xref: /linux/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c (revision f2ee442115c9b6219083c019939a9cc0c9abb2f8)
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 #define MAJ 3
59 #define MIN 6
60 #define BUILD 7
61 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
62 	__stringify(BUILD) "-k"
63 const char ixgbe_driver_version[] = DRV_VERSION;
64 static const char ixgbe_copyright[] =
65 				"Copyright (c) 1999-2011 Intel Corporation.";
66 
67 static const struct ixgbe_info *ixgbe_info_tbl[] = {
68 	[board_82598] = &ixgbe_82598_info,
69 	[board_82599] = &ixgbe_82599_info,
70 	[board_X540] = &ixgbe_X540_info,
71 };
72 
73 /* ixgbe_pci_tbl - PCI Device ID Table
74  *
75  * Wildcard entries (PCI_ANY_ID) should come last
76  * Last entry must be all 0s
77  *
78  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
79  *   Class, Class Mask, private data (not used) }
80  */
81 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
82 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
83 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
84 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
85 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
86 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
87 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
88 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
89 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
90 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
91 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
92 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
93 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
94 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
95 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
96 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
97 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
98 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
99 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
100 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
101 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
102 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
103 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
104 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
105 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
106 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
107 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
108 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
109 	/* required last entry */
110 	{0, }
111 };
112 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
113 
114 #ifdef CONFIG_IXGBE_DCA
115 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
116 			    void *p);
117 static struct notifier_block dca_notifier = {
118 	.notifier_call = ixgbe_notify_dca,
119 	.next          = NULL,
120 	.priority      = 0
121 };
122 #endif
123 
124 #ifdef CONFIG_PCI_IOV
125 static unsigned int max_vfs;
126 module_param(max_vfs, uint, 0);
127 MODULE_PARM_DESC(max_vfs,
128 		 "Maximum number of virtual functions to allocate per physical function");
129 #endif /* CONFIG_PCI_IOV */
130 
131 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
132 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
133 MODULE_LICENSE("GPL");
134 MODULE_VERSION(DRV_VERSION);
135 
136 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
137 
138 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
139 {
140 	if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
141 	    !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
142 		schedule_work(&adapter->service_task);
143 }
144 
145 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
146 {
147 	BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
148 
149 	/* flush memory to make sure state is correct before next watchog */
150 	smp_mb__before_clear_bit();
151 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
152 }
153 
154 struct ixgbe_reg_info {
155 	u32 ofs;
156 	char *name;
157 };
158 
159 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
160 
161 	/* General Registers */
162 	{IXGBE_CTRL, "CTRL"},
163 	{IXGBE_STATUS, "STATUS"},
164 	{IXGBE_CTRL_EXT, "CTRL_EXT"},
165 
166 	/* Interrupt Registers */
167 	{IXGBE_EICR, "EICR"},
168 
169 	/* RX Registers */
170 	{IXGBE_SRRCTL(0), "SRRCTL"},
171 	{IXGBE_DCA_RXCTRL(0), "DRXCTL"},
172 	{IXGBE_RDLEN(0), "RDLEN"},
173 	{IXGBE_RDH(0), "RDH"},
174 	{IXGBE_RDT(0), "RDT"},
175 	{IXGBE_RXDCTL(0), "RXDCTL"},
176 	{IXGBE_RDBAL(0), "RDBAL"},
177 	{IXGBE_RDBAH(0), "RDBAH"},
178 
179 	/* TX Registers */
180 	{IXGBE_TDBAL(0), "TDBAL"},
181 	{IXGBE_TDBAH(0), "TDBAH"},
182 	{IXGBE_TDLEN(0), "TDLEN"},
183 	{IXGBE_TDH(0), "TDH"},
184 	{IXGBE_TDT(0), "TDT"},
185 	{IXGBE_TXDCTL(0), "TXDCTL"},
186 
187 	/* List Terminator */
188 	{}
189 };
190 
191 
192 /*
193  * ixgbe_regdump - register printout routine
194  */
195 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
196 {
197 	int i = 0, j = 0;
198 	char rname[16];
199 	u32 regs[64];
200 
201 	switch (reginfo->ofs) {
202 	case IXGBE_SRRCTL(0):
203 		for (i = 0; i < 64; i++)
204 			regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
205 		break;
206 	case IXGBE_DCA_RXCTRL(0):
207 		for (i = 0; i < 64; i++)
208 			regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
209 		break;
210 	case IXGBE_RDLEN(0):
211 		for (i = 0; i < 64; i++)
212 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
213 		break;
214 	case IXGBE_RDH(0):
215 		for (i = 0; i < 64; i++)
216 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
217 		break;
218 	case IXGBE_RDT(0):
219 		for (i = 0; i < 64; i++)
220 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
221 		break;
222 	case IXGBE_RXDCTL(0):
223 		for (i = 0; i < 64; i++)
224 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
225 		break;
226 	case IXGBE_RDBAL(0):
227 		for (i = 0; i < 64; i++)
228 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
229 		break;
230 	case IXGBE_RDBAH(0):
231 		for (i = 0; i < 64; i++)
232 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
233 		break;
234 	case IXGBE_TDBAL(0):
235 		for (i = 0; i < 64; i++)
236 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
237 		break;
238 	case IXGBE_TDBAH(0):
239 		for (i = 0; i < 64; i++)
240 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
241 		break;
242 	case IXGBE_TDLEN(0):
243 		for (i = 0; i < 64; i++)
244 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
245 		break;
246 	case IXGBE_TDH(0):
247 		for (i = 0; i < 64; i++)
248 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
249 		break;
250 	case IXGBE_TDT(0):
251 		for (i = 0; i < 64; i++)
252 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
253 		break;
254 	case IXGBE_TXDCTL(0):
255 		for (i = 0; i < 64; i++)
256 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
257 		break;
258 	default:
259 		pr_info("%-15s %08x\n", reginfo->name,
260 			IXGBE_READ_REG(hw, reginfo->ofs));
261 		return;
262 	}
263 
264 	for (i = 0; i < 8; i++) {
265 		snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
266 		pr_err("%-15s", rname);
267 		for (j = 0; j < 8; j++)
268 			pr_cont(" %08x", regs[i*8+j]);
269 		pr_cont("\n");
270 	}
271 
272 }
273 
274 /*
275  * ixgbe_dump - Print registers, tx-rings and rx-rings
276  */
277 static void ixgbe_dump(struct ixgbe_adapter *adapter)
278 {
279 	struct net_device *netdev = adapter->netdev;
280 	struct ixgbe_hw *hw = &adapter->hw;
281 	struct ixgbe_reg_info *reginfo;
282 	int n = 0;
283 	struct ixgbe_ring *tx_ring;
284 	struct ixgbe_tx_buffer *tx_buffer_info;
285 	union ixgbe_adv_tx_desc *tx_desc;
286 	struct my_u0 { u64 a; u64 b; } *u0;
287 	struct ixgbe_ring *rx_ring;
288 	union ixgbe_adv_rx_desc *rx_desc;
289 	struct ixgbe_rx_buffer *rx_buffer_info;
290 	u32 staterr;
291 	int i = 0;
292 
293 	if (!netif_msg_hw(adapter))
294 		return;
295 
296 	/* Print netdevice Info */
297 	if (netdev) {
298 		dev_info(&adapter->pdev->dev, "Net device Info\n");
299 		pr_info("Device Name     state            "
300 			"trans_start      last_rx\n");
301 		pr_info("%-15s %016lX %016lX %016lX\n",
302 			netdev->name,
303 			netdev->state,
304 			netdev->trans_start,
305 			netdev->last_rx);
306 	}
307 
308 	/* Print Registers */
309 	dev_info(&adapter->pdev->dev, "Register Dump\n");
310 	pr_info(" Register Name   Value\n");
311 	for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
312 	     reginfo->name; reginfo++) {
313 		ixgbe_regdump(hw, reginfo);
314 	}
315 
316 	/* Print TX Ring Summary */
317 	if (!netdev || !netif_running(netdev))
318 		goto exit;
319 
320 	dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
321 	pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
322 	for (n = 0; n < adapter->num_tx_queues; n++) {
323 		tx_ring = adapter->tx_ring[n];
324 		tx_buffer_info =
325 			&tx_ring->tx_buffer_info[tx_ring->next_to_clean];
326 		pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
327 			   n, tx_ring->next_to_use, tx_ring->next_to_clean,
328 			   (u64)tx_buffer_info->dma,
329 			   tx_buffer_info->length,
330 			   tx_buffer_info->next_to_watch,
331 			   (u64)tx_buffer_info->time_stamp);
332 	}
333 
334 	/* Print TX Rings */
335 	if (!netif_msg_tx_done(adapter))
336 		goto rx_ring_summary;
337 
338 	dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
339 
340 	/* Transmit Descriptor Formats
341 	 *
342 	 * Advanced Transmit Descriptor
343 	 *   +--------------------------------------------------------------+
344 	 * 0 |         Buffer Address [63:0]                                |
345 	 *   +--------------------------------------------------------------+
346 	 * 8 |  PAYLEN  | PORTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
347 	 *   +--------------------------------------------------------------+
348 	 *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
349 	 */
350 
351 	for (n = 0; n < adapter->num_tx_queues; n++) {
352 		tx_ring = adapter->tx_ring[n];
353 		pr_info("------------------------------------\n");
354 		pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
355 		pr_info("------------------------------------\n");
356 		pr_info("T [desc]     [address 63:0  ] "
357 			"[PlPOIdStDDt Ln] [bi->dma       ] "
358 			"leng  ntw timestamp        bi->skb\n");
359 
360 		for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
361 			tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
362 			tx_buffer_info = &tx_ring->tx_buffer_info[i];
363 			u0 = (struct my_u0 *)tx_desc;
364 			pr_info("T [0x%03X]    %016llX %016llX %016llX"
365 				" %04X  %p %016llX %p", i,
366 				le64_to_cpu(u0->a),
367 				le64_to_cpu(u0->b),
368 				(u64)tx_buffer_info->dma,
369 				tx_buffer_info->length,
370 				tx_buffer_info->next_to_watch,
371 				(u64)tx_buffer_info->time_stamp,
372 				tx_buffer_info->skb);
373 			if (i == tx_ring->next_to_use &&
374 				i == tx_ring->next_to_clean)
375 				pr_cont(" NTC/U\n");
376 			else if (i == tx_ring->next_to_use)
377 				pr_cont(" NTU\n");
378 			else if (i == tx_ring->next_to_clean)
379 				pr_cont(" NTC\n");
380 			else
381 				pr_cont("\n");
382 
383 			if (netif_msg_pktdata(adapter) &&
384 				tx_buffer_info->dma != 0)
385 				print_hex_dump(KERN_INFO, "",
386 					DUMP_PREFIX_ADDRESS, 16, 1,
387 					phys_to_virt(tx_buffer_info->dma),
388 					tx_buffer_info->length, true);
389 		}
390 	}
391 
392 	/* Print RX Rings Summary */
393 rx_ring_summary:
394 	dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
395 	pr_info("Queue [NTU] [NTC]\n");
396 	for (n = 0; n < adapter->num_rx_queues; n++) {
397 		rx_ring = adapter->rx_ring[n];
398 		pr_info("%5d %5X %5X\n",
399 			n, rx_ring->next_to_use, rx_ring->next_to_clean);
400 	}
401 
402 	/* Print RX Rings */
403 	if (!netif_msg_rx_status(adapter))
404 		goto exit;
405 
406 	dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
407 
408 	/* Advanced Receive Descriptor (Read) Format
409 	 *    63                                           1        0
410 	 *    +-----------------------------------------------------+
411 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
412 	 *    +----------------------------------------------+------+
413 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
414 	 *    +-----------------------------------------------------+
415 	 *
416 	 *
417 	 * Advanced Receive Descriptor (Write-Back) Format
418 	 *
419 	 *   63       48 47    32 31  30      21 20 16 15   4 3     0
420 	 *   +------------------------------------------------------+
421 	 * 0 | Packet     IP     |SPH| HDR_LEN   | RSV|Packet|  RSS |
422 	 *   | Checksum   Ident  |   |           |    | Type | Type |
423 	 *   +------------------------------------------------------+
424 	 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
425 	 *   +------------------------------------------------------+
426 	 *   63       48 47    32 31            20 19               0
427 	 */
428 	for (n = 0; n < adapter->num_rx_queues; n++) {
429 		rx_ring = adapter->rx_ring[n];
430 		pr_info("------------------------------------\n");
431 		pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
432 		pr_info("------------------------------------\n");
433 		pr_info("R  [desc]      [ PktBuf     A0] "
434 			"[  HeadBuf   DD] [bi->dma       ] [bi->skb] "
435 			"<-- Adv Rx Read format\n");
436 		pr_info("RWB[desc]      [PcsmIpSHl PtRs] "
437 			"[vl er S cks ln] ---------------- [bi->skb] "
438 			"<-- Adv Rx Write-Back format\n");
439 
440 		for (i = 0; i < rx_ring->count; i++) {
441 			rx_buffer_info = &rx_ring->rx_buffer_info[i];
442 			rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
443 			u0 = (struct my_u0 *)rx_desc;
444 			staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
445 			if (staterr & IXGBE_RXD_STAT_DD) {
446 				/* Descriptor Done */
447 				pr_info("RWB[0x%03X]     %016llX "
448 					"%016llX ---------------- %p", i,
449 					le64_to_cpu(u0->a),
450 					le64_to_cpu(u0->b),
451 					rx_buffer_info->skb);
452 			} else {
453 				pr_info("R  [0x%03X]     %016llX "
454 					"%016llX %016llX %p", i,
455 					le64_to_cpu(u0->a),
456 					le64_to_cpu(u0->b),
457 					(u64)rx_buffer_info->dma,
458 					rx_buffer_info->skb);
459 
460 				if (netif_msg_pktdata(adapter)) {
461 					print_hex_dump(KERN_INFO, "",
462 					   DUMP_PREFIX_ADDRESS, 16, 1,
463 					   phys_to_virt(rx_buffer_info->dma),
464 					   rx_ring->rx_buf_len, true);
465 
466 					if (rx_ring->rx_buf_len
467 						< IXGBE_RXBUFFER_2K)
468 						print_hex_dump(KERN_INFO, "",
469 						  DUMP_PREFIX_ADDRESS, 16, 1,
470 						  phys_to_virt(
471 						    rx_buffer_info->page_dma +
472 						    rx_buffer_info->page_offset
473 						  ),
474 						  PAGE_SIZE/2, true);
475 				}
476 			}
477 
478 			if (i == rx_ring->next_to_use)
479 				pr_cont(" NTU\n");
480 			else if (i == rx_ring->next_to_clean)
481 				pr_cont(" NTC\n");
482 			else
483 				pr_cont("\n");
484 
485 		}
486 	}
487 
488 exit:
489 	return;
490 }
491 
492 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
493 {
494 	u32 ctrl_ext;
495 
496 	/* Let firmware take over control of h/w */
497 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
498 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
499 			ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
500 }
501 
502 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
503 {
504 	u32 ctrl_ext;
505 
506 	/* Let firmware know the driver has taken over */
507 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
508 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
509 			ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
510 }
511 
512 /*
513  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
514  * @adapter: pointer to adapter struct
515  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
516  * @queue: queue to map the corresponding interrupt to
517  * @msix_vector: the vector to map to the corresponding queue
518  *
519  */
520 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
521 			   u8 queue, u8 msix_vector)
522 {
523 	u32 ivar, index;
524 	struct ixgbe_hw *hw = &adapter->hw;
525 	switch (hw->mac.type) {
526 	case ixgbe_mac_82598EB:
527 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
528 		if (direction == -1)
529 			direction = 0;
530 		index = (((direction * 64) + queue) >> 2) & 0x1F;
531 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
532 		ivar &= ~(0xFF << (8 * (queue & 0x3)));
533 		ivar |= (msix_vector << (8 * (queue & 0x3)));
534 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
535 		break;
536 	case ixgbe_mac_82599EB:
537 	case ixgbe_mac_X540:
538 		if (direction == -1) {
539 			/* other causes */
540 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
541 			index = ((queue & 1) * 8);
542 			ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
543 			ivar &= ~(0xFF << index);
544 			ivar |= (msix_vector << index);
545 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
546 			break;
547 		} else {
548 			/* tx or rx causes */
549 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
550 			index = ((16 * (queue & 1)) + (8 * direction));
551 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
552 			ivar &= ~(0xFF << index);
553 			ivar |= (msix_vector << index);
554 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
555 			break;
556 		}
557 	default:
558 		break;
559 	}
560 }
561 
562 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
563 					  u64 qmask)
564 {
565 	u32 mask;
566 
567 	switch (adapter->hw.mac.type) {
568 	case ixgbe_mac_82598EB:
569 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
570 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
571 		break;
572 	case ixgbe_mac_82599EB:
573 	case ixgbe_mac_X540:
574 		mask = (qmask & 0xFFFFFFFF);
575 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
576 		mask = (qmask >> 32);
577 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
578 		break;
579 	default:
580 		break;
581 	}
582 }
583 
584 static inline void ixgbe_unmap_tx_resource(struct ixgbe_ring *ring,
585 					   struct ixgbe_tx_buffer *tx_buffer)
586 {
587 	if (tx_buffer->dma) {
588 		if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_MAPPED_AS_PAGE)
589 			dma_unmap_page(ring->dev,
590 			               tx_buffer->dma,
591 			               tx_buffer->length,
592 			               DMA_TO_DEVICE);
593 		else
594 			dma_unmap_single(ring->dev,
595 			                 tx_buffer->dma,
596 			                 tx_buffer->length,
597 			                 DMA_TO_DEVICE);
598 	}
599 	tx_buffer->dma = 0;
600 }
601 
602 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *tx_ring,
603 				      struct ixgbe_tx_buffer *tx_buffer_info)
604 {
605 	ixgbe_unmap_tx_resource(tx_ring, tx_buffer_info);
606 	if (tx_buffer_info->skb)
607 		dev_kfree_skb_any(tx_buffer_info->skb);
608 	tx_buffer_info->skb = NULL;
609 	/* tx_buffer_info must be completely set up in the transmit path */
610 }
611 
612 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
613 {
614 	struct ixgbe_hw *hw = &adapter->hw;
615 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
616 	u32 data = 0;
617 	u32 xoff[8] = {0};
618 	int i;
619 
620 	if ((hw->fc.current_mode == ixgbe_fc_full) ||
621 	    (hw->fc.current_mode == ixgbe_fc_rx_pause)) {
622 		switch (hw->mac.type) {
623 		case ixgbe_mac_82598EB:
624 			data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
625 			break;
626 		default:
627 			data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
628 		}
629 		hwstats->lxoffrxc += data;
630 
631 		/* refill credits (no tx hang) if we received xoff */
632 		if (!data)
633 			return;
634 
635 		for (i = 0; i < adapter->num_tx_queues; i++)
636 			clear_bit(__IXGBE_HANG_CHECK_ARMED,
637 				  &adapter->tx_ring[i]->state);
638 		return;
639 	} else if (!(adapter->dcb_cfg.pfc_mode_enable))
640 		return;
641 
642 	/* update stats for each tc, only valid with PFC enabled */
643 	for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
644 		switch (hw->mac.type) {
645 		case ixgbe_mac_82598EB:
646 			xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
647 			break;
648 		default:
649 			xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
650 		}
651 		hwstats->pxoffrxc[i] += xoff[i];
652 	}
653 
654 	/* disarm tx queues that have received xoff frames */
655 	for (i = 0; i < adapter->num_tx_queues; i++) {
656 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
657 		u8 tc = tx_ring->dcb_tc;
658 
659 		if (xoff[tc])
660 			clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
661 	}
662 }
663 
664 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
665 {
666 	return ring->tx_stats.completed;
667 }
668 
669 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
670 {
671 	struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
672 	struct ixgbe_hw *hw = &adapter->hw;
673 
674 	u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
675 	u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
676 
677 	if (head != tail)
678 		return (head < tail) ?
679 			tail - head : (tail + ring->count - head);
680 
681 	return 0;
682 }
683 
684 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
685 {
686 	u32 tx_done = ixgbe_get_tx_completed(tx_ring);
687 	u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
688 	u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
689 	bool ret = false;
690 
691 	clear_check_for_tx_hang(tx_ring);
692 
693 	/*
694 	 * Check for a hung queue, but be thorough. This verifies
695 	 * that a transmit has been completed since the previous
696 	 * check AND there is at least one packet pending. The
697 	 * ARMED bit is set to indicate a potential hang. The
698 	 * bit is cleared if a pause frame is received to remove
699 	 * false hang detection due to PFC or 802.3x frames. By
700 	 * requiring this to fail twice we avoid races with
701 	 * pfc clearing the ARMED bit and conditions where we
702 	 * run the check_tx_hang logic with a transmit completion
703 	 * pending but without time to complete it yet.
704 	 */
705 	if ((tx_done_old == tx_done) && tx_pending) {
706 		/* make sure it is true for two checks in a row */
707 		ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
708 				       &tx_ring->state);
709 	} else {
710 		/* update completed stats and continue */
711 		tx_ring->tx_stats.tx_done_old = tx_done;
712 		/* reset the countdown */
713 		clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
714 	}
715 
716 	return ret;
717 }
718 
719 /**
720  * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
721  * @adapter: driver private struct
722  **/
723 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
724 {
725 
726 	/* Do the reset outside of interrupt context */
727 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
728 		adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
729 		ixgbe_service_event_schedule(adapter);
730 	}
731 }
732 
733 /**
734  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
735  * @q_vector: structure containing interrupt and ring information
736  * @tx_ring: tx ring to clean
737  **/
738 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
739 			       struct ixgbe_ring *tx_ring)
740 {
741 	struct ixgbe_adapter *adapter = q_vector->adapter;
742 	struct ixgbe_tx_buffer *tx_buffer;
743 	union ixgbe_adv_tx_desc *tx_desc;
744 	unsigned int total_bytes = 0, total_packets = 0;
745 	unsigned int budget = q_vector->tx.work_limit;
746 	u16 i = tx_ring->next_to_clean;
747 
748 	tx_buffer = &tx_ring->tx_buffer_info[i];
749 	tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
750 
751 	for (; budget; budget--) {
752 		union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
753 
754 		/* if next_to_watch is not set then there is no work pending */
755 		if (!eop_desc)
756 			break;
757 
758 		/* if DD is not set pending work has not been completed */
759 		if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
760 			break;
761 
762 		/* count the packet as being completed */
763 		tx_ring->tx_stats.completed++;
764 
765 		/* clear next_to_watch to prevent false hangs */
766 		tx_buffer->next_to_watch = NULL;
767 
768 		/* prevent any other reads prior to eop_desc being verified */
769 		rmb();
770 
771 		do {
772 			ixgbe_unmap_tx_resource(tx_ring, tx_buffer);
773 			tx_desc->wb.status = 0;
774 			if (likely(tx_desc == eop_desc)) {
775 				eop_desc = NULL;
776 				dev_kfree_skb_any(tx_buffer->skb);
777 				tx_buffer->skb = NULL;
778 
779 				total_bytes += tx_buffer->bytecount;
780 				total_packets += tx_buffer->gso_segs;
781 			}
782 
783 			tx_buffer++;
784 			tx_desc++;
785 			i++;
786 			if (unlikely(i == tx_ring->count)) {
787 				i = 0;
788 
789 				tx_buffer = tx_ring->tx_buffer_info;
790 				tx_desc = IXGBE_TX_DESC_ADV(tx_ring, 0);
791 			}
792 
793 		} while (eop_desc);
794 	}
795 
796 	tx_ring->next_to_clean = i;
797 	u64_stats_update_begin(&tx_ring->syncp);
798 	tx_ring->stats.bytes += total_bytes;
799 	tx_ring->stats.packets += total_packets;
800 	u64_stats_update_end(&tx_ring->syncp);
801 	q_vector->tx.total_bytes += total_bytes;
802 	q_vector->tx.total_packets += total_packets;
803 
804 	if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
805 		/* schedule immediate reset if we believe we hung */
806 		struct ixgbe_hw *hw = &adapter->hw;
807 		tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
808 		e_err(drv, "Detected Tx Unit Hang\n"
809 			"  Tx Queue             <%d>\n"
810 			"  TDH, TDT             <%x>, <%x>\n"
811 			"  next_to_use          <%x>\n"
812 			"  next_to_clean        <%x>\n"
813 			"tx_buffer_info[next_to_clean]\n"
814 			"  time_stamp           <%lx>\n"
815 			"  jiffies              <%lx>\n",
816 			tx_ring->queue_index,
817 			IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
818 			IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
819 			tx_ring->next_to_use, i,
820 			tx_ring->tx_buffer_info[i].time_stamp, jiffies);
821 
822 		netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
823 
824 		e_info(probe,
825 		       "tx hang %d detected on queue %d, resetting adapter\n",
826 			adapter->tx_timeout_count + 1, tx_ring->queue_index);
827 
828 		/* schedule immediate reset if we believe we hung */
829 		ixgbe_tx_timeout_reset(adapter);
830 
831 		/* the adapter is about to reset, no point in enabling stuff */
832 		return true;
833 	}
834 
835 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
836 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
837 		     (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
838 		/* Make sure that anybody stopping the queue after this
839 		 * sees the new next_to_clean.
840 		 */
841 		smp_mb();
842 		if (__netif_subqueue_stopped(tx_ring->netdev, tx_ring->queue_index) &&
843 		    !test_bit(__IXGBE_DOWN, &adapter->state)) {
844 			netif_wake_subqueue(tx_ring->netdev, tx_ring->queue_index);
845 			++tx_ring->tx_stats.restart_queue;
846 		}
847 	}
848 
849 	return !!budget;
850 }
851 
852 #ifdef CONFIG_IXGBE_DCA
853 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
854 				struct ixgbe_ring *rx_ring,
855 				int cpu)
856 {
857 	struct ixgbe_hw *hw = &adapter->hw;
858 	u32 rxctrl;
859 	u8 reg_idx = rx_ring->reg_idx;
860 
861 	rxctrl = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(reg_idx));
862 	switch (hw->mac.type) {
863 	case ixgbe_mac_82598EB:
864 		rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
865 		rxctrl |= dca3_get_tag(rx_ring->dev, cpu);
866 		break;
867 	case ixgbe_mac_82599EB:
868 	case ixgbe_mac_X540:
869 		rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
870 		rxctrl |= (dca3_get_tag(rx_ring->dev, cpu) <<
871 			   IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
872 		break;
873 	default:
874 		break;
875 	}
876 	rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
877 	rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
878 	rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
879 	IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
880 }
881 
882 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
883 				struct ixgbe_ring *tx_ring,
884 				int cpu)
885 {
886 	struct ixgbe_hw *hw = &adapter->hw;
887 	u32 txctrl;
888 	u8 reg_idx = tx_ring->reg_idx;
889 
890 	switch (hw->mac.type) {
891 	case ixgbe_mac_82598EB:
892 		txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(reg_idx));
893 		txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
894 		txctrl |= dca3_get_tag(tx_ring->dev, cpu);
895 		txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
896 		IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(reg_idx), txctrl);
897 		break;
898 	case ixgbe_mac_82599EB:
899 	case ixgbe_mac_X540:
900 		txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx));
901 		txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
902 		txctrl |= (dca3_get_tag(tx_ring->dev, cpu) <<
903 			   IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
904 		txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
905 		IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx), txctrl);
906 		break;
907 	default:
908 		break;
909 	}
910 }
911 
912 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
913 {
914 	struct ixgbe_adapter *adapter = q_vector->adapter;
915 	struct ixgbe_ring *ring;
916 	int cpu = get_cpu();
917 
918 	if (q_vector->cpu == cpu)
919 		goto out_no_update;
920 
921 	for (ring = q_vector->tx.ring; ring != NULL; ring = ring->next)
922 		ixgbe_update_tx_dca(adapter, ring, cpu);
923 
924 	for (ring = q_vector->rx.ring; ring != NULL; ring = ring->next)
925 		ixgbe_update_rx_dca(adapter, ring, cpu);
926 
927 	q_vector->cpu = cpu;
928 out_no_update:
929 	put_cpu();
930 }
931 
932 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
933 {
934 	int num_q_vectors;
935 	int i;
936 
937 	if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
938 		return;
939 
940 	/* always use CB2 mode, difference is masked in the CB driver */
941 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
942 
943 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
944 		num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
945 	else
946 		num_q_vectors = 1;
947 
948 	for (i = 0; i < num_q_vectors; i++) {
949 		adapter->q_vector[i]->cpu = -1;
950 		ixgbe_update_dca(adapter->q_vector[i]);
951 	}
952 }
953 
954 static int __ixgbe_notify_dca(struct device *dev, void *data)
955 {
956 	struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
957 	unsigned long event = *(unsigned long *)data;
958 
959 	if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
960 		return 0;
961 
962 	switch (event) {
963 	case DCA_PROVIDER_ADD:
964 		/* if we're already enabled, don't do it again */
965 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
966 			break;
967 		if (dca_add_requester(dev) == 0) {
968 			adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
969 			ixgbe_setup_dca(adapter);
970 			break;
971 		}
972 		/* Fall Through since DCA is disabled. */
973 	case DCA_PROVIDER_REMOVE:
974 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
975 			dca_remove_requester(dev);
976 			adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
977 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
978 		}
979 		break;
980 	}
981 
982 	return 0;
983 }
984 #endif /* CONFIG_IXGBE_DCA */
985 
986 static inline void ixgbe_rx_hash(union ixgbe_adv_rx_desc *rx_desc,
987 				 struct sk_buff *skb)
988 {
989 	skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
990 }
991 
992 /**
993  * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
994  * @adapter: address of board private structure
995  * @rx_desc: advanced rx descriptor
996  *
997  * Returns : true if it is FCoE pkt
998  */
999 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_adapter *adapter,
1000 				    union ixgbe_adv_rx_desc *rx_desc)
1001 {
1002 	__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1003 
1004 	return (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
1005 	       ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1006 		(cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1007 			     IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1008 }
1009 
1010 /**
1011  * ixgbe_receive_skb - Send a completed packet up the stack
1012  * @adapter: board private structure
1013  * @skb: packet to send up
1014  * @status: hardware indication of status of receive
1015  * @rx_ring: rx descriptor ring (for a specific queue) to setup
1016  * @rx_desc: rx descriptor
1017  **/
1018 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
1019 			      struct sk_buff *skb, u8 status,
1020 			      struct ixgbe_ring *ring,
1021 			      union ixgbe_adv_rx_desc *rx_desc)
1022 {
1023 	struct ixgbe_adapter *adapter = q_vector->adapter;
1024 	struct napi_struct *napi = &q_vector->napi;
1025 	bool is_vlan = (status & IXGBE_RXD_STAT_VP);
1026 	u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
1027 
1028 	if (is_vlan && (tag & VLAN_VID_MASK))
1029 		__vlan_hwaccel_put_tag(skb, tag);
1030 
1031 	if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1032 		napi_gro_receive(napi, skb);
1033 	else
1034 		netif_rx(skb);
1035 }
1036 
1037 /**
1038  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1039  * @adapter: address of board private structure
1040  * @status_err: hardware indication of status of receive
1041  * @skb: skb currently being received and modified
1042  * @status_err: status error value of last descriptor in packet
1043  **/
1044 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
1045 				     union ixgbe_adv_rx_desc *rx_desc,
1046 				     struct sk_buff *skb,
1047 				     u32 status_err)
1048 {
1049 	skb->ip_summed = CHECKSUM_NONE;
1050 
1051 	/* Rx csum disabled */
1052 	if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
1053 		return;
1054 
1055 	/* if IP and error */
1056 	if ((status_err & IXGBE_RXD_STAT_IPCS) &&
1057 	    (status_err & IXGBE_RXDADV_ERR_IPE)) {
1058 		adapter->hw_csum_rx_error++;
1059 		return;
1060 	}
1061 
1062 	if (!(status_err & IXGBE_RXD_STAT_L4CS))
1063 		return;
1064 
1065 	if (status_err & IXGBE_RXDADV_ERR_TCPE) {
1066 		u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1067 
1068 		/*
1069 		 * 82599 errata, UDP frames with a 0 checksum can be marked as
1070 		 * checksum errors.
1071 		 */
1072 		if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
1073 		    (adapter->hw.mac.type == ixgbe_mac_82599EB))
1074 			return;
1075 
1076 		adapter->hw_csum_rx_error++;
1077 		return;
1078 	}
1079 
1080 	/* It must be a TCP or UDP packet with a valid checksum */
1081 	skb->ip_summed = CHECKSUM_UNNECESSARY;
1082 }
1083 
1084 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1085 {
1086 	/*
1087 	 * Force memory writes to complete before letting h/w
1088 	 * know there are new descriptors to fetch.  (Only
1089 	 * applicable for weak-ordered memory model archs,
1090 	 * such as IA-64).
1091 	 */
1092 	wmb();
1093 	writel(val, rx_ring->tail);
1094 }
1095 
1096 /**
1097  * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1098  * @rx_ring: ring to place buffers on
1099  * @cleaned_count: number of buffers to replace
1100  **/
1101 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1102 {
1103 	union ixgbe_adv_rx_desc *rx_desc;
1104 	struct ixgbe_rx_buffer *bi;
1105 	struct sk_buff *skb;
1106 	u16 i = rx_ring->next_to_use;
1107 
1108 	/* do nothing if no valid netdev defined */
1109 	if (!rx_ring->netdev)
1110 		return;
1111 
1112 	while (cleaned_count--) {
1113 		rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1114 		bi = &rx_ring->rx_buffer_info[i];
1115 		skb = bi->skb;
1116 
1117 		if (!skb) {
1118 			skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1119 							rx_ring->rx_buf_len);
1120 			if (!skb) {
1121 				rx_ring->rx_stats.alloc_rx_buff_failed++;
1122 				goto no_buffers;
1123 			}
1124 			/* initialize queue mapping */
1125 			skb_record_rx_queue(skb, rx_ring->queue_index);
1126 			bi->skb = skb;
1127 		}
1128 
1129 		if (!bi->dma) {
1130 			bi->dma = dma_map_single(rx_ring->dev,
1131 						 skb->data,
1132 						 rx_ring->rx_buf_len,
1133 						 DMA_FROM_DEVICE);
1134 			if (dma_mapping_error(rx_ring->dev, bi->dma)) {
1135 				rx_ring->rx_stats.alloc_rx_buff_failed++;
1136 				bi->dma = 0;
1137 				goto no_buffers;
1138 			}
1139 		}
1140 
1141 		if (ring_is_ps_enabled(rx_ring)) {
1142 			if (!bi->page) {
1143 				bi->page = netdev_alloc_page(rx_ring->netdev);
1144 				if (!bi->page) {
1145 					rx_ring->rx_stats.alloc_rx_page_failed++;
1146 					goto no_buffers;
1147 				}
1148 			}
1149 
1150 			if (!bi->page_dma) {
1151 				/* use a half page if we're re-using */
1152 				bi->page_offset ^= PAGE_SIZE / 2;
1153 				bi->page_dma = dma_map_page(rx_ring->dev,
1154 							    bi->page,
1155 							    bi->page_offset,
1156 							    PAGE_SIZE / 2,
1157 							    DMA_FROM_DEVICE);
1158 				if (dma_mapping_error(rx_ring->dev,
1159 						      bi->page_dma)) {
1160 					rx_ring->rx_stats.alloc_rx_page_failed++;
1161 					bi->page_dma = 0;
1162 					goto no_buffers;
1163 				}
1164 			}
1165 
1166 			/* Refresh the desc even if buffer_addrs didn't change
1167 			 * because each write-back erases this info. */
1168 			rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1169 			rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1170 		} else {
1171 			rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1172 			rx_desc->read.hdr_addr = 0;
1173 		}
1174 
1175 		i++;
1176 		if (i == rx_ring->count)
1177 			i = 0;
1178 	}
1179 
1180 no_buffers:
1181 	if (rx_ring->next_to_use != i) {
1182 		rx_ring->next_to_use = i;
1183 		ixgbe_release_rx_desc(rx_ring, i);
1184 	}
1185 }
1186 
1187 static inline u16 ixgbe_get_hlen(union ixgbe_adv_rx_desc *rx_desc)
1188 {
1189 	/* HW will not DMA in data larger than the given buffer, even if it
1190 	 * parses the (NFS, of course) header to be larger.  In that case, it
1191 	 * fills the header buffer and spills the rest into the page.
1192 	 */
1193 	u16 hdr_info = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info);
1194 	u16 hlen = (hdr_info &  IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1195 		    IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1196 	if (hlen > IXGBE_RX_HDR_SIZE)
1197 		hlen = IXGBE_RX_HDR_SIZE;
1198 	return hlen;
1199 }
1200 
1201 /**
1202  * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1203  * @skb: pointer to the last skb in the rsc queue
1204  *
1205  * This function changes a queue full of hw rsc buffers into a completed
1206  * packet.  It uses the ->prev pointers to find the first packet and then
1207  * turns it into the frag list owner.
1208  **/
1209 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb)
1210 {
1211 	unsigned int frag_list_size = 0;
1212 	unsigned int skb_cnt = 1;
1213 
1214 	while (skb->prev) {
1215 		struct sk_buff *prev = skb->prev;
1216 		frag_list_size += skb->len;
1217 		skb->prev = NULL;
1218 		skb = prev;
1219 		skb_cnt++;
1220 	}
1221 
1222 	skb_shinfo(skb)->frag_list = skb->next;
1223 	skb->next = NULL;
1224 	skb->len += frag_list_size;
1225 	skb->data_len += frag_list_size;
1226 	skb->truesize += frag_list_size;
1227 	IXGBE_RSC_CB(skb)->skb_cnt = skb_cnt;
1228 
1229 	return skb;
1230 }
1231 
1232 static inline bool ixgbe_get_rsc_state(union ixgbe_adv_rx_desc *rx_desc)
1233 {
1234 	return !!(le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1235 		IXGBE_RXDADV_RSCCNT_MASK);
1236 }
1237 
1238 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1239 			       struct ixgbe_ring *rx_ring,
1240 			       int budget)
1241 {
1242 	struct ixgbe_adapter *adapter = q_vector->adapter;
1243 	union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1244 	struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1245 	struct sk_buff *skb;
1246 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1247 	const int current_node = numa_node_id();
1248 #ifdef IXGBE_FCOE
1249 	int ddp_bytes = 0;
1250 #endif /* IXGBE_FCOE */
1251 	u32 staterr;
1252 	u16 i;
1253 	u16 cleaned_count = 0;
1254 	bool pkt_is_rsc = false;
1255 
1256 	i = rx_ring->next_to_clean;
1257 	rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1258 	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1259 
1260 	while (staterr & IXGBE_RXD_STAT_DD) {
1261 		u32 upper_len = 0;
1262 
1263 		rmb(); /* read descriptor and rx_buffer_info after status DD */
1264 
1265 		rx_buffer_info = &rx_ring->rx_buffer_info[i];
1266 
1267 		skb = rx_buffer_info->skb;
1268 		rx_buffer_info->skb = NULL;
1269 		prefetch(skb->data);
1270 
1271 		if (ring_is_rsc_enabled(rx_ring))
1272 			pkt_is_rsc = ixgbe_get_rsc_state(rx_desc);
1273 
1274 		/* linear means we are building an skb from multiple pages */
1275 		if (!skb_is_nonlinear(skb)) {
1276 			u16 hlen;
1277 			if (pkt_is_rsc &&
1278 			    !(staterr & IXGBE_RXD_STAT_EOP) &&
1279 			    !skb->prev) {
1280 				/*
1281 				 * When HWRSC is enabled, delay unmapping
1282 				 * of the first packet. It carries the
1283 				 * header information, HW may still
1284 				 * access the header after the writeback.
1285 				 * Only unmap it when EOP is reached
1286 				 */
1287 				IXGBE_RSC_CB(skb)->delay_unmap = true;
1288 				IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1289 			} else {
1290 				dma_unmap_single(rx_ring->dev,
1291 						 rx_buffer_info->dma,
1292 						 rx_ring->rx_buf_len,
1293 						 DMA_FROM_DEVICE);
1294 			}
1295 			rx_buffer_info->dma = 0;
1296 
1297 			if (ring_is_ps_enabled(rx_ring)) {
1298 				hlen = ixgbe_get_hlen(rx_desc);
1299 				upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1300 			} else {
1301 				hlen = le16_to_cpu(rx_desc->wb.upper.length);
1302 			}
1303 
1304 			skb_put(skb, hlen);
1305 		} else {
1306 			/* assume packet split since header is unmapped */
1307 			upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1308 		}
1309 
1310 		if (upper_len) {
1311 			dma_unmap_page(rx_ring->dev,
1312 				       rx_buffer_info->page_dma,
1313 				       PAGE_SIZE / 2,
1314 				       DMA_FROM_DEVICE);
1315 			rx_buffer_info->page_dma = 0;
1316 			skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1317 					   rx_buffer_info->page,
1318 					   rx_buffer_info->page_offset,
1319 					   upper_len);
1320 
1321 			if ((page_count(rx_buffer_info->page) == 1) &&
1322 			    (page_to_nid(rx_buffer_info->page) == current_node))
1323 				get_page(rx_buffer_info->page);
1324 			else
1325 				rx_buffer_info->page = NULL;
1326 
1327 			skb->len += upper_len;
1328 			skb->data_len += upper_len;
1329 			skb->truesize += PAGE_SIZE / 2;
1330 		}
1331 
1332 		i++;
1333 		if (i == rx_ring->count)
1334 			i = 0;
1335 
1336 		next_rxd = IXGBE_RX_DESC_ADV(rx_ring, i);
1337 		prefetch(next_rxd);
1338 		cleaned_count++;
1339 
1340 		if (pkt_is_rsc) {
1341 			u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1342 				     IXGBE_RXDADV_NEXTP_SHIFT;
1343 			next_buffer = &rx_ring->rx_buffer_info[nextp];
1344 		} else {
1345 			next_buffer = &rx_ring->rx_buffer_info[i];
1346 		}
1347 
1348 		if (!(staterr & IXGBE_RXD_STAT_EOP)) {
1349 			if (ring_is_ps_enabled(rx_ring)) {
1350 				rx_buffer_info->skb = next_buffer->skb;
1351 				rx_buffer_info->dma = next_buffer->dma;
1352 				next_buffer->skb = skb;
1353 				next_buffer->dma = 0;
1354 			} else {
1355 				skb->next = next_buffer->skb;
1356 				skb->next->prev = skb;
1357 			}
1358 			rx_ring->rx_stats.non_eop_descs++;
1359 			goto next_desc;
1360 		}
1361 
1362 		if (skb->prev) {
1363 			skb = ixgbe_transform_rsc_queue(skb);
1364 			/* if we got here without RSC the packet is invalid */
1365 			if (!pkt_is_rsc) {
1366 				__pskb_trim(skb, 0);
1367 				rx_buffer_info->skb = skb;
1368 				goto next_desc;
1369 			}
1370 		}
1371 
1372 		if (ring_is_rsc_enabled(rx_ring)) {
1373 			if (IXGBE_RSC_CB(skb)->delay_unmap) {
1374 				dma_unmap_single(rx_ring->dev,
1375 						 IXGBE_RSC_CB(skb)->dma,
1376 						 rx_ring->rx_buf_len,
1377 						 DMA_FROM_DEVICE);
1378 				IXGBE_RSC_CB(skb)->dma = 0;
1379 				IXGBE_RSC_CB(skb)->delay_unmap = false;
1380 			}
1381 		}
1382 		if (pkt_is_rsc) {
1383 			if (ring_is_ps_enabled(rx_ring))
1384 				rx_ring->rx_stats.rsc_count +=
1385 					skb_shinfo(skb)->nr_frags;
1386 			else
1387 				rx_ring->rx_stats.rsc_count +=
1388 					IXGBE_RSC_CB(skb)->skb_cnt;
1389 			rx_ring->rx_stats.rsc_flush++;
1390 		}
1391 
1392 		/* ERR_MASK will only have valid bits if EOP set */
1393 		if (unlikely(staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK)) {
1394 			dev_kfree_skb_any(skb);
1395 			goto next_desc;
1396 		}
1397 
1398 		ixgbe_rx_checksum(adapter, rx_desc, skb, staterr);
1399 		if (adapter->netdev->features & NETIF_F_RXHASH)
1400 			ixgbe_rx_hash(rx_desc, skb);
1401 
1402 		/* probably a little skewed due to removing CRC */
1403 		total_rx_bytes += skb->len;
1404 		total_rx_packets++;
1405 
1406 		skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1407 #ifdef IXGBE_FCOE
1408 		/* if ddp, not passing to ULD unless for FCP_RSP or error */
1409 		if (ixgbe_rx_is_fcoe(adapter, rx_desc)) {
1410 			ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb,
1411 						   staterr);
1412 			if (!ddp_bytes) {
1413 				dev_kfree_skb_any(skb);
1414 				goto next_desc;
1415 			}
1416 		}
1417 #endif /* IXGBE_FCOE */
1418 		ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1419 
1420 		budget--;
1421 next_desc:
1422 		rx_desc->wb.upper.status_error = 0;
1423 
1424 		if (!budget)
1425 			break;
1426 
1427 		/* return some buffers to hardware, one at a time is too slow */
1428 		if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1429 			ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1430 			cleaned_count = 0;
1431 		}
1432 
1433 		/* use prefetched values */
1434 		rx_desc = next_rxd;
1435 		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1436 	}
1437 
1438 	rx_ring->next_to_clean = i;
1439 	cleaned_count = ixgbe_desc_unused(rx_ring);
1440 
1441 	if (cleaned_count)
1442 		ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1443 
1444 #ifdef IXGBE_FCOE
1445 	/* include DDPed FCoE data */
1446 	if (ddp_bytes > 0) {
1447 		unsigned int mss;
1448 
1449 		mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1450 			sizeof(struct fc_frame_header) -
1451 			sizeof(struct fcoe_crc_eof);
1452 		if (mss > 512)
1453 			mss &= ~511;
1454 		total_rx_bytes += ddp_bytes;
1455 		total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1456 	}
1457 #endif /* IXGBE_FCOE */
1458 
1459 	u64_stats_update_begin(&rx_ring->syncp);
1460 	rx_ring->stats.packets += total_rx_packets;
1461 	rx_ring->stats.bytes += total_rx_bytes;
1462 	u64_stats_update_end(&rx_ring->syncp);
1463 	q_vector->rx.total_packets += total_rx_packets;
1464 	q_vector->rx.total_bytes += total_rx_bytes;
1465 
1466 	return !!budget;
1467 }
1468 
1469 /**
1470  * ixgbe_configure_msix - Configure MSI-X hardware
1471  * @adapter: board private structure
1472  *
1473  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1474  * interrupts.
1475  **/
1476 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1477 {
1478 	struct ixgbe_q_vector *q_vector;
1479 	int q_vectors, v_idx;
1480 	u32 mask;
1481 
1482 	q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1483 
1484 	/* Populate MSIX to EITR Select */
1485 	if (adapter->num_vfs > 32) {
1486 		u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
1487 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
1488 	}
1489 
1490 	/*
1491 	 * Populate the IVAR table and set the ITR values to the
1492 	 * corresponding register.
1493 	 */
1494 	for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1495 		struct ixgbe_ring *ring;
1496 		q_vector = adapter->q_vector[v_idx];
1497 
1498 		for (ring = q_vector->rx.ring; ring != NULL; ring = ring->next)
1499 			ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1500 
1501 		for (ring = q_vector->tx.ring; ring != NULL; ring = ring->next)
1502 			ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1503 
1504 		if (q_vector->tx.ring && !q_vector->rx.ring) {
1505 			/* tx only vector */
1506 			if (adapter->tx_itr_setting == 1)
1507 				q_vector->itr = IXGBE_10K_ITR;
1508 			else
1509 				q_vector->itr = adapter->tx_itr_setting;
1510 		} else {
1511 			/* rx or rx/tx vector */
1512 			if (adapter->rx_itr_setting == 1)
1513 				q_vector->itr = IXGBE_20K_ITR;
1514 			else
1515 				q_vector->itr = adapter->rx_itr_setting;
1516 		}
1517 
1518 		ixgbe_write_eitr(q_vector);
1519 	}
1520 
1521 	switch (adapter->hw.mac.type) {
1522 	case ixgbe_mac_82598EB:
1523 		ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1524 			       v_idx);
1525 		break;
1526 	case ixgbe_mac_82599EB:
1527 	case ixgbe_mac_X540:
1528 		ixgbe_set_ivar(adapter, -1, 1, v_idx);
1529 		break;
1530 	default:
1531 		break;
1532 	}
1533 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1534 
1535 	/* set up to autoclear timer, and the vectors */
1536 	mask = IXGBE_EIMS_ENABLE_MASK;
1537 	mask &= ~(IXGBE_EIMS_OTHER |
1538 		  IXGBE_EIMS_MAILBOX |
1539 		  IXGBE_EIMS_LSC);
1540 
1541 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1542 }
1543 
1544 enum latency_range {
1545 	lowest_latency = 0,
1546 	low_latency = 1,
1547 	bulk_latency = 2,
1548 	latency_invalid = 255
1549 };
1550 
1551 /**
1552  * ixgbe_update_itr - update the dynamic ITR value based on statistics
1553  * @q_vector: structure containing interrupt and ring information
1554  * @ring_container: structure containing ring performance data
1555  *
1556  *      Stores a new ITR value based on packets and byte
1557  *      counts during the last interrupt.  The advantage of per interrupt
1558  *      computation is faster updates and more accurate ITR for the current
1559  *      traffic pattern.  Constants in this function were computed
1560  *      based on theoretical maximum wire speed and thresholds were set based
1561  *      on testing data as well as attempting to minimize response time
1562  *      while increasing bulk throughput.
1563  *      this functionality is controlled by the InterruptThrottleRate module
1564  *      parameter (see ixgbe_param.c)
1565  **/
1566 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
1567 			     struct ixgbe_ring_container *ring_container)
1568 {
1569 	u64 bytes_perint;
1570 	struct ixgbe_adapter *adapter = q_vector->adapter;
1571 	int bytes = ring_container->total_bytes;
1572 	int packets = ring_container->total_packets;
1573 	u32 timepassed_us;
1574 	u8 itr_setting = ring_container->itr;
1575 
1576 	if (packets == 0)
1577 		return;
1578 
1579 	/* simple throttlerate management
1580 	 *    0-20MB/s lowest (100000 ints/s)
1581 	 *   20-100MB/s low   (20000 ints/s)
1582 	 *  100-1249MB/s bulk (8000 ints/s)
1583 	 */
1584 	/* what was last interrupt timeslice? */
1585 	timepassed_us = q_vector->itr >> 2;
1586 	bytes_perint = bytes / timepassed_us; /* bytes/usec */
1587 
1588 	switch (itr_setting) {
1589 	case lowest_latency:
1590 		if (bytes_perint > adapter->eitr_low)
1591 			itr_setting = low_latency;
1592 		break;
1593 	case low_latency:
1594 		if (bytes_perint > adapter->eitr_high)
1595 			itr_setting = bulk_latency;
1596 		else if (bytes_perint <= adapter->eitr_low)
1597 			itr_setting = lowest_latency;
1598 		break;
1599 	case bulk_latency:
1600 		if (bytes_perint <= adapter->eitr_high)
1601 			itr_setting = low_latency;
1602 		break;
1603 	}
1604 
1605 	/* clear work counters since we have the values we need */
1606 	ring_container->total_bytes = 0;
1607 	ring_container->total_packets = 0;
1608 
1609 	/* write updated itr to ring container */
1610 	ring_container->itr = itr_setting;
1611 }
1612 
1613 /**
1614  * ixgbe_write_eitr - write EITR register in hardware specific way
1615  * @q_vector: structure containing interrupt and ring information
1616  *
1617  * This function is made to be called by ethtool and by the driver
1618  * when it needs to update EITR registers at runtime.  Hardware
1619  * specific quirks/differences are taken care of here.
1620  */
1621 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1622 {
1623 	struct ixgbe_adapter *adapter = q_vector->adapter;
1624 	struct ixgbe_hw *hw = &adapter->hw;
1625 	int v_idx = q_vector->v_idx;
1626 	u32 itr_reg = q_vector->itr;
1627 
1628 	switch (adapter->hw.mac.type) {
1629 	case ixgbe_mac_82598EB:
1630 		/* must write high and low 16 bits to reset counter */
1631 		itr_reg |= (itr_reg << 16);
1632 		break;
1633 	case ixgbe_mac_82599EB:
1634 	case ixgbe_mac_X540:
1635 		/*
1636 		 * set the WDIS bit to not clear the timer bits and cause an
1637 		 * immediate assertion of the interrupt
1638 		 */
1639 		itr_reg |= IXGBE_EITR_CNT_WDIS;
1640 		break;
1641 	default:
1642 		break;
1643 	}
1644 	IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1645 }
1646 
1647 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
1648 {
1649 	u32 new_itr = q_vector->itr;
1650 	u8 current_itr;
1651 
1652 	ixgbe_update_itr(q_vector, &q_vector->tx);
1653 	ixgbe_update_itr(q_vector, &q_vector->rx);
1654 
1655 	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1656 
1657 	switch (current_itr) {
1658 	/* counts and packets in update_itr are dependent on these numbers */
1659 	case lowest_latency:
1660 		new_itr = IXGBE_100K_ITR;
1661 		break;
1662 	case low_latency:
1663 		new_itr = IXGBE_20K_ITR;
1664 		break;
1665 	case bulk_latency:
1666 		new_itr = IXGBE_8K_ITR;
1667 		break;
1668 	default:
1669 		break;
1670 	}
1671 
1672 	if (new_itr != q_vector->itr) {
1673 		/* do an exponential smoothing */
1674 		new_itr = (10 * new_itr * q_vector->itr) /
1675 			  ((9 * new_itr) + q_vector->itr);
1676 
1677 		/* save the algorithm value here */
1678 		q_vector->itr = new_itr & IXGBE_MAX_EITR;
1679 
1680 		ixgbe_write_eitr(q_vector);
1681 	}
1682 }
1683 
1684 /**
1685  * ixgbe_check_overtemp_subtask - check for over tempurature
1686  * @adapter: pointer to adapter
1687  **/
1688 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
1689 {
1690 	struct ixgbe_hw *hw = &adapter->hw;
1691 	u32 eicr = adapter->interrupt_event;
1692 
1693 	if (test_bit(__IXGBE_DOWN, &adapter->state))
1694 		return;
1695 
1696 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1697 	    !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
1698 		return;
1699 
1700 	adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
1701 
1702 	switch (hw->device_id) {
1703 	case IXGBE_DEV_ID_82599_T3_LOM:
1704 		/*
1705 		 * Since the warning interrupt is for both ports
1706 		 * we don't have to check if:
1707 		 *  - This interrupt wasn't for our port.
1708 		 *  - We may have missed the interrupt so always have to
1709 		 *    check if we  got a LSC
1710 		 */
1711 		if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
1712 		    !(eicr & IXGBE_EICR_LSC))
1713 			return;
1714 
1715 		if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
1716 			u32 autoneg;
1717 			bool link_up = false;
1718 
1719 			hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1720 
1721 			if (link_up)
1722 				return;
1723 		}
1724 
1725 		/* Check if this is not due to overtemp */
1726 		if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
1727 			return;
1728 
1729 		break;
1730 	default:
1731 		if (!(eicr & IXGBE_EICR_GPI_SDP0))
1732 			return;
1733 		break;
1734 	}
1735 	e_crit(drv,
1736 	       "Network adapter has been stopped because it has over heated. "
1737 	       "Restart the computer. If the problem persists, "
1738 	       "power off the system and replace the adapter\n");
1739 
1740 	adapter->interrupt_event = 0;
1741 }
1742 
1743 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1744 {
1745 	struct ixgbe_hw *hw = &adapter->hw;
1746 
1747 	if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1748 	    (eicr & IXGBE_EICR_GPI_SDP1)) {
1749 		e_crit(probe, "Fan has stopped, replace the adapter\n");
1750 		/* write to clear the interrupt */
1751 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1752 	}
1753 }
1754 
1755 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
1756 {
1757 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
1758 		return;
1759 
1760 	switch (adapter->hw.mac.type) {
1761 	case ixgbe_mac_82599EB:
1762 		/*
1763 		 * Need to check link state so complete overtemp check
1764 		 * on service task
1765 		 */
1766 		if (((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)) &&
1767 		    (!test_bit(__IXGBE_DOWN, &adapter->state))) {
1768 			adapter->interrupt_event = eicr;
1769 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
1770 			ixgbe_service_event_schedule(adapter);
1771 			return;
1772 		}
1773 		return;
1774 	case ixgbe_mac_X540:
1775 		if (!(eicr & IXGBE_EICR_TS))
1776 			return;
1777 		break;
1778 	default:
1779 		return;
1780 	}
1781 
1782 	e_crit(drv,
1783 	       "Network adapter has been stopped because it has over heated. "
1784 	       "Restart the computer. If the problem persists, "
1785 	       "power off the system and replace the adapter\n");
1786 }
1787 
1788 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1789 {
1790 	struct ixgbe_hw *hw = &adapter->hw;
1791 
1792 	if (eicr & IXGBE_EICR_GPI_SDP2) {
1793 		/* Clear the interrupt */
1794 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1795 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1796 			adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
1797 			ixgbe_service_event_schedule(adapter);
1798 		}
1799 	}
1800 
1801 	if (eicr & IXGBE_EICR_GPI_SDP1) {
1802 		/* Clear the interrupt */
1803 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1804 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1805 			adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
1806 			ixgbe_service_event_schedule(adapter);
1807 		}
1808 	}
1809 }
1810 
1811 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1812 {
1813 	struct ixgbe_hw *hw = &adapter->hw;
1814 
1815 	adapter->lsc_int++;
1816 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1817 	adapter->link_check_timeout = jiffies;
1818 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1819 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1820 		IXGBE_WRITE_FLUSH(hw);
1821 		ixgbe_service_event_schedule(adapter);
1822 	}
1823 }
1824 
1825 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1826 					   u64 qmask)
1827 {
1828 	u32 mask;
1829 	struct ixgbe_hw *hw = &adapter->hw;
1830 
1831 	switch (hw->mac.type) {
1832 	case ixgbe_mac_82598EB:
1833 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1834 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
1835 		break;
1836 	case ixgbe_mac_82599EB:
1837 	case ixgbe_mac_X540:
1838 		mask = (qmask & 0xFFFFFFFF);
1839 		if (mask)
1840 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
1841 		mask = (qmask >> 32);
1842 		if (mask)
1843 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
1844 		break;
1845 	default:
1846 		break;
1847 	}
1848 	/* skip the flush */
1849 }
1850 
1851 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1852 					    u64 qmask)
1853 {
1854 	u32 mask;
1855 	struct ixgbe_hw *hw = &adapter->hw;
1856 
1857 	switch (hw->mac.type) {
1858 	case ixgbe_mac_82598EB:
1859 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1860 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
1861 		break;
1862 	case ixgbe_mac_82599EB:
1863 	case ixgbe_mac_X540:
1864 		mask = (qmask & 0xFFFFFFFF);
1865 		if (mask)
1866 			IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
1867 		mask = (qmask >> 32);
1868 		if (mask)
1869 			IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
1870 		break;
1871 	default:
1872 		break;
1873 	}
1874 	/* skip the flush */
1875 }
1876 
1877 /**
1878  * ixgbe_irq_enable - Enable default interrupt generation settings
1879  * @adapter: board private structure
1880  **/
1881 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
1882 				    bool flush)
1883 {
1884 	u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
1885 
1886 	/* don't reenable LSC while waiting for link */
1887 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
1888 		mask &= ~IXGBE_EIMS_LSC;
1889 
1890 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
1891 		switch (adapter->hw.mac.type) {
1892 		case ixgbe_mac_82599EB:
1893 			mask |= IXGBE_EIMS_GPI_SDP0;
1894 			break;
1895 		case ixgbe_mac_X540:
1896 			mask |= IXGBE_EIMS_TS;
1897 			break;
1898 		default:
1899 			break;
1900 		}
1901 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
1902 		mask |= IXGBE_EIMS_GPI_SDP1;
1903 	switch (adapter->hw.mac.type) {
1904 	case ixgbe_mac_82599EB:
1905 		mask |= IXGBE_EIMS_GPI_SDP1;
1906 		mask |= IXGBE_EIMS_GPI_SDP2;
1907 	case ixgbe_mac_X540:
1908 		mask |= IXGBE_EIMS_ECC;
1909 		mask |= IXGBE_EIMS_MAILBOX;
1910 		break;
1911 	default:
1912 		break;
1913 	}
1914 	if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
1915 	    !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
1916 		mask |= IXGBE_EIMS_FLOW_DIR;
1917 
1918 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1919 	if (queues)
1920 		ixgbe_irq_enable_queues(adapter, ~0);
1921 	if (flush)
1922 		IXGBE_WRITE_FLUSH(&adapter->hw);
1923 }
1924 
1925 static irqreturn_t ixgbe_msix_other(int irq, void *data)
1926 {
1927 	struct ixgbe_adapter *adapter = data;
1928 	struct ixgbe_hw *hw = &adapter->hw;
1929 	u32 eicr;
1930 
1931 	/*
1932 	 * Workaround for Silicon errata.  Use clear-by-write instead
1933 	 * of clear-by-read.  Reading with EICS will return the
1934 	 * interrupt causes without clearing, which later be done
1935 	 * with the write to EICR.
1936 	 */
1937 	eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1938 	IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1939 
1940 	if (eicr & IXGBE_EICR_LSC)
1941 		ixgbe_check_lsc(adapter);
1942 
1943 	if (eicr & IXGBE_EICR_MAILBOX)
1944 		ixgbe_msg_task(adapter);
1945 
1946 	switch (hw->mac.type) {
1947 	case ixgbe_mac_82599EB:
1948 	case ixgbe_mac_X540:
1949 		if (eicr & IXGBE_EICR_ECC)
1950 			e_info(link, "Received unrecoverable ECC Err, please "
1951 			       "reboot\n");
1952 		/* Handle Flow Director Full threshold interrupt */
1953 		if (eicr & IXGBE_EICR_FLOW_DIR) {
1954 			int reinit_count = 0;
1955 			int i;
1956 			for (i = 0; i < adapter->num_tx_queues; i++) {
1957 				struct ixgbe_ring *ring = adapter->tx_ring[i];
1958 				if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
1959 						       &ring->state))
1960 					reinit_count++;
1961 			}
1962 			if (reinit_count) {
1963 				/* no more flow director interrupts until after init */
1964 				IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
1965 				adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
1966 				ixgbe_service_event_schedule(adapter);
1967 			}
1968 		}
1969 		ixgbe_check_sfp_event(adapter, eicr);
1970 		ixgbe_check_overtemp_event(adapter, eicr);
1971 		break;
1972 	default:
1973 		break;
1974 	}
1975 
1976 	ixgbe_check_fan_failure(adapter, eicr);
1977 
1978 	/* re-enable the original interrupt state, no lsc, no queues */
1979 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
1980 		ixgbe_irq_enable(adapter, false, false);
1981 
1982 	return IRQ_HANDLED;
1983 }
1984 
1985 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
1986 {
1987 	struct ixgbe_q_vector *q_vector = data;
1988 
1989 	/* EIAM disabled interrupts (on this vector) for us */
1990 
1991 	if (q_vector->rx.ring || q_vector->tx.ring)
1992 		napi_schedule(&q_vector->napi);
1993 
1994 	return IRQ_HANDLED;
1995 }
1996 
1997 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
1998 				     int r_idx)
1999 {
2000 	struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2001 	struct ixgbe_ring *rx_ring = a->rx_ring[r_idx];
2002 
2003 	rx_ring->q_vector = q_vector;
2004 	rx_ring->next = q_vector->rx.ring;
2005 	q_vector->rx.ring = rx_ring;
2006 	q_vector->rx.count++;
2007 }
2008 
2009 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2010 				     int t_idx)
2011 {
2012 	struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2013 	struct ixgbe_ring *tx_ring = a->tx_ring[t_idx];
2014 
2015 	tx_ring->q_vector = q_vector;
2016 	tx_ring->next = q_vector->tx.ring;
2017 	q_vector->tx.ring = tx_ring;
2018 	q_vector->tx.count++;
2019 	q_vector->tx.work_limit = a->tx_work_limit;
2020 }
2021 
2022 /**
2023  * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2024  * @adapter: board private structure to initialize
2025  *
2026  * This function maps descriptor rings to the queue-specific vectors
2027  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
2028  * one vector per ring/queue, but on a constrained vector budget, we
2029  * group the rings as "efficiently" as possible.  You would add new
2030  * mapping configurations in here.
2031  **/
2032 static void ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter)
2033 {
2034 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2035 	int rxr_remaining = adapter->num_rx_queues, rxr_idx = 0;
2036 	int txr_remaining = adapter->num_tx_queues, txr_idx = 0;
2037 	int v_start = 0;
2038 
2039 	/* only one q_vector if MSI-X is disabled. */
2040 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2041 		q_vectors = 1;
2042 
2043 	/*
2044 	 * If we don't have enough vectors for a 1-to-1 mapping, we'll have to
2045 	 * group them so there are multiple queues per vector.
2046 	 *
2047 	 * Re-adjusting *qpv takes care of the remainder.
2048 	 */
2049 	for (; v_start < q_vectors && rxr_remaining; v_start++) {
2050 		int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_start);
2051 		for (; rqpv; rqpv--, rxr_idx++, rxr_remaining--)
2052 			map_vector_to_rxq(adapter, v_start, rxr_idx);
2053 	}
2054 
2055 	/*
2056 	 * If there are not enough q_vectors for each ring to have it's own
2057 	 * vector then we must pair up Rx/Tx on a each vector
2058 	 */
2059 	if ((v_start + txr_remaining) > q_vectors)
2060 		v_start = 0;
2061 
2062 	for (; v_start < q_vectors && txr_remaining; v_start++) {
2063 		int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_start);
2064 		for (; tqpv; tqpv--, txr_idx++, txr_remaining--)
2065 			map_vector_to_txq(adapter, v_start, txr_idx);
2066 	}
2067 }
2068 
2069 /**
2070  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2071  * @adapter: board private structure
2072  *
2073  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2074  * interrupts from the kernel.
2075  **/
2076 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2077 {
2078 	struct net_device *netdev = adapter->netdev;
2079 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2080 	int vector, err;
2081 	int ri = 0, ti = 0;
2082 
2083 	for (vector = 0; vector < q_vectors; vector++) {
2084 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2085 		struct msix_entry *entry = &adapter->msix_entries[vector];
2086 
2087 		if (q_vector->tx.ring && q_vector->rx.ring) {
2088 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2089 				 "%s-%s-%d", netdev->name, "TxRx", ri++);
2090 			ti++;
2091 		} else if (q_vector->rx.ring) {
2092 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2093 				 "%s-%s-%d", netdev->name, "rx", ri++);
2094 		} else if (q_vector->tx.ring) {
2095 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2096 				 "%s-%s-%d", netdev->name, "tx", ti++);
2097 		} else {
2098 			/* skip this unused q_vector */
2099 			continue;
2100 		}
2101 		err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2102 				  q_vector->name, q_vector);
2103 		if (err) {
2104 			e_err(probe, "request_irq failed for MSIX interrupt "
2105 			      "Error: %d\n", err);
2106 			goto free_queue_irqs;
2107 		}
2108 		/* If Flow Director is enabled, set interrupt affinity */
2109 		if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2110 			/* assign the mask for this irq */
2111 			irq_set_affinity_hint(entry->vector,
2112 					      q_vector->affinity_mask);
2113 		}
2114 	}
2115 
2116 	err = request_irq(adapter->msix_entries[vector].vector,
2117 			  ixgbe_msix_other, 0, netdev->name, adapter);
2118 	if (err) {
2119 		e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2120 		goto free_queue_irqs;
2121 	}
2122 
2123 	return 0;
2124 
2125 free_queue_irqs:
2126 	while (vector) {
2127 		vector--;
2128 		irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2129 				      NULL);
2130 		free_irq(adapter->msix_entries[vector].vector,
2131 			 adapter->q_vector[vector]);
2132 	}
2133 	adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2134 	pci_disable_msix(adapter->pdev);
2135 	kfree(adapter->msix_entries);
2136 	adapter->msix_entries = NULL;
2137 	return err;
2138 }
2139 
2140 /**
2141  * ixgbe_intr - legacy mode Interrupt Handler
2142  * @irq: interrupt number
2143  * @data: pointer to a network interface device structure
2144  **/
2145 static irqreturn_t ixgbe_intr(int irq, void *data)
2146 {
2147 	struct ixgbe_adapter *adapter = data;
2148 	struct ixgbe_hw *hw = &adapter->hw;
2149 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2150 	u32 eicr;
2151 
2152 	/*
2153 	 * Workaround for silicon errata on 82598.  Mask the interrupts
2154 	 * before the read of EICR.
2155 	 */
2156 	IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2157 
2158 	/* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2159 	 * therefore no explict interrupt disable is necessary */
2160 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2161 	if (!eicr) {
2162 		/*
2163 		 * shared interrupt alert!
2164 		 * make sure interrupts are enabled because the read will
2165 		 * have disabled interrupts due to EIAM
2166 		 * finish the workaround of silicon errata on 82598.  Unmask
2167 		 * the interrupt that we masked before the EICR read.
2168 		 */
2169 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
2170 			ixgbe_irq_enable(adapter, true, true);
2171 		return IRQ_NONE;	/* Not our interrupt */
2172 	}
2173 
2174 	if (eicr & IXGBE_EICR_LSC)
2175 		ixgbe_check_lsc(adapter);
2176 
2177 	switch (hw->mac.type) {
2178 	case ixgbe_mac_82599EB:
2179 		ixgbe_check_sfp_event(adapter, eicr);
2180 		/* Fall through */
2181 	case ixgbe_mac_X540:
2182 		if (eicr & IXGBE_EICR_ECC)
2183 			e_info(link, "Received unrecoverable ECC err, please "
2184 				     "reboot\n");
2185 		ixgbe_check_overtemp_event(adapter, eicr);
2186 		break;
2187 	default:
2188 		break;
2189 	}
2190 
2191 	ixgbe_check_fan_failure(adapter, eicr);
2192 
2193 	if (napi_schedule_prep(&(q_vector->napi))) {
2194 		/* would disable interrupts here but EIAM disabled it */
2195 		__napi_schedule(&(q_vector->napi));
2196 	}
2197 
2198 	/*
2199 	 * re-enable link(maybe) and non-queue interrupts, no flush.
2200 	 * ixgbe_poll will re-enable the queue interrupts
2201 	 */
2202 
2203 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
2204 		ixgbe_irq_enable(adapter, false, false);
2205 
2206 	return IRQ_HANDLED;
2207 }
2208 
2209 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2210 {
2211 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2212 	int i;
2213 
2214 	/* legacy and MSI only use one vector */
2215 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2216 		q_vectors = 1;
2217 
2218 	for (i = 0; i < adapter->num_rx_queues; i++) {
2219 		adapter->rx_ring[i]->q_vector = NULL;
2220 		adapter->rx_ring[i]->next = NULL;
2221 	}
2222 	for (i = 0; i < adapter->num_tx_queues; i++) {
2223 		adapter->tx_ring[i]->q_vector = NULL;
2224 		adapter->tx_ring[i]->next = NULL;
2225 	}
2226 
2227 	for (i = 0; i < q_vectors; i++) {
2228 		struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2229 		memset(&q_vector->rx, 0, sizeof(struct ixgbe_ring_container));
2230 		memset(&q_vector->tx, 0, sizeof(struct ixgbe_ring_container));
2231 	}
2232 }
2233 
2234 /**
2235  * ixgbe_request_irq - initialize interrupts
2236  * @adapter: board private structure
2237  *
2238  * Attempts to configure interrupts using the best available
2239  * capabilities of the hardware and kernel.
2240  **/
2241 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2242 {
2243 	struct net_device *netdev = adapter->netdev;
2244 	int err;
2245 
2246 	/* map all of the rings to the q_vectors */
2247 	ixgbe_map_rings_to_vectors(adapter);
2248 
2249 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2250 		err = ixgbe_request_msix_irqs(adapter);
2251 	else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2252 		err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2253 				  netdev->name, adapter);
2254 	else
2255 		err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2256 				  netdev->name, adapter);
2257 
2258 	if (err) {
2259 		e_err(probe, "request_irq failed, Error %d\n", err);
2260 
2261 		/* place q_vectors and rings back into a known good state */
2262 		ixgbe_reset_q_vectors(adapter);
2263 	}
2264 
2265 	return err;
2266 }
2267 
2268 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2269 {
2270 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2271 		int i, q_vectors;
2272 
2273 		q_vectors = adapter->num_msix_vectors;
2274 		i = q_vectors - 1;
2275 		free_irq(adapter->msix_entries[i].vector, adapter);
2276 		i--;
2277 
2278 		for (; i >= 0; i--) {
2279 			/* free only the irqs that were actually requested */
2280 			if (!adapter->q_vector[i]->rx.ring &&
2281 			    !adapter->q_vector[i]->tx.ring)
2282 				continue;
2283 
2284 			/* clear the affinity_mask in the IRQ descriptor */
2285 			irq_set_affinity_hint(adapter->msix_entries[i].vector,
2286 					      NULL);
2287 
2288 			free_irq(adapter->msix_entries[i].vector,
2289 				 adapter->q_vector[i]);
2290 		}
2291 	} else {
2292 		free_irq(adapter->pdev->irq, adapter);
2293 	}
2294 
2295 	/* clear q_vector state information */
2296 	ixgbe_reset_q_vectors(adapter);
2297 }
2298 
2299 /**
2300  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2301  * @adapter: board private structure
2302  **/
2303 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2304 {
2305 	switch (adapter->hw.mac.type) {
2306 	case ixgbe_mac_82598EB:
2307 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2308 		break;
2309 	case ixgbe_mac_82599EB:
2310 	case ixgbe_mac_X540:
2311 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2312 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2313 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2314 		break;
2315 	default:
2316 		break;
2317 	}
2318 	IXGBE_WRITE_FLUSH(&adapter->hw);
2319 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2320 		int i;
2321 		for (i = 0; i < adapter->num_msix_vectors; i++)
2322 			synchronize_irq(adapter->msix_entries[i].vector);
2323 	} else {
2324 		synchronize_irq(adapter->pdev->irq);
2325 	}
2326 }
2327 
2328 /**
2329  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2330  *
2331  **/
2332 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2333 {
2334 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2335 
2336 	/* rx/tx vector */
2337 	if (adapter->rx_itr_setting == 1)
2338 		q_vector->itr = IXGBE_20K_ITR;
2339 	else
2340 		q_vector->itr = adapter->rx_itr_setting;
2341 
2342 	ixgbe_write_eitr(q_vector);
2343 
2344 	ixgbe_set_ivar(adapter, 0, 0, 0);
2345 	ixgbe_set_ivar(adapter, 1, 0, 0);
2346 
2347 	e_info(hw, "Legacy interrupt IVAR setup done\n");
2348 }
2349 
2350 /**
2351  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2352  * @adapter: board private structure
2353  * @ring: structure containing ring specific data
2354  *
2355  * Configure the Tx descriptor ring after a reset.
2356  **/
2357 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2358 			     struct ixgbe_ring *ring)
2359 {
2360 	struct ixgbe_hw *hw = &adapter->hw;
2361 	u64 tdba = ring->dma;
2362 	int wait_loop = 10;
2363 	u32 txdctl = IXGBE_TXDCTL_ENABLE;
2364 	u8 reg_idx = ring->reg_idx;
2365 
2366 	/* disable queue to avoid issues while updating state */
2367 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
2368 	IXGBE_WRITE_FLUSH(hw);
2369 
2370 	IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2371 			(tdba & DMA_BIT_MASK(32)));
2372 	IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2373 	IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2374 			ring->count * sizeof(union ixgbe_adv_tx_desc));
2375 	IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2376 	IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2377 	ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2378 
2379 	/*
2380 	 * set WTHRESH to encourage burst writeback, it should not be set
2381 	 * higher than 1 when ITR is 0 as it could cause false TX hangs
2382 	 *
2383 	 * In order to avoid issues WTHRESH + PTHRESH should always be equal
2384 	 * to or less than the number of on chip descriptors, which is
2385 	 * currently 40.
2386 	 */
2387 	if (!adapter->tx_itr_setting || !adapter->rx_itr_setting)
2388 		txdctl |= (1 << 16);	/* WTHRESH = 1 */
2389 	else
2390 		txdctl |= (8 << 16);	/* WTHRESH = 8 */
2391 
2392 	/* PTHRESH=32 is needed to avoid a Tx hang with DFP enabled. */
2393 	txdctl |= (1 << 8) |	/* HTHRESH = 1 */
2394 		   32;		/* PTHRESH = 32 */
2395 
2396 	/* reinitialize flowdirector state */
2397 	if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2398 	    adapter->atr_sample_rate) {
2399 		ring->atr_sample_rate = adapter->atr_sample_rate;
2400 		ring->atr_count = 0;
2401 		set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2402 	} else {
2403 		ring->atr_sample_rate = 0;
2404 	}
2405 
2406 	clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2407 
2408 	/* enable queue */
2409 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2410 
2411 	/* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2412 	if (hw->mac.type == ixgbe_mac_82598EB &&
2413 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2414 		return;
2415 
2416 	/* poll to verify queue is enabled */
2417 	do {
2418 		usleep_range(1000, 2000);
2419 		txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2420 	} while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2421 	if (!wait_loop)
2422 		e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2423 }
2424 
2425 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2426 {
2427 	struct ixgbe_hw *hw = &adapter->hw;
2428 	u32 rttdcs;
2429 	u32 reg;
2430 	u8 tcs = netdev_get_num_tc(adapter->netdev);
2431 
2432 	if (hw->mac.type == ixgbe_mac_82598EB)
2433 		return;
2434 
2435 	/* disable the arbiter while setting MTQC */
2436 	rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2437 	rttdcs |= IXGBE_RTTDCS_ARBDIS;
2438 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2439 
2440 	/* set transmit pool layout */
2441 	switch (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2442 	case (IXGBE_FLAG_SRIOV_ENABLED):
2443 		IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2444 				(IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2445 		break;
2446 	default:
2447 		if (!tcs)
2448 			reg = IXGBE_MTQC_64Q_1PB;
2449 		else if (tcs <= 4)
2450 			reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2451 		else
2452 			reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2453 
2454 		IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
2455 
2456 		/* Enable Security TX Buffer IFG for multiple pb */
2457 		if (tcs) {
2458 			reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
2459 			reg |= IXGBE_SECTX_DCB;
2460 			IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
2461 		}
2462 		break;
2463 	}
2464 
2465 	/* re-enable the arbiter */
2466 	rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2467 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2468 }
2469 
2470 /**
2471  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2472  * @adapter: board private structure
2473  *
2474  * Configure the Tx unit of the MAC after a reset.
2475  **/
2476 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2477 {
2478 	struct ixgbe_hw *hw = &adapter->hw;
2479 	u32 dmatxctl;
2480 	u32 i;
2481 
2482 	ixgbe_setup_mtqc(adapter);
2483 
2484 	if (hw->mac.type != ixgbe_mac_82598EB) {
2485 		/* DMATXCTL.EN must be before Tx queues are enabled */
2486 		dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2487 		dmatxctl |= IXGBE_DMATXCTL_TE;
2488 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2489 	}
2490 
2491 	/* Setup the HW Tx Head and Tail descriptor pointers */
2492 	for (i = 0; i < adapter->num_tx_queues; i++)
2493 		ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2494 }
2495 
2496 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2497 
2498 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2499 				   struct ixgbe_ring *rx_ring)
2500 {
2501 	u32 srrctl;
2502 	u8 reg_idx = rx_ring->reg_idx;
2503 
2504 	switch (adapter->hw.mac.type) {
2505 	case ixgbe_mac_82598EB: {
2506 		struct ixgbe_ring_feature *feature = adapter->ring_feature;
2507 		const int mask = feature[RING_F_RSS].mask;
2508 		reg_idx = reg_idx & mask;
2509 	}
2510 		break;
2511 	case ixgbe_mac_82599EB:
2512 	case ixgbe_mac_X540:
2513 	default:
2514 		break;
2515 	}
2516 
2517 	srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx));
2518 
2519 	srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2520 	srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2521 	if (adapter->num_vfs)
2522 		srrctl |= IXGBE_SRRCTL_DROP_EN;
2523 
2524 	srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2525 		  IXGBE_SRRCTL_BSIZEHDR_MASK;
2526 
2527 	if (ring_is_ps_enabled(rx_ring)) {
2528 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2529 		srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2530 #else
2531 		srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2532 #endif
2533 		srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2534 	} else {
2535 		srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2536 			  IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2537 		srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2538 	}
2539 
2540 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx), srrctl);
2541 }
2542 
2543 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2544 {
2545 	struct ixgbe_hw *hw = &adapter->hw;
2546 	static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2547 			  0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2548 			  0x6A3E67EA, 0x14364D17, 0x3BED200D};
2549 	u32 mrqc = 0, reta = 0;
2550 	u32 rxcsum;
2551 	int i, j;
2552 	u8 tcs = netdev_get_num_tc(adapter->netdev);
2553 	int maxq = adapter->ring_feature[RING_F_RSS].indices;
2554 
2555 	if (tcs)
2556 		maxq = min(maxq, adapter->num_tx_queues / tcs);
2557 
2558 	/* Fill out hash function seeds */
2559 	for (i = 0; i < 10; i++)
2560 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2561 
2562 	/* Fill out redirection table */
2563 	for (i = 0, j = 0; i < 128; i++, j++) {
2564 		if (j == maxq)
2565 			j = 0;
2566 		/* reta = 4-byte sliding window of
2567 		 * 0x00..(indices-1)(indices-1)00..etc. */
2568 		reta = (reta << 8) | (j * 0x11);
2569 		if ((i & 3) == 3)
2570 			IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2571 	}
2572 
2573 	/* Disable indicating checksum in descriptor, enables RSS hash */
2574 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2575 	rxcsum |= IXGBE_RXCSUM_PCSD;
2576 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2577 
2578 	if (adapter->hw.mac.type == ixgbe_mac_82598EB &&
2579 	    (adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
2580 		mrqc = IXGBE_MRQC_RSSEN;
2581 	} else {
2582 		int mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2583 					     | IXGBE_FLAG_SRIOV_ENABLED);
2584 
2585 		switch (mask) {
2586 		case (IXGBE_FLAG_RSS_ENABLED):
2587 			if (!tcs)
2588 				mrqc = IXGBE_MRQC_RSSEN;
2589 			else if (tcs <= 4)
2590 				mrqc = IXGBE_MRQC_RTRSS4TCEN;
2591 			else
2592 				mrqc = IXGBE_MRQC_RTRSS8TCEN;
2593 			break;
2594 		case (IXGBE_FLAG_SRIOV_ENABLED):
2595 			mrqc = IXGBE_MRQC_VMDQEN;
2596 			break;
2597 		default:
2598 			break;
2599 		}
2600 	}
2601 
2602 	/* Perform hash on these packet types */
2603 	mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2604 	      | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2605 	      | IXGBE_MRQC_RSS_FIELD_IPV6
2606 	      | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2607 
2608 	IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2609 }
2610 
2611 /**
2612  * ixgbe_configure_rscctl - enable RSC for the indicated ring
2613  * @adapter:    address of board private structure
2614  * @index:      index of ring to set
2615  **/
2616 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2617 				   struct ixgbe_ring *ring)
2618 {
2619 	struct ixgbe_hw *hw = &adapter->hw;
2620 	u32 rscctrl;
2621 	int rx_buf_len;
2622 	u8 reg_idx = ring->reg_idx;
2623 
2624 	if (!ring_is_rsc_enabled(ring))
2625 		return;
2626 
2627 	rx_buf_len = ring->rx_buf_len;
2628 	rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2629 	rscctrl |= IXGBE_RSCCTL_RSCEN;
2630 	/*
2631 	 * we must limit the number of descriptors so that the
2632 	 * total size of max desc * buf_len is not greater
2633 	 * than 65535
2634 	 */
2635 	if (ring_is_ps_enabled(ring)) {
2636 #if (MAX_SKB_FRAGS > 16)
2637 		rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2638 #elif (MAX_SKB_FRAGS > 8)
2639 		rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2640 #elif (MAX_SKB_FRAGS > 4)
2641 		rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2642 #else
2643 		rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2644 #endif
2645 	} else {
2646 		if (rx_buf_len < IXGBE_RXBUFFER_4K)
2647 			rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2648 		else if (rx_buf_len < IXGBE_RXBUFFER_8K)
2649 			rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2650 		else
2651 			rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2652 	}
2653 	IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2654 }
2655 
2656 /**
2657  *  ixgbe_set_uta - Set unicast filter table address
2658  *  @adapter: board private structure
2659  *
2660  *  The unicast table address is a register array of 32-bit registers.
2661  *  The table is meant to be used in a way similar to how the MTA is used
2662  *  however due to certain limitations in the hardware it is necessary to
2663  *  set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2664  *  enable bit to allow vlan tag stripping when promiscuous mode is enabled
2665  **/
2666 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2667 {
2668 	struct ixgbe_hw *hw = &adapter->hw;
2669 	int i;
2670 
2671 	/* The UTA table only exists on 82599 hardware and newer */
2672 	if (hw->mac.type < ixgbe_mac_82599EB)
2673 		return;
2674 
2675 	/* we only need to do this if VMDq is enabled */
2676 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2677 		return;
2678 
2679 	for (i = 0; i < 128; i++)
2680 		IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
2681 }
2682 
2683 #define IXGBE_MAX_RX_DESC_POLL 10
2684 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2685 				       struct ixgbe_ring *ring)
2686 {
2687 	struct ixgbe_hw *hw = &adapter->hw;
2688 	int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2689 	u32 rxdctl;
2690 	u8 reg_idx = ring->reg_idx;
2691 
2692 	/* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2693 	if (hw->mac.type == ixgbe_mac_82598EB &&
2694 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2695 		return;
2696 
2697 	do {
2698 		usleep_range(1000, 2000);
2699 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2700 	} while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
2701 
2702 	if (!wait_loop) {
2703 		e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
2704 		      "the polling period\n", reg_idx);
2705 	}
2706 }
2707 
2708 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
2709 			    struct ixgbe_ring *ring)
2710 {
2711 	struct ixgbe_hw *hw = &adapter->hw;
2712 	int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2713 	u32 rxdctl;
2714 	u8 reg_idx = ring->reg_idx;
2715 
2716 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2717 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
2718 
2719 	/* write value back with RXDCTL.ENABLE bit cleared */
2720 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2721 
2722 	if (hw->mac.type == ixgbe_mac_82598EB &&
2723 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2724 		return;
2725 
2726 	/* the hardware may take up to 100us to really disable the rx queue */
2727 	do {
2728 		udelay(10);
2729 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2730 	} while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
2731 
2732 	if (!wait_loop) {
2733 		e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
2734 		      "the polling period\n", reg_idx);
2735 	}
2736 }
2737 
2738 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
2739 			     struct ixgbe_ring *ring)
2740 {
2741 	struct ixgbe_hw *hw = &adapter->hw;
2742 	u64 rdba = ring->dma;
2743 	u32 rxdctl;
2744 	u8 reg_idx = ring->reg_idx;
2745 
2746 	/* disable queue to avoid issues while updating state */
2747 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2748 	ixgbe_disable_rx_queue(adapter, ring);
2749 
2750 	IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
2751 	IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
2752 	IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
2753 			ring->count * sizeof(union ixgbe_adv_rx_desc));
2754 	IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
2755 	IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
2756 	ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
2757 
2758 	ixgbe_configure_srrctl(adapter, ring);
2759 	ixgbe_configure_rscctl(adapter, ring);
2760 
2761 	/* If operating in IOV mode set RLPML for X540 */
2762 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
2763 	    hw->mac.type == ixgbe_mac_X540) {
2764 		rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
2765 		rxdctl |= ((ring->netdev->mtu + ETH_HLEN +
2766 			    ETH_FCS_LEN + VLAN_HLEN) | IXGBE_RXDCTL_RLPML_EN);
2767 	}
2768 
2769 	if (hw->mac.type == ixgbe_mac_82598EB) {
2770 		/*
2771 		 * enable cache line friendly hardware writes:
2772 		 * PTHRESH=32 descriptors (half the internal cache),
2773 		 * this also removes ugly rx_no_buffer_count increment
2774 		 * HTHRESH=4 descriptors (to minimize latency on fetch)
2775 		 * WTHRESH=8 burst writeback up to two cache lines
2776 		 */
2777 		rxdctl &= ~0x3FFFFF;
2778 		rxdctl |=  0x080420;
2779 	}
2780 
2781 	/* enable receive descriptor ring */
2782 	rxdctl |= IXGBE_RXDCTL_ENABLE;
2783 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2784 
2785 	ixgbe_rx_desc_queue_enable(adapter, ring);
2786 	ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
2787 }
2788 
2789 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
2790 {
2791 	struct ixgbe_hw *hw = &adapter->hw;
2792 	int p;
2793 
2794 	/* PSRTYPE must be initialized in non 82598 adapters */
2795 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2796 		      IXGBE_PSRTYPE_UDPHDR |
2797 		      IXGBE_PSRTYPE_IPV4HDR |
2798 		      IXGBE_PSRTYPE_L2HDR |
2799 		      IXGBE_PSRTYPE_IPV6HDR;
2800 
2801 	if (hw->mac.type == ixgbe_mac_82598EB)
2802 		return;
2803 
2804 	if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
2805 		psrtype |= (adapter->num_rx_queues_per_pool << 29);
2806 
2807 	for (p = 0; p < adapter->num_rx_pools; p++)
2808 		IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
2809 				psrtype);
2810 }
2811 
2812 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
2813 {
2814 	struct ixgbe_hw *hw = &adapter->hw;
2815 	u32 gcr_ext;
2816 	u32 vt_reg_bits;
2817 	u32 reg_offset, vf_shift;
2818 	u32 vmdctl;
2819 	int i;
2820 
2821 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2822 		return;
2823 
2824 	vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2825 	vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
2826 	vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
2827 	IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2828 
2829 	vf_shift = adapter->num_vfs % 32;
2830 	reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
2831 
2832 	/* Enable only the PF's pool for Tx/Rx */
2833 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2834 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
2835 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2836 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
2837 	IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2838 
2839 	/* Map PF MAC address in RAR Entry 0 to first pool following VFs */
2840 	hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2841 
2842 	/*
2843 	 * Set up VF register offsets for selected VT Mode,
2844 	 * i.e. 32 or 64 VFs for SR-IOV
2845 	 */
2846 	gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2847 	gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
2848 	gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
2849 	IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
2850 
2851 	/* enable Tx loopback for VF/PF communication */
2852 	IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2853 	/* Enable MAC Anti-Spoofing */
2854 	hw->mac.ops.set_mac_anti_spoofing(hw,
2855 					   (adapter->num_vfs != 0),
2856 					  adapter->num_vfs);
2857 	/* For VFs that have spoof checking turned off */
2858 	for (i = 0; i < adapter->num_vfs; i++) {
2859 		if (!adapter->vfinfo[i].spoofchk_enabled)
2860 			ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
2861 	}
2862 }
2863 
2864 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
2865 {
2866 	struct ixgbe_hw *hw = &adapter->hw;
2867 	struct net_device *netdev = adapter->netdev;
2868 	int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2869 	int rx_buf_len;
2870 	struct ixgbe_ring *rx_ring;
2871 	int i;
2872 	u32 mhadd, hlreg0;
2873 
2874 	/* Decide whether to use packet split mode or not */
2875 	/* On by default */
2876 	adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2877 
2878 	/* Do not use packet split if we're in SR-IOV Mode */
2879 	if (adapter->num_vfs)
2880 		adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
2881 
2882 	/* Disable packet split due to 82599 erratum #45 */
2883 	if (hw->mac.type == ixgbe_mac_82599EB)
2884 		adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
2885 
2886 #ifdef IXGBE_FCOE
2887 	/* adjust max frame to be able to do baby jumbo for FCoE */
2888 	if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
2889 	    (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2890 		max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2891 
2892 #endif /* IXGBE_FCOE */
2893 	mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2894 	if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2895 		mhadd &= ~IXGBE_MHADD_MFS_MASK;
2896 		mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2897 
2898 		IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2899 	}
2900 
2901 	/* MHADD will allow an extra 4 bytes past for vlan tagged frames */
2902 	max_frame += VLAN_HLEN;
2903 
2904 	/* Set the RX buffer length according to the mode */
2905 	if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2906 		rx_buf_len = IXGBE_RX_HDR_SIZE;
2907 	} else {
2908 		if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2909 		    (netdev->mtu <= ETH_DATA_LEN))
2910 			rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2911 		/*
2912 		 * Make best use of allocation by using all but 1K of a
2913 		 * power of 2 allocation that will be used for skb->head.
2914 		 */
2915 		else if (max_frame <= IXGBE_RXBUFFER_3K)
2916 			rx_buf_len = IXGBE_RXBUFFER_3K;
2917 		else if (max_frame <= IXGBE_RXBUFFER_7K)
2918 			rx_buf_len = IXGBE_RXBUFFER_7K;
2919 		else if (max_frame <= IXGBE_RXBUFFER_15K)
2920 			rx_buf_len = IXGBE_RXBUFFER_15K;
2921 		else
2922 			rx_buf_len = IXGBE_MAX_RXBUFFER;
2923 	}
2924 
2925 	hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2926 	/* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
2927 	hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2928 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2929 
2930 	/*
2931 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
2932 	 * the Base and Length of the Rx Descriptor Ring
2933 	 */
2934 	for (i = 0; i < adapter->num_rx_queues; i++) {
2935 		rx_ring = adapter->rx_ring[i];
2936 		rx_ring->rx_buf_len = rx_buf_len;
2937 
2938 		if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2939 			set_ring_ps_enabled(rx_ring);
2940 		else
2941 			clear_ring_ps_enabled(rx_ring);
2942 
2943 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
2944 			set_ring_rsc_enabled(rx_ring);
2945 		else
2946 			clear_ring_rsc_enabled(rx_ring);
2947 
2948 #ifdef IXGBE_FCOE
2949 		if (netdev->features & NETIF_F_FCOE_MTU) {
2950 			struct ixgbe_ring_feature *f;
2951 			f = &adapter->ring_feature[RING_F_FCOE];
2952 			if ((i >= f->mask) && (i < f->mask + f->indices)) {
2953 				clear_ring_ps_enabled(rx_ring);
2954 				if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2955 					rx_ring->rx_buf_len =
2956 						IXGBE_FCOE_JUMBO_FRAME_SIZE;
2957 			} else if (!ring_is_rsc_enabled(rx_ring) &&
2958 				   !ring_is_ps_enabled(rx_ring)) {
2959 				rx_ring->rx_buf_len =
2960 						IXGBE_FCOE_JUMBO_FRAME_SIZE;
2961 			}
2962 		}
2963 #endif /* IXGBE_FCOE */
2964 	}
2965 }
2966 
2967 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
2968 {
2969 	struct ixgbe_hw *hw = &adapter->hw;
2970 	u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2971 
2972 	switch (hw->mac.type) {
2973 	case ixgbe_mac_82598EB:
2974 		/*
2975 		 * For VMDq support of different descriptor types or
2976 		 * buffer sizes through the use of multiple SRRCTL
2977 		 * registers, RDRXCTL.MVMEN must be set to 1
2978 		 *
2979 		 * also, the manual doesn't mention it clearly but DCA hints
2980 		 * will only use queue 0's tags unless this bit is set.  Side
2981 		 * effects of setting this bit are only that SRRCTL must be
2982 		 * fully programmed [0..15]
2983 		 */
2984 		rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2985 		break;
2986 	case ixgbe_mac_82599EB:
2987 	case ixgbe_mac_X540:
2988 		/* Disable RSC for ACK packets */
2989 		IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2990 		   (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2991 		rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2992 		/* hardware requires some bits to be set by default */
2993 		rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
2994 		rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
2995 		break;
2996 	default:
2997 		/* We should do nothing since we don't know this hardware */
2998 		return;
2999 	}
3000 
3001 	IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3002 }
3003 
3004 /**
3005  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3006  * @adapter: board private structure
3007  *
3008  * Configure the Rx unit of the MAC after a reset.
3009  **/
3010 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3011 {
3012 	struct ixgbe_hw *hw = &adapter->hw;
3013 	int i;
3014 	u32 rxctrl;
3015 
3016 	/* disable receives while setting up the descriptors */
3017 	rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3018 	IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3019 
3020 	ixgbe_setup_psrtype(adapter);
3021 	ixgbe_setup_rdrxctl(adapter);
3022 
3023 	/* Program registers for the distribution of queues */
3024 	ixgbe_setup_mrqc(adapter);
3025 
3026 	ixgbe_set_uta(adapter);
3027 
3028 	/* set_rx_buffer_len must be called before ring initialization */
3029 	ixgbe_set_rx_buffer_len(adapter);
3030 
3031 	/*
3032 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
3033 	 * the Base and Length of the Rx Descriptor Ring
3034 	 */
3035 	for (i = 0; i < adapter->num_rx_queues; i++)
3036 		ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3037 
3038 	/* disable drop enable for 82598 parts */
3039 	if (hw->mac.type == ixgbe_mac_82598EB)
3040 		rxctrl |= IXGBE_RXCTRL_DMBYPS;
3041 
3042 	/* enable all receives */
3043 	rxctrl |= IXGBE_RXCTRL_RXEN;
3044 	hw->mac.ops.enable_rx_dma(hw, rxctrl);
3045 }
3046 
3047 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3048 {
3049 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3050 	struct ixgbe_hw *hw = &adapter->hw;
3051 	int pool_ndx = adapter->num_vfs;
3052 
3053 	/* add VID to filter table */
3054 	hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3055 	set_bit(vid, adapter->active_vlans);
3056 }
3057 
3058 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3059 {
3060 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3061 	struct ixgbe_hw *hw = &adapter->hw;
3062 	int pool_ndx = adapter->num_vfs;
3063 
3064 	/* remove VID from filter table */
3065 	hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3066 	clear_bit(vid, adapter->active_vlans);
3067 }
3068 
3069 /**
3070  * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3071  * @adapter: driver data
3072  */
3073 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3074 {
3075 	struct ixgbe_hw *hw = &adapter->hw;
3076 	u32 vlnctrl;
3077 
3078 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3079 	vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3080 	IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3081 }
3082 
3083 /**
3084  * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3085  * @adapter: driver data
3086  */
3087 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3088 {
3089 	struct ixgbe_hw *hw = &adapter->hw;
3090 	u32 vlnctrl;
3091 
3092 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3093 	vlnctrl |= IXGBE_VLNCTRL_VFE;
3094 	vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3095 	IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3096 }
3097 
3098 /**
3099  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3100  * @adapter: driver data
3101  */
3102 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3103 {
3104 	struct ixgbe_hw *hw = &adapter->hw;
3105 	u32 vlnctrl;
3106 	int i, j;
3107 
3108 	switch (hw->mac.type) {
3109 	case ixgbe_mac_82598EB:
3110 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3111 		vlnctrl &= ~IXGBE_VLNCTRL_VME;
3112 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3113 		break;
3114 	case ixgbe_mac_82599EB:
3115 	case ixgbe_mac_X540:
3116 		for (i = 0; i < adapter->num_rx_queues; i++) {
3117 			j = adapter->rx_ring[i]->reg_idx;
3118 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3119 			vlnctrl &= ~IXGBE_RXDCTL_VME;
3120 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3121 		}
3122 		break;
3123 	default:
3124 		break;
3125 	}
3126 }
3127 
3128 /**
3129  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3130  * @adapter: driver data
3131  */
3132 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3133 {
3134 	struct ixgbe_hw *hw = &adapter->hw;
3135 	u32 vlnctrl;
3136 	int i, j;
3137 
3138 	switch (hw->mac.type) {
3139 	case ixgbe_mac_82598EB:
3140 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3141 		vlnctrl |= IXGBE_VLNCTRL_VME;
3142 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3143 		break;
3144 	case ixgbe_mac_82599EB:
3145 	case ixgbe_mac_X540:
3146 		for (i = 0; i < adapter->num_rx_queues; i++) {
3147 			j = adapter->rx_ring[i]->reg_idx;
3148 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3149 			vlnctrl |= IXGBE_RXDCTL_VME;
3150 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3151 		}
3152 		break;
3153 	default:
3154 		break;
3155 	}
3156 }
3157 
3158 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3159 {
3160 	u16 vid;
3161 
3162 	ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3163 
3164 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3165 		ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3166 }
3167 
3168 /**
3169  * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3170  * @netdev: network interface device structure
3171  *
3172  * Writes unicast address list to the RAR table.
3173  * Returns: -ENOMEM on failure/insufficient address space
3174  *                0 on no addresses written
3175  *                X on writing X addresses to the RAR table
3176  **/
3177 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3178 {
3179 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3180 	struct ixgbe_hw *hw = &adapter->hw;
3181 	unsigned int vfn = adapter->num_vfs;
3182 	unsigned int rar_entries = IXGBE_MAX_PF_MACVLANS;
3183 	int count = 0;
3184 
3185 	/* return ENOMEM indicating insufficient memory for addresses */
3186 	if (netdev_uc_count(netdev) > rar_entries)
3187 		return -ENOMEM;
3188 
3189 	if (!netdev_uc_empty(netdev) && rar_entries) {
3190 		struct netdev_hw_addr *ha;
3191 		/* return error if we do not support writing to RAR table */
3192 		if (!hw->mac.ops.set_rar)
3193 			return -ENOMEM;
3194 
3195 		netdev_for_each_uc_addr(ha, netdev) {
3196 			if (!rar_entries)
3197 				break;
3198 			hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3199 					    vfn, IXGBE_RAH_AV);
3200 			count++;
3201 		}
3202 	}
3203 	/* write the addresses in reverse order to avoid write combining */
3204 	for (; rar_entries > 0 ; rar_entries--)
3205 		hw->mac.ops.clear_rar(hw, rar_entries);
3206 
3207 	return count;
3208 }
3209 
3210 /**
3211  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3212  * @netdev: network interface device structure
3213  *
3214  * The set_rx_method entry point is called whenever the unicast/multicast
3215  * address list or the network interface flags are updated.  This routine is
3216  * responsible for configuring the hardware for proper unicast, multicast and
3217  * promiscuous mode.
3218  **/
3219 void ixgbe_set_rx_mode(struct net_device *netdev)
3220 {
3221 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3222 	struct ixgbe_hw *hw = &adapter->hw;
3223 	u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3224 	int count;
3225 
3226 	/* Check for Promiscuous and All Multicast modes */
3227 
3228 	fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3229 
3230 	/* set all bits that we expect to always be set */
3231 	fctrl |= IXGBE_FCTRL_BAM;
3232 	fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3233 	fctrl |= IXGBE_FCTRL_PMCF;
3234 
3235 	/* clear the bits we are changing the status of */
3236 	fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3237 
3238 	if (netdev->flags & IFF_PROMISC) {
3239 		hw->addr_ctrl.user_set_promisc = true;
3240 		fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3241 		vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3242 		/* don't hardware filter vlans in promisc mode */
3243 		ixgbe_vlan_filter_disable(adapter);
3244 	} else {
3245 		if (netdev->flags & IFF_ALLMULTI) {
3246 			fctrl |= IXGBE_FCTRL_MPE;
3247 			vmolr |= IXGBE_VMOLR_MPE;
3248 		} else {
3249 			/*
3250 			 * Write addresses to the MTA, if the attempt fails
3251 			 * then we should just turn on promiscuous mode so
3252 			 * that we can at least receive multicast traffic
3253 			 */
3254 			hw->mac.ops.update_mc_addr_list(hw, netdev);
3255 			vmolr |= IXGBE_VMOLR_ROMPE;
3256 		}
3257 		ixgbe_vlan_filter_enable(adapter);
3258 		hw->addr_ctrl.user_set_promisc = false;
3259 		/*
3260 		 * Write addresses to available RAR registers, if there is not
3261 		 * sufficient space to store all the addresses then enable
3262 		 * unicast promiscuous mode
3263 		 */
3264 		count = ixgbe_write_uc_addr_list(netdev);
3265 		if (count < 0) {
3266 			fctrl |= IXGBE_FCTRL_UPE;
3267 			vmolr |= IXGBE_VMOLR_ROPE;
3268 		}
3269 	}
3270 
3271 	if (adapter->num_vfs) {
3272 		ixgbe_restore_vf_multicasts(adapter);
3273 		vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3274 			 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3275 			   IXGBE_VMOLR_ROPE);
3276 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3277 	}
3278 
3279 	IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3280 
3281 	if (netdev->features & NETIF_F_HW_VLAN_RX)
3282 		ixgbe_vlan_strip_enable(adapter);
3283 	else
3284 		ixgbe_vlan_strip_disable(adapter);
3285 }
3286 
3287 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3288 {
3289 	int q_idx;
3290 	struct ixgbe_q_vector *q_vector;
3291 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3292 
3293 	/* legacy and MSI only use one vector */
3294 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3295 		q_vectors = 1;
3296 
3297 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3298 		q_vector = adapter->q_vector[q_idx];
3299 		napi_enable(&q_vector->napi);
3300 	}
3301 }
3302 
3303 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3304 {
3305 	int q_idx;
3306 	struct ixgbe_q_vector *q_vector;
3307 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3308 
3309 	/* legacy and MSI only use one vector */
3310 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3311 		q_vectors = 1;
3312 
3313 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3314 		q_vector = adapter->q_vector[q_idx];
3315 		napi_disable(&q_vector->napi);
3316 	}
3317 }
3318 
3319 #ifdef CONFIG_IXGBE_DCB
3320 /*
3321  * ixgbe_configure_dcb - Configure DCB hardware
3322  * @adapter: ixgbe adapter struct
3323  *
3324  * This is called by the driver on open to configure the DCB hardware.
3325  * This is also called by the gennetlink interface when reconfiguring
3326  * the DCB state.
3327  */
3328 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3329 {
3330 	struct ixgbe_hw *hw = &adapter->hw;
3331 	int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3332 
3333 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3334 		if (hw->mac.type == ixgbe_mac_82598EB)
3335 			netif_set_gso_max_size(adapter->netdev, 65536);
3336 		return;
3337 	}
3338 
3339 	if (hw->mac.type == ixgbe_mac_82598EB)
3340 		netif_set_gso_max_size(adapter->netdev, 32768);
3341 
3342 
3343 	/* Enable VLAN tag insert/strip */
3344 	adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3345 
3346 	hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3347 
3348 #ifdef IXGBE_FCOE
3349 	if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3350 		max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3351 #endif
3352 
3353 	/* reconfigure the hardware */
3354 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
3355 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3356 						DCB_TX_CONFIG);
3357 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3358 						DCB_RX_CONFIG);
3359 		ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3360 	} else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
3361 		ixgbe_dcb_hw_ets(&adapter->hw,
3362 				 adapter->ixgbe_ieee_ets,
3363 				 max_frame);
3364 		ixgbe_dcb_hw_pfc_config(&adapter->hw,
3365 					adapter->ixgbe_ieee_pfc->pfc_en,
3366 					adapter->ixgbe_ieee_ets->prio_tc);
3367 	}
3368 
3369 	/* Enable RSS Hash per TC */
3370 	if (hw->mac.type != ixgbe_mac_82598EB) {
3371 		int i;
3372 		u32 reg = 0;
3373 
3374 		for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
3375 			u8 msb = 0;
3376 			u8 cnt = adapter->netdev->tc_to_txq[i].count;
3377 
3378 			while (cnt >>= 1)
3379 				msb++;
3380 
3381 			reg |= msb << IXGBE_RQTC_SHIFT_TC(i);
3382 		}
3383 		IXGBE_WRITE_REG(hw, IXGBE_RQTC, reg);
3384 	}
3385 }
3386 #endif
3387 
3388 /* Additional bittime to account for IXGBE framing */
3389 #define IXGBE_ETH_FRAMING 20
3390 
3391 /*
3392  * ixgbe_hpbthresh - calculate high water mark for flow control
3393  *
3394  * @adapter: board private structure to calculate for
3395  * @pb - packet buffer to calculate
3396  */
3397 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
3398 {
3399 	struct ixgbe_hw *hw = &adapter->hw;
3400 	struct net_device *dev = adapter->netdev;
3401 	int link, tc, kb, marker;
3402 	u32 dv_id, rx_pba;
3403 
3404 	/* Calculate max LAN frame size */
3405 	tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
3406 
3407 #ifdef IXGBE_FCOE
3408 	/* FCoE traffic class uses FCOE jumbo frames */
3409 	if (dev->features & NETIF_F_FCOE_MTU) {
3410 		int fcoe_pb = 0;
3411 
3412 #ifdef CONFIG_IXGBE_DCB
3413 		fcoe_pb = netdev_get_prio_tc_map(dev, adapter->fcoe.up);
3414 
3415 #endif
3416 		if (fcoe_pb == pb && tc < IXGBE_FCOE_JUMBO_FRAME_SIZE)
3417 			tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3418 	}
3419 #endif
3420 
3421 	/* Calculate delay value for device */
3422 	switch (hw->mac.type) {
3423 	case ixgbe_mac_X540:
3424 		dv_id = IXGBE_DV_X540(link, tc);
3425 		break;
3426 	default:
3427 		dv_id = IXGBE_DV(link, tc);
3428 		break;
3429 	}
3430 
3431 	/* Loopback switch introduces additional latency */
3432 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3433 		dv_id += IXGBE_B2BT(tc);
3434 
3435 	/* Delay value is calculated in bit times convert to KB */
3436 	kb = IXGBE_BT2KB(dv_id);
3437 	rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
3438 
3439 	marker = rx_pba - kb;
3440 
3441 	/* It is possible that the packet buffer is not large enough
3442 	 * to provide required headroom. In this case throw an error
3443 	 * to user and a do the best we can.
3444 	 */
3445 	if (marker < 0) {
3446 		e_warn(drv, "Packet Buffer(%i) can not provide enough"
3447 			    "headroom to support flow control."
3448 			    "Decrease MTU or number of traffic classes\n", pb);
3449 		marker = tc + 1;
3450 	}
3451 
3452 	return marker;
3453 }
3454 
3455 /*
3456  * ixgbe_lpbthresh - calculate low water mark for for flow control
3457  *
3458  * @adapter: board private structure to calculate for
3459  * @pb - packet buffer to calculate
3460  */
3461 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter)
3462 {
3463 	struct ixgbe_hw *hw = &adapter->hw;
3464 	struct net_device *dev = adapter->netdev;
3465 	int tc;
3466 	u32 dv_id;
3467 
3468 	/* Calculate max LAN frame size */
3469 	tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
3470 
3471 	/* Calculate delay value for device */
3472 	switch (hw->mac.type) {
3473 	case ixgbe_mac_X540:
3474 		dv_id = IXGBE_LOW_DV_X540(tc);
3475 		break;
3476 	default:
3477 		dv_id = IXGBE_LOW_DV(tc);
3478 		break;
3479 	}
3480 
3481 	/* Delay value is calculated in bit times convert to KB */
3482 	return IXGBE_BT2KB(dv_id);
3483 }
3484 
3485 /*
3486  * ixgbe_pbthresh_setup - calculate and setup high low water marks
3487  */
3488 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
3489 {
3490 	struct ixgbe_hw *hw = &adapter->hw;
3491 	int num_tc = netdev_get_num_tc(adapter->netdev);
3492 	int i;
3493 
3494 	if (!num_tc)
3495 		num_tc = 1;
3496 
3497 	hw->fc.low_water = ixgbe_lpbthresh(adapter);
3498 
3499 	for (i = 0; i < num_tc; i++) {
3500 		hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
3501 
3502 		/* Low water marks must not be larger than high water marks */
3503 		if (hw->fc.low_water > hw->fc.high_water[i])
3504 			hw->fc.low_water = 0;
3505 	}
3506 }
3507 
3508 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
3509 {
3510 	struct ixgbe_hw *hw = &adapter->hw;
3511 	int hdrm;
3512 	u8 tc = netdev_get_num_tc(adapter->netdev);
3513 
3514 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3515 	    adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3516 		hdrm = 32 << adapter->fdir_pballoc;
3517 	else
3518 		hdrm = 0;
3519 
3520 	hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
3521 	ixgbe_pbthresh_setup(adapter);
3522 }
3523 
3524 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
3525 {
3526 	struct ixgbe_hw *hw = &adapter->hw;
3527 	struct hlist_node *node, *node2;
3528 	struct ixgbe_fdir_filter *filter;
3529 
3530 	spin_lock(&adapter->fdir_perfect_lock);
3531 
3532 	if (!hlist_empty(&adapter->fdir_filter_list))
3533 		ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
3534 
3535 	hlist_for_each_entry_safe(filter, node, node2,
3536 				  &adapter->fdir_filter_list, fdir_node) {
3537 		ixgbe_fdir_write_perfect_filter_82599(hw,
3538 				&filter->filter,
3539 				filter->sw_idx,
3540 				(filter->action == IXGBE_FDIR_DROP_QUEUE) ?
3541 				IXGBE_FDIR_DROP_QUEUE :
3542 				adapter->rx_ring[filter->action]->reg_idx);
3543 	}
3544 
3545 	spin_unlock(&adapter->fdir_perfect_lock);
3546 }
3547 
3548 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3549 {
3550 	ixgbe_configure_pb(adapter);
3551 #ifdef CONFIG_IXGBE_DCB
3552 	ixgbe_configure_dcb(adapter);
3553 #endif
3554 
3555 	ixgbe_set_rx_mode(adapter->netdev);
3556 	ixgbe_restore_vlan(adapter);
3557 
3558 #ifdef IXGBE_FCOE
3559 	if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3560 		ixgbe_configure_fcoe(adapter);
3561 
3562 #endif /* IXGBE_FCOE */
3563 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3564 		ixgbe_init_fdir_signature_82599(&adapter->hw,
3565 						adapter->fdir_pballoc);
3566 	} else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3567 		ixgbe_init_fdir_perfect_82599(&adapter->hw,
3568 					      adapter->fdir_pballoc);
3569 		ixgbe_fdir_filter_restore(adapter);
3570 	}
3571 
3572 	ixgbe_configure_virtualization(adapter);
3573 
3574 	ixgbe_configure_tx(adapter);
3575 	ixgbe_configure_rx(adapter);
3576 }
3577 
3578 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3579 {
3580 	switch (hw->phy.type) {
3581 	case ixgbe_phy_sfp_avago:
3582 	case ixgbe_phy_sfp_ftl:
3583 	case ixgbe_phy_sfp_intel:
3584 	case ixgbe_phy_sfp_unknown:
3585 	case ixgbe_phy_sfp_passive_tyco:
3586 	case ixgbe_phy_sfp_passive_unknown:
3587 	case ixgbe_phy_sfp_active_unknown:
3588 	case ixgbe_phy_sfp_ftl_active:
3589 		return true;
3590 	case ixgbe_phy_nl:
3591 		if (hw->mac.type == ixgbe_mac_82598EB)
3592 			return true;
3593 	default:
3594 		return false;
3595 	}
3596 }
3597 
3598 /**
3599  * ixgbe_sfp_link_config - set up SFP+ link
3600  * @adapter: pointer to private adapter struct
3601  **/
3602 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3603 {
3604 	/*
3605 	 * We are assuming the worst case scenerio here, and that
3606 	 * is that an SFP was inserted/removed after the reset
3607 	 * but before SFP detection was enabled.  As such the best
3608 	 * solution is to just start searching as soon as we start
3609 	 */
3610 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3611 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
3612 
3613 	adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
3614 }
3615 
3616 /**
3617  * ixgbe_non_sfp_link_config - set up non-SFP+ link
3618  * @hw: pointer to private hardware struct
3619  *
3620  * Returns 0 on success, negative on failure
3621  **/
3622 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3623 {
3624 	u32 autoneg;
3625 	bool negotiation, link_up = false;
3626 	u32 ret = IXGBE_ERR_LINK_SETUP;
3627 
3628 	if (hw->mac.ops.check_link)
3629 		ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3630 
3631 	if (ret)
3632 		goto link_cfg_out;
3633 
3634 	autoneg = hw->phy.autoneg_advertised;
3635 	if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
3636 		ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3637 							&negotiation);
3638 	if (ret)
3639 		goto link_cfg_out;
3640 
3641 	if (hw->mac.ops.setup_link)
3642 		ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3643 link_cfg_out:
3644 	return ret;
3645 }
3646 
3647 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3648 {
3649 	struct ixgbe_hw *hw = &adapter->hw;
3650 	u32 gpie = 0;
3651 
3652 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3653 		gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3654 		       IXGBE_GPIE_OCD;
3655 		gpie |= IXGBE_GPIE_EIAME;
3656 		/*
3657 		 * use EIAM to auto-mask when MSI-X interrupt is asserted
3658 		 * this saves a register write for every interrupt
3659 		 */
3660 		switch (hw->mac.type) {
3661 		case ixgbe_mac_82598EB:
3662 			IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3663 			break;
3664 		case ixgbe_mac_82599EB:
3665 		case ixgbe_mac_X540:
3666 		default:
3667 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3668 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3669 			break;
3670 		}
3671 	} else {
3672 		/* legacy interrupts, use EIAM to auto-mask when reading EICR,
3673 		 * specifically only auto mask tx and rx interrupts */
3674 		IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3675 	}
3676 
3677 	/* XXX: to interrupt immediately for EICS writes, enable this */
3678 	/* gpie |= IXGBE_GPIE_EIMEN; */
3679 
3680 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3681 		gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3682 		gpie |= IXGBE_GPIE_VTMODE_64;
3683 	}
3684 
3685 	/* Enable Thermal over heat sensor interrupt */
3686 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
3687 		switch (adapter->hw.mac.type) {
3688 		case ixgbe_mac_82599EB:
3689 			gpie |= IXGBE_SDP0_GPIEN;
3690 			break;
3691 		case ixgbe_mac_X540:
3692 			gpie |= IXGBE_EIMS_TS;
3693 			break;
3694 		default:
3695 			break;
3696 		}
3697 	}
3698 
3699 	/* Enable fan failure interrupt */
3700 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3701 		gpie |= IXGBE_SDP1_GPIEN;
3702 
3703 	if (hw->mac.type == ixgbe_mac_82599EB) {
3704 		gpie |= IXGBE_SDP1_GPIEN;
3705 		gpie |= IXGBE_SDP2_GPIEN;
3706 	}
3707 
3708 	IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3709 }
3710 
3711 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
3712 {
3713 	struct ixgbe_hw *hw = &adapter->hw;
3714 	int err;
3715 	u32 ctrl_ext;
3716 
3717 	ixgbe_get_hw_control(adapter);
3718 	ixgbe_setup_gpie(adapter);
3719 
3720 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3721 		ixgbe_configure_msix(adapter);
3722 	else
3723 		ixgbe_configure_msi_and_legacy(adapter);
3724 
3725 	/* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
3726 	if (hw->mac.ops.enable_tx_laser &&
3727 	    ((hw->phy.multispeed_fiber) ||
3728 	     ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3729 	      (hw->mac.type == ixgbe_mac_82599EB))))
3730 		hw->mac.ops.enable_tx_laser(hw);
3731 
3732 	clear_bit(__IXGBE_DOWN, &adapter->state);
3733 	ixgbe_napi_enable_all(adapter);
3734 
3735 	if (ixgbe_is_sfp(hw)) {
3736 		ixgbe_sfp_link_config(adapter);
3737 	} else {
3738 		err = ixgbe_non_sfp_link_config(hw);
3739 		if (err)
3740 			e_err(probe, "link_config FAILED %d\n", err);
3741 	}
3742 
3743 	/* clear any pending interrupts, may auto mask */
3744 	IXGBE_READ_REG(hw, IXGBE_EICR);
3745 	ixgbe_irq_enable(adapter, true, true);
3746 
3747 	/*
3748 	 * If this adapter has a fan, check to see if we had a failure
3749 	 * before we enabled the interrupt.
3750 	 */
3751 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3752 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3753 		if (esdp & IXGBE_ESDP_SDP1)
3754 			e_crit(drv, "Fan has stopped, replace the adapter\n");
3755 	}
3756 
3757 	/* enable transmits */
3758 	netif_tx_start_all_queues(adapter->netdev);
3759 
3760 	/* bring the link up in the watchdog, this could race with our first
3761 	 * link up interrupt but shouldn't be a problem */
3762 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3763 	adapter->link_check_timeout = jiffies;
3764 	mod_timer(&adapter->service_timer, jiffies);
3765 
3766 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
3767 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3768 	ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3769 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3770 }
3771 
3772 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3773 {
3774 	WARN_ON(in_interrupt());
3775 	/* put off any impending NetWatchDogTimeout */
3776 	adapter->netdev->trans_start = jiffies;
3777 
3778 	while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3779 		usleep_range(1000, 2000);
3780 	ixgbe_down(adapter);
3781 	/*
3782 	 * If SR-IOV enabled then wait a bit before bringing the adapter
3783 	 * back up to give the VFs time to respond to the reset.  The
3784 	 * two second wait is based upon the watchdog timer cycle in
3785 	 * the VF driver.
3786 	 */
3787 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3788 		msleep(2000);
3789 	ixgbe_up(adapter);
3790 	clear_bit(__IXGBE_RESETTING, &adapter->state);
3791 }
3792 
3793 void ixgbe_up(struct ixgbe_adapter *adapter)
3794 {
3795 	/* hardware has been reset, we need to reload some things */
3796 	ixgbe_configure(adapter);
3797 
3798 	ixgbe_up_complete(adapter);
3799 }
3800 
3801 void ixgbe_reset(struct ixgbe_adapter *adapter)
3802 {
3803 	struct ixgbe_hw *hw = &adapter->hw;
3804 	int err;
3805 
3806 	/* lock SFP init bit to prevent race conditions with the watchdog */
3807 	while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
3808 		usleep_range(1000, 2000);
3809 
3810 	/* clear all SFP and link config related flags while holding SFP_INIT */
3811 	adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
3812 			     IXGBE_FLAG2_SFP_NEEDS_RESET);
3813 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
3814 
3815 	err = hw->mac.ops.init_hw(hw);
3816 	switch (err) {
3817 	case 0:
3818 	case IXGBE_ERR_SFP_NOT_PRESENT:
3819 	case IXGBE_ERR_SFP_NOT_SUPPORTED:
3820 		break;
3821 	case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3822 		e_dev_err("master disable timed out\n");
3823 		break;
3824 	case IXGBE_ERR_EEPROM_VERSION:
3825 		/* We are running on a pre-production device, log a warning */
3826 		e_dev_warn("This device is a pre-production adapter/LOM. "
3827 			   "Please be aware there may be issuesassociated with "
3828 			   "your hardware.  If you are experiencing problems "
3829 			   "please contact your Intel or hardware "
3830 			   "representative who provided you with this "
3831 			   "hardware.\n");
3832 		break;
3833 	default:
3834 		e_dev_err("Hardware Error: %d\n", err);
3835 	}
3836 
3837 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
3838 
3839 	/* reprogram the RAR[0] in case user changed it. */
3840 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3841 			    IXGBE_RAH_AV);
3842 }
3843 
3844 /**
3845  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3846  * @rx_ring: ring to free buffers from
3847  **/
3848 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
3849 {
3850 	struct device *dev = rx_ring->dev;
3851 	unsigned long size;
3852 	u16 i;
3853 
3854 	/* ring already cleared, nothing to do */
3855 	if (!rx_ring->rx_buffer_info)
3856 		return;
3857 
3858 	/* Free all the Rx ring sk_buffs */
3859 	for (i = 0; i < rx_ring->count; i++) {
3860 		struct ixgbe_rx_buffer *rx_buffer_info;
3861 
3862 		rx_buffer_info = &rx_ring->rx_buffer_info[i];
3863 		if (rx_buffer_info->dma) {
3864 			dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
3865 					 rx_ring->rx_buf_len,
3866 					 DMA_FROM_DEVICE);
3867 			rx_buffer_info->dma = 0;
3868 		}
3869 		if (rx_buffer_info->skb) {
3870 			struct sk_buff *skb = rx_buffer_info->skb;
3871 			rx_buffer_info->skb = NULL;
3872 			do {
3873 				struct sk_buff *this = skb;
3874 				if (IXGBE_RSC_CB(this)->delay_unmap) {
3875 					dma_unmap_single(dev,
3876 							 IXGBE_RSC_CB(this)->dma,
3877 							 rx_ring->rx_buf_len,
3878 							 DMA_FROM_DEVICE);
3879 					IXGBE_RSC_CB(this)->dma = 0;
3880 					IXGBE_RSC_CB(skb)->delay_unmap = false;
3881 				}
3882 				skb = skb->prev;
3883 				dev_kfree_skb(this);
3884 			} while (skb);
3885 		}
3886 		if (!rx_buffer_info->page)
3887 			continue;
3888 		if (rx_buffer_info->page_dma) {
3889 			dma_unmap_page(dev, rx_buffer_info->page_dma,
3890 				       PAGE_SIZE / 2, DMA_FROM_DEVICE);
3891 			rx_buffer_info->page_dma = 0;
3892 		}
3893 		put_page(rx_buffer_info->page);
3894 		rx_buffer_info->page = NULL;
3895 		rx_buffer_info->page_offset = 0;
3896 	}
3897 
3898 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3899 	memset(rx_ring->rx_buffer_info, 0, size);
3900 
3901 	/* Zero out the descriptor ring */
3902 	memset(rx_ring->desc, 0, rx_ring->size);
3903 
3904 	rx_ring->next_to_clean = 0;
3905 	rx_ring->next_to_use = 0;
3906 }
3907 
3908 /**
3909  * ixgbe_clean_tx_ring - Free Tx Buffers
3910  * @tx_ring: ring to be cleaned
3911  **/
3912 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
3913 {
3914 	struct ixgbe_tx_buffer *tx_buffer_info;
3915 	unsigned long size;
3916 	u16 i;
3917 
3918 	/* ring already cleared, nothing to do */
3919 	if (!tx_ring->tx_buffer_info)
3920 		return;
3921 
3922 	/* Free all the Tx ring sk_buffs */
3923 	for (i = 0; i < tx_ring->count; i++) {
3924 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
3925 		ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
3926 	}
3927 
3928 	size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3929 	memset(tx_ring->tx_buffer_info, 0, size);
3930 
3931 	/* Zero out the descriptor ring */
3932 	memset(tx_ring->desc, 0, tx_ring->size);
3933 
3934 	tx_ring->next_to_use = 0;
3935 	tx_ring->next_to_clean = 0;
3936 }
3937 
3938 /**
3939  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3940  * @adapter: board private structure
3941  **/
3942 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3943 {
3944 	int i;
3945 
3946 	for (i = 0; i < adapter->num_rx_queues; i++)
3947 		ixgbe_clean_rx_ring(adapter->rx_ring[i]);
3948 }
3949 
3950 /**
3951  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3952  * @adapter: board private structure
3953  **/
3954 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3955 {
3956 	int i;
3957 
3958 	for (i = 0; i < adapter->num_tx_queues; i++)
3959 		ixgbe_clean_tx_ring(adapter->tx_ring[i]);
3960 }
3961 
3962 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
3963 {
3964 	struct hlist_node *node, *node2;
3965 	struct ixgbe_fdir_filter *filter;
3966 
3967 	spin_lock(&adapter->fdir_perfect_lock);
3968 
3969 	hlist_for_each_entry_safe(filter, node, node2,
3970 				  &adapter->fdir_filter_list, fdir_node) {
3971 		hlist_del(&filter->fdir_node);
3972 		kfree(filter);
3973 	}
3974 	adapter->fdir_filter_count = 0;
3975 
3976 	spin_unlock(&adapter->fdir_perfect_lock);
3977 }
3978 
3979 void ixgbe_down(struct ixgbe_adapter *adapter)
3980 {
3981 	struct net_device *netdev = adapter->netdev;
3982 	struct ixgbe_hw *hw = &adapter->hw;
3983 	u32 rxctrl;
3984 	int i;
3985 
3986 	/* signal that we are down to the interrupt handler */
3987 	set_bit(__IXGBE_DOWN, &adapter->state);
3988 
3989 	/* disable receives */
3990 	rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3991 	IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3992 
3993 	/* disable all enabled rx queues */
3994 	for (i = 0; i < adapter->num_rx_queues; i++)
3995 		/* this call also flushes the previous write */
3996 		ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
3997 
3998 	usleep_range(10000, 20000);
3999 
4000 	netif_tx_stop_all_queues(netdev);
4001 
4002 	/* call carrier off first to avoid false dev_watchdog timeouts */
4003 	netif_carrier_off(netdev);
4004 	netif_tx_disable(netdev);
4005 
4006 	ixgbe_irq_disable(adapter);
4007 
4008 	ixgbe_napi_disable_all(adapter);
4009 
4010 	adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
4011 			     IXGBE_FLAG2_RESET_REQUESTED);
4012 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4013 
4014 	del_timer_sync(&adapter->service_timer);
4015 
4016 	if (adapter->num_vfs) {
4017 		/* Clear EITR Select mapping */
4018 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
4019 
4020 		/* Mark all the VFs as inactive */
4021 		for (i = 0 ; i < adapter->num_vfs; i++)
4022 			adapter->vfinfo[i].clear_to_send = 0;
4023 
4024 		/* ping all the active vfs to let them know we are going down */
4025 		ixgbe_ping_all_vfs(adapter);
4026 
4027 		/* Disable all VFTE/VFRE TX/RX */
4028 		ixgbe_disable_tx_rx(adapter);
4029 	}
4030 
4031 	/* disable transmits in the hardware now that interrupts are off */
4032 	for (i = 0; i < adapter->num_tx_queues; i++) {
4033 		u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4034 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
4035 	}
4036 
4037 	/* Disable the Tx DMA engine on 82599 and X540 */
4038 	switch (hw->mac.type) {
4039 	case ixgbe_mac_82599EB:
4040 	case ixgbe_mac_X540:
4041 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4042 				(IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4043 				 ~IXGBE_DMATXCTL_TE));
4044 		break;
4045 	default:
4046 		break;
4047 	}
4048 
4049 	if (!pci_channel_offline(adapter->pdev))
4050 		ixgbe_reset(adapter);
4051 
4052 	/* power down the optics for multispeed fiber and 82599 SFP+ fiber */
4053 	if (hw->mac.ops.disable_tx_laser &&
4054 	    ((hw->phy.multispeed_fiber) ||
4055 	     ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4056 	      (hw->mac.type == ixgbe_mac_82599EB))))
4057 		hw->mac.ops.disable_tx_laser(hw);
4058 
4059 	ixgbe_clean_all_tx_rings(adapter);
4060 	ixgbe_clean_all_rx_rings(adapter);
4061 
4062 #ifdef CONFIG_IXGBE_DCA
4063 	/* since we reset the hardware DCA settings were cleared */
4064 	ixgbe_setup_dca(adapter);
4065 #endif
4066 }
4067 
4068 /**
4069  * ixgbe_poll - NAPI Rx polling callback
4070  * @napi: structure for representing this polling device
4071  * @budget: how many packets driver is allowed to clean
4072  *
4073  * This function is used for legacy and MSI, NAPI mode
4074  **/
4075 static int ixgbe_poll(struct napi_struct *napi, int budget)
4076 {
4077 	struct ixgbe_q_vector *q_vector =
4078 				container_of(napi, struct ixgbe_q_vector, napi);
4079 	struct ixgbe_adapter *adapter = q_vector->adapter;
4080 	struct ixgbe_ring *ring;
4081 	int per_ring_budget;
4082 	bool clean_complete = true;
4083 
4084 #ifdef CONFIG_IXGBE_DCA
4085 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
4086 		ixgbe_update_dca(q_vector);
4087 #endif
4088 
4089 	for (ring = q_vector->tx.ring; ring != NULL; ring = ring->next)
4090 		clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
4091 
4092 	/* attempt to distribute budget to each queue fairly, but don't allow
4093 	 * the budget to go below 1 because we'll exit polling */
4094 	if (q_vector->rx.count > 1)
4095 		per_ring_budget = max(budget/q_vector->rx.count, 1);
4096 	else
4097 		per_ring_budget = budget;
4098 
4099 	for (ring = q_vector->rx.ring; ring != NULL; ring = ring->next)
4100 		clean_complete &= ixgbe_clean_rx_irq(q_vector, ring,
4101 						     per_ring_budget);
4102 
4103 	/* If all work not completed, return budget and keep polling */
4104 	if (!clean_complete)
4105 		return budget;
4106 
4107 	/* all work done, exit the polling mode */
4108 	napi_complete(napi);
4109 	if (adapter->rx_itr_setting & 1)
4110 		ixgbe_set_itr(q_vector);
4111 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
4112 		ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
4113 
4114 	return 0;
4115 }
4116 
4117 /**
4118  * ixgbe_tx_timeout - Respond to a Tx Hang
4119  * @netdev: network interface device structure
4120  **/
4121 static void ixgbe_tx_timeout(struct net_device *netdev)
4122 {
4123 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4124 
4125 	/* Do the reset outside of interrupt context */
4126 	ixgbe_tx_timeout_reset(adapter);
4127 }
4128 
4129 /**
4130  * ixgbe_set_rss_queues: Allocate queues for RSS
4131  * @adapter: board private structure to initialize
4132  *
4133  * This is our "base" multiqueue mode.  RSS (Receive Side Scaling) will try
4134  * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
4135  *
4136  **/
4137 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
4138 {
4139 	bool ret = false;
4140 	struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
4141 
4142 	if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4143 		f->mask = 0xF;
4144 		adapter->num_rx_queues = f->indices;
4145 		adapter->num_tx_queues = f->indices;
4146 		ret = true;
4147 	} else {
4148 		ret = false;
4149 	}
4150 
4151 	return ret;
4152 }
4153 
4154 /**
4155  * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4156  * @adapter: board private structure to initialize
4157  *
4158  * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4159  * to the original CPU that initiated the Tx session.  This runs in addition
4160  * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4161  * Rx load across CPUs using RSS.
4162  *
4163  **/
4164 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4165 {
4166 	bool ret = false;
4167 	struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4168 
4169 	f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4170 	f_fdir->mask = 0;
4171 
4172 	/* Flow Director must have RSS enabled */
4173 	if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
4174 	    (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)) {
4175 		adapter->num_tx_queues = f_fdir->indices;
4176 		adapter->num_rx_queues = f_fdir->indices;
4177 		ret = true;
4178 	} else {
4179 		adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4180 	}
4181 	return ret;
4182 }
4183 
4184 #ifdef IXGBE_FCOE
4185 /**
4186  * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4187  * @adapter: board private structure to initialize
4188  *
4189  * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4190  * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4191  * rx queues out of the max number of rx queues, instead, it is used as the
4192  * index of the first rx queue used by FCoE.
4193  *
4194  **/
4195 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4196 {
4197 	struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4198 
4199 	if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4200 		return false;
4201 
4202 	f->indices = min((int)num_online_cpus(), f->indices);
4203 
4204 	adapter->num_rx_queues = 1;
4205 	adapter->num_tx_queues = 1;
4206 
4207 	if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4208 		e_info(probe, "FCoE enabled with RSS\n");
4209 		if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
4210 			ixgbe_set_fdir_queues(adapter);
4211 		else
4212 			ixgbe_set_rss_queues(adapter);
4213 	}
4214 
4215 	/* adding FCoE rx rings to the end */
4216 	f->mask = adapter->num_rx_queues;
4217 	adapter->num_rx_queues += f->indices;
4218 	adapter->num_tx_queues += f->indices;
4219 
4220 	return true;
4221 }
4222 #endif /* IXGBE_FCOE */
4223 
4224 /* Artificial max queue cap per traffic class in DCB mode */
4225 #define DCB_QUEUE_CAP 8
4226 
4227 #ifdef CONFIG_IXGBE_DCB
4228 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
4229 {
4230 	int per_tc_q, q, i, offset = 0;
4231 	struct net_device *dev = adapter->netdev;
4232 	int tcs = netdev_get_num_tc(dev);
4233 
4234 	if (!tcs)
4235 		return false;
4236 
4237 	/* Map queue offset and counts onto allocated tx queues */
4238 	per_tc_q = min(dev->num_tx_queues / tcs, (unsigned int)DCB_QUEUE_CAP);
4239 	q = min((int)num_online_cpus(), per_tc_q);
4240 
4241 	for (i = 0; i < tcs; i++) {
4242 		netdev_set_tc_queue(dev, i, q, offset);
4243 		offset += q;
4244 	}
4245 
4246 	adapter->num_tx_queues = q * tcs;
4247 	adapter->num_rx_queues = q * tcs;
4248 
4249 #ifdef IXGBE_FCOE
4250 	/* FCoE enabled queues require special configuration indexed
4251 	 * by feature specific indices and mask. Here we map FCoE
4252 	 * indices onto the DCB queue pairs allowing FCoE to own
4253 	 * configuration later.
4254 	 */
4255 	if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4256 		int tc;
4257 		struct ixgbe_ring_feature *f =
4258 					&adapter->ring_feature[RING_F_FCOE];
4259 
4260 		tc = netdev_get_prio_tc_map(dev, adapter->fcoe.up);
4261 		f->indices = dev->tc_to_txq[tc].count;
4262 		f->mask = dev->tc_to_txq[tc].offset;
4263 	}
4264 #endif
4265 
4266 	return true;
4267 }
4268 #endif
4269 
4270 /**
4271  * ixgbe_set_sriov_queues: Allocate queues for IOV use
4272  * @adapter: board private structure to initialize
4273  *
4274  * IOV doesn't actually use anything, so just NAK the
4275  * request for now and let the other queue routines
4276  * figure out what to do.
4277  */
4278 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4279 {
4280 	return false;
4281 }
4282 
4283 /*
4284  * ixgbe_set_num_queues: Allocate queues for device, feature dependent
4285  * @adapter: board private structure to initialize
4286  *
4287  * This is the top level queue allocation routine.  The order here is very
4288  * important, starting with the "most" number of features turned on at once,
4289  * and ending with the smallest set of features.  This way large combinations
4290  * can be allocated if they're turned on, and smaller combinations are the
4291  * fallthrough conditions.
4292  *
4293  **/
4294 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4295 {
4296 	/* Start with base case */
4297 	adapter->num_rx_queues = 1;
4298 	adapter->num_tx_queues = 1;
4299 	adapter->num_rx_pools = adapter->num_rx_queues;
4300 	adapter->num_rx_queues_per_pool = 1;
4301 
4302 	if (ixgbe_set_sriov_queues(adapter))
4303 		goto done;
4304 
4305 #ifdef CONFIG_IXGBE_DCB
4306 	if (ixgbe_set_dcb_queues(adapter))
4307 		goto done;
4308 
4309 #endif
4310 #ifdef IXGBE_FCOE
4311 	if (ixgbe_set_fcoe_queues(adapter))
4312 		goto done;
4313 
4314 #endif /* IXGBE_FCOE */
4315 	if (ixgbe_set_fdir_queues(adapter))
4316 		goto done;
4317 
4318 	if (ixgbe_set_rss_queues(adapter))
4319 		goto done;
4320 
4321 	/* fallback to base case */
4322 	adapter->num_rx_queues = 1;
4323 	adapter->num_tx_queues = 1;
4324 
4325 done:
4326 	/* Notify the stack of the (possibly) reduced queue counts. */
4327 	netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4328 	return netif_set_real_num_rx_queues(adapter->netdev,
4329 					    adapter->num_rx_queues);
4330 }
4331 
4332 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4333 				       int vectors)
4334 {
4335 	int err, vector_threshold;
4336 
4337 	/* We'll want at least 3 (vector_threshold):
4338 	 * 1) TxQ[0] Cleanup
4339 	 * 2) RxQ[0] Cleanup
4340 	 * 3) Other (Link Status Change, etc.)
4341 	 * 4) TCP Timer (optional)
4342 	 */
4343 	vector_threshold = MIN_MSIX_COUNT;
4344 
4345 	/* The more we get, the more we will assign to Tx/Rx Cleanup
4346 	 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4347 	 * Right now, we simply care about how many we'll get; we'll
4348 	 * set them up later while requesting irq's.
4349 	 */
4350 	while (vectors >= vector_threshold) {
4351 		err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4352 				      vectors);
4353 		if (!err) /* Success in acquiring all requested vectors. */
4354 			break;
4355 		else if (err < 0)
4356 			vectors = 0; /* Nasty failure, quit now */
4357 		else /* err == number of vectors we should try again with */
4358 			vectors = err;
4359 	}
4360 
4361 	if (vectors < vector_threshold) {
4362 		/* Can't allocate enough MSI-X interrupts?  Oh well.
4363 		 * This just means we'll go with either a single MSI
4364 		 * vector or fall back to legacy interrupts.
4365 		 */
4366 		netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4367 			     "Unable to allocate MSI-X interrupts\n");
4368 		adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4369 		kfree(adapter->msix_entries);
4370 		adapter->msix_entries = NULL;
4371 	} else {
4372 		adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4373 		/*
4374 		 * Adjust for only the vectors we'll use, which is minimum
4375 		 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4376 		 * vectors we were allocated.
4377 		 */
4378 		adapter->num_msix_vectors = min(vectors,
4379 				   adapter->max_msix_q_vectors + NON_Q_VECTORS);
4380 	}
4381 }
4382 
4383 /**
4384  * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4385  * @adapter: board private structure to initialize
4386  *
4387  * Cache the descriptor ring offsets for RSS to the assigned rings.
4388  *
4389  **/
4390 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4391 {
4392 	int i;
4393 
4394 	if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
4395 		return false;
4396 
4397 	for (i = 0; i < adapter->num_rx_queues; i++)
4398 		adapter->rx_ring[i]->reg_idx = i;
4399 	for (i = 0; i < adapter->num_tx_queues; i++)
4400 		adapter->tx_ring[i]->reg_idx = i;
4401 
4402 	return true;
4403 }
4404 
4405 #ifdef CONFIG_IXGBE_DCB
4406 
4407 /* ixgbe_get_first_reg_idx - Return first register index associated with ring */
4408 static void ixgbe_get_first_reg_idx(struct ixgbe_adapter *adapter, u8 tc,
4409 				    unsigned int *tx, unsigned int *rx)
4410 {
4411 	struct net_device *dev = adapter->netdev;
4412 	struct ixgbe_hw *hw = &adapter->hw;
4413 	u8 num_tcs = netdev_get_num_tc(dev);
4414 
4415 	*tx = 0;
4416 	*rx = 0;
4417 
4418 	switch (hw->mac.type) {
4419 	case ixgbe_mac_82598EB:
4420 		*tx = tc << 2;
4421 		*rx = tc << 3;
4422 		break;
4423 	case ixgbe_mac_82599EB:
4424 	case ixgbe_mac_X540:
4425 		if (num_tcs > 4) {
4426 			if (tc < 3) {
4427 				*tx = tc << 5;
4428 				*rx = tc << 4;
4429 			} else if (tc <  5) {
4430 				*tx = ((tc + 2) << 4);
4431 				*rx = tc << 4;
4432 			} else if (tc < num_tcs) {
4433 				*tx = ((tc + 8) << 3);
4434 				*rx = tc << 4;
4435 			}
4436 		} else {
4437 			*rx =  tc << 5;
4438 			switch (tc) {
4439 			case 0:
4440 				*tx =  0;
4441 				break;
4442 			case 1:
4443 				*tx = 64;
4444 				break;
4445 			case 2:
4446 				*tx = 96;
4447 				break;
4448 			case 3:
4449 				*tx = 112;
4450 				break;
4451 			default:
4452 				break;
4453 			}
4454 		}
4455 		break;
4456 	default:
4457 		break;
4458 	}
4459 }
4460 
4461 /**
4462  * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4463  * @adapter: board private structure to initialize
4464  *
4465  * Cache the descriptor ring offsets for DCB to the assigned rings.
4466  *
4467  **/
4468 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4469 {
4470 	struct net_device *dev = adapter->netdev;
4471 	int i, j, k;
4472 	u8 num_tcs = netdev_get_num_tc(dev);
4473 
4474 	if (!num_tcs)
4475 		return false;
4476 
4477 	for (i = 0, k = 0; i < num_tcs; i++) {
4478 		unsigned int tx_s, rx_s;
4479 		u16 count = dev->tc_to_txq[i].count;
4480 
4481 		ixgbe_get_first_reg_idx(adapter, i, &tx_s, &rx_s);
4482 		for (j = 0; j < count; j++, k++) {
4483 			adapter->tx_ring[k]->reg_idx = tx_s + j;
4484 			adapter->rx_ring[k]->reg_idx = rx_s + j;
4485 			adapter->tx_ring[k]->dcb_tc = i;
4486 			adapter->rx_ring[k]->dcb_tc = i;
4487 		}
4488 	}
4489 
4490 	return true;
4491 }
4492 #endif
4493 
4494 /**
4495  * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4496  * @adapter: board private structure to initialize
4497  *
4498  * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4499  *
4500  **/
4501 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4502 {
4503 	int i;
4504 	bool ret = false;
4505 
4506 	if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
4507 	    (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)) {
4508 		for (i = 0; i < adapter->num_rx_queues; i++)
4509 			adapter->rx_ring[i]->reg_idx = i;
4510 		for (i = 0; i < adapter->num_tx_queues; i++)
4511 			adapter->tx_ring[i]->reg_idx = i;
4512 		ret = true;
4513 	}
4514 
4515 	return ret;
4516 }
4517 
4518 #ifdef IXGBE_FCOE
4519 /**
4520  * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4521  * @adapter: board private structure to initialize
4522  *
4523  * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4524  *
4525  */
4526 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4527 {
4528 	struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4529 	int i;
4530 	u8 fcoe_rx_i = 0, fcoe_tx_i = 0;
4531 
4532 	if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4533 		return false;
4534 
4535 	if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4536 		if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
4537 			ixgbe_cache_ring_fdir(adapter);
4538 		else
4539 			ixgbe_cache_ring_rss(adapter);
4540 
4541 		fcoe_rx_i = f->mask;
4542 		fcoe_tx_i = f->mask;
4543 	}
4544 	for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4545 		adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4546 		adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4547 	}
4548 	return true;
4549 }
4550 
4551 #endif /* IXGBE_FCOE */
4552 /**
4553  * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4554  * @adapter: board private structure to initialize
4555  *
4556  * SR-IOV doesn't use any descriptor rings but changes the default if
4557  * no other mapping is used.
4558  *
4559  */
4560 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4561 {
4562 	adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4563 	adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4564 	if (adapter->num_vfs)
4565 		return true;
4566 	else
4567 		return false;
4568 }
4569 
4570 /**
4571  * ixgbe_cache_ring_register - Descriptor ring to register mapping
4572  * @adapter: board private structure to initialize
4573  *
4574  * Once we know the feature-set enabled for the device, we'll cache
4575  * the register offset the descriptor ring is assigned to.
4576  *
4577  * Note, the order the various feature calls is important.  It must start with
4578  * the "most" features enabled at the same time, then trickle down to the
4579  * least amount of features turned on at once.
4580  **/
4581 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4582 {
4583 	/* start with default case */
4584 	adapter->rx_ring[0]->reg_idx = 0;
4585 	adapter->tx_ring[0]->reg_idx = 0;
4586 
4587 	if (ixgbe_cache_ring_sriov(adapter))
4588 		return;
4589 
4590 #ifdef CONFIG_IXGBE_DCB
4591 	if (ixgbe_cache_ring_dcb(adapter))
4592 		return;
4593 #endif
4594 
4595 #ifdef IXGBE_FCOE
4596 	if (ixgbe_cache_ring_fcoe(adapter))
4597 		return;
4598 #endif /* IXGBE_FCOE */
4599 
4600 	if (ixgbe_cache_ring_fdir(adapter))
4601 		return;
4602 
4603 	if (ixgbe_cache_ring_rss(adapter))
4604 		return;
4605 }
4606 
4607 /**
4608  * ixgbe_alloc_queues - Allocate memory for all rings
4609  * @adapter: board private structure to initialize
4610  *
4611  * We allocate one ring per queue at run-time since we don't know the
4612  * number of queues at compile-time.  The polling_netdev array is
4613  * intended for Multiqueue, but should work fine with a single queue.
4614  **/
4615 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4616 {
4617 	int rx = 0, tx = 0, nid = adapter->node;
4618 
4619 	if (nid < 0 || !node_online(nid))
4620 		nid = first_online_node;
4621 
4622 	for (; tx < adapter->num_tx_queues; tx++) {
4623 		struct ixgbe_ring *ring;
4624 
4625 		ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4626 		if (!ring)
4627 			ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4628 		if (!ring)
4629 			goto err_allocation;
4630 		ring->count = adapter->tx_ring_count;
4631 		ring->queue_index = tx;
4632 		ring->numa_node = nid;
4633 		ring->dev = &adapter->pdev->dev;
4634 		ring->netdev = adapter->netdev;
4635 
4636 		adapter->tx_ring[tx] = ring;
4637 	}
4638 
4639 	for (; rx < adapter->num_rx_queues; rx++) {
4640 		struct ixgbe_ring *ring;
4641 
4642 		ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4643 		if (!ring)
4644 			ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4645 		if (!ring)
4646 			goto err_allocation;
4647 		ring->count = adapter->rx_ring_count;
4648 		ring->queue_index = rx;
4649 		ring->numa_node = nid;
4650 		ring->dev = &adapter->pdev->dev;
4651 		ring->netdev = adapter->netdev;
4652 
4653 		adapter->rx_ring[rx] = ring;
4654 	}
4655 
4656 	ixgbe_cache_ring_register(adapter);
4657 
4658 	return 0;
4659 
4660 err_allocation:
4661 	while (tx)
4662 		kfree(adapter->tx_ring[--tx]);
4663 
4664 	while (rx)
4665 		kfree(adapter->rx_ring[--rx]);
4666 	return -ENOMEM;
4667 }
4668 
4669 /**
4670  * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4671  * @adapter: board private structure to initialize
4672  *
4673  * Attempt to configure the interrupts using the best available
4674  * capabilities of the hardware and the kernel.
4675  **/
4676 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4677 {
4678 	struct ixgbe_hw *hw = &adapter->hw;
4679 	int err = 0;
4680 	int vector, v_budget;
4681 
4682 	/*
4683 	 * It's easy to be greedy for MSI-X vectors, but it really
4684 	 * doesn't do us much good if we have a lot more vectors
4685 	 * than CPU's.  So let's be conservative and only ask for
4686 	 * (roughly) the same number of vectors as there are CPU's.
4687 	 */
4688 	v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4689 		       (int)num_online_cpus()) + NON_Q_VECTORS;
4690 
4691 	/*
4692 	 * At the same time, hardware can only support a maximum of
4693 	 * hw.mac->max_msix_vectors vectors.  With features
4694 	 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4695 	 * descriptor queues supported by our device.  Thus, we cap it off in
4696 	 * those rare cases where the cpu count also exceeds our vector limit.
4697 	 */
4698 	v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4699 
4700 	/* A failure in MSI-X entry allocation isn't fatal, but it does
4701 	 * mean we disable MSI-X capabilities of the adapter. */
4702 	adapter->msix_entries = kcalloc(v_budget,
4703 					sizeof(struct msix_entry), GFP_KERNEL);
4704 	if (adapter->msix_entries) {
4705 		for (vector = 0; vector < v_budget; vector++)
4706 			adapter->msix_entries[vector].entry = vector;
4707 
4708 		ixgbe_acquire_msix_vectors(adapter, v_budget);
4709 
4710 		if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4711 			goto out;
4712 	}
4713 
4714 	adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4715 	adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4716 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
4717 		e_err(probe,
4718 		      "ATR is not supported while multiple "
4719 		      "queues are disabled.  Disabling Flow Director\n");
4720 	}
4721 	adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4722 	adapter->atr_sample_rate = 0;
4723 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4724 		ixgbe_disable_sriov(adapter);
4725 
4726 	err = ixgbe_set_num_queues(adapter);
4727 	if (err)
4728 		return err;
4729 
4730 	err = pci_enable_msi(adapter->pdev);
4731 	if (!err) {
4732 		adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4733 	} else {
4734 		netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4735 			     "Unable to allocate MSI interrupt, "
4736 			     "falling back to legacy.  Error: %d\n", err);
4737 		/* reset err */
4738 		err = 0;
4739 	}
4740 
4741 out:
4742 	return err;
4743 }
4744 
4745 /**
4746  * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4747  * @adapter: board private structure to initialize
4748  *
4749  * We allocate one q_vector per queue interrupt.  If allocation fails we
4750  * return -ENOMEM.
4751  **/
4752 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4753 {
4754 	int v_idx, num_q_vectors;
4755 	struct ixgbe_q_vector *q_vector;
4756 
4757 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4758 		num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4759 	else
4760 		num_q_vectors = 1;
4761 
4762 	for (v_idx = 0; v_idx < num_q_vectors; v_idx++) {
4763 		q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4764 					GFP_KERNEL, adapter->node);
4765 		if (!q_vector)
4766 			q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4767 					   GFP_KERNEL);
4768 		if (!q_vector)
4769 			goto err_out;
4770 
4771 		q_vector->adapter = adapter;
4772 		q_vector->v_idx = v_idx;
4773 
4774 		/* Allocate the affinity_hint cpumask, configure the mask */
4775 		if (!alloc_cpumask_var(&q_vector->affinity_mask, GFP_KERNEL))
4776 			goto err_out;
4777 		cpumask_set_cpu(v_idx, q_vector->affinity_mask);
4778 		netif_napi_add(adapter->netdev, &q_vector->napi,
4779 			       ixgbe_poll, 64);
4780 		adapter->q_vector[v_idx] = q_vector;
4781 	}
4782 
4783 	return 0;
4784 
4785 err_out:
4786 	while (v_idx) {
4787 		v_idx--;
4788 		q_vector = adapter->q_vector[v_idx];
4789 		netif_napi_del(&q_vector->napi);
4790 		free_cpumask_var(q_vector->affinity_mask);
4791 		kfree(q_vector);
4792 		adapter->q_vector[v_idx] = NULL;
4793 	}
4794 	return -ENOMEM;
4795 }
4796 
4797 /**
4798  * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4799  * @adapter: board private structure to initialize
4800  *
4801  * This function frees the memory allocated to the q_vectors.  In addition if
4802  * NAPI is enabled it will delete any references to the NAPI struct prior
4803  * to freeing the q_vector.
4804  **/
4805 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4806 {
4807 	int v_idx, num_q_vectors;
4808 
4809 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4810 		num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4811 	else
4812 		num_q_vectors = 1;
4813 
4814 	for (v_idx = 0; v_idx < num_q_vectors; v_idx++) {
4815 		struct ixgbe_q_vector *q_vector = adapter->q_vector[v_idx];
4816 		adapter->q_vector[v_idx] = NULL;
4817 		netif_napi_del(&q_vector->napi);
4818 		free_cpumask_var(q_vector->affinity_mask);
4819 		kfree(q_vector);
4820 	}
4821 }
4822 
4823 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4824 {
4825 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4826 		adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4827 		pci_disable_msix(adapter->pdev);
4828 		kfree(adapter->msix_entries);
4829 		adapter->msix_entries = NULL;
4830 	} else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4831 		adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4832 		pci_disable_msi(adapter->pdev);
4833 	}
4834 }
4835 
4836 /**
4837  * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4838  * @adapter: board private structure to initialize
4839  *
4840  * We determine which interrupt scheme to use based on...
4841  * - Kernel support (MSI, MSI-X)
4842  *   - which can be user-defined (via MODULE_PARAM)
4843  * - Hardware queue count (num_*_queues)
4844  *   - defined by miscellaneous hardware support/features (RSS, etc.)
4845  **/
4846 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4847 {
4848 	int err;
4849 
4850 	/* Number of supported queues */
4851 	err = ixgbe_set_num_queues(adapter);
4852 	if (err)
4853 		return err;
4854 
4855 	err = ixgbe_set_interrupt_capability(adapter);
4856 	if (err) {
4857 		e_dev_err("Unable to setup interrupt capabilities\n");
4858 		goto err_set_interrupt;
4859 	}
4860 
4861 	err = ixgbe_alloc_q_vectors(adapter);
4862 	if (err) {
4863 		e_dev_err("Unable to allocate memory for queue vectors\n");
4864 		goto err_alloc_q_vectors;
4865 	}
4866 
4867 	err = ixgbe_alloc_queues(adapter);
4868 	if (err) {
4869 		e_dev_err("Unable to allocate memory for queues\n");
4870 		goto err_alloc_queues;
4871 	}
4872 
4873 	e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4874 		   (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4875 		   adapter->num_rx_queues, adapter->num_tx_queues);
4876 
4877 	set_bit(__IXGBE_DOWN, &adapter->state);
4878 
4879 	return 0;
4880 
4881 err_alloc_queues:
4882 	ixgbe_free_q_vectors(adapter);
4883 err_alloc_q_vectors:
4884 	ixgbe_reset_interrupt_capability(adapter);
4885 err_set_interrupt:
4886 	return err;
4887 }
4888 
4889 /**
4890  * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4891  * @adapter: board private structure to clear interrupt scheme on
4892  *
4893  * We go through and clear interrupt specific resources and reset the structure
4894  * to pre-load conditions
4895  **/
4896 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4897 {
4898 	int i;
4899 
4900 	for (i = 0; i < adapter->num_tx_queues; i++) {
4901 		kfree(adapter->tx_ring[i]);
4902 		adapter->tx_ring[i] = NULL;
4903 	}
4904 	for (i = 0; i < adapter->num_rx_queues; i++) {
4905 		struct ixgbe_ring *ring = adapter->rx_ring[i];
4906 
4907 		/* ixgbe_get_stats64() might access this ring, we must wait
4908 		 * a grace period before freeing it.
4909 		 */
4910 		kfree_rcu(ring, rcu);
4911 		adapter->rx_ring[i] = NULL;
4912 	}
4913 
4914 	adapter->num_tx_queues = 0;
4915 	adapter->num_rx_queues = 0;
4916 
4917 	ixgbe_free_q_vectors(adapter);
4918 	ixgbe_reset_interrupt_capability(adapter);
4919 }
4920 
4921 /**
4922  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4923  * @adapter: board private structure to initialize
4924  *
4925  * ixgbe_sw_init initializes the Adapter private data structure.
4926  * Fields are initialized based on PCI device information and
4927  * OS network device settings (MTU size).
4928  **/
4929 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4930 {
4931 	struct ixgbe_hw *hw = &adapter->hw;
4932 	struct pci_dev *pdev = adapter->pdev;
4933 	unsigned int rss;
4934 #ifdef CONFIG_IXGBE_DCB
4935 	int j;
4936 	struct tc_configuration *tc;
4937 #endif
4938 
4939 	/* PCI config space info */
4940 
4941 	hw->vendor_id = pdev->vendor;
4942 	hw->device_id = pdev->device;
4943 	hw->revision_id = pdev->revision;
4944 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
4945 	hw->subsystem_device_id = pdev->subsystem_device;
4946 
4947 	/* Set capability flags */
4948 	rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4949 	adapter->ring_feature[RING_F_RSS].indices = rss;
4950 	adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4951 	switch (hw->mac.type) {
4952 	case ixgbe_mac_82598EB:
4953 		if (hw->device_id == IXGBE_DEV_ID_82598AT)
4954 			adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4955 		adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4956 		break;
4957 	case ixgbe_mac_X540:
4958 		adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4959 	case ixgbe_mac_82599EB:
4960 		adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4961 		adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4962 		adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4963 		if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4964 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4965 		/* Flow Director hash filters enabled */
4966 		adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4967 		adapter->atr_sample_rate = 20;
4968 		adapter->ring_feature[RING_F_FDIR].indices =
4969 							 IXGBE_MAX_FDIR_INDICES;
4970 		adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
4971 #ifdef IXGBE_FCOE
4972 		adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4973 		adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4974 		adapter->ring_feature[RING_F_FCOE].indices = 0;
4975 #ifdef CONFIG_IXGBE_DCB
4976 		/* Default traffic class to use for FCoE */
4977 		adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4978 #endif
4979 #endif /* IXGBE_FCOE */
4980 		break;
4981 	default:
4982 		break;
4983 	}
4984 
4985 	/* n-tuple support exists, always init our spinlock */
4986 	spin_lock_init(&adapter->fdir_perfect_lock);
4987 
4988 #ifdef CONFIG_IXGBE_DCB
4989 	switch (hw->mac.type) {
4990 	case ixgbe_mac_X540:
4991 		adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
4992 		adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
4993 		break;
4994 	default:
4995 		adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
4996 		adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
4997 		break;
4998 	}
4999 
5000 	/* Configure DCB traffic classes */
5001 	for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5002 		tc = &adapter->dcb_cfg.tc_config[j];
5003 		tc->path[DCB_TX_CONFIG].bwg_id = 0;
5004 		tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5005 		tc->path[DCB_RX_CONFIG].bwg_id = 0;
5006 		tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5007 		tc->dcb_pfc = pfc_disabled;
5008 	}
5009 
5010 	/* Initialize default user to priority mapping, UPx->TC0 */
5011 	tc = &adapter->dcb_cfg.tc_config[0];
5012 	tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
5013 	tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
5014 
5015 	adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5016 	adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5017 	adapter->dcb_cfg.pfc_mode_enable = false;
5018 	adapter->dcb_set_bitmap = 0x00;
5019 	adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
5020 	ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
5021 			   MAX_TRAFFIC_CLASS);
5022 
5023 #endif
5024 
5025 	/* default flow control settings */
5026 	hw->fc.requested_mode = ixgbe_fc_full;
5027 	hw->fc.current_mode = ixgbe_fc_full;	/* init for ethtool output */
5028 #ifdef CONFIG_DCB
5029 	adapter->last_lfc_mode = hw->fc.current_mode;
5030 #endif
5031 	ixgbe_pbthresh_setup(adapter);
5032 	hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5033 	hw->fc.send_xon = true;
5034 	hw->fc.disable_fc_autoneg = false;
5035 
5036 	/* enable itr by default in dynamic mode */
5037 	adapter->rx_itr_setting = 1;
5038 	adapter->tx_itr_setting = 1;
5039 
5040 	/* set defaults for eitr in MegaBytes */
5041 	adapter->eitr_low = 10;
5042 	adapter->eitr_high = 20;
5043 
5044 	/* set default ring sizes */
5045 	adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5046 	adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5047 
5048 	/* set default work limits */
5049 	adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
5050 
5051 	/* initialize eeprom parameters */
5052 	if (ixgbe_init_eeprom_params_generic(hw)) {
5053 		e_dev_err("EEPROM initialization failed\n");
5054 		return -EIO;
5055 	}
5056 
5057 	/* enable rx csum by default */
5058 	adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
5059 
5060 	/* get assigned NUMA node */
5061 	adapter->node = dev_to_node(&pdev->dev);
5062 
5063 	set_bit(__IXGBE_DOWN, &adapter->state);
5064 
5065 	return 0;
5066 }
5067 
5068 /**
5069  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5070  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
5071  *
5072  * Return 0 on success, negative on failure
5073  **/
5074 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5075 {
5076 	struct device *dev = tx_ring->dev;
5077 	int size;
5078 
5079 	size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5080 	tx_ring->tx_buffer_info = vzalloc_node(size, tx_ring->numa_node);
5081 	if (!tx_ring->tx_buffer_info)
5082 		tx_ring->tx_buffer_info = vzalloc(size);
5083 	if (!tx_ring->tx_buffer_info)
5084 		goto err;
5085 
5086 	/* round up to nearest 4K */
5087 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5088 	tx_ring->size = ALIGN(tx_ring->size, 4096);
5089 
5090 	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5091 					   &tx_ring->dma, GFP_KERNEL);
5092 	if (!tx_ring->desc)
5093 		goto err;
5094 
5095 	tx_ring->next_to_use = 0;
5096 	tx_ring->next_to_clean = 0;
5097 	return 0;
5098 
5099 err:
5100 	vfree(tx_ring->tx_buffer_info);
5101 	tx_ring->tx_buffer_info = NULL;
5102 	dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5103 	return -ENOMEM;
5104 }
5105 
5106 /**
5107  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5108  * @adapter: board private structure
5109  *
5110  * If this function returns with an error, then it's possible one or
5111  * more of the rings is populated (while the rest are not).  It is the
5112  * callers duty to clean those orphaned rings.
5113  *
5114  * Return 0 on success, negative on failure
5115  **/
5116 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5117 {
5118 	int i, err = 0;
5119 
5120 	for (i = 0; i < adapter->num_tx_queues; i++) {
5121 		err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5122 		if (!err)
5123 			continue;
5124 		e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5125 		break;
5126 	}
5127 
5128 	return err;
5129 }
5130 
5131 /**
5132  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5133  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
5134  *
5135  * Returns 0 on success, negative on failure
5136  **/
5137 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5138 {
5139 	struct device *dev = rx_ring->dev;
5140 	int size;
5141 
5142 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5143 	rx_ring->rx_buffer_info = vzalloc_node(size, rx_ring->numa_node);
5144 	if (!rx_ring->rx_buffer_info)
5145 		rx_ring->rx_buffer_info = vzalloc(size);
5146 	if (!rx_ring->rx_buffer_info)
5147 		goto err;
5148 
5149 	/* Round up to nearest 4K */
5150 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5151 	rx_ring->size = ALIGN(rx_ring->size, 4096);
5152 
5153 	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5154 					   &rx_ring->dma, GFP_KERNEL);
5155 
5156 	if (!rx_ring->desc)
5157 		goto err;
5158 
5159 	rx_ring->next_to_clean = 0;
5160 	rx_ring->next_to_use = 0;
5161 
5162 	return 0;
5163 err:
5164 	vfree(rx_ring->rx_buffer_info);
5165 	rx_ring->rx_buffer_info = NULL;
5166 	dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5167 	return -ENOMEM;
5168 }
5169 
5170 /**
5171  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5172  * @adapter: board private structure
5173  *
5174  * If this function returns with an error, then it's possible one or
5175  * more of the rings is populated (while the rest are not).  It is the
5176  * callers duty to clean those orphaned rings.
5177  *
5178  * Return 0 on success, negative on failure
5179  **/
5180 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5181 {
5182 	int i, err = 0;
5183 
5184 	for (i = 0; i < adapter->num_rx_queues; i++) {
5185 		err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5186 		if (!err)
5187 			continue;
5188 		e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5189 		break;
5190 	}
5191 
5192 	return err;
5193 }
5194 
5195 /**
5196  * ixgbe_free_tx_resources - Free Tx Resources per Queue
5197  * @tx_ring: Tx descriptor ring for a specific queue
5198  *
5199  * Free all transmit software resources
5200  **/
5201 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5202 {
5203 	ixgbe_clean_tx_ring(tx_ring);
5204 
5205 	vfree(tx_ring->tx_buffer_info);
5206 	tx_ring->tx_buffer_info = NULL;
5207 
5208 	/* if not set, then don't free */
5209 	if (!tx_ring->desc)
5210 		return;
5211 
5212 	dma_free_coherent(tx_ring->dev, tx_ring->size,
5213 			  tx_ring->desc, tx_ring->dma);
5214 
5215 	tx_ring->desc = NULL;
5216 }
5217 
5218 /**
5219  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5220  * @adapter: board private structure
5221  *
5222  * Free all transmit software resources
5223  **/
5224 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5225 {
5226 	int i;
5227 
5228 	for (i = 0; i < adapter->num_tx_queues; i++)
5229 		if (adapter->tx_ring[i]->desc)
5230 			ixgbe_free_tx_resources(adapter->tx_ring[i]);
5231 }
5232 
5233 /**
5234  * ixgbe_free_rx_resources - Free Rx Resources
5235  * @rx_ring: ring to clean the resources from
5236  *
5237  * Free all receive software resources
5238  **/
5239 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5240 {
5241 	ixgbe_clean_rx_ring(rx_ring);
5242 
5243 	vfree(rx_ring->rx_buffer_info);
5244 	rx_ring->rx_buffer_info = NULL;
5245 
5246 	/* if not set, then don't free */
5247 	if (!rx_ring->desc)
5248 		return;
5249 
5250 	dma_free_coherent(rx_ring->dev, rx_ring->size,
5251 			  rx_ring->desc, rx_ring->dma);
5252 
5253 	rx_ring->desc = NULL;
5254 }
5255 
5256 /**
5257  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5258  * @adapter: board private structure
5259  *
5260  * Free all receive software resources
5261  **/
5262 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5263 {
5264 	int i;
5265 
5266 	for (i = 0; i < adapter->num_rx_queues; i++)
5267 		if (adapter->rx_ring[i]->desc)
5268 			ixgbe_free_rx_resources(adapter->rx_ring[i]);
5269 }
5270 
5271 /**
5272  * ixgbe_change_mtu - Change the Maximum Transfer Unit
5273  * @netdev: network interface device structure
5274  * @new_mtu: new value for maximum frame size
5275  *
5276  * Returns 0 on success, negative on failure
5277  **/
5278 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5279 {
5280 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
5281 	struct ixgbe_hw *hw = &adapter->hw;
5282 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5283 
5284 	/* MTU < 68 is an error and causes problems on some kernels */
5285 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED &&
5286 	    hw->mac.type != ixgbe_mac_X540) {
5287 		if ((new_mtu < 68) || (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
5288 			return -EINVAL;
5289 	} else {
5290 		if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5291 			return -EINVAL;
5292 	}
5293 
5294 	e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5295 	/* must set new MTU before calling down or up */
5296 	netdev->mtu = new_mtu;
5297 
5298 	if (netif_running(netdev))
5299 		ixgbe_reinit_locked(adapter);
5300 
5301 	return 0;
5302 }
5303 
5304 /**
5305  * ixgbe_open - Called when a network interface is made active
5306  * @netdev: network interface device structure
5307  *
5308  * Returns 0 on success, negative value on failure
5309  *
5310  * The open entry point is called when a network interface is made
5311  * active by the system (IFF_UP).  At this point all resources needed
5312  * for transmit and receive operations are allocated, the interrupt
5313  * handler is registered with the OS, the watchdog timer is started,
5314  * and the stack is notified that the interface is ready.
5315  **/
5316 static int ixgbe_open(struct net_device *netdev)
5317 {
5318 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
5319 	int err;
5320 
5321 	/* disallow open during test */
5322 	if (test_bit(__IXGBE_TESTING, &adapter->state))
5323 		return -EBUSY;
5324 
5325 	netif_carrier_off(netdev);
5326 
5327 	/* allocate transmit descriptors */
5328 	err = ixgbe_setup_all_tx_resources(adapter);
5329 	if (err)
5330 		goto err_setup_tx;
5331 
5332 	/* allocate receive descriptors */
5333 	err = ixgbe_setup_all_rx_resources(adapter);
5334 	if (err)
5335 		goto err_setup_rx;
5336 
5337 	ixgbe_configure(adapter);
5338 
5339 	err = ixgbe_request_irq(adapter);
5340 	if (err)
5341 		goto err_req_irq;
5342 
5343 	ixgbe_up_complete(adapter);
5344 
5345 	return 0;
5346 
5347 err_req_irq:
5348 err_setup_rx:
5349 	ixgbe_free_all_rx_resources(adapter);
5350 err_setup_tx:
5351 	ixgbe_free_all_tx_resources(adapter);
5352 	ixgbe_reset(adapter);
5353 
5354 	return err;
5355 }
5356 
5357 /**
5358  * ixgbe_close - Disables a network interface
5359  * @netdev: network interface device structure
5360  *
5361  * Returns 0, this is not allowed to fail
5362  *
5363  * The close entry point is called when an interface is de-activated
5364  * by the OS.  The hardware is still under the drivers control, but
5365  * needs to be disabled.  A global MAC reset is issued to stop the
5366  * hardware, and all transmit and receive resources are freed.
5367  **/
5368 static int ixgbe_close(struct net_device *netdev)
5369 {
5370 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
5371 
5372 	ixgbe_down(adapter);
5373 	ixgbe_free_irq(adapter);
5374 
5375 	ixgbe_fdir_filter_exit(adapter);
5376 
5377 	ixgbe_free_all_tx_resources(adapter);
5378 	ixgbe_free_all_rx_resources(adapter);
5379 
5380 	ixgbe_release_hw_control(adapter);
5381 
5382 	return 0;
5383 }
5384 
5385 #ifdef CONFIG_PM
5386 static int ixgbe_resume(struct pci_dev *pdev)
5387 {
5388 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5389 	struct net_device *netdev = adapter->netdev;
5390 	u32 err;
5391 
5392 	pci_set_power_state(pdev, PCI_D0);
5393 	pci_restore_state(pdev);
5394 	/*
5395 	 * pci_restore_state clears dev->state_saved so call
5396 	 * pci_save_state to restore it.
5397 	 */
5398 	pci_save_state(pdev);
5399 
5400 	err = pci_enable_device_mem(pdev);
5401 	if (err) {
5402 		e_dev_err("Cannot enable PCI device from suspend\n");
5403 		return err;
5404 	}
5405 	pci_set_master(pdev);
5406 
5407 	pci_wake_from_d3(pdev, false);
5408 
5409 	err = ixgbe_init_interrupt_scheme(adapter);
5410 	if (err) {
5411 		e_dev_err("Cannot initialize interrupts for device\n");
5412 		return err;
5413 	}
5414 
5415 	ixgbe_reset(adapter);
5416 
5417 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5418 
5419 	if (netif_running(netdev)) {
5420 		err = ixgbe_open(netdev);
5421 		if (err)
5422 			return err;
5423 	}
5424 
5425 	netif_device_attach(netdev);
5426 
5427 	return 0;
5428 }
5429 #endif /* CONFIG_PM */
5430 
5431 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5432 {
5433 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5434 	struct net_device *netdev = adapter->netdev;
5435 	struct ixgbe_hw *hw = &adapter->hw;
5436 	u32 ctrl, fctrl;
5437 	u32 wufc = adapter->wol;
5438 #ifdef CONFIG_PM
5439 	int retval = 0;
5440 #endif
5441 
5442 	netif_device_detach(netdev);
5443 
5444 	if (netif_running(netdev)) {
5445 		ixgbe_down(adapter);
5446 		ixgbe_free_irq(adapter);
5447 		ixgbe_free_all_tx_resources(adapter);
5448 		ixgbe_free_all_rx_resources(adapter);
5449 	}
5450 
5451 	ixgbe_clear_interrupt_scheme(adapter);
5452 #ifdef CONFIG_DCB
5453 	kfree(adapter->ixgbe_ieee_pfc);
5454 	kfree(adapter->ixgbe_ieee_ets);
5455 #endif
5456 
5457 #ifdef CONFIG_PM
5458 	retval = pci_save_state(pdev);
5459 	if (retval)
5460 		return retval;
5461 
5462 #endif
5463 	if (wufc) {
5464 		ixgbe_set_rx_mode(netdev);
5465 
5466 		/* turn on all-multi mode if wake on multicast is enabled */
5467 		if (wufc & IXGBE_WUFC_MC) {
5468 			fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5469 			fctrl |= IXGBE_FCTRL_MPE;
5470 			IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5471 		}
5472 
5473 		ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5474 		ctrl |= IXGBE_CTRL_GIO_DIS;
5475 		IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5476 
5477 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5478 	} else {
5479 		IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5480 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5481 	}
5482 
5483 	switch (hw->mac.type) {
5484 	case ixgbe_mac_82598EB:
5485 		pci_wake_from_d3(pdev, false);
5486 		break;
5487 	case ixgbe_mac_82599EB:
5488 	case ixgbe_mac_X540:
5489 		pci_wake_from_d3(pdev, !!wufc);
5490 		break;
5491 	default:
5492 		break;
5493 	}
5494 
5495 	*enable_wake = !!wufc;
5496 
5497 	ixgbe_release_hw_control(adapter);
5498 
5499 	pci_disable_device(pdev);
5500 
5501 	return 0;
5502 }
5503 
5504 #ifdef CONFIG_PM
5505 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5506 {
5507 	int retval;
5508 	bool wake;
5509 
5510 	retval = __ixgbe_shutdown(pdev, &wake);
5511 	if (retval)
5512 		return retval;
5513 
5514 	if (wake) {
5515 		pci_prepare_to_sleep(pdev);
5516 	} else {
5517 		pci_wake_from_d3(pdev, false);
5518 		pci_set_power_state(pdev, PCI_D3hot);
5519 	}
5520 
5521 	return 0;
5522 }
5523 #endif /* CONFIG_PM */
5524 
5525 static void ixgbe_shutdown(struct pci_dev *pdev)
5526 {
5527 	bool wake;
5528 
5529 	__ixgbe_shutdown(pdev, &wake);
5530 
5531 	if (system_state == SYSTEM_POWER_OFF) {
5532 		pci_wake_from_d3(pdev, wake);
5533 		pci_set_power_state(pdev, PCI_D3hot);
5534 	}
5535 }
5536 
5537 /**
5538  * ixgbe_update_stats - Update the board statistics counters.
5539  * @adapter: board private structure
5540  **/
5541 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5542 {
5543 	struct net_device *netdev = adapter->netdev;
5544 	struct ixgbe_hw *hw = &adapter->hw;
5545 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
5546 	u64 total_mpc = 0;
5547 	u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5548 	u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5549 	u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5550 	u64 bytes = 0, packets = 0;
5551 #ifdef IXGBE_FCOE
5552 	struct ixgbe_fcoe *fcoe = &adapter->fcoe;
5553 	unsigned int cpu;
5554 	u64 fcoe_noddp_counts_sum = 0, fcoe_noddp_ext_buff_counts_sum = 0;
5555 #endif /* IXGBE_FCOE */
5556 
5557 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5558 	    test_bit(__IXGBE_RESETTING, &adapter->state))
5559 		return;
5560 
5561 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5562 		u64 rsc_count = 0;
5563 		u64 rsc_flush = 0;
5564 		for (i = 0; i < 16; i++)
5565 			adapter->hw_rx_no_dma_resources +=
5566 				IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5567 		for (i = 0; i < adapter->num_rx_queues; i++) {
5568 			rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5569 			rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5570 		}
5571 		adapter->rsc_total_count = rsc_count;
5572 		adapter->rsc_total_flush = rsc_flush;
5573 	}
5574 
5575 	for (i = 0; i < adapter->num_rx_queues; i++) {
5576 		struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5577 		non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5578 		alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5579 		alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5580 		bytes += rx_ring->stats.bytes;
5581 		packets += rx_ring->stats.packets;
5582 	}
5583 	adapter->non_eop_descs = non_eop_descs;
5584 	adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5585 	adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5586 	netdev->stats.rx_bytes = bytes;
5587 	netdev->stats.rx_packets = packets;
5588 
5589 	bytes = 0;
5590 	packets = 0;
5591 	/* gather some stats to the adapter struct that are per queue */
5592 	for (i = 0; i < adapter->num_tx_queues; i++) {
5593 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5594 		restart_queue += tx_ring->tx_stats.restart_queue;
5595 		tx_busy += tx_ring->tx_stats.tx_busy;
5596 		bytes += tx_ring->stats.bytes;
5597 		packets += tx_ring->stats.packets;
5598 	}
5599 	adapter->restart_queue = restart_queue;
5600 	adapter->tx_busy = tx_busy;
5601 	netdev->stats.tx_bytes = bytes;
5602 	netdev->stats.tx_packets = packets;
5603 
5604 	hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5605 
5606 	/* 8 register reads */
5607 	for (i = 0; i < 8; i++) {
5608 		/* for packet buffers not used, the register should read 0 */
5609 		mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5610 		missed_rx += mpc;
5611 		hwstats->mpc[i] += mpc;
5612 		total_mpc += hwstats->mpc[i];
5613 		hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5614 		hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5615 		switch (hw->mac.type) {
5616 		case ixgbe_mac_82598EB:
5617 			hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5618 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5619 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5620 			hwstats->pxonrxc[i] +=
5621 				IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5622 			break;
5623 		case ixgbe_mac_82599EB:
5624 		case ixgbe_mac_X540:
5625 			hwstats->pxonrxc[i] +=
5626 				IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5627 			break;
5628 		default:
5629 			break;
5630 		}
5631 	}
5632 
5633 	/*16 register reads */
5634 	for (i = 0; i < 16; i++) {
5635 		hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5636 		hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5637 		if ((hw->mac.type == ixgbe_mac_82599EB) ||
5638 		    (hw->mac.type == ixgbe_mac_X540)) {
5639 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
5640 			IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
5641 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
5642 			IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
5643 		}
5644 	}
5645 
5646 	hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5647 	/* work around hardware counting issue */
5648 	hwstats->gprc -= missed_rx;
5649 
5650 	ixgbe_update_xoff_received(adapter);
5651 
5652 	/* 82598 hardware only has a 32 bit counter in the high register */
5653 	switch (hw->mac.type) {
5654 	case ixgbe_mac_82598EB:
5655 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5656 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5657 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5658 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5659 		break;
5660 	case ixgbe_mac_X540:
5661 		/* OS2BMC stats are X540 only*/
5662 		hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5663 		hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5664 		hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5665 		hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5666 	case ixgbe_mac_82599EB:
5667 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5668 		IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5669 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5670 		IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5671 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5672 		IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5673 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5674 		hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5675 		hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5676 #ifdef IXGBE_FCOE
5677 		hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5678 		hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5679 		hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5680 		hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5681 		hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5682 		hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5683 		/* Add up per cpu counters for total ddp aloc fail */
5684 		if (fcoe->pcpu_noddp && fcoe->pcpu_noddp_ext_buff) {
5685 			for_each_possible_cpu(cpu) {
5686 				fcoe_noddp_counts_sum +=
5687 					*per_cpu_ptr(fcoe->pcpu_noddp, cpu);
5688 				fcoe_noddp_ext_buff_counts_sum +=
5689 					*per_cpu_ptr(fcoe->
5690 						pcpu_noddp_ext_buff, cpu);
5691 			}
5692 		}
5693 		hwstats->fcoe_noddp = fcoe_noddp_counts_sum;
5694 		hwstats->fcoe_noddp_ext_buff = fcoe_noddp_ext_buff_counts_sum;
5695 #endif /* IXGBE_FCOE */
5696 		break;
5697 	default:
5698 		break;
5699 	}
5700 	bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5701 	hwstats->bprc += bprc;
5702 	hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5703 	if (hw->mac.type == ixgbe_mac_82598EB)
5704 		hwstats->mprc -= bprc;
5705 	hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5706 	hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5707 	hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5708 	hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5709 	hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5710 	hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5711 	hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5712 	hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5713 	lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5714 	hwstats->lxontxc += lxon;
5715 	lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5716 	hwstats->lxofftxc += lxoff;
5717 	hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5718 	hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5719 	/*
5720 	 * 82598 errata - tx of flow control packets is included in tx counters
5721 	 */
5722 	xon_off_tot = lxon + lxoff;
5723 	hwstats->gptc -= xon_off_tot;
5724 	hwstats->mptc -= xon_off_tot;
5725 	hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5726 	hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5727 	hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5728 	hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5729 	hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5730 	hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5731 	hwstats->ptc64 -= xon_off_tot;
5732 	hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5733 	hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5734 	hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5735 	hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5736 	hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5737 	hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5738 
5739 	/* Fill out the OS statistics structure */
5740 	netdev->stats.multicast = hwstats->mprc;
5741 
5742 	/* Rx Errors */
5743 	netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5744 	netdev->stats.rx_dropped = 0;
5745 	netdev->stats.rx_length_errors = hwstats->rlec;
5746 	netdev->stats.rx_crc_errors = hwstats->crcerrs;
5747 	netdev->stats.rx_missed_errors = total_mpc;
5748 }
5749 
5750 /**
5751  * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5752  * @adapter - pointer to the device adapter structure
5753  **/
5754 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5755 {
5756 	struct ixgbe_hw *hw = &adapter->hw;
5757 	int i;
5758 
5759 	if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5760 		return;
5761 
5762 	adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5763 
5764 	/* if interface is down do nothing */
5765 	if (test_bit(__IXGBE_DOWN, &adapter->state))
5766 		return;
5767 
5768 	/* do nothing if we are not using signature filters */
5769 	if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5770 		return;
5771 
5772 	adapter->fdir_overflow++;
5773 
5774 	if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5775 		for (i = 0; i < adapter->num_tx_queues; i++)
5776 			set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5777 			        &(adapter->tx_ring[i]->state));
5778 		/* re-enable flow director interrupts */
5779 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5780 	} else {
5781 		e_err(probe, "failed to finish FDIR re-initialization, "
5782 		      "ignored adding FDIR ATR filters\n");
5783 	}
5784 }
5785 
5786 /**
5787  * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5788  * @adapter - pointer to the device adapter structure
5789  *
5790  * This function serves two purposes.  First it strobes the interrupt lines
5791  * in order to make certain interrupts are occuring.  Secondly it sets the
5792  * bits needed to check for TX hangs.  As a result we should immediately
5793  * determine if a hang has occured.
5794  */
5795 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5796 {
5797 	struct ixgbe_hw *hw = &adapter->hw;
5798 	u64 eics = 0;
5799 	int i;
5800 
5801 	/* If we're down or resetting, just bail */
5802 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5803 	    test_bit(__IXGBE_RESETTING, &adapter->state))
5804 		return;
5805 
5806 	/* Force detection of hung controller */
5807 	if (netif_carrier_ok(adapter->netdev)) {
5808 		for (i = 0; i < adapter->num_tx_queues; i++)
5809 			set_check_for_tx_hang(adapter->tx_ring[i]);
5810 	}
5811 
5812 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5813 		/*
5814 		 * for legacy and MSI interrupts don't set any bits
5815 		 * that are enabled for EIAM, because this operation
5816 		 * would set *both* EIMS and EICS for any bit in EIAM
5817 		 */
5818 		IXGBE_WRITE_REG(hw, IXGBE_EICS,
5819 			(IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5820 	} else {
5821 		/* get one bit for every active tx/rx interrupt vector */
5822 		for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5823 			struct ixgbe_q_vector *qv = adapter->q_vector[i];
5824 			if (qv->rx.ring || qv->tx.ring)
5825 				eics |= ((u64)1 << i);
5826 		}
5827 	}
5828 
5829 	/* Cause software interrupt to ensure rings are cleaned */
5830 	ixgbe_irq_rearm_queues(adapter, eics);
5831 
5832 }
5833 
5834 /**
5835  * ixgbe_watchdog_update_link - update the link status
5836  * @adapter - pointer to the device adapter structure
5837  * @link_speed - pointer to a u32 to store the link_speed
5838  **/
5839 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
5840 {
5841 	struct ixgbe_hw *hw = &adapter->hw;
5842 	u32 link_speed = adapter->link_speed;
5843 	bool link_up = adapter->link_up;
5844 	int i;
5845 
5846 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5847 		return;
5848 
5849 	if (hw->mac.ops.check_link) {
5850 		hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5851 	} else {
5852 		/* always assume link is up, if no check link function */
5853 		link_speed = IXGBE_LINK_SPEED_10GB_FULL;
5854 		link_up = true;
5855 	}
5856 	if (link_up) {
5857 		if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5858 			for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5859 				hw->mac.ops.fc_enable(hw, i);
5860 		} else {
5861 			hw->mac.ops.fc_enable(hw, 0);
5862 		}
5863 	}
5864 
5865 	if (link_up ||
5866 	    time_after(jiffies, (adapter->link_check_timeout +
5867 				 IXGBE_TRY_LINK_TIMEOUT))) {
5868 		adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5869 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5870 		IXGBE_WRITE_FLUSH(hw);
5871 	}
5872 
5873 	adapter->link_up = link_up;
5874 	adapter->link_speed = link_speed;
5875 }
5876 
5877 /**
5878  * ixgbe_watchdog_link_is_up - update netif_carrier status and
5879  *                             print link up message
5880  * @adapter - pointer to the device adapter structure
5881  **/
5882 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
5883 {
5884 	struct net_device *netdev = adapter->netdev;
5885 	struct ixgbe_hw *hw = &adapter->hw;
5886 	u32 link_speed = adapter->link_speed;
5887 	bool flow_rx, flow_tx;
5888 
5889 	/* only continue if link was previously down */
5890 	if (netif_carrier_ok(netdev))
5891 		return;
5892 
5893 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
5894 
5895 	switch (hw->mac.type) {
5896 	case ixgbe_mac_82598EB: {
5897 		u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5898 		u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5899 		flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5900 		flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5901 	}
5902 		break;
5903 	case ixgbe_mac_X540:
5904 	case ixgbe_mac_82599EB: {
5905 		u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5906 		u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5907 		flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5908 		flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5909 	}
5910 		break;
5911 	default:
5912 		flow_tx = false;
5913 		flow_rx = false;
5914 		break;
5915 	}
5916 	e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5917 	       (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5918 	       "10 Gbps" :
5919 	       (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5920 	       "1 Gbps" :
5921 	       (link_speed == IXGBE_LINK_SPEED_100_FULL ?
5922 	       "100 Mbps" :
5923 	       "unknown speed"))),
5924 	       ((flow_rx && flow_tx) ? "RX/TX" :
5925 	       (flow_rx ? "RX" :
5926 	       (flow_tx ? "TX" : "None"))));
5927 
5928 	netif_carrier_on(netdev);
5929 	ixgbe_check_vf_rate_limit(adapter);
5930 }
5931 
5932 /**
5933  * ixgbe_watchdog_link_is_down - update netif_carrier status and
5934  *                               print link down message
5935  * @adapter - pointer to the adapter structure
5936  **/
5937 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter* adapter)
5938 {
5939 	struct net_device *netdev = adapter->netdev;
5940 	struct ixgbe_hw *hw = &adapter->hw;
5941 
5942 	adapter->link_up = false;
5943 	adapter->link_speed = 0;
5944 
5945 	/* only continue if link was up previously */
5946 	if (!netif_carrier_ok(netdev))
5947 		return;
5948 
5949 	/* poll for SFP+ cable when link is down */
5950 	if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
5951 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5952 
5953 	e_info(drv, "NIC Link is Down\n");
5954 	netif_carrier_off(netdev);
5955 }
5956 
5957 /**
5958  * ixgbe_watchdog_flush_tx - flush queues on link down
5959  * @adapter - pointer to the device adapter structure
5960  **/
5961 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
5962 {
5963 	int i;
5964 	int some_tx_pending = 0;
5965 
5966 	if (!netif_carrier_ok(adapter->netdev)) {
5967 		for (i = 0; i < adapter->num_tx_queues; i++) {
5968 			struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5969 			if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5970 				some_tx_pending = 1;
5971 				break;
5972 			}
5973 		}
5974 
5975 		if (some_tx_pending) {
5976 			/* We've lost link, so the controller stops DMA,
5977 			 * but we've got queued Tx work that's never going
5978 			 * to get done, so reset controller to flush Tx.
5979 			 * (Do the reset outside of interrupt context).
5980 			 */
5981 			adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
5982 		}
5983 	}
5984 }
5985 
5986 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
5987 {
5988 	u32 ssvpc;
5989 
5990 	/* Do not perform spoof check for 82598 */
5991 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5992 		return;
5993 
5994 	ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
5995 
5996 	/*
5997 	 * ssvpc register is cleared on read, if zero then no
5998 	 * spoofed packets in the last interval.
5999 	 */
6000 	if (!ssvpc)
6001 		return;
6002 
6003 	e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
6004 }
6005 
6006 /**
6007  * ixgbe_watchdog_subtask - check and bring link up
6008  * @adapter - pointer to the device adapter structure
6009  **/
6010 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
6011 {
6012 	/* if interface is down do nothing */
6013 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6014 	    test_bit(__IXGBE_RESETTING, &adapter->state))
6015 		return;
6016 
6017 	ixgbe_watchdog_update_link(adapter);
6018 
6019 	if (adapter->link_up)
6020 		ixgbe_watchdog_link_is_up(adapter);
6021 	else
6022 		ixgbe_watchdog_link_is_down(adapter);
6023 
6024 	ixgbe_spoof_check(adapter);
6025 	ixgbe_update_stats(adapter);
6026 
6027 	ixgbe_watchdog_flush_tx(adapter);
6028 }
6029 
6030 /**
6031  * ixgbe_sfp_detection_subtask - poll for SFP+ cable
6032  * @adapter - the ixgbe adapter structure
6033  **/
6034 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
6035 {
6036 	struct ixgbe_hw *hw = &adapter->hw;
6037 	s32 err;
6038 
6039 	/* not searching for SFP so there is nothing to do here */
6040 	if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
6041 	    !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6042 		return;
6043 
6044 	/* someone else is in init, wait until next service event */
6045 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6046 		return;
6047 
6048 	err = hw->phy.ops.identify_sfp(hw);
6049 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6050 		goto sfp_out;
6051 
6052 	if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
6053 		/* If no cable is present, then we need to reset
6054 		 * the next time we find a good cable. */
6055 		adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
6056 	}
6057 
6058 	/* exit on error */
6059 	if (err)
6060 		goto sfp_out;
6061 
6062 	/* exit if reset not needed */
6063 	if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6064 		goto sfp_out;
6065 
6066 	adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
6067 
6068 	/*
6069 	 * A module may be identified correctly, but the EEPROM may not have
6070 	 * support for that module.  setup_sfp() will fail in that case, so
6071 	 * we should not allow that module to load.
6072 	 */
6073 	if (hw->mac.type == ixgbe_mac_82598EB)
6074 		err = hw->phy.ops.reset(hw);
6075 	else
6076 		err = hw->mac.ops.setup_sfp(hw);
6077 
6078 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6079 		goto sfp_out;
6080 
6081 	adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
6082 	e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
6083 
6084 sfp_out:
6085 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6086 
6087 	if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
6088 	    (adapter->netdev->reg_state == NETREG_REGISTERED)) {
6089 		e_dev_err("failed to initialize because an unsupported "
6090 			  "SFP+ module type was detected.\n");
6091 		e_dev_err("Reload the driver after installing a "
6092 			  "supported module.\n");
6093 		unregister_netdev(adapter->netdev);
6094 	}
6095 }
6096 
6097 /**
6098  * ixgbe_sfp_link_config_subtask - set up link SFP after module install
6099  * @adapter - the ixgbe adapter structure
6100  **/
6101 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
6102 {
6103 	struct ixgbe_hw *hw = &adapter->hw;
6104 	u32 autoneg;
6105 	bool negotiation;
6106 
6107 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
6108 		return;
6109 
6110 	/* someone else is in init, wait until next service event */
6111 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6112 		return;
6113 
6114 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
6115 
6116 	autoneg = hw->phy.autoneg_advertised;
6117 	if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
6118 		hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
6119 	if (hw->mac.ops.setup_link)
6120 		hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
6121 
6122 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
6123 	adapter->link_check_timeout = jiffies;
6124 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6125 }
6126 
6127 #ifdef CONFIG_PCI_IOV
6128 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
6129 {
6130 	int vf;
6131 	struct ixgbe_hw *hw = &adapter->hw;
6132 	struct net_device *netdev = adapter->netdev;
6133 	u32 gpc;
6134 	u32 ciaa, ciad;
6135 
6136 	gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
6137 	if (gpc) /* If incrementing then no need for the check below */
6138 		return;
6139 	/*
6140 	 * Check to see if a bad DMA write target from an errant or
6141 	 * malicious VF has caused a PCIe error.  If so then we can
6142 	 * issue a VFLR to the offending VF(s) and then resume without
6143 	 * requesting a full slot reset.
6144 	 */
6145 
6146 	for (vf = 0; vf < adapter->num_vfs; vf++) {
6147 		ciaa = (vf << 16) | 0x80000000;
6148 		/* 32 bit read so align, we really want status at offset 6 */
6149 		ciaa |= PCI_COMMAND;
6150 		IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6151 		ciad = IXGBE_READ_REG(hw, IXGBE_CIAD_82599);
6152 		ciaa &= 0x7FFFFFFF;
6153 		/* disable debug mode asap after reading data */
6154 		IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6155 		/* Get the upper 16 bits which will be the PCI status reg */
6156 		ciad >>= 16;
6157 		if (ciad & PCI_STATUS_REC_MASTER_ABORT) {
6158 			netdev_err(netdev, "VF %d Hung DMA\n", vf);
6159 			/* Issue VFLR */
6160 			ciaa = (vf << 16) | 0x80000000;
6161 			ciaa |= 0xA8;
6162 			IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6163 			ciad = 0x00008000;  /* VFLR */
6164 			IXGBE_WRITE_REG(hw, IXGBE_CIAD_82599, ciad);
6165 			ciaa &= 0x7FFFFFFF;
6166 			IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6167 		}
6168 	}
6169 }
6170 
6171 #endif
6172 /**
6173  * ixgbe_service_timer - Timer Call-back
6174  * @data: pointer to adapter cast into an unsigned long
6175  **/
6176 static void ixgbe_service_timer(unsigned long data)
6177 {
6178 	struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
6179 	unsigned long next_event_offset;
6180 	bool ready = true;
6181 
6182 #ifdef CONFIG_PCI_IOV
6183 	ready = false;
6184 
6185 	/*
6186 	 * don't bother with SR-IOV VF DMA hang check if there are
6187 	 * no VFs or the link is down
6188 	 */
6189 	if (!adapter->num_vfs ||
6190 	    (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)) {
6191 		ready = true;
6192 		goto normal_timer_service;
6193 	}
6194 
6195 	/* If we have VFs allocated then we must check for DMA hangs */
6196 	ixgbe_check_for_bad_vf(adapter);
6197 	next_event_offset = HZ / 50;
6198 	adapter->timer_event_accumulator++;
6199 
6200 	if (adapter->timer_event_accumulator >= 100) {
6201 		ready = true;
6202 		adapter->timer_event_accumulator = 0;
6203 	}
6204 
6205 	goto schedule_event;
6206 
6207 normal_timer_service:
6208 #endif
6209 	/* poll faster when waiting for link */
6210 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
6211 		next_event_offset = HZ / 10;
6212 	else
6213 		next_event_offset = HZ * 2;
6214 
6215 #ifdef CONFIG_PCI_IOV
6216 schedule_event:
6217 #endif
6218 	/* Reset the timer */
6219 	mod_timer(&adapter->service_timer, next_event_offset + jiffies);
6220 
6221 	if (ready)
6222 		ixgbe_service_event_schedule(adapter);
6223 }
6224 
6225 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
6226 {
6227 	if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
6228 		return;
6229 
6230 	adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
6231 
6232 	/* If we're already down or resetting, just bail */
6233 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6234 	    test_bit(__IXGBE_RESETTING, &adapter->state))
6235 		return;
6236 
6237 	ixgbe_dump(adapter);
6238 	netdev_err(adapter->netdev, "Reset adapter\n");
6239 	adapter->tx_timeout_count++;
6240 
6241 	ixgbe_reinit_locked(adapter);
6242 }
6243 
6244 /**
6245  * ixgbe_service_task - manages and runs subtasks
6246  * @work: pointer to work_struct containing our data
6247  **/
6248 static void ixgbe_service_task(struct work_struct *work)
6249 {
6250 	struct ixgbe_adapter *adapter = container_of(work,
6251 						     struct ixgbe_adapter,
6252 						     service_task);
6253 
6254 	ixgbe_reset_subtask(adapter);
6255 	ixgbe_sfp_detection_subtask(adapter);
6256 	ixgbe_sfp_link_config_subtask(adapter);
6257 	ixgbe_check_overtemp_subtask(adapter);
6258 	ixgbe_watchdog_subtask(adapter);
6259 	ixgbe_fdir_reinit_subtask(adapter);
6260 	ixgbe_check_hang_subtask(adapter);
6261 
6262 	ixgbe_service_event_complete(adapter);
6263 }
6264 
6265 void ixgbe_tx_ctxtdesc(struct ixgbe_ring *tx_ring, u32 vlan_macip_lens,
6266 		       u32 fcoe_sof_eof, u32 type_tucmd, u32 mss_l4len_idx)
6267 {
6268 	struct ixgbe_adv_tx_context_desc *context_desc;
6269 	u16 i = tx_ring->next_to_use;
6270 
6271 	context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6272 
6273 	i++;
6274 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
6275 
6276 	/* set bits to identify this as an advanced context descriptor */
6277 	type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
6278 
6279 	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
6280 	context_desc->seqnum_seed	= cpu_to_le32(fcoe_sof_eof);
6281 	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
6282 	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
6283 }
6284 
6285 static int ixgbe_tso(struct ixgbe_ring *tx_ring, struct sk_buff *skb,
6286 		     u32 tx_flags, __be16 protocol, u8 *hdr_len)
6287 {
6288 	int err;
6289 	u32 vlan_macip_lens, type_tucmd;
6290 	u32 mss_l4len_idx, l4len;
6291 
6292 	if (!skb_is_gso(skb))
6293 		return 0;
6294 
6295 	if (skb_header_cloned(skb)) {
6296 		err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
6297 		if (err)
6298 			return err;
6299 	}
6300 
6301 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6302 	type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
6303 
6304 	if (protocol == __constant_htons(ETH_P_IP)) {
6305 		struct iphdr *iph = ip_hdr(skb);
6306 		iph->tot_len = 0;
6307 		iph->check = 0;
6308 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6309 							 iph->daddr, 0,
6310 							 IPPROTO_TCP,
6311 							 0);
6312 		type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6313 	} else if (skb_is_gso_v6(skb)) {
6314 		ipv6_hdr(skb)->payload_len = 0;
6315 		tcp_hdr(skb)->check =
6316 		    ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6317 				     &ipv6_hdr(skb)->daddr,
6318 				     0, IPPROTO_TCP, 0);
6319 	}
6320 
6321 	l4len = tcp_hdrlen(skb);
6322 	*hdr_len = skb_transport_offset(skb) + l4len;
6323 
6324 	/* mss_l4len_id: use 1 as index for TSO */
6325 	mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
6326 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
6327 	mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
6328 
6329 	/* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
6330 	vlan_macip_lens = skb_network_header_len(skb);
6331 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6332 	vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6333 
6334 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
6335 	                  mss_l4len_idx);
6336 
6337 	return 1;
6338 }
6339 
6340 static bool ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
6341 			  struct sk_buff *skb, u32 tx_flags,
6342 			  __be16 protocol)
6343 {
6344 	u32 vlan_macip_lens = 0;
6345 	u32 mss_l4len_idx = 0;
6346 	u32 type_tucmd = 0;
6347 
6348 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
6349 	    if (!(tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
6350 		!(tx_flags & IXGBE_TX_FLAGS_TXSW))
6351 			return false;
6352 	} else {
6353 		u8 l4_hdr = 0;
6354 		switch (protocol) {
6355 		case __constant_htons(ETH_P_IP):
6356 			vlan_macip_lens |= skb_network_header_len(skb);
6357 			type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6358 			l4_hdr = ip_hdr(skb)->protocol;
6359 			break;
6360 		case __constant_htons(ETH_P_IPV6):
6361 			vlan_macip_lens |= skb_network_header_len(skb);
6362 			l4_hdr = ipv6_hdr(skb)->nexthdr;
6363 			break;
6364 		default:
6365 			if (unlikely(net_ratelimit())) {
6366 				dev_warn(tx_ring->dev,
6367 				 "partial checksum but proto=%x!\n",
6368 				 skb->protocol);
6369 			}
6370 			break;
6371 		}
6372 
6373 		switch (l4_hdr) {
6374 		case IPPROTO_TCP:
6375 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6376 			mss_l4len_idx = tcp_hdrlen(skb) <<
6377 					IXGBE_ADVTXD_L4LEN_SHIFT;
6378 			break;
6379 		case IPPROTO_SCTP:
6380 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6381 			mss_l4len_idx = sizeof(struct sctphdr) <<
6382 					IXGBE_ADVTXD_L4LEN_SHIFT;
6383 			break;
6384 		case IPPROTO_UDP:
6385 			mss_l4len_idx = sizeof(struct udphdr) <<
6386 					IXGBE_ADVTXD_L4LEN_SHIFT;
6387 			break;
6388 		default:
6389 			if (unlikely(net_ratelimit())) {
6390 				dev_warn(tx_ring->dev,
6391 				 "partial checksum but l4 proto=%x!\n",
6392 				 skb->protocol);
6393 			}
6394 			break;
6395 		}
6396 	}
6397 
6398 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6399 	vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6400 
6401 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
6402 			  type_tucmd, mss_l4len_idx);
6403 
6404 	return (skb->ip_summed == CHECKSUM_PARTIAL);
6405 }
6406 
6407 static __le32 ixgbe_tx_cmd_type(u32 tx_flags)
6408 {
6409 	/* set type for advanced descriptor with frame checksum insertion */
6410 	__le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
6411 				      IXGBE_ADVTXD_DCMD_IFCS |
6412 				      IXGBE_ADVTXD_DCMD_DEXT);
6413 
6414 	/* set HW vlan bit if vlan is present */
6415 	if (tx_flags & IXGBE_TX_FLAGS_HW_VLAN)
6416 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
6417 
6418 	/* set segmentation enable bits for TSO/FSO */
6419 #ifdef IXGBE_FCOE
6420 	if ((tx_flags & IXGBE_TX_FLAGS_TSO) || (tx_flags & IXGBE_TX_FLAGS_FSO))
6421 #else
6422 	if (tx_flags & IXGBE_TX_FLAGS_TSO)
6423 #endif
6424 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
6425 
6426 	return cmd_type;
6427 }
6428 
6429 static __le32 ixgbe_tx_olinfo_status(u32 tx_flags, unsigned int paylen)
6430 {
6431 	__le32 olinfo_status =
6432 		cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
6433 
6434 	if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6435 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM |
6436 					    (1 << IXGBE_ADVTXD_IDX_SHIFT));
6437 		/* enble IPv4 checksum for TSO */
6438 		if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6439 			olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
6440 	}
6441 
6442 	/* enable L4 checksum for TSO and TX checksum offload */
6443 	if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6444 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
6445 
6446 #ifdef IXGBE_FCOE
6447 	/* use index 1 context for FCOE/FSO */
6448 	if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6449 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC |
6450 					    (1 << IXGBE_ADVTXD_IDX_SHIFT));
6451 
6452 #endif
6453 	/*
6454 	 * Check Context must be set if Tx switch is enabled, which it
6455 	 * always is for case where virtual functions are running
6456 	 */
6457 	if (tx_flags & IXGBE_TX_FLAGS_TXSW)
6458 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
6459 
6460 	return olinfo_status;
6461 }
6462 
6463 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
6464 		       IXGBE_TXD_CMD_RS)
6465 
6466 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
6467 			 struct sk_buff *skb,
6468 			 struct ixgbe_tx_buffer *first,
6469 			 u32 tx_flags,
6470 			 const u8 hdr_len)
6471 {
6472 	struct device *dev = tx_ring->dev;
6473 	struct ixgbe_tx_buffer *tx_buffer_info;
6474 	union ixgbe_adv_tx_desc *tx_desc;
6475 	dma_addr_t dma;
6476 	__le32 cmd_type, olinfo_status;
6477 	struct skb_frag_struct *frag;
6478 	unsigned int f = 0;
6479 	unsigned int data_len = skb->data_len;
6480 	unsigned int size = skb_headlen(skb);
6481 	u32 offset = 0;
6482 	u32 paylen = skb->len - hdr_len;
6483 	u16 i = tx_ring->next_to_use;
6484 	u16 gso_segs;
6485 
6486 #ifdef IXGBE_FCOE
6487 	if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6488 		if (data_len >= sizeof(struct fcoe_crc_eof)) {
6489 			data_len -= sizeof(struct fcoe_crc_eof);
6490 		} else {
6491 			size -= sizeof(struct fcoe_crc_eof) - data_len;
6492 			data_len = 0;
6493 		}
6494 	}
6495 
6496 #endif
6497 	dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
6498 	if (dma_mapping_error(dev, dma))
6499 		goto dma_error;
6500 
6501 	cmd_type = ixgbe_tx_cmd_type(tx_flags);
6502 	olinfo_status = ixgbe_tx_olinfo_status(tx_flags, paylen);
6503 
6504 	tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6505 
6506 	for (;;) {
6507 		while (size > IXGBE_MAX_DATA_PER_TXD) {
6508 			tx_desc->read.buffer_addr = cpu_to_le64(dma + offset);
6509 			tx_desc->read.cmd_type_len =
6510 				cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
6511 			tx_desc->read.olinfo_status = olinfo_status;
6512 
6513 			offset += IXGBE_MAX_DATA_PER_TXD;
6514 			size -= IXGBE_MAX_DATA_PER_TXD;
6515 
6516 			tx_desc++;
6517 			i++;
6518 			if (i == tx_ring->count) {
6519 				tx_desc = IXGBE_TX_DESC_ADV(tx_ring, 0);
6520 				i = 0;
6521 			}
6522 		}
6523 
6524 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
6525 		tx_buffer_info->length = offset + size;
6526 		tx_buffer_info->tx_flags = tx_flags;
6527 		tx_buffer_info->dma = dma;
6528 
6529 		tx_desc->read.buffer_addr = cpu_to_le64(dma + offset);
6530 		tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
6531 		tx_desc->read.olinfo_status = olinfo_status;
6532 
6533 		if (!data_len)
6534 			break;
6535 
6536 		frag = &skb_shinfo(skb)->frags[f];
6537 #ifdef IXGBE_FCOE
6538 		size = min_t(unsigned int, data_len, skb_frag_size(frag));
6539 #else
6540 		size = skb_frag_size(frag);
6541 #endif
6542 		data_len -= size;
6543 		f++;
6544 
6545 		offset = 0;
6546 		tx_flags |= IXGBE_TX_FLAGS_MAPPED_AS_PAGE;
6547 
6548 		dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
6549 		if (dma_mapping_error(dev, dma))
6550 			goto dma_error;
6551 
6552 		tx_desc++;
6553 		i++;
6554 		if (i == tx_ring->count) {
6555 			tx_desc = IXGBE_TX_DESC_ADV(tx_ring, 0);
6556 			i = 0;
6557 		}
6558 	}
6559 
6560 	tx_desc->read.cmd_type_len |= cpu_to_le32(IXGBE_TXD_CMD);
6561 
6562 	i++;
6563 	if (i == tx_ring->count)
6564 		i = 0;
6565 
6566 	tx_ring->next_to_use = i;
6567 
6568 	if (tx_flags & IXGBE_TX_FLAGS_TSO)
6569 		gso_segs = skb_shinfo(skb)->gso_segs;
6570 #ifdef IXGBE_FCOE
6571 	/* adjust for FCoE Sequence Offload */
6572 	else if (tx_flags & IXGBE_TX_FLAGS_FSO)
6573 		gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
6574 					skb_shinfo(skb)->gso_size);
6575 #endif /* IXGBE_FCOE */
6576 	else
6577 		gso_segs = 1;
6578 
6579 	/* multiply data chunks by size of headers */
6580 	tx_buffer_info->bytecount = paylen + (gso_segs * hdr_len);
6581 	tx_buffer_info->gso_segs = gso_segs;
6582 	tx_buffer_info->skb = skb;
6583 
6584 	/* set the timestamp */
6585 	first->time_stamp = jiffies;
6586 
6587 	/*
6588 	 * Force memory writes to complete before letting h/w
6589 	 * know there are new descriptors to fetch.  (Only
6590 	 * applicable for weak-ordered memory model archs,
6591 	 * such as IA-64).
6592 	 */
6593 	wmb();
6594 
6595 	/* set next_to_watch value indicating a packet is present */
6596 	first->next_to_watch = tx_desc;
6597 
6598 	/* notify HW of packet */
6599 	writel(i, tx_ring->tail);
6600 
6601 	return;
6602 dma_error:
6603 	dev_err(dev, "TX DMA map failed\n");
6604 
6605 	/* clear dma mappings for failed tx_buffer_info map */
6606 	for (;;) {
6607 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
6608 		ixgbe_unmap_tx_resource(tx_ring, tx_buffer_info);
6609 		if (tx_buffer_info == first)
6610 			break;
6611 		if (i == 0)
6612 			i = tx_ring->count;
6613 		i--;
6614 	}
6615 
6616 	dev_kfree_skb_any(skb);
6617 
6618 	tx_ring->next_to_use = i;
6619 }
6620 
6621 static void ixgbe_atr(struct ixgbe_ring *ring, struct sk_buff *skb,
6622 		      u32 tx_flags, __be16 protocol)
6623 {
6624 	struct ixgbe_q_vector *q_vector = ring->q_vector;
6625 	union ixgbe_atr_hash_dword input = { .dword = 0 };
6626 	union ixgbe_atr_hash_dword common = { .dword = 0 };
6627 	union {
6628 		unsigned char *network;
6629 		struct iphdr *ipv4;
6630 		struct ipv6hdr *ipv6;
6631 	} hdr;
6632 	struct tcphdr *th;
6633 	__be16 vlan_id;
6634 
6635 	/* if ring doesn't have a interrupt vector, cannot perform ATR */
6636 	if (!q_vector)
6637 		return;
6638 
6639 	/* do nothing if sampling is disabled */
6640 	if (!ring->atr_sample_rate)
6641 		return;
6642 
6643 	ring->atr_count++;
6644 
6645 	/* snag network header to get L4 type and address */
6646 	hdr.network = skb_network_header(skb);
6647 
6648 	/* Currently only IPv4/IPv6 with TCP is supported */
6649 	if ((protocol != __constant_htons(ETH_P_IPV6) ||
6650 	     hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6651 	    (protocol != __constant_htons(ETH_P_IP) ||
6652 	     hdr.ipv4->protocol != IPPROTO_TCP))
6653 		return;
6654 
6655 	th = tcp_hdr(skb);
6656 
6657 	/* skip this packet since it is invalid or the socket is closing */
6658 	if (!th || th->fin)
6659 		return;
6660 
6661 	/* sample on all syn packets or once every atr sample count */
6662 	if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6663 		return;
6664 
6665 	/* reset sample count */
6666 	ring->atr_count = 0;
6667 
6668 	vlan_id = htons(tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6669 
6670 	/*
6671 	 * src and dst are inverted, think how the receiver sees them
6672 	 *
6673 	 * The input is broken into two sections, a non-compressed section
6674 	 * containing vm_pool, vlan_id, and flow_type.  The rest of the data
6675 	 * is XORed together and stored in the compressed dword.
6676 	 */
6677 	input.formatted.vlan_id = vlan_id;
6678 
6679 	/*
6680 	 * since src port and flex bytes occupy the same word XOR them together
6681 	 * and write the value to source port portion of compressed dword
6682 	 */
6683 	if (tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6684 		common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6685 	else
6686 		common.port.src ^= th->dest ^ protocol;
6687 	common.port.dst ^= th->source;
6688 
6689 	if (protocol == __constant_htons(ETH_P_IP)) {
6690 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6691 		common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6692 	} else {
6693 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6694 		common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6695 			     hdr.ipv6->saddr.s6_addr32[1] ^
6696 			     hdr.ipv6->saddr.s6_addr32[2] ^
6697 			     hdr.ipv6->saddr.s6_addr32[3] ^
6698 			     hdr.ipv6->daddr.s6_addr32[0] ^
6699 			     hdr.ipv6->daddr.s6_addr32[1] ^
6700 			     hdr.ipv6->daddr.s6_addr32[2] ^
6701 			     hdr.ipv6->daddr.s6_addr32[3];
6702 	}
6703 
6704 	/* This assumes the Rx queue and Tx queue are bound to the same CPU */
6705 	ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6706 					      input, common, ring->queue_index);
6707 }
6708 
6709 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6710 {
6711 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6712 	/* Herbert's original patch had:
6713 	 *  smp_mb__after_netif_stop_queue();
6714 	 * but since that doesn't exist yet, just open code it. */
6715 	smp_mb();
6716 
6717 	/* We need to check again in a case another CPU has just
6718 	 * made room available. */
6719 	if (likely(ixgbe_desc_unused(tx_ring) < size))
6720 		return -EBUSY;
6721 
6722 	/* A reprieve! - use start_queue because it doesn't call schedule */
6723 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6724 	++tx_ring->tx_stats.restart_queue;
6725 	return 0;
6726 }
6727 
6728 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6729 {
6730 	if (likely(ixgbe_desc_unused(tx_ring) >= size))
6731 		return 0;
6732 	return __ixgbe_maybe_stop_tx(tx_ring, size);
6733 }
6734 
6735 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6736 {
6737 	struct ixgbe_adapter *adapter = netdev_priv(dev);
6738 	int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6739 					       smp_processor_id();
6740 #ifdef IXGBE_FCOE
6741 	__be16 protocol = vlan_get_protocol(skb);
6742 
6743 	if (((protocol == htons(ETH_P_FCOE)) ||
6744 	    (protocol == htons(ETH_P_FIP))) &&
6745 	    (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6746 		txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6747 		txq += adapter->ring_feature[RING_F_FCOE].mask;
6748 		return txq;
6749 	}
6750 #endif
6751 
6752 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6753 		while (unlikely(txq >= dev->real_num_tx_queues))
6754 			txq -= dev->real_num_tx_queues;
6755 		return txq;
6756 	}
6757 
6758 	return skb_tx_hash(dev, skb);
6759 }
6760 
6761 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6762 			  struct ixgbe_adapter *adapter,
6763 			  struct ixgbe_ring *tx_ring)
6764 {
6765 	struct ixgbe_tx_buffer *first;
6766 	int tso;
6767 	u32 tx_flags = 0;
6768 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6769 	unsigned short f;
6770 #endif
6771 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
6772 	__be16 protocol = skb->protocol;
6773 	u8 hdr_len = 0;
6774 
6775 	/*
6776 	 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6777 	 *       + 1 desc for skb_head_len/IXGBE_MAX_DATA_PER_TXD,
6778 	 *       + 2 desc gap to keep tail from touching head,
6779 	 *       + 1 desc for context descriptor,
6780 	 * otherwise try next time
6781 	 */
6782 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6783 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6784 		count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6785 #else
6786 	count += skb_shinfo(skb)->nr_frags;
6787 #endif
6788 	if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
6789 		tx_ring->tx_stats.tx_busy++;
6790 		return NETDEV_TX_BUSY;
6791 	}
6792 
6793 #ifdef CONFIG_PCI_IOV
6794 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6795 		tx_flags |= IXGBE_TX_FLAGS_TXSW;
6796 
6797 #endif
6798 	/* if we have a HW VLAN tag being added default to the HW one */
6799 	if (vlan_tx_tag_present(skb)) {
6800 		tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6801 		tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6802 	/* else if it is a SW VLAN check the next protocol and store the tag */
6803 	} else if (protocol == __constant_htons(ETH_P_8021Q)) {
6804 		struct vlan_hdr *vhdr, _vhdr;
6805 		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
6806 		if (!vhdr)
6807 			goto out_drop;
6808 
6809 		protocol = vhdr->h_vlan_encapsulated_proto;
6810 		tx_flags |= ntohs(vhdr->h_vlan_TCI) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6811 		tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
6812 	}
6813 
6814 	/* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
6815 	if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
6816 	    ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
6817 	     (skb->priority != TC_PRIO_CONTROL))) {
6818 		tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6819 		tx_flags |= (skb->priority & 0x7) <<
6820 					IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
6821 		if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
6822 			struct vlan_ethhdr *vhdr;
6823 			if (skb_header_cloned(skb) &&
6824 			    pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
6825 				goto out_drop;
6826 			vhdr = (struct vlan_ethhdr *)skb->data;
6827 			vhdr->h_vlan_TCI = htons(tx_flags >>
6828 						 IXGBE_TX_FLAGS_VLAN_SHIFT);
6829 		} else {
6830 			tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6831 		}
6832 	}
6833 
6834 	/* record the location of the first descriptor for this packet */
6835 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6836 
6837 #ifdef IXGBE_FCOE
6838 	/* setup tx offload for FCoE */
6839 	if ((protocol == __constant_htons(ETH_P_FCOE)) &&
6840 	    (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6841 		tso = ixgbe_fso(tx_ring, skb, tx_flags, &hdr_len);
6842 		if (tso < 0)
6843 			goto out_drop;
6844 		else if (tso)
6845 			tx_flags |= IXGBE_TX_FLAGS_FSO |
6846 				    IXGBE_TX_FLAGS_FCOE;
6847 		else
6848 			tx_flags |= IXGBE_TX_FLAGS_FCOE;
6849 
6850 		goto xmit_fcoe;
6851 	}
6852 
6853 #endif /* IXGBE_FCOE */
6854 	/* setup IPv4/IPv6 offloads */
6855 	if (protocol == __constant_htons(ETH_P_IP))
6856 		tx_flags |= IXGBE_TX_FLAGS_IPV4;
6857 
6858 	tso = ixgbe_tso(tx_ring, skb, tx_flags, protocol, &hdr_len);
6859 	if (tso < 0)
6860 		goto out_drop;
6861 	else if (tso)
6862 		tx_flags |= IXGBE_TX_FLAGS_TSO;
6863 	else if (ixgbe_tx_csum(tx_ring, skb, tx_flags, protocol))
6864 		tx_flags |= IXGBE_TX_FLAGS_CSUM;
6865 
6866 	/* add the ATR filter if ATR is on */
6867 	if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6868 		ixgbe_atr(tx_ring, skb, tx_flags, protocol);
6869 
6870 #ifdef IXGBE_FCOE
6871 xmit_fcoe:
6872 #endif /* IXGBE_FCOE */
6873 	ixgbe_tx_map(tx_ring, skb, first, tx_flags, hdr_len);
6874 
6875 	ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6876 
6877 	return NETDEV_TX_OK;
6878 
6879 out_drop:
6880 	dev_kfree_skb_any(skb);
6881 	return NETDEV_TX_OK;
6882 }
6883 
6884 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6885 {
6886 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6887 	struct ixgbe_ring *tx_ring;
6888 
6889 	tx_ring = adapter->tx_ring[skb->queue_mapping];
6890 	return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6891 }
6892 
6893 /**
6894  * ixgbe_set_mac - Change the Ethernet Address of the NIC
6895  * @netdev: network interface device structure
6896  * @p: pointer to an address structure
6897  *
6898  * Returns 0 on success, negative on failure
6899  **/
6900 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6901 {
6902 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6903 	struct ixgbe_hw *hw = &adapter->hw;
6904 	struct sockaddr *addr = p;
6905 
6906 	if (!is_valid_ether_addr(addr->sa_data))
6907 		return -EADDRNOTAVAIL;
6908 
6909 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6910 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6911 
6912 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6913 			    IXGBE_RAH_AV);
6914 
6915 	return 0;
6916 }
6917 
6918 static int
6919 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6920 {
6921 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6922 	struct ixgbe_hw *hw = &adapter->hw;
6923 	u16 value;
6924 	int rc;
6925 
6926 	if (prtad != hw->phy.mdio.prtad)
6927 		return -EINVAL;
6928 	rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6929 	if (!rc)
6930 		rc = value;
6931 	return rc;
6932 }
6933 
6934 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6935 			    u16 addr, u16 value)
6936 {
6937 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6938 	struct ixgbe_hw *hw = &adapter->hw;
6939 
6940 	if (prtad != hw->phy.mdio.prtad)
6941 		return -EINVAL;
6942 	return hw->phy.ops.write_reg(hw, addr, devad, value);
6943 }
6944 
6945 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6946 {
6947 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6948 
6949 	return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6950 }
6951 
6952 /**
6953  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6954  * netdev->dev_addrs
6955  * @netdev: network interface device structure
6956  *
6957  * Returns non-zero on failure
6958  **/
6959 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6960 {
6961 	int err = 0;
6962 	struct ixgbe_adapter *adapter = netdev_priv(dev);
6963 	struct ixgbe_mac_info *mac = &adapter->hw.mac;
6964 
6965 	if (is_valid_ether_addr(mac->san_addr)) {
6966 		rtnl_lock();
6967 		err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6968 		rtnl_unlock();
6969 	}
6970 	return err;
6971 }
6972 
6973 /**
6974  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6975  * netdev->dev_addrs
6976  * @netdev: network interface device structure
6977  *
6978  * Returns non-zero on failure
6979  **/
6980 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6981 {
6982 	int err = 0;
6983 	struct ixgbe_adapter *adapter = netdev_priv(dev);
6984 	struct ixgbe_mac_info *mac = &adapter->hw.mac;
6985 
6986 	if (is_valid_ether_addr(mac->san_addr)) {
6987 		rtnl_lock();
6988 		err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6989 		rtnl_unlock();
6990 	}
6991 	return err;
6992 }
6993 
6994 #ifdef CONFIG_NET_POLL_CONTROLLER
6995 /*
6996  * Polling 'interrupt' - used by things like netconsole to send skbs
6997  * without having to re-enable interrupts. It's not called while
6998  * the interrupt routine is executing.
6999  */
7000 static void ixgbe_netpoll(struct net_device *netdev)
7001 {
7002 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7003 	int i;
7004 
7005 	/* if interface is down do nothing */
7006 	if (test_bit(__IXGBE_DOWN, &adapter->state))
7007 		return;
7008 
7009 	adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
7010 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
7011 		int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
7012 		for (i = 0; i < num_q_vectors; i++) {
7013 			struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
7014 			ixgbe_msix_clean_rings(0, q_vector);
7015 		}
7016 	} else {
7017 		ixgbe_intr(adapter->pdev->irq, netdev);
7018 	}
7019 	adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
7020 }
7021 #endif
7022 
7023 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
7024 						   struct rtnl_link_stats64 *stats)
7025 {
7026 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7027 	int i;
7028 
7029 	rcu_read_lock();
7030 	for (i = 0; i < adapter->num_rx_queues; i++) {
7031 		struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
7032 		u64 bytes, packets;
7033 		unsigned int start;
7034 
7035 		if (ring) {
7036 			do {
7037 				start = u64_stats_fetch_begin_bh(&ring->syncp);
7038 				packets = ring->stats.packets;
7039 				bytes   = ring->stats.bytes;
7040 			} while (u64_stats_fetch_retry_bh(&ring->syncp, start));
7041 			stats->rx_packets += packets;
7042 			stats->rx_bytes   += bytes;
7043 		}
7044 	}
7045 
7046 	for (i = 0; i < adapter->num_tx_queues; i++) {
7047 		struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
7048 		u64 bytes, packets;
7049 		unsigned int start;
7050 
7051 		if (ring) {
7052 			do {
7053 				start = u64_stats_fetch_begin_bh(&ring->syncp);
7054 				packets = ring->stats.packets;
7055 				bytes   = ring->stats.bytes;
7056 			} while (u64_stats_fetch_retry_bh(&ring->syncp, start));
7057 			stats->tx_packets += packets;
7058 			stats->tx_bytes   += bytes;
7059 		}
7060 	}
7061 	rcu_read_unlock();
7062 	/* following stats updated by ixgbe_watchdog_task() */
7063 	stats->multicast	= netdev->stats.multicast;
7064 	stats->rx_errors	= netdev->stats.rx_errors;
7065 	stats->rx_length_errors	= netdev->stats.rx_length_errors;
7066 	stats->rx_crc_errors	= netdev->stats.rx_crc_errors;
7067 	stats->rx_missed_errors	= netdev->stats.rx_missed_errors;
7068 	return stats;
7069 }
7070 
7071 /* ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
7072  * #adapter: pointer to ixgbe_adapter
7073  * @tc: number of traffic classes currently enabled
7074  *
7075  * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
7076  * 802.1Q priority maps to a packet buffer that exists.
7077  */
7078 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
7079 {
7080 	struct ixgbe_hw *hw = &adapter->hw;
7081 	u32 reg, rsave;
7082 	int i;
7083 
7084 	/* 82598 have a static priority to TC mapping that can not
7085 	 * be changed so no validation is needed.
7086 	 */
7087 	if (hw->mac.type == ixgbe_mac_82598EB)
7088 		return;
7089 
7090 	reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
7091 	rsave = reg;
7092 
7093 	for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
7094 		u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
7095 
7096 		/* If up2tc is out of bounds default to zero */
7097 		if (up2tc > tc)
7098 			reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
7099 	}
7100 
7101 	if (reg != rsave)
7102 		IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
7103 
7104 	return;
7105 }
7106 
7107 
7108 /* ixgbe_setup_tc - routine to configure net_device for multiple traffic
7109  * classes.
7110  *
7111  * @netdev: net device to configure
7112  * @tc: number of traffic classes to enable
7113  */
7114 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
7115 {
7116 	struct ixgbe_adapter *adapter = netdev_priv(dev);
7117 	struct ixgbe_hw *hw = &adapter->hw;
7118 
7119 	/* Multiple traffic classes requires multiple queues */
7120 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
7121 		e_err(drv, "Enable failed, needs MSI-X\n");
7122 		return -EINVAL;
7123 	}
7124 
7125 	/* Hardware supports up to 8 traffic classes */
7126 	if (tc > adapter->dcb_cfg.num_tcs.pg_tcs ||
7127 	    (hw->mac.type == ixgbe_mac_82598EB && tc < MAX_TRAFFIC_CLASS))
7128 		return -EINVAL;
7129 
7130 	/* Hardware has to reinitialize queues and interrupts to
7131 	 * match packet buffer alignment. Unfortunantly, the
7132 	 * hardware is not flexible enough to do this dynamically.
7133 	 */
7134 	if (netif_running(dev))
7135 		ixgbe_close(dev);
7136 	ixgbe_clear_interrupt_scheme(adapter);
7137 
7138 	if (tc) {
7139 		netdev_set_num_tc(dev, tc);
7140 		adapter->last_lfc_mode = adapter->hw.fc.current_mode;
7141 
7142 		adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
7143 		adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
7144 
7145 		if (adapter->hw.mac.type == ixgbe_mac_82598EB)
7146 			adapter->hw.fc.requested_mode = ixgbe_fc_none;
7147 	} else {
7148 		netdev_reset_tc(dev);
7149 
7150 		adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
7151 
7152 		adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
7153 		adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
7154 
7155 		adapter->temp_dcb_cfg.pfc_mode_enable = false;
7156 		adapter->dcb_cfg.pfc_mode_enable = false;
7157 	}
7158 
7159 	ixgbe_init_interrupt_scheme(adapter);
7160 	ixgbe_validate_rtr(adapter, tc);
7161 	if (netif_running(dev))
7162 		ixgbe_open(dev);
7163 
7164 	return 0;
7165 }
7166 
7167 void ixgbe_do_reset(struct net_device *netdev)
7168 {
7169 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7170 
7171 	if (netif_running(netdev))
7172 		ixgbe_reinit_locked(adapter);
7173 	else
7174 		ixgbe_reset(adapter);
7175 }
7176 
7177 static u32 ixgbe_fix_features(struct net_device *netdev, u32 data)
7178 {
7179 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7180 
7181 #ifdef CONFIG_DCB
7182 	if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
7183 		data &= ~NETIF_F_HW_VLAN_RX;
7184 #endif
7185 
7186 	/* return error if RXHASH is being enabled when RSS is not supported */
7187 	if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
7188 		data &= ~NETIF_F_RXHASH;
7189 
7190 	/* If Rx checksum is disabled, then RSC/LRO should also be disabled */
7191 	if (!(data & NETIF_F_RXCSUM))
7192 		data &= ~NETIF_F_LRO;
7193 
7194 	/* Turn off LRO if not RSC capable or invalid ITR settings */
7195 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)) {
7196 		data &= ~NETIF_F_LRO;
7197 	} else if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
7198 		   (adapter->rx_itr_setting != 1 &&
7199 		    adapter->rx_itr_setting > IXGBE_MAX_RSC_INT_RATE)) {
7200 		data &= ~NETIF_F_LRO;
7201 		e_info(probe, "rx-usecs set too low, not enabling RSC\n");
7202 	}
7203 
7204 	return data;
7205 }
7206 
7207 static int ixgbe_set_features(struct net_device *netdev, u32 data)
7208 {
7209 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7210 	bool need_reset = false;
7211 
7212 	/* If Rx checksum is disabled, then RSC/LRO should also be disabled */
7213 	if (!(data & NETIF_F_RXCSUM))
7214 		adapter->flags &= ~IXGBE_FLAG_RX_CSUM_ENABLED;
7215 	else
7216 		adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
7217 
7218 	/* Make sure RSC matches LRO, reset if change */
7219 	if (!!(data & NETIF_F_LRO) !=
7220 	     !!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
7221 		adapter->flags2 ^= IXGBE_FLAG2_RSC_ENABLED;
7222 		switch (adapter->hw.mac.type) {
7223 		case ixgbe_mac_X540:
7224 		case ixgbe_mac_82599EB:
7225 			need_reset = true;
7226 			break;
7227 		default:
7228 			break;
7229 		}
7230 	}
7231 
7232 	/*
7233 	 * Check if Flow Director n-tuple support was enabled or disabled.  If
7234 	 * the state changed, we need to reset.
7235 	 */
7236 	if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
7237 		/* turn off ATR, enable perfect filters and reset */
7238 		if (data & NETIF_F_NTUPLE) {
7239 			adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
7240 			adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7241 			need_reset = true;
7242 		}
7243 	} else if (!(data & NETIF_F_NTUPLE)) {
7244 		/* turn off Flow Director, set ATR and reset */
7245 		adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7246 		if ((adapter->flags &  IXGBE_FLAG_RSS_ENABLED) &&
7247 		    !(adapter->flags &  IXGBE_FLAG_DCB_ENABLED))
7248 			adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
7249 		need_reset = true;
7250 	}
7251 
7252 	if (need_reset)
7253 		ixgbe_do_reset(netdev);
7254 
7255 	return 0;
7256 
7257 }
7258 
7259 static const struct net_device_ops ixgbe_netdev_ops = {
7260 	.ndo_open		= ixgbe_open,
7261 	.ndo_stop		= ixgbe_close,
7262 	.ndo_start_xmit		= ixgbe_xmit_frame,
7263 	.ndo_select_queue	= ixgbe_select_queue,
7264 	.ndo_set_rx_mode        = ixgbe_set_rx_mode,
7265 	.ndo_validate_addr	= eth_validate_addr,
7266 	.ndo_set_mac_address	= ixgbe_set_mac,
7267 	.ndo_change_mtu		= ixgbe_change_mtu,
7268 	.ndo_tx_timeout		= ixgbe_tx_timeout,
7269 	.ndo_vlan_rx_add_vid	= ixgbe_vlan_rx_add_vid,
7270 	.ndo_vlan_rx_kill_vid	= ixgbe_vlan_rx_kill_vid,
7271 	.ndo_do_ioctl		= ixgbe_ioctl,
7272 	.ndo_set_vf_mac		= ixgbe_ndo_set_vf_mac,
7273 	.ndo_set_vf_vlan	= ixgbe_ndo_set_vf_vlan,
7274 	.ndo_set_vf_tx_rate	= ixgbe_ndo_set_vf_bw,
7275 	.ndo_set_vf_spoofchk    = ixgbe_ndo_set_vf_spoofchk,
7276 	.ndo_get_vf_config	= ixgbe_ndo_get_vf_config,
7277 	.ndo_get_stats64	= ixgbe_get_stats64,
7278 	.ndo_setup_tc		= ixgbe_setup_tc,
7279 #ifdef CONFIG_NET_POLL_CONTROLLER
7280 	.ndo_poll_controller	= ixgbe_netpoll,
7281 #endif
7282 #ifdef IXGBE_FCOE
7283 	.ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
7284 	.ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
7285 	.ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
7286 	.ndo_fcoe_enable = ixgbe_fcoe_enable,
7287 	.ndo_fcoe_disable = ixgbe_fcoe_disable,
7288 	.ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
7289 #endif /* IXGBE_FCOE */
7290 	.ndo_set_features = ixgbe_set_features,
7291 	.ndo_fix_features = ixgbe_fix_features,
7292 };
7293 
7294 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
7295 			   const struct ixgbe_info *ii)
7296 {
7297 #ifdef CONFIG_PCI_IOV
7298 	struct ixgbe_hw *hw = &adapter->hw;
7299 
7300 	if (hw->mac.type == ixgbe_mac_82598EB)
7301 		return;
7302 
7303 	/* The 82599 supports up to 64 VFs per physical function
7304 	 * but this implementation limits allocation to 63 so that
7305 	 * basic networking resources are still available to the
7306 	 * physical function
7307 	 */
7308 	adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
7309 	ixgbe_enable_sriov(adapter, ii);
7310 #endif /* CONFIG_PCI_IOV */
7311 }
7312 
7313 /**
7314  * ixgbe_probe - Device Initialization Routine
7315  * @pdev: PCI device information struct
7316  * @ent: entry in ixgbe_pci_tbl
7317  *
7318  * Returns 0 on success, negative on failure
7319  *
7320  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7321  * The OS initialization, configuring of the adapter private structure,
7322  * and a hardware reset occur.
7323  **/
7324 static int __devinit ixgbe_probe(struct pci_dev *pdev,
7325 				 const struct pci_device_id *ent)
7326 {
7327 	struct net_device *netdev;
7328 	struct ixgbe_adapter *adapter = NULL;
7329 	struct ixgbe_hw *hw;
7330 	const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7331 	static int cards_found;
7332 	int i, err, pci_using_dac;
7333 	u8 part_str[IXGBE_PBANUM_LENGTH];
7334 	unsigned int indices = num_possible_cpus();
7335 #ifdef IXGBE_FCOE
7336 	u16 device_caps;
7337 #endif
7338 	u32 eec;
7339 	u16 wol_cap;
7340 
7341 	/* Catch broken hardware that put the wrong VF device ID in
7342 	 * the PCIe SR-IOV capability.
7343 	 */
7344 	if (pdev->is_virtfn) {
7345 		WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7346 		     pci_name(pdev), pdev->vendor, pdev->device);
7347 		return -EINVAL;
7348 	}
7349 
7350 	err = pci_enable_device_mem(pdev);
7351 	if (err)
7352 		return err;
7353 
7354 	if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7355 	    !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7356 		pci_using_dac = 1;
7357 	} else {
7358 		err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7359 		if (err) {
7360 			err = dma_set_coherent_mask(&pdev->dev,
7361 						    DMA_BIT_MASK(32));
7362 			if (err) {
7363 				dev_err(&pdev->dev,
7364 					"No usable DMA configuration, aborting\n");
7365 				goto err_dma;
7366 			}
7367 		}
7368 		pci_using_dac = 0;
7369 	}
7370 
7371 	err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7372 					   IORESOURCE_MEM), ixgbe_driver_name);
7373 	if (err) {
7374 		dev_err(&pdev->dev,
7375 			"pci_request_selected_regions failed 0x%x\n", err);
7376 		goto err_pci_reg;
7377 	}
7378 
7379 	pci_enable_pcie_error_reporting(pdev);
7380 
7381 	pci_set_master(pdev);
7382 	pci_save_state(pdev);
7383 
7384 #ifdef CONFIG_IXGBE_DCB
7385 	indices *= MAX_TRAFFIC_CLASS;
7386 #endif
7387 
7388 	if (ii->mac == ixgbe_mac_82598EB)
7389 		indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7390 	else
7391 		indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7392 
7393 #ifdef IXGBE_FCOE
7394 	indices += min_t(unsigned int, num_possible_cpus(),
7395 			 IXGBE_MAX_FCOE_INDICES);
7396 #endif
7397 	netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7398 	if (!netdev) {
7399 		err = -ENOMEM;
7400 		goto err_alloc_etherdev;
7401 	}
7402 
7403 	SET_NETDEV_DEV(netdev, &pdev->dev);
7404 
7405 	adapter = netdev_priv(netdev);
7406 	pci_set_drvdata(pdev, adapter);
7407 
7408 	adapter->netdev = netdev;
7409 	adapter->pdev = pdev;
7410 	hw = &adapter->hw;
7411 	hw->back = adapter;
7412 	adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
7413 
7414 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7415 			      pci_resource_len(pdev, 0));
7416 	if (!hw->hw_addr) {
7417 		err = -EIO;
7418 		goto err_ioremap;
7419 	}
7420 
7421 	for (i = 1; i <= 5; i++) {
7422 		if (pci_resource_len(pdev, i) == 0)
7423 			continue;
7424 	}
7425 
7426 	netdev->netdev_ops = &ixgbe_netdev_ops;
7427 	ixgbe_set_ethtool_ops(netdev);
7428 	netdev->watchdog_timeo = 5 * HZ;
7429 	strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7430 
7431 	adapter->bd_number = cards_found;
7432 
7433 	/* Setup hw api */
7434 	memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7435 	hw->mac.type  = ii->mac;
7436 
7437 	/* EEPROM */
7438 	memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7439 	eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7440 	/* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7441 	if (!(eec & (1 << 8)))
7442 		hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7443 
7444 	/* PHY */
7445 	memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7446 	hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7447 	/* ixgbe_identify_phy_generic will set prtad and mmds properly */
7448 	hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7449 	hw->phy.mdio.mmds = 0;
7450 	hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7451 	hw->phy.mdio.dev = netdev;
7452 	hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7453 	hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7454 
7455 	ii->get_invariants(hw);
7456 
7457 	/* setup the private structure */
7458 	err = ixgbe_sw_init(adapter);
7459 	if (err)
7460 		goto err_sw_init;
7461 
7462 	/* Make it possible the adapter to be woken up via WOL */
7463 	switch (adapter->hw.mac.type) {
7464 	case ixgbe_mac_82599EB:
7465 	case ixgbe_mac_X540:
7466 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7467 		break;
7468 	default:
7469 		break;
7470 	}
7471 
7472 	/*
7473 	 * If there is a fan on this device and it has failed log the
7474 	 * failure.
7475 	 */
7476 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7477 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7478 		if (esdp & IXGBE_ESDP_SDP1)
7479 			e_crit(probe, "Fan has stopped, replace the adapter\n");
7480 	}
7481 
7482 	/* reset_hw fills in the perm_addr as well */
7483 	hw->phy.reset_if_overtemp = true;
7484 	err = hw->mac.ops.reset_hw(hw);
7485 	hw->phy.reset_if_overtemp = false;
7486 	if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7487 	    hw->mac.type == ixgbe_mac_82598EB) {
7488 		err = 0;
7489 	} else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7490 		e_dev_err("failed to load because an unsupported SFP+ "
7491 			  "module type was detected.\n");
7492 		e_dev_err("Reload the driver after installing a supported "
7493 			  "module.\n");
7494 		goto err_sw_init;
7495 	} else if (err) {
7496 		e_dev_err("HW Init failed: %d\n", err);
7497 		goto err_sw_init;
7498 	}
7499 
7500 	ixgbe_probe_vf(adapter, ii);
7501 
7502 	netdev->features = NETIF_F_SG |
7503 			   NETIF_F_IP_CSUM |
7504 			   NETIF_F_IPV6_CSUM |
7505 			   NETIF_F_HW_VLAN_TX |
7506 			   NETIF_F_HW_VLAN_RX |
7507 			   NETIF_F_HW_VLAN_FILTER |
7508 			   NETIF_F_TSO |
7509 			   NETIF_F_TSO6 |
7510 			   NETIF_F_RXHASH |
7511 			   NETIF_F_RXCSUM;
7512 
7513 	netdev->hw_features = netdev->features;
7514 
7515 	switch (adapter->hw.mac.type) {
7516 	case ixgbe_mac_82599EB:
7517 	case ixgbe_mac_X540:
7518 		netdev->features |= NETIF_F_SCTP_CSUM;
7519 		netdev->hw_features |= NETIF_F_SCTP_CSUM |
7520 				       NETIF_F_NTUPLE;
7521 		break;
7522 	default:
7523 		break;
7524 	}
7525 
7526 	netdev->vlan_features |= NETIF_F_TSO;
7527 	netdev->vlan_features |= NETIF_F_TSO6;
7528 	netdev->vlan_features |= NETIF_F_IP_CSUM;
7529 	netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7530 	netdev->vlan_features |= NETIF_F_SG;
7531 
7532 	netdev->priv_flags |= IFF_UNICAST_FLT;
7533 
7534 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7535 		adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
7536 				    IXGBE_FLAG_DCB_ENABLED);
7537 
7538 #ifdef CONFIG_IXGBE_DCB
7539 	netdev->dcbnl_ops = &dcbnl_ops;
7540 #endif
7541 
7542 #ifdef IXGBE_FCOE
7543 	if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7544 		if (hw->mac.ops.get_device_caps) {
7545 			hw->mac.ops.get_device_caps(hw, &device_caps);
7546 			if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7547 				adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7548 		}
7549 	}
7550 	if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7551 		netdev->vlan_features |= NETIF_F_FCOE_CRC;
7552 		netdev->vlan_features |= NETIF_F_FSO;
7553 		netdev->vlan_features |= NETIF_F_FCOE_MTU;
7554 	}
7555 #endif /* IXGBE_FCOE */
7556 	if (pci_using_dac) {
7557 		netdev->features |= NETIF_F_HIGHDMA;
7558 		netdev->vlan_features |= NETIF_F_HIGHDMA;
7559 	}
7560 
7561 	if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
7562 		netdev->hw_features |= NETIF_F_LRO;
7563 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7564 		netdev->features |= NETIF_F_LRO;
7565 
7566 	/* make sure the EEPROM is good */
7567 	if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7568 		e_dev_err("The EEPROM Checksum Is Not Valid\n");
7569 		err = -EIO;
7570 		goto err_eeprom;
7571 	}
7572 
7573 	memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7574 	memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7575 
7576 	if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7577 		e_dev_err("invalid MAC address\n");
7578 		err = -EIO;
7579 		goto err_eeprom;
7580 	}
7581 
7582 	setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7583 	            (unsigned long) adapter);
7584 
7585 	INIT_WORK(&adapter->service_task, ixgbe_service_task);
7586 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7587 
7588 	err = ixgbe_init_interrupt_scheme(adapter);
7589 	if (err)
7590 		goto err_sw_init;
7591 
7592 	if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
7593 		netdev->hw_features &= ~NETIF_F_RXHASH;
7594 		netdev->features &= ~NETIF_F_RXHASH;
7595 	}
7596 
7597 	/* WOL not supported for all but the following */
7598 	adapter->wol = 0;
7599 	switch (pdev->device) {
7600 	case IXGBE_DEV_ID_82599_SFP:
7601 		/* Only this subdevice supports WOL */
7602 		if (pdev->subsystem_device == IXGBE_SUBDEV_ID_82599_SFP)
7603 			adapter->wol = IXGBE_WUFC_MAG;
7604 		break;
7605 	case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7606 		/* All except this subdevice support WOL */
7607 		if (pdev->subsystem_device != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7608 			adapter->wol = IXGBE_WUFC_MAG;
7609 		break;
7610 	case IXGBE_DEV_ID_82599_KX4:
7611 		adapter->wol = IXGBE_WUFC_MAG;
7612 		break;
7613 	case IXGBE_DEV_ID_X540T:
7614 		/* Check eeprom to see if it is enabled */
7615 		hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
7616 		wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
7617 
7618 		if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
7619 		    ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
7620 		     (hw->bus.func == 0)))
7621 			adapter->wol = IXGBE_WUFC_MAG;
7622 		break;
7623 	}
7624 	device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7625 
7626 	/* save off EEPROM version number */
7627 	hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
7628 	hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
7629 
7630 	/* pick up the PCI bus settings for reporting later */
7631 	hw->mac.ops.get_bus_info(hw);
7632 
7633 	/* print bus type/speed/width info */
7634 	e_dev_info("(PCI Express:%s:%s) %pM\n",
7635 		   (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7636 		    hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7637 		    "Unknown"),
7638 		   (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7639 		    hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7640 		    hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7641 		    "Unknown"),
7642 		   netdev->dev_addr);
7643 
7644 	err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7645 	if (err)
7646 		strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7647 	if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7648 		e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7649 			   hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7650 		           part_str);
7651 	else
7652 		e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7653 			   hw->mac.type, hw->phy.type, part_str);
7654 
7655 	if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7656 		e_dev_warn("PCI-Express bandwidth available for this card is "
7657 			   "not sufficient for optimal performance.\n");
7658 		e_dev_warn("For optimal performance a x8 PCI-Express slot "
7659 			   "is required.\n");
7660 	}
7661 
7662 	/* reset the hardware with the new settings */
7663 	err = hw->mac.ops.start_hw(hw);
7664 
7665 	if (err == IXGBE_ERR_EEPROM_VERSION) {
7666 		/* We are running on a pre-production device, log a warning */
7667 		e_dev_warn("This device is a pre-production adapter/LOM. "
7668 			   "Please be aware there may be issues associated "
7669 			   "with your hardware.  If you are experiencing "
7670 			   "problems please contact your Intel or hardware "
7671 			   "representative who provided you with this "
7672 			   "hardware.\n");
7673 	}
7674 	strcpy(netdev->name, "eth%d");
7675 	err = register_netdev(netdev);
7676 	if (err)
7677 		goto err_register;
7678 
7679 	/* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7680 	if (hw->mac.ops.disable_tx_laser &&
7681 	    ((hw->phy.multispeed_fiber) ||
7682 	     ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7683 	      (hw->mac.type == ixgbe_mac_82599EB))))
7684 		hw->mac.ops.disable_tx_laser(hw);
7685 
7686 	/* carrier off reporting is important to ethtool even BEFORE open */
7687 	netif_carrier_off(netdev);
7688 
7689 #ifdef CONFIG_IXGBE_DCA
7690 	if (dca_add_requester(&pdev->dev) == 0) {
7691 		adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7692 		ixgbe_setup_dca(adapter);
7693 	}
7694 #endif
7695 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7696 		e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7697 		for (i = 0; i < adapter->num_vfs; i++)
7698 			ixgbe_vf_configuration(pdev, (i | 0x10000000));
7699 	}
7700 
7701 	/* firmware requires driver version to be 0xFFFFFFFF
7702 	 * since os does not support feature
7703 	 */
7704 	if (hw->mac.ops.set_fw_drv_ver)
7705 		hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
7706 					   0xFF);
7707 
7708 	/* add san mac addr to netdev */
7709 	ixgbe_add_sanmac_netdev(netdev);
7710 
7711 	e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
7712 	cards_found++;
7713 	return 0;
7714 
7715 err_register:
7716 	ixgbe_release_hw_control(adapter);
7717 	ixgbe_clear_interrupt_scheme(adapter);
7718 err_sw_init:
7719 err_eeprom:
7720 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7721 		ixgbe_disable_sriov(adapter);
7722 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7723 	iounmap(hw->hw_addr);
7724 err_ioremap:
7725 	free_netdev(netdev);
7726 err_alloc_etherdev:
7727 	pci_release_selected_regions(pdev,
7728 				     pci_select_bars(pdev, IORESOURCE_MEM));
7729 err_pci_reg:
7730 err_dma:
7731 	pci_disable_device(pdev);
7732 	return err;
7733 }
7734 
7735 /**
7736  * ixgbe_remove - Device Removal Routine
7737  * @pdev: PCI device information struct
7738  *
7739  * ixgbe_remove is called by the PCI subsystem to alert the driver
7740  * that it should release a PCI device.  The could be caused by a
7741  * Hot-Plug event, or because the driver is going to be removed from
7742  * memory.
7743  **/
7744 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7745 {
7746 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7747 	struct net_device *netdev = adapter->netdev;
7748 
7749 	set_bit(__IXGBE_DOWN, &adapter->state);
7750 	cancel_work_sync(&adapter->service_task);
7751 
7752 #ifdef CONFIG_IXGBE_DCA
7753 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7754 		adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7755 		dca_remove_requester(&pdev->dev);
7756 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7757 	}
7758 
7759 #endif
7760 #ifdef IXGBE_FCOE
7761 	if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7762 		ixgbe_cleanup_fcoe(adapter);
7763 
7764 #endif /* IXGBE_FCOE */
7765 
7766 	/* remove the added san mac */
7767 	ixgbe_del_sanmac_netdev(netdev);
7768 
7769 	if (netdev->reg_state == NETREG_REGISTERED)
7770 		unregister_netdev(netdev);
7771 
7772 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7773 		if (!(ixgbe_check_vf_assignment(adapter)))
7774 			ixgbe_disable_sriov(adapter);
7775 		else
7776 			e_dev_warn("Unloading driver while VFs are assigned "
7777 				   "- VFs will not be deallocated\n");
7778 	}
7779 
7780 	ixgbe_clear_interrupt_scheme(adapter);
7781 
7782 	ixgbe_release_hw_control(adapter);
7783 
7784 	iounmap(adapter->hw.hw_addr);
7785 	pci_release_selected_regions(pdev, pci_select_bars(pdev,
7786 				     IORESOURCE_MEM));
7787 
7788 	e_dev_info("complete\n");
7789 
7790 	free_netdev(netdev);
7791 
7792 	pci_disable_pcie_error_reporting(pdev);
7793 
7794 	pci_disable_device(pdev);
7795 }
7796 
7797 /**
7798  * ixgbe_io_error_detected - called when PCI error is detected
7799  * @pdev: Pointer to PCI device
7800  * @state: The current pci connection state
7801  *
7802  * This function is called after a PCI bus error affecting
7803  * this device has been detected.
7804  */
7805 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7806 						pci_channel_state_t state)
7807 {
7808 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7809 	struct net_device *netdev = adapter->netdev;
7810 
7811 #ifdef CONFIG_PCI_IOV
7812 	struct pci_dev *bdev, *vfdev;
7813 	u32 dw0, dw1, dw2, dw3;
7814 	int vf, pos;
7815 	u16 req_id, pf_func;
7816 
7817 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7818 	    adapter->num_vfs == 0)
7819 		goto skip_bad_vf_detection;
7820 
7821 	bdev = pdev->bus->self;
7822 	while (bdev && (bdev->pcie_type != PCI_EXP_TYPE_ROOT_PORT))
7823 		bdev = bdev->bus->self;
7824 
7825 	if (!bdev)
7826 		goto skip_bad_vf_detection;
7827 
7828 	pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
7829 	if (!pos)
7830 		goto skip_bad_vf_detection;
7831 
7832 	pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG, &dw0);
7833 	pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 4, &dw1);
7834 	pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 8, &dw2);
7835 	pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 12, &dw3);
7836 
7837 	req_id = dw1 >> 16;
7838 	/* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
7839 	if (!(req_id & 0x0080))
7840 		goto skip_bad_vf_detection;
7841 
7842 	pf_func = req_id & 0x01;
7843 	if ((pf_func & 1) == (pdev->devfn & 1)) {
7844 		unsigned int device_id;
7845 
7846 		vf = (req_id & 0x7F) >> 1;
7847 		e_dev_err("VF %d has caused a PCIe error\n", vf);
7848 		e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
7849 				"%8.8x\tdw3: %8.8x\n",
7850 		dw0, dw1, dw2, dw3);
7851 		switch (adapter->hw.mac.type) {
7852 		case ixgbe_mac_82599EB:
7853 			device_id = IXGBE_82599_VF_DEVICE_ID;
7854 			break;
7855 		case ixgbe_mac_X540:
7856 			device_id = IXGBE_X540_VF_DEVICE_ID;
7857 			break;
7858 		default:
7859 			device_id = 0;
7860 			break;
7861 		}
7862 
7863 		/* Find the pci device of the offending VF */
7864 		vfdev = pci_get_device(IXGBE_INTEL_VENDOR_ID, device_id, NULL);
7865 		while (vfdev) {
7866 			if (vfdev->devfn == (req_id & 0xFF))
7867 				break;
7868 			vfdev = pci_get_device(IXGBE_INTEL_VENDOR_ID,
7869 					       device_id, vfdev);
7870 		}
7871 		/*
7872 		 * There's a slim chance the VF could have been hot plugged,
7873 		 * so if it is no longer present we don't need to issue the
7874 		 * VFLR.  Just clean up the AER in that case.
7875 		 */
7876 		if (vfdev) {
7877 			e_dev_err("Issuing VFLR to VF %d\n", vf);
7878 			pci_write_config_dword(vfdev, 0xA8, 0x00008000);
7879 		}
7880 
7881 		pci_cleanup_aer_uncorrect_error_status(pdev);
7882 	}
7883 
7884 	/*
7885 	 * Even though the error may have occurred on the other port
7886 	 * we still need to increment the vf error reference count for
7887 	 * both ports because the I/O resume function will be called
7888 	 * for both of them.
7889 	 */
7890 	adapter->vferr_refcount++;
7891 
7892 	return PCI_ERS_RESULT_RECOVERED;
7893 
7894 skip_bad_vf_detection:
7895 #endif /* CONFIG_PCI_IOV */
7896 	netif_device_detach(netdev);
7897 
7898 	if (state == pci_channel_io_perm_failure)
7899 		return PCI_ERS_RESULT_DISCONNECT;
7900 
7901 	if (netif_running(netdev))
7902 		ixgbe_down(adapter);
7903 	pci_disable_device(pdev);
7904 
7905 	/* Request a slot reset. */
7906 	return PCI_ERS_RESULT_NEED_RESET;
7907 }
7908 
7909 /**
7910  * ixgbe_io_slot_reset - called after the pci bus has been reset.
7911  * @pdev: Pointer to PCI device
7912  *
7913  * Restart the card from scratch, as if from a cold-boot.
7914  */
7915 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7916 {
7917 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7918 	pci_ers_result_t result;
7919 	int err;
7920 
7921 	if (pci_enable_device_mem(pdev)) {
7922 		e_err(probe, "Cannot re-enable PCI device after reset.\n");
7923 		result = PCI_ERS_RESULT_DISCONNECT;
7924 	} else {
7925 		pci_set_master(pdev);
7926 		pci_restore_state(pdev);
7927 		pci_save_state(pdev);
7928 
7929 		pci_wake_from_d3(pdev, false);
7930 
7931 		ixgbe_reset(adapter);
7932 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7933 		result = PCI_ERS_RESULT_RECOVERED;
7934 	}
7935 
7936 	err = pci_cleanup_aer_uncorrect_error_status(pdev);
7937 	if (err) {
7938 		e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7939 			  "failed 0x%0x\n", err);
7940 		/* non-fatal, continue */
7941 	}
7942 
7943 	return result;
7944 }
7945 
7946 /**
7947  * ixgbe_io_resume - called when traffic can start flowing again.
7948  * @pdev: Pointer to PCI device
7949  *
7950  * This callback is called when the error recovery driver tells us that
7951  * its OK to resume normal operation.
7952  */
7953 static void ixgbe_io_resume(struct pci_dev *pdev)
7954 {
7955 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7956 	struct net_device *netdev = adapter->netdev;
7957 
7958 #ifdef CONFIG_PCI_IOV
7959 	if (adapter->vferr_refcount) {
7960 		e_info(drv, "Resuming after VF err\n");
7961 		adapter->vferr_refcount--;
7962 		return;
7963 	}
7964 
7965 #endif
7966 	if (netif_running(netdev))
7967 		ixgbe_up(adapter);
7968 
7969 	netif_device_attach(netdev);
7970 }
7971 
7972 static struct pci_error_handlers ixgbe_err_handler = {
7973 	.error_detected = ixgbe_io_error_detected,
7974 	.slot_reset = ixgbe_io_slot_reset,
7975 	.resume = ixgbe_io_resume,
7976 };
7977 
7978 static struct pci_driver ixgbe_driver = {
7979 	.name     = ixgbe_driver_name,
7980 	.id_table = ixgbe_pci_tbl,
7981 	.probe    = ixgbe_probe,
7982 	.remove   = __devexit_p(ixgbe_remove),
7983 #ifdef CONFIG_PM
7984 	.suspend  = ixgbe_suspend,
7985 	.resume   = ixgbe_resume,
7986 #endif
7987 	.shutdown = ixgbe_shutdown,
7988 	.err_handler = &ixgbe_err_handler
7989 };
7990 
7991 /**
7992  * ixgbe_init_module - Driver Registration Routine
7993  *
7994  * ixgbe_init_module is the first routine called when the driver is
7995  * loaded. All it does is register with the PCI subsystem.
7996  **/
7997 static int __init ixgbe_init_module(void)
7998 {
7999 	int ret;
8000 	pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
8001 	pr_info("%s\n", ixgbe_copyright);
8002 
8003 #ifdef CONFIG_IXGBE_DCA
8004 	dca_register_notify(&dca_notifier);
8005 #endif
8006 
8007 	ret = pci_register_driver(&ixgbe_driver);
8008 	return ret;
8009 }
8010 
8011 module_init(ixgbe_init_module);
8012 
8013 /**
8014  * ixgbe_exit_module - Driver Exit Cleanup Routine
8015  *
8016  * ixgbe_exit_module is called just before the driver is removed
8017  * from memory.
8018  **/
8019 static void __exit ixgbe_exit_module(void)
8020 {
8021 #ifdef CONFIG_IXGBE_DCA
8022 	dca_unregister_notify(&dca_notifier);
8023 #endif
8024 	pci_unregister_driver(&ixgbe_driver);
8025 	rcu_barrier(); /* Wait for completion of call_rcu()'s */
8026 }
8027 
8028 #ifdef CONFIG_IXGBE_DCA
8029 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
8030 			    void *p)
8031 {
8032 	int ret_val;
8033 
8034 	ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
8035 					 __ixgbe_notify_dca);
8036 
8037 	return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
8038 }
8039 
8040 #endif /* CONFIG_IXGBE_DCA */
8041 
8042 module_exit(ixgbe_exit_module);
8043 
8044 /* ixgbe_main.c */
8045