: board private structure to initialize
4385  *
4386  * ixgbe_sw_init initializes the Adapter private data structure.
4387  * Fields are initialized based on PCI device information and
4388  * OS network device settings (MTU size).
4389  **/
4390 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4391 {
4392 	struct ixgbe_hw *hw = &adapter->hw;
4393 	struct pci_dev *pdev = adapter->pdev;
4394 	unsigned int rss;
4395 #ifdef CONFIG_IXGBE_DCB
4396 	int j;
4397 	struct tc_configuration *tc;
4398 #endif
4399 
4400 	/* PCI config space info */
4401 
4402 	hw->vendor_id = pdev->vendor;
4403 	hw->device_id = pdev->device;
4404 	hw->revision_id = pdev->revision;
4405 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
4406 	hw->subsystem_device_id = pdev->subsystem_device;
4407 
4408 	/* Set capability flags */
4409 	rss = min_t(int, IXGBE_MAX_RSS_INDICES, num_online_cpus());
4410 	adapter->ring_feature[RING_F_RSS].limit = rss;
4411 	switch (hw->mac.type) {
4412 	case ixgbe_mac_82598EB:
4413 		if (hw->device_id == IXGBE_DEV_ID_82598AT)
4414 			adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4415 		adapter->max_q_vectors = MAX_Q_VECTORS_82598;
4416 		break;
4417 	case ixgbe_mac_X540:
4418 		adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4419 	case ixgbe_mac_82599EB:
4420 		adapter->max_q_vectors = MAX_Q_VECTORS_82599;
4421 		adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4422 		adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4423 		if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4424 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4425 		/* Flow Director hash filters enabled */
4426 		adapter->atr_sample_rate = 20;
4427 		adapter->ring_feature[RING_F_FDIR].limit =
4428 							 IXGBE_MAX_FDIR_INDICES;
4429 		adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
4430 #ifdef IXGBE_FCOE
4431 		adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4432 		adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4433 #ifdef CONFIG_IXGBE_DCB
4434 		/* Default traffic class to use for FCoE */
4435 		adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4436 #endif
4437 #endif /* IXGBE_FCOE */
4438 		break;
4439 	default:
4440 		break;
4441 	}
4442 
4443 #ifdef IXGBE_FCOE
4444 	/* FCoE support exists, always init the FCoE lock */
4445 	spin_lock_init(&adapter->fcoe.lock);
4446 
4447 #endif
4448 	/* n-tuple support exists, always init our spinlock */
4449 	spin_lock_init(&adapter->fdir_perfect_lock);
4450 
4451 #ifdef CONFIG_IXGBE_DCB
4452 	switch (hw->mac.type) {
4453 	case ixgbe_mac_X540:
4454 		adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
4455 		adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
4456 		break;
4457 	default:
4458 		adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
4459 		adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
4460 		break;
4461 	}
4462 
4463 	/* Configure DCB traffic classes */
4464 	for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4465 		tc = &adapter->dcb_cfg.tc_config[j];
4466 		tc->path[DCB_TX_CONFIG].bwg_id = 0;
4467 		tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4468 		tc->path[DCB_RX_CONFIG].bwg_id = 0;
4469 		tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4470 		tc->dcb_pfc = pfc_disabled;
4471 	}
4472 
4473 	/* Initialize default user to priority mapping, UPx->TC0 */
4474 	tc = &adapter->dcb_cfg.tc_config[0];
4475 	tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
4476 	tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
4477 
4478 	adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4479 	adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4480 	adapter->dcb_cfg.pfc_mode_enable = false;
4481 	adapter->dcb_set_bitmap = 0x00;
4482 	adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
4483 	memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
4484 	       sizeof(adapter->temp_dcb_cfg));
4485 
4486 #endif
4487 
4488 	/* default flow control settings */
4489 	hw->fc.requested_mode = ixgbe_fc_full;
4490 	hw->fc.current_mode = ixgbe_fc_full;	/* init for ethtool output */
4491 	ixgbe_pbthresh_setup(adapter);
4492 	hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4493 	hw->fc.send_xon = true;
4494 	hw->fc.disable_fc_autoneg = false;
4495 
4496 #ifdef CONFIG_PCI_IOV
4497 	/* assign number of SR-IOV VFs */
4498 	if (hw->mac.type != ixgbe_mac_82598EB)
4499 		adapter->num_vfs = (max_vfs > 63) ? 0 : max_vfs;
4500 
4501 #endif
4502 	/* enable itr by default in dynamic mode */
4503 	adapter->rx_itr_setting = 1;
4504 	adapter->tx_itr_setting = 1;
4505 
4506 	/* set default ring sizes */
4507 	adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4508 	adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4509 
4510 	/* set default work limits */
4511 	adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
4512 
4513 	/* initialize eeprom parameters */
4514 	if (ixgbe_init_eeprom_params_generic(hw)) {
4515 		e_dev_err("EEPROM initialization failed\n");
4516 		return -EIO;
4517 	}
4518 
4519 	set_bit(__IXGBE_DOWN, &adapter->state);
4520 
4521 	return 0;
4522 }
4523 
4524 /**
4525  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4526  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
4527  *
4528  * Return 0 on success, negative on failure
4529  **/
4530 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
4531 {
4532 	struct device *dev = tx_ring->dev;
4533 	int orig_node = dev_to_node(dev);
4534 	int numa_node = -1;
4535 	int size;
4536 
4537 	size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4538 
4539 	if (tx_ring->q_vector)
4540 		numa_node = tx_ring->q_vector->numa_node;
4541 
4542 	tx_ring->tx_buffer_info = vzalloc_node(size, numa_node);
4543 	if (!tx_ring->tx_buffer_info)
4544 		tx_ring->tx_buffer_info = vzalloc(size);
4545 	if (!tx_ring->tx_buffer_info)
4546 		goto err;
4547 
4548 	/* round up to nearest 4K */
4549 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4550 	tx_ring->size = ALIGN(tx_ring->size, 4096);
4551 
4552 	set_dev_node(dev, numa_node);
4553 	tx_ring->desc = dma_alloc_coherent(dev,
4554 					   tx_ring->size,
4555 					   &tx_ring->dma,
4556 					   GFP_KERNEL);
4557 	set_dev_node(dev, orig_node);
4558 	if (!tx_ring->desc)
4559 		tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4560 						   &tx_ring->dma, GFP_KERNEL);
4561 	if (!tx_ring->desc)
4562 		goto err;
4563 
4564 	tx_ring->next_to_use = 0;
4565 	tx_ring->next_to_clean = 0;
4566 	return 0;
4567 
4568 err:
4569 	vfree(tx_ring->tx_buffer_info);
4570 	tx_ring->tx_buffer_info = NULL;
4571 	dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4572 	return -ENOMEM;
4573 }
4574 
4575 /**
4576  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4577  * @adapter: board private structure
4578  *
4579  * If this function returns with an error, then it's possible one or
4580  * more of the rings is populated (while the rest are not).  It is the
4581  * callers duty to clean those orphaned rings.
4582  *
4583  * Return 0 on success, negative on failure
4584  **/
4585 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4586 {
4587 	int i, err = 0;
4588 
4589 	for (i = 0; i < adapter->num_tx_queues; i++) {
4590 		err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
4591 		if (!err)
4592 			continue;
4593 
4594 		e_err(probe, "Allocation for Tx Queue %u failed\n", i);
4595 		goto err_setup_tx;
4596 	}
4597 
4598 	return 0;
4599 err_setup_tx:
4600 	/* rewind the index freeing the rings as we go */
4601 	while (i--)
4602 		ixgbe_free_tx_resources(adapter->tx_ring[i]);
4603 	return err;
4604 }
4605 
4606 /**
4607  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4608  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
4609  *
4610  * Returns 0 on success, negative on failure
4611  **/
4612 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
4613 {
4614 	struct device *dev = rx_ring->dev;
4615 	int orig_node = dev_to_node(dev);
4616 	int numa_node = -1;
4617 	int size;
4618 
4619 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4620 
4621 	if (rx_ring->q_vector)
4622 		numa_node = rx_ring->q_vector->numa_node;
4623 
4624 	rx_ring->rx_buffer_info = vzalloc_node(size, numa_node);
4625 	if (!rx_ring->rx_buffer_info)
4626 		rx_ring->rx_buffer_info = vzalloc(size);
4627 	if (!rx_ring->rx_buffer_info)
4628 		goto err;
4629 
4630 	/* Round up to nearest 4K */
4631 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4632 	rx_ring->size = ALIGN(rx_ring->size, 4096);
4633 
4634 	set_dev_node(dev, numa_node);
4635 	rx_ring->desc = dma_alloc_coherent(dev,
4636 					   rx_ring->size,
4637 					   &rx_ring->dma,
4638 					   GFP_KERNEL);
4639 	set_dev_node(dev, orig_node);
4640 	if (!rx_ring->desc)
4641 		rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
4642 						   &rx_ring->dma, GFP_KERNEL);
4643 	if (!rx_ring->desc)
4644 		goto err;
4645 
4646 	rx_ring->next_to_clean = 0;
4647 	rx_ring->next_to_use = 0;
4648 
4649 	ixgbe_init_rx_page_offset(rx_ring);
4650 
4651 	return 0;
4652 err:
4653 	vfree(rx_ring->rx_buffer_info);
4654 	rx_ring->rx_buffer_info = NULL;
4655 	dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
4656 	return -ENOMEM;
4657 }
4658 
4659 /**
4660  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4661  * @adapter: board private structure
4662  *
4663  * If this function returns with an error, then it's possible one or
4664  * more of the rings is populated (while the rest are not).  It is the
4665  * callers duty to clean those orphaned rings.
4666  *
4667  * Return 0 on success, negative on failure
4668  **/
4669 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4670 {
4671 	int i, err = 0;
4672 
4673 	for (i = 0; i < adapter->num_rx_queues; i++) {
4674 		err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
4675 		if (!err)
4676 			continue;
4677 
4678 		e_err(probe, "Allocation for Rx Queue %u failed\n", i);
4679 		goto err_setup_rx;
4680 	}
4681 
4682 #ifdef IXGBE_FCOE
4683 	err = ixgbe_setup_fcoe_ddp_resources(adapter);
4684 	if (!err)
4685 #endif
4686 		return 0;
4687 err_setup_rx:
4688 	/* rewind the index freeing the rings as we go */
4689 	while (i--)
4690 		ixgbe_free_rx_resources(adapter->rx_ring[i]);
4691 	return err;
4692 }
4693 
4694 /**
4695  * ixgbe_free_tx_resources - Free Tx Resources per Queue
4696  * @tx_ring: Tx descriptor ring for a specific queue
4697  *
4698  * Free all transmit software resources
4699  **/
4700 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
4701 {
4702 	ixgbe_clean_tx_ring(tx_ring);
4703 
4704 	vfree(tx_ring->tx_buffer_info);
4705 	tx_ring->tx_buffer_info = NULL;
4706 
4707 	/* if not set, then don't free */
4708 	if (!tx_ring->desc)
4709 		return;
4710 
4711 	dma_free_coherent(tx_ring->dev, tx_ring->size,
4712 			  tx_ring->desc, tx_ring->dma);
4713 
4714 	tx_ring->desc = NULL;
4715 }
4716 
4717 /**
4718  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4719  * @adapter: board private structure
4720  *
4721  * Free all transmit software resources
4722  **/
4723 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4724 {
4725 	int i;
4726 
4727 	for (i = 0; i < adapter->num_tx_queues; i++)
4728 		if (adapter->tx_ring[i]->desc)
4729 			ixgbe_free_tx_resources(adapter->tx_ring[i]);
4730 }
4731 
4732 /**
4733  * ixgbe_free_rx_resources - Free Rx Resources
4734  * @rx_ring: ring to clean the resources from
4735  *
4736  * Free all receive software resources
4737  **/
4738 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
4739 {
4740 	ixgbe_clean_rx_ring(rx_ring);
4741 
4742 	vfree(rx_ring->rx_buffer_info);
4743 	rx_ring->rx_buffer_info = NULL;
4744 
4745 	/* if not set, then don't free */
4746 	if (!rx_ring->desc)
4747 		return;
4748 
4749 	dma_free_coherent(rx_ring->dev, rx_ring->size,
4750 			  rx_ring->desc, rx_ring->dma);
4751 
4752 	rx_ring->desc = NULL;
4753 }
4754 
4755 /**
4756  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4757  * @adapter: board private structure
4758  *
4759  * Free all receive software resources
4760  **/
4761 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4762 {
4763 	int i;
4764 
4765 #ifdef IXGBE_FCOE
4766 	ixgbe_free_fcoe_ddp_resources(adapter);
4767 
4768 #endif
4769 	for (i = 0; i < adapter->num_rx_queues; i++)
4770 		if (adapter->rx_ring[i]->desc)
4771 			ixgbe_free_rx_resources(adapter->rx_ring[i]);
4772 }
4773 
4774 /**
4775  * ixgbe_change_mtu - Change the Maximum Transfer Unit
4776  * @netdev: network interface device structure
4777  * @new_mtu: new value for maximum frame size
4778  *
4779  * Returns 0 on success, negative on failure
4780  **/
4781 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4782 {
4783 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4784 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4785 
4786 	/* MTU < 68 is an error and causes problems on some kernels */
4787 	if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4788 		return -EINVAL;
4789 
4790 	/*
4791 	 * For 82599EB we cannot allow PF to change MTU greater than 1500
4792 	 * in SR-IOV mode as it may cause buffer overruns in guest VFs that
4793 	 * don't allocate and chain buffers correctly.
4794 	 */
4795 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
4796 	    (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
4797 	    (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
4798 			return -EINVAL;
4799 
4800 	e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4801 
4802 	/* must set new MTU before calling down or up */
4803 	netdev->mtu = new_mtu;
4804 
4805 	if (netif_running(netdev))
4806 		ixgbe_reinit_locked(adapter);
4807 
4808 	return 0;
4809 }
4810 
4811 /**
4812  * ixgbe_open - Called when a network interface is made active
4813  * @netdev: network interface device structure
4814  *
4815  * Returns 0 on success, negative value on failure
4816  *
4817  * The open entry point is called when a network interface is made
4818  * active by the system (IFF_UP).  At this point all resources needed
4819  * for transmit and receive operations are allocated, the interrupt
4820  * handler is registered with the OS, the watchdog timer is started,
4821  * and the stack is notified that the interface is ready.
4822  **/
4823 static int ixgbe_open(struct net_device *netdev)
4824 {
4825 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4826 	int err;
4827 
4828 	/* disallow open during test */
4829 	if (test_bit(__IXGBE_TESTING, &adapter->state))
4830 		return -EBUSY;
4831 
4832 	netif_carrier_off(netdev);
4833 
4834 	/* allocate transmit descriptors */
4835 	err = ixgbe_setup_all_tx_resources(adapter);
4836 	if (err)
4837 		goto err_setup_tx;
4838 
4839 	/* allocate receive descriptors */
4840 	err = ixgbe_setup_all_rx_resources(adapter);
4841 	if (err)
4842 		goto err_setup_rx;
4843 
4844 	ixgbe_configure(adapter);
4845 
4846 	err = ixgbe_request_irq(adapter);
4847 	if (err)
4848 		goto err_req_irq;
4849 
4850 	/* Notify the stack of the actual queue counts. */
4851 	err = netif_set_real_num_tx_queues(netdev,
4852 					   adapter->num_rx_pools > 1 ? 1 :
4853 					   adapter->num_tx_queues);
4854 	if (err)
4855 		goto err_set_queues;
4856 
4857 
4858 	err = netif_set_real_num_rx_queues(netdev,
4859 					   adapter->num_rx_pools > 1 ? 1 :
4860 					   adapter->num_rx_queues);
4861 	if (err)
4862 		goto err_set_queues;
4863 
4864 	ixgbe_up_complete(adapter);
4865 
4866 	return 0;
4867 
4868 err_set_queues:
4869 	ixgbe_free_irq(adapter);
4870 err_req_irq:
4871 	ixgbe_free_all_rx_resources(adapter);
4872 err_setup_rx:
4873 	ixgbe_free_all_tx_resources(adapter);
4874 err_setup_tx:
4875 	ixgbe_reset(adapter);
4876 
4877 	return err;
4878 }
4879 
4880 /**
4881  * ixgbe_close - Disables a network interface
4882  * @netdev: network interface device structure
4883  *
4884  * Returns 0, this is not allowed to fail
4885  *
4886  * The close entry point is called when an interface is de-activated
4887  * by the OS.  The hardware is still under the drivers control, but
4888  * needs to be disabled.  A global MAC reset is issued to stop the
4889  * hardware, and all transmit and receive resources are freed.
4890  **/
4891 static int ixgbe_close(struct net_device *netdev)
4892 {
4893 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4894 
4895 	ixgbe_down(adapter);
4896 	ixgbe_free_irq(adapter);
4897 
4898 	ixgbe_fdir_filter_exit(adapter);
4899 
4900 	ixgbe_free_all_tx_resources(adapter);
4901 	ixgbe_free_all_rx_resources(adapter);
4902 
4903 	ixgbe_release_hw_control(adapter);
4904 
4905 	return 0;
4906 }
4907 
4908 #ifdef CONFIG_PM
4909 static int ixgbe_resume(struct pci_dev *pdev)
4910 {
4911 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
4912 	struct net_device *netdev = adapter->netdev;
4913 	u32 err;
4914 
4915 	pci_set_power_state(pdev, PCI_D0);
4916 	pci_restore_state(pdev);
4917 	/*
4918 	 * pci_restore_state clears dev->state_saved so call
4919 	 * pci_save_state to restore it.
4920 	 */
4921 	pci_save_state(pdev);
4922 
4923 	err = pci_enable_device_mem(pdev);
4924 	if (err) {
4925 		e_dev_err("Cannot enable PCI device from suspend\n");
4926 		return err;
4927 	}
4928 	pci_set_master(pdev);
4929 
4930 	pci_wake_from_d3(pdev, false);
4931 
4932 	ixgbe_reset(adapter);
4933 
4934 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4935 
4936 	rtnl_lock();
4937 	err = ixgbe_init_interrupt_scheme(adapter);
4938 	if (!err && netif_running(netdev))
4939 		err = ixgbe_open(netdev);
4940 
4941 	rtnl_unlock();
4942 
4943 	if (err)
4944 		return err;
4945 
4946 	netif_device_attach(netdev);
4947 
4948 	return 0;
4949 }
4950 #endif /* CONFIG_PM */
4951 
4952 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
4953 {
4954 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
4955 	struct net_device *netdev = adapter->netdev;
4956 	struct ixgbe_hw *hw = &adapter->hw;
4957 	u32 ctrl, fctrl;
4958 	u32 wufc = adapter->wol;
4959 #ifdef CONFIG_PM
4960 	int retval = 0;
4961 #endif
4962 
4963 	netif_device_detach(netdev);
4964 
4965 	if (netif_running(netdev)) {
4966 		rtnl_lock();
4967 		ixgbe_down(adapter);
4968 		ixgbe_free_irq(adapter);
4969 		ixgbe_free_all_tx_resources(adapter);
4970 		ixgbe_free_all_rx_resources(adapter);
4971 		rtnl_unlock();
4972 	}
4973 
4974 	ixgbe_clear_interrupt_scheme(adapter);
4975 
4976 #ifdef CONFIG_PM
4977 	retval = pci_save_state(pdev);
4978 	if (retval)
4979 		return retval;
4980 
4981 #endif
4982 	if (wufc) {
4983 		ixgbe_set_rx_mode(netdev);
4984 
4985 		/*
4986 		 * enable the optics for both mult-speed fiber and
4987 		 * 82599 SFP+ fiber as we can WoL.
4988 		 */
4989 		if (hw->mac.ops.enable_tx_laser &&
4990 		    (hw->phy.multispeed_fiber ||
4991 		    (hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber &&
4992 		     hw->mac.type == ixgbe_mac_82599EB)))
4993 			hw->mac.ops.enable_tx_laser(hw);
4994 
4995 		/* turn on all-multi mode if wake on multicast is enabled */
4996 		if (wufc & IXGBE_WUFC_MC) {
4997 			fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4998 			fctrl |= IXGBE_FCTRL_MPE;
4999 			IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5000 		}
5001 
5002 		ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5003 		ctrl |= IXGBE_CTRL_GIO_DIS;
5004 		IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5005 
5006 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5007 	} else {
5008 		IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5009 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5010 	}
5011 
5012 	switch (hw->mac.type) {
5013 	case ixgbe_mac_82598EB:
5014 		pci_wake_from_d3(pdev, false);
5015 		break;
5016 	case ixgbe_mac_82599EB:
5017 	case ixgbe_mac_X540:
5018 		pci_wake_from_d3(pdev, !!wufc);
5019 		break;
5020 	default:
5021 		break;
5022 	}
5023 
5024 	*enable_wake = !!wufc;
5025 
5026 	ixgbe_release_hw_control(adapter);
5027 
5028 	pci_disable_device(pdev);
5029 
5030 	return 0;
5031 }
5032 
5033 #ifdef CONFIG_PM
5034 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5035 {
5036 	int retval;
5037 	bool wake;
5038 
5039 	retval = __ixgbe_shutdown(pdev, &wake);
5040 	if (retval)
5041 		return retval;
5042 
5043 	if (wake) {
5044 		pci_prepare_to_sleep(pdev);
5045 	} else {
5046 		pci_wake_from_d3(pdev, false);
5047 		pci_set_power_state(pdev, PCI_D3hot);
5048 	}
5049 
5050 	return 0;
5051 }
5052 #endif /* CONFIG_PM */
5053 
5054 static void ixgbe_shutdown(struct pci_dev *pdev)
5055 {
5056 	bool wake;
5057 
5058 	__ixgbe_shutdown(pdev, &wake);
5059 
5060 	if (system_state == SYSTEM_POWER_OFF) {
5061 		pci_wake_from_d3(pdev, wake);
5062 		pci_set_power_state(pdev, PCI_D3hot);
5063 	}
5064 }
5065 
5066 /**
5067  * ixgbe_update_stats - Update the board statistics counters.
5068  * @adapter: board private structure
5069  **/
5070 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5071 {
5072 	struct net_device *netdev = adapter->netdev;
5073 	struct ixgbe_hw *hw = &adapter->hw;
5074 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
5075 	u64 total_mpc = 0;
5076 	u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5077 	u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5078 	u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5079 	u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
5080 
5081 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5082 	    test_bit(__IXGBE_RESETTING, &adapter->state))
5083 		return;
5084 
5085 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5086 		u64 rsc_count = 0;
5087 		u64 rsc_flush = 0;
5088 		for (i = 0; i < adapter->num_rx_queues; i++) {
5089 			rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5090 			rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5091 		}
5092 		adapter->rsc_total_count = rsc_count;
5093 		adapter->rsc_total_flush = rsc_flush;
5094 	}
5095 
5096 	for (i = 0; i < adapter->num_rx_queues; i++) {
5097 		struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5098 		non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5099 		alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5100 		alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5101 		hw_csum_rx_error += rx_ring->rx_stats.csum_err;
5102 		bytes += rx_ring->stats.bytes;
5103 		packets += rx_ring->stats.packets;
5104 	}
5105 	adapter->non_eop_descs = non_eop_descs;
5106 	adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5107 	adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5108 	adapter->hw_csum_rx_error = hw_csum_rx_error;
5109 	netdev->stats.rx_bytes = bytes;
5110 	netdev->stats.rx_packets = packets;
5111 
5112 	bytes = 0;
5113 	packets = 0;
5114 	/* gather some stats to the adapter struct that are per queue */
5115 	for (i = 0; i < adapter->num_tx_queues; i++) {
5116 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5117 		restart_queue += tx_ring->tx_stats.restart_queue;
5118 		tx_busy += tx_ring->tx_stats.tx_busy;
5119 		bytes += tx_ring->stats.bytes;
5120 		packets += tx_ring->stats.packets;
5121 	}
5122 	adapter->restart_queue = restart_queue;
5123 	adapter->tx_busy = tx_busy;
5124 	netdev->stats.tx_bytes = bytes;
5125 	netdev->stats.tx_packets = packets;
5126 
5127 	hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5128 
5129 	/* 8 register reads */
5130 	for (i = 0; i < 8; i++) {
5131 		/* for packet buffers not used, the register should read 0 */
5132 		mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5133 		missed_rx += mpc;
5134 		hwstats->mpc[i] += mpc;
5135 		total_mpc += hwstats->mpc[i];
5136 		hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5137 		hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5138 		switch (hw->mac.type) {
5139 		case ixgbe_mac_82598EB:
5140 			hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5141 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5142 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5143 			hwstats->pxonrxc[i] +=
5144 				IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5145 			break;
5146 		case ixgbe_mac_82599EB:
5147 		case ixgbe_mac_X540:
5148 			hwstats->pxonrxc[i] +=
5149 				IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5150 			break;
5151 		default:
5152 			break;
5153 		}
5154 	}
5155 
5156 	/*16 register reads */
5157 	for (i = 0; i < 16; i++) {
5158 		hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5159 		hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5160 		if ((hw->mac.type == ixgbe_mac_82599EB) ||
5161 		    (hw->mac.type == ixgbe_mac_X540)) {
5162 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
5163 			IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
5164 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
5165 			IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
5166 		}
5167 	}
5168 
5169 	hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5170 	/* work around hardware counting issue */
5171 	hwstats->gprc -= missed_rx;
5172 
5173 	ixgbe_update_xoff_received(adapter);
5174 
5175 	/* 82598 hardware only has a 32 bit counter in the high register */
5176 	switch (hw->mac.type) {
5177 	case ixgbe_mac_82598EB:
5178 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5179 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5180 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5181 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5182 		break;
5183 	case ixgbe_mac_X540:
5184 		/* OS2BMC stats are X540 only*/
5185 		hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5186 		hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5187 		hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5188 		hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5189 	case ixgbe_mac_82599EB:
5190 		for (i = 0; i < 16; i++)
5191 			adapter->hw_rx_no_dma_resources +=
5192 					     IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5193 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5194 		IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5195 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5196 		IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5197 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5198 		IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5199 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5200 		hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5201 		hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5202 #ifdef IXGBE_FCOE
5203 		hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5204 		hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5205 		hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5206 		hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5207 		hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5208 		hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5209 		/* Add up per cpu counters for total ddp aloc fail */
5210 		if (adapter->fcoe.ddp_pool) {
5211 			struct ixgbe_fcoe *fcoe = &adapter->fcoe;
5212 			struct ixgbe_fcoe_ddp_pool *ddp_pool;
5213 			unsigned int cpu;
5214 			u64 noddp = 0, noddp_ext_buff = 0;
5215 			for_each_possible_cpu(cpu) {
5216 				ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
5217 				noddp += ddp_pool->noddp;
5218 				noddp_ext_buff += ddp_pool->noddp_ext_buff;
5219 			}
5220 			hwstats->fcoe_noddp = noddp;
5221 			hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
5222 		}
5223 #endif /* IXGBE_FCOE */
5224 		break;
5225 	default:
5226 		break;
5227 	}
5228 	bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5229 	hwstats->bprc += bprc;
5230 	hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5231 	if (hw->mac.type == ixgbe_mac_82598EB)
5232 		hwstats->mprc -= bprc;
5233 	hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5234 	hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5235 	hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5236 	hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5237 	hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5238 	hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5239 	hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5240 	hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5241 	lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5242 	hwstats->lxontxc += lxon;
5243 	lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5244 	hwstats->lxofftxc += lxoff;
5245 	hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5246 	hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5247 	/*
5248 	 * 82598 errata - tx of flow control packets is included in tx counters
5249 	 */
5250 	xon_off_tot = lxon + lxoff;
5251 	hwstats->gptc -= xon_off_tot;
5252 	hwstats->mptc -= xon_off_tot;
5253 	hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5254 	hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5255 	hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5256 	hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5257 	hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5258 	hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5259 	hwstats->ptc64 -= xon_off_tot;
5260 	hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5261 	hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5262 	hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5263 	hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5264 	hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5265 	hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5266 
5267 	/* Fill out the OS statistics structure */
5268 	netdev->stats.multicast = hwstats->mprc;
5269 
5270 	/* Rx Errors */
5271 	netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5272 	netdev->stats.rx_dropped = 0;
5273 	netdev->stats.rx_length_errors = hwstats->rlec;
5274 	netdev->stats.rx_crc_errors = hwstats->crcerrs;
5275 	netdev->stats.rx_missed_errors = total_mpc;
5276 }
5277 
5278 /**
5279  * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5280  * @adapter: pointer to the device adapter structure
5281  **/
5282 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5283 {
5284 	struct ixgbe_hw *hw = &adapter->hw;
5285 	int i;
5286 
5287 	if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5288 		return;
5289 
5290 	adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5291 
5292 	/* if interface is down do nothing */
5293 	if (test_bit(__IXGBE_DOWN, &adapter->state))
5294 		return;
5295 
5296 	/* do nothing if we are not using signature filters */
5297 	if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5298 		return;
5299 
5300 	adapter->fdir_overflow++;
5301 
5302 	if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5303 		for (i = 0; i < adapter->num_tx_queues; i++)
5304 			set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5305 			        &(adapter->tx_ring[i]->state));
5306 		/* re-enable flow director interrupts */
5307 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5308 	} else {
5309 		e_err(probe, "failed to finish FDIR re-initialization, "
5310 		      "ignored adding FDIR ATR filters\n");
5311 	}
5312 }
5313 
5314 /**
5315  * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5316  * @adapter: pointer to the device adapter structure
5317  *
5318  * This function serves two purposes.  First it strobes the interrupt lines
5319  * in order to make certain interrupts are occurring.  Secondly it sets the
5320  * bits needed to check for TX hangs.  As a result we should immediately
5321  * determine if a hang has occurred.
5322  */
5323 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5324 {
5325 	struct ixgbe_hw *hw = &adapter->hw;
5326 	u64 eics = 0;
5327 	int i;
5328 
5329 	/* If we're down or resetting, just bail */
5330 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5331 	    test_bit(__IXGBE_RESETTING, &adapter->state))
5332 		return;
5333 
5334 	/* Force detection of hung controller */
5335 	if (netif_carrier_ok(adapter->netdev)) {
5336 		for (i = 0; i < adapter->num_tx_queues; i++)
5337 			set_check_for_tx_hang(adapter->tx_ring[i]);
5338 	}
5339 
5340 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5341 		/*
5342 		 * for legacy and MSI interrupts don't set any bits
5343 		 * that are enabled for EIAM, because this operation
5344 		 * would set *both* EIMS and EICS for any bit in EIAM
5345 		 */
5346 		IXGBE_WRITE_REG(hw, IXGBE_EICS,
5347 			(IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5348 	} else {
5349 		/* get one bit for every active tx/rx interrupt vector */
5350 		for (i = 0; i < adapter->num_q_vectors; i++) {
5351 			struct ixgbe_q_vector *qv = adapter->q_vector[i];
5352 			if (qv->rx.ring || qv->tx.ring)
5353 				eics |= ((u64)1 << i);
5354 		}
5355 	}
5356 
5357 	/* Cause software interrupt to ensure rings are cleaned */
5358 	ixgbe_irq_rearm_queues(adapter, eics);
5359 
5360 }
5361 
5362 /**
5363  * ixgbe_watchdog_update_link - update the link status
5364  * @adapter: pointer to the device adapter structure
5365  * @link_speed: pointer to a u32 to store the link_speed
5366  **/
5367 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
5368 {
5369 	struct ixgbe_hw *hw = &adapter->hw;
5370 	u32 link_speed = adapter->link_speed;
5371 	bool link_up = adapter->link_up;
5372 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
5373 
5374 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5375 		return;
5376 
5377 	if (hw->mac.ops.check_link) {
5378 		hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5379 	} else {
5380 		/* always assume link is up, if no check link function */
5381 		link_speed = IXGBE_LINK_SPEED_10GB_FULL;
5382 		link_up = true;
5383 	}
5384 
5385 	if (adapter->ixgbe_ieee_pfc)
5386 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
5387 
5388 	if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
5389 		hw->mac.ops.fc_enable(hw);
5390 		ixgbe_set_rx_drop_en(adapter);
5391 	}
5392 
5393 	if (link_up ||
5394 	    time_after(jiffies, (adapter->link_check_timeout +
5395 				 IXGBE_TRY_LINK_TIMEOUT))) {
5396 		adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5397 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5398 		IXGBE_WRITE_FLUSH(hw);
5399 	}
5400 
5401 	adapter->link_up = link_up;
5402 	adapter->link_speed = link_speed;
5403 }
5404 
5405 /**
5406  * ixgbe_watchdog_link_is_up - update netif_carrier status and
5407  *                             print link up message
5408  * @adapter: pointer to the device adapter structure
5409  **/
5410 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
5411 {
5412 	struct net_device *netdev = adapter->netdev;
5413 	struct ixgbe_hw *hw = &adapter->hw;
5414 	u32 link_speed = adapter->link_speed;
5415 	bool flow_rx, flow_tx;
5416 
5417 	/* only continue if link was previously down */
5418 	if (netif_carrier_ok(netdev))
5419 		return;
5420 
5421 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
5422 
5423 	switch (hw->mac.type) {
5424 	case ixgbe_mac_82598EB: {
5425 		u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5426 		u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5427 		flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5428 		flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5429 	}
5430 		break;
5431 	case ixgbe_mac_X540:
5432 	case ixgbe_mac_82599EB: {
5433 		u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5434 		u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5435 		flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5436 		flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5437 	}
5438 		break;
5439 	default:
5440 		flow_tx = false;
5441 		flow_rx = false;
5442 		break;
5443 	}
5444 
5445 #ifdef CONFIG_IXGBE_PTP
5446 	ixgbe_ptp_start_cyclecounter(adapter);
5447 #endif
5448 
5449 	e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5450 	       (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5451 	       "10 Gbps" :
5452 	       (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5453 	       "1 Gbps" :
5454 	       (link_speed == IXGBE_LINK_SPEED_100_FULL ?
5455 	       "100 Mbps" :
5456 	       "unknown speed"))),
5457 	       ((flow_rx && flow_tx) ? "RX/TX" :
5458 	       (flow_rx ? "RX" :
5459 	       (flow_tx ? "TX" : "None"))));
5460 
5461 	netif_carrier_on(netdev);
5462 	ixgbe_check_vf_rate_limit(adapter);
5463 
5464 	/* ping all the active vfs to let them know link has changed */
5465 	ixgbe_ping_all_vfs(adapter);
5466 }
5467 
5468 /**
5469  * ixgbe_watchdog_link_is_down - update netif_carrier status and
5470  *                               print link down message
5471  * @adapter: pointer to the adapter structure
5472  **/
5473 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
5474 {
5475 	struct net_device *netdev = adapter->netdev;
5476 	struct ixgbe_hw *hw = &adapter->hw;
5477 
5478 	adapter->link_up = false;
5479 	adapter->link_speed = 0;
5480 
5481 	/* only continue if link was up previously */
5482 	if (!netif_carrier_ok(netdev))
5483 		return;
5484 
5485 	/* poll for SFP+ cable when link is down */
5486 	if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
5487 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5488 
5489 #ifdef CONFIG_IXGBE_PTP
5490 	ixgbe_ptp_start_cyclecounter(adapter);
5491 #endif
5492 
5493 	e_info(drv, "NIC Link is Down\n");
5494 	netif_carrier_off(netdev);
5495 
5496 	/* ping all the active vfs to let them know link has changed */
5497 	ixgbe_ping_all_vfs(adapter);
5498 }
5499 
5500 /**
5501  * ixgbe_watchdog_flush_tx - flush queues on link down
5502  * @adapter: pointer to the device adapter structure
5503  **/
5504 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
5505 {
5506 	int i;
5507 	int some_tx_pending = 0;
5508 
5509 	if (!netif_carrier_ok(adapter->netdev)) {
5510 		for (i = 0; i < adapter->num_tx_queues; i++) {
5511 			struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5512 			if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5513 				some_tx_pending = 1;
5514 				break;
5515 			}
5516 		}
5517 
5518 		if (some_tx_pending) {
5519 			/* We've lost link, so the controller stops DMA,
5520 			 * but we've got queued Tx work that's never going
5521 			 * to get done, so reset controller to flush Tx.
5522 			 * (Do the reset outside of interrupt context).
5523 			 */
5524 			adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
5525 		}
5526 	}
5527 }
5528 
5529 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
5530 {
5531 	u32 ssvpc;
5532 
5533 	/* Do not perform spoof check for 82598 */
5534 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5535 		return;
5536 
5537 	ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
5538 
5539 	/*
5540 	 * ssvpc register is cleared on read, if zero then no
5541 	 * spoofed packets in the last interval.
5542 	 */
5543 	if (!ssvpc)
5544 		return;
5545 
5546 	e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
5547 }
5548 
5549 /**
5550  * ixgbe_watchdog_subtask - check and bring link up
5551  * @adapter: pointer to the device adapter structure
5552  **/
5553 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
5554 {
5555 	/* if interface is down do nothing */
5556 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5557 	    test_bit(__IXGBE_RESETTING, &adapter->state))
5558 		return;
5559 
5560 	ixgbe_watchdog_update_link(adapter);
5561 
5562 	if (adapter->link_up)
5563 		ixgbe_watchdog_link_is_up(adapter);
5564 	else
5565 		ixgbe_watchdog_link_is_down(adapter);
5566 
5567 	ixgbe_spoof_check(adapter);
5568 	ixgbe_update_stats(adapter);
5569 
5570 	ixgbe_watchdog_flush_tx(adapter);
5571 }
5572 
5573 /**
5574  * ixgbe_sfp_detection_subtask - poll for SFP+ cable
5575  * @adapter: the ixgbe adapter structure
5576  **/
5577 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
5578 {
5579 	struct ixgbe_hw *hw = &adapter->hw;
5580 	s32 err;
5581 
5582 	/* not searching for SFP so there is nothing to do here */
5583 	if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
5584 	    !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5585 		return;
5586 
5587 	/* someone else is in init, wait until next service event */
5588 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5589 		return;
5590 
5591 	err = hw->phy.ops.identify_sfp(hw);
5592 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5593 		goto sfp_out;
5594 
5595 	if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
5596 		/* If no cable is present, then we need to reset
5597 		 * the next time we find a good cable. */
5598 		adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5599 	}
5600 
5601 	/* exit on error */
5602 	if (err)
5603 		goto sfp_out;
5604 
5605 	/* exit if reset not needed */
5606 	if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5607 		goto sfp_out;
5608 
5609 	adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
5610 
5611 	/*
5612 	 * A module may be identified correctly, but the EEPROM may not have
5613 	 * support for that module.  setup_sfp() will fail in that case, so
5614 	 * we should not allow that module to load.
5615 	 */
5616 	if (hw->mac.type == ixgbe_mac_82598EB)
5617 		err = hw->phy.ops.reset(hw);
5618 	else
5619 		err = hw->mac.ops.setup_sfp(hw);
5620 
5621 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5622 		goto sfp_out;
5623 
5624 	adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
5625 	e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5626 
5627 sfp_out:
5628 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5629 
5630 	if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
5631 	    (adapter->netdev->reg_state == NETREG_REGISTERED)) {
5632 		e_dev_err("failed to initialize because an unsupported "
5633 			  "SFP+ module type was detected.\n");
5634 		e_dev_err("Reload the driver after installing a "
5635 			  "supported module.\n");
5636 		unregister_netdev(adapter->netdev);
5637 	}
5638 }
5639 
5640 /**
5641  * ixgbe_sfp_link_config_subtask - set up link SFP after module install
5642  * @adapter: the ixgbe adapter structure
5643  **/
5644 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
5645 {
5646 	struct ixgbe_hw *hw = &adapter->hw;
5647 	u32 autoneg;
5648 	bool negotiation;
5649 
5650 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
5651 		return;
5652 
5653 	/* someone else is in init, wait until next service event */
5654 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5655 		return;
5656 
5657 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5658 
5659 	autoneg = hw->phy.autoneg_advertised;
5660 	if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5661 		hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5662 	if (hw->mac.ops.setup_link)
5663 		hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5664 
5665 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5666 	adapter->link_check_timeout = jiffies;
5667 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5668 }
5669 
5670 #ifdef CONFIG_PCI_IOV
5671 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
5672 {
5673 	int vf;
5674 	struct ixgbe_hw *hw = &adapter->hw;
5675 	struct net_device *netdev = adapter->netdev;
5676 	u32 gpc;
5677 	u32 ciaa, ciad;
5678 
5679 	gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
5680 	if (gpc) /* If incrementing then no need for the check below */
5681 		return;
5682 	/*
5683 	 * Check to see if a bad DMA write target from an errant or
5684 	 * malicious VF has caused a PCIe error.  If so then we can
5685 	 * issue a VFLR to the offending VF(s) and then resume without
5686 	 * requesting a full slot reset.
5687 	 */
5688 
5689 	for (vf = 0; vf < adapter->num_vfs; vf++) {
5690 		ciaa = (vf << 16) | 0x80000000;
5691 		/* 32 bit read so align, we really want status at offset 6 */
5692 		ciaa |= PCI_COMMAND;
5693 		IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5694 		ciad = IXGBE_READ_REG(hw, IXGBE_CIAD_82599);
5695 		ciaa &= 0x7FFFFFFF;
5696 		/* disable debug mode asap after reading data */
5697 		IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5698 		/* Get the upper 16 bits which will be the PCI status reg */
5699 		ciad >>= 16;
5700 		if (ciad & PCI_STATUS_REC_MASTER_ABORT) {
5701 			netdev_err(netdev, "VF %d Hung DMA\n", vf);
5702 			/* Issue VFLR */
5703 			ciaa = (vf << 16) | 0x80000000;
5704 			ciaa |= 0xA8;
5705 			IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5706 			ciad = 0x00008000;  /* VFLR */
5707 			IXGBE_WRITE_REG(hw, IXGBE_CIAD_82599, ciad);
5708 			ciaa &= 0x7FFFFFFF;
5709 			IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5710 		}
5711 	}
5712 }
5713 
5714 #endif
5715 /**
5716  * ixgbe_service_timer - Timer Call-back
5717  * @data: pointer to adapter cast into an unsigned long
5718  **/
5719 static void ixgbe_service_timer(unsigned long data)
5720 {
5721 	struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5722 	unsigned long next_event_offset;
5723 	bool ready = true;
5724 
5725 	/* poll faster when waiting for link */
5726 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
5727 		next_event_offset = HZ / 10;
5728 	else
5729 		next_event_offset = HZ * 2;
5730 
5731 #ifdef CONFIG_PCI_IOV
5732 	/*
5733 	 * don't bother with SR-IOV VF DMA hang check if there are
5734 	 * no VFs or the link is down
5735 	 */
5736 	if (!adapter->num_vfs ||
5737 	    (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5738 		goto normal_timer_service;
5739 
5740 	/* If we have VFs allocated then we must check for DMA hangs */
5741 	ixgbe_check_for_bad_vf(adapter);
5742 	next_event_offset = HZ / 50;
5743 	adapter->timer_event_accumulator++;
5744 
5745 	if (adapter->timer_event_accumulator >= 100)
5746 		adapter->timer_event_accumulator = 0;
5747 	else
5748 		ready = false;
5749 
5750 normal_timer_service:
5751 #endif
5752 	/* Reset the timer */
5753 	mod_timer(&adapter->service_timer, next_event_offset + jiffies);
5754 
5755 	if (ready)
5756 		ixgbe_service_event_schedule(adapter);
5757 }
5758 
5759 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
5760 {
5761 	if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
5762 		return;
5763 
5764 	adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
5765 
5766 	/* If we're already down or resetting, just bail */
5767 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5768 	    test_bit(__IXGBE_RESETTING, &adapter->state))
5769 		return;
5770 
5771 	ixgbe_dump(adapter);
5772 	netdev_err(adapter->netdev, "Reset adapter\n");
5773 	adapter->tx_timeout_count++;
5774 
5775 	ixgbe_reinit_locked(adapter);
5776 }
5777 
5778 /**
5779  * ixgbe_service_task - manages and runs subtasks
5780  * @work: pointer to work_struct containing our data
5781  **/
5782 static void ixgbe_service_task(struct work_struct *work)
5783 {
5784 	struct ixgbe_adapter *adapter = container_of(work,
5785 						     struct ixgbe_adapter,
5786 						     service_task);
5787 
5788 	ixgbe_reset_subtask(adapter);
5789 	ixgbe_sfp_detection_subtask(adapter);
5790 	ixgbe_sfp_link_config_subtask(adapter);
5791 	ixgbe_check_overtemp_subtask(adapter);
5792 	ixgbe_watchdog_subtask(adapter);
5793 	ixgbe_fdir_reinit_subtask(adapter);
5794 	ixgbe_check_hang_subtask(adapter);
5795 #ifdef CONFIG_IXGBE_PTP
5796 	ixgbe_ptp_overflow_check(adapter);
5797 #endif
5798 
5799 	ixgbe_service_event_complete(adapter);
5800 }
5801 
5802 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
5803 		     struct ixgbe_tx_buffer *first,
5804 		     u8 *hdr_len)
5805 {
5806 	struct sk_buff *skb = first->skb;
5807 	u32 vlan_macip_lens, type_tucmd;
5808 	u32 mss_l4len_idx, l4len;
5809 
5810 	if (!skb_is_gso(skb))
5811 		return 0;
5812 
5813 	if (skb_header_cloned(skb)) {
5814 		int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5815 		if (err)
5816 			return err;
5817 	}
5818 
5819 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5820 	type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
5821 
5822 	if (first->protocol == __constant_htons(ETH_P_IP)) {
5823 		struct iphdr *iph = ip_hdr(skb);
5824 		iph->tot_len = 0;
5825 		iph->check = 0;
5826 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5827 							 iph->daddr, 0,
5828 							 IPPROTO_TCP,
5829 							 0);
5830 		type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5831 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5832 				   IXGBE_TX_FLAGS_CSUM |
5833 				   IXGBE_TX_FLAGS_IPV4;
5834 	} else if (skb_is_gso_v6(skb)) {
5835 		ipv6_hdr(skb)->payload_len = 0;
5836 		tcp_hdr(skb)->check =
5837 		    ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5838 				     &ipv6_hdr(skb)->daddr,
5839 				     0, IPPROTO_TCP, 0);
5840 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5841 				   IXGBE_TX_FLAGS_CSUM;
5842 	}
5843 
5844 	/* compute header lengths */
5845 	l4len = tcp_hdrlen(skb);
5846 	*hdr_len = skb_transport_offset(skb) + l4len;
5847 
5848 	/* update gso size and bytecount with header size */
5849 	first->gso_segs = skb_shinfo(skb)->gso_segs;
5850 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
5851 
5852 	/* mss_l4len_id: use 1 as index for TSO */
5853 	mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
5854 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
5855 	mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
5856 
5857 	/* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
5858 	vlan_macip_lens = skb_network_header_len(skb);
5859 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
5860 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
5861 
5862 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
5863 			  mss_l4len_idx);
5864 
5865 	return 1;
5866 }
5867 
5868 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
5869 			  struct ixgbe_tx_buffer *first)
5870 {
5871 	struct sk_buff *skb = first->skb;
5872 	u32 vlan_macip_lens = 0;
5873 	u32 mss_l4len_idx = 0;
5874 	u32 type_tucmd = 0;
5875 
5876 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
5877 		if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
5878 		    !(first->tx_flags & IXGBE_TX_FLAGS_TXSW))
5879 			return;
5880 	} else {
5881 		u8 l4_hdr = 0;
5882 		switch (first->protocol) {
5883 		case __constant_htons(ETH_P_IP):
5884 			vlan_macip_lens |= skb_network_header_len(skb);
5885 			type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5886 			l4_hdr = ip_hdr(skb)->protocol;
5887 			break;
5888 		case __constant_htons(ETH_P_IPV6):
5889 			vlan_macip_lens |= skb_network_header_len(skb);
5890 			l4_hdr = ipv6_hdr(skb)->nexthdr;
5891 			break;
5892 		default:
5893 			if (unlikely(net_ratelimit())) {
5894 				dev_warn(tx_ring->dev,
5895 				 "partial checksum but proto=%x!\n",
5896 				 first->protocol);
5897 			}
5898 			break;
5899 		}
5900 
5901 		switch (l4_hdr) {
5902 		case IPPROTO_TCP:
5903 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5904 			mss_l4len_idx = tcp_hdrlen(skb) <<
5905 					IXGBE_ADVTXD_L4LEN_SHIFT;
5906 			break;
5907 		case IPPROTO_SCTP:
5908 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5909 			mss_l4len_idx = sizeof(struct sctphdr) <<
5910 					IXGBE_ADVTXD_L4LEN_SHIFT;
5911 			break;
5912 		case IPPROTO_UDP:
5913 			mss_l4len_idx = sizeof(struct udphdr) <<
5914 					IXGBE_ADVTXD_L4LEN_SHIFT;
5915 			break;
5916 		default:
5917 			if (unlikely(net_ratelimit())) {
5918 				dev_warn(tx_ring->dev,
5919 				 "partial checksum but l4 proto=%x!\n",
5920 				 l4_hdr);
5921 			}
5922 			break;
5923 		}
5924 
5925 		/* update TX checksum flag */
5926 		first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
5927 	}
5928 
5929 	/* vlan_macip_lens: MACLEN, VLAN tag */
5930 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
5931 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
5932 
5933 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
5934 			  type_tucmd, mss_l4len_idx);
5935 }
5936 
5937 static __le32 ixgbe_tx_cmd_type(u32 tx_flags)
5938 {
5939 	/* set type for advanced descriptor with frame checksum insertion */
5940 	__le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
5941 				      IXGBE_ADVTXD_DCMD_IFCS |
5942 				      IXGBE_ADVTXD_DCMD_DEXT);
5943 
5944 	/* set HW vlan bit if vlan is present */
5945 	if (tx_flags & IXGBE_TX_FLAGS_HW_VLAN)
5946 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
5947 
5948 #ifdef CONFIG_IXGBE_PTP
5949 	if (tx_flags & IXGBE_TX_FLAGS_TSTAMP)
5950 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_MAC_TSTAMP);
5951 #endif
5952 
5953 	/* set segmentation enable bits for TSO/FSO */
5954 #ifdef IXGBE_FCOE
5955 	if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FSO))
5956 #else
5957 	if (tx_flags & IXGBE_TX_FLAGS_TSO)
5958 #endif
5959 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
5960 
5961 	return cmd_type;
5962 }
5963 
5964 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
5965 				   u32 tx_flags, unsigned int paylen)
5966 {
5967 	__le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
5968 
5969 	/* enable L4 checksum for TSO and TX checksum offload */
5970 	if (tx_flags & IXGBE_TX_FLAGS_CSUM)
5971 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
5972 
5973 	/* enble IPv4 checksum for TSO */
5974 	if (tx_flags & IXGBE_TX_FLAGS_IPV4)
5975 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
5976 
5977 	/* use index 1 context for TSO/FSO/FCOE */
5978 #ifdef IXGBE_FCOE
5979 	if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FCOE))
5980 #else
5981 	if (tx_flags & IXGBE_TX_FLAGS_TSO)
5982 #endif
5983 		olinfo_status |= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT);
5984 
5985 	/*
5986 	 * Check Context must be set if Tx switch is enabled, which it
5987 	 * always is for case where virtual functions are running
5988 	 */
5989 #ifdef IXGBE_FCOE
5990 	if (tx_flags & (IXGBE_TX_FLAGS_TXSW | IXGBE_TX_FLAGS_FCOE))
5991 #else
5992 	if (tx_flags & IXGBE_TX_FLAGS_TXSW)
5993 #endif
5994 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
5995 
5996 	tx_desc->read.olinfo_status = olinfo_status;
5997 }
5998 
5999 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
6000 		       IXGBE_TXD_CMD_RS)
6001 
6002 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
6003 			 struct ixgbe_tx_buffer *first,
6004 			 const u8 hdr_len)
6005 {
6006 	dma_addr_t dma;
6007 	struct sk_buff *skb = first->skb;
6008 	struct ixgbe_tx_buffer *tx_buffer;
6009 	union ixgbe_adv_tx_desc *tx_desc;
6010 	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
6011 	unsigned int data_len = skb->data_len;
6012 	unsigned int size = skb_headlen(skb);
6013 	unsigned int paylen = skb->len - hdr_len;
6014 	u32 tx_flags = first->tx_flags;
6015 	__le32 cmd_type;
6016 	u16 i = tx_ring->next_to_use;
6017 
6018 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
6019 
6020 	ixgbe_tx_olinfo_status(tx_desc, tx_flags, paylen);
6021 	cmd_type = ixgbe_tx_cmd_type(tx_flags);
6022 
6023 #ifdef IXGBE_FCOE
6024 	if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6025 		if (data_len < sizeof(struct fcoe_crc_eof)) {
6026 			size -= sizeof(struct fcoe_crc_eof) - data_len;
6027 			data_len = 0;
6028 		} else {
6029 			data_len -= sizeof(struct fcoe_crc_eof);
6030 		}
6031 	}
6032 
6033 #endif
6034 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
6035 	if (dma_mapping_error(tx_ring->dev, dma))
6036 		goto dma_error;
6037 
6038 	/* record length, and DMA address */
6039 	dma_unmap_len_set(first, len, size);
6040 	dma_unmap_addr_set(first, dma, dma);
6041 
6042 	tx_desc->read.buffer_addr = cpu_to_le64(dma);
6043 
6044 	for (;;) {
6045 		while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
6046 			tx_desc->read.cmd_type_len =
6047 				cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
6048 
6049 			i++;
6050 			tx_desc++;
6051 			if (i == tx_ring->count) {
6052 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6053 				i = 0;
6054 			}
6055 
6056 			dma += IXGBE_MAX_DATA_PER_TXD;
6057 			size -= IXGBE_MAX_DATA_PER_TXD;
6058 
6059 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
6060 			tx_desc->read.olinfo_status = 0;
6061 		}
6062 
6063 		if (likely(!data_len))
6064 			break;
6065 
6066 		if (unlikely(skb->no_fcs))
6067 			cmd_type &= ~(cpu_to_le32(IXGBE_ADVTXD_DCMD_IFCS));
6068 		tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
6069 
6070 		i++;
6071 		tx_desc++;
6072 		if (i == tx_ring->count) {
6073 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6074 			i = 0;
6075 		}
6076 
6077 #ifdef IXGBE_FCOE
6078 		size = min_t(unsigned int, data_len, skb_frag_size(frag));
6079 #else
6080 		size = skb_frag_size(frag);
6081 #endif
6082 		data_len -= size;
6083 
6084 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
6085 				       DMA_TO_DEVICE);
6086 		if (dma_mapping_error(tx_ring->dev, dma))
6087 			goto dma_error;
6088 
6089 		tx_buffer = &tx_ring->tx_buffer_info[i];
6090 		dma_unmap_len_set(tx_buffer, len, size);
6091 		dma_unmap_addr_set(tx_buffer, dma, dma);
6092 
6093 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
6094 		tx_desc->read.olinfo_status = 0;
6095 
6096 		frag++;
6097 	}
6098 
6099 	/* write last descriptor with RS and EOP bits */
6100 	cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
6101 	tx_desc->read.cmd_type_len = cmd_type;
6102 
6103 	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
6104 
6105 	/* set the timestamp */
6106 	first->time_stamp = jiffies;
6107 
6108 	/*
6109 	 * Force memory writes to complete before letting h/w know there
6110 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
6111 	 * memory model archs, such as IA-64).
6112 	 *
6113 	 * We also need this memory barrier to make certain all of the
6114 	 * status bits have been updated before next_to_watch is written.
6115 	 */
6116 	wmb();
6117 
6118 	/* set next_to_watch value indicating a packet is present */
6119 	first->next_to_watch = tx_desc;
6120 
6121 	i++;
6122 	if (i == tx_ring->count)
6123 		i = 0;
6124 
6125 	tx_ring->next_to_use = i;
6126 
6127 	/* notify HW of packet */
6128 	writel(i, tx_ring->tail);
6129 
6130 	return;
6131 dma_error:
6132 	dev_err(tx_ring->dev, "TX DMA map failed\n");
6133 
6134 	/* clear dma mappings for failed tx_buffer_info map */
6135 	for (;;) {
6136 		tx_buffer = &tx_ring->tx_buffer_info[i];
6137 		ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
6138 		if (tx_buffer == first)
6139 			break;
6140 		if (i == 0)
6141 			i = tx_ring->count;
6142 		i--;
6143 	}
6144 
6145 	tx_ring->next_to_use = i;
6146 }
6147 
6148 static void ixgbe_atr(struct ixgbe_ring *ring,
6149 		      struct ixgbe_tx_buffer *first)
6150 {
6151 	struct ixgbe_q_vector *q_vector = ring->q_vector;
6152 	union ixgbe_atr_hash_dword input = { .dword = 0 };
6153 	union ixgbe_atr_hash_dword common = { .dword = 0 };
6154 	union {
6155 		unsigned char *network;
6156 		struct iphdr *ipv4;
6157 		struct ipv6hdr *ipv6;
6158 	} hdr;
6159 	struct tcphdr *th;
6160 	__be16 vlan_id;
6161 
6162 	/* if ring doesn't have a interrupt vector, cannot perform ATR */
6163 	if (!q_vector)
6164 		return;
6165 
6166 	/* do nothing if sampling is disabled */
6167 	if (!ring->atr_sample_rate)
6168 		return;
6169 
6170 	ring->atr_count++;
6171 
6172 	/* snag network header to get L4 type and address */
6173 	hdr.network = skb_network_header(first->skb);
6174 
6175 	/* Currently only IPv4/IPv6 with TCP is supported */
6176 	if ((first->protocol != __constant_htons(ETH_P_IPV6) ||
6177 	     hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6178 	    (first->protocol != __constant_htons(ETH_P_IP) ||
6179 	     hdr.ipv4->protocol != IPPROTO_TCP))
6180 		return;
6181 
6182 	th = tcp_hdr(first->skb);
6183 
6184 	/* skip this packet since it is invalid or the socket is closing */
6185 	if (!th || th->fin)
6186 		return;
6187 
6188 	/* sample on all syn packets or once every atr sample count */
6189 	if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6190 		return;
6191 
6192 	/* reset sample count */
6193 	ring->atr_count = 0;
6194 
6195 	vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6196 
6197 	/*
6198 	 * src and dst are inverted, think how the receiver sees them
6199 	 *
6200 	 * The input is broken into two sections, a non-compressed section
6201 	 * containing vm_pool, vlan_id, and flow_type.  The rest of the data
6202 	 * is XORed together and stored in the compressed dword.
6203 	 */
6204 	input.formatted.vlan_id = vlan_id;
6205 
6206 	/*
6207 	 * since src port and flex bytes occupy the same word XOR them together
6208 	 * and write the value to source port portion of compressed dword
6209 	 */
6210 	if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6211 		common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6212 	else
6213 		common.port.src ^= th->dest ^ first->protocol;
6214 	common.port.dst ^= th->source;
6215 
6216 	if (first->protocol == __constant_htons(ETH_P_IP)) {
6217 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6218 		common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6219 	} else {
6220 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6221 		common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6222 			     hdr.ipv6->saddr.s6_addr32[1] ^
6223 			     hdr.ipv6->saddr.s6_addr32[2] ^
6224 			     hdr.ipv6->saddr.s6_addr32[3] ^
6225 			     hdr.ipv6->daddr.s6_addr32[0] ^
6226 			     hdr.ipv6->daddr.s6_addr32[1] ^
6227 			     hdr.ipv6->daddr.s6_addr32[2] ^
6228 			     hdr.ipv6->daddr.s6_addr32[3];
6229 	}
6230 
6231 	/* This assumes the Rx queue and Tx queue are bound to the same CPU */
6232 	ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6233 					      input, common, ring->queue_index);
6234 }
6235 
6236 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6237 {
6238 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6239 	/* Herbert's original patch had:
6240 	 *  smp_mb__after_netif_stop_queue();
6241 	 * but since that doesn't exist yet, just open code it. */
6242 	smp_mb();
6243 
6244 	/* We need to check again in a case another CPU has just
6245 	 * made room available. */
6246 	if (likely(ixgbe_desc_unused(tx_ring) < size))
6247 		return -EBUSY;
6248 
6249 	/* A reprieve! - use start_queue because it doesn't call schedule */
6250 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6251 	++tx_ring->tx_stats.restart_queue;
6252 	return 0;
6253 }
6254 
6255 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6256 {
6257 	if (likely(ixgbe_desc_unused(tx_ring) >= size))
6258 		return 0;
6259 	return __ixgbe_maybe_stop_tx(tx_ring, size);
6260 }
6261 
6262 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6263 {
6264 	struct ixgbe_adapter *adapter = netdev_priv(dev);
6265 	int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6266 					       smp_processor_id();
6267 #ifdef IXGBE_FCOE
6268 	__be16 protocol = vlan_get_protocol(skb);
6269 
6270 	if (((protocol == htons(ETH_P_FCOE)) ||
6271 	    (protocol == htons(ETH_P_FIP))) &&
6272 	    (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6273 		struct ixgbe_ring_feature *f;
6274 
6275 		f = &adapter->ring_feature[RING_F_FCOE];
6276 
6277 		while (txq >= f->indices)
6278 			txq -= f->indices;
6279 		txq += adapter->ring_feature[RING_F_FCOE].offset;
6280 
6281 		return txq;
6282 	}
6283 #endif
6284 
6285 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6286 		while (unlikely(txq >= dev->real_num_tx_queues))
6287 			txq -= dev->real_num_tx_queues;
6288 		return txq;
6289 	}
6290 
6291 	return skb_tx_hash(dev, skb);
6292 }
6293 
6294 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6295 			  struct ixgbe_adapter *adapter,
6296 			  struct ixgbe_ring *tx_ring)
6297 {
6298 	struct ixgbe_tx_buffer *first;
6299 	int tso;
6300 	u32 tx_flags = 0;
6301 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6302 	unsigned short f;
6303 #endif
6304 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
6305 	__be16 protocol = skb->protocol;
6306 	u8 hdr_len = 0;
6307 
6308 	/*
6309 	 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6310 	 *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
6311 	 *       + 2 desc gap to keep tail from touching head,
6312 	 *       + 1 desc for context descriptor,
6313 	 * otherwise try next time
6314 	 */
6315 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6316 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6317 		count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6318 #else
6319 	count += skb_shinfo(skb)->nr_frags;
6320 #endif
6321 	if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
6322 		tx_ring->tx_stats.tx_busy++;
6323 		return NETDEV_TX_BUSY;
6324 	}
6325 
6326 	/* record the location of the first descriptor for this packet */
6327 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6328 	first->skb = skb;
6329 	first->bytecount = skb->len;
6330 	first->gso_segs = 1;
6331 
6332 	/* if we have a HW VLAN tag being added default to the HW one */
6333 	if (vlan_tx_tag_present(skb)) {
6334 		tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6335 		tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6336 	/* else if it is a SW VLAN check the next protocol and store the tag */
6337 	} else if (protocol == __constant_htons(ETH_P_8021Q)) {
6338 		struct vlan_hdr *vhdr, _vhdr;
6339 		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
6340 		if (!vhdr)
6341 			goto out_drop;
6342 
6343 		protocol = vhdr->h_vlan_encapsulated_proto;
6344 		tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
6345 				  IXGBE_TX_FLAGS_VLAN_SHIFT;
6346 		tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
6347 	}
6348 
6349 	skb_tx_timestamp(skb);
6350 
6351 #ifdef CONFIG_IXGBE_PTP
6352 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
6353 		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
6354 		tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
6355 	}
6356 #endif
6357 
6358 #ifdef CONFIG_PCI_IOV
6359 	/*
6360 	 * Use the l2switch_enable flag - would be false if the DMA
6361 	 * Tx switch had been disabled.
6362 	 */
6363 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6364 		tx_flags |= IXGBE_TX_FLAGS_TXSW;
6365 
6366 #endif
6367 	/* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
6368 	if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
6369 	    ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
6370 	     (skb->priority != TC_PRIO_CONTROL))) {
6371 		tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6372 		tx_flags |= (skb->priority & 0x7) <<
6373 					IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
6374 		if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
6375 			struct vlan_ethhdr *vhdr;
6376 			if (skb_header_cloned(skb) &&
6377 			    pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
6378 				goto out_drop;
6379 			vhdr = (struct vlan_ethhdr *)skb->data;
6380 			vhdr->h_vlan_TCI = htons(tx_flags >>
6381 						 IXGBE_TX_FLAGS_VLAN_SHIFT);
6382 		} else {
6383 			tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6384 		}
6385 	}
6386 
6387 	/* record initial flags and protocol */
6388 	first->tx_flags = tx_flags;
6389 	first->protocol = protocol;
6390 
6391 #ifdef IXGBE_FCOE
6392 	/* setup tx offload for FCoE */
6393 	if ((protocol == __constant_htons(ETH_P_FCOE)) &&
6394 	    (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
6395 		tso = ixgbe_fso(tx_ring, first, &hdr_len);
6396 		if (tso < 0)
6397 			goto out_drop;
6398 
6399 		goto xmit_fcoe;
6400 	}
6401 
6402 #endif /* IXGBE_FCOE */
6403 	tso = ixgbe_tso(tx_ring, first, &hdr_len);
6404 	if (tso < 0)
6405 		goto out_drop;
6406 	else if (!tso)
6407 		ixgbe_tx_csum(tx_ring, first);
6408 
6409 	/* add the ATR filter if ATR is on */
6410 	if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6411 		ixgbe_atr(tx_ring, first);
6412 
6413 #ifdef IXGBE_FCOE
6414 xmit_fcoe:
6415 #endif /* IXGBE_FCOE */
6416 	ixgbe_tx_map(tx_ring, first, hdr_len);
6417 
6418 	ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6419 
6420 	return NETDEV_TX_OK;
6421 
6422 out_drop:
6423 	dev_kfree_skb_any(first->skb);
6424 	first->skb = NULL;
6425 
6426 	return NETDEV_TX_OK;
6427 }
6428 
6429 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
6430 				    struct net_device *netdev)
6431 {
6432 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6433 	struct ixgbe_ring *tx_ring;
6434 
6435 	/*
6436 	 * The minimum packet size for olinfo paylen is 17 so pad the skb
6437 	 * in order to meet this minimum size requirement.
6438 	 */
6439 	if (unlikely(skb->len < 17)) {
6440 		if (skb_pad(skb, 17 - skb->len))
6441 			return NETDEV_TX_OK;
6442 		skb->len = 17;
6443 	}
6444 
6445 	tx_ring = adapter->tx_ring[skb->queue_mapping];
6446 	return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6447 }
6448 
6449 /**
6450  * ixgbe_set_mac - Change the Ethernet Address of the NIC
6451  * @netdev: network interface device structure
6452  * @p: pointer to an address structure
6453  *
6454  * Returns 0 on success, negative on failure
6455  **/
6456 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6457 {
6458 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6459 	struct ixgbe_hw *hw = &adapter->hw;
6460 	struct sockaddr *addr = p;
6461 
6462 	if (!is_valid_ether_addr(addr->sa_data))
6463 		return -EADDRNOTAVAIL;
6464 
6465 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6466 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6467 
6468 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, VMDQ_P(0), IXGBE_RAH_AV);
6469 
6470 	return 0;
6471 }
6472 
6473 static int
6474 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6475 {
6476 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6477 	struct ixgbe_hw *hw = &adapter->hw;
6478 	u16 value;
6479 	int rc;
6480 
6481 	if (prtad != hw->phy.mdio.prtad)
6482 		return -EINVAL;
6483 	rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6484 	if (!rc)
6485 		rc = value;
6486 	return rc;
6487 }
6488 
6489 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6490 			    u16 addr, u16 value)
6491 {
6492 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6493 	struct ixgbe_hw *hw = &adapter->hw;
6494 
6495 	if (prtad != hw->phy.mdio.prtad)
6496 		return -EINVAL;
6497 	return hw->phy.ops.write_reg(hw, addr, devad, value);
6498 }
6499 
6500 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6501 {
6502 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6503 
6504 	switch (cmd) {
6505 #ifdef CONFIG_IXGBE_PTP
6506 	case SIOCSHWTSTAMP:
6507 		return ixgbe_ptp_hwtstamp_ioctl(adapter, req, cmd);
6508 #endif
6509 	default:
6510 		return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6511 	}
6512 }
6513 
6514 /**
6515  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6516  * netdev->dev_addrs
6517  * @netdev: network interface device structure
6518  *
6519  * Returns non-zero on failure
6520  **/
6521 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6522 {
6523 	int err = 0;
6524 	struct ixgbe_adapter *adapter = netdev_priv(dev);
6525 	struct ixgbe_hw *hw = &adapter->hw;
6526 
6527 	if (is_valid_ether_addr(hw->mac.san_addr)) {
6528 		rtnl_lock();
6529 		err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
6530 		rtnl_unlock();
6531 
6532 		/* update SAN MAC vmdq pool selection */
6533 		hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
6534 	}
6535 	return err;
6536 }
6537 
6538 /**
6539  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6540  * netdev->dev_addrs
6541  * @netdev: network interface device structure
6542  *
6543  * Returns non-zero on failure
6544  **/
6545 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6546 {
6547 	int err = 0;
6548 	struct ixgbe_adapter *adapter = netdev_priv(dev);
6549 	struct ixgbe_mac_info *mac = &adapter->hw.mac;
6550 
6551 	if (is_valid_ether_addr(mac->san_addr)) {
6552 		rtnl_lock();
6553 		err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6554 		rtnl_unlock();
6555 	}
6556 	return err;
6557 }
6558 
6559 #ifdef CONFIG_NET_POLL_CONTROLLER
6560 /*
6561  * Polling 'interrupt' - used by things like netconsole to send skbs
6562  * without having to re-enable interrupts. It's not called while
6563  * the interrupt routine is executing.
6564  */
6565 static void ixgbe_netpoll(struct net_device *netdev)
6566 {
6567 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6568 	int i;
6569 
6570 	/* if interface is down do nothing */
6571 	if (test_bit(__IXGBE_DOWN, &adapter->state))
6572 		return;
6573 
6574 	adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6575 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6576 		for (i = 0; i < adapter->num_q_vectors; i++)
6577 			ixgbe_msix_clean_rings(0, adapter->q_vector[i]);
6578 	} else {
6579 		ixgbe_intr(adapter->pdev->irq, netdev);
6580 	}
6581 	adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6582 }
6583 
6584 #endif
6585 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6586 						   struct rtnl_link_stats64 *stats)
6587 {
6588 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6589 	int i;
6590 
6591 	rcu_read_lock();
6592 	for (i = 0; i < adapter->num_rx_queues; i++) {
6593 		struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6594 		u64 bytes, packets;
6595 		unsigned int start;
6596 
6597 		if (ring) {
6598 			do {
6599 				start = u64_stats_fetch_begin_bh(&ring->syncp);
6600 				packets = ring->stats.packets;
6601 				bytes   = ring->stats.bytes;
6602 			} while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6603 			stats->rx_packets += packets;
6604 			stats->rx_bytes   += bytes;
6605 		}
6606 	}
6607 
6608 	for (i = 0; i < adapter->num_tx_queues; i++) {
6609 		struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
6610 		u64 bytes, packets;
6611 		unsigned int start;
6612 
6613 		if (ring) {
6614 			do {
6615 				start = u64_stats_fetch_begin_bh(&ring->syncp);
6616 				packets = ring->stats.packets;
6617 				bytes   = ring->stats.bytes;
6618 			} while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6619 			stats->tx_packets += packets;
6620 			stats->tx_bytes   += bytes;
6621 		}
6622 	}
6623 	rcu_read_unlock();
6624 	/* following stats updated by ixgbe_watchdog_task() */
6625 	stats->multicast	= netdev->stats.multicast;
6626 	stats->rx_errors	= netdev->stats.rx_errors;
6627 	stats->rx_length_errors	= netdev->stats.rx_length_errors;
6628 	stats->rx_crc_errors	= netdev->stats.rx_crc_errors;
6629 	stats->rx_missed_errors	= netdev->stats.rx_missed_errors;
6630 	return stats;
6631 }
6632 
6633 #ifdef CONFIG_IXGBE_DCB
6634 /**
6635  * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
6636  * @adapter: pointer to ixgbe_adapter
6637  * @tc: number of traffic classes currently enabled
6638  *
6639  * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
6640  * 802.1Q priority maps to a packet buffer that exists.
6641  */
6642 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
6643 {
6644 	struct ixgbe_hw *hw = &adapter->hw;
6645 	u32 reg, rsave;
6646 	int i;
6647 
6648 	/* 82598 have a static priority to TC mapping that can not
6649 	 * be changed so no validation is needed.
6650 	 */
6651 	if (hw->mac.type == ixgbe_mac_82598EB)
6652 		return;
6653 
6654 	reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
6655 	rsave = reg;
6656 
6657 	for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
6658 		u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
6659 
6660 		/* If up2tc is out of bounds default to zero */
6661 		if (up2tc > tc)
6662 			reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
6663 	}
6664 
6665 	if (reg != rsave)
6666 		IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
6667 
6668 	return;
6669 }
6670 
6671 /**
6672  * ixgbe_set_prio_tc_map - Configure netdev prio tc map
6673  * @adapter: Pointer to adapter struct
6674  *
6675  * Populate the netdev user priority to tc map
6676  */
6677 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
6678 {
6679 	struct net_device *dev = adapter->netdev;
6680 	struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
6681 	struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
6682 	u8 prio;
6683 
6684 	for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
6685 		u8 tc = 0;
6686 
6687 		if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
6688 			tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
6689 		else if (ets)
6690 			tc = ets->prio_tc[prio];
6691 
6692 		netdev_set_prio_tc_map(dev, prio, tc);
6693 	}
6694 }
6695 
6696 /**
6697  * ixgbe_setup_tc - configure net_device for multiple traffic classes
6698  *
6699  * @netdev: net device to configure
6700  * @tc: number of traffic classes to enable
6701  */
6702 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
6703 {
6704 	struct ixgbe_adapter *adapter = netdev_priv(dev);
6705 	struct ixgbe_hw *hw = &adapter->hw;
6706 
6707 	/* Hardware supports up to 8 traffic classes */
6708 	if (tc > adapter->dcb_cfg.num_tcs.pg_tcs ||
6709 	    (hw->mac.type == ixgbe_mac_82598EB &&
6710 	     tc < MAX_TRAFFIC_CLASS))
6711 		return -EINVAL;
6712 
6713 	/* Hardware has to reinitialize queues and interrupts to
6714 	 * match packet buffer alignment. Unfortunately, the
6715 	 * hardware is not flexible enough to do this dynamically.
6716 	 */
6717 	if (netif_running(dev))
6718 		ixgbe_close(dev);
6719 	ixgbe_clear_interrupt_scheme(adapter);
6720 
6721 	if (tc) {
6722 		netdev_set_num_tc(dev, tc);
6723 		ixgbe_set_prio_tc_map(adapter);
6724 
6725 		adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
6726 
6727 		if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
6728 			adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
6729 			adapter->hw.fc.requested_mode = ixgbe_fc_none;
6730 		}
6731 	} else {
6732 		netdev_reset_tc(dev);
6733 
6734 		if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6735 			adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
6736 
6737 		adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6738 
6739 		adapter->temp_dcb_cfg.pfc_mode_enable = false;
6740 		adapter->dcb_cfg.pfc_mode_enable = false;
6741 	}
6742 
6743 	ixgbe_init_interrupt_scheme(adapter);
6744 	ixgbe_validate_rtr(adapter, tc);
6745 	if (netif_running(dev))
6746 		ixgbe_open(dev);
6747 
6748 	return 0;
6749 }
6750 
6751 #endif /* CONFIG_IXGBE_DCB */
6752 void ixgbe_do_reset(struct net_device *netdev)
6753 {
6754 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6755 
6756 	if (netif_running(netdev))
6757 		ixgbe_reinit_locked(adapter);
6758 	else
6759 		ixgbe_reset(adapter);
6760 }
6761 
6762 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
6763 					    netdev_features_t features)
6764 {
6765 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6766 
6767 	/* If Rx checksum is disabled, then RSC/LRO should also be disabled */
6768 	if (!(features & NETIF_F_RXCSUM))
6769 		features &= ~NETIF_F_LRO;
6770 
6771 	/* Turn off LRO if not RSC capable */
6772 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
6773 		features &= ~NETIF_F_LRO;
6774 
6775 	return features;
6776 }
6777 
6778 static int ixgbe_set_features(struct net_device *netdev,
6779 			      netdev_features_t features)
6780 {
6781 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6782 	netdev_features_t changed = netdev->features ^ features;
6783 	bool need_reset = false;
6784 
6785 	/* Make sure RSC matches LRO, reset if change */
6786 	if (!(features & NETIF_F_LRO)) {
6787 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6788 			need_reset = true;
6789 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
6790 	} else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
6791 		   !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
6792 		if (adapter->rx_itr_setting == 1 ||
6793 		    adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
6794 			adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
6795 			need_reset = true;
6796 		} else if ((changed ^ features) & NETIF_F_LRO) {
6797 			e_info(probe, "rx-usecs set too low, "
6798 			       "disabling RSC\n");
6799 		}
6800 	}
6801 
6802 	/*
6803 	 * Check if Flow Director n-tuple support was enabled or disabled.  If
6804 	 * the state changed, we need to reset.
6805 	 */
6806 	switch (features & NETIF_F_NTUPLE) {
6807 	case NETIF_F_NTUPLE:
6808 		/* turn off ATR, enable perfect filters and reset */
6809 		if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
6810 			need_reset = true;
6811 
6812 		adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6813 		adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6814 		break;
6815 	default:
6816 		/* turn off perfect filters, enable ATR and reset */
6817 		if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6818 			need_reset = true;
6819 
6820 		adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6821 
6822 		/* We cannot enable ATR if SR-IOV is enabled */
6823 		if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6824 			break;
6825 
6826 		/* We cannot enable ATR if we have 2 or more traffic classes */
6827 		if (netdev_get_num_tc(netdev) > 1)
6828 			break;
6829 
6830 		/* We cannot enable ATR if RSS is disabled */
6831 		if (adapter->ring_feature[RING_F_RSS].limit <= 1)
6832 			break;
6833 
6834 		/* A sample rate of 0 indicates ATR disabled */
6835 		if (!adapter->atr_sample_rate)
6836 			break;
6837 
6838 		adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
6839 		break;
6840 	}
6841 
6842 	if (features & NETIF_F_HW_VLAN_RX)
6843 		ixgbe_vlan_strip_enable(adapter);
6844 	else
6845 		ixgbe_vlan_strip_disable(adapter);
6846 
6847 	if (changed & NETIF_F_RXALL)
6848 		need_reset = true;
6849 
6850 	netdev->features = features;
6851 	if (need_reset)
6852 		ixgbe_do_reset(netdev);
6853 
6854 	return 0;
6855 }
6856 
6857 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm,
6858 			     struct net_device *dev,
6859 			     unsigned char *addr,
6860 			     u16 flags)
6861 {
6862 	struct ixgbe_adapter *adapter = netdev_priv(dev);
6863 	int err;
6864 
6865 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
6866 		return -EOPNOTSUPP;
6867 
6868 	if (ndm->ndm_state & NUD_PERMANENT) {
6869 		pr_info("%s: FDB only supports static addresses\n",
6870 			ixgbe_driver_name);
6871 		return -EINVAL;
6872 	}
6873 
6874 	if (is_unicast_ether_addr(addr)) {
6875 		u32 rar_uc_entries = IXGBE_MAX_PF_MACVLANS;
6876 
6877 		if (netdev_uc_count(dev) < rar_uc_entries)
6878 			err = dev_uc_add_excl(dev, addr);
6879 		else
6880 			err = -ENOMEM;
6881 	} else if (is_multicast_ether_addr(addr)) {
6882 		err = dev_mc_add_excl(dev, addr);
6883 	} else {
6884 		err = -EINVAL;
6885 	}
6886 
6887 	/* Only return duplicate errors if NLM_F_EXCL is set */
6888 	if (err == -EEXIST && !(flags & NLM_F_EXCL))
6889 		err = 0;
6890 
6891 	return err;
6892 }
6893 
6894 static int ixgbe_ndo_fdb_del(struct ndmsg *ndm,
6895 			     struct net_device *dev,
6896 			     unsigned char *addr)
6897 {
6898 	struct ixgbe_adapter *adapter = netdev_priv(dev);
6899 	int err = -EOPNOTSUPP;
6900 
6901 	if (ndm->ndm_state & NUD_PERMANENT) {
6902 		pr_info("%s: FDB only supports static addresses\n",
6903 			ixgbe_driver_name);
6904 		return -EINVAL;
6905 	}
6906 
6907 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6908 		if (is_unicast_ether_addr(addr))
6909 			err = dev_uc_del(dev, addr);
6910 		else if (is_multicast_ether_addr(addr))
6911 			err = dev_mc_del(dev, addr);
6912 		else
6913 			err = -EINVAL;
6914 	}
6915 
6916 	return err;
6917 }
6918 
6919 static int ixgbe_ndo_fdb_dump(struct sk_buff *skb,
6920 			      struct netlink_callback *cb,
6921 			      struct net_device *dev,
6922 			      int idx)
6923 {
6924 	struct ixgbe_adapter *adapter = netdev_priv(dev);
6925 
6926 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6927 		idx = ndo_dflt_fdb_dump(skb, cb, dev, idx);
6928 
6929 	return idx;
6930 }
6931 
6932 static const struct net_device_ops ixgbe_netdev_ops = {
6933 	.ndo_open		= ixgbe_open,
6934 	.ndo_stop		= ixgbe_close,
6935 	.ndo_start_xmit		= ixgbe_xmit_frame,
6936 	.ndo_select_queue	= ixgbe_select_queue,
6937 	.ndo_set_rx_mode	= ixgbe_set_rx_mode,
6938 	.ndo_validate_addr	= eth_validate_addr,
6939 	.ndo_set_mac_address	= ixgbe_set_mac,
6940 	.ndo_change_mtu		= ixgbe_change_mtu,
6941 	.ndo_tx_timeout		= ixgbe_tx_timeout,
6942 	.ndo_vlan_rx_add_vid	= ixgbe_vlan_rx_add_vid,
6943 	.ndo_vlan_rx_kill_vid	= ixgbe_vlan_rx_kill_vid,
6944 	.ndo_do_ioctl		= ixgbe_ioctl,
6945 	.ndo_set_vf_mac		= ixgbe_ndo_set_vf_mac,
6946 	.ndo_set_vf_vlan	= ixgbe_ndo_set_vf_vlan,
6947 	.ndo_set_vf_tx_rate	= ixgbe_ndo_set_vf_bw,
6948 	.ndo_set_vf_spoofchk	= ixgbe_ndo_set_vf_spoofchk,
6949 	.ndo_get_vf_config	= ixgbe_ndo_get_vf_config,
6950 	.ndo_get_stats64	= ixgbe_get_stats64,
6951 #ifdef CONFIG_IXGBE_DCB
6952 	.ndo_setup_tc		= ixgbe_setup_tc,
6953 #endif
6954 #ifdef CONFIG_NET_POLL_CONTROLLER
6955 	.ndo_poll_controller	= ixgbe_netpoll,
6956 #endif
6957 #ifdef IXGBE_FCOE
6958 	.ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6959 	.ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
6960 	.ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6961 	.ndo_fcoe_enable = ixgbe_fcoe_enable,
6962 	.ndo_fcoe_disable = ixgbe_fcoe_disable,
6963 	.ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6964 	.ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
6965 #endif /* IXGBE_FCOE */
6966 	.ndo_set_features = ixgbe_set_features,
6967 	.ndo_fix_features = ixgbe_fix_features,
6968 	.ndo_fdb_add		= ixgbe_ndo_fdb_add,
6969 	.ndo_fdb_del		= ixgbe_ndo_fdb_del,
6970 	.ndo_fdb_dump		= ixgbe_ndo_fdb_dump,
6971 };
6972 
6973 /**
6974  * ixgbe_wol_supported - Check whether device supports WoL
6975  * @hw: hw specific details
6976  * @device_id: the device ID
6977  * @subdev_id: the subsystem device ID
6978  *
6979  * This function is used by probe and ethtool to determine
6980  * which devices have WoL support
6981  *
6982  **/
6983 int ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
6984 			u16 subdevice_id)
6985 {
6986 	struct ixgbe_hw *hw = &adapter->hw;
6987 	u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
6988 	int is_wol_supported = 0;
6989 
6990 	switch (device_id) {
6991 	case IXGBE_DEV_ID_82599_SFP:
6992 		/* Only these subdevices could supports WOL */
6993 		switch (subdevice_id) {
6994 		case IXGBE_SUBDEV_ID_82599_560FLR:
6995 			/* only support first port */
6996 			if (hw->bus.func != 0)
6997 				break;
6998 		case IXGBE_SUBDEV_ID_82599_SFP:
6999 		case IXGBE_SUBDEV_ID_82599_RNDC:
7000 			is_wol_supported = 1;
7001 			break;
7002 		}
7003 		break;
7004 	case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7005 		/* All except this subdevice support WOL */
7006 		if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7007 			is_wol_supported = 1;
7008 		break;
7009 	case IXGBE_DEV_ID_82599_KX4:
7010 		is_wol_supported = 1;
7011 		break;
7012 	case IXGBE_DEV_ID_X540T:
7013 		/* check eeprom to see if enabled wol */
7014 		if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
7015 		    ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
7016 		     (hw->bus.func == 0))) {
7017 			is_wol_supported = 1;
7018 		}
7019 		break;
7020 	}
7021 
7022 	return is_wol_supported;
7023 }
7024 
7025 /**
7026  * ixgbe_probe - Device Initialization Routine
7027  * @pdev: PCI device information struct
7028  * @ent: entry in ixgbe_pci_tbl
7029  *
7030  * Returns 0 on success, negative on failure
7031  *
7032  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7033  * The OS initialization, configuring of the adapter private structure,
7034  * and a hardware reset occur.
7035  **/
7036 static int __devinit ixgbe_probe(struct pci_dev *pdev,
7037 				 const struct pci_device_id *ent)
7038 {
7039 	struct net_device *netdev;
7040 	struct ixgbe_adapter *adapter = NULL;
7041 	struct ixgbe_hw *hw;
7042 	const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7043 	static int cards_found;
7044 	int i, err, pci_using_dac;
7045 	u8 part_str[IXGBE_PBANUM_LENGTH];
7046 	unsigned int indices = num_possible_cpus();
7047 	unsigned int dcb_max = 0;
7048 #ifdef IXGBE_FCOE
7049 	u16 device_caps;
7050 #endif
7051 	u32 eec;
7052 
7053 	/* Catch broken hardware that put the wrong VF device ID in
7054 	 * the PCIe SR-IOV capability.
7055 	 */
7056 	if (pdev->is_virtfn) {
7057 		WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7058 		     pci_name(pdev), pdev->vendor, pdev->device);
7059 		return -EINVAL;
7060 	}
7061 
7062 	err = pci_enable_device_mem(pdev);
7063 	if (err)
7064 		return err;
7065 
7066 	if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7067 	    !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7068 		pci_using_dac = 1;
7069 	} else {
7070 		err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7071 		if (err) {
7072 			err = dma_set_coherent_mask(&pdev->dev,
7073 						    DMA_BIT_MASK(32));
7074 			if (err) {
7075 				dev_err(&pdev->dev,
7076 					"No usable DMA configuration, aborting\n");
7077 				goto err_dma;
7078 			}
7079 		}
7080 		pci_using_dac = 0;
7081 	}
7082 
7083 	err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7084 					   IORESOURCE_MEM), ixgbe_driver_name);
7085 	if (err) {
7086 		dev_err(&pdev->dev,
7087 			"pci_request_selected_regions failed 0x%x\n", err);
7088 		goto err_pci_reg;
7089 	}
7090 
7091 	pci_enable_pcie_error_reporting(pdev);
7092 
7093 	pci_set_master(pdev);
7094 	pci_save_state(pdev);
7095 
7096 #ifdef CONFIG_IXGBE_DCB
7097 	if (ii->mac == ixgbe_mac_82598EB)
7098 		dcb_max = min_t(unsigned int, indices * MAX_TRAFFIC_CLASS,
7099 				IXGBE_MAX_RSS_INDICES);
7100 	else
7101 		dcb_max = min_t(unsigned int, indices * MAX_TRAFFIC_CLASS,
7102 				IXGBE_MAX_FDIR_INDICES);
7103 #endif
7104 
7105 	if (ii->mac == ixgbe_mac_82598EB)
7106 		indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7107 	else
7108 		indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7109 
7110 #ifdef IXGBE_FCOE
7111 	indices += min_t(unsigned int, num_possible_cpus(),
7112 			 IXGBE_MAX_FCOE_INDICES);
7113 #endif
7114 	indices = max_t(unsigned int, dcb_max, indices);
7115 	netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7116 	if (!netdev) {
7117 		err = -ENOMEM;
7118 		goto err_alloc_etherdev;
7119 	}
7120 
7121 	SET_NETDEV_DEV(netdev, &pdev->dev);
7122 
7123 	adapter = netdev_priv(netdev);
7124 	pci_set_drvdata(pdev, adapter);
7125 
7126 	adapter->netdev = netdev;
7127 	adapter->pdev = pdev;
7128 	hw = &adapter->hw;
7129 	hw->back = adapter;
7130 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
7131 
7132 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7133 			      pci_resource_len(pdev, 0));
7134 	if (!hw->hw_addr) {
7135 		err = -EIO;
7136 		goto err_ioremap;
7137 	}
7138 
7139 	for (i = 1; i <= 5; i++) {
7140 		if (pci_resource_len(pdev, i) == 0)
7141 			continue;
7142 	}
7143 
7144 	netdev->netdev_ops = &ixgbe_netdev_ops;
7145 	ixgbe_set_ethtool_ops(netdev);
7146 	netdev->watchdog_timeo = 5 * HZ;
7147 	strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7148 
7149 	adapter->bd_number = cards_found;
7150 
7151 	/* Setup hw api */
7152 	memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7153 	hw->mac.type  = ii->mac;
7154 
7155 	/* EEPROM */
7156 	memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7157 	eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7158 	/* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7159 	if (!(eec & (1 << 8)))
7160 		hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7161 
7162 	/* PHY */
7163 	memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7164 	hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7165 	/* ixgbe_identify_phy_generic will set prtad and mmds properly */
7166 	hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7167 	hw->phy.mdio.mmds = 0;
7168 	hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7169 	hw->phy.mdio.dev = netdev;
7170 	hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7171 	hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7172 
7173 	ii->get_invariants(hw);
7174 
7175 	/* setup the private structure */
7176 	err = ixgbe_sw_init(adapter);
7177 	if (err)
7178 		goto err_sw_init;
7179 
7180 	/* Make it possible the adapter to be woken up via WOL */
7181 	switch (adapter->hw.mac.type) {
7182 	case ixgbe_mac_82599EB:
7183 	case ixgbe_mac_X540:
7184 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7185 		break;
7186 	default:
7187 		break;
7188 	}
7189 
7190 	/*
7191 	 * If there is a fan on this device and it has failed log the
7192 	 * failure.
7193 	 */
7194 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7195 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7196 		if (esdp & IXGBE_ESDP_SDP1)
7197 			e_crit(probe, "Fan has stopped, replace the adapter\n");
7198 	}
7199 
7200 	if (allow_unsupported_sfp)
7201 		hw->allow_unsupported_sfp = allow_unsupported_sfp;
7202 
7203 	/* reset_hw fills in the perm_addr as well */
7204 	hw->phy.reset_if_overtemp = true;
7205 	err = hw->mac.ops.reset_hw(hw);
7206 	hw->phy.reset_if_overtemp = false;
7207 	if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7208 	    hw->mac.type == ixgbe_mac_82598EB) {
7209 		err = 0;
7210 	} else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7211 		e_dev_err("failed to load because an unsupported SFP+ "
7212 			  "module type was detected.\n");
7213 		e_dev_err("Reload the driver after installing a supported "
7214 			  "module.\n");
7215 		goto err_sw_init;
7216 	} else if (err) {
7217 		e_dev_err("HW Init failed: %d\n", err);
7218 		goto err_sw_init;
7219 	}
7220 
7221 #ifdef CONFIG_PCI_IOV
7222 	ixgbe_enable_sriov(adapter, ii);
7223 
7224 #endif
7225 	netdev->features = NETIF_F_SG |
7226 			   NETIF_F_IP_CSUM |
7227 			   NETIF_F_IPV6_CSUM |
7228 			   NETIF_F_HW_VLAN_TX |
7229 			   NETIF_F_HW_VLAN_RX |
7230 			   NETIF_F_HW_VLAN_FILTER |
7231 			   NETIF_F_TSO |
7232 			   NETIF_F_TSO6 |
7233 			   NETIF_F_RXHASH |
7234 			   NETIF_F_RXCSUM;
7235 
7236 	netdev->hw_features = netdev->features;
7237 
7238 	switch (adapter->hw.mac.type) {
7239 	case ixgbe_mac_82599EB:
7240 	case ixgbe_mac_X540:
7241 		netdev->features |= NETIF_F_SCTP_CSUM;
7242 		netdev->hw_features |= NETIF_F_SCTP_CSUM |
7243 				       NETIF_F_NTUPLE;
7244 		break;
7245 	default:
7246 		break;
7247 	}
7248 
7249 	netdev->hw_features |= NETIF_F_RXALL;
7250 
7251 	netdev->vlan_features |= NETIF_F_TSO;
7252 	netdev->vlan_features |= NETIF_F_TSO6;
7253 	netdev->vlan_features |= NETIF_F_IP_CSUM;
7254 	netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7255 	netdev->vlan_features |= NETIF_F_SG;
7256 
7257 	netdev->priv_flags |= IFF_UNICAST_FLT;
7258 	netdev->priv_flags |= IFF_SUPP_NOFCS;
7259 
7260 #ifdef CONFIG_IXGBE_DCB
7261 	netdev->dcbnl_ops = &dcbnl_ops;
7262 #endif
7263 
7264 #ifdef IXGBE_FCOE
7265 	if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7266 		if (hw->mac.ops.get_device_caps) {
7267 			hw->mac.ops.get_device_caps(hw, &device_caps);
7268 			if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7269 				adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7270 		}
7271 
7272 		adapter->ring_feature[RING_F_FCOE].limit = IXGBE_FCRETA_SIZE;
7273 
7274 		netdev->features |= NETIF_F_FSO |
7275 				    NETIF_F_FCOE_CRC;
7276 
7277 		netdev->vlan_features |= NETIF_F_FSO |
7278 					 NETIF_F_FCOE_CRC |
7279 					 NETIF_F_FCOE_MTU;
7280 	}
7281 #endif /* IXGBE_FCOE */
7282 	if (pci_using_dac) {
7283 		netdev->features |= NETIF_F_HIGHDMA;
7284 		netdev->vlan_features |= NETIF_F_HIGHDMA;
7285 	}
7286 
7287 	if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
7288 		netdev->hw_features |= NETIF_F_LRO;
7289 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7290 		netdev->features |= NETIF_F_LRO;
7291 
7292 	/* make sure the EEPROM is good */
7293 	if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7294 		e_dev_err("The EEPROM Checksum Is Not Valid\n");
7295 		err = -EIO;
7296 		goto err_sw_init;
7297 	}
7298 
7299 	memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7300 	memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7301 
7302 	if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7303 		e_dev_err("invalid MAC address\n");
7304 		err = -EIO;
7305 		goto err_sw_init;
7306 	}
7307 
7308 	setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7309 		    (unsigned long) adapter);
7310 
7311 	INIT_WORK(&adapter->service_task, ixgbe_service_task);
7312 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7313 
7314 	err = ixgbe_init_interrupt_scheme(adapter);
7315 	if (err)
7316 		goto err_sw_init;
7317 
7318 	/* WOL not supported for all devices */
7319 	adapter->wol = 0;
7320 	hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
7321 	if (ixgbe_wol_supported(adapter, pdev->device, pdev->subsystem_device))
7322 		adapter->wol = IXGBE_WUFC_MAG;
7323 
7324 	device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7325 
7326 #ifdef CONFIG_IXGBE_PTP
7327 	ixgbe_ptp_init(adapter);
7328 #endif /* CONFIG_IXGBE_PTP*/
7329 
7330 	/* save off EEPROM version number */
7331 	hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
7332 	hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
7333 
7334 	/* pick up the PCI bus settings for reporting later */
7335 	hw->mac.ops.get_bus_info(hw);
7336 
7337 	/* print bus type/speed/width info */
7338 	e_dev_info("(PCI Express:%s:%s) %pM\n",
7339 		   (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7340 		    hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7341 		    "Unknown"),
7342 		   (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7343 		    hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7344 		    hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7345 		    "Unknown"),
7346 		   netdev->dev_addr);
7347 
7348 	err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7349 	if (err)
7350 		strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7351 	if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7352 		e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7353 			   hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7354 		           part_str);
7355 	else
7356 		e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7357 			   hw->mac.type, hw->phy.type, part_str);
7358 
7359 	if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7360 		e_dev_warn("PCI-Express bandwidth available for this card is "
7361 			   "not sufficient for optimal performance.\n");
7362 		e_dev_warn("For optimal performance a x8 PCI-Express slot "
7363 			   "is required.\n");
7364 	}
7365 
7366 	/* reset the hardware with the new settings */
7367 	err = hw->mac.ops.start_hw(hw);
7368 	if (err == IXGBE_ERR_EEPROM_VERSION) {
7369 		/* We are running on a pre-production device, log a warning */
7370 		e_dev_warn("This device is a pre-production adapter/LOM. "
7371 			   "Please be aware there may be issues associated "
7372 			   "with your hardware.  If you are experiencing "
7373 			   "problems please contact your Intel or hardware "
7374 			   "representative who provided you with this "
7375 			   "hardware.\n");
7376 	}
7377 	strcpy(netdev->name, "eth%d");
7378 	err = register_netdev(netdev);
7379 	if (err)
7380 		goto err_register;
7381 
7382 	/* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7383 	if (hw->mac.ops.disable_tx_laser &&
7384 	    ((hw->phy.multispeed_fiber) ||
7385 	     ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7386 	      (hw->mac.type == ixgbe_mac_82599EB))))
7387 		hw->mac.ops.disable_tx_laser(hw);
7388 
7389 	/* carrier off reporting is important to ethtool even BEFORE open */
7390 	netif_carrier_off(netdev);
7391 
7392 #ifdef CONFIG_IXGBE_DCA
7393 	if (dca_add_requester(&pdev->dev) == 0) {
7394 		adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7395 		ixgbe_setup_dca(adapter);
7396 	}
7397 #endif
7398 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7399 		e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7400 		for (i = 0; i < adapter->num_vfs; i++)
7401 			ixgbe_vf_configuration(pdev, (i | 0x10000000));
7402 	}
7403 
7404 	/* firmware requires driver version to be 0xFFFFFFFF
7405 	 * since os does not support feature
7406 	 */
7407 	if (hw->mac.ops.set_fw_drv_ver)
7408 		hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
7409 					   0xFF);
7410 
7411 	/* add san mac addr to netdev */
7412 	ixgbe_add_sanmac_netdev(netdev);
7413 
7414 	e_dev_info("%s\n", ixgbe_default_device_descr);
7415 	cards_found++;
7416 
7417 #ifdef CONFIG_IXGBE_HWMON
7418 	if (ixgbe_sysfs_init(adapter))
7419 		e_err(probe, "failed to allocate sysfs resources\n");
7420 #endif /* CONFIG_IXGBE_HWMON */
7421 
7422 	return 0;
7423 
7424 err_register:
7425 	ixgbe_release_hw_control(adapter);
7426 	ixgbe_clear_interrupt_scheme(adapter);
7427 err_sw_init:
7428 	ixgbe_disable_sriov(adapter);
7429 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7430 	iounmap(hw->hw_addr);
7431 err_ioremap:
7432 	free_netdev(netdev);
7433 err_alloc_etherdev:
7434 	pci_release_selected_regions(pdev,
7435 				     pci_select_bars(pdev, IORESOURCE_MEM));
7436 err_pci_reg:
7437 err_dma:
7438 	pci_disable_device(pdev);
7439 	return err;
7440 }
7441 
7442 /**
7443  * ixgbe_remove - Device Removal Routine
7444  * @pdev: PCI device information struct
7445  *
7446  * ixgbe_remove is called by the PCI subsystem to alert the driver
7447  * that it should release a PCI device.  The could be caused by a
7448  * Hot-Plug event, or because the driver is going to be removed from
7449  * memory.
7450  **/
7451 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7452 {
7453 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7454 	struct net_device *netdev = adapter->netdev;
7455 
7456 	set_bit(__IXGBE_DOWN, &adapter->state);
7457 	cancel_work_sync(&adapter->service_task);
7458 
7459 #ifdef CONFIG_IXGBE_PTP
7460 	ixgbe_ptp_stop(adapter);
7461 #endif
7462 
7463 #ifdef CONFIG_IXGBE_DCA
7464 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7465 		adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7466 		dca_remove_requester(&pdev->dev);
7467 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7468 	}
7469 
7470 #endif
7471 #ifdef CONFIG_IXGBE_HWMON
7472 	ixgbe_sysfs_exit(adapter);
7473 #endif /* CONFIG_IXGBE_HWMON */
7474 
7475 	/* remove the added san mac */
7476 	ixgbe_del_sanmac_netdev(netdev);
7477 
7478 	if (netdev->reg_state == NETREG_REGISTERED)
7479 		unregister_netdev(netdev);
7480 
7481 	ixgbe_disable_sriov(adapter);
7482 
7483 	ixgbe_clear_interrupt_scheme(adapter);
7484 
7485 	ixgbe_release_hw_control(adapter);
7486 
7487 #ifdef CONFIG_DCB
7488 	kfree(adapter->ixgbe_ieee_pfc);
7489 	kfree(adapter->ixgbe_ieee_ets);
7490 
7491 #endif
7492 	iounmap(adapter->hw.hw_addr);
7493 	pci_release_selected_regions(pdev, pci_select_bars(pdev,
7494 				     IORESOURCE_MEM));
7495 
7496 	e_dev_info("complete\n");
7497 
7498 	free_netdev(netdev);
7499 
7500 	pci_disable_pcie_error_reporting(pdev);
7501 
7502 	pci_disable_device(pdev);
7503 }
7504 
7505 /**
7506  * ixgbe_io_error_detected - called when PCI error is detected
7507  * @pdev: Pointer to PCI device
7508  * @state: The current pci connection state
7509  *
7510  * This function is called after a PCI bus error affecting
7511  * this device has been detected.
7512  */
7513 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7514 						pci_channel_state_t state)
7515 {
7516 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7517 	struct net_device *netdev = adapter->netdev;
7518 
7519 #ifdef CONFIG_PCI_IOV
7520 	struct pci_dev *bdev, *vfdev;
7521 	u32 dw0, dw1, dw2, dw3;
7522 	int vf, pos;
7523 	u16 req_id, pf_func;
7524 
7525 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7526 	    adapter->num_vfs == 0)
7527 		goto skip_bad_vf_detection;
7528 
7529 	bdev = pdev->bus->self;
7530 	while (bdev && (bdev->pcie_type != PCI_EXP_TYPE_ROOT_PORT))
7531 		bdev = bdev->bus->self;
7532 
7533 	if (!bdev)
7534 		goto skip_bad_vf_detection;
7535 
7536 	pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
7537 	if (!pos)
7538 		goto skip_bad_vf_detection;
7539 
7540 	pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG, &dw0);
7541 	pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 4, &dw1);
7542 	pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 8, &dw2);
7543 	pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 12, &dw3);
7544 
7545 	req_id = dw1 >> 16;
7546 	/* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
7547 	if (!(req_id & 0x0080))
7548 		goto skip_bad_vf_detection;
7549 
7550 	pf_func = req_id & 0x01;
7551 	if ((pf_func & 1) == (pdev->devfn & 1)) {
7552 		unsigned int device_id;
7553 
7554 		vf = (req_id & 0x7F) >> 1;
7555 		e_dev_err("VF %d has caused a PCIe error\n", vf);
7556 		e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
7557 				"%8.8x\tdw3: %8.8x\n",
7558 		dw0, dw1, dw2, dw3);
7559 		switch (adapter->hw.mac.type) {
7560 		case ixgbe_mac_82599EB:
7561 			device_id = IXGBE_82599_VF_DEVICE_ID;
7562 			break;
7563 		case ixgbe_mac_X540:
7564 			device_id = IXGBE_X540_VF_DEVICE_ID;
7565 			break;
7566 		default:
7567 			device_id = 0;
7568 			break;
7569 		}
7570 
7571 		/* Find the pci device of the offending VF */
7572 		vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
7573 		while (vfdev) {
7574 			if (vfdev->devfn == (req_id & 0xFF))
7575 				break;
7576 			vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
7577 					       device_id, vfdev);
7578 		}
7579 		/*
7580 		 * There's a slim chance the VF could have been hot plugged,
7581 		 * so if it is no longer present we don't need to issue the
7582 		 * VFLR.  Just clean up the AER in that case.
7583 		 */
7584 		if (vfdev) {
7585 			e_dev_err("Issuing VFLR to VF %d\n", vf);
7586 			pci_write_config_dword(vfdev, 0xA8, 0x00008000);
7587 		}
7588 
7589 		pci_cleanup_aer_uncorrect_error_status(pdev);
7590 	}
7591 
7592 	/*
7593 	 * Even though the error may have occurred on the other port
7594 	 * we still need to increment the vf error reference count for
7595 	 * both ports because the I/O resume function will be called
7596 	 * for both of them.
7597 	 */
7598 	adapter->vferr_refcount++;
7599 
7600 	return PCI_ERS_RESULT_RECOVERED;
7601 
7602 skip_bad_vf_detection:
7603 #endif /* CONFIG_PCI_IOV */
7604 	netif_device_detach(netdev);
7605 
7606 	if (state == pci_channel_io_perm_failure)
7607 		return PCI_ERS_RESULT_DISCONNECT;
7608 
7609 	if (netif_running(netdev))
7610 		ixgbe_down(adapter);
7611 	pci_disable_device(pdev);
7612 
7613 	/* Request a slot reset. */
7614 	return PCI_ERS_RESULT_NEED_RESET;
7615 }
7616 
7617 /**
7618  * ixgbe_io_slot_reset - called after the pci bus has been reset.
7619  * @pdev: Pointer to PCI device
7620  *
7621  * Restart the card from scratch, as if from a cold-boot.
7622  */
7623 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7624 {
7625 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7626 	pci_ers_result_t result;
7627 	int err;
7628 
7629 	if (pci_enable_device_mem(pdev)) {
7630 		e_err(probe, "Cannot re-enable PCI device after reset.\n");
7631 		result = PCI_ERS_RESULT_DISCONNECT;
7632 	} else {
7633 		pci_set_master(pdev);
7634 		pci_restore_state(pdev);
7635 		pci_save_state(pdev);
7636 
7637 		pci_wake_from_d3(pdev, false);
7638 
7639 		ixgbe_reset(adapter);
7640 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7641 		result = PCI_ERS_RESULT_RECOVERED;
7642 	}
7643 
7644 	err = pci_cleanup_aer_uncorrect_error_status(pdev);
7645 	if (err) {
7646 		e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7647 			  "failed 0x%0x\n", err);
7648 		/* non-fatal, continue */
7649 	}
7650 
7651 	return result;
7652 }
7653 
7654 /**
7655  * ixgbe_io_resume - called when traffic can start flowing again.
7656  * @pdev: Pointer to PCI device
7657  *
7658  * This callback is called when the error recovery driver tells us that
7659  * its OK to resume normal operation.
7660  */
7661 static void ixgbe_io_resume(struct pci_dev *pdev)
7662 {
7663 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7664 	struct net_device *netdev = adapter->netdev;
7665 
7666 #ifdef CONFIG_PCI_IOV
7667 	if (adapter->vferr_refcount) {
7668 		e_info(drv, "Resuming after VF err\n");
7669 		adapter->vferr_refcount--;
7670 		return;
7671 	}
7672 
7673 #endif
7674 	if (netif_running(netdev))
7675 		ixgbe_up(adapter);
7676 
7677 	netif_device_attach(netdev);
7678 }
7679 
7680 static struct pci_error_handlers ixgbe_err_handler = {
7681 	.error_detected = ixgbe_io_error_detected,
7682 	.slot_reset = ixgbe_io_slot_reset,
7683 	.resume = ixgbe_io_resume,
7684 };
7685 
7686 static struct pci_driver ixgbe_driver = {
7687 	.name     = ixgbe_driver_name,
7688 	.id_table = ixgbe_pci_tbl,
7689 	.probe    = ixgbe_probe,
7690 	.remove   = __devexit_p(ixgbe_remove),
7691 #ifdef CONFIG_PM
7692 	.suspend  = ixgbe_suspend,
7693 	.resume   = ixgbe_resume,
7694 #endif
7695 	.shutdown = ixgbe_shutdown,
7696 	.err_handler = &ixgbe_err_handler
7697 };
7698 
7699 /**
7700  * ixgbe_init_module - Driver Registration Routine
7701  *
7702  * ixgbe_init_module is the first routine called when the driver is
7703  * loaded. All it does is register with the PCI subsystem.
7704  **/
7705 static int __init ixgbe_init_module(void)
7706 {
7707 	int ret;
7708 	pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7709 	pr_info("%s\n", ixgbe_copyright);
7710 
7711 #ifdef CONFIG_IXGBE_DCA
7712 	dca_register_notify(&dca_notifier);
7713 #endif
7714 
7715 	ret = pci_register_driver(&ixgbe_driver);
7716 	return ret;
7717 }
7718 
7719 module_init(ixgbe_init_module);
7720 
7721 /**
7722  * ixgbe_exit_module - Driver Exit Cleanup Routine
7723  *
7724  * ixgbe_exit_module is called just before the driver is removed
7725  * from memory.
7726  **/
7727 static void __exit ixgbe_exit_module(void)
7728 {
7729 #ifdef CONFIG_IXGBE_DCA
7730 	dca_unregister_notify(&dca_notifier);
7731 #endif
7732 	pci_unregister_driver(&ixgbe_driver);
7733 	rcu_barrier(); /* Wait for completion of call_rcu()'s */
7734 }
7735 
7736 #ifdef CONFIG_IXGBE_DCA
7737 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7738 			    void *p)
7739 {
7740 	int ret_val;
7741 
7742 	ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7743 					 __ixgbe_notify_dca);
7744 
7745 	return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7746 }
7747 
7748 #endif /* CONFIG_IXGBE_DCA */
7749 
7750 module_exit(ixgbe_exit_module);
7751 
7752 /* ixgbe_main.c */
7753