xref: /linux/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c (revision b43ab901d671e3e3cad425ea5e9a3c74e266dcdd)
1 /*******************************************************************************
2 
3   Intel 10 Gigabit PCI Express Linux driver
4   Copyright(c) 1999 - 2011 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 char ixgbe_default_device_descr[] =
59 			      "Intel(R) 10 Gigabit Network Connection";
60 #define MAJ 3
61 #define MIN 6
62 #define BUILD 7
63 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
64 	__stringify(BUILD) "-k"
65 const char ixgbe_driver_version[] = DRV_VERSION;
66 static const char ixgbe_copyright[] =
67 				"Copyright (c) 1999-2011 Intel Corporation.";
68 
69 static const struct ixgbe_info *ixgbe_info_tbl[] = {
70 	[board_82598] = &ixgbe_82598_info,
71 	[board_82599] = &ixgbe_82599_info,
72 	[board_X540] = &ixgbe_X540_info,
73 };
74 
75 /* ixgbe_pci_tbl - PCI Device ID Table
76  *
77  * Wildcard entries (PCI_ANY_ID) should come last
78  * Last entry must be all 0s
79  *
80  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
81  *   Class, Class Mask, private data (not used) }
82  */
83 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
84 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
85 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
86 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
87 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
88 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
89 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
90 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
91 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
92 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
93 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
94 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
95 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
96 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
97 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
98 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
99 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
100 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
101 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
102 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
103 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
104 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
105 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
106 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
107 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
108 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
109 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
110 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
111 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
112 	/* required last entry */
113 	{0, }
114 };
115 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
116 
117 #ifdef CONFIG_IXGBE_DCA
118 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
119 			    void *p);
120 static struct notifier_block dca_notifier = {
121 	.notifier_call = ixgbe_notify_dca,
122 	.next          = NULL,
123 	.priority      = 0
124 };
125 #endif
126 
127 #ifdef CONFIG_PCI_IOV
128 static unsigned int max_vfs;
129 module_param(max_vfs, uint, 0);
130 MODULE_PARM_DESC(max_vfs,
131 		 "Maximum number of virtual functions to allocate per physical function");
132 #endif /* CONFIG_PCI_IOV */
133 
134 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
135 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
136 MODULE_LICENSE("GPL");
137 MODULE_VERSION(DRV_VERSION);
138 
139 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
140 
141 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
142 {
143 	if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
144 	    !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
145 		schedule_work(&adapter->service_task);
146 }
147 
148 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
149 {
150 	BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
151 
152 	/* flush memory to make sure state is correct before next watchdog */
153 	smp_mb__before_clear_bit();
154 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
155 }
156 
157 struct ixgbe_reg_info {
158 	u32 ofs;
159 	char *name;
160 };
161 
162 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
163 
164 	/* General Registers */
165 	{IXGBE_CTRL, "CTRL"},
166 	{IXGBE_STATUS, "STATUS"},
167 	{IXGBE_CTRL_EXT, "CTRL_EXT"},
168 
169 	/* Interrupt Registers */
170 	{IXGBE_EICR, "EICR"},
171 
172 	/* RX Registers */
173 	{IXGBE_SRRCTL(0), "SRRCTL"},
174 	{IXGBE_DCA_RXCTRL(0), "DRXCTL"},
175 	{IXGBE_RDLEN(0), "RDLEN"},
176 	{IXGBE_RDH(0), "RDH"},
177 	{IXGBE_RDT(0), "RDT"},
178 	{IXGBE_RXDCTL(0), "RXDCTL"},
179 	{IXGBE_RDBAL(0), "RDBAL"},
180 	{IXGBE_RDBAH(0), "RDBAH"},
181 
182 	/* TX Registers */
183 	{IXGBE_TDBAL(0), "TDBAL"},
184 	{IXGBE_TDBAH(0), "TDBAH"},
185 	{IXGBE_TDLEN(0), "TDLEN"},
186 	{IXGBE_TDH(0), "TDH"},
187 	{IXGBE_TDT(0), "TDT"},
188 	{IXGBE_TXDCTL(0), "TXDCTL"},
189 
190 	/* List Terminator */
191 	{}
192 };
193 
194 
195 /*
196  * ixgbe_regdump - register printout routine
197  */
198 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
199 {
200 	int i = 0, j = 0;
201 	char rname[16];
202 	u32 regs[64];
203 
204 	switch (reginfo->ofs) {
205 	case IXGBE_SRRCTL(0):
206 		for (i = 0; i < 64; i++)
207 			regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
208 		break;
209 	case IXGBE_DCA_RXCTRL(0):
210 		for (i = 0; i < 64; i++)
211 			regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
212 		break;
213 	case IXGBE_RDLEN(0):
214 		for (i = 0; i < 64; i++)
215 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
216 		break;
217 	case IXGBE_RDH(0):
218 		for (i = 0; i < 64; i++)
219 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
220 		break;
221 	case IXGBE_RDT(0):
222 		for (i = 0; i < 64; i++)
223 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
224 		break;
225 	case IXGBE_RXDCTL(0):
226 		for (i = 0; i < 64; i++)
227 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
228 		break;
229 	case IXGBE_RDBAL(0):
230 		for (i = 0; i < 64; i++)
231 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
232 		break;
233 	case IXGBE_RDBAH(0):
234 		for (i = 0; i < 64; i++)
235 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
236 		break;
237 	case IXGBE_TDBAL(0):
238 		for (i = 0; i < 64; i++)
239 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
240 		break;
241 	case IXGBE_TDBAH(0):
242 		for (i = 0; i < 64; i++)
243 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
244 		break;
245 	case IXGBE_TDLEN(0):
246 		for (i = 0; i < 64; i++)
247 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
248 		break;
249 	case IXGBE_TDH(0):
250 		for (i = 0; i < 64; i++)
251 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
252 		break;
253 	case IXGBE_TDT(0):
254 		for (i = 0; i < 64; i++)
255 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
256 		break;
257 	case IXGBE_TXDCTL(0):
258 		for (i = 0; i < 64; i++)
259 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
260 		break;
261 	default:
262 		pr_info("%-15s %08x\n", reginfo->name,
263 			IXGBE_READ_REG(hw, reginfo->ofs));
264 		return;
265 	}
266 
267 	for (i = 0; i < 8; i++) {
268 		snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
269 		pr_err("%-15s", rname);
270 		for (j = 0; j < 8; j++)
271 			pr_cont(" %08x", regs[i*8+j]);
272 		pr_cont("\n");
273 	}
274 
275 }
276 
277 /*
278  * ixgbe_dump - Print registers, tx-rings and rx-rings
279  */
280 static void ixgbe_dump(struct ixgbe_adapter *adapter)
281 {
282 	struct net_device *netdev = adapter->netdev;
283 	struct ixgbe_hw *hw = &adapter->hw;
284 	struct ixgbe_reg_info *reginfo;
285 	int n = 0;
286 	struct ixgbe_ring *tx_ring;
287 	struct ixgbe_tx_buffer *tx_buffer_info;
288 	union ixgbe_adv_tx_desc *tx_desc;
289 	struct my_u0 { u64 a; u64 b; } *u0;
290 	struct ixgbe_ring *rx_ring;
291 	union ixgbe_adv_rx_desc *rx_desc;
292 	struct ixgbe_rx_buffer *rx_buffer_info;
293 	u32 staterr;
294 	int i = 0;
295 
296 	if (!netif_msg_hw(adapter))
297 		return;
298 
299 	/* Print netdevice Info */
300 	if (netdev) {
301 		dev_info(&adapter->pdev->dev, "Net device Info\n");
302 		pr_info("Device Name     state            "
303 			"trans_start      last_rx\n");
304 		pr_info("%-15s %016lX %016lX %016lX\n",
305 			netdev->name,
306 			netdev->state,
307 			netdev->trans_start,
308 			netdev->last_rx);
309 	}
310 
311 	/* Print Registers */
312 	dev_info(&adapter->pdev->dev, "Register Dump\n");
313 	pr_info(" Register Name   Value\n");
314 	for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
315 	     reginfo->name; reginfo++) {
316 		ixgbe_regdump(hw, reginfo);
317 	}
318 
319 	/* Print TX Ring Summary */
320 	if (!netdev || !netif_running(netdev))
321 		goto exit;
322 
323 	dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
324 	pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
325 	for (n = 0; n < adapter->num_tx_queues; n++) {
326 		tx_ring = adapter->tx_ring[n];
327 		tx_buffer_info =
328 			&tx_ring->tx_buffer_info[tx_ring->next_to_clean];
329 		pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
330 			   n, tx_ring->next_to_use, tx_ring->next_to_clean,
331 			   (u64)tx_buffer_info->dma,
332 			   tx_buffer_info->length,
333 			   tx_buffer_info->next_to_watch,
334 			   (u64)tx_buffer_info->time_stamp);
335 	}
336 
337 	/* Print TX Rings */
338 	if (!netif_msg_tx_done(adapter))
339 		goto rx_ring_summary;
340 
341 	dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
342 
343 	/* Transmit Descriptor Formats
344 	 *
345 	 * Advanced Transmit Descriptor
346 	 *   +--------------------------------------------------------------+
347 	 * 0 |         Buffer Address [63:0]                                |
348 	 *   +--------------------------------------------------------------+
349 	 * 8 |  PAYLEN  | PORTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
350 	 *   +--------------------------------------------------------------+
351 	 *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
352 	 */
353 
354 	for (n = 0; n < adapter->num_tx_queues; n++) {
355 		tx_ring = adapter->tx_ring[n];
356 		pr_info("------------------------------------\n");
357 		pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
358 		pr_info("------------------------------------\n");
359 		pr_info("T [desc]     [address 63:0  ] "
360 			"[PlPOIdStDDt Ln] [bi->dma       ] "
361 			"leng  ntw timestamp        bi->skb\n");
362 
363 		for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
364 			tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
365 			tx_buffer_info = &tx_ring->tx_buffer_info[i];
366 			u0 = (struct my_u0 *)tx_desc;
367 			pr_info("T [0x%03X]    %016llX %016llX %016llX"
368 				" %04X  %p %016llX %p", i,
369 				le64_to_cpu(u0->a),
370 				le64_to_cpu(u0->b),
371 				(u64)tx_buffer_info->dma,
372 				tx_buffer_info->length,
373 				tx_buffer_info->next_to_watch,
374 				(u64)tx_buffer_info->time_stamp,
375 				tx_buffer_info->skb);
376 			if (i == tx_ring->next_to_use &&
377 				i == tx_ring->next_to_clean)
378 				pr_cont(" NTC/U\n");
379 			else if (i == tx_ring->next_to_use)
380 				pr_cont(" NTU\n");
381 			else if (i == tx_ring->next_to_clean)
382 				pr_cont(" NTC\n");
383 			else
384 				pr_cont("\n");
385 
386 			if (netif_msg_pktdata(adapter) &&
387 				tx_buffer_info->dma != 0)
388 				print_hex_dump(KERN_INFO, "",
389 					DUMP_PREFIX_ADDRESS, 16, 1,
390 					phys_to_virt(tx_buffer_info->dma),
391 					tx_buffer_info->length, true);
392 		}
393 	}
394 
395 	/* Print RX Rings Summary */
396 rx_ring_summary:
397 	dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
398 	pr_info("Queue [NTU] [NTC]\n");
399 	for (n = 0; n < adapter->num_rx_queues; n++) {
400 		rx_ring = adapter->rx_ring[n];
401 		pr_info("%5d %5X %5X\n",
402 			n, rx_ring->next_to_use, rx_ring->next_to_clean);
403 	}
404 
405 	/* Print RX Rings */
406 	if (!netif_msg_rx_status(adapter))
407 		goto exit;
408 
409 	dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
410 
411 	/* Advanced Receive Descriptor (Read) Format
412 	 *    63                                           1        0
413 	 *    +-----------------------------------------------------+
414 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
415 	 *    +----------------------------------------------+------+
416 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
417 	 *    +-----------------------------------------------------+
418 	 *
419 	 *
420 	 * Advanced Receive Descriptor (Write-Back) Format
421 	 *
422 	 *   63       48 47    32 31  30      21 20 16 15   4 3     0
423 	 *   +------------------------------------------------------+
424 	 * 0 | Packet     IP     |SPH| HDR_LEN   | RSV|Packet|  RSS |
425 	 *   | Checksum   Ident  |   |           |    | Type | Type |
426 	 *   +------------------------------------------------------+
427 	 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
428 	 *   +------------------------------------------------------+
429 	 *   63       48 47    32 31            20 19               0
430 	 */
431 	for (n = 0; n < adapter->num_rx_queues; n++) {
432 		rx_ring = adapter->rx_ring[n];
433 		pr_info("------------------------------------\n");
434 		pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
435 		pr_info("------------------------------------\n");
436 		pr_info("R  [desc]      [ PktBuf     A0] "
437 			"[  HeadBuf   DD] [bi->dma       ] [bi->skb] "
438 			"<-- Adv Rx Read format\n");
439 		pr_info("RWB[desc]      [PcsmIpSHl PtRs] "
440 			"[vl er S cks ln] ---------------- [bi->skb] "
441 			"<-- Adv Rx Write-Back format\n");
442 
443 		for (i = 0; i < rx_ring->count; i++) {
444 			rx_buffer_info = &rx_ring->rx_buffer_info[i];
445 			rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
446 			u0 = (struct my_u0 *)rx_desc;
447 			staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
448 			if (staterr & IXGBE_RXD_STAT_DD) {
449 				/* Descriptor Done */
450 				pr_info("RWB[0x%03X]     %016llX "
451 					"%016llX ---------------- %p", i,
452 					le64_to_cpu(u0->a),
453 					le64_to_cpu(u0->b),
454 					rx_buffer_info->skb);
455 			} else {
456 				pr_info("R  [0x%03X]     %016llX "
457 					"%016llX %016llX %p", i,
458 					le64_to_cpu(u0->a),
459 					le64_to_cpu(u0->b),
460 					(u64)rx_buffer_info->dma,
461 					rx_buffer_info->skb);
462 
463 				if (netif_msg_pktdata(adapter)) {
464 					print_hex_dump(KERN_INFO, "",
465 					   DUMP_PREFIX_ADDRESS, 16, 1,
466 					   phys_to_virt(rx_buffer_info->dma),
467 					   rx_ring->rx_buf_len, true);
468 
469 					if (rx_ring->rx_buf_len
470 						< IXGBE_RXBUFFER_2K)
471 						print_hex_dump(KERN_INFO, "",
472 						  DUMP_PREFIX_ADDRESS, 16, 1,
473 						  phys_to_virt(
474 						    rx_buffer_info->page_dma +
475 						    rx_buffer_info->page_offset
476 						  ),
477 						  PAGE_SIZE/2, true);
478 				}
479 			}
480 
481 			if (i == rx_ring->next_to_use)
482 				pr_cont(" NTU\n");
483 			else if (i == rx_ring->next_to_clean)
484 				pr_cont(" NTC\n");
485 			else
486 				pr_cont("\n");
487 
488 		}
489 	}
490 
491 exit:
492 	return;
493 }
494 
495 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
496 {
497 	u32 ctrl_ext;
498 
499 	/* Let firmware take over control of h/w */
500 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
501 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
502 			ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
503 }
504 
505 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
506 {
507 	u32 ctrl_ext;
508 
509 	/* Let firmware know the driver has taken over */
510 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
511 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
512 			ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
513 }
514 
515 /*
516  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
517  * @adapter: pointer to adapter struct
518  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
519  * @queue: queue to map the corresponding interrupt to
520  * @msix_vector: the vector to map to the corresponding queue
521  *
522  */
523 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
524 			   u8 queue, u8 msix_vector)
525 {
526 	u32 ivar, index;
527 	struct ixgbe_hw *hw = &adapter->hw;
528 	switch (hw->mac.type) {
529 	case ixgbe_mac_82598EB:
530 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
531 		if (direction == -1)
532 			direction = 0;
533 		index = (((direction * 64) + queue) >> 2) & 0x1F;
534 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
535 		ivar &= ~(0xFF << (8 * (queue & 0x3)));
536 		ivar |= (msix_vector << (8 * (queue & 0x3)));
537 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
538 		break;
539 	case ixgbe_mac_82599EB:
540 	case ixgbe_mac_X540:
541 		if (direction == -1) {
542 			/* other causes */
543 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
544 			index = ((queue & 1) * 8);
545 			ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
546 			ivar &= ~(0xFF << index);
547 			ivar |= (msix_vector << index);
548 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
549 			break;
550 		} else {
551 			/* tx or rx causes */
552 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
553 			index = ((16 * (queue & 1)) + (8 * direction));
554 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
555 			ivar &= ~(0xFF << index);
556 			ivar |= (msix_vector << index);
557 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
558 			break;
559 		}
560 	default:
561 		break;
562 	}
563 }
564 
565 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
566 					  u64 qmask)
567 {
568 	u32 mask;
569 
570 	switch (adapter->hw.mac.type) {
571 	case ixgbe_mac_82598EB:
572 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
573 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
574 		break;
575 	case ixgbe_mac_82599EB:
576 	case ixgbe_mac_X540:
577 		mask = (qmask & 0xFFFFFFFF);
578 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
579 		mask = (qmask >> 32);
580 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
581 		break;
582 	default:
583 		break;
584 	}
585 }
586 
587 static inline void ixgbe_unmap_tx_resource(struct ixgbe_ring *ring,
588 					   struct ixgbe_tx_buffer *tx_buffer)
589 {
590 	if (tx_buffer->dma) {
591 		if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_MAPPED_AS_PAGE)
592 			dma_unmap_page(ring->dev,
593 			               tx_buffer->dma,
594 			               tx_buffer->length,
595 			               DMA_TO_DEVICE);
596 		else
597 			dma_unmap_single(ring->dev,
598 			                 tx_buffer->dma,
599 			                 tx_buffer->length,
600 			                 DMA_TO_DEVICE);
601 	}
602 	tx_buffer->dma = 0;
603 }
604 
605 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *tx_ring,
606 				      struct ixgbe_tx_buffer *tx_buffer_info)
607 {
608 	ixgbe_unmap_tx_resource(tx_ring, tx_buffer_info);
609 	if (tx_buffer_info->skb)
610 		dev_kfree_skb_any(tx_buffer_info->skb);
611 	tx_buffer_info->skb = NULL;
612 	/* tx_buffer_info must be completely set up in the transmit path */
613 }
614 
615 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
616 {
617 	struct ixgbe_hw *hw = &adapter->hw;
618 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
619 	u32 data = 0;
620 	u32 xoff[8] = {0};
621 	int i;
622 
623 	if ((hw->fc.current_mode == ixgbe_fc_full) ||
624 	    (hw->fc.current_mode == ixgbe_fc_rx_pause)) {
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 		return;
642 	} else if (!(adapter->dcb_cfg.pfc_mode_enable))
643 		return;
644 
645 	/* update stats for each tc, only valid with PFC enabled */
646 	for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
647 		switch (hw->mac.type) {
648 		case ixgbe_mac_82598EB:
649 			xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
650 			break;
651 		default:
652 			xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
653 		}
654 		hwstats->pxoffrxc[i] += xoff[i];
655 	}
656 
657 	/* disarm tx queues that have received xoff frames */
658 	for (i = 0; i < adapter->num_tx_queues; i++) {
659 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
660 		u8 tc = tx_ring->dcb_tc;
661 
662 		if (xoff[tc])
663 			clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
664 	}
665 }
666 
667 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
668 {
669 	return ring->tx_stats.completed;
670 }
671 
672 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
673 {
674 	struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
675 	struct ixgbe_hw *hw = &adapter->hw;
676 
677 	u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
678 	u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
679 
680 	if (head != tail)
681 		return (head < tail) ?
682 			tail - head : (tail + ring->count - head);
683 
684 	return 0;
685 }
686 
687 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
688 {
689 	u32 tx_done = ixgbe_get_tx_completed(tx_ring);
690 	u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
691 	u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
692 	bool ret = false;
693 
694 	clear_check_for_tx_hang(tx_ring);
695 
696 	/*
697 	 * Check for a hung queue, but be thorough. This verifies
698 	 * that a transmit has been completed since the previous
699 	 * check AND there is at least one packet pending. The
700 	 * ARMED bit is set to indicate a potential hang. The
701 	 * bit is cleared if a pause frame is received to remove
702 	 * false hang detection due to PFC or 802.3x frames. By
703 	 * requiring this to fail twice we avoid races with
704 	 * pfc clearing the ARMED bit and conditions where we
705 	 * run the check_tx_hang logic with a transmit completion
706 	 * pending but without time to complete it yet.
707 	 */
708 	if ((tx_done_old == tx_done) && tx_pending) {
709 		/* make sure it is true for two checks in a row */
710 		ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
711 				       &tx_ring->state);
712 	} else {
713 		/* update completed stats and continue */
714 		tx_ring->tx_stats.tx_done_old = tx_done;
715 		/* reset the countdown */
716 		clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
717 	}
718 
719 	return ret;
720 }
721 
722 /**
723  * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
724  * @adapter: driver private struct
725  **/
726 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
727 {
728 
729 	/* Do the reset outside of interrupt context */
730 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
731 		adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
732 		ixgbe_service_event_schedule(adapter);
733 	}
734 }
735 
736 /**
737  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
738  * @q_vector: structure containing interrupt and ring information
739  * @tx_ring: tx ring to clean
740  **/
741 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
742 			       struct ixgbe_ring *tx_ring)
743 {
744 	struct ixgbe_adapter *adapter = q_vector->adapter;
745 	struct ixgbe_tx_buffer *tx_buffer;
746 	union ixgbe_adv_tx_desc *tx_desc;
747 	unsigned int total_bytes = 0, total_packets = 0;
748 	unsigned int budget = q_vector->tx.work_limit;
749 	u16 i = tx_ring->next_to_clean;
750 
751 	tx_buffer = &tx_ring->tx_buffer_info[i];
752 	tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
753 
754 	for (; budget; budget--) {
755 		union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
756 
757 		/* if next_to_watch is not set then there is no work pending */
758 		if (!eop_desc)
759 			break;
760 
761 		/* if DD is not set pending work has not been completed */
762 		if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
763 			break;
764 
765 		/* count the packet as being completed */
766 		tx_ring->tx_stats.completed++;
767 
768 		/* clear next_to_watch to prevent false hangs */
769 		tx_buffer->next_to_watch = NULL;
770 
771 		/* prevent any other reads prior to eop_desc being verified */
772 		rmb();
773 
774 		do {
775 			ixgbe_unmap_tx_resource(tx_ring, tx_buffer);
776 			tx_desc->wb.status = 0;
777 			if (likely(tx_desc == eop_desc)) {
778 				eop_desc = NULL;
779 				dev_kfree_skb_any(tx_buffer->skb);
780 				tx_buffer->skb = NULL;
781 
782 				total_bytes += tx_buffer->bytecount;
783 				total_packets += tx_buffer->gso_segs;
784 			}
785 
786 			tx_buffer++;
787 			tx_desc++;
788 			i++;
789 			if (unlikely(i == tx_ring->count)) {
790 				i = 0;
791 
792 				tx_buffer = tx_ring->tx_buffer_info;
793 				tx_desc = IXGBE_TX_DESC_ADV(tx_ring, 0);
794 			}
795 
796 		} while (eop_desc);
797 	}
798 
799 	tx_ring->next_to_clean = i;
800 	u64_stats_update_begin(&tx_ring->syncp);
801 	tx_ring->stats.bytes += total_bytes;
802 	tx_ring->stats.packets += total_packets;
803 	u64_stats_update_end(&tx_ring->syncp);
804 	q_vector->tx.total_bytes += total_bytes;
805 	q_vector->tx.total_packets += total_packets;
806 
807 	if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
808 		/* schedule immediate reset if we believe we hung */
809 		struct ixgbe_hw *hw = &adapter->hw;
810 		tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
811 		e_err(drv, "Detected Tx Unit Hang\n"
812 			"  Tx Queue             <%d>\n"
813 			"  TDH, TDT             <%x>, <%x>\n"
814 			"  next_to_use          <%x>\n"
815 			"  next_to_clean        <%x>\n"
816 			"tx_buffer_info[next_to_clean]\n"
817 			"  time_stamp           <%lx>\n"
818 			"  jiffies              <%lx>\n",
819 			tx_ring->queue_index,
820 			IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
821 			IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
822 			tx_ring->next_to_use, i,
823 			tx_ring->tx_buffer_info[i].time_stamp, jiffies);
824 
825 		netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
826 
827 		e_info(probe,
828 		       "tx hang %d detected on queue %d, resetting adapter\n",
829 			adapter->tx_timeout_count + 1, tx_ring->queue_index);
830 
831 		/* schedule immediate reset if we believe we hung */
832 		ixgbe_tx_timeout_reset(adapter);
833 
834 		/* the adapter is about to reset, no point in enabling stuff */
835 		return true;
836 	}
837 
838 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
839 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
840 		     (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
841 		/* Make sure that anybody stopping the queue after this
842 		 * sees the new next_to_clean.
843 		 */
844 		smp_mb();
845 		if (__netif_subqueue_stopped(tx_ring->netdev, tx_ring->queue_index) &&
846 		    !test_bit(__IXGBE_DOWN, &adapter->state)) {
847 			netif_wake_subqueue(tx_ring->netdev, tx_ring->queue_index);
848 			++tx_ring->tx_stats.restart_queue;
849 		}
850 	}
851 
852 	return !!budget;
853 }
854 
855 #ifdef CONFIG_IXGBE_DCA
856 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
857 				struct ixgbe_ring *rx_ring,
858 				int cpu)
859 {
860 	struct ixgbe_hw *hw = &adapter->hw;
861 	u32 rxctrl;
862 	u8 reg_idx = rx_ring->reg_idx;
863 
864 	rxctrl = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(reg_idx));
865 	switch (hw->mac.type) {
866 	case ixgbe_mac_82598EB:
867 		rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
868 		rxctrl |= dca3_get_tag(rx_ring->dev, cpu);
869 		break;
870 	case ixgbe_mac_82599EB:
871 	case ixgbe_mac_X540:
872 		rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
873 		rxctrl |= (dca3_get_tag(rx_ring->dev, cpu) <<
874 			   IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
875 		break;
876 	default:
877 		break;
878 	}
879 	rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
880 	rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
881 	rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
882 	IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
883 }
884 
885 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
886 				struct ixgbe_ring *tx_ring,
887 				int cpu)
888 {
889 	struct ixgbe_hw *hw = &adapter->hw;
890 	u32 txctrl;
891 	u8 reg_idx = tx_ring->reg_idx;
892 
893 	switch (hw->mac.type) {
894 	case ixgbe_mac_82598EB:
895 		txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(reg_idx));
896 		txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
897 		txctrl |= dca3_get_tag(tx_ring->dev, cpu);
898 		txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
899 		IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(reg_idx), txctrl);
900 		break;
901 	case ixgbe_mac_82599EB:
902 	case ixgbe_mac_X540:
903 		txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx));
904 		txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
905 		txctrl |= (dca3_get_tag(tx_ring->dev, cpu) <<
906 			   IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
907 		txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
908 		IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx), txctrl);
909 		break;
910 	default:
911 		break;
912 	}
913 }
914 
915 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
916 {
917 	struct ixgbe_adapter *adapter = q_vector->adapter;
918 	struct ixgbe_ring *ring;
919 	int cpu = get_cpu();
920 
921 	if (q_vector->cpu == cpu)
922 		goto out_no_update;
923 
924 	for (ring = q_vector->tx.ring; ring != NULL; ring = ring->next)
925 		ixgbe_update_tx_dca(adapter, ring, cpu);
926 
927 	for (ring = q_vector->rx.ring; ring != NULL; ring = ring->next)
928 		ixgbe_update_rx_dca(adapter, ring, cpu);
929 
930 	q_vector->cpu = cpu;
931 out_no_update:
932 	put_cpu();
933 }
934 
935 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
936 {
937 	int num_q_vectors;
938 	int i;
939 
940 	if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
941 		return;
942 
943 	/* always use CB2 mode, difference is masked in the CB driver */
944 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
945 
946 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
947 		num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
948 	else
949 		num_q_vectors = 1;
950 
951 	for (i = 0; i < num_q_vectors; i++) {
952 		adapter->q_vector[i]->cpu = -1;
953 		ixgbe_update_dca(adapter->q_vector[i]);
954 	}
955 }
956 
957 static int __ixgbe_notify_dca(struct device *dev, void *data)
958 {
959 	struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
960 	unsigned long event = *(unsigned long *)data;
961 
962 	if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
963 		return 0;
964 
965 	switch (event) {
966 	case DCA_PROVIDER_ADD:
967 		/* if we're already enabled, don't do it again */
968 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
969 			break;
970 		if (dca_add_requester(dev) == 0) {
971 			adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
972 			ixgbe_setup_dca(adapter);
973 			break;
974 		}
975 		/* Fall Through since DCA is disabled. */
976 	case DCA_PROVIDER_REMOVE:
977 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
978 			dca_remove_requester(dev);
979 			adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
980 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
981 		}
982 		break;
983 	}
984 
985 	return 0;
986 }
987 #endif /* CONFIG_IXGBE_DCA */
988 
989 static inline void ixgbe_rx_hash(union ixgbe_adv_rx_desc *rx_desc,
990 				 struct sk_buff *skb)
991 {
992 	skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
993 }
994 
995 /**
996  * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
997  * @adapter: address of board private structure
998  * @rx_desc: advanced rx descriptor
999  *
1000  * Returns : true if it is FCoE pkt
1001  */
1002 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_adapter *adapter,
1003 				    union ixgbe_adv_rx_desc *rx_desc)
1004 {
1005 	__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1006 
1007 	return (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
1008 	       ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1009 		(cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1010 			     IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1011 }
1012 
1013 /**
1014  * ixgbe_receive_skb - Send a completed packet up the stack
1015  * @adapter: board private structure
1016  * @skb: packet to send up
1017  * @status: hardware indication of status of receive
1018  * @rx_ring: rx descriptor ring (for a specific queue) to setup
1019  * @rx_desc: rx descriptor
1020  **/
1021 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
1022 			      struct sk_buff *skb, u8 status,
1023 			      struct ixgbe_ring *ring,
1024 			      union ixgbe_adv_rx_desc *rx_desc)
1025 {
1026 	struct ixgbe_adapter *adapter = q_vector->adapter;
1027 	struct napi_struct *napi = &q_vector->napi;
1028 	bool is_vlan = (status & IXGBE_RXD_STAT_VP);
1029 	u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
1030 
1031 	if (is_vlan && (tag & VLAN_VID_MASK))
1032 		__vlan_hwaccel_put_tag(skb, tag);
1033 
1034 	if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1035 		napi_gro_receive(napi, skb);
1036 	else
1037 		netif_rx(skb);
1038 }
1039 
1040 /**
1041  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1042  * @adapter: address of board private structure
1043  * @status_err: hardware indication of status of receive
1044  * @skb: skb currently being received and modified
1045  * @status_err: status error value of last descriptor in packet
1046  **/
1047 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
1048 				     union ixgbe_adv_rx_desc *rx_desc,
1049 				     struct sk_buff *skb,
1050 				     u32 status_err)
1051 {
1052 	skb->ip_summed = CHECKSUM_NONE;
1053 
1054 	/* Rx csum disabled */
1055 	if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
1056 		return;
1057 
1058 	/* if IP and error */
1059 	if ((status_err & IXGBE_RXD_STAT_IPCS) &&
1060 	    (status_err & IXGBE_RXDADV_ERR_IPE)) {
1061 		adapter->hw_csum_rx_error++;
1062 		return;
1063 	}
1064 
1065 	if (!(status_err & IXGBE_RXD_STAT_L4CS))
1066 		return;
1067 
1068 	if (status_err & IXGBE_RXDADV_ERR_TCPE) {
1069 		u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1070 
1071 		/*
1072 		 * 82599 errata, UDP frames with a 0 checksum can be marked as
1073 		 * checksum errors.
1074 		 */
1075 		if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
1076 		    (adapter->hw.mac.type == ixgbe_mac_82599EB))
1077 			return;
1078 
1079 		adapter->hw_csum_rx_error++;
1080 		return;
1081 	}
1082 
1083 	/* It must be a TCP or UDP packet with a valid checksum */
1084 	skb->ip_summed = CHECKSUM_UNNECESSARY;
1085 }
1086 
1087 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1088 {
1089 	/*
1090 	 * Force memory writes to complete before letting h/w
1091 	 * know there are new descriptors to fetch.  (Only
1092 	 * applicable for weak-ordered memory model archs,
1093 	 * such as IA-64).
1094 	 */
1095 	wmb();
1096 	writel(val, rx_ring->tail);
1097 }
1098 
1099 /**
1100  * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1101  * @rx_ring: ring to place buffers on
1102  * @cleaned_count: number of buffers to replace
1103  **/
1104 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1105 {
1106 	union ixgbe_adv_rx_desc *rx_desc;
1107 	struct ixgbe_rx_buffer *bi;
1108 	struct sk_buff *skb;
1109 	u16 i = rx_ring->next_to_use;
1110 
1111 	/* do nothing if no valid netdev defined */
1112 	if (!rx_ring->netdev)
1113 		return;
1114 
1115 	while (cleaned_count--) {
1116 		rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1117 		bi = &rx_ring->rx_buffer_info[i];
1118 		skb = bi->skb;
1119 
1120 		if (!skb) {
1121 			skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1122 							rx_ring->rx_buf_len);
1123 			if (!skb) {
1124 				rx_ring->rx_stats.alloc_rx_buff_failed++;
1125 				goto no_buffers;
1126 			}
1127 			/* initialize queue mapping */
1128 			skb_record_rx_queue(skb, rx_ring->queue_index);
1129 			bi->skb = skb;
1130 		}
1131 
1132 		if (!bi->dma) {
1133 			bi->dma = dma_map_single(rx_ring->dev,
1134 						 skb->data,
1135 						 rx_ring->rx_buf_len,
1136 						 DMA_FROM_DEVICE);
1137 			if (dma_mapping_error(rx_ring->dev, bi->dma)) {
1138 				rx_ring->rx_stats.alloc_rx_buff_failed++;
1139 				bi->dma = 0;
1140 				goto no_buffers;
1141 			}
1142 		}
1143 
1144 		if (ring_is_ps_enabled(rx_ring)) {
1145 			if (!bi->page) {
1146 				bi->page = alloc_page(GFP_ATOMIC | __GFP_COLD);
1147 				if (!bi->page) {
1148 					rx_ring->rx_stats.alloc_rx_page_failed++;
1149 					goto no_buffers;
1150 				}
1151 			}
1152 
1153 			if (!bi->page_dma) {
1154 				/* use a half page if we're re-using */
1155 				bi->page_offset ^= PAGE_SIZE / 2;
1156 				bi->page_dma = dma_map_page(rx_ring->dev,
1157 							    bi->page,
1158 							    bi->page_offset,
1159 							    PAGE_SIZE / 2,
1160 							    DMA_FROM_DEVICE);
1161 				if (dma_mapping_error(rx_ring->dev,
1162 						      bi->page_dma)) {
1163 					rx_ring->rx_stats.alloc_rx_page_failed++;
1164 					bi->page_dma = 0;
1165 					goto no_buffers;
1166 				}
1167 			}
1168 
1169 			/* Refresh the desc even if buffer_addrs didn't change
1170 			 * because each write-back erases this info. */
1171 			rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1172 			rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1173 		} else {
1174 			rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1175 			rx_desc->read.hdr_addr = 0;
1176 		}
1177 
1178 		i++;
1179 		if (i == rx_ring->count)
1180 			i = 0;
1181 	}
1182 
1183 no_buffers:
1184 	if (rx_ring->next_to_use != i) {
1185 		rx_ring->next_to_use = i;
1186 		ixgbe_release_rx_desc(rx_ring, i);
1187 	}
1188 }
1189 
1190 static inline u16 ixgbe_get_hlen(union ixgbe_adv_rx_desc *rx_desc)
1191 {
1192 	/* HW will not DMA in data larger than the given buffer, even if it
1193 	 * parses the (NFS, of course) header to be larger.  In that case, it
1194 	 * fills the header buffer and spills the rest into the page.
1195 	 */
1196 	u16 hdr_info = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info);
1197 	u16 hlen = (hdr_info &  IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1198 		    IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1199 	if (hlen > IXGBE_RX_HDR_SIZE)
1200 		hlen = IXGBE_RX_HDR_SIZE;
1201 	return hlen;
1202 }
1203 
1204 /**
1205  * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1206  * @skb: pointer to the last skb in the rsc queue
1207  *
1208  * This function changes a queue full of hw rsc buffers into a completed
1209  * packet.  It uses the ->prev pointers to find the first packet and then
1210  * turns it into the frag list owner.
1211  **/
1212 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb)
1213 {
1214 	unsigned int frag_list_size = 0;
1215 	unsigned int skb_cnt = 1;
1216 
1217 	while (skb->prev) {
1218 		struct sk_buff *prev = skb->prev;
1219 		frag_list_size += skb->len;
1220 		skb->prev = NULL;
1221 		skb = prev;
1222 		skb_cnt++;
1223 	}
1224 
1225 	skb_shinfo(skb)->frag_list = skb->next;
1226 	skb->next = NULL;
1227 	skb->len += frag_list_size;
1228 	skb->data_len += frag_list_size;
1229 	skb->truesize += frag_list_size;
1230 	IXGBE_RSC_CB(skb)->skb_cnt = skb_cnt;
1231 
1232 	return skb;
1233 }
1234 
1235 static inline bool ixgbe_get_rsc_state(union ixgbe_adv_rx_desc *rx_desc)
1236 {
1237 	return !!(le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1238 		IXGBE_RXDADV_RSCCNT_MASK);
1239 }
1240 
1241 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1242 			       struct ixgbe_ring *rx_ring,
1243 			       int budget)
1244 {
1245 	struct ixgbe_adapter *adapter = q_vector->adapter;
1246 	union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1247 	struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1248 	struct sk_buff *skb;
1249 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1250 	const int current_node = numa_node_id();
1251 #ifdef IXGBE_FCOE
1252 	int ddp_bytes = 0;
1253 #endif /* IXGBE_FCOE */
1254 	u32 staterr;
1255 	u16 i;
1256 	u16 cleaned_count = 0;
1257 	bool pkt_is_rsc = false;
1258 
1259 	i = rx_ring->next_to_clean;
1260 	rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1261 	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1262 
1263 	while (staterr & IXGBE_RXD_STAT_DD) {
1264 		u32 upper_len = 0;
1265 
1266 		rmb(); /* read descriptor and rx_buffer_info after status DD */
1267 
1268 		rx_buffer_info = &rx_ring->rx_buffer_info[i];
1269 
1270 		skb = rx_buffer_info->skb;
1271 		rx_buffer_info->skb = NULL;
1272 		prefetch(skb->data);
1273 
1274 		if (ring_is_rsc_enabled(rx_ring))
1275 			pkt_is_rsc = ixgbe_get_rsc_state(rx_desc);
1276 
1277 		/* linear means we are building an skb from multiple pages */
1278 		if (!skb_is_nonlinear(skb)) {
1279 			u16 hlen;
1280 			if (pkt_is_rsc &&
1281 			    !(staterr & IXGBE_RXD_STAT_EOP) &&
1282 			    !skb->prev) {
1283 				/*
1284 				 * When HWRSC is enabled, delay unmapping
1285 				 * of the first packet. It carries the
1286 				 * header information, HW may still
1287 				 * access the header after the writeback.
1288 				 * Only unmap it when EOP is reached
1289 				 */
1290 				IXGBE_RSC_CB(skb)->delay_unmap = true;
1291 				IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1292 			} else {
1293 				dma_unmap_single(rx_ring->dev,
1294 						 rx_buffer_info->dma,
1295 						 rx_ring->rx_buf_len,
1296 						 DMA_FROM_DEVICE);
1297 			}
1298 			rx_buffer_info->dma = 0;
1299 
1300 			if (ring_is_ps_enabled(rx_ring)) {
1301 				hlen = ixgbe_get_hlen(rx_desc);
1302 				upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1303 			} else {
1304 				hlen = le16_to_cpu(rx_desc->wb.upper.length);
1305 			}
1306 
1307 			skb_put(skb, hlen);
1308 		} else {
1309 			/* assume packet split since header is unmapped */
1310 			upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1311 		}
1312 
1313 		if (upper_len) {
1314 			dma_unmap_page(rx_ring->dev,
1315 				       rx_buffer_info->page_dma,
1316 				       PAGE_SIZE / 2,
1317 				       DMA_FROM_DEVICE);
1318 			rx_buffer_info->page_dma = 0;
1319 			skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1320 					   rx_buffer_info->page,
1321 					   rx_buffer_info->page_offset,
1322 					   upper_len);
1323 
1324 			if ((page_count(rx_buffer_info->page) == 1) &&
1325 			    (page_to_nid(rx_buffer_info->page) == current_node))
1326 				get_page(rx_buffer_info->page);
1327 			else
1328 				rx_buffer_info->page = NULL;
1329 
1330 			skb->len += upper_len;
1331 			skb->data_len += upper_len;
1332 			skb->truesize += PAGE_SIZE / 2;
1333 		}
1334 
1335 		i++;
1336 		if (i == rx_ring->count)
1337 			i = 0;
1338 
1339 		next_rxd = IXGBE_RX_DESC_ADV(rx_ring, i);
1340 		prefetch(next_rxd);
1341 		cleaned_count++;
1342 
1343 		if (pkt_is_rsc) {
1344 			u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1345 				     IXGBE_RXDADV_NEXTP_SHIFT;
1346 			next_buffer = &rx_ring->rx_buffer_info[nextp];
1347 		} else {
1348 			next_buffer = &rx_ring->rx_buffer_info[i];
1349 		}
1350 
1351 		if (!(staterr & IXGBE_RXD_STAT_EOP)) {
1352 			if (ring_is_ps_enabled(rx_ring)) {
1353 				rx_buffer_info->skb = next_buffer->skb;
1354 				rx_buffer_info->dma = next_buffer->dma;
1355 				next_buffer->skb = skb;
1356 				next_buffer->dma = 0;
1357 			} else {
1358 				skb->next = next_buffer->skb;
1359 				skb->next->prev = skb;
1360 			}
1361 			rx_ring->rx_stats.non_eop_descs++;
1362 			goto next_desc;
1363 		}
1364 
1365 		if (skb->prev) {
1366 			skb = ixgbe_transform_rsc_queue(skb);
1367 			/* if we got here without RSC the packet is invalid */
1368 			if (!pkt_is_rsc) {
1369 				__pskb_trim(skb, 0);
1370 				rx_buffer_info->skb = skb;
1371 				goto next_desc;
1372 			}
1373 		}
1374 
1375 		if (ring_is_rsc_enabled(rx_ring)) {
1376 			if (IXGBE_RSC_CB(skb)->delay_unmap) {
1377 				dma_unmap_single(rx_ring->dev,
1378 						 IXGBE_RSC_CB(skb)->dma,
1379 						 rx_ring->rx_buf_len,
1380 						 DMA_FROM_DEVICE);
1381 				IXGBE_RSC_CB(skb)->dma = 0;
1382 				IXGBE_RSC_CB(skb)->delay_unmap = false;
1383 			}
1384 		}
1385 		if (pkt_is_rsc) {
1386 			if (ring_is_ps_enabled(rx_ring))
1387 				rx_ring->rx_stats.rsc_count +=
1388 					skb_shinfo(skb)->nr_frags;
1389 			else
1390 				rx_ring->rx_stats.rsc_count +=
1391 					IXGBE_RSC_CB(skb)->skb_cnt;
1392 			rx_ring->rx_stats.rsc_flush++;
1393 		}
1394 
1395 		/* ERR_MASK will only have valid bits if EOP set */
1396 		if (unlikely(staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK)) {
1397 			dev_kfree_skb_any(skb);
1398 			goto next_desc;
1399 		}
1400 
1401 		ixgbe_rx_checksum(adapter, rx_desc, skb, staterr);
1402 		if (adapter->netdev->features & NETIF_F_RXHASH)
1403 			ixgbe_rx_hash(rx_desc, skb);
1404 
1405 		/* probably a little skewed due to removing CRC */
1406 		total_rx_bytes += skb->len;
1407 		total_rx_packets++;
1408 
1409 		skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1410 #ifdef IXGBE_FCOE
1411 		/* if ddp, not passing to ULD unless for FCP_RSP or error */
1412 		if (ixgbe_rx_is_fcoe(adapter, rx_desc)) {
1413 			ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb,
1414 						   staterr);
1415 			if (!ddp_bytes) {
1416 				dev_kfree_skb_any(skb);
1417 				goto next_desc;
1418 			}
1419 		}
1420 #endif /* IXGBE_FCOE */
1421 		ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1422 
1423 		budget--;
1424 next_desc:
1425 		rx_desc->wb.upper.status_error = 0;
1426 
1427 		if (!budget)
1428 			break;
1429 
1430 		/* return some buffers to hardware, one at a time is too slow */
1431 		if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1432 			ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1433 			cleaned_count = 0;
1434 		}
1435 
1436 		/* use prefetched values */
1437 		rx_desc = next_rxd;
1438 		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1439 	}
1440 
1441 	rx_ring->next_to_clean = i;
1442 	cleaned_count = ixgbe_desc_unused(rx_ring);
1443 
1444 	if (cleaned_count)
1445 		ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1446 
1447 #ifdef IXGBE_FCOE
1448 	/* include DDPed FCoE data */
1449 	if (ddp_bytes > 0) {
1450 		unsigned int mss;
1451 
1452 		mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1453 			sizeof(struct fc_frame_header) -
1454 			sizeof(struct fcoe_crc_eof);
1455 		if (mss > 512)
1456 			mss &= ~511;
1457 		total_rx_bytes += ddp_bytes;
1458 		total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1459 	}
1460 #endif /* IXGBE_FCOE */
1461 
1462 	u64_stats_update_begin(&rx_ring->syncp);
1463 	rx_ring->stats.packets += total_rx_packets;
1464 	rx_ring->stats.bytes += total_rx_bytes;
1465 	u64_stats_update_end(&rx_ring->syncp);
1466 	q_vector->rx.total_packets += total_rx_packets;
1467 	q_vector->rx.total_bytes += total_rx_bytes;
1468 
1469 	return !!budget;
1470 }
1471 
1472 /**
1473  * ixgbe_configure_msix - Configure MSI-X hardware
1474  * @adapter: board private structure
1475  *
1476  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1477  * interrupts.
1478  **/
1479 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1480 {
1481 	struct ixgbe_q_vector *q_vector;
1482 	int q_vectors, v_idx;
1483 	u32 mask;
1484 
1485 	q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1486 
1487 	/* Populate MSIX to EITR Select */
1488 	if (adapter->num_vfs > 32) {
1489 		u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
1490 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
1491 	}
1492 
1493 	/*
1494 	 * Populate the IVAR table and set the ITR values to the
1495 	 * corresponding register.
1496 	 */
1497 	for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1498 		struct ixgbe_ring *ring;
1499 		q_vector = adapter->q_vector[v_idx];
1500 
1501 		for (ring = q_vector->rx.ring; ring != NULL; ring = ring->next)
1502 			ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1503 
1504 		for (ring = q_vector->tx.ring; ring != NULL; ring = ring->next)
1505 			ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1506 
1507 		if (q_vector->tx.ring && !q_vector->rx.ring) {
1508 			/* tx only vector */
1509 			if (adapter->tx_itr_setting == 1)
1510 				q_vector->itr = IXGBE_10K_ITR;
1511 			else
1512 				q_vector->itr = adapter->tx_itr_setting;
1513 		} else {
1514 			/* rx or rx/tx vector */
1515 			if (adapter->rx_itr_setting == 1)
1516 				q_vector->itr = IXGBE_20K_ITR;
1517 			else
1518 				q_vector->itr = adapter->rx_itr_setting;
1519 		}
1520 
1521 		ixgbe_write_eitr(q_vector);
1522 	}
1523 
1524 	switch (adapter->hw.mac.type) {
1525 	case ixgbe_mac_82598EB:
1526 		ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1527 			       v_idx);
1528 		break;
1529 	case ixgbe_mac_82599EB:
1530 	case ixgbe_mac_X540:
1531 		ixgbe_set_ivar(adapter, -1, 1, v_idx);
1532 		break;
1533 	default:
1534 		break;
1535 	}
1536 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1537 
1538 	/* set up to autoclear timer, and the vectors */
1539 	mask = IXGBE_EIMS_ENABLE_MASK;
1540 	mask &= ~(IXGBE_EIMS_OTHER |
1541 		  IXGBE_EIMS_MAILBOX |
1542 		  IXGBE_EIMS_LSC);
1543 
1544 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1545 }
1546 
1547 enum latency_range {
1548 	lowest_latency = 0,
1549 	low_latency = 1,
1550 	bulk_latency = 2,
1551 	latency_invalid = 255
1552 };
1553 
1554 /**
1555  * ixgbe_update_itr - update the dynamic ITR value based on statistics
1556  * @q_vector: structure containing interrupt and ring information
1557  * @ring_container: structure containing ring performance data
1558  *
1559  *      Stores a new ITR value based on packets and byte
1560  *      counts during the last interrupt.  The advantage of per interrupt
1561  *      computation is faster updates and more accurate ITR for the current
1562  *      traffic pattern.  Constants in this function were computed
1563  *      based on theoretical maximum wire speed and thresholds were set based
1564  *      on testing data as well as attempting to minimize response time
1565  *      while increasing bulk throughput.
1566  *      this functionality is controlled by the InterruptThrottleRate module
1567  *      parameter (see ixgbe_param.c)
1568  **/
1569 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
1570 			     struct ixgbe_ring_container *ring_container)
1571 {
1572 	u64 bytes_perint;
1573 	struct ixgbe_adapter *adapter = q_vector->adapter;
1574 	int bytes = ring_container->total_bytes;
1575 	int packets = ring_container->total_packets;
1576 	u32 timepassed_us;
1577 	u8 itr_setting = ring_container->itr;
1578 
1579 	if (packets == 0)
1580 		return;
1581 
1582 	/* simple throttlerate management
1583 	 *    0-20MB/s lowest (100000 ints/s)
1584 	 *   20-100MB/s low   (20000 ints/s)
1585 	 *  100-1249MB/s bulk (8000 ints/s)
1586 	 */
1587 	/* what was last interrupt timeslice? */
1588 	timepassed_us = q_vector->itr >> 2;
1589 	bytes_perint = bytes / timepassed_us; /* bytes/usec */
1590 
1591 	switch (itr_setting) {
1592 	case lowest_latency:
1593 		if (bytes_perint > adapter->eitr_low)
1594 			itr_setting = low_latency;
1595 		break;
1596 	case low_latency:
1597 		if (bytes_perint > adapter->eitr_high)
1598 			itr_setting = bulk_latency;
1599 		else if (bytes_perint <= adapter->eitr_low)
1600 			itr_setting = lowest_latency;
1601 		break;
1602 	case bulk_latency:
1603 		if (bytes_perint <= adapter->eitr_high)
1604 			itr_setting = low_latency;
1605 		break;
1606 	}
1607 
1608 	/* clear work counters since we have the values we need */
1609 	ring_container->total_bytes = 0;
1610 	ring_container->total_packets = 0;
1611 
1612 	/* write updated itr to ring container */
1613 	ring_container->itr = itr_setting;
1614 }
1615 
1616 /**
1617  * ixgbe_write_eitr - write EITR register in hardware specific way
1618  * @q_vector: structure containing interrupt and ring information
1619  *
1620  * This function is made to be called by ethtool and by the driver
1621  * when it needs to update EITR registers at runtime.  Hardware
1622  * specific quirks/differences are taken care of here.
1623  */
1624 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1625 {
1626 	struct ixgbe_adapter *adapter = q_vector->adapter;
1627 	struct ixgbe_hw *hw = &adapter->hw;
1628 	int v_idx = q_vector->v_idx;
1629 	u32 itr_reg = q_vector->itr;
1630 
1631 	switch (adapter->hw.mac.type) {
1632 	case ixgbe_mac_82598EB:
1633 		/* must write high and low 16 bits to reset counter */
1634 		itr_reg |= (itr_reg << 16);
1635 		break;
1636 	case ixgbe_mac_82599EB:
1637 	case ixgbe_mac_X540:
1638 		/*
1639 		 * set the WDIS bit to not clear the timer bits and cause an
1640 		 * immediate assertion of the interrupt
1641 		 */
1642 		itr_reg |= IXGBE_EITR_CNT_WDIS;
1643 		break;
1644 	default:
1645 		break;
1646 	}
1647 	IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1648 }
1649 
1650 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
1651 {
1652 	u32 new_itr = q_vector->itr;
1653 	u8 current_itr;
1654 
1655 	ixgbe_update_itr(q_vector, &q_vector->tx);
1656 	ixgbe_update_itr(q_vector, &q_vector->rx);
1657 
1658 	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1659 
1660 	switch (current_itr) {
1661 	/* counts and packets in update_itr are dependent on these numbers */
1662 	case lowest_latency:
1663 		new_itr = IXGBE_100K_ITR;
1664 		break;
1665 	case low_latency:
1666 		new_itr = IXGBE_20K_ITR;
1667 		break;
1668 	case bulk_latency:
1669 		new_itr = IXGBE_8K_ITR;
1670 		break;
1671 	default:
1672 		break;
1673 	}
1674 
1675 	if (new_itr != q_vector->itr) {
1676 		/* do an exponential smoothing */
1677 		new_itr = (10 * new_itr * q_vector->itr) /
1678 			  ((9 * new_itr) + q_vector->itr);
1679 
1680 		/* save the algorithm value here */
1681 		q_vector->itr = new_itr & IXGBE_MAX_EITR;
1682 
1683 		ixgbe_write_eitr(q_vector);
1684 	}
1685 }
1686 
1687 /**
1688  * ixgbe_check_overtemp_subtask - check for over tempurature
1689  * @adapter: pointer to adapter
1690  **/
1691 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
1692 {
1693 	struct ixgbe_hw *hw = &adapter->hw;
1694 	u32 eicr = adapter->interrupt_event;
1695 
1696 	if (test_bit(__IXGBE_DOWN, &adapter->state))
1697 		return;
1698 
1699 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1700 	    !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
1701 		return;
1702 
1703 	adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
1704 
1705 	switch (hw->device_id) {
1706 	case IXGBE_DEV_ID_82599_T3_LOM:
1707 		/*
1708 		 * Since the warning interrupt is for both ports
1709 		 * we don't have to check if:
1710 		 *  - This interrupt wasn't for our port.
1711 		 *  - We may have missed the interrupt so always have to
1712 		 *    check if we  got a LSC
1713 		 */
1714 		if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
1715 		    !(eicr & IXGBE_EICR_LSC))
1716 			return;
1717 
1718 		if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
1719 			u32 autoneg;
1720 			bool link_up = false;
1721 
1722 			hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1723 
1724 			if (link_up)
1725 				return;
1726 		}
1727 
1728 		/* Check if this is not due to overtemp */
1729 		if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
1730 			return;
1731 
1732 		break;
1733 	default:
1734 		if (!(eicr & IXGBE_EICR_GPI_SDP0))
1735 			return;
1736 		break;
1737 	}
1738 	e_crit(drv,
1739 	       "Network adapter has been stopped because it has over heated. "
1740 	       "Restart the computer. If the problem persists, "
1741 	       "power off the system and replace the adapter\n");
1742 
1743 	adapter->interrupt_event = 0;
1744 }
1745 
1746 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1747 {
1748 	struct ixgbe_hw *hw = &adapter->hw;
1749 
1750 	if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1751 	    (eicr & IXGBE_EICR_GPI_SDP1)) {
1752 		e_crit(probe, "Fan has stopped, replace the adapter\n");
1753 		/* write to clear the interrupt */
1754 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1755 	}
1756 }
1757 
1758 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
1759 {
1760 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
1761 		return;
1762 
1763 	switch (adapter->hw.mac.type) {
1764 	case ixgbe_mac_82599EB:
1765 		/*
1766 		 * Need to check link state so complete overtemp check
1767 		 * on service task
1768 		 */
1769 		if (((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)) &&
1770 		    (!test_bit(__IXGBE_DOWN, &adapter->state))) {
1771 			adapter->interrupt_event = eicr;
1772 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
1773 			ixgbe_service_event_schedule(adapter);
1774 			return;
1775 		}
1776 		return;
1777 	case ixgbe_mac_X540:
1778 		if (!(eicr & IXGBE_EICR_TS))
1779 			return;
1780 		break;
1781 	default:
1782 		return;
1783 	}
1784 
1785 	e_crit(drv,
1786 	       "Network adapter has been stopped because it has over heated. "
1787 	       "Restart the computer. If the problem persists, "
1788 	       "power off the system and replace the adapter\n");
1789 }
1790 
1791 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1792 {
1793 	struct ixgbe_hw *hw = &adapter->hw;
1794 
1795 	if (eicr & IXGBE_EICR_GPI_SDP2) {
1796 		/* Clear the interrupt */
1797 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1798 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1799 			adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
1800 			ixgbe_service_event_schedule(adapter);
1801 		}
1802 	}
1803 
1804 	if (eicr & IXGBE_EICR_GPI_SDP1) {
1805 		/* Clear the interrupt */
1806 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1807 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1808 			adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
1809 			ixgbe_service_event_schedule(adapter);
1810 		}
1811 	}
1812 }
1813 
1814 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1815 {
1816 	struct ixgbe_hw *hw = &adapter->hw;
1817 
1818 	adapter->lsc_int++;
1819 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1820 	adapter->link_check_timeout = jiffies;
1821 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1822 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1823 		IXGBE_WRITE_FLUSH(hw);
1824 		ixgbe_service_event_schedule(adapter);
1825 	}
1826 }
1827 
1828 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1829 					   u64 qmask)
1830 {
1831 	u32 mask;
1832 	struct ixgbe_hw *hw = &adapter->hw;
1833 
1834 	switch (hw->mac.type) {
1835 	case ixgbe_mac_82598EB:
1836 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1837 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
1838 		break;
1839 	case ixgbe_mac_82599EB:
1840 	case ixgbe_mac_X540:
1841 		mask = (qmask & 0xFFFFFFFF);
1842 		if (mask)
1843 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
1844 		mask = (qmask >> 32);
1845 		if (mask)
1846 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
1847 		break;
1848 	default:
1849 		break;
1850 	}
1851 	/* skip the flush */
1852 }
1853 
1854 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1855 					    u64 qmask)
1856 {
1857 	u32 mask;
1858 	struct ixgbe_hw *hw = &adapter->hw;
1859 
1860 	switch (hw->mac.type) {
1861 	case ixgbe_mac_82598EB:
1862 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1863 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
1864 		break;
1865 	case ixgbe_mac_82599EB:
1866 	case ixgbe_mac_X540:
1867 		mask = (qmask & 0xFFFFFFFF);
1868 		if (mask)
1869 			IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
1870 		mask = (qmask >> 32);
1871 		if (mask)
1872 			IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
1873 		break;
1874 	default:
1875 		break;
1876 	}
1877 	/* skip the flush */
1878 }
1879 
1880 /**
1881  * ixgbe_irq_enable - Enable default interrupt generation settings
1882  * @adapter: board private structure
1883  **/
1884 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
1885 				    bool flush)
1886 {
1887 	u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
1888 
1889 	/* don't reenable LSC while waiting for link */
1890 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
1891 		mask &= ~IXGBE_EIMS_LSC;
1892 
1893 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
1894 		switch (adapter->hw.mac.type) {
1895 		case ixgbe_mac_82599EB:
1896 			mask |= IXGBE_EIMS_GPI_SDP0;
1897 			break;
1898 		case ixgbe_mac_X540:
1899 			mask |= IXGBE_EIMS_TS;
1900 			break;
1901 		default:
1902 			break;
1903 		}
1904 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
1905 		mask |= IXGBE_EIMS_GPI_SDP1;
1906 	switch (adapter->hw.mac.type) {
1907 	case ixgbe_mac_82599EB:
1908 		mask |= IXGBE_EIMS_GPI_SDP1;
1909 		mask |= IXGBE_EIMS_GPI_SDP2;
1910 	case ixgbe_mac_X540:
1911 		mask |= IXGBE_EIMS_ECC;
1912 		mask |= IXGBE_EIMS_MAILBOX;
1913 		break;
1914 	default:
1915 		break;
1916 	}
1917 	if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
1918 	    !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
1919 		mask |= IXGBE_EIMS_FLOW_DIR;
1920 
1921 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1922 	if (queues)
1923 		ixgbe_irq_enable_queues(adapter, ~0);
1924 	if (flush)
1925 		IXGBE_WRITE_FLUSH(&adapter->hw);
1926 }
1927 
1928 static irqreturn_t ixgbe_msix_other(int irq, void *data)
1929 {
1930 	struct ixgbe_adapter *adapter = data;
1931 	struct ixgbe_hw *hw = &adapter->hw;
1932 	u32 eicr;
1933 
1934 	/*
1935 	 * Workaround for Silicon errata.  Use clear-by-write instead
1936 	 * of clear-by-read.  Reading with EICS will return the
1937 	 * interrupt causes without clearing, which later be done
1938 	 * with the write to EICR.
1939 	 */
1940 	eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1941 	IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1942 
1943 	if (eicr & IXGBE_EICR_LSC)
1944 		ixgbe_check_lsc(adapter);
1945 
1946 	if (eicr & IXGBE_EICR_MAILBOX)
1947 		ixgbe_msg_task(adapter);
1948 
1949 	switch (hw->mac.type) {
1950 	case ixgbe_mac_82599EB:
1951 	case ixgbe_mac_X540:
1952 		if (eicr & IXGBE_EICR_ECC)
1953 			e_info(link, "Received unrecoverable ECC Err, please "
1954 			       "reboot\n");
1955 		/* Handle Flow Director Full threshold interrupt */
1956 		if (eicr & IXGBE_EICR_FLOW_DIR) {
1957 			int reinit_count = 0;
1958 			int i;
1959 			for (i = 0; i < adapter->num_tx_queues; i++) {
1960 				struct ixgbe_ring *ring = adapter->tx_ring[i];
1961 				if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
1962 						       &ring->state))
1963 					reinit_count++;
1964 			}
1965 			if (reinit_count) {
1966 				/* no more flow director interrupts until after init */
1967 				IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
1968 				adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
1969 				ixgbe_service_event_schedule(adapter);
1970 			}
1971 		}
1972 		ixgbe_check_sfp_event(adapter, eicr);
1973 		ixgbe_check_overtemp_event(adapter, eicr);
1974 		break;
1975 	default:
1976 		break;
1977 	}
1978 
1979 	ixgbe_check_fan_failure(adapter, eicr);
1980 
1981 	/* re-enable the original interrupt state, no lsc, no queues */
1982 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
1983 		ixgbe_irq_enable(adapter, false, false);
1984 
1985 	return IRQ_HANDLED;
1986 }
1987 
1988 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
1989 {
1990 	struct ixgbe_q_vector *q_vector = data;
1991 
1992 	/* EIAM disabled interrupts (on this vector) for us */
1993 
1994 	if (q_vector->rx.ring || q_vector->tx.ring)
1995 		napi_schedule(&q_vector->napi);
1996 
1997 	return IRQ_HANDLED;
1998 }
1999 
2000 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2001 				     int r_idx)
2002 {
2003 	struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2004 	struct ixgbe_ring *rx_ring = a->rx_ring[r_idx];
2005 
2006 	rx_ring->q_vector = q_vector;
2007 	rx_ring->next = q_vector->rx.ring;
2008 	q_vector->rx.ring = rx_ring;
2009 	q_vector->rx.count++;
2010 }
2011 
2012 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2013 				     int t_idx)
2014 {
2015 	struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2016 	struct ixgbe_ring *tx_ring = a->tx_ring[t_idx];
2017 
2018 	tx_ring->q_vector = q_vector;
2019 	tx_ring->next = q_vector->tx.ring;
2020 	q_vector->tx.ring = tx_ring;
2021 	q_vector->tx.count++;
2022 	q_vector->tx.work_limit = a->tx_work_limit;
2023 }
2024 
2025 /**
2026  * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2027  * @adapter: board private structure to initialize
2028  *
2029  * This function maps descriptor rings to the queue-specific vectors
2030  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
2031  * one vector per ring/queue, but on a constrained vector budget, we
2032  * group the rings as "efficiently" as possible.  You would add new
2033  * mapping configurations in here.
2034  **/
2035 static void ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter)
2036 {
2037 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2038 	int rxr_remaining = adapter->num_rx_queues, rxr_idx = 0;
2039 	int txr_remaining = adapter->num_tx_queues, txr_idx = 0;
2040 	int v_start = 0;
2041 
2042 	/* only one q_vector if MSI-X is disabled. */
2043 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2044 		q_vectors = 1;
2045 
2046 	/*
2047 	 * If we don't have enough vectors for a 1-to-1 mapping, we'll have to
2048 	 * group them so there are multiple queues per vector.
2049 	 *
2050 	 * Re-adjusting *qpv takes care of the remainder.
2051 	 */
2052 	for (; v_start < q_vectors && rxr_remaining; v_start++) {
2053 		int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_start);
2054 		for (; rqpv; rqpv--, rxr_idx++, rxr_remaining--)
2055 			map_vector_to_rxq(adapter, v_start, rxr_idx);
2056 	}
2057 
2058 	/*
2059 	 * If there are not enough q_vectors for each ring to have it's own
2060 	 * vector then we must pair up Rx/Tx on a each vector
2061 	 */
2062 	if ((v_start + txr_remaining) > q_vectors)
2063 		v_start = 0;
2064 
2065 	for (; v_start < q_vectors && txr_remaining; v_start++) {
2066 		int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_start);
2067 		for (; tqpv; tqpv--, txr_idx++, txr_remaining--)
2068 			map_vector_to_txq(adapter, v_start, txr_idx);
2069 	}
2070 }
2071 
2072 /**
2073  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2074  * @adapter: board private structure
2075  *
2076  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2077  * interrupts from the kernel.
2078  **/
2079 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2080 {
2081 	struct net_device *netdev = adapter->netdev;
2082 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2083 	int vector, err;
2084 	int ri = 0, ti = 0;
2085 
2086 	for (vector = 0; vector < q_vectors; vector++) {
2087 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2088 		struct msix_entry *entry = &adapter->msix_entries[vector];
2089 
2090 		if (q_vector->tx.ring && q_vector->rx.ring) {
2091 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2092 				 "%s-%s-%d", netdev->name, "TxRx", ri++);
2093 			ti++;
2094 		} else if (q_vector->rx.ring) {
2095 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2096 				 "%s-%s-%d", netdev->name, "rx", ri++);
2097 		} else if (q_vector->tx.ring) {
2098 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2099 				 "%s-%s-%d", netdev->name, "tx", ti++);
2100 		} else {
2101 			/* skip this unused q_vector */
2102 			continue;
2103 		}
2104 		err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2105 				  q_vector->name, q_vector);
2106 		if (err) {
2107 			e_err(probe, "request_irq failed for MSIX interrupt "
2108 			      "Error: %d\n", err);
2109 			goto free_queue_irqs;
2110 		}
2111 		/* If Flow Director is enabled, set interrupt affinity */
2112 		if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2113 			/* assign the mask for this irq */
2114 			irq_set_affinity_hint(entry->vector,
2115 					      q_vector->affinity_mask);
2116 		}
2117 	}
2118 
2119 	err = request_irq(adapter->msix_entries[vector].vector,
2120 			  ixgbe_msix_other, 0, netdev->name, adapter);
2121 	if (err) {
2122 		e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2123 		goto free_queue_irqs;
2124 	}
2125 
2126 	return 0;
2127 
2128 free_queue_irqs:
2129 	while (vector) {
2130 		vector--;
2131 		irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2132 				      NULL);
2133 		free_irq(adapter->msix_entries[vector].vector,
2134 			 adapter->q_vector[vector]);
2135 	}
2136 	adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2137 	pci_disable_msix(adapter->pdev);
2138 	kfree(adapter->msix_entries);
2139 	adapter->msix_entries = NULL;
2140 	return err;
2141 }
2142 
2143 /**
2144  * ixgbe_intr - legacy mode Interrupt Handler
2145  * @irq: interrupt number
2146  * @data: pointer to a network interface device structure
2147  **/
2148 static irqreturn_t ixgbe_intr(int irq, void *data)
2149 {
2150 	struct ixgbe_adapter *adapter = data;
2151 	struct ixgbe_hw *hw = &adapter->hw;
2152 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2153 	u32 eicr;
2154 
2155 	/*
2156 	 * Workaround for silicon errata on 82598.  Mask the interrupts
2157 	 * before the read of EICR.
2158 	 */
2159 	IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2160 
2161 	/* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2162 	 * therefore no explicit interrupt disable is necessary */
2163 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2164 	if (!eicr) {
2165 		/*
2166 		 * shared interrupt alert!
2167 		 * make sure interrupts are enabled because the read will
2168 		 * have disabled interrupts due to EIAM
2169 		 * finish the workaround of silicon errata on 82598.  Unmask
2170 		 * the interrupt that we masked before the EICR read.
2171 		 */
2172 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
2173 			ixgbe_irq_enable(adapter, true, true);
2174 		return IRQ_NONE;	/* Not our interrupt */
2175 	}
2176 
2177 	if (eicr & IXGBE_EICR_LSC)
2178 		ixgbe_check_lsc(adapter);
2179 
2180 	switch (hw->mac.type) {
2181 	case ixgbe_mac_82599EB:
2182 		ixgbe_check_sfp_event(adapter, eicr);
2183 		/* Fall through */
2184 	case ixgbe_mac_X540:
2185 		if (eicr & IXGBE_EICR_ECC)
2186 			e_info(link, "Received unrecoverable ECC err, please "
2187 				     "reboot\n");
2188 		ixgbe_check_overtemp_event(adapter, eicr);
2189 		break;
2190 	default:
2191 		break;
2192 	}
2193 
2194 	ixgbe_check_fan_failure(adapter, eicr);
2195 
2196 	if (napi_schedule_prep(&(q_vector->napi))) {
2197 		/* would disable interrupts here but EIAM disabled it */
2198 		__napi_schedule(&(q_vector->napi));
2199 	}
2200 
2201 	/*
2202 	 * re-enable link(maybe) and non-queue interrupts, no flush.
2203 	 * ixgbe_poll will re-enable the queue interrupts
2204 	 */
2205 
2206 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
2207 		ixgbe_irq_enable(adapter, false, false);
2208 
2209 	return IRQ_HANDLED;
2210 }
2211 
2212 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2213 {
2214 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2215 	int i;
2216 
2217 	/* legacy and MSI only use one vector */
2218 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2219 		q_vectors = 1;
2220 
2221 	for (i = 0; i < adapter->num_rx_queues; i++) {
2222 		adapter->rx_ring[i]->q_vector = NULL;
2223 		adapter->rx_ring[i]->next = NULL;
2224 	}
2225 	for (i = 0; i < adapter->num_tx_queues; i++) {
2226 		adapter->tx_ring[i]->q_vector = NULL;
2227 		adapter->tx_ring[i]->next = NULL;
2228 	}
2229 
2230 	for (i = 0; i < q_vectors; i++) {
2231 		struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2232 		memset(&q_vector->rx, 0, sizeof(struct ixgbe_ring_container));
2233 		memset(&q_vector->tx, 0, sizeof(struct ixgbe_ring_container));
2234 	}
2235 }
2236 
2237 /**
2238  * ixgbe_request_irq - initialize interrupts
2239  * @adapter: board private structure
2240  *
2241  * Attempts to configure interrupts using the best available
2242  * capabilities of the hardware and kernel.
2243  **/
2244 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2245 {
2246 	struct net_device *netdev = adapter->netdev;
2247 	int err;
2248 
2249 	/* map all of the rings to the q_vectors */
2250 	ixgbe_map_rings_to_vectors(adapter);
2251 
2252 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2253 		err = ixgbe_request_msix_irqs(adapter);
2254 	else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2255 		err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2256 				  netdev->name, adapter);
2257 	else
2258 		err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2259 				  netdev->name, adapter);
2260 
2261 	if (err) {
2262 		e_err(probe, "request_irq failed, Error %d\n", err);
2263 
2264 		/* place q_vectors and rings back into a known good state */
2265 		ixgbe_reset_q_vectors(adapter);
2266 	}
2267 
2268 	return err;
2269 }
2270 
2271 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2272 {
2273 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2274 		int i, q_vectors;
2275 
2276 		q_vectors = adapter->num_msix_vectors;
2277 		i = q_vectors - 1;
2278 		free_irq(adapter->msix_entries[i].vector, adapter);
2279 		i--;
2280 
2281 		for (; i >= 0; i--) {
2282 			/* free only the irqs that were actually requested */
2283 			if (!adapter->q_vector[i]->rx.ring &&
2284 			    !adapter->q_vector[i]->tx.ring)
2285 				continue;
2286 
2287 			/* clear the affinity_mask in the IRQ descriptor */
2288 			irq_set_affinity_hint(adapter->msix_entries[i].vector,
2289 					      NULL);
2290 
2291 			free_irq(adapter->msix_entries[i].vector,
2292 				 adapter->q_vector[i]);
2293 		}
2294 	} else {
2295 		free_irq(adapter->pdev->irq, adapter);
2296 	}
2297 
2298 	/* clear q_vector state information */
2299 	ixgbe_reset_q_vectors(adapter);
2300 }
2301 
2302 /**
2303  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2304  * @adapter: board private structure
2305  **/
2306 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2307 {
2308 	switch (adapter->hw.mac.type) {
2309 	case ixgbe_mac_82598EB:
2310 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2311 		break;
2312 	case ixgbe_mac_82599EB:
2313 	case ixgbe_mac_X540:
2314 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2315 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2316 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2317 		break;
2318 	default:
2319 		break;
2320 	}
2321 	IXGBE_WRITE_FLUSH(&adapter->hw);
2322 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2323 		int i;
2324 		for (i = 0; i < adapter->num_msix_vectors; i++)
2325 			synchronize_irq(adapter->msix_entries[i].vector);
2326 	} else {
2327 		synchronize_irq(adapter->pdev->irq);
2328 	}
2329 }
2330 
2331 /**
2332  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2333  *
2334  **/
2335 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2336 {
2337 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2338 
2339 	/* rx/tx vector */
2340 	if (adapter->rx_itr_setting == 1)
2341 		q_vector->itr = IXGBE_20K_ITR;
2342 	else
2343 		q_vector->itr = adapter->rx_itr_setting;
2344 
2345 	ixgbe_write_eitr(q_vector);
2346 
2347 	ixgbe_set_ivar(adapter, 0, 0, 0);
2348 	ixgbe_set_ivar(adapter, 1, 0, 0);
2349 
2350 	e_info(hw, "Legacy interrupt IVAR setup done\n");
2351 }
2352 
2353 /**
2354  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2355  * @adapter: board private structure
2356  * @ring: structure containing ring specific data
2357  *
2358  * Configure the Tx descriptor ring after a reset.
2359  **/
2360 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2361 			     struct ixgbe_ring *ring)
2362 {
2363 	struct ixgbe_hw *hw = &adapter->hw;
2364 	u64 tdba = ring->dma;
2365 	int wait_loop = 10;
2366 	u32 txdctl = IXGBE_TXDCTL_ENABLE;
2367 	u8 reg_idx = ring->reg_idx;
2368 
2369 	/* disable queue to avoid issues while updating state */
2370 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
2371 	IXGBE_WRITE_FLUSH(hw);
2372 
2373 	IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2374 			(tdba & DMA_BIT_MASK(32)));
2375 	IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2376 	IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2377 			ring->count * sizeof(union ixgbe_adv_tx_desc));
2378 	IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2379 	IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2380 	ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2381 
2382 	/*
2383 	 * set WTHRESH to encourage burst writeback, it should not be set
2384 	 * higher than 1 when ITR is 0 as it could cause false TX hangs
2385 	 *
2386 	 * In order to avoid issues WTHRESH + PTHRESH should always be equal
2387 	 * to or less than the number of on chip descriptors, which is
2388 	 * currently 40.
2389 	 */
2390 	if (!adapter->tx_itr_setting || !adapter->rx_itr_setting)
2391 		txdctl |= (1 << 16);	/* WTHRESH = 1 */
2392 	else
2393 		txdctl |= (8 << 16);	/* WTHRESH = 8 */
2394 
2395 	/* PTHRESH=32 is needed to avoid a Tx hang with DFP enabled. */
2396 	txdctl |= (1 << 8) |	/* HTHRESH = 1 */
2397 		   32;		/* PTHRESH = 32 */
2398 
2399 	/* reinitialize flowdirector state */
2400 	if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2401 	    adapter->atr_sample_rate) {
2402 		ring->atr_sample_rate = adapter->atr_sample_rate;
2403 		ring->atr_count = 0;
2404 		set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2405 	} else {
2406 		ring->atr_sample_rate = 0;
2407 	}
2408 
2409 	clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2410 
2411 	/* enable queue */
2412 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2413 
2414 	/* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2415 	if (hw->mac.type == ixgbe_mac_82598EB &&
2416 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2417 		return;
2418 
2419 	/* poll to verify queue is enabled */
2420 	do {
2421 		usleep_range(1000, 2000);
2422 		txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2423 	} while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2424 	if (!wait_loop)
2425 		e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2426 }
2427 
2428 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2429 {
2430 	struct ixgbe_hw *hw = &adapter->hw;
2431 	u32 rttdcs;
2432 	u32 reg;
2433 	u8 tcs = netdev_get_num_tc(adapter->netdev);
2434 
2435 	if (hw->mac.type == ixgbe_mac_82598EB)
2436 		return;
2437 
2438 	/* disable the arbiter while setting MTQC */
2439 	rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2440 	rttdcs |= IXGBE_RTTDCS_ARBDIS;
2441 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2442 
2443 	/* set transmit pool layout */
2444 	switch (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2445 	case (IXGBE_FLAG_SRIOV_ENABLED):
2446 		IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2447 				(IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2448 		break;
2449 	default:
2450 		if (!tcs)
2451 			reg = IXGBE_MTQC_64Q_1PB;
2452 		else if (tcs <= 4)
2453 			reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2454 		else
2455 			reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2456 
2457 		IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
2458 
2459 		/* Enable Security TX Buffer IFG for multiple pb */
2460 		if (tcs) {
2461 			reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
2462 			reg |= IXGBE_SECTX_DCB;
2463 			IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
2464 		}
2465 		break;
2466 	}
2467 
2468 	/* re-enable the arbiter */
2469 	rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2470 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2471 }
2472 
2473 /**
2474  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2475  * @adapter: board private structure
2476  *
2477  * Configure the Tx unit of the MAC after a reset.
2478  **/
2479 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2480 {
2481 	struct ixgbe_hw *hw = &adapter->hw;
2482 	u32 dmatxctl;
2483 	u32 i;
2484 
2485 	ixgbe_setup_mtqc(adapter);
2486 
2487 	if (hw->mac.type != ixgbe_mac_82598EB) {
2488 		/* DMATXCTL.EN must be before Tx queues are enabled */
2489 		dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2490 		dmatxctl |= IXGBE_DMATXCTL_TE;
2491 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2492 	}
2493 
2494 	/* Setup the HW Tx Head and Tail descriptor pointers */
2495 	for (i = 0; i < adapter->num_tx_queues; i++)
2496 		ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2497 }
2498 
2499 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2500 
2501 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2502 				   struct ixgbe_ring *rx_ring)
2503 {
2504 	u32 srrctl;
2505 	u8 reg_idx = rx_ring->reg_idx;
2506 
2507 	switch (adapter->hw.mac.type) {
2508 	case ixgbe_mac_82598EB: {
2509 		struct ixgbe_ring_feature *feature = adapter->ring_feature;
2510 		const int mask = feature[RING_F_RSS].mask;
2511 		reg_idx = reg_idx & mask;
2512 	}
2513 		break;
2514 	case ixgbe_mac_82599EB:
2515 	case ixgbe_mac_X540:
2516 	default:
2517 		break;
2518 	}
2519 
2520 	srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx));
2521 
2522 	srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2523 	srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2524 	if (adapter->num_vfs)
2525 		srrctl |= IXGBE_SRRCTL_DROP_EN;
2526 
2527 	srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2528 		  IXGBE_SRRCTL_BSIZEHDR_MASK;
2529 
2530 	if (ring_is_ps_enabled(rx_ring)) {
2531 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2532 		srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2533 #else
2534 		srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2535 #endif
2536 		srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2537 	} else {
2538 		srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2539 			  IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2540 		srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2541 	}
2542 
2543 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx), srrctl);
2544 }
2545 
2546 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2547 {
2548 	struct ixgbe_hw *hw = &adapter->hw;
2549 	static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2550 			  0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2551 			  0x6A3E67EA, 0x14364D17, 0x3BED200D};
2552 	u32 mrqc = 0, reta = 0;
2553 	u32 rxcsum;
2554 	int i, j;
2555 	u8 tcs = netdev_get_num_tc(adapter->netdev);
2556 	int maxq = adapter->ring_feature[RING_F_RSS].indices;
2557 
2558 	if (tcs)
2559 		maxq = min(maxq, adapter->num_tx_queues / tcs);
2560 
2561 	/* Fill out hash function seeds */
2562 	for (i = 0; i < 10; i++)
2563 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2564 
2565 	/* Fill out redirection table */
2566 	for (i = 0, j = 0; i < 128; i++, j++) {
2567 		if (j == maxq)
2568 			j = 0;
2569 		/* reta = 4-byte sliding window of
2570 		 * 0x00..(indices-1)(indices-1)00..etc. */
2571 		reta = (reta << 8) | (j * 0x11);
2572 		if ((i & 3) == 3)
2573 			IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2574 	}
2575 
2576 	/* Disable indicating checksum in descriptor, enables RSS hash */
2577 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2578 	rxcsum |= IXGBE_RXCSUM_PCSD;
2579 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2580 
2581 	if (adapter->hw.mac.type == ixgbe_mac_82598EB &&
2582 	    (adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
2583 		mrqc = IXGBE_MRQC_RSSEN;
2584 	} else {
2585 		int mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2586 					     | IXGBE_FLAG_SRIOV_ENABLED);
2587 
2588 		switch (mask) {
2589 		case (IXGBE_FLAG_RSS_ENABLED):
2590 			if (!tcs)
2591 				mrqc = IXGBE_MRQC_RSSEN;
2592 			else if (tcs <= 4)
2593 				mrqc = IXGBE_MRQC_RTRSS4TCEN;
2594 			else
2595 				mrqc = IXGBE_MRQC_RTRSS8TCEN;
2596 			break;
2597 		case (IXGBE_FLAG_SRIOV_ENABLED):
2598 			mrqc = IXGBE_MRQC_VMDQEN;
2599 			break;
2600 		default:
2601 			break;
2602 		}
2603 	}
2604 
2605 	/* Perform hash on these packet types */
2606 	mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2607 	      | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2608 	      | IXGBE_MRQC_RSS_FIELD_IPV6
2609 	      | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2610 
2611 	IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2612 }
2613 
2614 /**
2615  * ixgbe_configure_rscctl - enable RSC for the indicated ring
2616  * @adapter:    address of board private structure
2617  * @index:      index of ring to set
2618  **/
2619 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2620 				   struct ixgbe_ring *ring)
2621 {
2622 	struct ixgbe_hw *hw = &adapter->hw;
2623 	u32 rscctrl;
2624 	int rx_buf_len;
2625 	u8 reg_idx = ring->reg_idx;
2626 
2627 	if (!ring_is_rsc_enabled(ring))
2628 		return;
2629 
2630 	rx_buf_len = ring->rx_buf_len;
2631 	rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2632 	rscctrl |= IXGBE_RSCCTL_RSCEN;
2633 	/*
2634 	 * we must limit the number of descriptors so that the
2635 	 * total size of max desc * buf_len is not greater
2636 	 * than 65535
2637 	 */
2638 	if (ring_is_ps_enabled(ring)) {
2639 #if (MAX_SKB_FRAGS > 16)
2640 		rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2641 #elif (MAX_SKB_FRAGS > 8)
2642 		rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2643 #elif (MAX_SKB_FRAGS > 4)
2644 		rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2645 #else
2646 		rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2647 #endif
2648 	} else {
2649 		if (rx_buf_len < IXGBE_RXBUFFER_4K)
2650 			rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2651 		else if (rx_buf_len < IXGBE_RXBUFFER_8K)
2652 			rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2653 		else
2654 			rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2655 	}
2656 	IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2657 }
2658 
2659 /**
2660  *  ixgbe_set_uta - Set unicast filter table address
2661  *  @adapter: board private structure
2662  *
2663  *  The unicast table address is a register array of 32-bit registers.
2664  *  The table is meant to be used in a way similar to how the MTA is used
2665  *  however due to certain limitations in the hardware it is necessary to
2666  *  set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2667  *  enable bit to allow vlan tag stripping when promiscuous mode is enabled
2668  **/
2669 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2670 {
2671 	struct ixgbe_hw *hw = &adapter->hw;
2672 	int i;
2673 
2674 	/* The UTA table only exists on 82599 hardware and newer */
2675 	if (hw->mac.type < ixgbe_mac_82599EB)
2676 		return;
2677 
2678 	/* we only need to do this if VMDq is enabled */
2679 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2680 		return;
2681 
2682 	for (i = 0; i < 128; i++)
2683 		IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
2684 }
2685 
2686 #define IXGBE_MAX_RX_DESC_POLL 10
2687 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2688 				       struct ixgbe_ring *ring)
2689 {
2690 	struct ixgbe_hw *hw = &adapter->hw;
2691 	int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2692 	u32 rxdctl;
2693 	u8 reg_idx = ring->reg_idx;
2694 
2695 	/* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2696 	if (hw->mac.type == ixgbe_mac_82598EB &&
2697 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2698 		return;
2699 
2700 	do {
2701 		usleep_range(1000, 2000);
2702 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2703 	} while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
2704 
2705 	if (!wait_loop) {
2706 		e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
2707 		      "the polling period\n", reg_idx);
2708 	}
2709 }
2710 
2711 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
2712 			    struct ixgbe_ring *ring)
2713 {
2714 	struct ixgbe_hw *hw = &adapter->hw;
2715 	int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2716 	u32 rxdctl;
2717 	u8 reg_idx = ring->reg_idx;
2718 
2719 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2720 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
2721 
2722 	/* write value back with RXDCTL.ENABLE bit cleared */
2723 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2724 
2725 	if (hw->mac.type == ixgbe_mac_82598EB &&
2726 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2727 		return;
2728 
2729 	/* the hardware may take up to 100us to really disable the rx queue */
2730 	do {
2731 		udelay(10);
2732 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2733 	} while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
2734 
2735 	if (!wait_loop) {
2736 		e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
2737 		      "the polling period\n", reg_idx);
2738 	}
2739 }
2740 
2741 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
2742 			     struct ixgbe_ring *ring)
2743 {
2744 	struct ixgbe_hw *hw = &adapter->hw;
2745 	u64 rdba = ring->dma;
2746 	u32 rxdctl;
2747 	u8 reg_idx = ring->reg_idx;
2748 
2749 	/* disable queue to avoid issues while updating state */
2750 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2751 	ixgbe_disable_rx_queue(adapter, ring);
2752 
2753 	IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
2754 	IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
2755 	IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
2756 			ring->count * sizeof(union ixgbe_adv_rx_desc));
2757 	IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
2758 	IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
2759 	ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
2760 
2761 	ixgbe_configure_srrctl(adapter, ring);
2762 	ixgbe_configure_rscctl(adapter, ring);
2763 
2764 	/* If operating in IOV mode set RLPML for X540 */
2765 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
2766 	    hw->mac.type == ixgbe_mac_X540) {
2767 		rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
2768 		rxdctl |= ((ring->netdev->mtu + ETH_HLEN +
2769 			    ETH_FCS_LEN + VLAN_HLEN) | IXGBE_RXDCTL_RLPML_EN);
2770 	}
2771 
2772 	if (hw->mac.type == ixgbe_mac_82598EB) {
2773 		/*
2774 		 * enable cache line friendly hardware writes:
2775 		 * PTHRESH=32 descriptors (half the internal cache),
2776 		 * this also removes ugly rx_no_buffer_count increment
2777 		 * HTHRESH=4 descriptors (to minimize latency on fetch)
2778 		 * WTHRESH=8 burst writeback up to two cache lines
2779 		 */
2780 		rxdctl &= ~0x3FFFFF;
2781 		rxdctl |=  0x080420;
2782 	}
2783 
2784 	/* enable receive descriptor ring */
2785 	rxdctl |= IXGBE_RXDCTL_ENABLE;
2786 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2787 
2788 	ixgbe_rx_desc_queue_enable(adapter, ring);
2789 	ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
2790 }
2791 
2792 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
2793 {
2794 	struct ixgbe_hw *hw = &adapter->hw;
2795 	int p;
2796 
2797 	/* PSRTYPE must be initialized in non 82598 adapters */
2798 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2799 		      IXGBE_PSRTYPE_UDPHDR |
2800 		      IXGBE_PSRTYPE_IPV4HDR |
2801 		      IXGBE_PSRTYPE_L2HDR |
2802 		      IXGBE_PSRTYPE_IPV6HDR;
2803 
2804 	if (hw->mac.type == ixgbe_mac_82598EB)
2805 		return;
2806 
2807 	if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
2808 		psrtype |= (adapter->num_rx_queues_per_pool << 29);
2809 
2810 	for (p = 0; p < adapter->num_rx_pools; p++)
2811 		IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
2812 				psrtype);
2813 }
2814 
2815 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
2816 {
2817 	struct ixgbe_hw *hw = &adapter->hw;
2818 	u32 gcr_ext;
2819 	u32 vt_reg_bits;
2820 	u32 reg_offset, vf_shift;
2821 	u32 vmdctl;
2822 	int i;
2823 
2824 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2825 		return;
2826 
2827 	vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2828 	vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
2829 	vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
2830 	IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2831 
2832 	vf_shift = adapter->num_vfs % 32;
2833 	reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
2834 
2835 	/* Enable only the PF's pool for Tx/Rx */
2836 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2837 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
2838 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2839 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
2840 	IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2841 
2842 	/* Map PF MAC address in RAR Entry 0 to first pool following VFs */
2843 	hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2844 
2845 	/*
2846 	 * Set up VF register offsets for selected VT Mode,
2847 	 * i.e. 32 or 64 VFs for SR-IOV
2848 	 */
2849 	gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2850 	gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
2851 	gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
2852 	IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
2853 
2854 	/* enable Tx loopback for VF/PF communication */
2855 	IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2856 	/* Enable MAC Anti-Spoofing */
2857 	hw->mac.ops.set_mac_anti_spoofing(hw,
2858 					   (adapter->num_vfs != 0),
2859 					  adapter->num_vfs);
2860 	/* For VFs that have spoof checking turned off */
2861 	for (i = 0; i < adapter->num_vfs; i++) {
2862 		if (!adapter->vfinfo[i].spoofchk_enabled)
2863 			ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
2864 	}
2865 }
2866 
2867 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
2868 {
2869 	struct ixgbe_hw *hw = &adapter->hw;
2870 	struct net_device *netdev = adapter->netdev;
2871 	int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2872 	int rx_buf_len;
2873 	struct ixgbe_ring *rx_ring;
2874 	int i;
2875 	u32 mhadd, hlreg0;
2876 
2877 	/* Decide whether to use packet split mode or not */
2878 	/* On by default */
2879 	adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2880 
2881 	/* Do not use packet split if we're in SR-IOV Mode */
2882 	if (adapter->num_vfs)
2883 		adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
2884 
2885 	/* Disable packet split due to 82599 erratum #45 */
2886 	if (hw->mac.type == ixgbe_mac_82599EB)
2887 		adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
2888 
2889 #ifdef IXGBE_FCOE
2890 	/* adjust max frame to be able to do baby jumbo for FCoE */
2891 	if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
2892 	    (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2893 		max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2894 
2895 #endif /* IXGBE_FCOE */
2896 	mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2897 	if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2898 		mhadd &= ~IXGBE_MHADD_MFS_MASK;
2899 		mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2900 
2901 		IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2902 	}
2903 
2904 	/* MHADD will allow an extra 4 bytes past for vlan tagged frames */
2905 	max_frame += VLAN_HLEN;
2906 
2907 	/* Set the RX buffer length according to the mode */
2908 	if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2909 		rx_buf_len = IXGBE_RX_HDR_SIZE;
2910 	} else {
2911 		if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2912 		    (netdev->mtu <= ETH_DATA_LEN))
2913 			rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2914 		/*
2915 		 * Make best use of allocation by using all but 1K of a
2916 		 * power of 2 allocation that will be used for skb->head.
2917 		 */
2918 		else if (max_frame <= IXGBE_RXBUFFER_3K)
2919 			rx_buf_len = IXGBE_RXBUFFER_3K;
2920 		else if (max_frame <= IXGBE_RXBUFFER_7K)
2921 			rx_buf_len = IXGBE_RXBUFFER_7K;
2922 		else if (max_frame <= IXGBE_RXBUFFER_15K)
2923 			rx_buf_len = IXGBE_RXBUFFER_15K;
2924 		else
2925 			rx_buf_len = IXGBE_MAX_RXBUFFER;
2926 	}
2927 
2928 	hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2929 	/* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
2930 	hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2931 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2932 
2933 	/*
2934 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
2935 	 * the Base and Length of the Rx Descriptor Ring
2936 	 */
2937 	for (i = 0; i < adapter->num_rx_queues; i++) {
2938 		rx_ring = adapter->rx_ring[i];
2939 		rx_ring->rx_buf_len = rx_buf_len;
2940 
2941 		if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2942 			set_ring_ps_enabled(rx_ring);
2943 		else
2944 			clear_ring_ps_enabled(rx_ring);
2945 
2946 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
2947 			set_ring_rsc_enabled(rx_ring);
2948 		else
2949 			clear_ring_rsc_enabled(rx_ring);
2950 
2951 #ifdef IXGBE_FCOE
2952 		if (netdev->features & NETIF_F_FCOE_MTU) {
2953 			struct ixgbe_ring_feature *f;
2954 			f = &adapter->ring_feature[RING_F_FCOE];
2955 			if ((i >= f->mask) && (i < f->mask + f->indices)) {
2956 				clear_ring_ps_enabled(rx_ring);
2957 				if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2958 					rx_ring->rx_buf_len =
2959 						IXGBE_FCOE_JUMBO_FRAME_SIZE;
2960 			} else if (!ring_is_rsc_enabled(rx_ring) &&
2961 				   !ring_is_ps_enabled(rx_ring)) {
2962 				rx_ring->rx_buf_len =
2963 						IXGBE_FCOE_JUMBO_FRAME_SIZE;
2964 			}
2965 		}
2966 #endif /* IXGBE_FCOE */
2967 	}
2968 }
2969 
2970 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
2971 {
2972 	struct ixgbe_hw *hw = &adapter->hw;
2973 	u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2974 
2975 	switch (hw->mac.type) {
2976 	case ixgbe_mac_82598EB:
2977 		/*
2978 		 * For VMDq support of different descriptor types or
2979 		 * buffer sizes through the use of multiple SRRCTL
2980 		 * registers, RDRXCTL.MVMEN must be set to 1
2981 		 *
2982 		 * also, the manual doesn't mention it clearly but DCA hints
2983 		 * will only use queue 0's tags unless this bit is set.  Side
2984 		 * effects of setting this bit are only that SRRCTL must be
2985 		 * fully programmed [0..15]
2986 		 */
2987 		rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2988 		break;
2989 	case ixgbe_mac_82599EB:
2990 	case ixgbe_mac_X540:
2991 		/* Disable RSC for ACK packets */
2992 		IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2993 		   (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2994 		rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2995 		/* hardware requires some bits to be set by default */
2996 		rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
2997 		rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
2998 		break;
2999 	default:
3000 		/* We should do nothing since we don't know this hardware */
3001 		return;
3002 	}
3003 
3004 	IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3005 }
3006 
3007 /**
3008  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3009  * @adapter: board private structure
3010  *
3011  * Configure the Rx unit of the MAC after a reset.
3012  **/
3013 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3014 {
3015 	struct ixgbe_hw *hw = &adapter->hw;
3016 	int i;
3017 	u32 rxctrl;
3018 
3019 	/* disable receives while setting up the descriptors */
3020 	rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3021 	IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3022 
3023 	ixgbe_setup_psrtype(adapter);
3024 	ixgbe_setup_rdrxctl(adapter);
3025 
3026 	/* Program registers for the distribution of queues */
3027 	ixgbe_setup_mrqc(adapter);
3028 
3029 	ixgbe_set_uta(adapter);
3030 
3031 	/* set_rx_buffer_len must be called before ring initialization */
3032 	ixgbe_set_rx_buffer_len(adapter);
3033 
3034 	/*
3035 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
3036 	 * the Base and Length of the Rx Descriptor Ring
3037 	 */
3038 	for (i = 0; i < adapter->num_rx_queues; i++)
3039 		ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3040 
3041 	/* disable drop enable for 82598 parts */
3042 	if (hw->mac.type == ixgbe_mac_82598EB)
3043 		rxctrl |= IXGBE_RXCTRL_DMBYPS;
3044 
3045 	/* enable all receives */
3046 	rxctrl |= IXGBE_RXCTRL_RXEN;
3047 	hw->mac.ops.enable_rx_dma(hw, rxctrl);
3048 }
3049 
3050 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3051 {
3052 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3053 	struct ixgbe_hw *hw = &adapter->hw;
3054 	int pool_ndx = adapter->num_vfs;
3055 
3056 	/* add VID to filter table */
3057 	hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3058 	set_bit(vid, adapter->active_vlans);
3059 
3060 	return 0;
3061 }
3062 
3063 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3064 {
3065 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3066 	struct ixgbe_hw *hw = &adapter->hw;
3067 	int pool_ndx = adapter->num_vfs;
3068 
3069 	/* remove VID from filter table */
3070 	hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3071 	clear_bit(vid, adapter->active_vlans);
3072 
3073 	return 0;
3074 }
3075 
3076 /**
3077  * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3078  * @adapter: driver data
3079  */
3080 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3081 {
3082 	struct ixgbe_hw *hw = &adapter->hw;
3083 	u32 vlnctrl;
3084 
3085 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3086 	vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3087 	IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3088 }
3089 
3090 /**
3091  * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3092  * @adapter: driver data
3093  */
3094 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3095 {
3096 	struct ixgbe_hw *hw = &adapter->hw;
3097 	u32 vlnctrl;
3098 
3099 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3100 	vlnctrl |= IXGBE_VLNCTRL_VFE;
3101 	vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3102 	IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3103 }
3104 
3105 /**
3106  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3107  * @adapter: driver data
3108  */
3109 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3110 {
3111 	struct ixgbe_hw *hw = &adapter->hw;
3112 	u32 vlnctrl;
3113 	int i, j;
3114 
3115 	switch (hw->mac.type) {
3116 	case ixgbe_mac_82598EB:
3117 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3118 		vlnctrl &= ~IXGBE_VLNCTRL_VME;
3119 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3120 		break;
3121 	case ixgbe_mac_82599EB:
3122 	case ixgbe_mac_X540:
3123 		for (i = 0; i < adapter->num_rx_queues; i++) {
3124 			j = adapter->rx_ring[i]->reg_idx;
3125 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3126 			vlnctrl &= ~IXGBE_RXDCTL_VME;
3127 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3128 		}
3129 		break;
3130 	default:
3131 		break;
3132 	}
3133 }
3134 
3135 /**
3136  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3137  * @adapter: driver data
3138  */
3139 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3140 {
3141 	struct ixgbe_hw *hw = &adapter->hw;
3142 	u32 vlnctrl;
3143 	int i, j;
3144 
3145 	switch (hw->mac.type) {
3146 	case ixgbe_mac_82598EB:
3147 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3148 		vlnctrl |= IXGBE_VLNCTRL_VME;
3149 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3150 		break;
3151 	case ixgbe_mac_82599EB:
3152 	case ixgbe_mac_X540:
3153 		for (i = 0; i < adapter->num_rx_queues; i++) {
3154 			j = adapter->rx_ring[i]->reg_idx;
3155 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3156 			vlnctrl |= IXGBE_RXDCTL_VME;
3157 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3158 		}
3159 		break;
3160 	default:
3161 		break;
3162 	}
3163 }
3164 
3165 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3166 {
3167 	u16 vid;
3168 
3169 	ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3170 
3171 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3172 		ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3173 }
3174 
3175 /**
3176  * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3177  * @netdev: network interface device structure
3178  *
3179  * Writes unicast address list to the RAR table.
3180  * Returns: -ENOMEM on failure/insufficient address space
3181  *                0 on no addresses written
3182  *                X on writing X addresses to the RAR table
3183  **/
3184 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3185 {
3186 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3187 	struct ixgbe_hw *hw = &adapter->hw;
3188 	unsigned int vfn = adapter->num_vfs;
3189 	unsigned int rar_entries = IXGBE_MAX_PF_MACVLANS;
3190 	int count = 0;
3191 
3192 	/* return ENOMEM indicating insufficient memory for addresses */
3193 	if (netdev_uc_count(netdev) > rar_entries)
3194 		return -ENOMEM;
3195 
3196 	if (!netdev_uc_empty(netdev) && rar_entries) {
3197 		struct netdev_hw_addr *ha;
3198 		/* return error if we do not support writing to RAR table */
3199 		if (!hw->mac.ops.set_rar)
3200 			return -ENOMEM;
3201 
3202 		netdev_for_each_uc_addr(ha, netdev) {
3203 			if (!rar_entries)
3204 				break;
3205 			hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3206 					    vfn, IXGBE_RAH_AV);
3207 			count++;
3208 		}
3209 	}
3210 	/* write the addresses in reverse order to avoid write combining */
3211 	for (; rar_entries > 0 ; rar_entries--)
3212 		hw->mac.ops.clear_rar(hw, rar_entries);
3213 
3214 	return count;
3215 }
3216 
3217 /**
3218  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3219  * @netdev: network interface device structure
3220  *
3221  * The set_rx_method entry point is called whenever the unicast/multicast
3222  * address list or the network interface flags are updated.  This routine is
3223  * responsible for configuring the hardware for proper unicast, multicast and
3224  * promiscuous mode.
3225  **/
3226 void ixgbe_set_rx_mode(struct net_device *netdev)
3227 {
3228 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3229 	struct ixgbe_hw *hw = &adapter->hw;
3230 	u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3231 	int count;
3232 
3233 	/* Check for Promiscuous and All Multicast modes */
3234 
3235 	fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3236 
3237 	/* set all bits that we expect to always be set */
3238 	fctrl |= IXGBE_FCTRL_BAM;
3239 	fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3240 	fctrl |= IXGBE_FCTRL_PMCF;
3241 
3242 	/* clear the bits we are changing the status of */
3243 	fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3244 
3245 	if (netdev->flags & IFF_PROMISC) {
3246 		hw->addr_ctrl.user_set_promisc = true;
3247 		fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3248 		vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3249 		/* don't hardware filter vlans in promisc mode */
3250 		ixgbe_vlan_filter_disable(adapter);
3251 	} else {
3252 		if (netdev->flags & IFF_ALLMULTI) {
3253 			fctrl |= IXGBE_FCTRL_MPE;
3254 			vmolr |= IXGBE_VMOLR_MPE;
3255 		} else {
3256 			/*
3257 			 * Write addresses to the MTA, if the attempt fails
3258 			 * then we should just turn on promiscuous mode so
3259 			 * that we can at least receive multicast traffic
3260 			 */
3261 			hw->mac.ops.update_mc_addr_list(hw, netdev);
3262 			vmolr |= IXGBE_VMOLR_ROMPE;
3263 		}
3264 		ixgbe_vlan_filter_enable(adapter);
3265 		hw->addr_ctrl.user_set_promisc = false;
3266 		/*
3267 		 * Write addresses to available RAR registers, if there is not
3268 		 * sufficient space to store all the addresses then enable
3269 		 * unicast promiscuous mode
3270 		 */
3271 		count = ixgbe_write_uc_addr_list(netdev);
3272 		if (count < 0) {
3273 			fctrl |= IXGBE_FCTRL_UPE;
3274 			vmolr |= IXGBE_VMOLR_ROPE;
3275 		}
3276 	}
3277 
3278 	if (adapter->num_vfs) {
3279 		ixgbe_restore_vf_multicasts(adapter);
3280 		vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3281 			 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3282 			   IXGBE_VMOLR_ROPE);
3283 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3284 	}
3285 
3286 	IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3287 
3288 	if (netdev->features & NETIF_F_HW_VLAN_RX)
3289 		ixgbe_vlan_strip_enable(adapter);
3290 	else
3291 		ixgbe_vlan_strip_disable(adapter);
3292 }
3293 
3294 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3295 {
3296 	int q_idx;
3297 	struct ixgbe_q_vector *q_vector;
3298 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3299 
3300 	/* legacy and MSI only use one vector */
3301 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3302 		q_vectors = 1;
3303 
3304 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3305 		q_vector = adapter->q_vector[q_idx];
3306 		napi_enable(&q_vector->napi);
3307 	}
3308 }
3309 
3310 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3311 {
3312 	int q_idx;
3313 	struct ixgbe_q_vector *q_vector;
3314 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3315 
3316 	/* legacy and MSI only use one vector */
3317 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3318 		q_vectors = 1;
3319 
3320 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3321 		q_vector = adapter->q_vector[q_idx];
3322 		napi_disable(&q_vector->napi);
3323 	}
3324 }
3325 
3326 #ifdef CONFIG_IXGBE_DCB
3327 /*
3328  * ixgbe_configure_dcb - Configure DCB hardware
3329  * @adapter: ixgbe adapter struct
3330  *
3331  * This is called by the driver on open to configure the DCB hardware.
3332  * This is also called by the gennetlink interface when reconfiguring
3333  * the DCB state.
3334  */
3335 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3336 {
3337 	struct ixgbe_hw *hw = &adapter->hw;
3338 	int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3339 
3340 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3341 		if (hw->mac.type == ixgbe_mac_82598EB)
3342 			netif_set_gso_max_size(adapter->netdev, 65536);
3343 		return;
3344 	}
3345 
3346 	if (hw->mac.type == ixgbe_mac_82598EB)
3347 		netif_set_gso_max_size(adapter->netdev, 32768);
3348 
3349 
3350 	/* Enable VLAN tag insert/strip */
3351 	adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3352 
3353 	hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3354 
3355 #ifdef IXGBE_FCOE
3356 	if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3357 		max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3358 #endif
3359 
3360 	/* reconfigure the hardware */
3361 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
3362 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3363 						DCB_TX_CONFIG);
3364 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3365 						DCB_RX_CONFIG);
3366 		ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3367 	} else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
3368 		ixgbe_dcb_hw_ets(&adapter->hw,
3369 				 adapter->ixgbe_ieee_ets,
3370 				 max_frame);
3371 		ixgbe_dcb_hw_pfc_config(&adapter->hw,
3372 					adapter->ixgbe_ieee_pfc->pfc_en,
3373 					adapter->ixgbe_ieee_ets->prio_tc);
3374 	}
3375 
3376 	/* Enable RSS Hash per TC */
3377 	if (hw->mac.type != ixgbe_mac_82598EB) {
3378 		int i;
3379 		u32 reg = 0;
3380 
3381 		for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
3382 			u8 msb = 0;
3383 			u8 cnt = adapter->netdev->tc_to_txq[i].count;
3384 
3385 			while (cnt >>= 1)
3386 				msb++;
3387 
3388 			reg |= msb << IXGBE_RQTC_SHIFT_TC(i);
3389 		}
3390 		IXGBE_WRITE_REG(hw, IXGBE_RQTC, reg);
3391 	}
3392 }
3393 #endif
3394 
3395 /* Additional bittime to account for IXGBE framing */
3396 #define IXGBE_ETH_FRAMING 20
3397 
3398 /*
3399  * ixgbe_hpbthresh - calculate high water mark for flow control
3400  *
3401  * @adapter: board private structure to calculate for
3402  * @pb - packet buffer to calculate
3403  */
3404 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
3405 {
3406 	struct ixgbe_hw *hw = &adapter->hw;
3407 	struct net_device *dev = adapter->netdev;
3408 	int link, tc, kb, marker;
3409 	u32 dv_id, rx_pba;
3410 
3411 	/* Calculate max LAN frame size */
3412 	tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
3413 
3414 #ifdef IXGBE_FCOE
3415 	/* FCoE traffic class uses FCOE jumbo frames */
3416 	if (dev->features & NETIF_F_FCOE_MTU) {
3417 		int fcoe_pb = 0;
3418 
3419 #ifdef CONFIG_IXGBE_DCB
3420 		fcoe_pb = netdev_get_prio_tc_map(dev, adapter->fcoe.up);
3421 
3422 #endif
3423 		if (fcoe_pb == pb && tc < IXGBE_FCOE_JUMBO_FRAME_SIZE)
3424 			tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3425 	}
3426 #endif
3427 
3428 	/* Calculate delay value for device */
3429 	switch (hw->mac.type) {
3430 	case ixgbe_mac_X540:
3431 		dv_id = IXGBE_DV_X540(link, tc);
3432 		break;
3433 	default:
3434 		dv_id = IXGBE_DV(link, tc);
3435 		break;
3436 	}
3437 
3438 	/* Loopback switch introduces additional latency */
3439 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3440 		dv_id += IXGBE_B2BT(tc);
3441 
3442 	/* Delay value is calculated in bit times convert to KB */
3443 	kb = IXGBE_BT2KB(dv_id);
3444 	rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
3445 
3446 	marker = rx_pba - kb;
3447 
3448 	/* It is possible that the packet buffer is not large enough
3449 	 * to provide required headroom. In this case throw an error
3450 	 * to user and a do the best we can.
3451 	 */
3452 	if (marker < 0) {
3453 		e_warn(drv, "Packet Buffer(%i) can not provide enough"
3454 			    "headroom to support flow control."
3455 			    "Decrease MTU or number of traffic classes\n", pb);
3456 		marker = tc + 1;
3457 	}
3458 
3459 	return marker;
3460 }
3461 
3462 /*
3463  * ixgbe_lpbthresh - calculate low water mark for for flow control
3464  *
3465  * @adapter: board private structure to calculate for
3466  * @pb - packet buffer to calculate
3467  */
3468 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter)
3469 {
3470 	struct ixgbe_hw *hw = &adapter->hw;
3471 	struct net_device *dev = adapter->netdev;
3472 	int tc;
3473 	u32 dv_id;
3474 
3475 	/* Calculate max LAN frame size */
3476 	tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
3477 
3478 	/* Calculate delay value for device */
3479 	switch (hw->mac.type) {
3480 	case ixgbe_mac_X540:
3481 		dv_id = IXGBE_LOW_DV_X540(tc);
3482 		break;
3483 	default:
3484 		dv_id = IXGBE_LOW_DV(tc);
3485 		break;
3486 	}
3487 
3488 	/* Delay value is calculated in bit times convert to KB */
3489 	return IXGBE_BT2KB(dv_id);
3490 }
3491 
3492 /*
3493  * ixgbe_pbthresh_setup - calculate and setup high low water marks
3494  */
3495 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
3496 {
3497 	struct ixgbe_hw *hw = &adapter->hw;
3498 	int num_tc = netdev_get_num_tc(adapter->netdev);
3499 	int i;
3500 
3501 	if (!num_tc)
3502 		num_tc = 1;
3503 
3504 	hw->fc.low_water = ixgbe_lpbthresh(adapter);
3505 
3506 	for (i = 0; i < num_tc; i++) {
3507 		hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
3508 
3509 		/* Low water marks must not be larger than high water marks */
3510 		if (hw->fc.low_water > hw->fc.high_water[i])
3511 			hw->fc.low_water = 0;
3512 	}
3513 }
3514 
3515 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
3516 {
3517 	struct ixgbe_hw *hw = &adapter->hw;
3518 	int hdrm;
3519 	u8 tc = netdev_get_num_tc(adapter->netdev);
3520 
3521 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3522 	    adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3523 		hdrm = 32 << adapter->fdir_pballoc;
3524 	else
3525 		hdrm = 0;
3526 
3527 	hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
3528 	ixgbe_pbthresh_setup(adapter);
3529 }
3530 
3531 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
3532 {
3533 	struct ixgbe_hw *hw = &adapter->hw;
3534 	struct hlist_node *node, *node2;
3535 	struct ixgbe_fdir_filter *filter;
3536 
3537 	spin_lock(&adapter->fdir_perfect_lock);
3538 
3539 	if (!hlist_empty(&adapter->fdir_filter_list))
3540 		ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
3541 
3542 	hlist_for_each_entry_safe(filter, node, node2,
3543 				  &adapter->fdir_filter_list, fdir_node) {
3544 		ixgbe_fdir_write_perfect_filter_82599(hw,
3545 				&filter->filter,
3546 				filter->sw_idx,
3547 				(filter->action == IXGBE_FDIR_DROP_QUEUE) ?
3548 				IXGBE_FDIR_DROP_QUEUE :
3549 				adapter->rx_ring[filter->action]->reg_idx);
3550 	}
3551 
3552 	spin_unlock(&adapter->fdir_perfect_lock);
3553 }
3554 
3555 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3556 {
3557 	ixgbe_configure_pb(adapter);
3558 #ifdef CONFIG_IXGBE_DCB
3559 	ixgbe_configure_dcb(adapter);
3560 #endif
3561 
3562 	ixgbe_set_rx_mode(adapter->netdev);
3563 	ixgbe_restore_vlan(adapter);
3564 
3565 #ifdef IXGBE_FCOE
3566 	if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3567 		ixgbe_configure_fcoe(adapter);
3568 
3569 #endif /* IXGBE_FCOE */
3570 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3571 		ixgbe_init_fdir_signature_82599(&adapter->hw,
3572 						adapter->fdir_pballoc);
3573 	} else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3574 		ixgbe_init_fdir_perfect_82599(&adapter->hw,
3575 					      adapter->fdir_pballoc);
3576 		ixgbe_fdir_filter_restore(adapter);
3577 	}
3578 
3579 	ixgbe_configure_virtualization(adapter);
3580 
3581 	ixgbe_configure_tx(adapter);
3582 	ixgbe_configure_rx(adapter);
3583 }
3584 
3585 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3586 {
3587 	switch (hw->phy.type) {
3588 	case ixgbe_phy_sfp_avago:
3589 	case ixgbe_phy_sfp_ftl:
3590 	case ixgbe_phy_sfp_intel:
3591 	case ixgbe_phy_sfp_unknown:
3592 	case ixgbe_phy_sfp_passive_tyco:
3593 	case ixgbe_phy_sfp_passive_unknown:
3594 	case ixgbe_phy_sfp_active_unknown:
3595 	case ixgbe_phy_sfp_ftl_active:
3596 		return true;
3597 	case ixgbe_phy_nl:
3598 		if (hw->mac.type == ixgbe_mac_82598EB)
3599 			return true;
3600 	default:
3601 		return false;
3602 	}
3603 }
3604 
3605 /**
3606  * ixgbe_sfp_link_config - set up SFP+ link
3607  * @adapter: pointer to private adapter struct
3608  **/
3609 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3610 {
3611 	/*
3612 	 * We are assuming the worst case scenario here, and that
3613 	 * is that an SFP was inserted/removed after the reset
3614 	 * but before SFP detection was enabled.  As such the best
3615 	 * solution is to just start searching as soon as we start
3616 	 */
3617 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3618 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
3619 
3620 	adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
3621 }
3622 
3623 /**
3624  * ixgbe_non_sfp_link_config - set up non-SFP+ link
3625  * @hw: pointer to private hardware struct
3626  *
3627  * Returns 0 on success, negative on failure
3628  **/
3629 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3630 {
3631 	u32 autoneg;
3632 	bool negotiation, link_up = false;
3633 	u32 ret = IXGBE_ERR_LINK_SETUP;
3634 
3635 	if (hw->mac.ops.check_link)
3636 		ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3637 
3638 	if (ret)
3639 		goto link_cfg_out;
3640 
3641 	autoneg = hw->phy.autoneg_advertised;
3642 	if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
3643 		ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3644 							&negotiation);
3645 	if (ret)
3646 		goto link_cfg_out;
3647 
3648 	if (hw->mac.ops.setup_link)
3649 		ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3650 link_cfg_out:
3651 	return ret;
3652 }
3653 
3654 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3655 {
3656 	struct ixgbe_hw *hw = &adapter->hw;
3657 	u32 gpie = 0;
3658 
3659 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3660 		gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3661 		       IXGBE_GPIE_OCD;
3662 		gpie |= IXGBE_GPIE_EIAME;
3663 		/*
3664 		 * use EIAM to auto-mask when MSI-X interrupt is asserted
3665 		 * this saves a register write for every interrupt
3666 		 */
3667 		switch (hw->mac.type) {
3668 		case ixgbe_mac_82598EB:
3669 			IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3670 			break;
3671 		case ixgbe_mac_82599EB:
3672 		case ixgbe_mac_X540:
3673 		default:
3674 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3675 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3676 			break;
3677 		}
3678 	} else {
3679 		/* legacy interrupts, use EIAM to auto-mask when reading EICR,
3680 		 * specifically only auto mask tx and rx interrupts */
3681 		IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3682 	}
3683 
3684 	/* XXX: to interrupt immediately for EICS writes, enable this */
3685 	/* gpie |= IXGBE_GPIE_EIMEN; */
3686 
3687 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3688 		gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3689 		gpie |= IXGBE_GPIE_VTMODE_64;
3690 	}
3691 
3692 	/* Enable Thermal over heat sensor interrupt */
3693 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
3694 		switch (adapter->hw.mac.type) {
3695 		case ixgbe_mac_82599EB:
3696 			gpie |= IXGBE_SDP0_GPIEN;
3697 			break;
3698 		case ixgbe_mac_X540:
3699 			gpie |= IXGBE_EIMS_TS;
3700 			break;
3701 		default:
3702 			break;
3703 		}
3704 	}
3705 
3706 	/* Enable fan failure interrupt */
3707 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3708 		gpie |= IXGBE_SDP1_GPIEN;
3709 
3710 	if (hw->mac.type == ixgbe_mac_82599EB) {
3711 		gpie |= IXGBE_SDP1_GPIEN;
3712 		gpie |= IXGBE_SDP2_GPIEN;
3713 	}
3714 
3715 	IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3716 }
3717 
3718 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
3719 {
3720 	struct ixgbe_hw *hw = &adapter->hw;
3721 	int err;
3722 	u32 ctrl_ext;
3723 
3724 	ixgbe_get_hw_control(adapter);
3725 	ixgbe_setup_gpie(adapter);
3726 
3727 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3728 		ixgbe_configure_msix(adapter);
3729 	else
3730 		ixgbe_configure_msi_and_legacy(adapter);
3731 
3732 	/* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
3733 	if (hw->mac.ops.enable_tx_laser &&
3734 	    ((hw->phy.multispeed_fiber) ||
3735 	     ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3736 	      (hw->mac.type == ixgbe_mac_82599EB))))
3737 		hw->mac.ops.enable_tx_laser(hw);
3738 
3739 	clear_bit(__IXGBE_DOWN, &adapter->state);
3740 	ixgbe_napi_enable_all(adapter);
3741 
3742 	if (ixgbe_is_sfp(hw)) {
3743 		ixgbe_sfp_link_config(adapter);
3744 	} else {
3745 		err = ixgbe_non_sfp_link_config(hw);
3746 		if (err)
3747 			e_err(probe, "link_config FAILED %d\n", err);
3748 	}
3749 
3750 	/* clear any pending interrupts, may auto mask */
3751 	IXGBE_READ_REG(hw, IXGBE_EICR);
3752 	ixgbe_irq_enable(adapter, true, true);
3753 
3754 	/*
3755 	 * If this adapter has a fan, check to see if we had a failure
3756 	 * before we enabled the interrupt.
3757 	 */
3758 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3759 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3760 		if (esdp & IXGBE_ESDP_SDP1)
3761 			e_crit(drv, "Fan has stopped, replace the adapter\n");
3762 	}
3763 
3764 	/* enable transmits */
3765 	netif_tx_start_all_queues(adapter->netdev);
3766 
3767 	/* bring the link up in the watchdog, this could race with our first
3768 	 * link up interrupt but shouldn't be a problem */
3769 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3770 	adapter->link_check_timeout = jiffies;
3771 	mod_timer(&adapter->service_timer, jiffies);
3772 
3773 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
3774 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3775 	ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3776 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3777 }
3778 
3779 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3780 {
3781 	WARN_ON(in_interrupt());
3782 	/* put off any impending NetWatchDogTimeout */
3783 	adapter->netdev->trans_start = jiffies;
3784 
3785 	while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3786 		usleep_range(1000, 2000);
3787 	ixgbe_down(adapter);
3788 	/*
3789 	 * If SR-IOV enabled then wait a bit before bringing the adapter
3790 	 * back up to give the VFs time to respond to the reset.  The
3791 	 * two second wait is based upon the watchdog timer cycle in
3792 	 * the VF driver.
3793 	 */
3794 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3795 		msleep(2000);
3796 	ixgbe_up(adapter);
3797 	clear_bit(__IXGBE_RESETTING, &adapter->state);
3798 }
3799 
3800 void ixgbe_up(struct ixgbe_adapter *adapter)
3801 {
3802 	/* hardware has been reset, we need to reload some things */
3803 	ixgbe_configure(adapter);
3804 
3805 	ixgbe_up_complete(adapter);
3806 }
3807 
3808 void ixgbe_reset(struct ixgbe_adapter *adapter)
3809 {
3810 	struct ixgbe_hw *hw = &adapter->hw;
3811 	int err;
3812 
3813 	/* lock SFP init bit to prevent race conditions with the watchdog */
3814 	while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
3815 		usleep_range(1000, 2000);
3816 
3817 	/* clear all SFP and link config related flags while holding SFP_INIT */
3818 	adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
3819 			     IXGBE_FLAG2_SFP_NEEDS_RESET);
3820 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
3821 
3822 	err = hw->mac.ops.init_hw(hw);
3823 	switch (err) {
3824 	case 0:
3825 	case IXGBE_ERR_SFP_NOT_PRESENT:
3826 	case IXGBE_ERR_SFP_NOT_SUPPORTED:
3827 		break;
3828 	case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3829 		e_dev_err("master disable timed out\n");
3830 		break;
3831 	case IXGBE_ERR_EEPROM_VERSION:
3832 		/* We are running on a pre-production device, log a warning */
3833 		e_dev_warn("This device is a pre-production adapter/LOM. "
3834 			   "Please be aware there may be issues associated with "
3835 			   "your hardware.  If you are experiencing problems "
3836 			   "please contact your Intel or hardware "
3837 			   "representative who provided you with this "
3838 			   "hardware.\n");
3839 		break;
3840 	default:
3841 		e_dev_err("Hardware Error: %d\n", err);
3842 	}
3843 
3844 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
3845 
3846 	/* reprogram the RAR[0] in case user changed it. */
3847 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3848 			    IXGBE_RAH_AV);
3849 }
3850 
3851 /**
3852  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3853  * @rx_ring: ring to free buffers from
3854  **/
3855 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
3856 {
3857 	struct device *dev = rx_ring->dev;
3858 	unsigned long size;
3859 	u16 i;
3860 
3861 	/* ring already cleared, nothing to do */
3862 	if (!rx_ring->rx_buffer_info)
3863 		return;
3864 
3865 	/* Free all the Rx ring sk_buffs */
3866 	for (i = 0; i < rx_ring->count; i++) {
3867 		struct ixgbe_rx_buffer *rx_buffer_info;
3868 
3869 		rx_buffer_info = &rx_ring->rx_buffer_info[i];
3870 		if (rx_buffer_info->dma) {
3871 			dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
3872 					 rx_ring->rx_buf_len,
3873 					 DMA_FROM_DEVICE);
3874 			rx_buffer_info->dma = 0;
3875 		}
3876 		if (rx_buffer_info->skb) {
3877 			struct sk_buff *skb = rx_buffer_info->skb;
3878 			rx_buffer_info->skb = NULL;
3879 			do {
3880 				struct sk_buff *this = skb;
3881 				if (IXGBE_RSC_CB(this)->delay_unmap) {
3882 					dma_unmap_single(dev,
3883 							 IXGBE_RSC_CB(this)->dma,
3884 							 rx_ring->rx_buf_len,
3885 							 DMA_FROM_DEVICE);
3886 					IXGBE_RSC_CB(this)->dma = 0;
3887 					IXGBE_RSC_CB(skb)->delay_unmap = false;
3888 				}
3889 				skb = skb->prev;
3890 				dev_kfree_skb(this);
3891 			} while (skb);
3892 		}
3893 		if (!rx_buffer_info->page)
3894 			continue;
3895 		if (rx_buffer_info->page_dma) {
3896 			dma_unmap_page(dev, rx_buffer_info->page_dma,
3897 				       PAGE_SIZE / 2, DMA_FROM_DEVICE);
3898 			rx_buffer_info->page_dma = 0;
3899 		}
3900 		put_page(rx_buffer_info->page);
3901 		rx_buffer_info->page = NULL;
3902 		rx_buffer_info->page_offset = 0;
3903 	}
3904 
3905 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3906 	memset(rx_ring->rx_buffer_info, 0, size);
3907 
3908 	/* Zero out the descriptor ring */
3909 	memset(rx_ring->desc, 0, rx_ring->size);
3910 
3911 	rx_ring->next_to_clean = 0;
3912 	rx_ring->next_to_use = 0;
3913 }
3914 
3915 /**
3916  * ixgbe_clean_tx_ring - Free Tx Buffers
3917  * @tx_ring: ring to be cleaned
3918  **/
3919 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
3920 {
3921 	struct ixgbe_tx_buffer *tx_buffer_info;
3922 	unsigned long size;
3923 	u16 i;
3924 
3925 	/* ring already cleared, nothing to do */
3926 	if (!tx_ring->tx_buffer_info)
3927 		return;
3928 
3929 	/* Free all the Tx ring sk_buffs */
3930 	for (i = 0; i < tx_ring->count; i++) {
3931 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
3932 		ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
3933 	}
3934 
3935 	size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3936 	memset(tx_ring->tx_buffer_info, 0, size);
3937 
3938 	/* Zero out the descriptor ring */
3939 	memset(tx_ring->desc, 0, tx_ring->size);
3940 
3941 	tx_ring->next_to_use = 0;
3942 	tx_ring->next_to_clean = 0;
3943 }
3944 
3945 /**
3946  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3947  * @adapter: board private structure
3948  **/
3949 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3950 {
3951 	int i;
3952 
3953 	for (i = 0; i < adapter->num_rx_queues; i++)
3954 		ixgbe_clean_rx_ring(adapter->rx_ring[i]);
3955 }
3956 
3957 /**
3958  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3959  * @adapter: board private structure
3960  **/
3961 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3962 {
3963 	int i;
3964 
3965 	for (i = 0; i < adapter->num_tx_queues; i++)
3966 		ixgbe_clean_tx_ring(adapter->tx_ring[i]);
3967 }
3968 
3969 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
3970 {
3971 	struct hlist_node *node, *node2;
3972 	struct ixgbe_fdir_filter *filter;
3973 
3974 	spin_lock(&adapter->fdir_perfect_lock);
3975 
3976 	hlist_for_each_entry_safe(filter, node, node2,
3977 				  &adapter->fdir_filter_list, fdir_node) {
3978 		hlist_del(&filter->fdir_node);
3979 		kfree(filter);
3980 	}
3981 	adapter->fdir_filter_count = 0;
3982 
3983 	spin_unlock(&adapter->fdir_perfect_lock);
3984 }
3985 
3986 void ixgbe_down(struct ixgbe_adapter *adapter)
3987 {
3988 	struct net_device *netdev = adapter->netdev;
3989 	struct ixgbe_hw *hw = &adapter->hw;
3990 	u32 rxctrl;
3991 	int i;
3992 
3993 	/* signal that we are down to the interrupt handler */
3994 	set_bit(__IXGBE_DOWN, &adapter->state);
3995 
3996 	/* disable receives */
3997 	rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3998 	IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3999 
4000 	/* disable all enabled rx queues */
4001 	for (i = 0; i < adapter->num_rx_queues; i++)
4002 		/* this call also flushes the previous write */
4003 		ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4004 
4005 	usleep_range(10000, 20000);
4006 
4007 	netif_tx_stop_all_queues(netdev);
4008 
4009 	/* call carrier off first to avoid false dev_watchdog timeouts */
4010 	netif_carrier_off(netdev);
4011 	netif_tx_disable(netdev);
4012 
4013 	ixgbe_irq_disable(adapter);
4014 
4015 	ixgbe_napi_disable_all(adapter);
4016 
4017 	adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
4018 			     IXGBE_FLAG2_RESET_REQUESTED);
4019 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4020 
4021 	del_timer_sync(&adapter->service_timer);
4022 
4023 	if (adapter->num_vfs) {
4024 		/* Clear EITR Select mapping */
4025 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
4026 
4027 		/* Mark all the VFs as inactive */
4028 		for (i = 0 ; i < adapter->num_vfs; i++)
4029 			adapter->vfinfo[i].clear_to_send = false;
4030 
4031 		/* ping all the active vfs to let them know we are going down */
4032 		ixgbe_ping_all_vfs(adapter);
4033 
4034 		/* Disable all VFTE/VFRE TX/RX */
4035 		ixgbe_disable_tx_rx(adapter);
4036 	}
4037 
4038 	/* disable transmits in the hardware now that interrupts are off */
4039 	for (i = 0; i < adapter->num_tx_queues; i++) {
4040 		u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4041 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
4042 	}
4043 
4044 	/* Disable the Tx DMA engine on 82599 and X540 */
4045 	switch (hw->mac.type) {
4046 	case ixgbe_mac_82599EB:
4047 	case ixgbe_mac_X540:
4048 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4049 				(IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4050 				 ~IXGBE_DMATXCTL_TE));
4051 		break;
4052 	default:
4053 		break;
4054 	}
4055 
4056 	if (!pci_channel_offline(adapter->pdev))
4057 		ixgbe_reset(adapter);
4058 
4059 	/* power down the optics for multispeed fiber and 82599 SFP+ fiber */
4060 	if (hw->mac.ops.disable_tx_laser &&
4061 	    ((hw->phy.multispeed_fiber) ||
4062 	     ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4063 	      (hw->mac.type == ixgbe_mac_82599EB))))
4064 		hw->mac.ops.disable_tx_laser(hw);
4065 
4066 	ixgbe_clean_all_tx_rings(adapter);
4067 	ixgbe_clean_all_rx_rings(adapter);
4068 
4069 #ifdef CONFIG_IXGBE_DCA
4070 	/* since we reset the hardware DCA settings were cleared */
4071 	ixgbe_setup_dca(adapter);
4072 #endif
4073 }
4074 
4075 /**
4076  * ixgbe_poll - NAPI Rx polling callback
4077  * @napi: structure for representing this polling device
4078  * @budget: how many packets driver is allowed to clean
4079  *
4080  * This function is used for legacy and MSI, NAPI mode
4081  **/
4082 static int ixgbe_poll(struct napi_struct *napi, int budget)
4083 {
4084 	struct ixgbe_q_vector *q_vector =
4085 				container_of(napi, struct ixgbe_q_vector, napi);
4086 	struct ixgbe_adapter *adapter = q_vector->adapter;
4087 	struct ixgbe_ring *ring;
4088 	int per_ring_budget;
4089 	bool clean_complete = true;
4090 
4091 #ifdef CONFIG_IXGBE_DCA
4092 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
4093 		ixgbe_update_dca(q_vector);
4094 #endif
4095 
4096 	for (ring = q_vector->tx.ring; ring != NULL; ring = ring->next)
4097 		clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
4098 
4099 	/* attempt to distribute budget to each queue fairly, but don't allow
4100 	 * the budget to go below 1 because we'll exit polling */
4101 	if (q_vector->rx.count > 1)
4102 		per_ring_budget = max(budget/q_vector->rx.count, 1);
4103 	else
4104 		per_ring_budget = budget;
4105 
4106 	for (ring = q_vector->rx.ring; ring != NULL; ring = ring->next)
4107 		clean_complete &= ixgbe_clean_rx_irq(q_vector, ring,
4108 						     per_ring_budget);
4109 
4110 	/* If all work not completed, return budget and keep polling */
4111 	if (!clean_complete)
4112 		return budget;
4113 
4114 	/* all work done, exit the polling mode */
4115 	napi_complete(napi);
4116 	if (adapter->rx_itr_setting & 1)
4117 		ixgbe_set_itr(q_vector);
4118 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
4119 		ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
4120 
4121 	return 0;
4122 }
4123 
4124 /**
4125  * ixgbe_tx_timeout - Respond to a Tx Hang
4126  * @netdev: network interface device structure
4127  **/
4128 static void ixgbe_tx_timeout(struct net_device *netdev)
4129 {
4130 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4131 
4132 	/* Do the reset outside of interrupt context */
4133 	ixgbe_tx_timeout_reset(adapter);
4134 }
4135 
4136 /**
4137  * ixgbe_set_rss_queues: Allocate queues for RSS
4138  * @adapter: board private structure to initialize
4139  *
4140  * This is our "base" multiqueue mode.  RSS (Receive Side Scaling) will try
4141  * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
4142  *
4143  **/
4144 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
4145 {
4146 	bool ret = false;
4147 	struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
4148 
4149 	if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4150 		f->mask = 0xF;
4151 		adapter->num_rx_queues = f->indices;
4152 		adapter->num_tx_queues = f->indices;
4153 		ret = true;
4154 	} else {
4155 		ret = false;
4156 	}
4157 
4158 	return ret;
4159 }
4160 
4161 /**
4162  * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4163  * @adapter: board private structure to initialize
4164  *
4165  * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4166  * to the original CPU that initiated the Tx session.  This runs in addition
4167  * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4168  * Rx load across CPUs using RSS.
4169  *
4170  **/
4171 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4172 {
4173 	bool ret = false;
4174 	struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4175 
4176 	f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4177 	f_fdir->mask = 0;
4178 
4179 	/* Flow Director must have RSS enabled */
4180 	if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
4181 	    (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)) {
4182 		adapter->num_tx_queues = f_fdir->indices;
4183 		adapter->num_rx_queues = f_fdir->indices;
4184 		ret = true;
4185 	} else {
4186 		adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4187 	}
4188 	return ret;
4189 }
4190 
4191 #ifdef IXGBE_FCOE
4192 /**
4193  * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4194  * @adapter: board private structure to initialize
4195  *
4196  * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4197  * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4198  * rx queues out of the max number of rx queues, instead, it is used as the
4199  * index of the first rx queue used by FCoE.
4200  *
4201  **/
4202 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4203 {
4204 	struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4205 
4206 	if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4207 		return false;
4208 
4209 	f->indices = min((int)num_online_cpus(), f->indices);
4210 
4211 	adapter->num_rx_queues = 1;
4212 	adapter->num_tx_queues = 1;
4213 
4214 	if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4215 		e_info(probe, "FCoE enabled with RSS\n");
4216 		if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
4217 			ixgbe_set_fdir_queues(adapter);
4218 		else
4219 			ixgbe_set_rss_queues(adapter);
4220 	}
4221 
4222 	/* adding FCoE rx rings to the end */
4223 	f->mask = adapter->num_rx_queues;
4224 	adapter->num_rx_queues += f->indices;
4225 	adapter->num_tx_queues += f->indices;
4226 
4227 	return true;
4228 }
4229 #endif /* IXGBE_FCOE */
4230 
4231 /* Artificial max queue cap per traffic class in DCB mode */
4232 #define DCB_QUEUE_CAP 8
4233 
4234 #ifdef CONFIG_IXGBE_DCB
4235 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
4236 {
4237 	int per_tc_q, q, i, offset = 0;
4238 	struct net_device *dev = adapter->netdev;
4239 	int tcs = netdev_get_num_tc(dev);
4240 
4241 	if (!tcs)
4242 		return false;
4243 
4244 	/* Map queue offset and counts onto allocated tx queues */
4245 	per_tc_q = min(dev->num_tx_queues / tcs, (unsigned int)DCB_QUEUE_CAP);
4246 	q = min((int)num_online_cpus(), per_tc_q);
4247 
4248 	for (i = 0; i < tcs; i++) {
4249 		netdev_set_tc_queue(dev, i, q, offset);
4250 		offset += q;
4251 	}
4252 
4253 	adapter->num_tx_queues = q * tcs;
4254 	adapter->num_rx_queues = q * tcs;
4255 
4256 #ifdef IXGBE_FCOE
4257 	/* FCoE enabled queues require special configuration indexed
4258 	 * by feature specific indices and mask. Here we map FCoE
4259 	 * indices onto the DCB queue pairs allowing FCoE to own
4260 	 * configuration later.
4261 	 */
4262 	if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4263 		int tc;
4264 		struct ixgbe_ring_feature *f =
4265 					&adapter->ring_feature[RING_F_FCOE];
4266 
4267 		tc = netdev_get_prio_tc_map(dev, adapter->fcoe.up);
4268 		f->indices = dev->tc_to_txq[tc].count;
4269 		f->mask = dev->tc_to_txq[tc].offset;
4270 	}
4271 #endif
4272 
4273 	return true;
4274 }
4275 #endif
4276 
4277 /**
4278  * ixgbe_set_sriov_queues: Allocate queues for IOV use
4279  * @adapter: board private structure to initialize
4280  *
4281  * IOV doesn't actually use anything, so just NAK the
4282  * request for now and let the other queue routines
4283  * figure out what to do.
4284  */
4285 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4286 {
4287 	return false;
4288 }
4289 
4290 /*
4291  * ixgbe_set_num_queues: Allocate queues for device, feature dependent
4292  * @adapter: board private structure to initialize
4293  *
4294  * This is the top level queue allocation routine.  The order here is very
4295  * important, starting with the "most" number of features turned on at once,
4296  * and ending with the smallest set of features.  This way large combinations
4297  * can be allocated if they're turned on, and smaller combinations are the
4298  * fallthrough conditions.
4299  *
4300  **/
4301 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4302 {
4303 	/* Start with base case */
4304 	adapter->num_rx_queues = 1;
4305 	adapter->num_tx_queues = 1;
4306 	adapter->num_rx_pools = adapter->num_rx_queues;
4307 	adapter->num_rx_queues_per_pool = 1;
4308 
4309 	if (ixgbe_set_sriov_queues(adapter))
4310 		goto done;
4311 
4312 #ifdef CONFIG_IXGBE_DCB
4313 	if (ixgbe_set_dcb_queues(adapter))
4314 		goto done;
4315 
4316 #endif
4317 #ifdef IXGBE_FCOE
4318 	if (ixgbe_set_fcoe_queues(adapter))
4319 		goto done;
4320 
4321 #endif /* IXGBE_FCOE */
4322 	if (ixgbe_set_fdir_queues(adapter))
4323 		goto done;
4324 
4325 	if (ixgbe_set_rss_queues(adapter))
4326 		goto done;
4327 
4328 	/* fallback to base case */
4329 	adapter->num_rx_queues = 1;
4330 	adapter->num_tx_queues = 1;
4331 
4332 done:
4333 	/* Notify the stack of the (possibly) reduced queue counts. */
4334 	netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4335 	return netif_set_real_num_rx_queues(adapter->netdev,
4336 					    adapter->num_rx_queues);
4337 }
4338 
4339 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4340 				       int vectors)
4341 {
4342 	int err, vector_threshold;
4343 
4344 	/* We'll want at least 3 (vector_threshold):
4345 	 * 1) TxQ[0] Cleanup
4346 	 * 2) RxQ[0] Cleanup
4347 	 * 3) Other (Link Status Change, etc.)
4348 	 * 4) TCP Timer (optional)
4349 	 */
4350 	vector_threshold = MIN_MSIX_COUNT;
4351 
4352 	/* The more we get, the more we will assign to Tx/Rx Cleanup
4353 	 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4354 	 * Right now, we simply care about how many we'll get; we'll
4355 	 * set them up later while requesting irq's.
4356 	 */
4357 	while (vectors >= vector_threshold) {
4358 		err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4359 				      vectors);
4360 		if (!err) /* Success in acquiring all requested vectors. */
4361 			break;
4362 		else if (err < 0)
4363 			vectors = 0; /* Nasty failure, quit now */
4364 		else /* err == number of vectors we should try again with */
4365 			vectors = err;
4366 	}
4367 
4368 	if (vectors < vector_threshold) {
4369 		/* Can't allocate enough MSI-X interrupts?  Oh well.
4370 		 * This just means we'll go with either a single MSI
4371 		 * vector or fall back to legacy interrupts.
4372 		 */
4373 		netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4374 			     "Unable to allocate MSI-X interrupts\n");
4375 		adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4376 		kfree(adapter->msix_entries);
4377 		adapter->msix_entries = NULL;
4378 	} else {
4379 		adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4380 		/*
4381 		 * Adjust for only the vectors we'll use, which is minimum
4382 		 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4383 		 * vectors we were allocated.
4384 		 */
4385 		adapter->num_msix_vectors = min(vectors,
4386 				   adapter->max_msix_q_vectors + NON_Q_VECTORS);
4387 	}
4388 }
4389 
4390 /**
4391  * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4392  * @adapter: board private structure to initialize
4393  *
4394  * Cache the descriptor ring offsets for RSS to the assigned rings.
4395  *
4396  **/
4397 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4398 {
4399 	int i;
4400 
4401 	if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
4402 		return false;
4403 
4404 	for (i = 0; i < adapter->num_rx_queues; i++)
4405 		adapter->rx_ring[i]->reg_idx = i;
4406 	for (i = 0; i < adapter->num_tx_queues; i++)
4407 		adapter->tx_ring[i]->reg_idx = i;
4408 
4409 	return true;
4410 }
4411 
4412 #ifdef CONFIG_IXGBE_DCB
4413 
4414 /* ixgbe_get_first_reg_idx - Return first register index associated with ring */
4415 static void ixgbe_get_first_reg_idx(struct ixgbe_adapter *adapter, u8 tc,
4416 				    unsigned int *tx, unsigned int *rx)
4417 {
4418 	struct net_device *dev = adapter->netdev;
4419 	struct ixgbe_hw *hw = &adapter->hw;
4420 	u8 num_tcs = netdev_get_num_tc(dev);
4421 
4422 	*tx = 0;
4423 	*rx = 0;
4424 
4425 	switch (hw->mac.type) {
4426 	case ixgbe_mac_82598EB:
4427 		*tx = tc << 2;
4428 		*rx = tc << 3;
4429 		break;
4430 	case ixgbe_mac_82599EB:
4431 	case ixgbe_mac_X540:
4432 		if (num_tcs > 4) {
4433 			if (tc < 3) {
4434 				*tx = tc << 5;
4435 				*rx = tc << 4;
4436 			} else if (tc <  5) {
4437 				*tx = ((tc + 2) << 4);
4438 				*rx = tc << 4;
4439 			} else if (tc < num_tcs) {
4440 				*tx = ((tc + 8) << 3);
4441 				*rx = tc << 4;
4442 			}
4443 		} else {
4444 			*rx =  tc << 5;
4445 			switch (tc) {
4446 			case 0:
4447 				*tx =  0;
4448 				break;
4449 			case 1:
4450 				*tx = 64;
4451 				break;
4452 			case 2:
4453 				*tx = 96;
4454 				break;
4455 			case 3:
4456 				*tx = 112;
4457 				break;
4458 			default:
4459 				break;
4460 			}
4461 		}
4462 		break;
4463 	default:
4464 		break;
4465 	}
4466 }
4467 
4468 /**
4469  * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4470  * @adapter: board private structure to initialize
4471  *
4472  * Cache the descriptor ring offsets for DCB to the assigned rings.
4473  *
4474  **/
4475 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4476 {
4477 	struct net_device *dev = adapter->netdev;
4478 	int i, j, k;
4479 	u8 num_tcs = netdev_get_num_tc(dev);
4480 
4481 	if (!num_tcs)
4482 		return false;
4483 
4484 	for (i = 0, k = 0; i < num_tcs; i++) {
4485 		unsigned int tx_s, rx_s;
4486 		u16 count = dev->tc_to_txq[i].count;
4487 
4488 		ixgbe_get_first_reg_idx(adapter, i, &tx_s, &rx_s);
4489 		for (j = 0; j < count; j++, k++) {
4490 			adapter->tx_ring[k]->reg_idx = tx_s + j;
4491 			adapter->rx_ring[k]->reg_idx = rx_s + j;
4492 			adapter->tx_ring[k]->dcb_tc = i;
4493 			adapter->rx_ring[k]->dcb_tc = i;
4494 		}
4495 	}
4496 
4497 	return true;
4498 }
4499 #endif
4500 
4501 /**
4502  * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4503  * @adapter: board private structure to initialize
4504  *
4505  * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4506  *
4507  **/
4508 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4509 {
4510 	int i;
4511 	bool ret = false;
4512 
4513 	if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
4514 	    (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)) {
4515 		for (i = 0; i < adapter->num_rx_queues; i++)
4516 			adapter->rx_ring[i]->reg_idx = i;
4517 		for (i = 0; i < adapter->num_tx_queues; i++)
4518 			adapter->tx_ring[i]->reg_idx = i;
4519 		ret = true;
4520 	}
4521 
4522 	return ret;
4523 }
4524 
4525 #ifdef IXGBE_FCOE
4526 /**
4527  * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4528  * @adapter: board private structure to initialize
4529  *
4530  * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4531  *
4532  */
4533 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4534 {
4535 	struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4536 	int i;
4537 	u8 fcoe_rx_i = 0, fcoe_tx_i = 0;
4538 
4539 	if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4540 		return false;
4541 
4542 	if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4543 		if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
4544 			ixgbe_cache_ring_fdir(adapter);
4545 		else
4546 			ixgbe_cache_ring_rss(adapter);
4547 
4548 		fcoe_rx_i = f->mask;
4549 		fcoe_tx_i = f->mask;
4550 	}
4551 	for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4552 		adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4553 		adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4554 	}
4555 	return true;
4556 }
4557 
4558 #endif /* IXGBE_FCOE */
4559 /**
4560  * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4561  * @adapter: board private structure to initialize
4562  *
4563  * SR-IOV doesn't use any descriptor rings but changes the default if
4564  * no other mapping is used.
4565  *
4566  */
4567 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4568 {
4569 	adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4570 	adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4571 	if (adapter->num_vfs)
4572 		return true;
4573 	else
4574 		return false;
4575 }
4576 
4577 /**
4578  * ixgbe_cache_ring_register - Descriptor ring to register mapping
4579  * @adapter: board private structure to initialize
4580  *
4581  * Once we know the feature-set enabled for the device, we'll cache
4582  * the register offset the descriptor ring is assigned to.
4583  *
4584  * Note, the order the various feature calls is important.  It must start with
4585  * the "most" features enabled at the same time, then trickle down to the
4586  * least amount of features turned on at once.
4587  **/
4588 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4589 {
4590 	/* start with default case */
4591 	adapter->rx_ring[0]->reg_idx = 0;
4592 	adapter->tx_ring[0]->reg_idx = 0;
4593 
4594 	if (ixgbe_cache_ring_sriov(adapter))
4595 		return;
4596 
4597 #ifdef CONFIG_IXGBE_DCB
4598 	if (ixgbe_cache_ring_dcb(adapter))
4599 		return;
4600 #endif
4601 
4602 #ifdef IXGBE_FCOE
4603 	if (ixgbe_cache_ring_fcoe(adapter))
4604 		return;
4605 #endif /* IXGBE_FCOE */
4606 
4607 	if (ixgbe_cache_ring_fdir(adapter))
4608 		return;
4609 
4610 	if (ixgbe_cache_ring_rss(adapter))
4611 		return;
4612 }
4613 
4614 /**
4615  * ixgbe_alloc_queues - Allocate memory for all rings
4616  * @adapter: board private structure to initialize
4617  *
4618  * We allocate one ring per queue at run-time since we don't know the
4619  * number of queues at compile-time.  The polling_netdev array is
4620  * intended for Multiqueue, but should work fine with a single queue.
4621  **/
4622 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4623 {
4624 	int rx = 0, tx = 0, nid = adapter->node;
4625 
4626 	if (nid < 0 || !node_online(nid))
4627 		nid = first_online_node;
4628 
4629 	for (; tx < adapter->num_tx_queues; tx++) {
4630 		struct ixgbe_ring *ring;
4631 
4632 		ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4633 		if (!ring)
4634 			ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4635 		if (!ring)
4636 			goto err_allocation;
4637 		ring->count = adapter->tx_ring_count;
4638 		ring->queue_index = tx;
4639 		ring->numa_node = nid;
4640 		ring->dev = &adapter->pdev->dev;
4641 		ring->netdev = adapter->netdev;
4642 
4643 		adapter->tx_ring[tx] = ring;
4644 	}
4645 
4646 	for (; rx < adapter->num_rx_queues; rx++) {
4647 		struct ixgbe_ring *ring;
4648 
4649 		ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4650 		if (!ring)
4651 			ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4652 		if (!ring)
4653 			goto err_allocation;
4654 		ring->count = adapter->rx_ring_count;
4655 		ring->queue_index = rx;
4656 		ring->numa_node = nid;
4657 		ring->dev = &adapter->pdev->dev;
4658 		ring->netdev = adapter->netdev;
4659 
4660 		adapter->rx_ring[rx] = ring;
4661 	}
4662 
4663 	ixgbe_cache_ring_register(adapter);
4664 
4665 	return 0;
4666 
4667 err_allocation:
4668 	while (tx)
4669 		kfree(adapter->tx_ring[--tx]);
4670 
4671 	while (rx)
4672 		kfree(adapter->rx_ring[--rx]);
4673 	return -ENOMEM;
4674 }
4675 
4676 /**
4677  * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4678  * @adapter: board private structure to initialize
4679  *
4680  * Attempt to configure the interrupts using the best available
4681  * capabilities of the hardware and the kernel.
4682  **/
4683 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4684 {
4685 	struct ixgbe_hw *hw = &adapter->hw;
4686 	int err = 0;
4687 	int vector, v_budget;
4688 
4689 	/*
4690 	 * It's easy to be greedy for MSI-X vectors, but it really
4691 	 * doesn't do us much good if we have a lot more vectors
4692 	 * than CPU's.  So let's be conservative and only ask for
4693 	 * (roughly) the same number of vectors as there are CPU's.
4694 	 */
4695 	v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4696 		       (int)num_online_cpus()) + NON_Q_VECTORS;
4697 
4698 	/*
4699 	 * At the same time, hardware can only support a maximum of
4700 	 * hw.mac->max_msix_vectors vectors.  With features
4701 	 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4702 	 * descriptor queues supported by our device.  Thus, we cap it off in
4703 	 * those rare cases where the cpu count also exceeds our vector limit.
4704 	 */
4705 	v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4706 
4707 	/* A failure in MSI-X entry allocation isn't fatal, but it does
4708 	 * mean we disable MSI-X capabilities of the adapter. */
4709 	adapter->msix_entries = kcalloc(v_budget,
4710 					sizeof(struct msix_entry), GFP_KERNEL);
4711 	if (adapter->msix_entries) {
4712 		for (vector = 0; vector < v_budget; vector++)
4713 			adapter->msix_entries[vector].entry = vector;
4714 
4715 		ixgbe_acquire_msix_vectors(adapter, v_budget);
4716 
4717 		if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4718 			goto out;
4719 	}
4720 
4721 	adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4722 	adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4723 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
4724 		e_err(probe,
4725 		      "ATR is not supported while multiple "
4726 		      "queues are disabled.  Disabling Flow Director\n");
4727 	}
4728 	adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4729 	adapter->atr_sample_rate = 0;
4730 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4731 		ixgbe_disable_sriov(adapter);
4732 
4733 	err = ixgbe_set_num_queues(adapter);
4734 	if (err)
4735 		return err;
4736 
4737 	err = pci_enable_msi(adapter->pdev);
4738 	if (!err) {
4739 		adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4740 	} else {
4741 		netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4742 			     "Unable to allocate MSI interrupt, "
4743 			     "falling back to legacy.  Error: %d\n", err);
4744 		/* reset err */
4745 		err = 0;
4746 	}
4747 
4748 out:
4749 	return err;
4750 }
4751 
4752 /**
4753  * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4754  * @adapter: board private structure to initialize
4755  *
4756  * We allocate one q_vector per queue interrupt.  If allocation fails we
4757  * return -ENOMEM.
4758  **/
4759 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4760 {
4761 	int v_idx, num_q_vectors;
4762 	struct ixgbe_q_vector *q_vector;
4763 
4764 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4765 		num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4766 	else
4767 		num_q_vectors = 1;
4768 
4769 	for (v_idx = 0; v_idx < num_q_vectors; v_idx++) {
4770 		q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4771 					GFP_KERNEL, adapter->node);
4772 		if (!q_vector)
4773 			q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4774 					   GFP_KERNEL);
4775 		if (!q_vector)
4776 			goto err_out;
4777 
4778 		q_vector->adapter = adapter;
4779 		q_vector->v_idx = v_idx;
4780 
4781 		/* Allocate the affinity_hint cpumask, configure the mask */
4782 		if (!alloc_cpumask_var(&q_vector->affinity_mask, GFP_KERNEL))
4783 			goto err_out;
4784 		cpumask_set_cpu(v_idx, q_vector->affinity_mask);
4785 		netif_napi_add(adapter->netdev, &q_vector->napi,
4786 			       ixgbe_poll, 64);
4787 		adapter->q_vector[v_idx] = q_vector;
4788 	}
4789 
4790 	return 0;
4791 
4792 err_out:
4793 	while (v_idx) {
4794 		v_idx--;
4795 		q_vector = adapter->q_vector[v_idx];
4796 		netif_napi_del(&q_vector->napi);
4797 		free_cpumask_var(q_vector->affinity_mask);
4798 		kfree(q_vector);
4799 		adapter->q_vector[v_idx] = NULL;
4800 	}
4801 	return -ENOMEM;
4802 }
4803 
4804 /**
4805  * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4806  * @adapter: board private structure to initialize
4807  *
4808  * This function frees the memory allocated to the q_vectors.  In addition if
4809  * NAPI is enabled it will delete any references to the NAPI struct prior
4810  * to freeing the q_vector.
4811  **/
4812 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4813 {
4814 	int v_idx, num_q_vectors;
4815 
4816 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4817 		num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4818 	else
4819 		num_q_vectors = 1;
4820 
4821 	for (v_idx = 0; v_idx < num_q_vectors; v_idx++) {
4822 		struct ixgbe_q_vector *q_vector = adapter->q_vector[v_idx];
4823 		adapter->q_vector[v_idx] = NULL;
4824 		netif_napi_del(&q_vector->napi);
4825 		free_cpumask_var(q_vector->affinity_mask);
4826 		kfree(q_vector);
4827 	}
4828 }
4829 
4830 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4831 {
4832 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4833 		adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4834 		pci_disable_msix(adapter->pdev);
4835 		kfree(adapter->msix_entries);
4836 		adapter->msix_entries = NULL;
4837 	} else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4838 		adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4839 		pci_disable_msi(adapter->pdev);
4840 	}
4841 }
4842 
4843 /**
4844  * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4845  * @adapter: board private structure to initialize
4846  *
4847  * We determine which interrupt scheme to use based on...
4848  * - Kernel support (MSI, MSI-X)
4849  *   - which can be user-defined (via MODULE_PARAM)
4850  * - Hardware queue count (num_*_queues)
4851  *   - defined by miscellaneous hardware support/features (RSS, etc.)
4852  **/
4853 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4854 {
4855 	int err;
4856 
4857 	/* Number of supported queues */
4858 	err = ixgbe_set_num_queues(adapter);
4859 	if (err)
4860 		return err;
4861 
4862 	err = ixgbe_set_interrupt_capability(adapter);
4863 	if (err) {
4864 		e_dev_err("Unable to setup interrupt capabilities\n");
4865 		goto err_set_interrupt;
4866 	}
4867 
4868 	err = ixgbe_alloc_q_vectors(adapter);
4869 	if (err) {
4870 		e_dev_err("Unable to allocate memory for queue vectors\n");
4871 		goto err_alloc_q_vectors;
4872 	}
4873 
4874 	err = ixgbe_alloc_queues(adapter);
4875 	if (err) {
4876 		e_dev_err("Unable to allocate memory for queues\n");
4877 		goto err_alloc_queues;
4878 	}
4879 
4880 	e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4881 		   (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4882 		   adapter->num_rx_queues, adapter->num_tx_queues);
4883 
4884 	set_bit(__IXGBE_DOWN, &adapter->state);
4885 
4886 	return 0;
4887 
4888 err_alloc_queues:
4889 	ixgbe_free_q_vectors(adapter);
4890 err_alloc_q_vectors:
4891 	ixgbe_reset_interrupt_capability(adapter);
4892 err_set_interrupt:
4893 	return err;
4894 }
4895 
4896 /**
4897  * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4898  * @adapter: board private structure to clear interrupt scheme on
4899  *
4900  * We go through and clear interrupt specific resources and reset the structure
4901  * to pre-load conditions
4902  **/
4903 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4904 {
4905 	int i;
4906 
4907 	for (i = 0; i < adapter->num_tx_queues; i++) {
4908 		kfree(adapter->tx_ring[i]);
4909 		adapter->tx_ring[i] = NULL;
4910 	}
4911 	for (i = 0; i < adapter->num_rx_queues; i++) {
4912 		struct ixgbe_ring *ring = adapter->rx_ring[i];
4913 
4914 		/* ixgbe_get_stats64() might access this ring, we must wait
4915 		 * a grace period before freeing it.
4916 		 */
4917 		kfree_rcu(ring, rcu);
4918 		adapter->rx_ring[i] = NULL;
4919 	}
4920 
4921 	adapter->num_tx_queues = 0;
4922 	adapter->num_rx_queues = 0;
4923 
4924 	ixgbe_free_q_vectors(adapter);
4925 	ixgbe_reset_interrupt_capability(adapter);
4926 }
4927 
4928 /**
4929  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4930  * @adapter: board private structure to initialize
4931  *
4932  * ixgbe_sw_init initializes the Adapter private data structure.
4933  * Fields are initialized based on PCI device information and
4934  * OS network device settings (MTU size).
4935  **/
4936 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4937 {
4938 	struct ixgbe_hw *hw = &adapter->hw;
4939 	struct pci_dev *pdev = adapter->pdev;
4940 	unsigned int rss;
4941 #ifdef CONFIG_IXGBE_DCB
4942 	int j;
4943 	struct tc_configuration *tc;
4944 #endif
4945 
4946 	/* PCI config space info */
4947 
4948 	hw->vendor_id = pdev->vendor;
4949 	hw->device_id = pdev->device;
4950 	hw->revision_id = pdev->revision;
4951 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
4952 	hw->subsystem_device_id = pdev->subsystem_device;
4953 
4954 	/* Set capability flags */
4955 	rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4956 	adapter->ring_feature[RING_F_RSS].indices = rss;
4957 	adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4958 	switch (hw->mac.type) {
4959 	case ixgbe_mac_82598EB:
4960 		if (hw->device_id == IXGBE_DEV_ID_82598AT)
4961 			adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4962 		adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4963 		break;
4964 	case ixgbe_mac_X540:
4965 		adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4966 	case ixgbe_mac_82599EB:
4967 		adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4968 		adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4969 		adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4970 		if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4971 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4972 		/* Flow Director hash filters enabled */
4973 		adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4974 		adapter->atr_sample_rate = 20;
4975 		adapter->ring_feature[RING_F_FDIR].indices =
4976 							 IXGBE_MAX_FDIR_INDICES;
4977 		adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
4978 #ifdef IXGBE_FCOE
4979 		adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4980 		adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4981 		adapter->ring_feature[RING_F_FCOE].indices = 0;
4982 #ifdef CONFIG_IXGBE_DCB
4983 		/* Default traffic class to use for FCoE */
4984 		adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4985 #endif
4986 #endif /* IXGBE_FCOE */
4987 		break;
4988 	default:
4989 		break;
4990 	}
4991 
4992 	/* n-tuple support exists, always init our spinlock */
4993 	spin_lock_init(&adapter->fdir_perfect_lock);
4994 
4995 #ifdef CONFIG_IXGBE_DCB
4996 	switch (hw->mac.type) {
4997 	case ixgbe_mac_X540:
4998 		adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
4999 		adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
5000 		break;
5001 	default:
5002 		adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
5003 		adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
5004 		break;
5005 	}
5006 
5007 	/* Configure DCB traffic classes */
5008 	for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5009 		tc = &adapter->dcb_cfg.tc_config[j];
5010 		tc->path[DCB_TX_CONFIG].bwg_id = 0;
5011 		tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5012 		tc->path[DCB_RX_CONFIG].bwg_id = 0;
5013 		tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5014 		tc->dcb_pfc = pfc_disabled;
5015 	}
5016 
5017 	/* Initialize default user to priority mapping, UPx->TC0 */
5018 	tc = &adapter->dcb_cfg.tc_config[0];
5019 	tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
5020 	tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
5021 
5022 	adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5023 	adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5024 	adapter->dcb_cfg.pfc_mode_enable = false;
5025 	adapter->dcb_set_bitmap = 0x00;
5026 	adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
5027 	ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
5028 			   MAX_TRAFFIC_CLASS);
5029 
5030 #endif
5031 
5032 	/* default flow control settings */
5033 	hw->fc.requested_mode = ixgbe_fc_full;
5034 	hw->fc.current_mode = ixgbe_fc_full;	/* init for ethtool output */
5035 #ifdef CONFIG_DCB
5036 	adapter->last_lfc_mode = hw->fc.current_mode;
5037 #endif
5038 	ixgbe_pbthresh_setup(adapter);
5039 	hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5040 	hw->fc.send_xon = true;
5041 	hw->fc.disable_fc_autoneg = false;
5042 
5043 	/* enable itr by default in dynamic mode */
5044 	adapter->rx_itr_setting = 1;
5045 	adapter->tx_itr_setting = 1;
5046 
5047 	/* set defaults for eitr in MegaBytes */
5048 	adapter->eitr_low = 10;
5049 	adapter->eitr_high = 20;
5050 
5051 	/* set default ring sizes */
5052 	adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5053 	adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5054 
5055 	/* set default work limits */
5056 	adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
5057 
5058 	/* initialize eeprom parameters */
5059 	if (ixgbe_init_eeprom_params_generic(hw)) {
5060 		e_dev_err("EEPROM initialization failed\n");
5061 		return -EIO;
5062 	}
5063 
5064 	/* enable rx csum by default */
5065 	adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
5066 
5067 	/* get assigned NUMA node */
5068 	adapter->node = dev_to_node(&pdev->dev);
5069 
5070 	set_bit(__IXGBE_DOWN, &adapter->state);
5071 
5072 	return 0;
5073 }
5074 
5075 /**
5076  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5077  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
5078  *
5079  * Return 0 on success, negative on failure
5080  **/
5081 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5082 {
5083 	struct device *dev = tx_ring->dev;
5084 	int size;
5085 
5086 	size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5087 	tx_ring->tx_buffer_info = vzalloc_node(size, tx_ring->numa_node);
5088 	if (!tx_ring->tx_buffer_info)
5089 		tx_ring->tx_buffer_info = vzalloc(size);
5090 	if (!tx_ring->tx_buffer_info)
5091 		goto err;
5092 
5093 	/* round up to nearest 4K */
5094 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5095 	tx_ring->size = ALIGN(tx_ring->size, 4096);
5096 
5097 	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5098 					   &tx_ring->dma, GFP_KERNEL);
5099 	if (!tx_ring->desc)
5100 		goto err;
5101 
5102 	tx_ring->next_to_use = 0;
5103 	tx_ring->next_to_clean = 0;
5104 	return 0;
5105 
5106 err:
5107 	vfree(tx_ring->tx_buffer_info);
5108 	tx_ring->tx_buffer_info = NULL;
5109 	dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5110 	return -ENOMEM;
5111 }
5112 
5113 /**
5114  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5115  * @adapter: board private structure
5116  *
5117  * If this function returns with an error, then it's possible one or
5118  * more of the rings is populated (while the rest are not).  It is the
5119  * callers duty to clean those orphaned rings.
5120  *
5121  * Return 0 on success, negative on failure
5122  **/
5123 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5124 {
5125 	int i, err = 0;
5126 
5127 	for (i = 0; i < adapter->num_tx_queues; i++) {
5128 		err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5129 		if (!err)
5130 			continue;
5131 		e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5132 		break;
5133 	}
5134 
5135 	return err;
5136 }
5137 
5138 /**
5139  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5140  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
5141  *
5142  * Returns 0 on success, negative on failure
5143  **/
5144 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5145 {
5146 	struct device *dev = rx_ring->dev;
5147 	int size;
5148 
5149 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5150 	rx_ring->rx_buffer_info = vzalloc_node(size, rx_ring->numa_node);
5151 	if (!rx_ring->rx_buffer_info)
5152 		rx_ring->rx_buffer_info = vzalloc(size);
5153 	if (!rx_ring->rx_buffer_info)
5154 		goto err;
5155 
5156 	/* Round up to nearest 4K */
5157 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5158 	rx_ring->size = ALIGN(rx_ring->size, 4096);
5159 
5160 	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5161 					   &rx_ring->dma, GFP_KERNEL);
5162 
5163 	if (!rx_ring->desc)
5164 		goto err;
5165 
5166 	rx_ring->next_to_clean = 0;
5167 	rx_ring->next_to_use = 0;
5168 
5169 	return 0;
5170 err:
5171 	vfree(rx_ring->rx_buffer_info);
5172 	rx_ring->rx_buffer_info = NULL;
5173 	dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5174 	return -ENOMEM;
5175 }
5176 
5177 /**
5178  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5179  * @adapter: board private structure
5180  *
5181  * If this function returns with an error, then it's possible one or
5182  * more of the rings is populated (while the rest are not).  It is the
5183  * callers duty to clean those orphaned rings.
5184  *
5185  * Return 0 on success, negative on failure
5186  **/
5187 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5188 {
5189 	int i, err = 0;
5190 
5191 	for (i = 0; i < adapter->num_rx_queues; i++) {
5192 		err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5193 		if (!err)
5194 			continue;
5195 		e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5196 		break;
5197 	}
5198 
5199 	return err;
5200 }
5201 
5202 /**
5203  * ixgbe_free_tx_resources - Free Tx Resources per Queue
5204  * @tx_ring: Tx descriptor ring for a specific queue
5205  *
5206  * Free all transmit software resources
5207  **/
5208 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5209 {
5210 	ixgbe_clean_tx_ring(tx_ring);
5211 
5212 	vfree(tx_ring->tx_buffer_info);
5213 	tx_ring->tx_buffer_info = NULL;
5214 
5215 	/* if not set, then don't free */
5216 	if (!tx_ring->desc)
5217 		return;
5218 
5219 	dma_free_coherent(tx_ring->dev, tx_ring->size,
5220 			  tx_ring->desc, tx_ring->dma);
5221 
5222 	tx_ring->desc = NULL;
5223 }
5224 
5225 /**
5226  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5227  * @adapter: board private structure
5228  *
5229  * Free all transmit software resources
5230  **/
5231 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5232 {
5233 	int i;
5234 
5235 	for (i = 0; i < adapter->num_tx_queues; i++)
5236 		if (adapter->tx_ring[i]->desc)
5237 			ixgbe_free_tx_resources(adapter->tx_ring[i]);
5238 }
5239 
5240 /**
5241  * ixgbe_free_rx_resources - Free Rx Resources
5242  * @rx_ring: ring to clean the resources from
5243  *
5244  * Free all receive software resources
5245  **/
5246 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5247 {
5248 	ixgbe_clean_rx_ring(rx_ring);
5249 
5250 	vfree(rx_ring->rx_buffer_info);
5251 	rx_ring->rx_buffer_info = NULL;
5252 
5253 	/* if not set, then don't free */
5254 	if (!rx_ring->desc)
5255 		return;
5256 
5257 	dma_free_coherent(rx_ring->dev, rx_ring->size,
5258 			  rx_ring->desc, rx_ring->dma);
5259 
5260 	rx_ring->desc = NULL;
5261 }
5262 
5263 /**
5264  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5265  * @adapter: board private structure
5266  *
5267  * Free all receive software resources
5268  **/
5269 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5270 {
5271 	int i;
5272 
5273 	for (i = 0; i < adapter->num_rx_queues; i++)
5274 		if (adapter->rx_ring[i]->desc)
5275 			ixgbe_free_rx_resources(adapter->rx_ring[i]);
5276 }
5277 
5278 /**
5279  * ixgbe_change_mtu - Change the Maximum Transfer Unit
5280  * @netdev: network interface device structure
5281  * @new_mtu: new value for maximum frame size
5282  *
5283  * Returns 0 on success, negative on failure
5284  **/
5285 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5286 {
5287 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
5288 	struct ixgbe_hw *hw = &adapter->hw;
5289 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5290 
5291 	/* MTU < 68 is an error and causes problems on some kernels */
5292 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED &&
5293 	    hw->mac.type != ixgbe_mac_X540) {
5294 		if ((new_mtu < 68) || (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
5295 			return -EINVAL;
5296 	} else {
5297 		if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5298 			return -EINVAL;
5299 	}
5300 
5301 	e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5302 	/* must set new MTU before calling down or up */
5303 	netdev->mtu = new_mtu;
5304 
5305 	if (netif_running(netdev))
5306 		ixgbe_reinit_locked(adapter);
5307 
5308 	return 0;
5309 }
5310 
5311 /**
5312  * ixgbe_open - Called when a network interface is made active
5313  * @netdev: network interface device structure
5314  *
5315  * Returns 0 on success, negative value on failure
5316  *
5317  * The open entry point is called when a network interface is made
5318  * active by the system (IFF_UP).  At this point all resources needed
5319  * for transmit and receive operations are allocated, the interrupt
5320  * handler is registered with the OS, the watchdog timer is started,
5321  * and the stack is notified that the interface is ready.
5322  **/
5323 static int ixgbe_open(struct net_device *netdev)
5324 {
5325 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
5326 	int err;
5327 
5328 	/* disallow open during test */
5329 	if (test_bit(__IXGBE_TESTING, &adapter->state))
5330 		return -EBUSY;
5331 
5332 	netif_carrier_off(netdev);
5333 
5334 	/* allocate transmit descriptors */
5335 	err = ixgbe_setup_all_tx_resources(adapter);
5336 	if (err)
5337 		goto err_setup_tx;
5338 
5339 	/* allocate receive descriptors */
5340 	err = ixgbe_setup_all_rx_resources(adapter);
5341 	if (err)
5342 		goto err_setup_rx;
5343 
5344 	ixgbe_configure(adapter);
5345 
5346 	err = ixgbe_request_irq(adapter);
5347 	if (err)
5348 		goto err_req_irq;
5349 
5350 	ixgbe_up_complete(adapter);
5351 
5352 	return 0;
5353 
5354 err_req_irq:
5355 err_setup_rx:
5356 	ixgbe_free_all_rx_resources(adapter);
5357 err_setup_tx:
5358 	ixgbe_free_all_tx_resources(adapter);
5359 	ixgbe_reset(adapter);
5360 
5361 	return err;
5362 }
5363 
5364 /**
5365  * ixgbe_close - Disables a network interface
5366  * @netdev: network interface device structure
5367  *
5368  * Returns 0, this is not allowed to fail
5369  *
5370  * The close entry point is called when an interface is de-activated
5371  * by the OS.  The hardware is still under the drivers control, but
5372  * needs to be disabled.  A global MAC reset is issued to stop the
5373  * hardware, and all transmit and receive resources are freed.
5374  **/
5375 static int ixgbe_close(struct net_device *netdev)
5376 {
5377 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
5378 
5379 	ixgbe_down(adapter);
5380 	ixgbe_free_irq(adapter);
5381 
5382 	ixgbe_fdir_filter_exit(adapter);
5383 
5384 	ixgbe_free_all_tx_resources(adapter);
5385 	ixgbe_free_all_rx_resources(adapter);
5386 
5387 	ixgbe_release_hw_control(adapter);
5388 
5389 	return 0;
5390 }
5391 
5392 #ifdef CONFIG_PM
5393 static int ixgbe_resume(struct pci_dev *pdev)
5394 {
5395 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5396 	struct net_device *netdev = adapter->netdev;
5397 	u32 err;
5398 
5399 	pci_set_power_state(pdev, PCI_D0);
5400 	pci_restore_state(pdev);
5401 	/*
5402 	 * pci_restore_state clears dev->state_saved so call
5403 	 * pci_save_state to restore it.
5404 	 */
5405 	pci_save_state(pdev);
5406 
5407 	err = pci_enable_device_mem(pdev);
5408 	if (err) {
5409 		e_dev_err("Cannot enable PCI device from suspend\n");
5410 		return err;
5411 	}
5412 	pci_set_master(pdev);
5413 
5414 	pci_wake_from_d3(pdev, false);
5415 
5416 	err = ixgbe_init_interrupt_scheme(adapter);
5417 	if (err) {
5418 		e_dev_err("Cannot initialize interrupts for device\n");
5419 		return err;
5420 	}
5421 
5422 	ixgbe_reset(adapter);
5423 
5424 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5425 
5426 	if (netif_running(netdev)) {
5427 		err = ixgbe_open(netdev);
5428 		if (err)
5429 			return err;
5430 	}
5431 
5432 	netif_device_attach(netdev);
5433 
5434 	return 0;
5435 }
5436 #endif /* CONFIG_PM */
5437 
5438 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5439 {
5440 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5441 	struct net_device *netdev = adapter->netdev;
5442 	struct ixgbe_hw *hw = &adapter->hw;
5443 	u32 ctrl, fctrl;
5444 	u32 wufc = adapter->wol;
5445 #ifdef CONFIG_PM
5446 	int retval = 0;
5447 #endif
5448 
5449 	netif_device_detach(netdev);
5450 
5451 	if (netif_running(netdev)) {
5452 		ixgbe_down(adapter);
5453 		ixgbe_free_irq(adapter);
5454 		ixgbe_free_all_tx_resources(adapter);
5455 		ixgbe_free_all_rx_resources(adapter);
5456 	}
5457 
5458 	ixgbe_clear_interrupt_scheme(adapter);
5459 #ifdef CONFIG_DCB
5460 	kfree(adapter->ixgbe_ieee_pfc);
5461 	kfree(adapter->ixgbe_ieee_ets);
5462 #endif
5463 
5464 #ifdef CONFIG_PM
5465 	retval = pci_save_state(pdev);
5466 	if (retval)
5467 		return retval;
5468 
5469 #endif
5470 	if (wufc) {
5471 		ixgbe_set_rx_mode(netdev);
5472 
5473 		/* turn on all-multi mode if wake on multicast is enabled */
5474 		if (wufc & IXGBE_WUFC_MC) {
5475 			fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5476 			fctrl |= IXGBE_FCTRL_MPE;
5477 			IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5478 		}
5479 
5480 		ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5481 		ctrl |= IXGBE_CTRL_GIO_DIS;
5482 		IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5483 
5484 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5485 	} else {
5486 		IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5487 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5488 	}
5489 
5490 	switch (hw->mac.type) {
5491 	case ixgbe_mac_82598EB:
5492 		pci_wake_from_d3(pdev, false);
5493 		break;
5494 	case ixgbe_mac_82599EB:
5495 	case ixgbe_mac_X540:
5496 		pci_wake_from_d3(pdev, !!wufc);
5497 		break;
5498 	default:
5499 		break;
5500 	}
5501 
5502 	*enable_wake = !!wufc;
5503 
5504 	ixgbe_release_hw_control(adapter);
5505 
5506 	pci_disable_device(pdev);
5507 
5508 	return 0;
5509 }
5510 
5511 #ifdef CONFIG_PM
5512 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5513 {
5514 	int retval;
5515 	bool wake;
5516 
5517 	retval = __ixgbe_shutdown(pdev, &wake);
5518 	if (retval)
5519 		return retval;
5520 
5521 	if (wake) {
5522 		pci_prepare_to_sleep(pdev);
5523 	} else {
5524 		pci_wake_from_d3(pdev, false);
5525 		pci_set_power_state(pdev, PCI_D3hot);
5526 	}
5527 
5528 	return 0;
5529 }
5530 #endif /* CONFIG_PM */
5531 
5532 static void ixgbe_shutdown(struct pci_dev *pdev)
5533 {
5534 	bool wake;
5535 
5536 	__ixgbe_shutdown(pdev, &wake);
5537 
5538 	if (system_state == SYSTEM_POWER_OFF) {
5539 		pci_wake_from_d3(pdev, wake);
5540 		pci_set_power_state(pdev, PCI_D3hot);
5541 	}
5542 }
5543 
5544 /**
5545  * ixgbe_update_stats - Update the board statistics counters.
5546  * @adapter: board private structure
5547  **/
5548 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5549 {
5550 	struct net_device *netdev = adapter->netdev;
5551 	struct ixgbe_hw *hw = &adapter->hw;
5552 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
5553 	u64 total_mpc = 0;
5554 	u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5555 	u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5556 	u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5557 	u64 bytes = 0, packets = 0;
5558 #ifdef IXGBE_FCOE
5559 	struct ixgbe_fcoe *fcoe = &adapter->fcoe;
5560 	unsigned int cpu;
5561 	u64 fcoe_noddp_counts_sum = 0, fcoe_noddp_ext_buff_counts_sum = 0;
5562 #endif /* IXGBE_FCOE */
5563 
5564 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5565 	    test_bit(__IXGBE_RESETTING, &adapter->state))
5566 		return;
5567 
5568 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5569 		u64 rsc_count = 0;
5570 		u64 rsc_flush = 0;
5571 		for (i = 0; i < 16; i++)
5572 			adapter->hw_rx_no_dma_resources +=
5573 				IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5574 		for (i = 0; i < adapter->num_rx_queues; i++) {
5575 			rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5576 			rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5577 		}
5578 		adapter->rsc_total_count = rsc_count;
5579 		adapter->rsc_total_flush = rsc_flush;
5580 	}
5581 
5582 	for (i = 0; i < adapter->num_rx_queues; i++) {
5583 		struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5584 		non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5585 		alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5586 		alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5587 		bytes += rx_ring->stats.bytes;
5588 		packets += rx_ring->stats.packets;
5589 	}
5590 	adapter->non_eop_descs = non_eop_descs;
5591 	adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5592 	adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5593 	netdev->stats.rx_bytes = bytes;
5594 	netdev->stats.rx_packets = packets;
5595 
5596 	bytes = 0;
5597 	packets = 0;
5598 	/* gather some stats to the adapter struct that are per queue */
5599 	for (i = 0; i < adapter->num_tx_queues; i++) {
5600 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5601 		restart_queue += tx_ring->tx_stats.restart_queue;
5602 		tx_busy += tx_ring->tx_stats.tx_busy;
5603 		bytes += tx_ring->stats.bytes;
5604 		packets += tx_ring->stats.packets;
5605 	}
5606 	adapter->restart_queue = restart_queue;
5607 	adapter->tx_busy = tx_busy;
5608 	netdev->stats.tx_bytes = bytes;
5609 	netdev->stats.tx_packets = packets;
5610 
5611 	hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5612 
5613 	/* 8 register reads */
5614 	for (i = 0; i < 8; i++) {
5615 		/* for packet buffers not used, the register should read 0 */
5616 		mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5617 		missed_rx += mpc;
5618 		hwstats->mpc[i] += mpc;
5619 		total_mpc += hwstats->mpc[i];
5620 		hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5621 		hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5622 		switch (hw->mac.type) {
5623 		case ixgbe_mac_82598EB:
5624 			hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5625 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5626 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5627 			hwstats->pxonrxc[i] +=
5628 				IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5629 			break;
5630 		case ixgbe_mac_82599EB:
5631 		case ixgbe_mac_X540:
5632 			hwstats->pxonrxc[i] +=
5633 				IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5634 			break;
5635 		default:
5636 			break;
5637 		}
5638 	}
5639 
5640 	/*16 register reads */
5641 	for (i = 0; i < 16; i++) {
5642 		hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5643 		hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5644 		if ((hw->mac.type == ixgbe_mac_82599EB) ||
5645 		    (hw->mac.type == ixgbe_mac_X540)) {
5646 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
5647 			IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
5648 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
5649 			IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
5650 		}
5651 	}
5652 
5653 	hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5654 	/* work around hardware counting issue */
5655 	hwstats->gprc -= missed_rx;
5656 
5657 	ixgbe_update_xoff_received(adapter);
5658 
5659 	/* 82598 hardware only has a 32 bit counter in the high register */
5660 	switch (hw->mac.type) {
5661 	case ixgbe_mac_82598EB:
5662 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5663 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5664 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5665 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5666 		break;
5667 	case ixgbe_mac_X540:
5668 		/* OS2BMC stats are X540 only*/
5669 		hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5670 		hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5671 		hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5672 		hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5673 	case ixgbe_mac_82599EB:
5674 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5675 		IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5676 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5677 		IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5678 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5679 		IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5680 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5681 		hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5682 		hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5683 #ifdef IXGBE_FCOE
5684 		hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5685 		hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5686 		hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5687 		hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5688 		hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5689 		hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5690 		/* Add up per cpu counters for total ddp aloc fail */
5691 		if (fcoe->pcpu_noddp && fcoe->pcpu_noddp_ext_buff) {
5692 			for_each_possible_cpu(cpu) {
5693 				fcoe_noddp_counts_sum +=
5694 					*per_cpu_ptr(fcoe->pcpu_noddp, cpu);
5695 				fcoe_noddp_ext_buff_counts_sum +=
5696 					*per_cpu_ptr(fcoe->
5697 						pcpu_noddp_ext_buff, cpu);
5698 			}
5699 		}
5700 		hwstats->fcoe_noddp = fcoe_noddp_counts_sum;
5701 		hwstats->fcoe_noddp_ext_buff = fcoe_noddp_ext_buff_counts_sum;
5702 #endif /* IXGBE_FCOE */
5703 		break;
5704 	default:
5705 		break;
5706 	}
5707 	bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5708 	hwstats->bprc += bprc;
5709 	hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5710 	if (hw->mac.type == ixgbe_mac_82598EB)
5711 		hwstats->mprc -= bprc;
5712 	hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5713 	hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5714 	hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5715 	hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5716 	hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5717 	hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5718 	hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5719 	hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5720 	lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5721 	hwstats->lxontxc += lxon;
5722 	lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5723 	hwstats->lxofftxc += lxoff;
5724 	hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5725 	hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5726 	/*
5727 	 * 82598 errata - tx of flow control packets is included in tx counters
5728 	 */
5729 	xon_off_tot = lxon + lxoff;
5730 	hwstats->gptc -= xon_off_tot;
5731 	hwstats->mptc -= xon_off_tot;
5732 	hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5733 	hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5734 	hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5735 	hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5736 	hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5737 	hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5738 	hwstats->ptc64 -= xon_off_tot;
5739 	hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5740 	hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5741 	hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5742 	hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5743 	hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5744 	hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5745 
5746 	/* Fill out the OS statistics structure */
5747 	netdev->stats.multicast = hwstats->mprc;
5748 
5749 	/* Rx Errors */
5750 	netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5751 	netdev->stats.rx_dropped = 0;
5752 	netdev->stats.rx_length_errors = hwstats->rlec;
5753 	netdev->stats.rx_crc_errors = hwstats->crcerrs;
5754 	netdev->stats.rx_missed_errors = total_mpc;
5755 }
5756 
5757 /**
5758  * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5759  * @adapter - pointer to the device adapter structure
5760  **/
5761 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5762 {
5763 	struct ixgbe_hw *hw = &adapter->hw;
5764 	int i;
5765 
5766 	if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5767 		return;
5768 
5769 	adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5770 
5771 	/* if interface is down do nothing */
5772 	if (test_bit(__IXGBE_DOWN, &adapter->state))
5773 		return;
5774 
5775 	/* do nothing if we are not using signature filters */
5776 	if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5777 		return;
5778 
5779 	adapter->fdir_overflow++;
5780 
5781 	if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5782 		for (i = 0; i < adapter->num_tx_queues; i++)
5783 			set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5784 			        &(adapter->tx_ring[i]->state));
5785 		/* re-enable flow director interrupts */
5786 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5787 	} else {
5788 		e_err(probe, "failed to finish FDIR re-initialization, "
5789 		      "ignored adding FDIR ATR filters\n");
5790 	}
5791 }
5792 
5793 /**
5794  * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5795  * @adapter - pointer to the device adapter structure
5796  *
5797  * This function serves two purposes.  First it strobes the interrupt lines
5798  * in order to make certain interrupts are occurring.  Secondly it sets the
5799  * bits needed to check for TX hangs.  As a result we should immediately
5800  * determine if a hang has occurred.
5801  */
5802 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5803 {
5804 	struct ixgbe_hw *hw = &adapter->hw;
5805 	u64 eics = 0;
5806 	int i;
5807 
5808 	/* If we're down or resetting, just bail */
5809 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5810 	    test_bit(__IXGBE_RESETTING, &adapter->state))
5811 		return;
5812 
5813 	/* Force detection of hung controller */
5814 	if (netif_carrier_ok(adapter->netdev)) {
5815 		for (i = 0; i < adapter->num_tx_queues; i++)
5816 			set_check_for_tx_hang(adapter->tx_ring[i]);
5817 	}
5818 
5819 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5820 		/*
5821 		 * for legacy and MSI interrupts don't set any bits
5822 		 * that are enabled for EIAM, because this operation
5823 		 * would set *both* EIMS and EICS for any bit in EIAM
5824 		 */
5825 		IXGBE_WRITE_REG(hw, IXGBE_EICS,
5826 			(IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5827 	} else {
5828 		/* get one bit for every active tx/rx interrupt vector */
5829 		for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5830 			struct ixgbe_q_vector *qv = adapter->q_vector[i];
5831 			if (qv->rx.ring || qv->tx.ring)
5832 				eics |= ((u64)1 << i);
5833 		}
5834 	}
5835 
5836 	/* Cause software interrupt to ensure rings are cleaned */
5837 	ixgbe_irq_rearm_queues(adapter, eics);
5838 
5839 }
5840 
5841 /**
5842  * ixgbe_watchdog_update_link - update the link status
5843  * @adapter - pointer to the device adapter structure
5844  * @link_speed - pointer to a u32 to store the link_speed
5845  **/
5846 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
5847 {
5848 	struct ixgbe_hw *hw = &adapter->hw;
5849 	u32 link_speed = adapter->link_speed;
5850 	bool link_up = adapter->link_up;
5851 	int i;
5852 
5853 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5854 		return;
5855 
5856 	if (hw->mac.ops.check_link) {
5857 		hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5858 	} else {
5859 		/* always assume link is up, if no check link function */
5860 		link_speed = IXGBE_LINK_SPEED_10GB_FULL;
5861 		link_up = true;
5862 	}
5863 	if (link_up) {
5864 		if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5865 			for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5866 				hw->mac.ops.fc_enable(hw, i);
5867 		} else {
5868 			hw->mac.ops.fc_enable(hw, 0);
5869 		}
5870 	}
5871 
5872 	if (link_up ||
5873 	    time_after(jiffies, (adapter->link_check_timeout +
5874 				 IXGBE_TRY_LINK_TIMEOUT))) {
5875 		adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5876 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5877 		IXGBE_WRITE_FLUSH(hw);
5878 	}
5879 
5880 	adapter->link_up = link_up;
5881 	adapter->link_speed = link_speed;
5882 }
5883 
5884 /**
5885  * ixgbe_watchdog_link_is_up - update netif_carrier status and
5886  *                             print link up message
5887  * @adapter - pointer to the device adapter structure
5888  **/
5889 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
5890 {
5891 	struct net_device *netdev = adapter->netdev;
5892 	struct ixgbe_hw *hw = &adapter->hw;
5893 	u32 link_speed = adapter->link_speed;
5894 	bool flow_rx, flow_tx;
5895 
5896 	/* only continue if link was previously down */
5897 	if (netif_carrier_ok(netdev))
5898 		return;
5899 
5900 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
5901 
5902 	switch (hw->mac.type) {
5903 	case ixgbe_mac_82598EB: {
5904 		u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5905 		u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5906 		flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5907 		flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5908 	}
5909 		break;
5910 	case ixgbe_mac_X540:
5911 	case ixgbe_mac_82599EB: {
5912 		u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5913 		u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5914 		flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5915 		flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5916 	}
5917 		break;
5918 	default:
5919 		flow_tx = false;
5920 		flow_rx = false;
5921 		break;
5922 	}
5923 	e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5924 	       (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5925 	       "10 Gbps" :
5926 	       (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5927 	       "1 Gbps" :
5928 	       (link_speed == IXGBE_LINK_SPEED_100_FULL ?
5929 	       "100 Mbps" :
5930 	       "unknown speed"))),
5931 	       ((flow_rx && flow_tx) ? "RX/TX" :
5932 	       (flow_rx ? "RX" :
5933 	       (flow_tx ? "TX" : "None"))));
5934 
5935 	netif_carrier_on(netdev);
5936 	ixgbe_check_vf_rate_limit(adapter);
5937 }
5938 
5939 /**
5940  * ixgbe_watchdog_link_is_down - update netif_carrier status and
5941  *                               print link down message
5942  * @adapter - pointer to the adapter structure
5943  **/
5944 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter* adapter)
5945 {
5946 	struct net_device *netdev = adapter->netdev;
5947 	struct ixgbe_hw *hw = &adapter->hw;
5948 
5949 	adapter->link_up = false;
5950 	adapter->link_speed = 0;
5951 
5952 	/* only continue if link was up previously */
5953 	if (!netif_carrier_ok(netdev))
5954 		return;
5955 
5956 	/* poll for SFP+ cable when link is down */
5957 	if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
5958 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5959 
5960 	e_info(drv, "NIC Link is Down\n");
5961 	netif_carrier_off(netdev);
5962 }
5963 
5964 /**
5965  * ixgbe_watchdog_flush_tx - flush queues on link down
5966  * @adapter - pointer to the device adapter structure
5967  **/
5968 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
5969 {
5970 	int i;
5971 	int some_tx_pending = 0;
5972 
5973 	if (!netif_carrier_ok(adapter->netdev)) {
5974 		for (i = 0; i < adapter->num_tx_queues; i++) {
5975 			struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5976 			if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5977 				some_tx_pending = 1;
5978 				break;
5979 			}
5980 		}
5981 
5982 		if (some_tx_pending) {
5983 			/* We've lost link, so the controller stops DMA,
5984 			 * but we've got queued Tx work that's never going
5985 			 * to get done, so reset controller to flush Tx.
5986 			 * (Do the reset outside of interrupt context).
5987 			 */
5988 			adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
5989 		}
5990 	}
5991 }
5992 
5993 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
5994 {
5995 	u32 ssvpc;
5996 
5997 	/* Do not perform spoof check for 82598 */
5998 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5999 		return;
6000 
6001 	ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
6002 
6003 	/*
6004 	 * ssvpc register is cleared on read, if zero then no
6005 	 * spoofed packets in the last interval.
6006 	 */
6007 	if (!ssvpc)
6008 		return;
6009 
6010 	e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
6011 }
6012 
6013 /**
6014  * ixgbe_watchdog_subtask - check and bring link up
6015  * @adapter - pointer to the device adapter structure
6016  **/
6017 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
6018 {
6019 	/* if interface is down do nothing */
6020 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6021 	    test_bit(__IXGBE_RESETTING, &adapter->state))
6022 		return;
6023 
6024 	ixgbe_watchdog_update_link(adapter);
6025 
6026 	if (adapter->link_up)
6027 		ixgbe_watchdog_link_is_up(adapter);
6028 	else
6029 		ixgbe_watchdog_link_is_down(adapter);
6030 
6031 	ixgbe_spoof_check(adapter);
6032 	ixgbe_update_stats(adapter);
6033 
6034 	ixgbe_watchdog_flush_tx(adapter);
6035 }
6036 
6037 /**
6038  * ixgbe_sfp_detection_subtask - poll for SFP+ cable
6039  * @adapter - the ixgbe adapter structure
6040  **/
6041 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
6042 {
6043 	struct ixgbe_hw *hw = &adapter->hw;
6044 	s32 err;
6045 
6046 	/* not searching for SFP so there is nothing to do here */
6047 	if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
6048 	    !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6049 		return;
6050 
6051 	/* someone else is in init, wait until next service event */
6052 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6053 		return;
6054 
6055 	err = hw->phy.ops.identify_sfp(hw);
6056 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6057 		goto sfp_out;
6058 
6059 	if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
6060 		/* If no cable is present, then we need to reset
6061 		 * the next time we find a good cable. */
6062 		adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
6063 	}
6064 
6065 	/* exit on error */
6066 	if (err)
6067 		goto sfp_out;
6068 
6069 	/* exit if reset not needed */
6070 	if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6071 		goto sfp_out;
6072 
6073 	adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
6074 
6075 	/*
6076 	 * A module may be identified correctly, but the EEPROM may not have
6077 	 * support for that module.  setup_sfp() will fail in that case, so
6078 	 * we should not allow that module to load.
6079 	 */
6080 	if (hw->mac.type == ixgbe_mac_82598EB)
6081 		err = hw->phy.ops.reset(hw);
6082 	else
6083 		err = hw->mac.ops.setup_sfp(hw);
6084 
6085 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6086 		goto sfp_out;
6087 
6088 	adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
6089 	e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
6090 
6091 sfp_out:
6092 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6093 
6094 	if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
6095 	    (adapter->netdev->reg_state == NETREG_REGISTERED)) {
6096 		e_dev_err("failed to initialize because an unsupported "
6097 			  "SFP+ module type was detected.\n");
6098 		e_dev_err("Reload the driver after installing a "
6099 			  "supported module.\n");
6100 		unregister_netdev(adapter->netdev);
6101 	}
6102 }
6103 
6104 /**
6105  * ixgbe_sfp_link_config_subtask - set up link SFP after module install
6106  * @adapter - the ixgbe adapter structure
6107  **/
6108 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
6109 {
6110 	struct ixgbe_hw *hw = &adapter->hw;
6111 	u32 autoneg;
6112 	bool negotiation;
6113 
6114 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
6115 		return;
6116 
6117 	/* someone else is in init, wait until next service event */
6118 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6119 		return;
6120 
6121 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
6122 
6123 	autoneg = hw->phy.autoneg_advertised;
6124 	if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
6125 		hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
6126 	if (hw->mac.ops.setup_link)
6127 		hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
6128 
6129 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
6130 	adapter->link_check_timeout = jiffies;
6131 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6132 }
6133 
6134 #ifdef CONFIG_PCI_IOV
6135 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
6136 {
6137 	int vf;
6138 	struct ixgbe_hw *hw = &adapter->hw;
6139 	struct net_device *netdev = adapter->netdev;
6140 	u32 gpc;
6141 	u32 ciaa, ciad;
6142 
6143 	gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
6144 	if (gpc) /* If incrementing then no need for the check below */
6145 		return;
6146 	/*
6147 	 * Check to see if a bad DMA write target from an errant or
6148 	 * malicious VF has caused a PCIe error.  If so then we can
6149 	 * issue a VFLR to the offending VF(s) and then resume without
6150 	 * requesting a full slot reset.
6151 	 */
6152 
6153 	for (vf = 0; vf < adapter->num_vfs; vf++) {
6154 		ciaa = (vf << 16) | 0x80000000;
6155 		/* 32 bit read so align, we really want status at offset 6 */
6156 		ciaa |= PCI_COMMAND;
6157 		IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6158 		ciad = IXGBE_READ_REG(hw, IXGBE_CIAD_82599);
6159 		ciaa &= 0x7FFFFFFF;
6160 		/* disable debug mode asap after reading data */
6161 		IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6162 		/* Get the upper 16 bits which will be the PCI status reg */
6163 		ciad >>= 16;
6164 		if (ciad & PCI_STATUS_REC_MASTER_ABORT) {
6165 			netdev_err(netdev, "VF %d Hung DMA\n", vf);
6166 			/* Issue VFLR */
6167 			ciaa = (vf << 16) | 0x80000000;
6168 			ciaa |= 0xA8;
6169 			IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6170 			ciad = 0x00008000;  /* VFLR */
6171 			IXGBE_WRITE_REG(hw, IXGBE_CIAD_82599, ciad);
6172 			ciaa &= 0x7FFFFFFF;
6173 			IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6174 		}
6175 	}
6176 }
6177 
6178 #endif
6179 /**
6180  * ixgbe_service_timer - Timer Call-back
6181  * @data: pointer to adapter cast into an unsigned long
6182  **/
6183 static void ixgbe_service_timer(unsigned long data)
6184 {
6185 	struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
6186 	unsigned long next_event_offset;
6187 	bool ready = true;
6188 
6189 #ifdef CONFIG_PCI_IOV
6190 	ready = false;
6191 
6192 	/*
6193 	 * don't bother with SR-IOV VF DMA hang check if there are
6194 	 * no VFs or the link is down
6195 	 */
6196 	if (!adapter->num_vfs ||
6197 	    (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)) {
6198 		ready = true;
6199 		goto normal_timer_service;
6200 	}
6201 
6202 	/* If we have VFs allocated then we must check for DMA hangs */
6203 	ixgbe_check_for_bad_vf(adapter);
6204 	next_event_offset = HZ / 50;
6205 	adapter->timer_event_accumulator++;
6206 
6207 	if (adapter->timer_event_accumulator >= 100) {
6208 		ready = true;
6209 		adapter->timer_event_accumulator = 0;
6210 	}
6211 
6212 	goto schedule_event;
6213 
6214 normal_timer_service:
6215 #endif
6216 	/* poll faster when waiting for link */
6217 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
6218 		next_event_offset = HZ / 10;
6219 	else
6220 		next_event_offset = HZ * 2;
6221 
6222 #ifdef CONFIG_PCI_IOV
6223 schedule_event:
6224 #endif
6225 	/* Reset the timer */
6226 	mod_timer(&adapter->service_timer, next_event_offset + jiffies);
6227 
6228 	if (ready)
6229 		ixgbe_service_event_schedule(adapter);
6230 }
6231 
6232 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
6233 {
6234 	if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
6235 		return;
6236 
6237 	adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
6238 
6239 	/* If we're already down or resetting, just bail */
6240 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6241 	    test_bit(__IXGBE_RESETTING, &adapter->state))
6242 		return;
6243 
6244 	ixgbe_dump(adapter);
6245 	netdev_err(adapter->netdev, "Reset adapter\n");
6246 	adapter->tx_timeout_count++;
6247 
6248 	ixgbe_reinit_locked(adapter);
6249 }
6250 
6251 /**
6252  * ixgbe_service_task - manages and runs subtasks
6253  * @work: pointer to work_struct containing our data
6254  **/
6255 static void ixgbe_service_task(struct work_struct *work)
6256 {
6257 	struct ixgbe_adapter *adapter = container_of(work,
6258 						     struct ixgbe_adapter,
6259 						     service_task);
6260 
6261 	ixgbe_reset_subtask(adapter);
6262 	ixgbe_sfp_detection_subtask(adapter);
6263 	ixgbe_sfp_link_config_subtask(adapter);
6264 	ixgbe_check_overtemp_subtask(adapter);
6265 	ixgbe_watchdog_subtask(adapter);
6266 	ixgbe_fdir_reinit_subtask(adapter);
6267 	ixgbe_check_hang_subtask(adapter);
6268 
6269 	ixgbe_service_event_complete(adapter);
6270 }
6271 
6272 void ixgbe_tx_ctxtdesc(struct ixgbe_ring *tx_ring, u32 vlan_macip_lens,
6273 		       u32 fcoe_sof_eof, u32 type_tucmd, u32 mss_l4len_idx)
6274 {
6275 	struct ixgbe_adv_tx_context_desc *context_desc;
6276 	u16 i = tx_ring->next_to_use;
6277 
6278 	context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6279 
6280 	i++;
6281 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
6282 
6283 	/* set bits to identify this as an advanced context descriptor */
6284 	type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
6285 
6286 	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
6287 	context_desc->seqnum_seed	= cpu_to_le32(fcoe_sof_eof);
6288 	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
6289 	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
6290 }
6291 
6292 static int ixgbe_tso(struct ixgbe_ring *tx_ring, struct sk_buff *skb,
6293 		     u32 tx_flags, __be16 protocol, u8 *hdr_len)
6294 {
6295 	int err;
6296 	u32 vlan_macip_lens, type_tucmd;
6297 	u32 mss_l4len_idx, l4len;
6298 
6299 	if (!skb_is_gso(skb))
6300 		return 0;
6301 
6302 	if (skb_header_cloned(skb)) {
6303 		err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
6304 		if (err)
6305 			return err;
6306 	}
6307 
6308 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6309 	type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
6310 
6311 	if (protocol == __constant_htons(ETH_P_IP)) {
6312 		struct iphdr *iph = ip_hdr(skb);
6313 		iph->tot_len = 0;
6314 		iph->check = 0;
6315 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6316 							 iph->daddr, 0,
6317 							 IPPROTO_TCP,
6318 							 0);
6319 		type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6320 	} else if (skb_is_gso_v6(skb)) {
6321 		ipv6_hdr(skb)->payload_len = 0;
6322 		tcp_hdr(skb)->check =
6323 		    ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6324 				     &ipv6_hdr(skb)->daddr,
6325 				     0, IPPROTO_TCP, 0);
6326 	}
6327 
6328 	l4len = tcp_hdrlen(skb);
6329 	*hdr_len = skb_transport_offset(skb) + l4len;
6330 
6331 	/* mss_l4len_id: use 1 as index for TSO */
6332 	mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
6333 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
6334 	mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
6335 
6336 	/* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
6337 	vlan_macip_lens = skb_network_header_len(skb);
6338 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6339 	vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6340 
6341 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
6342 	                  mss_l4len_idx);
6343 
6344 	return 1;
6345 }
6346 
6347 static bool ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
6348 			  struct sk_buff *skb, u32 tx_flags,
6349 			  __be16 protocol)
6350 {
6351 	u32 vlan_macip_lens = 0;
6352 	u32 mss_l4len_idx = 0;
6353 	u32 type_tucmd = 0;
6354 
6355 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
6356 	    if (!(tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
6357 		!(tx_flags & IXGBE_TX_FLAGS_TXSW))
6358 			return false;
6359 	} else {
6360 		u8 l4_hdr = 0;
6361 		switch (protocol) {
6362 		case __constant_htons(ETH_P_IP):
6363 			vlan_macip_lens |= skb_network_header_len(skb);
6364 			type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6365 			l4_hdr = ip_hdr(skb)->protocol;
6366 			break;
6367 		case __constant_htons(ETH_P_IPV6):
6368 			vlan_macip_lens |= skb_network_header_len(skb);
6369 			l4_hdr = ipv6_hdr(skb)->nexthdr;
6370 			break;
6371 		default:
6372 			if (unlikely(net_ratelimit())) {
6373 				dev_warn(tx_ring->dev,
6374 				 "partial checksum but proto=%x!\n",
6375 				 skb->protocol);
6376 			}
6377 			break;
6378 		}
6379 
6380 		switch (l4_hdr) {
6381 		case IPPROTO_TCP:
6382 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6383 			mss_l4len_idx = tcp_hdrlen(skb) <<
6384 					IXGBE_ADVTXD_L4LEN_SHIFT;
6385 			break;
6386 		case IPPROTO_SCTP:
6387 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6388 			mss_l4len_idx = sizeof(struct sctphdr) <<
6389 					IXGBE_ADVTXD_L4LEN_SHIFT;
6390 			break;
6391 		case IPPROTO_UDP:
6392 			mss_l4len_idx = sizeof(struct udphdr) <<
6393 					IXGBE_ADVTXD_L4LEN_SHIFT;
6394 			break;
6395 		default:
6396 			if (unlikely(net_ratelimit())) {
6397 				dev_warn(tx_ring->dev,
6398 				 "partial checksum but l4 proto=%x!\n",
6399 				 skb->protocol);
6400 			}
6401 			break;
6402 		}
6403 	}
6404 
6405 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6406 	vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6407 
6408 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
6409 			  type_tucmd, mss_l4len_idx);
6410 
6411 	return (skb->ip_summed == CHECKSUM_PARTIAL);
6412 }
6413 
6414 static __le32 ixgbe_tx_cmd_type(u32 tx_flags)
6415 {
6416 	/* set type for advanced descriptor with frame checksum insertion */
6417 	__le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
6418 				      IXGBE_ADVTXD_DCMD_IFCS |
6419 				      IXGBE_ADVTXD_DCMD_DEXT);
6420 
6421 	/* set HW vlan bit if vlan is present */
6422 	if (tx_flags & IXGBE_TX_FLAGS_HW_VLAN)
6423 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
6424 
6425 	/* set segmentation enable bits for TSO/FSO */
6426 #ifdef IXGBE_FCOE
6427 	if ((tx_flags & IXGBE_TX_FLAGS_TSO) || (tx_flags & IXGBE_TX_FLAGS_FSO))
6428 #else
6429 	if (tx_flags & IXGBE_TX_FLAGS_TSO)
6430 #endif
6431 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
6432 
6433 	return cmd_type;
6434 }
6435 
6436 static __le32 ixgbe_tx_olinfo_status(u32 tx_flags, unsigned int paylen)
6437 {
6438 	__le32 olinfo_status =
6439 		cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
6440 
6441 	if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6442 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM |
6443 					    (1 << IXGBE_ADVTXD_IDX_SHIFT));
6444 		/* enble IPv4 checksum for TSO */
6445 		if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6446 			olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
6447 	}
6448 
6449 	/* enable L4 checksum for TSO and TX checksum offload */
6450 	if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6451 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
6452 
6453 #ifdef IXGBE_FCOE
6454 	/* use index 1 context for FCOE/FSO */
6455 	if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6456 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC |
6457 					    (1 << IXGBE_ADVTXD_IDX_SHIFT));
6458 
6459 #endif
6460 	/*
6461 	 * Check Context must be set if Tx switch is enabled, which it
6462 	 * always is for case where virtual functions are running
6463 	 */
6464 	if (tx_flags & IXGBE_TX_FLAGS_TXSW)
6465 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
6466 
6467 	return olinfo_status;
6468 }
6469 
6470 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
6471 		       IXGBE_TXD_CMD_RS)
6472 
6473 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
6474 			 struct sk_buff *skb,
6475 			 struct ixgbe_tx_buffer *first,
6476 			 u32 tx_flags,
6477 			 const u8 hdr_len)
6478 {
6479 	struct device *dev = tx_ring->dev;
6480 	struct ixgbe_tx_buffer *tx_buffer_info;
6481 	union ixgbe_adv_tx_desc *tx_desc;
6482 	dma_addr_t dma;
6483 	__le32 cmd_type, olinfo_status;
6484 	struct skb_frag_struct *frag;
6485 	unsigned int f = 0;
6486 	unsigned int data_len = skb->data_len;
6487 	unsigned int size = skb_headlen(skb);
6488 	u32 offset = 0;
6489 	u32 paylen = skb->len - hdr_len;
6490 	u16 i = tx_ring->next_to_use;
6491 	u16 gso_segs;
6492 
6493 #ifdef IXGBE_FCOE
6494 	if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6495 		if (data_len >= sizeof(struct fcoe_crc_eof)) {
6496 			data_len -= sizeof(struct fcoe_crc_eof);
6497 		} else {
6498 			size -= sizeof(struct fcoe_crc_eof) - data_len;
6499 			data_len = 0;
6500 		}
6501 	}
6502 
6503 #endif
6504 	dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
6505 	if (dma_mapping_error(dev, dma))
6506 		goto dma_error;
6507 
6508 	cmd_type = ixgbe_tx_cmd_type(tx_flags);
6509 	olinfo_status = ixgbe_tx_olinfo_status(tx_flags, paylen);
6510 
6511 	tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6512 
6513 	for (;;) {
6514 		while (size > IXGBE_MAX_DATA_PER_TXD) {
6515 			tx_desc->read.buffer_addr = cpu_to_le64(dma + offset);
6516 			tx_desc->read.cmd_type_len =
6517 				cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
6518 			tx_desc->read.olinfo_status = olinfo_status;
6519 
6520 			offset += IXGBE_MAX_DATA_PER_TXD;
6521 			size -= IXGBE_MAX_DATA_PER_TXD;
6522 
6523 			tx_desc++;
6524 			i++;
6525 			if (i == tx_ring->count) {
6526 				tx_desc = IXGBE_TX_DESC_ADV(tx_ring, 0);
6527 				i = 0;
6528 			}
6529 		}
6530 
6531 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
6532 		tx_buffer_info->length = offset + size;
6533 		tx_buffer_info->tx_flags = tx_flags;
6534 		tx_buffer_info->dma = dma;
6535 
6536 		tx_desc->read.buffer_addr = cpu_to_le64(dma + offset);
6537 		tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
6538 		tx_desc->read.olinfo_status = olinfo_status;
6539 
6540 		if (!data_len)
6541 			break;
6542 
6543 		frag = &skb_shinfo(skb)->frags[f];
6544 #ifdef IXGBE_FCOE
6545 		size = min_t(unsigned int, data_len, skb_frag_size(frag));
6546 #else
6547 		size = skb_frag_size(frag);
6548 #endif
6549 		data_len -= size;
6550 		f++;
6551 
6552 		offset = 0;
6553 		tx_flags |= IXGBE_TX_FLAGS_MAPPED_AS_PAGE;
6554 
6555 		dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
6556 		if (dma_mapping_error(dev, dma))
6557 			goto dma_error;
6558 
6559 		tx_desc++;
6560 		i++;
6561 		if (i == tx_ring->count) {
6562 			tx_desc = IXGBE_TX_DESC_ADV(tx_ring, 0);
6563 			i = 0;
6564 		}
6565 	}
6566 
6567 	tx_desc->read.cmd_type_len |= cpu_to_le32(IXGBE_TXD_CMD);
6568 
6569 	i++;
6570 	if (i == tx_ring->count)
6571 		i = 0;
6572 
6573 	tx_ring->next_to_use = i;
6574 
6575 	if (tx_flags & IXGBE_TX_FLAGS_TSO)
6576 		gso_segs = skb_shinfo(skb)->gso_segs;
6577 #ifdef IXGBE_FCOE
6578 	/* adjust for FCoE Sequence Offload */
6579 	else if (tx_flags & IXGBE_TX_FLAGS_FSO)
6580 		gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
6581 					skb_shinfo(skb)->gso_size);
6582 #endif /* IXGBE_FCOE */
6583 	else
6584 		gso_segs = 1;
6585 
6586 	/* multiply data chunks by size of headers */
6587 	tx_buffer_info->bytecount = paylen + (gso_segs * hdr_len);
6588 	tx_buffer_info->gso_segs = gso_segs;
6589 	tx_buffer_info->skb = skb;
6590 
6591 	/* set the timestamp */
6592 	first->time_stamp = jiffies;
6593 
6594 	/*
6595 	 * Force memory writes to complete before letting h/w
6596 	 * know there are new descriptors to fetch.  (Only
6597 	 * applicable for weak-ordered memory model archs,
6598 	 * such as IA-64).
6599 	 */
6600 	wmb();
6601 
6602 	/* set next_to_watch value indicating a packet is present */
6603 	first->next_to_watch = tx_desc;
6604 
6605 	/* notify HW of packet */
6606 	writel(i, tx_ring->tail);
6607 
6608 	return;
6609 dma_error:
6610 	dev_err(dev, "TX DMA map failed\n");
6611 
6612 	/* clear dma mappings for failed tx_buffer_info map */
6613 	for (;;) {
6614 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
6615 		ixgbe_unmap_tx_resource(tx_ring, tx_buffer_info);
6616 		if (tx_buffer_info == first)
6617 			break;
6618 		if (i == 0)
6619 			i = tx_ring->count;
6620 		i--;
6621 	}
6622 
6623 	dev_kfree_skb_any(skb);
6624 
6625 	tx_ring->next_to_use = i;
6626 }
6627 
6628 static void ixgbe_atr(struct ixgbe_ring *ring, struct sk_buff *skb,
6629 		      u32 tx_flags, __be16 protocol)
6630 {
6631 	struct ixgbe_q_vector *q_vector = ring->q_vector;
6632 	union ixgbe_atr_hash_dword input = { .dword = 0 };
6633 	union ixgbe_atr_hash_dword common = { .dword = 0 };
6634 	union {
6635 		unsigned char *network;
6636 		struct iphdr *ipv4;
6637 		struct ipv6hdr *ipv6;
6638 	} hdr;
6639 	struct tcphdr *th;
6640 	__be16 vlan_id;
6641 
6642 	/* if ring doesn't have a interrupt vector, cannot perform ATR */
6643 	if (!q_vector)
6644 		return;
6645 
6646 	/* do nothing if sampling is disabled */
6647 	if (!ring->atr_sample_rate)
6648 		return;
6649 
6650 	ring->atr_count++;
6651 
6652 	/* snag network header to get L4 type and address */
6653 	hdr.network = skb_network_header(skb);
6654 
6655 	/* Currently only IPv4/IPv6 with TCP is supported */
6656 	if ((protocol != __constant_htons(ETH_P_IPV6) ||
6657 	     hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6658 	    (protocol != __constant_htons(ETH_P_IP) ||
6659 	     hdr.ipv4->protocol != IPPROTO_TCP))
6660 		return;
6661 
6662 	th = tcp_hdr(skb);
6663 
6664 	/* skip this packet since it is invalid or the socket is closing */
6665 	if (!th || th->fin)
6666 		return;
6667 
6668 	/* sample on all syn packets or once every atr sample count */
6669 	if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6670 		return;
6671 
6672 	/* reset sample count */
6673 	ring->atr_count = 0;
6674 
6675 	vlan_id = htons(tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6676 
6677 	/*
6678 	 * src and dst are inverted, think how the receiver sees them
6679 	 *
6680 	 * The input is broken into two sections, a non-compressed section
6681 	 * containing vm_pool, vlan_id, and flow_type.  The rest of the data
6682 	 * is XORed together and stored in the compressed dword.
6683 	 */
6684 	input.formatted.vlan_id = vlan_id;
6685 
6686 	/*
6687 	 * since src port and flex bytes occupy the same word XOR them together
6688 	 * and write the value to source port portion of compressed dword
6689 	 */
6690 	if (tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6691 		common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6692 	else
6693 		common.port.src ^= th->dest ^ protocol;
6694 	common.port.dst ^= th->source;
6695 
6696 	if (protocol == __constant_htons(ETH_P_IP)) {
6697 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6698 		common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6699 	} else {
6700 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6701 		common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6702 			     hdr.ipv6->saddr.s6_addr32[1] ^
6703 			     hdr.ipv6->saddr.s6_addr32[2] ^
6704 			     hdr.ipv6->saddr.s6_addr32[3] ^
6705 			     hdr.ipv6->daddr.s6_addr32[0] ^
6706 			     hdr.ipv6->daddr.s6_addr32[1] ^
6707 			     hdr.ipv6->daddr.s6_addr32[2] ^
6708 			     hdr.ipv6->daddr.s6_addr32[3];
6709 	}
6710 
6711 	/* This assumes the Rx queue and Tx queue are bound to the same CPU */
6712 	ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6713 					      input, common, ring->queue_index);
6714 }
6715 
6716 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6717 {
6718 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6719 	/* Herbert's original patch had:
6720 	 *  smp_mb__after_netif_stop_queue();
6721 	 * but since that doesn't exist yet, just open code it. */
6722 	smp_mb();
6723 
6724 	/* We need to check again in a case another CPU has just
6725 	 * made room available. */
6726 	if (likely(ixgbe_desc_unused(tx_ring) < size))
6727 		return -EBUSY;
6728 
6729 	/* A reprieve! - use start_queue because it doesn't call schedule */
6730 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6731 	++tx_ring->tx_stats.restart_queue;
6732 	return 0;
6733 }
6734 
6735 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6736 {
6737 	if (likely(ixgbe_desc_unused(tx_ring) >= size))
6738 		return 0;
6739 	return __ixgbe_maybe_stop_tx(tx_ring, size);
6740 }
6741 
6742 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6743 {
6744 	struct ixgbe_adapter *adapter = netdev_priv(dev);
6745 	int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6746 					       smp_processor_id();
6747 #ifdef IXGBE_FCOE
6748 	__be16 protocol = vlan_get_protocol(skb);
6749 
6750 	if (((protocol == htons(ETH_P_FCOE)) ||
6751 	    (protocol == htons(ETH_P_FIP))) &&
6752 	    (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6753 		txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6754 		txq += adapter->ring_feature[RING_F_FCOE].mask;
6755 		return txq;
6756 	}
6757 #endif
6758 
6759 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6760 		while (unlikely(txq >= dev->real_num_tx_queues))
6761 			txq -= dev->real_num_tx_queues;
6762 		return txq;
6763 	}
6764 
6765 	return skb_tx_hash(dev, skb);
6766 }
6767 
6768 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6769 			  struct ixgbe_adapter *adapter,
6770 			  struct ixgbe_ring *tx_ring)
6771 {
6772 	struct ixgbe_tx_buffer *first;
6773 	int tso;
6774 	u32 tx_flags = 0;
6775 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6776 	unsigned short f;
6777 #endif
6778 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
6779 	__be16 protocol = skb->protocol;
6780 	u8 hdr_len = 0;
6781 
6782 	/*
6783 	 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6784 	 *       + 1 desc for skb_head_len/IXGBE_MAX_DATA_PER_TXD,
6785 	 *       + 2 desc gap to keep tail from touching head,
6786 	 *       + 1 desc for context descriptor,
6787 	 * otherwise try next time
6788 	 */
6789 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6790 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6791 		count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6792 #else
6793 	count += skb_shinfo(skb)->nr_frags;
6794 #endif
6795 	if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
6796 		tx_ring->tx_stats.tx_busy++;
6797 		return NETDEV_TX_BUSY;
6798 	}
6799 
6800 #ifdef CONFIG_PCI_IOV
6801 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6802 		tx_flags |= IXGBE_TX_FLAGS_TXSW;
6803 
6804 #endif
6805 	/* if we have a HW VLAN tag being added default to the HW one */
6806 	if (vlan_tx_tag_present(skb)) {
6807 		tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6808 		tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6809 	/* else if it is a SW VLAN check the next protocol and store the tag */
6810 	} else if (protocol == __constant_htons(ETH_P_8021Q)) {
6811 		struct vlan_hdr *vhdr, _vhdr;
6812 		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
6813 		if (!vhdr)
6814 			goto out_drop;
6815 
6816 		protocol = vhdr->h_vlan_encapsulated_proto;
6817 		tx_flags |= ntohs(vhdr->h_vlan_TCI) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6818 		tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
6819 	}
6820 
6821 	/* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
6822 	if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
6823 	    ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
6824 	     (skb->priority != TC_PRIO_CONTROL))) {
6825 		tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6826 		tx_flags |= (skb->priority & 0x7) <<
6827 					IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
6828 		if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
6829 			struct vlan_ethhdr *vhdr;
6830 			if (skb_header_cloned(skb) &&
6831 			    pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
6832 				goto out_drop;
6833 			vhdr = (struct vlan_ethhdr *)skb->data;
6834 			vhdr->h_vlan_TCI = htons(tx_flags >>
6835 						 IXGBE_TX_FLAGS_VLAN_SHIFT);
6836 		} else {
6837 			tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6838 		}
6839 	}
6840 
6841 	/* record the location of the first descriptor for this packet */
6842 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6843 
6844 #ifdef IXGBE_FCOE
6845 	/* setup tx offload for FCoE */
6846 	if ((protocol == __constant_htons(ETH_P_FCOE)) &&
6847 	    (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6848 		tso = ixgbe_fso(tx_ring, skb, tx_flags, &hdr_len);
6849 		if (tso < 0)
6850 			goto out_drop;
6851 		else if (tso)
6852 			tx_flags |= IXGBE_TX_FLAGS_FSO |
6853 				    IXGBE_TX_FLAGS_FCOE;
6854 		else
6855 			tx_flags |= IXGBE_TX_FLAGS_FCOE;
6856 
6857 		goto xmit_fcoe;
6858 	}
6859 
6860 #endif /* IXGBE_FCOE */
6861 	/* setup IPv4/IPv6 offloads */
6862 	if (protocol == __constant_htons(ETH_P_IP))
6863 		tx_flags |= IXGBE_TX_FLAGS_IPV4;
6864 
6865 	tso = ixgbe_tso(tx_ring, skb, tx_flags, protocol, &hdr_len);
6866 	if (tso < 0)
6867 		goto out_drop;
6868 	else if (tso)
6869 		tx_flags |= IXGBE_TX_FLAGS_TSO;
6870 	else if (ixgbe_tx_csum(tx_ring, skb, tx_flags, protocol))
6871 		tx_flags |= IXGBE_TX_FLAGS_CSUM;
6872 
6873 	/* add the ATR filter if ATR is on */
6874 	if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6875 		ixgbe_atr(tx_ring, skb, tx_flags, protocol);
6876 
6877 #ifdef IXGBE_FCOE
6878 xmit_fcoe:
6879 #endif /* IXGBE_FCOE */
6880 	ixgbe_tx_map(tx_ring, skb, first, tx_flags, hdr_len);
6881 
6882 	ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6883 
6884 	return NETDEV_TX_OK;
6885 
6886 out_drop:
6887 	dev_kfree_skb_any(skb);
6888 	return NETDEV_TX_OK;
6889 }
6890 
6891 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6892 {
6893 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6894 	struct ixgbe_ring *tx_ring;
6895 
6896 	tx_ring = adapter->tx_ring[skb->queue_mapping];
6897 	return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6898 }
6899 
6900 /**
6901  * ixgbe_set_mac - Change the Ethernet Address of the NIC
6902  * @netdev: network interface device structure
6903  * @p: pointer to an address structure
6904  *
6905  * Returns 0 on success, negative on failure
6906  **/
6907 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6908 {
6909 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6910 	struct ixgbe_hw *hw = &adapter->hw;
6911 	struct sockaddr *addr = p;
6912 
6913 	if (!is_valid_ether_addr(addr->sa_data))
6914 		return -EADDRNOTAVAIL;
6915 
6916 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6917 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6918 
6919 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6920 			    IXGBE_RAH_AV);
6921 
6922 	return 0;
6923 }
6924 
6925 static int
6926 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6927 {
6928 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6929 	struct ixgbe_hw *hw = &adapter->hw;
6930 	u16 value;
6931 	int rc;
6932 
6933 	if (prtad != hw->phy.mdio.prtad)
6934 		return -EINVAL;
6935 	rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6936 	if (!rc)
6937 		rc = value;
6938 	return rc;
6939 }
6940 
6941 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6942 			    u16 addr, u16 value)
6943 {
6944 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6945 	struct ixgbe_hw *hw = &adapter->hw;
6946 
6947 	if (prtad != hw->phy.mdio.prtad)
6948 		return -EINVAL;
6949 	return hw->phy.ops.write_reg(hw, addr, devad, value);
6950 }
6951 
6952 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6953 {
6954 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6955 
6956 	return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6957 }
6958 
6959 /**
6960  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6961  * netdev->dev_addrs
6962  * @netdev: network interface device structure
6963  *
6964  * Returns non-zero on failure
6965  **/
6966 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6967 {
6968 	int err = 0;
6969 	struct ixgbe_adapter *adapter = netdev_priv(dev);
6970 	struct ixgbe_mac_info *mac = &adapter->hw.mac;
6971 
6972 	if (is_valid_ether_addr(mac->san_addr)) {
6973 		rtnl_lock();
6974 		err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6975 		rtnl_unlock();
6976 	}
6977 	return err;
6978 }
6979 
6980 /**
6981  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6982  * netdev->dev_addrs
6983  * @netdev: network interface device structure
6984  *
6985  * Returns non-zero on failure
6986  **/
6987 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6988 {
6989 	int err = 0;
6990 	struct ixgbe_adapter *adapter = netdev_priv(dev);
6991 	struct ixgbe_mac_info *mac = &adapter->hw.mac;
6992 
6993 	if (is_valid_ether_addr(mac->san_addr)) {
6994 		rtnl_lock();
6995 		err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6996 		rtnl_unlock();
6997 	}
6998 	return err;
6999 }
7000 
7001 #ifdef CONFIG_NET_POLL_CONTROLLER
7002 /*
7003  * Polling 'interrupt' - used by things like netconsole to send skbs
7004  * without having to re-enable interrupts. It's not called while
7005  * the interrupt routine is executing.
7006  */
7007 static void ixgbe_netpoll(struct net_device *netdev)
7008 {
7009 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7010 	int i;
7011 
7012 	/* if interface is down do nothing */
7013 	if (test_bit(__IXGBE_DOWN, &adapter->state))
7014 		return;
7015 
7016 	adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
7017 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
7018 		int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
7019 		for (i = 0; i < num_q_vectors; i++) {
7020 			struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
7021 			ixgbe_msix_clean_rings(0, q_vector);
7022 		}
7023 	} else {
7024 		ixgbe_intr(adapter->pdev->irq, netdev);
7025 	}
7026 	adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
7027 }
7028 #endif
7029 
7030 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
7031 						   struct rtnl_link_stats64 *stats)
7032 {
7033 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7034 	int i;
7035 
7036 	rcu_read_lock();
7037 	for (i = 0; i < adapter->num_rx_queues; i++) {
7038 		struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
7039 		u64 bytes, packets;
7040 		unsigned int start;
7041 
7042 		if (ring) {
7043 			do {
7044 				start = u64_stats_fetch_begin_bh(&ring->syncp);
7045 				packets = ring->stats.packets;
7046 				bytes   = ring->stats.bytes;
7047 			} while (u64_stats_fetch_retry_bh(&ring->syncp, start));
7048 			stats->rx_packets += packets;
7049 			stats->rx_bytes   += bytes;
7050 		}
7051 	}
7052 
7053 	for (i = 0; i < adapter->num_tx_queues; i++) {
7054 		struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
7055 		u64 bytes, packets;
7056 		unsigned int start;
7057 
7058 		if (ring) {
7059 			do {
7060 				start = u64_stats_fetch_begin_bh(&ring->syncp);
7061 				packets = ring->stats.packets;
7062 				bytes   = ring->stats.bytes;
7063 			} while (u64_stats_fetch_retry_bh(&ring->syncp, start));
7064 			stats->tx_packets += packets;
7065 			stats->tx_bytes   += bytes;
7066 		}
7067 	}
7068 	rcu_read_unlock();
7069 	/* following stats updated by ixgbe_watchdog_task() */
7070 	stats->multicast	= netdev->stats.multicast;
7071 	stats->rx_errors	= netdev->stats.rx_errors;
7072 	stats->rx_length_errors	= netdev->stats.rx_length_errors;
7073 	stats->rx_crc_errors	= netdev->stats.rx_crc_errors;
7074 	stats->rx_missed_errors	= netdev->stats.rx_missed_errors;
7075 	return stats;
7076 }
7077 
7078 /* ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
7079  * #adapter: pointer to ixgbe_adapter
7080  * @tc: number of traffic classes currently enabled
7081  *
7082  * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
7083  * 802.1Q priority maps to a packet buffer that exists.
7084  */
7085 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
7086 {
7087 	struct ixgbe_hw *hw = &adapter->hw;
7088 	u32 reg, rsave;
7089 	int i;
7090 
7091 	/* 82598 have a static priority to TC mapping that can not
7092 	 * be changed so no validation is needed.
7093 	 */
7094 	if (hw->mac.type == ixgbe_mac_82598EB)
7095 		return;
7096 
7097 	reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
7098 	rsave = reg;
7099 
7100 	for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
7101 		u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
7102 
7103 		/* If up2tc is out of bounds default to zero */
7104 		if (up2tc > tc)
7105 			reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
7106 	}
7107 
7108 	if (reg != rsave)
7109 		IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
7110 
7111 	return;
7112 }
7113 
7114 
7115 /* ixgbe_setup_tc - routine to configure net_device for multiple traffic
7116  * classes.
7117  *
7118  * @netdev: net device to configure
7119  * @tc: number of traffic classes to enable
7120  */
7121 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
7122 {
7123 	struct ixgbe_adapter *adapter = netdev_priv(dev);
7124 	struct ixgbe_hw *hw = &adapter->hw;
7125 
7126 	/* Multiple traffic classes requires multiple queues */
7127 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
7128 		e_err(drv, "Enable failed, needs MSI-X\n");
7129 		return -EINVAL;
7130 	}
7131 
7132 	/* Hardware supports up to 8 traffic classes */
7133 	if (tc > adapter->dcb_cfg.num_tcs.pg_tcs ||
7134 	    (hw->mac.type == ixgbe_mac_82598EB && tc < MAX_TRAFFIC_CLASS))
7135 		return -EINVAL;
7136 
7137 	/* Hardware has to reinitialize queues and interrupts to
7138 	 * match packet buffer alignment. Unfortunately, the
7139 	 * hardware is not flexible enough to do this dynamically.
7140 	 */
7141 	if (netif_running(dev))
7142 		ixgbe_close(dev);
7143 	ixgbe_clear_interrupt_scheme(adapter);
7144 
7145 	if (tc) {
7146 		netdev_set_num_tc(dev, tc);
7147 		adapter->last_lfc_mode = adapter->hw.fc.current_mode;
7148 
7149 		adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
7150 		adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
7151 
7152 		if (adapter->hw.mac.type == ixgbe_mac_82598EB)
7153 			adapter->hw.fc.requested_mode = ixgbe_fc_none;
7154 	} else {
7155 		netdev_reset_tc(dev);
7156 
7157 		adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
7158 
7159 		adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
7160 		adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
7161 
7162 		adapter->temp_dcb_cfg.pfc_mode_enable = false;
7163 		adapter->dcb_cfg.pfc_mode_enable = false;
7164 	}
7165 
7166 	ixgbe_init_interrupt_scheme(adapter);
7167 	ixgbe_validate_rtr(adapter, tc);
7168 	if (netif_running(dev))
7169 		ixgbe_open(dev);
7170 
7171 	return 0;
7172 }
7173 
7174 void ixgbe_do_reset(struct net_device *netdev)
7175 {
7176 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7177 
7178 	if (netif_running(netdev))
7179 		ixgbe_reinit_locked(adapter);
7180 	else
7181 		ixgbe_reset(adapter);
7182 }
7183 
7184 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
7185 	netdev_features_t data)
7186 {
7187 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7188 
7189 #ifdef CONFIG_DCB
7190 	if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
7191 		data &= ~NETIF_F_HW_VLAN_RX;
7192 #endif
7193 
7194 	/* return error if RXHASH is being enabled when RSS is not supported */
7195 	if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
7196 		data &= ~NETIF_F_RXHASH;
7197 
7198 	/* If Rx checksum is disabled, then RSC/LRO should also be disabled */
7199 	if (!(data & NETIF_F_RXCSUM))
7200 		data &= ~NETIF_F_LRO;
7201 
7202 	/* Turn off LRO if not RSC capable or invalid ITR settings */
7203 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)) {
7204 		data &= ~NETIF_F_LRO;
7205 	} else if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
7206 		   (adapter->rx_itr_setting != 1 &&
7207 		    adapter->rx_itr_setting > IXGBE_MAX_RSC_INT_RATE)) {
7208 		data &= ~NETIF_F_LRO;
7209 		e_info(probe, "rx-usecs set too low, not enabling RSC\n");
7210 	}
7211 
7212 	return data;
7213 }
7214 
7215 static int ixgbe_set_features(struct net_device *netdev,
7216 	netdev_features_t data)
7217 {
7218 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7219 	bool need_reset = false;
7220 
7221 	/* If Rx checksum is disabled, then RSC/LRO should also be disabled */
7222 	if (!(data & NETIF_F_RXCSUM))
7223 		adapter->flags &= ~IXGBE_FLAG_RX_CSUM_ENABLED;
7224 	else
7225 		adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
7226 
7227 	/* Make sure RSC matches LRO, reset if change */
7228 	if (!!(data & NETIF_F_LRO) !=
7229 	     !!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
7230 		adapter->flags2 ^= IXGBE_FLAG2_RSC_ENABLED;
7231 		switch (adapter->hw.mac.type) {
7232 		case ixgbe_mac_X540:
7233 		case ixgbe_mac_82599EB:
7234 			need_reset = true;
7235 			break;
7236 		default:
7237 			break;
7238 		}
7239 	}
7240 
7241 	/*
7242 	 * Check if Flow Director n-tuple support was enabled or disabled.  If
7243 	 * the state changed, we need to reset.
7244 	 */
7245 	if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
7246 		/* turn off ATR, enable perfect filters and reset */
7247 		if (data & NETIF_F_NTUPLE) {
7248 			adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
7249 			adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7250 			need_reset = true;
7251 		}
7252 	} else if (!(data & NETIF_F_NTUPLE)) {
7253 		/* turn off Flow Director, set ATR and reset */
7254 		adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7255 		if ((adapter->flags &  IXGBE_FLAG_RSS_ENABLED) &&
7256 		    !(adapter->flags &  IXGBE_FLAG_DCB_ENABLED))
7257 			adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
7258 		need_reset = true;
7259 	}
7260 
7261 	if (need_reset)
7262 		ixgbe_do_reset(netdev);
7263 
7264 	return 0;
7265 
7266 }
7267 
7268 static const struct net_device_ops ixgbe_netdev_ops = {
7269 	.ndo_open		= ixgbe_open,
7270 	.ndo_stop		= ixgbe_close,
7271 	.ndo_start_xmit		= ixgbe_xmit_frame,
7272 	.ndo_select_queue	= ixgbe_select_queue,
7273 	.ndo_set_rx_mode        = ixgbe_set_rx_mode,
7274 	.ndo_validate_addr	= eth_validate_addr,
7275 	.ndo_set_mac_address	= ixgbe_set_mac,
7276 	.ndo_change_mtu		= ixgbe_change_mtu,
7277 	.ndo_tx_timeout		= ixgbe_tx_timeout,
7278 	.ndo_vlan_rx_add_vid	= ixgbe_vlan_rx_add_vid,
7279 	.ndo_vlan_rx_kill_vid	= ixgbe_vlan_rx_kill_vid,
7280 	.ndo_do_ioctl		= ixgbe_ioctl,
7281 	.ndo_set_vf_mac		= ixgbe_ndo_set_vf_mac,
7282 	.ndo_set_vf_vlan	= ixgbe_ndo_set_vf_vlan,
7283 	.ndo_set_vf_tx_rate	= ixgbe_ndo_set_vf_bw,
7284 	.ndo_set_vf_spoofchk    = ixgbe_ndo_set_vf_spoofchk,
7285 	.ndo_get_vf_config	= ixgbe_ndo_get_vf_config,
7286 	.ndo_get_stats64	= ixgbe_get_stats64,
7287 	.ndo_setup_tc		= ixgbe_setup_tc,
7288 #ifdef CONFIG_NET_POLL_CONTROLLER
7289 	.ndo_poll_controller	= ixgbe_netpoll,
7290 #endif
7291 #ifdef IXGBE_FCOE
7292 	.ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
7293 	.ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
7294 	.ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
7295 	.ndo_fcoe_enable = ixgbe_fcoe_enable,
7296 	.ndo_fcoe_disable = ixgbe_fcoe_disable,
7297 	.ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
7298 	.ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
7299 #endif /* IXGBE_FCOE */
7300 	.ndo_set_features = ixgbe_set_features,
7301 	.ndo_fix_features = ixgbe_fix_features,
7302 };
7303 
7304 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
7305 			   const struct ixgbe_info *ii)
7306 {
7307 #ifdef CONFIG_PCI_IOV
7308 	struct ixgbe_hw *hw = &adapter->hw;
7309 
7310 	if (hw->mac.type == ixgbe_mac_82598EB)
7311 		return;
7312 
7313 	/* The 82599 supports up to 64 VFs per physical function
7314 	 * but this implementation limits allocation to 63 so that
7315 	 * basic networking resources are still available to the
7316 	 * physical function
7317 	 */
7318 	adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
7319 	ixgbe_enable_sriov(adapter, ii);
7320 #endif /* CONFIG_PCI_IOV */
7321 }
7322 
7323 /**
7324  * ixgbe_probe - Device Initialization Routine
7325  * @pdev: PCI device information struct
7326  * @ent: entry in ixgbe_pci_tbl
7327  *
7328  * Returns 0 on success, negative on failure
7329  *
7330  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7331  * The OS initialization, configuring of the adapter private structure,
7332  * and a hardware reset occur.
7333  **/
7334 static int __devinit ixgbe_probe(struct pci_dev *pdev,
7335 				 const struct pci_device_id *ent)
7336 {
7337 	struct net_device *netdev;
7338 	struct ixgbe_adapter *adapter = NULL;
7339 	struct ixgbe_hw *hw;
7340 	const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7341 	static int cards_found;
7342 	int i, err, pci_using_dac;
7343 	u8 part_str[IXGBE_PBANUM_LENGTH];
7344 	unsigned int indices = num_possible_cpus();
7345 #ifdef IXGBE_FCOE
7346 	u16 device_caps;
7347 #endif
7348 	u32 eec;
7349 	u16 wol_cap;
7350 
7351 	/* Catch broken hardware that put the wrong VF device ID in
7352 	 * the PCIe SR-IOV capability.
7353 	 */
7354 	if (pdev->is_virtfn) {
7355 		WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7356 		     pci_name(pdev), pdev->vendor, pdev->device);
7357 		return -EINVAL;
7358 	}
7359 
7360 	err = pci_enable_device_mem(pdev);
7361 	if (err)
7362 		return err;
7363 
7364 	if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7365 	    !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7366 		pci_using_dac = 1;
7367 	} else {
7368 		err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7369 		if (err) {
7370 			err = dma_set_coherent_mask(&pdev->dev,
7371 						    DMA_BIT_MASK(32));
7372 			if (err) {
7373 				dev_err(&pdev->dev,
7374 					"No usable DMA configuration, aborting\n");
7375 				goto err_dma;
7376 			}
7377 		}
7378 		pci_using_dac = 0;
7379 	}
7380 
7381 	err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7382 					   IORESOURCE_MEM), ixgbe_driver_name);
7383 	if (err) {
7384 		dev_err(&pdev->dev,
7385 			"pci_request_selected_regions failed 0x%x\n", err);
7386 		goto err_pci_reg;
7387 	}
7388 
7389 	pci_enable_pcie_error_reporting(pdev);
7390 
7391 	pci_set_master(pdev);
7392 	pci_save_state(pdev);
7393 
7394 #ifdef CONFIG_IXGBE_DCB
7395 	indices *= MAX_TRAFFIC_CLASS;
7396 #endif
7397 
7398 	if (ii->mac == ixgbe_mac_82598EB)
7399 		indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7400 	else
7401 		indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7402 
7403 #ifdef IXGBE_FCOE
7404 	indices += min_t(unsigned int, num_possible_cpus(),
7405 			 IXGBE_MAX_FCOE_INDICES);
7406 #endif
7407 	netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7408 	if (!netdev) {
7409 		err = -ENOMEM;
7410 		goto err_alloc_etherdev;
7411 	}
7412 
7413 	SET_NETDEV_DEV(netdev, &pdev->dev);
7414 
7415 	adapter = netdev_priv(netdev);
7416 	pci_set_drvdata(pdev, adapter);
7417 
7418 	adapter->netdev = netdev;
7419 	adapter->pdev = pdev;
7420 	hw = &adapter->hw;
7421 	hw->back = adapter;
7422 	adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
7423 
7424 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7425 			      pci_resource_len(pdev, 0));
7426 	if (!hw->hw_addr) {
7427 		err = -EIO;
7428 		goto err_ioremap;
7429 	}
7430 
7431 	for (i = 1; i <= 5; i++) {
7432 		if (pci_resource_len(pdev, i) == 0)
7433 			continue;
7434 	}
7435 
7436 	netdev->netdev_ops = &ixgbe_netdev_ops;
7437 	ixgbe_set_ethtool_ops(netdev);
7438 	netdev->watchdog_timeo = 5 * HZ;
7439 	strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7440 
7441 	adapter->bd_number = cards_found;
7442 
7443 	/* Setup hw api */
7444 	memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7445 	hw->mac.type  = ii->mac;
7446 
7447 	/* EEPROM */
7448 	memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7449 	eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7450 	/* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7451 	if (!(eec & (1 << 8)))
7452 		hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7453 
7454 	/* PHY */
7455 	memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7456 	hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7457 	/* ixgbe_identify_phy_generic will set prtad and mmds properly */
7458 	hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7459 	hw->phy.mdio.mmds = 0;
7460 	hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7461 	hw->phy.mdio.dev = netdev;
7462 	hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7463 	hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7464 
7465 	ii->get_invariants(hw);
7466 
7467 	/* setup the private structure */
7468 	err = ixgbe_sw_init(adapter);
7469 	if (err)
7470 		goto err_sw_init;
7471 
7472 	/* Make it possible the adapter to be woken up via WOL */
7473 	switch (adapter->hw.mac.type) {
7474 	case ixgbe_mac_82599EB:
7475 	case ixgbe_mac_X540:
7476 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7477 		break;
7478 	default:
7479 		break;
7480 	}
7481 
7482 	/*
7483 	 * If there is a fan on this device and it has failed log the
7484 	 * failure.
7485 	 */
7486 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7487 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7488 		if (esdp & IXGBE_ESDP_SDP1)
7489 			e_crit(probe, "Fan has stopped, replace the adapter\n");
7490 	}
7491 
7492 	/* reset_hw fills in the perm_addr as well */
7493 	hw->phy.reset_if_overtemp = true;
7494 	err = hw->mac.ops.reset_hw(hw);
7495 	hw->phy.reset_if_overtemp = false;
7496 	if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7497 	    hw->mac.type == ixgbe_mac_82598EB) {
7498 		err = 0;
7499 	} else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7500 		e_dev_err("failed to load because an unsupported SFP+ "
7501 			  "module type was detected.\n");
7502 		e_dev_err("Reload the driver after installing a supported "
7503 			  "module.\n");
7504 		goto err_sw_init;
7505 	} else if (err) {
7506 		e_dev_err("HW Init failed: %d\n", err);
7507 		goto err_sw_init;
7508 	}
7509 
7510 	ixgbe_probe_vf(adapter, ii);
7511 
7512 	netdev->features = NETIF_F_SG |
7513 			   NETIF_F_IP_CSUM |
7514 			   NETIF_F_IPV6_CSUM |
7515 			   NETIF_F_HW_VLAN_TX |
7516 			   NETIF_F_HW_VLAN_RX |
7517 			   NETIF_F_HW_VLAN_FILTER |
7518 			   NETIF_F_TSO |
7519 			   NETIF_F_TSO6 |
7520 			   NETIF_F_RXHASH |
7521 			   NETIF_F_RXCSUM;
7522 
7523 	netdev->hw_features = netdev->features;
7524 
7525 	switch (adapter->hw.mac.type) {
7526 	case ixgbe_mac_82599EB:
7527 	case ixgbe_mac_X540:
7528 		netdev->features |= NETIF_F_SCTP_CSUM;
7529 		netdev->hw_features |= NETIF_F_SCTP_CSUM |
7530 				       NETIF_F_NTUPLE;
7531 		break;
7532 	default:
7533 		break;
7534 	}
7535 
7536 	netdev->vlan_features |= NETIF_F_TSO;
7537 	netdev->vlan_features |= NETIF_F_TSO6;
7538 	netdev->vlan_features |= NETIF_F_IP_CSUM;
7539 	netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7540 	netdev->vlan_features |= NETIF_F_SG;
7541 
7542 	netdev->priv_flags |= IFF_UNICAST_FLT;
7543 
7544 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7545 		adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
7546 				    IXGBE_FLAG_DCB_ENABLED);
7547 
7548 #ifdef CONFIG_IXGBE_DCB
7549 	netdev->dcbnl_ops = &dcbnl_ops;
7550 #endif
7551 
7552 #ifdef IXGBE_FCOE
7553 	if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7554 		if (hw->mac.ops.get_device_caps) {
7555 			hw->mac.ops.get_device_caps(hw, &device_caps);
7556 			if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7557 				adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7558 		}
7559 	}
7560 	if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7561 		netdev->vlan_features |= NETIF_F_FCOE_CRC;
7562 		netdev->vlan_features |= NETIF_F_FSO;
7563 		netdev->vlan_features |= NETIF_F_FCOE_MTU;
7564 	}
7565 #endif /* IXGBE_FCOE */
7566 	if (pci_using_dac) {
7567 		netdev->features |= NETIF_F_HIGHDMA;
7568 		netdev->vlan_features |= NETIF_F_HIGHDMA;
7569 	}
7570 
7571 	if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
7572 		netdev->hw_features |= NETIF_F_LRO;
7573 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7574 		netdev->features |= NETIF_F_LRO;
7575 
7576 	/* make sure the EEPROM is good */
7577 	if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7578 		e_dev_err("The EEPROM Checksum Is Not Valid\n");
7579 		err = -EIO;
7580 		goto err_eeprom;
7581 	}
7582 
7583 	memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7584 	memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7585 
7586 	if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7587 		e_dev_err("invalid MAC address\n");
7588 		err = -EIO;
7589 		goto err_eeprom;
7590 	}
7591 
7592 	setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7593 	            (unsigned long) adapter);
7594 
7595 	INIT_WORK(&adapter->service_task, ixgbe_service_task);
7596 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7597 
7598 	err = ixgbe_init_interrupt_scheme(adapter);
7599 	if (err)
7600 		goto err_sw_init;
7601 
7602 	if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
7603 		netdev->hw_features &= ~NETIF_F_RXHASH;
7604 		netdev->features &= ~NETIF_F_RXHASH;
7605 	}
7606 
7607 	/* WOL not supported for all but the following */
7608 	adapter->wol = 0;
7609 	switch (pdev->device) {
7610 	case IXGBE_DEV_ID_82599_SFP:
7611 		/* Only these subdevice supports WOL */
7612 		switch (pdev->subsystem_device) {
7613 		case IXGBE_SUBDEV_ID_82599_560FLR:
7614 			/* only support first port */
7615 			if (hw->bus.func != 0)
7616 				break;
7617 		case IXGBE_SUBDEV_ID_82599_SFP:
7618 			adapter->wol = IXGBE_WUFC_MAG;
7619 			break;
7620 		}
7621 		break;
7622 	case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7623 		/* All except this subdevice support WOL */
7624 		if (pdev->subsystem_device != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7625 			adapter->wol = IXGBE_WUFC_MAG;
7626 		break;
7627 	case IXGBE_DEV_ID_82599_KX4:
7628 		adapter->wol = IXGBE_WUFC_MAG;
7629 		break;
7630 	case IXGBE_DEV_ID_X540T:
7631 		/* Check eeprom to see if it is enabled */
7632 		hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
7633 		wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
7634 
7635 		if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
7636 		    ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
7637 		     (hw->bus.func == 0)))
7638 			adapter->wol = IXGBE_WUFC_MAG;
7639 		break;
7640 	}
7641 	device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7642 
7643 	/* save off EEPROM version number */
7644 	hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
7645 	hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
7646 
7647 	/* pick up the PCI bus settings for reporting later */
7648 	hw->mac.ops.get_bus_info(hw);
7649 
7650 	/* print bus type/speed/width info */
7651 	e_dev_info("(PCI Express:%s:%s) %pM\n",
7652 		   (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7653 		    hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7654 		    "Unknown"),
7655 		   (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7656 		    hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7657 		    hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7658 		    "Unknown"),
7659 		   netdev->dev_addr);
7660 
7661 	err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7662 	if (err)
7663 		strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7664 	if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7665 		e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7666 			   hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7667 		           part_str);
7668 	else
7669 		e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7670 			   hw->mac.type, hw->phy.type, part_str);
7671 
7672 	if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7673 		e_dev_warn("PCI-Express bandwidth available for this card is "
7674 			   "not sufficient for optimal performance.\n");
7675 		e_dev_warn("For optimal performance a x8 PCI-Express slot "
7676 			   "is required.\n");
7677 	}
7678 
7679 	/* reset the hardware with the new settings */
7680 	err = hw->mac.ops.start_hw(hw);
7681 
7682 	if (err == IXGBE_ERR_EEPROM_VERSION) {
7683 		/* We are running on a pre-production device, log a warning */
7684 		e_dev_warn("This device is a pre-production adapter/LOM. "
7685 			   "Please be aware there may be issues associated "
7686 			   "with your hardware.  If you are experiencing "
7687 			   "problems please contact your Intel or hardware "
7688 			   "representative who provided you with this "
7689 			   "hardware.\n");
7690 	}
7691 	strcpy(netdev->name, "eth%d");
7692 	err = register_netdev(netdev);
7693 	if (err)
7694 		goto err_register;
7695 
7696 	/* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7697 	if (hw->mac.ops.disable_tx_laser &&
7698 	    ((hw->phy.multispeed_fiber) ||
7699 	     ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7700 	      (hw->mac.type == ixgbe_mac_82599EB))))
7701 		hw->mac.ops.disable_tx_laser(hw);
7702 
7703 	/* carrier off reporting is important to ethtool even BEFORE open */
7704 	netif_carrier_off(netdev);
7705 
7706 #ifdef CONFIG_IXGBE_DCA
7707 	if (dca_add_requester(&pdev->dev) == 0) {
7708 		adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7709 		ixgbe_setup_dca(adapter);
7710 	}
7711 #endif
7712 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7713 		e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7714 		for (i = 0; i < adapter->num_vfs; i++)
7715 			ixgbe_vf_configuration(pdev, (i | 0x10000000));
7716 	}
7717 
7718 	/* firmware requires driver version to be 0xFFFFFFFF
7719 	 * since os does not support feature
7720 	 */
7721 	if (hw->mac.ops.set_fw_drv_ver)
7722 		hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
7723 					   0xFF);
7724 
7725 	/* add san mac addr to netdev */
7726 	ixgbe_add_sanmac_netdev(netdev);
7727 
7728 	e_dev_info("%s\n", ixgbe_default_device_descr);
7729 	cards_found++;
7730 	return 0;
7731 
7732 err_register:
7733 	ixgbe_release_hw_control(adapter);
7734 	ixgbe_clear_interrupt_scheme(adapter);
7735 err_sw_init:
7736 err_eeprom:
7737 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7738 		ixgbe_disable_sriov(adapter);
7739 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7740 	iounmap(hw->hw_addr);
7741 err_ioremap:
7742 	free_netdev(netdev);
7743 err_alloc_etherdev:
7744 	pci_release_selected_regions(pdev,
7745 				     pci_select_bars(pdev, IORESOURCE_MEM));
7746 err_pci_reg:
7747 err_dma:
7748 	pci_disable_device(pdev);
7749 	return err;
7750 }
7751 
7752 /**
7753  * ixgbe_remove - Device Removal Routine
7754  * @pdev: PCI device information struct
7755  *
7756  * ixgbe_remove is called by the PCI subsystem to alert the driver
7757  * that it should release a PCI device.  The could be caused by a
7758  * Hot-Plug event, or because the driver is going to be removed from
7759  * memory.
7760  **/
7761 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7762 {
7763 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7764 	struct net_device *netdev = adapter->netdev;
7765 
7766 	set_bit(__IXGBE_DOWN, &adapter->state);
7767 	cancel_work_sync(&adapter->service_task);
7768 
7769 #ifdef CONFIG_IXGBE_DCA
7770 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7771 		adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7772 		dca_remove_requester(&pdev->dev);
7773 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7774 	}
7775 
7776 #endif
7777 #ifdef IXGBE_FCOE
7778 	if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7779 		ixgbe_cleanup_fcoe(adapter);
7780 
7781 #endif /* IXGBE_FCOE */
7782 
7783 	/* remove the added san mac */
7784 	ixgbe_del_sanmac_netdev(netdev);
7785 
7786 	if (netdev->reg_state == NETREG_REGISTERED)
7787 		unregister_netdev(netdev);
7788 
7789 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7790 		if (!(ixgbe_check_vf_assignment(adapter)))
7791 			ixgbe_disable_sriov(adapter);
7792 		else
7793 			e_dev_warn("Unloading driver while VFs are assigned "
7794 				   "- VFs will not be deallocated\n");
7795 	}
7796 
7797 	ixgbe_clear_interrupt_scheme(adapter);
7798 
7799 	ixgbe_release_hw_control(adapter);
7800 
7801 	iounmap(adapter->hw.hw_addr);
7802 	pci_release_selected_regions(pdev, pci_select_bars(pdev,
7803 				     IORESOURCE_MEM));
7804 
7805 	e_dev_info("complete\n");
7806 
7807 	free_netdev(netdev);
7808 
7809 	pci_disable_pcie_error_reporting(pdev);
7810 
7811 	pci_disable_device(pdev);
7812 }
7813 
7814 /**
7815  * ixgbe_io_error_detected - called when PCI error is detected
7816  * @pdev: Pointer to PCI device
7817  * @state: The current pci connection state
7818  *
7819  * This function is called after a PCI bus error affecting
7820  * this device has been detected.
7821  */
7822 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7823 						pci_channel_state_t state)
7824 {
7825 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7826 	struct net_device *netdev = adapter->netdev;
7827 
7828 #ifdef CONFIG_PCI_IOV
7829 	struct pci_dev *bdev, *vfdev;
7830 	u32 dw0, dw1, dw2, dw3;
7831 	int vf, pos;
7832 	u16 req_id, pf_func;
7833 
7834 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7835 	    adapter->num_vfs == 0)
7836 		goto skip_bad_vf_detection;
7837 
7838 	bdev = pdev->bus->self;
7839 	while (bdev && (bdev->pcie_type != PCI_EXP_TYPE_ROOT_PORT))
7840 		bdev = bdev->bus->self;
7841 
7842 	if (!bdev)
7843 		goto skip_bad_vf_detection;
7844 
7845 	pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
7846 	if (!pos)
7847 		goto skip_bad_vf_detection;
7848 
7849 	pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG, &dw0);
7850 	pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 4, &dw1);
7851 	pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 8, &dw2);
7852 	pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 12, &dw3);
7853 
7854 	req_id = dw1 >> 16;
7855 	/* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
7856 	if (!(req_id & 0x0080))
7857 		goto skip_bad_vf_detection;
7858 
7859 	pf_func = req_id & 0x01;
7860 	if ((pf_func & 1) == (pdev->devfn & 1)) {
7861 		unsigned int device_id;
7862 
7863 		vf = (req_id & 0x7F) >> 1;
7864 		e_dev_err("VF %d has caused a PCIe error\n", vf);
7865 		e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
7866 				"%8.8x\tdw3: %8.8x\n",
7867 		dw0, dw1, dw2, dw3);
7868 		switch (adapter->hw.mac.type) {
7869 		case ixgbe_mac_82599EB:
7870 			device_id = IXGBE_82599_VF_DEVICE_ID;
7871 			break;
7872 		case ixgbe_mac_X540:
7873 			device_id = IXGBE_X540_VF_DEVICE_ID;
7874 			break;
7875 		default:
7876 			device_id = 0;
7877 			break;
7878 		}
7879 
7880 		/* Find the pci device of the offending VF */
7881 		vfdev = pci_get_device(IXGBE_INTEL_VENDOR_ID, device_id, NULL);
7882 		while (vfdev) {
7883 			if (vfdev->devfn == (req_id & 0xFF))
7884 				break;
7885 			vfdev = pci_get_device(IXGBE_INTEL_VENDOR_ID,
7886 					       device_id, vfdev);
7887 		}
7888 		/*
7889 		 * There's a slim chance the VF could have been hot plugged,
7890 		 * so if it is no longer present we don't need to issue the
7891 		 * VFLR.  Just clean up the AER in that case.
7892 		 */
7893 		if (vfdev) {
7894 			e_dev_err("Issuing VFLR to VF %d\n", vf);
7895 			pci_write_config_dword(vfdev, 0xA8, 0x00008000);
7896 		}
7897 
7898 		pci_cleanup_aer_uncorrect_error_status(pdev);
7899 	}
7900 
7901 	/*
7902 	 * Even though the error may have occurred on the other port
7903 	 * we still need to increment the vf error reference count for
7904 	 * both ports because the I/O resume function will be called
7905 	 * for both of them.
7906 	 */
7907 	adapter->vferr_refcount++;
7908 
7909 	return PCI_ERS_RESULT_RECOVERED;
7910 
7911 skip_bad_vf_detection:
7912 #endif /* CONFIG_PCI_IOV */
7913 	netif_device_detach(netdev);
7914 
7915 	if (state == pci_channel_io_perm_failure)
7916 		return PCI_ERS_RESULT_DISCONNECT;
7917 
7918 	if (netif_running(netdev))
7919 		ixgbe_down(adapter);
7920 	pci_disable_device(pdev);
7921 
7922 	/* Request a slot reset. */
7923 	return PCI_ERS_RESULT_NEED_RESET;
7924 }
7925 
7926 /**
7927  * ixgbe_io_slot_reset - called after the pci bus has been reset.
7928  * @pdev: Pointer to PCI device
7929  *
7930  * Restart the card from scratch, as if from a cold-boot.
7931  */
7932 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7933 {
7934 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7935 	pci_ers_result_t result;
7936 	int err;
7937 
7938 	if (pci_enable_device_mem(pdev)) {
7939 		e_err(probe, "Cannot re-enable PCI device after reset.\n");
7940 		result = PCI_ERS_RESULT_DISCONNECT;
7941 	} else {
7942 		pci_set_master(pdev);
7943 		pci_restore_state(pdev);
7944 		pci_save_state(pdev);
7945 
7946 		pci_wake_from_d3(pdev, false);
7947 
7948 		ixgbe_reset(adapter);
7949 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7950 		result = PCI_ERS_RESULT_RECOVERED;
7951 	}
7952 
7953 	err = pci_cleanup_aer_uncorrect_error_status(pdev);
7954 	if (err) {
7955 		e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7956 			  "failed 0x%0x\n", err);
7957 		/* non-fatal, continue */
7958 	}
7959 
7960 	return result;
7961 }
7962 
7963 /**
7964  * ixgbe_io_resume - called when traffic can start flowing again.
7965  * @pdev: Pointer to PCI device
7966  *
7967  * This callback is called when the error recovery driver tells us that
7968  * its OK to resume normal operation.
7969  */
7970 static void ixgbe_io_resume(struct pci_dev *pdev)
7971 {
7972 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7973 	struct net_device *netdev = adapter->netdev;
7974 
7975 #ifdef CONFIG_PCI_IOV
7976 	if (adapter->vferr_refcount) {
7977 		e_info(drv, "Resuming after VF err\n");
7978 		adapter->vferr_refcount--;
7979 		return;
7980 	}
7981 
7982 #endif
7983 	if (netif_running(netdev))
7984 		ixgbe_up(adapter);
7985 
7986 	netif_device_attach(netdev);
7987 }
7988 
7989 static struct pci_error_handlers ixgbe_err_handler = {
7990 	.error_detected = ixgbe_io_error_detected,
7991 	.slot_reset = ixgbe_io_slot_reset,
7992 	.resume = ixgbe_io_resume,
7993 };
7994 
7995 static struct pci_driver ixgbe_driver = {
7996 	.name     = ixgbe_driver_name,
7997 	.id_table = ixgbe_pci_tbl,
7998 	.probe    = ixgbe_probe,
7999 	.remove   = __devexit_p(ixgbe_remove),
8000 #ifdef CONFIG_PM
8001 	.suspend  = ixgbe_suspend,
8002 	.resume   = ixgbe_resume,
8003 #endif
8004 	.shutdown = ixgbe_shutdown,
8005 	.err_handler = &ixgbe_err_handler
8006 };
8007 
8008 /**
8009  * ixgbe_init_module - Driver Registration Routine
8010  *
8011  * ixgbe_init_module is the first routine called when the driver is
8012  * loaded. All it does is register with the PCI subsystem.
8013  **/
8014 static int __init ixgbe_init_module(void)
8015 {
8016 	int ret;
8017 	pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
8018 	pr_info("%s\n", ixgbe_copyright);
8019 
8020 #ifdef CONFIG_IXGBE_DCA
8021 	dca_register_notify(&dca_notifier);
8022 #endif
8023 
8024 	ret = pci_register_driver(&ixgbe_driver);
8025 	return ret;
8026 }
8027 
8028 module_init(ixgbe_init_module);
8029 
8030 /**
8031  * ixgbe_exit_module - Driver Exit Cleanup Routine
8032  *
8033  * ixgbe_exit_module is called just before the driver is removed
8034  * from memory.
8035  **/
8036 static void __exit ixgbe_exit_module(void)
8037 {
8038 #ifdef CONFIG_IXGBE_DCA
8039 	dca_unregister_notify(&dca_notifier);
8040 #endif
8041 	pci_unregister_driver(&ixgbe_driver);
8042 	rcu_barrier(); /* Wait for completion of call_rcu()'s */
8043 }
8044 
8045 #ifdef CONFIG_IXGBE_DCA
8046 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
8047 			    void *p)
8048 {
8049 	int ret_val;
8050 
8051 	ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
8052 					 __ixgbe_notify_dca);
8053 
8054 	return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
8055 }
8056 
8057 #endif /* CONFIG_IXGBE_DCA */
8058 
8059 module_exit(ixgbe_exit_module);
8060 
8061 /* ixgbe_main.c */
8062