1 /*******************************************************************************
2 
3   Intel 10 Gigabit PCI Express Linux driver
4   Copyright(c) 1999 - 2013 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/if_macvlan.h>
48 #include <linux/if_bridge.h>
49 #include <linux/prefetch.h>
50 #include <scsi/fc/fc_fcoe.h>
51 
52 #include "ixgbe.h"
53 #include "ixgbe_common.h"
54 #include "ixgbe_dcb_82599.h"
55 #include "ixgbe_sriov.h"
56 
57 char ixgbe_driver_name[] = "ixgbe";
58 static const char ixgbe_driver_string[] =
59 			      "Intel(R) 10 Gigabit PCI Express Network Driver";
60 #ifdef IXGBE_FCOE
61 char ixgbe_default_device_descr[] =
62 			      "Intel(R) 10 Gigabit Network Connection";
63 #else
64 static char ixgbe_default_device_descr[] =
65 			      "Intel(R) 10 Gigabit Network Connection";
66 #endif
67 #define DRV_VERSION "3.19.1-k"
68 const char ixgbe_driver_version[] = DRV_VERSION;
69 static const char ixgbe_copyright[] =
70 				"Copyright (c) 1999-2013 Intel Corporation.";
71 
72 static const struct ixgbe_info *ixgbe_info_tbl[] = {
73 	[board_82598] = &ixgbe_82598_info,
74 	[board_82599] = &ixgbe_82599_info,
75 	[board_X540] = &ixgbe_X540_info,
76 };
77 
78 /* ixgbe_pci_tbl - PCI Device ID Table
79  *
80  * Wildcard entries (PCI_ANY_ID) should come last
81  * Last entry must be all 0s
82  *
83  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
84  *   Class, Class Mask, private data (not used) }
85  */
86 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
87 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
88 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
89 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
90 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
91 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
92 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
93 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
94 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
95 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
96 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
97 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
98 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
99 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
100 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
101 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
102 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
103 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
104 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
105 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
106 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
107 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
108 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
109 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
110 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
111 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
112 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
113 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_QSFP_SF_QP), board_82599 },
114 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
115 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
116 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
117 	/* required last entry */
118 	{0, }
119 };
120 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
121 
122 #ifdef CONFIG_IXGBE_DCA
123 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
124 			    void *p);
125 static struct notifier_block dca_notifier = {
126 	.notifier_call = ixgbe_notify_dca,
127 	.next          = NULL,
128 	.priority      = 0
129 };
130 #endif
131 
132 #ifdef CONFIG_PCI_IOV
133 static unsigned int max_vfs;
134 module_param(max_vfs, uint, 0);
135 MODULE_PARM_DESC(max_vfs,
136 		 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63. (Deprecated)");
137 #endif /* CONFIG_PCI_IOV */
138 
139 static unsigned int allow_unsupported_sfp;
140 module_param(allow_unsupported_sfp, uint, 0);
141 MODULE_PARM_DESC(allow_unsupported_sfp,
142 		 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
143 
144 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
145 static int debug = -1;
146 module_param(debug, int, 0);
147 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
148 
149 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
150 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
151 MODULE_LICENSE("GPL");
152 MODULE_VERSION(DRV_VERSION);
153 
154 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
155 					  u32 reg, u16 *value)
156 {
157 	struct pci_dev *parent_dev;
158 	struct pci_bus *parent_bus;
159 
160 	parent_bus = adapter->pdev->bus->parent;
161 	if (!parent_bus)
162 		return -1;
163 
164 	parent_dev = parent_bus->self;
165 	if (!parent_dev)
166 		return -1;
167 
168 	if (!pci_is_pcie(parent_dev))
169 		return -1;
170 
171 	pcie_capability_read_word(parent_dev, reg, value);
172 	return 0;
173 }
174 
175 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
176 {
177 	struct ixgbe_hw *hw = &adapter->hw;
178 	u16 link_status = 0;
179 	int err;
180 
181 	hw->bus.type = ixgbe_bus_type_pci_express;
182 
183 	/* Get the negotiated link width and speed from PCI config space of the
184 	 * parent, as this device is behind a switch
185 	 */
186 	err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
187 
188 	/* assume caller will handle error case */
189 	if (err)
190 		return err;
191 
192 	hw->bus.width = ixgbe_convert_bus_width(link_status);
193 	hw->bus.speed = ixgbe_convert_bus_speed(link_status);
194 
195 	return 0;
196 }
197 
198 /**
199  * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
200  * @hw: hw specific details
201  *
202  * This function is used by probe to determine whether a device's PCI-Express
203  * bandwidth details should be gathered from the parent bus instead of from the
204  * device. Used to ensure that various locations all have the correct device ID
205  * checks.
206  */
207 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
208 {
209 	switch (hw->device_id) {
210 	case IXGBE_DEV_ID_82599_SFP_SF_QP:
211 	case IXGBE_DEV_ID_82599_QSFP_SF_QP:
212 		return true;
213 	default:
214 		return false;
215 	}
216 }
217 
218 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
219 				     int expected_gts)
220 {
221 	int max_gts = 0;
222 	enum pci_bus_speed speed = PCI_SPEED_UNKNOWN;
223 	enum pcie_link_width width = PCIE_LNK_WIDTH_UNKNOWN;
224 	struct pci_dev *pdev;
225 
226 	/* determine whether to use the the parent device
227 	 */
228 	if (ixgbe_pcie_from_parent(&adapter->hw))
229 		pdev = adapter->pdev->bus->parent->self;
230 	else
231 		pdev = adapter->pdev;
232 
233 	if (pcie_get_minimum_link(pdev, &speed, &width) ||
234 	    speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN) {
235 		e_dev_warn("Unable to determine PCI Express bandwidth.\n");
236 		return;
237 	}
238 
239 	switch (speed) {
240 	case PCIE_SPEED_2_5GT:
241 		/* 8b/10b encoding reduces max throughput by 20% */
242 		max_gts = 2 * width;
243 		break;
244 	case PCIE_SPEED_5_0GT:
245 		/* 8b/10b encoding reduces max throughput by 20% */
246 		max_gts = 4 * width;
247 		break;
248 	case PCIE_SPEED_8_0GT:
249 		/* 128b/130b encoding reduces throughput by less than 2% */
250 		max_gts = 8 * width;
251 		break;
252 	default:
253 		e_dev_warn("Unable to determine PCI Express bandwidth.\n");
254 		return;
255 	}
256 
257 	e_dev_info("PCI Express bandwidth of %dGT/s available\n",
258 		   max_gts);
259 	e_dev_info("(Speed:%s, Width: x%d, Encoding Loss:%s)\n",
260 		   (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" :
261 		    speed == PCIE_SPEED_5_0GT ? "5.0GT/s" :
262 		    speed == PCIE_SPEED_2_5GT ? "2.5GT/s" :
263 		    "Unknown"),
264 		   width,
265 		   (speed == PCIE_SPEED_2_5GT ? "20%" :
266 		    speed == PCIE_SPEED_5_0GT ? "20%" :
267 		    speed == PCIE_SPEED_8_0GT ? "<2%" :
268 		    "Unknown"));
269 
270 	if (max_gts < expected_gts) {
271 		e_dev_warn("This is not sufficient for optimal performance of this card.\n");
272 		e_dev_warn("For optimal performance, at least %dGT/s of bandwidth is required.\n",
273 			expected_gts);
274 		e_dev_warn("A slot with more lanes and/or higher speed is suggested.\n");
275 	}
276 }
277 
278 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
279 {
280 	if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
281 	    !test_bit(__IXGBE_REMOVING, &adapter->state) &&
282 	    !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
283 		schedule_work(&adapter->service_task);
284 }
285 
286 static void ixgbe_remove_adapter(struct ixgbe_hw *hw)
287 {
288 	struct ixgbe_adapter *adapter = hw->back;
289 
290 	if (!hw->hw_addr)
291 		return;
292 	hw->hw_addr = NULL;
293 	e_dev_err("Adapter removed\n");
294 	ixgbe_service_event_schedule(adapter);
295 }
296 
297 void ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg)
298 {
299 	u32 value;
300 
301 	/* The following check not only optimizes a bit by not
302 	 * performing a read on the status register when the
303 	 * register just read was a status register read that
304 	 * returned IXGBE_FAILED_READ_REG. It also blocks any
305 	 * potential recursion.
306 	 */
307 	if (reg == IXGBE_STATUS) {
308 		ixgbe_remove_adapter(hw);
309 		return;
310 	}
311 	value = ixgbe_read_reg(hw, IXGBE_STATUS);
312 	if (value == IXGBE_FAILED_READ_REG)
313 		ixgbe_remove_adapter(hw);
314 }
315 
316 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
317 {
318 	BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
319 
320 	/* flush memory to make sure state is correct before next watchdog */
321 	smp_mb__before_clear_bit();
322 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
323 }
324 
325 struct ixgbe_reg_info {
326 	u32 ofs;
327 	char *name;
328 };
329 
330 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
331 
332 	/* General Registers */
333 	{IXGBE_CTRL, "CTRL"},
334 	{IXGBE_STATUS, "STATUS"},
335 	{IXGBE_CTRL_EXT, "CTRL_EXT"},
336 
337 	/* Interrupt Registers */
338 	{IXGBE_EICR, "EICR"},
339 
340 	/* RX Registers */
341 	{IXGBE_SRRCTL(0), "SRRCTL"},
342 	{IXGBE_DCA_RXCTRL(0), "DRXCTL"},
343 	{IXGBE_RDLEN(0), "RDLEN"},
344 	{IXGBE_RDH(0), "RDH"},
345 	{IXGBE_RDT(0), "RDT"},
346 	{IXGBE_RXDCTL(0), "RXDCTL"},
347 	{IXGBE_RDBAL(0), "RDBAL"},
348 	{IXGBE_RDBAH(0), "RDBAH"},
349 
350 	/* TX Registers */
351 	{IXGBE_TDBAL(0), "TDBAL"},
352 	{IXGBE_TDBAH(0), "TDBAH"},
353 	{IXGBE_TDLEN(0), "TDLEN"},
354 	{IXGBE_TDH(0), "TDH"},
355 	{IXGBE_TDT(0), "TDT"},
356 	{IXGBE_TXDCTL(0), "TXDCTL"},
357 
358 	/* List Terminator */
359 	{}
360 };
361 
362 
363 /*
364  * ixgbe_regdump - register printout routine
365  */
366 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
367 {
368 	int i = 0, j = 0;
369 	char rname[16];
370 	u32 regs[64];
371 
372 	switch (reginfo->ofs) {
373 	case IXGBE_SRRCTL(0):
374 		for (i = 0; i < 64; i++)
375 			regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
376 		break;
377 	case IXGBE_DCA_RXCTRL(0):
378 		for (i = 0; i < 64; i++)
379 			regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
380 		break;
381 	case IXGBE_RDLEN(0):
382 		for (i = 0; i < 64; i++)
383 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
384 		break;
385 	case IXGBE_RDH(0):
386 		for (i = 0; i < 64; i++)
387 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
388 		break;
389 	case IXGBE_RDT(0):
390 		for (i = 0; i < 64; i++)
391 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
392 		break;
393 	case IXGBE_RXDCTL(0):
394 		for (i = 0; i < 64; i++)
395 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
396 		break;
397 	case IXGBE_RDBAL(0):
398 		for (i = 0; i < 64; i++)
399 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
400 		break;
401 	case IXGBE_RDBAH(0):
402 		for (i = 0; i < 64; i++)
403 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
404 		break;
405 	case IXGBE_TDBAL(0):
406 		for (i = 0; i < 64; i++)
407 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
408 		break;
409 	case IXGBE_TDBAH(0):
410 		for (i = 0; i < 64; i++)
411 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
412 		break;
413 	case IXGBE_TDLEN(0):
414 		for (i = 0; i < 64; i++)
415 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
416 		break;
417 	case IXGBE_TDH(0):
418 		for (i = 0; i < 64; i++)
419 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
420 		break;
421 	case IXGBE_TDT(0):
422 		for (i = 0; i < 64; i++)
423 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
424 		break;
425 	case IXGBE_TXDCTL(0):
426 		for (i = 0; i < 64; i++)
427 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
428 		break;
429 	default:
430 		pr_info("%-15s %08x\n", reginfo->name,
431 			IXGBE_READ_REG(hw, reginfo->ofs));
432 		return;
433 	}
434 
435 	for (i = 0; i < 8; i++) {
436 		snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
437 		pr_err("%-15s", rname);
438 		for (j = 0; j < 8; j++)
439 			pr_cont(" %08x", regs[i*8+j]);
440 		pr_cont("\n");
441 	}
442 
443 }
444 
445 /*
446  * ixgbe_dump - Print registers, tx-rings and rx-rings
447  */
448 static void ixgbe_dump(struct ixgbe_adapter *adapter)
449 {
450 	struct net_device *netdev = adapter->netdev;
451 	struct ixgbe_hw *hw = &adapter->hw;
452 	struct ixgbe_reg_info *reginfo;
453 	int n = 0;
454 	struct ixgbe_ring *tx_ring;
455 	struct ixgbe_tx_buffer *tx_buffer;
456 	union ixgbe_adv_tx_desc *tx_desc;
457 	struct my_u0 { u64 a; u64 b; } *u0;
458 	struct ixgbe_ring *rx_ring;
459 	union ixgbe_adv_rx_desc *rx_desc;
460 	struct ixgbe_rx_buffer *rx_buffer_info;
461 	u32 staterr;
462 	int i = 0;
463 
464 	if (!netif_msg_hw(adapter))
465 		return;
466 
467 	/* Print netdevice Info */
468 	if (netdev) {
469 		dev_info(&adapter->pdev->dev, "Net device Info\n");
470 		pr_info("Device Name     state            "
471 			"trans_start      last_rx\n");
472 		pr_info("%-15s %016lX %016lX %016lX\n",
473 			netdev->name,
474 			netdev->state,
475 			netdev->trans_start,
476 			netdev->last_rx);
477 	}
478 
479 	/* Print Registers */
480 	dev_info(&adapter->pdev->dev, "Register Dump\n");
481 	pr_info(" Register Name   Value\n");
482 	for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
483 	     reginfo->name; reginfo++) {
484 		ixgbe_regdump(hw, reginfo);
485 	}
486 
487 	/* Print TX Ring Summary */
488 	if (!netdev || !netif_running(netdev))
489 		goto exit;
490 
491 	dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
492 	pr_info(" %s     %s              %s        %s\n",
493 		"Queue [NTU] [NTC] [bi(ntc)->dma  ]",
494 		"leng", "ntw", "timestamp");
495 	for (n = 0; n < adapter->num_tx_queues; n++) {
496 		tx_ring = adapter->tx_ring[n];
497 		tx_buffer = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
498 		pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
499 			   n, tx_ring->next_to_use, tx_ring->next_to_clean,
500 			   (u64)dma_unmap_addr(tx_buffer, dma),
501 			   dma_unmap_len(tx_buffer, len),
502 			   tx_buffer->next_to_watch,
503 			   (u64)tx_buffer->time_stamp);
504 	}
505 
506 	/* Print TX Rings */
507 	if (!netif_msg_tx_done(adapter))
508 		goto rx_ring_summary;
509 
510 	dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
511 
512 	/* Transmit Descriptor Formats
513 	 *
514 	 * 82598 Advanced Transmit Descriptor
515 	 *   +--------------------------------------------------------------+
516 	 * 0 |         Buffer Address [63:0]                                |
517 	 *   +--------------------------------------------------------------+
518 	 * 8 |  PAYLEN  | POPTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
519 	 *   +--------------------------------------------------------------+
520 	 *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
521 	 *
522 	 * 82598 Advanced Transmit Descriptor (Write-Back Format)
523 	 *   +--------------------------------------------------------------+
524 	 * 0 |                          RSV [63:0]                          |
525 	 *   +--------------------------------------------------------------+
526 	 * 8 |            RSV           |  STA  |          NXTSEQ           |
527 	 *   +--------------------------------------------------------------+
528 	 *   63                       36 35   32 31                         0
529 	 *
530 	 * 82599+ Advanced Transmit Descriptor
531 	 *   +--------------------------------------------------------------+
532 	 * 0 |         Buffer Address [63:0]                                |
533 	 *   +--------------------------------------------------------------+
534 	 * 8 |PAYLEN  |POPTS|CC|IDX  |STA  |DCMD  |DTYP |MAC  |RSV  |DTALEN |
535 	 *   +--------------------------------------------------------------+
536 	 *   63     46 45 40 39 38 36 35 32 31  24 23 20 19 18 17 16 15     0
537 	 *
538 	 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
539 	 *   +--------------------------------------------------------------+
540 	 * 0 |                          RSV [63:0]                          |
541 	 *   +--------------------------------------------------------------+
542 	 * 8 |            RSV           |  STA  |           RSV             |
543 	 *   +--------------------------------------------------------------+
544 	 *   63                       36 35   32 31                         0
545 	 */
546 
547 	for (n = 0; n < adapter->num_tx_queues; n++) {
548 		tx_ring = adapter->tx_ring[n];
549 		pr_info("------------------------------------\n");
550 		pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
551 		pr_info("------------------------------------\n");
552 		pr_info("%s%s    %s              %s        %s          %s\n",
553 			"T [desc]     [address 63:0  ] ",
554 			"[PlPOIdStDDt Ln] [bi->dma       ] ",
555 			"leng", "ntw", "timestamp", "bi->skb");
556 
557 		for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
558 			tx_desc = IXGBE_TX_DESC(tx_ring, i);
559 			tx_buffer = &tx_ring->tx_buffer_info[i];
560 			u0 = (struct my_u0 *)tx_desc;
561 			if (dma_unmap_len(tx_buffer, len) > 0) {
562 				pr_info("T [0x%03X]    %016llX %016llX %016llX %08X %p %016llX %p",
563 					i,
564 					le64_to_cpu(u0->a),
565 					le64_to_cpu(u0->b),
566 					(u64)dma_unmap_addr(tx_buffer, dma),
567 					dma_unmap_len(tx_buffer, len),
568 					tx_buffer->next_to_watch,
569 					(u64)tx_buffer->time_stamp,
570 					tx_buffer->skb);
571 				if (i == tx_ring->next_to_use &&
572 					i == tx_ring->next_to_clean)
573 					pr_cont(" NTC/U\n");
574 				else if (i == tx_ring->next_to_use)
575 					pr_cont(" NTU\n");
576 				else if (i == tx_ring->next_to_clean)
577 					pr_cont(" NTC\n");
578 				else
579 					pr_cont("\n");
580 
581 				if (netif_msg_pktdata(adapter) &&
582 				    tx_buffer->skb)
583 					print_hex_dump(KERN_INFO, "",
584 						DUMP_PREFIX_ADDRESS, 16, 1,
585 						tx_buffer->skb->data,
586 						dma_unmap_len(tx_buffer, len),
587 						true);
588 			}
589 		}
590 	}
591 
592 	/* Print RX Rings Summary */
593 rx_ring_summary:
594 	dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
595 	pr_info("Queue [NTU] [NTC]\n");
596 	for (n = 0; n < adapter->num_rx_queues; n++) {
597 		rx_ring = adapter->rx_ring[n];
598 		pr_info("%5d %5X %5X\n",
599 			n, rx_ring->next_to_use, rx_ring->next_to_clean);
600 	}
601 
602 	/* Print RX Rings */
603 	if (!netif_msg_rx_status(adapter))
604 		goto exit;
605 
606 	dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
607 
608 	/* Receive Descriptor Formats
609 	 *
610 	 * 82598 Advanced Receive Descriptor (Read) Format
611 	 *    63                                           1        0
612 	 *    +-----------------------------------------------------+
613 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
614 	 *    +----------------------------------------------+------+
615 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
616 	 *    +-----------------------------------------------------+
617 	 *
618 	 *
619 	 * 82598 Advanced Receive Descriptor (Write-Back) Format
620 	 *
621 	 *   63       48 47    32 31  30      21 20 16 15   4 3     0
622 	 *   +------------------------------------------------------+
623 	 * 0 |       RSS Hash /  |SPH| HDR_LEN  | RSV |Packet|  RSS |
624 	 *   | Packet   | IP     |   |          |     | Type | Type |
625 	 *   | Checksum | Ident  |   |          |     |      |      |
626 	 *   +------------------------------------------------------+
627 	 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
628 	 *   +------------------------------------------------------+
629 	 *   63       48 47    32 31            20 19               0
630 	 *
631 	 * 82599+ Advanced Receive Descriptor (Read) Format
632 	 *    63                                           1        0
633 	 *    +-----------------------------------------------------+
634 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
635 	 *    +----------------------------------------------+------+
636 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
637 	 *    +-----------------------------------------------------+
638 	 *
639 	 *
640 	 * 82599+ Advanced Receive Descriptor (Write-Back) Format
641 	 *
642 	 *   63       48 47    32 31  30      21 20 17 16   4 3     0
643 	 *   +------------------------------------------------------+
644 	 * 0 |RSS / Frag Checksum|SPH| HDR_LEN  |RSC- |Packet|  RSS |
645 	 *   |/ RTT / PCoE_PARAM |   |          | CNT | Type | Type |
646 	 *   |/ Flow Dir Flt ID  |   |          |     |      |      |
647 	 *   +------------------------------------------------------+
648 	 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
649 	 *   +------------------------------------------------------+
650 	 *   63       48 47    32 31          20 19                 0
651 	 */
652 
653 	for (n = 0; n < adapter->num_rx_queues; n++) {
654 		rx_ring = adapter->rx_ring[n];
655 		pr_info("------------------------------------\n");
656 		pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
657 		pr_info("------------------------------------\n");
658 		pr_info("%s%s%s",
659 			"R  [desc]      [ PktBuf     A0] ",
660 			"[  HeadBuf   DD] [bi->dma       ] [bi->skb       ] ",
661 			"<-- Adv Rx Read format\n");
662 		pr_info("%s%s%s",
663 			"RWB[desc]      [PcsmIpSHl PtRs] ",
664 			"[vl er S cks ln] ---------------- [bi->skb       ] ",
665 			"<-- Adv Rx Write-Back format\n");
666 
667 		for (i = 0; i < rx_ring->count; i++) {
668 			rx_buffer_info = &rx_ring->rx_buffer_info[i];
669 			rx_desc = IXGBE_RX_DESC(rx_ring, i);
670 			u0 = (struct my_u0 *)rx_desc;
671 			staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
672 			if (staterr & IXGBE_RXD_STAT_DD) {
673 				/* Descriptor Done */
674 				pr_info("RWB[0x%03X]     %016llX "
675 					"%016llX ---------------- %p", i,
676 					le64_to_cpu(u0->a),
677 					le64_to_cpu(u0->b),
678 					rx_buffer_info->skb);
679 			} else {
680 				pr_info("R  [0x%03X]     %016llX "
681 					"%016llX %016llX %p", i,
682 					le64_to_cpu(u0->a),
683 					le64_to_cpu(u0->b),
684 					(u64)rx_buffer_info->dma,
685 					rx_buffer_info->skb);
686 
687 				if (netif_msg_pktdata(adapter) &&
688 				    rx_buffer_info->dma) {
689 					print_hex_dump(KERN_INFO, "",
690 					   DUMP_PREFIX_ADDRESS, 16, 1,
691 					   page_address(rx_buffer_info->page) +
692 						    rx_buffer_info->page_offset,
693 					   ixgbe_rx_bufsz(rx_ring), true);
694 				}
695 			}
696 
697 			if (i == rx_ring->next_to_use)
698 				pr_cont(" NTU\n");
699 			else if (i == rx_ring->next_to_clean)
700 				pr_cont(" NTC\n");
701 			else
702 				pr_cont("\n");
703 
704 		}
705 	}
706 
707 exit:
708 	return;
709 }
710 
711 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
712 {
713 	u32 ctrl_ext;
714 
715 	/* Let firmware take over control of h/w */
716 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
717 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
718 			ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
719 }
720 
721 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
722 {
723 	u32 ctrl_ext;
724 
725 	/* Let firmware know the driver has taken over */
726 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
727 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
728 			ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
729 }
730 
731 /**
732  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
733  * @adapter: pointer to adapter struct
734  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
735  * @queue: queue to map the corresponding interrupt to
736  * @msix_vector: the vector to map to the corresponding queue
737  *
738  */
739 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
740 			   u8 queue, u8 msix_vector)
741 {
742 	u32 ivar, index;
743 	struct ixgbe_hw *hw = &adapter->hw;
744 	switch (hw->mac.type) {
745 	case ixgbe_mac_82598EB:
746 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
747 		if (direction == -1)
748 			direction = 0;
749 		index = (((direction * 64) + queue) >> 2) & 0x1F;
750 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
751 		ivar &= ~(0xFF << (8 * (queue & 0x3)));
752 		ivar |= (msix_vector << (8 * (queue & 0x3)));
753 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
754 		break;
755 	case ixgbe_mac_82599EB:
756 	case ixgbe_mac_X540:
757 		if (direction == -1) {
758 			/* other causes */
759 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
760 			index = ((queue & 1) * 8);
761 			ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
762 			ivar &= ~(0xFF << index);
763 			ivar |= (msix_vector << index);
764 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
765 			break;
766 		} else {
767 			/* tx or rx causes */
768 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
769 			index = ((16 * (queue & 1)) + (8 * direction));
770 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
771 			ivar &= ~(0xFF << index);
772 			ivar |= (msix_vector << index);
773 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
774 			break;
775 		}
776 	default:
777 		break;
778 	}
779 }
780 
781 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
782 					  u64 qmask)
783 {
784 	u32 mask;
785 
786 	switch (adapter->hw.mac.type) {
787 	case ixgbe_mac_82598EB:
788 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
789 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
790 		break;
791 	case ixgbe_mac_82599EB:
792 	case ixgbe_mac_X540:
793 		mask = (qmask & 0xFFFFFFFF);
794 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
795 		mask = (qmask >> 32);
796 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
797 		break;
798 	default:
799 		break;
800 	}
801 }
802 
803 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *ring,
804 				      struct ixgbe_tx_buffer *tx_buffer)
805 {
806 	if (tx_buffer->skb) {
807 		dev_kfree_skb_any(tx_buffer->skb);
808 		if (dma_unmap_len(tx_buffer, len))
809 			dma_unmap_single(ring->dev,
810 					 dma_unmap_addr(tx_buffer, dma),
811 					 dma_unmap_len(tx_buffer, len),
812 					 DMA_TO_DEVICE);
813 	} else if (dma_unmap_len(tx_buffer, len)) {
814 		dma_unmap_page(ring->dev,
815 			       dma_unmap_addr(tx_buffer, dma),
816 			       dma_unmap_len(tx_buffer, len),
817 			       DMA_TO_DEVICE);
818 	}
819 	tx_buffer->next_to_watch = NULL;
820 	tx_buffer->skb = NULL;
821 	dma_unmap_len_set(tx_buffer, len, 0);
822 	/* tx_buffer must be completely set up in the transmit path */
823 }
824 
825 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
826 {
827 	struct ixgbe_hw *hw = &adapter->hw;
828 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
829 	int i;
830 	u32 data;
831 
832 	if ((hw->fc.current_mode != ixgbe_fc_full) &&
833 	    (hw->fc.current_mode != ixgbe_fc_rx_pause))
834 		return;
835 
836 	switch (hw->mac.type) {
837 	case ixgbe_mac_82598EB:
838 		data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
839 		break;
840 	default:
841 		data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
842 	}
843 	hwstats->lxoffrxc += data;
844 
845 	/* refill credits (no tx hang) if we received xoff */
846 	if (!data)
847 		return;
848 
849 	for (i = 0; i < adapter->num_tx_queues; i++)
850 		clear_bit(__IXGBE_HANG_CHECK_ARMED,
851 			  &adapter->tx_ring[i]->state);
852 }
853 
854 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
855 {
856 	struct ixgbe_hw *hw = &adapter->hw;
857 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
858 	u32 xoff[8] = {0};
859 	u8 tc;
860 	int i;
861 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
862 
863 	if (adapter->ixgbe_ieee_pfc)
864 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
865 
866 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
867 		ixgbe_update_xoff_rx_lfc(adapter);
868 		return;
869 	}
870 
871 	/* update stats for each tc, only valid with PFC enabled */
872 	for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
873 		u32 pxoffrxc;
874 
875 		switch (hw->mac.type) {
876 		case ixgbe_mac_82598EB:
877 			pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
878 			break;
879 		default:
880 			pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
881 		}
882 		hwstats->pxoffrxc[i] += pxoffrxc;
883 		/* Get the TC for given UP */
884 		tc = netdev_get_prio_tc_map(adapter->netdev, i);
885 		xoff[tc] += pxoffrxc;
886 	}
887 
888 	/* disarm tx queues that have received xoff frames */
889 	for (i = 0; i < adapter->num_tx_queues; i++) {
890 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
891 
892 		tc = tx_ring->dcb_tc;
893 		if (xoff[tc])
894 			clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
895 	}
896 }
897 
898 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
899 {
900 	return ring->stats.packets;
901 }
902 
903 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
904 {
905 	struct ixgbe_adapter *adapter;
906 	struct ixgbe_hw *hw;
907 	u32 head, tail;
908 
909 	if (ring->l2_accel_priv)
910 		adapter = ring->l2_accel_priv->real_adapter;
911 	else
912 		adapter = netdev_priv(ring->netdev);
913 
914 	hw = &adapter->hw;
915 	head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
916 	tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
917 
918 	if (head != tail)
919 		return (head < tail) ?
920 			tail - head : (tail + ring->count - head);
921 
922 	return 0;
923 }
924 
925 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
926 {
927 	u32 tx_done = ixgbe_get_tx_completed(tx_ring);
928 	u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
929 	u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
930 	bool ret = false;
931 
932 	clear_check_for_tx_hang(tx_ring);
933 
934 	/*
935 	 * Check for a hung queue, but be thorough. This verifies
936 	 * that a transmit has been completed since the previous
937 	 * check AND there is at least one packet pending. The
938 	 * ARMED bit is set to indicate a potential hang. The
939 	 * bit is cleared if a pause frame is received to remove
940 	 * false hang detection due to PFC or 802.3x frames. By
941 	 * requiring this to fail twice we avoid races with
942 	 * pfc clearing the ARMED bit and conditions where we
943 	 * run the check_tx_hang logic with a transmit completion
944 	 * pending but without time to complete it yet.
945 	 */
946 	if ((tx_done_old == tx_done) && tx_pending) {
947 		/* make sure it is true for two checks in a row */
948 		ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
949 				       &tx_ring->state);
950 	} else {
951 		/* update completed stats and continue */
952 		tx_ring->tx_stats.tx_done_old = tx_done;
953 		/* reset the countdown */
954 		clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
955 	}
956 
957 	return ret;
958 }
959 
960 /**
961  * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
962  * @adapter: driver private struct
963  **/
964 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
965 {
966 
967 	/* Do the reset outside of interrupt context */
968 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
969 		adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
970 		e_warn(drv, "initiating reset due to tx timeout\n");
971 		ixgbe_service_event_schedule(adapter);
972 	}
973 }
974 
975 /**
976  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
977  * @q_vector: structure containing interrupt and ring information
978  * @tx_ring: tx ring to clean
979  **/
980 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
981 			       struct ixgbe_ring *tx_ring)
982 {
983 	struct ixgbe_adapter *adapter = q_vector->adapter;
984 	struct ixgbe_tx_buffer *tx_buffer;
985 	union ixgbe_adv_tx_desc *tx_desc;
986 	unsigned int total_bytes = 0, total_packets = 0;
987 	unsigned int budget = q_vector->tx.work_limit;
988 	unsigned int i = tx_ring->next_to_clean;
989 
990 	if (test_bit(__IXGBE_DOWN, &adapter->state))
991 		return true;
992 
993 	tx_buffer = &tx_ring->tx_buffer_info[i];
994 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
995 	i -= tx_ring->count;
996 
997 	do {
998 		union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
999 
1000 		/* if next_to_watch is not set then there is no work pending */
1001 		if (!eop_desc)
1002 			break;
1003 
1004 		/* prevent any other reads prior to eop_desc */
1005 		read_barrier_depends();
1006 
1007 		/* if DD is not set pending work has not been completed */
1008 		if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
1009 			break;
1010 
1011 		/* clear next_to_watch to prevent false hangs */
1012 		tx_buffer->next_to_watch = NULL;
1013 
1014 		/* update the statistics for this packet */
1015 		total_bytes += tx_buffer->bytecount;
1016 		total_packets += tx_buffer->gso_segs;
1017 
1018 		/* free the skb */
1019 		dev_kfree_skb_any(tx_buffer->skb);
1020 
1021 		/* unmap skb header data */
1022 		dma_unmap_single(tx_ring->dev,
1023 				 dma_unmap_addr(tx_buffer, dma),
1024 				 dma_unmap_len(tx_buffer, len),
1025 				 DMA_TO_DEVICE);
1026 
1027 		/* clear tx_buffer data */
1028 		tx_buffer->skb = NULL;
1029 		dma_unmap_len_set(tx_buffer, len, 0);
1030 
1031 		/* unmap remaining buffers */
1032 		while (tx_desc != eop_desc) {
1033 			tx_buffer++;
1034 			tx_desc++;
1035 			i++;
1036 			if (unlikely(!i)) {
1037 				i -= tx_ring->count;
1038 				tx_buffer = tx_ring->tx_buffer_info;
1039 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1040 			}
1041 
1042 			/* unmap any remaining paged data */
1043 			if (dma_unmap_len(tx_buffer, len)) {
1044 				dma_unmap_page(tx_ring->dev,
1045 					       dma_unmap_addr(tx_buffer, dma),
1046 					       dma_unmap_len(tx_buffer, len),
1047 					       DMA_TO_DEVICE);
1048 				dma_unmap_len_set(tx_buffer, len, 0);
1049 			}
1050 		}
1051 
1052 		/* move us one more past the eop_desc for start of next pkt */
1053 		tx_buffer++;
1054 		tx_desc++;
1055 		i++;
1056 		if (unlikely(!i)) {
1057 			i -= tx_ring->count;
1058 			tx_buffer = tx_ring->tx_buffer_info;
1059 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1060 		}
1061 
1062 		/* issue prefetch for next Tx descriptor */
1063 		prefetch(tx_desc);
1064 
1065 		/* update budget accounting */
1066 		budget--;
1067 	} while (likely(budget));
1068 
1069 	i += tx_ring->count;
1070 	tx_ring->next_to_clean = i;
1071 	u64_stats_update_begin(&tx_ring->syncp);
1072 	tx_ring->stats.bytes += total_bytes;
1073 	tx_ring->stats.packets += total_packets;
1074 	u64_stats_update_end(&tx_ring->syncp);
1075 	q_vector->tx.total_bytes += total_bytes;
1076 	q_vector->tx.total_packets += total_packets;
1077 
1078 	if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1079 		/* schedule immediate reset if we believe we hung */
1080 		struct ixgbe_hw *hw = &adapter->hw;
1081 		e_err(drv, "Detected Tx Unit Hang\n"
1082 			"  Tx Queue             <%d>\n"
1083 			"  TDH, TDT             <%x>, <%x>\n"
1084 			"  next_to_use          <%x>\n"
1085 			"  next_to_clean        <%x>\n"
1086 			"tx_buffer_info[next_to_clean]\n"
1087 			"  time_stamp           <%lx>\n"
1088 			"  jiffies              <%lx>\n",
1089 			tx_ring->queue_index,
1090 			IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1091 			IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1092 			tx_ring->next_to_use, i,
1093 			tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1094 
1095 		netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
1096 
1097 		e_info(probe,
1098 		       "tx hang %d detected on queue %d, resetting adapter\n",
1099 			adapter->tx_timeout_count + 1, tx_ring->queue_index);
1100 
1101 		/* schedule immediate reset if we believe we hung */
1102 		ixgbe_tx_timeout_reset(adapter);
1103 
1104 		/* the adapter is about to reset, no point in enabling stuff */
1105 		return true;
1106 	}
1107 
1108 	netdev_tx_completed_queue(txring_txq(tx_ring),
1109 				  total_packets, total_bytes);
1110 
1111 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1112 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1113 		     (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1114 		/* Make sure that anybody stopping the queue after this
1115 		 * sees the new next_to_clean.
1116 		 */
1117 		smp_mb();
1118 		if (__netif_subqueue_stopped(tx_ring->netdev,
1119 					     tx_ring->queue_index)
1120 		    && !test_bit(__IXGBE_DOWN, &adapter->state)) {
1121 			netif_wake_subqueue(tx_ring->netdev,
1122 					    tx_ring->queue_index);
1123 			++tx_ring->tx_stats.restart_queue;
1124 		}
1125 	}
1126 
1127 	return !!budget;
1128 }
1129 
1130 #ifdef CONFIG_IXGBE_DCA
1131 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1132 				struct ixgbe_ring *tx_ring,
1133 				int cpu)
1134 {
1135 	struct ixgbe_hw *hw = &adapter->hw;
1136 	u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
1137 	u16 reg_offset;
1138 
1139 	switch (hw->mac.type) {
1140 	case ixgbe_mac_82598EB:
1141 		reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1142 		break;
1143 	case ixgbe_mac_82599EB:
1144 	case ixgbe_mac_X540:
1145 		reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1146 		txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1147 		break;
1148 	default:
1149 		/* for unknown hardware do not write register */
1150 		return;
1151 	}
1152 
1153 	/*
1154 	 * We can enable relaxed ordering for reads, but not writes when
1155 	 * DCA is enabled.  This is due to a known issue in some chipsets
1156 	 * which will cause the DCA tag to be cleared.
1157 	 */
1158 	txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1159 		  IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1160 		  IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1161 
1162 	IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1163 }
1164 
1165 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1166 				struct ixgbe_ring *rx_ring,
1167 				int cpu)
1168 {
1169 	struct ixgbe_hw *hw = &adapter->hw;
1170 	u32 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1171 	u8 reg_idx = rx_ring->reg_idx;
1172 
1173 
1174 	switch (hw->mac.type) {
1175 	case ixgbe_mac_82599EB:
1176 	case ixgbe_mac_X540:
1177 		rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1178 		break;
1179 	default:
1180 		break;
1181 	}
1182 
1183 	/*
1184 	 * We can enable relaxed ordering for reads, but not writes when
1185 	 * DCA is enabled.  This is due to a known issue in some chipsets
1186 	 * which will cause the DCA tag to be cleared.
1187 	 */
1188 	rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1189 		  IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1190 
1191 	IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1192 }
1193 
1194 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1195 {
1196 	struct ixgbe_adapter *adapter = q_vector->adapter;
1197 	struct ixgbe_ring *ring;
1198 	int cpu = get_cpu();
1199 
1200 	if (q_vector->cpu == cpu)
1201 		goto out_no_update;
1202 
1203 	ixgbe_for_each_ring(ring, q_vector->tx)
1204 		ixgbe_update_tx_dca(adapter, ring, cpu);
1205 
1206 	ixgbe_for_each_ring(ring, q_vector->rx)
1207 		ixgbe_update_rx_dca(adapter, ring, cpu);
1208 
1209 	q_vector->cpu = cpu;
1210 out_no_update:
1211 	put_cpu();
1212 }
1213 
1214 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1215 {
1216 	int i;
1217 
1218 	if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1219 		return;
1220 
1221 	/* always use CB2 mode, difference is masked in the CB driver */
1222 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1223 
1224 	for (i = 0; i < adapter->num_q_vectors; i++) {
1225 		adapter->q_vector[i]->cpu = -1;
1226 		ixgbe_update_dca(adapter->q_vector[i]);
1227 	}
1228 }
1229 
1230 static int __ixgbe_notify_dca(struct device *dev, void *data)
1231 {
1232 	struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1233 	unsigned long event = *(unsigned long *)data;
1234 
1235 	if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1236 		return 0;
1237 
1238 	switch (event) {
1239 	case DCA_PROVIDER_ADD:
1240 		/* if we're already enabled, don't do it again */
1241 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1242 			break;
1243 		if (dca_add_requester(dev) == 0) {
1244 			adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1245 			ixgbe_setup_dca(adapter);
1246 			break;
1247 		}
1248 		/* Fall Through since DCA is disabled. */
1249 	case DCA_PROVIDER_REMOVE:
1250 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1251 			dca_remove_requester(dev);
1252 			adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1253 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1254 		}
1255 		break;
1256 	}
1257 
1258 	return 0;
1259 }
1260 
1261 #endif /* CONFIG_IXGBE_DCA */
1262 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1263 				 union ixgbe_adv_rx_desc *rx_desc,
1264 				 struct sk_buff *skb)
1265 {
1266 	if (ring->netdev->features & NETIF_F_RXHASH)
1267 		skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
1268 }
1269 
1270 #ifdef IXGBE_FCOE
1271 /**
1272  * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1273  * @ring: structure containing ring specific data
1274  * @rx_desc: advanced rx descriptor
1275  *
1276  * Returns : true if it is FCoE pkt
1277  */
1278 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1279 				    union ixgbe_adv_rx_desc *rx_desc)
1280 {
1281 	__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1282 
1283 	return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1284 	       ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1285 		(cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1286 			     IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1287 }
1288 
1289 #endif /* IXGBE_FCOE */
1290 /**
1291  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1292  * @ring: structure containing ring specific data
1293  * @rx_desc: current Rx descriptor being processed
1294  * @skb: skb currently being received and modified
1295  **/
1296 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1297 				     union ixgbe_adv_rx_desc *rx_desc,
1298 				     struct sk_buff *skb)
1299 {
1300 	skb_checksum_none_assert(skb);
1301 
1302 	/* Rx csum disabled */
1303 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
1304 		return;
1305 
1306 	/* if IP and error */
1307 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1308 	    ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1309 		ring->rx_stats.csum_err++;
1310 		return;
1311 	}
1312 
1313 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1314 		return;
1315 
1316 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1317 		__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1318 
1319 		/*
1320 		 * 82599 errata, UDP frames with a 0 checksum can be marked as
1321 		 * checksum errors.
1322 		 */
1323 		if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1324 		    test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1325 			return;
1326 
1327 		ring->rx_stats.csum_err++;
1328 		return;
1329 	}
1330 
1331 	/* It must be a TCP or UDP packet with a valid checksum */
1332 	skb->ip_summed = CHECKSUM_UNNECESSARY;
1333 }
1334 
1335 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1336 {
1337 	rx_ring->next_to_use = val;
1338 
1339 	/* update next to alloc since we have filled the ring */
1340 	rx_ring->next_to_alloc = val;
1341 	/*
1342 	 * Force memory writes to complete before letting h/w
1343 	 * know there are new descriptors to fetch.  (Only
1344 	 * applicable for weak-ordered memory model archs,
1345 	 * such as IA-64).
1346 	 */
1347 	wmb();
1348 	ixgbe_write_tail(rx_ring, val);
1349 }
1350 
1351 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1352 				    struct ixgbe_rx_buffer *bi)
1353 {
1354 	struct page *page = bi->page;
1355 	dma_addr_t dma = bi->dma;
1356 
1357 	/* since we are recycling buffers we should seldom need to alloc */
1358 	if (likely(dma))
1359 		return true;
1360 
1361 	/* alloc new page for storage */
1362 	if (likely(!page)) {
1363 		page = __skb_alloc_pages(GFP_ATOMIC | __GFP_COLD | __GFP_COMP,
1364 					 bi->skb, ixgbe_rx_pg_order(rx_ring));
1365 		if (unlikely(!page)) {
1366 			rx_ring->rx_stats.alloc_rx_page_failed++;
1367 			return false;
1368 		}
1369 		bi->page = page;
1370 	}
1371 
1372 	/* map page for use */
1373 	dma = dma_map_page(rx_ring->dev, page, 0,
1374 			   ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1375 
1376 	/*
1377 	 * if mapping failed free memory back to system since
1378 	 * there isn't much point in holding memory we can't use
1379 	 */
1380 	if (dma_mapping_error(rx_ring->dev, dma)) {
1381 		__free_pages(page, ixgbe_rx_pg_order(rx_ring));
1382 		bi->page = NULL;
1383 
1384 		rx_ring->rx_stats.alloc_rx_page_failed++;
1385 		return false;
1386 	}
1387 
1388 	bi->dma = dma;
1389 	bi->page_offset = 0;
1390 
1391 	return true;
1392 }
1393 
1394 /**
1395  * ixgbe_alloc_rx_buffers - Replace used receive buffers
1396  * @rx_ring: ring to place buffers on
1397  * @cleaned_count: number of buffers to replace
1398  **/
1399 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1400 {
1401 	union ixgbe_adv_rx_desc *rx_desc;
1402 	struct ixgbe_rx_buffer *bi;
1403 	u16 i = rx_ring->next_to_use;
1404 
1405 	/* nothing to do */
1406 	if (!cleaned_count)
1407 		return;
1408 
1409 	rx_desc = IXGBE_RX_DESC(rx_ring, i);
1410 	bi = &rx_ring->rx_buffer_info[i];
1411 	i -= rx_ring->count;
1412 
1413 	do {
1414 		if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1415 			break;
1416 
1417 		/*
1418 		 * Refresh the desc even if buffer_addrs didn't change
1419 		 * because each write-back erases this info.
1420 		 */
1421 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1422 
1423 		rx_desc++;
1424 		bi++;
1425 		i++;
1426 		if (unlikely(!i)) {
1427 			rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1428 			bi = rx_ring->rx_buffer_info;
1429 			i -= rx_ring->count;
1430 		}
1431 
1432 		/* clear the hdr_addr for the next_to_use descriptor */
1433 		rx_desc->read.hdr_addr = 0;
1434 
1435 		cleaned_count--;
1436 	} while (cleaned_count);
1437 
1438 	i += rx_ring->count;
1439 
1440 	if (rx_ring->next_to_use != i)
1441 		ixgbe_release_rx_desc(rx_ring, i);
1442 }
1443 
1444 /**
1445  * ixgbe_get_headlen - determine size of header for RSC/LRO/GRO/FCOE
1446  * @data: pointer to the start of the headers
1447  * @max_len: total length of section to find headers in
1448  *
1449  * This function is meant to determine the length of headers that will
1450  * be recognized by hardware for LRO, GRO, and RSC offloads.  The main
1451  * motivation of doing this is to only perform one pull for IPv4 TCP
1452  * packets so that we can do basic things like calculating the gso_size
1453  * based on the average data per packet.
1454  **/
1455 static unsigned int ixgbe_get_headlen(unsigned char *data,
1456 				      unsigned int max_len)
1457 {
1458 	union {
1459 		unsigned char *network;
1460 		/* l2 headers */
1461 		struct ethhdr *eth;
1462 		struct vlan_hdr *vlan;
1463 		/* l3 headers */
1464 		struct iphdr *ipv4;
1465 		struct ipv6hdr *ipv6;
1466 	} hdr;
1467 	__be16 protocol;
1468 	u8 nexthdr = 0;	/* default to not TCP */
1469 	u8 hlen;
1470 
1471 	/* this should never happen, but better safe than sorry */
1472 	if (max_len < ETH_HLEN)
1473 		return max_len;
1474 
1475 	/* initialize network frame pointer */
1476 	hdr.network = data;
1477 
1478 	/* set first protocol and move network header forward */
1479 	protocol = hdr.eth->h_proto;
1480 	hdr.network += ETH_HLEN;
1481 
1482 	/* handle any vlan tag if present */
1483 	if (protocol == __constant_htons(ETH_P_8021Q)) {
1484 		if ((hdr.network - data) > (max_len - VLAN_HLEN))
1485 			return max_len;
1486 
1487 		protocol = hdr.vlan->h_vlan_encapsulated_proto;
1488 		hdr.network += VLAN_HLEN;
1489 	}
1490 
1491 	/* handle L3 protocols */
1492 	if (protocol == __constant_htons(ETH_P_IP)) {
1493 		if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
1494 			return max_len;
1495 
1496 		/* access ihl as a u8 to avoid unaligned access on ia64 */
1497 		hlen = (hdr.network[0] & 0x0F) << 2;
1498 
1499 		/* verify hlen meets minimum size requirements */
1500 		if (hlen < sizeof(struct iphdr))
1501 			return hdr.network - data;
1502 
1503 		/* record next protocol if header is present */
1504 		if (!(hdr.ipv4->frag_off & htons(IP_OFFSET)))
1505 			nexthdr = hdr.ipv4->protocol;
1506 	} else if (protocol == __constant_htons(ETH_P_IPV6)) {
1507 		if ((hdr.network - data) > (max_len - sizeof(struct ipv6hdr)))
1508 			return max_len;
1509 
1510 		/* record next protocol */
1511 		nexthdr = hdr.ipv6->nexthdr;
1512 		hlen = sizeof(struct ipv6hdr);
1513 #ifdef IXGBE_FCOE
1514 	} else if (protocol == __constant_htons(ETH_P_FCOE)) {
1515 		if ((hdr.network - data) > (max_len - FCOE_HEADER_LEN))
1516 			return max_len;
1517 		hlen = FCOE_HEADER_LEN;
1518 #endif
1519 	} else {
1520 		return hdr.network - data;
1521 	}
1522 
1523 	/* relocate pointer to start of L4 header */
1524 	hdr.network += hlen;
1525 
1526 	/* finally sort out TCP/UDP */
1527 	if (nexthdr == IPPROTO_TCP) {
1528 		if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
1529 			return max_len;
1530 
1531 		/* access doff as a u8 to avoid unaligned access on ia64 */
1532 		hlen = (hdr.network[12] & 0xF0) >> 2;
1533 
1534 		/* verify hlen meets minimum size requirements */
1535 		if (hlen < sizeof(struct tcphdr))
1536 			return hdr.network - data;
1537 
1538 		hdr.network += hlen;
1539 	} else if (nexthdr == IPPROTO_UDP) {
1540 		if ((hdr.network - data) > (max_len - sizeof(struct udphdr)))
1541 			return max_len;
1542 
1543 		hdr.network += sizeof(struct udphdr);
1544 	}
1545 
1546 	/*
1547 	 * If everything has gone correctly hdr.network should be the
1548 	 * data section of the packet and will be the end of the header.
1549 	 * If not then it probably represents the end of the last recognized
1550 	 * header.
1551 	 */
1552 	if ((hdr.network - data) < max_len)
1553 		return hdr.network - data;
1554 	else
1555 		return max_len;
1556 }
1557 
1558 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1559 				   struct sk_buff *skb)
1560 {
1561 	u16 hdr_len = skb_headlen(skb);
1562 
1563 	/* set gso_size to avoid messing up TCP MSS */
1564 	skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1565 						 IXGBE_CB(skb)->append_cnt);
1566 	skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1567 }
1568 
1569 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1570 				   struct sk_buff *skb)
1571 {
1572 	/* if append_cnt is 0 then frame is not RSC */
1573 	if (!IXGBE_CB(skb)->append_cnt)
1574 		return;
1575 
1576 	rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1577 	rx_ring->rx_stats.rsc_flush++;
1578 
1579 	ixgbe_set_rsc_gso_size(rx_ring, skb);
1580 
1581 	/* gso_size is computed using append_cnt so always clear it last */
1582 	IXGBE_CB(skb)->append_cnt = 0;
1583 }
1584 
1585 /**
1586  * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1587  * @rx_ring: rx descriptor ring packet is being transacted on
1588  * @rx_desc: pointer to the EOP Rx descriptor
1589  * @skb: pointer to current skb being populated
1590  *
1591  * This function checks the ring, descriptor, and packet information in
1592  * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1593  * other fields within the skb.
1594  **/
1595 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1596 				     union ixgbe_adv_rx_desc *rx_desc,
1597 				     struct sk_buff *skb)
1598 {
1599 	struct net_device *dev = rx_ring->netdev;
1600 
1601 	ixgbe_update_rsc_stats(rx_ring, skb);
1602 
1603 	ixgbe_rx_hash(rx_ring, rx_desc, skb);
1604 
1605 	ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1606 
1607 	ixgbe_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
1608 
1609 	if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1610 	    ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1611 		u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1612 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1613 	}
1614 
1615 	skb_record_rx_queue(skb, rx_ring->queue_index);
1616 
1617 	skb->protocol = eth_type_trans(skb, dev);
1618 }
1619 
1620 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1621 			 struct sk_buff *skb)
1622 {
1623 	struct ixgbe_adapter *adapter = q_vector->adapter;
1624 
1625 	if (ixgbe_qv_busy_polling(q_vector))
1626 		netif_receive_skb(skb);
1627 	else if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1628 		napi_gro_receive(&q_vector->napi, skb);
1629 	else
1630 		netif_rx(skb);
1631 }
1632 
1633 /**
1634  * ixgbe_is_non_eop - process handling of non-EOP buffers
1635  * @rx_ring: Rx ring being processed
1636  * @rx_desc: Rx descriptor for current buffer
1637  * @skb: Current socket buffer containing buffer in progress
1638  *
1639  * This function updates next to clean.  If the buffer is an EOP buffer
1640  * this function exits returning false, otherwise it will place the
1641  * sk_buff in the next buffer to be chained and return true indicating
1642  * that this is in fact a non-EOP buffer.
1643  **/
1644 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1645 			     union ixgbe_adv_rx_desc *rx_desc,
1646 			     struct sk_buff *skb)
1647 {
1648 	u32 ntc = rx_ring->next_to_clean + 1;
1649 
1650 	/* fetch, update, and store next to clean */
1651 	ntc = (ntc < rx_ring->count) ? ntc : 0;
1652 	rx_ring->next_to_clean = ntc;
1653 
1654 	prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1655 
1656 	/* update RSC append count if present */
1657 	if (ring_is_rsc_enabled(rx_ring)) {
1658 		__le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1659 				     cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1660 
1661 		if (unlikely(rsc_enabled)) {
1662 			u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1663 
1664 			rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1665 			IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1666 
1667 			/* update ntc based on RSC value */
1668 			ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1669 			ntc &= IXGBE_RXDADV_NEXTP_MASK;
1670 			ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1671 		}
1672 	}
1673 
1674 	/* if we are the last buffer then there is nothing else to do */
1675 	if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1676 		return false;
1677 
1678 	/* place skb in next buffer to be received */
1679 	rx_ring->rx_buffer_info[ntc].skb = skb;
1680 	rx_ring->rx_stats.non_eop_descs++;
1681 
1682 	return true;
1683 }
1684 
1685 /**
1686  * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1687  * @rx_ring: rx descriptor ring packet is being transacted on
1688  * @skb: pointer to current skb being adjusted
1689  *
1690  * This function is an ixgbe specific version of __pskb_pull_tail.  The
1691  * main difference between this version and the original function is that
1692  * this function can make several assumptions about the state of things
1693  * that allow for significant optimizations versus the standard function.
1694  * As a result we can do things like drop a frag and maintain an accurate
1695  * truesize for the skb.
1696  */
1697 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1698 			    struct sk_buff *skb)
1699 {
1700 	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1701 	unsigned char *va;
1702 	unsigned int pull_len;
1703 
1704 	/*
1705 	 * it is valid to use page_address instead of kmap since we are
1706 	 * working with pages allocated out of the lomem pool per
1707 	 * alloc_page(GFP_ATOMIC)
1708 	 */
1709 	va = skb_frag_address(frag);
1710 
1711 	/*
1712 	 * we need the header to contain the greater of either ETH_HLEN or
1713 	 * 60 bytes if the skb->len is less than 60 for skb_pad.
1714 	 */
1715 	pull_len = ixgbe_get_headlen(va, IXGBE_RX_HDR_SIZE);
1716 
1717 	/* align pull length to size of long to optimize memcpy performance */
1718 	skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1719 
1720 	/* update all of the pointers */
1721 	skb_frag_size_sub(frag, pull_len);
1722 	frag->page_offset += pull_len;
1723 	skb->data_len -= pull_len;
1724 	skb->tail += pull_len;
1725 }
1726 
1727 /**
1728  * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1729  * @rx_ring: rx descriptor ring packet is being transacted on
1730  * @skb: pointer to current skb being updated
1731  *
1732  * This function provides a basic DMA sync up for the first fragment of an
1733  * skb.  The reason for doing this is that the first fragment cannot be
1734  * unmapped until we have reached the end of packet descriptor for a buffer
1735  * chain.
1736  */
1737 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1738 				struct sk_buff *skb)
1739 {
1740 	/* if the page was released unmap it, else just sync our portion */
1741 	if (unlikely(IXGBE_CB(skb)->page_released)) {
1742 		dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1743 			       ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1744 		IXGBE_CB(skb)->page_released = false;
1745 	} else {
1746 		struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1747 
1748 		dma_sync_single_range_for_cpu(rx_ring->dev,
1749 					      IXGBE_CB(skb)->dma,
1750 					      frag->page_offset,
1751 					      ixgbe_rx_bufsz(rx_ring),
1752 					      DMA_FROM_DEVICE);
1753 	}
1754 	IXGBE_CB(skb)->dma = 0;
1755 }
1756 
1757 /**
1758  * ixgbe_cleanup_headers - Correct corrupted or empty headers
1759  * @rx_ring: rx descriptor ring packet is being transacted on
1760  * @rx_desc: pointer to the EOP Rx descriptor
1761  * @skb: pointer to current skb being fixed
1762  *
1763  * Check for corrupted packet headers caused by senders on the local L2
1764  * embedded NIC switch not setting up their Tx Descriptors right.  These
1765  * should be very rare.
1766  *
1767  * Also address the case where we are pulling data in on pages only
1768  * and as such no data is present in the skb header.
1769  *
1770  * In addition if skb is not at least 60 bytes we need to pad it so that
1771  * it is large enough to qualify as a valid Ethernet frame.
1772  *
1773  * Returns true if an error was encountered and skb was freed.
1774  **/
1775 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1776 				  union ixgbe_adv_rx_desc *rx_desc,
1777 				  struct sk_buff *skb)
1778 {
1779 	struct net_device *netdev = rx_ring->netdev;
1780 
1781 	/* verify that the packet does not have any known errors */
1782 	if (unlikely(ixgbe_test_staterr(rx_desc,
1783 					IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1784 	    !(netdev->features & NETIF_F_RXALL))) {
1785 		dev_kfree_skb_any(skb);
1786 		return true;
1787 	}
1788 
1789 	/* place header in linear portion of buffer */
1790 	if (skb_is_nonlinear(skb))
1791 		ixgbe_pull_tail(rx_ring, skb);
1792 
1793 #ifdef IXGBE_FCOE
1794 	/* do not attempt to pad FCoE Frames as this will disrupt DDP */
1795 	if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1796 		return false;
1797 
1798 #endif
1799 	/* if skb_pad returns an error the skb was freed */
1800 	if (unlikely(skb->len < 60)) {
1801 		int pad_len = 60 - skb->len;
1802 
1803 		if (skb_pad(skb, pad_len))
1804 			return true;
1805 		__skb_put(skb, pad_len);
1806 	}
1807 
1808 	return false;
1809 }
1810 
1811 /**
1812  * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1813  * @rx_ring: rx descriptor ring to store buffers on
1814  * @old_buff: donor buffer to have page reused
1815  *
1816  * Synchronizes page for reuse by the adapter
1817  **/
1818 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1819 				struct ixgbe_rx_buffer *old_buff)
1820 {
1821 	struct ixgbe_rx_buffer *new_buff;
1822 	u16 nta = rx_ring->next_to_alloc;
1823 
1824 	new_buff = &rx_ring->rx_buffer_info[nta];
1825 
1826 	/* update, and store next to alloc */
1827 	nta++;
1828 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1829 
1830 	/* transfer page from old buffer to new buffer */
1831 	new_buff->page = old_buff->page;
1832 	new_buff->dma = old_buff->dma;
1833 	new_buff->page_offset = old_buff->page_offset;
1834 
1835 	/* sync the buffer for use by the device */
1836 	dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1837 					 new_buff->page_offset,
1838 					 ixgbe_rx_bufsz(rx_ring),
1839 					 DMA_FROM_DEVICE);
1840 }
1841 
1842 /**
1843  * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1844  * @rx_ring: rx descriptor ring to transact packets on
1845  * @rx_buffer: buffer containing page to add
1846  * @rx_desc: descriptor containing length of buffer written by hardware
1847  * @skb: sk_buff to place the data into
1848  *
1849  * This function will add the data contained in rx_buffer->page to the skb.
1850  * This is done either through a direct copy if the data in the buffer is
1851  * less than the skb header size, otherwise it will just attach the page as
1852  * a frag to the skb.
1853  *
1854  * The function will then update the page offset if necessary and return
1855  * true if the buffer can be reused by the adapter.
1856  **/
1857 static bool ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1858 			      struct ixgbe_rx_buffer *rx_buffer,
1859 			      union ixgbe_adv_rx_desc *rx_desc,
1860 			      struct sk_buff *skb)
1861 {
1862 	struct page *page = rx_buffer->page;
1863 	unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
1864 #if (PAGE_SIZE < 8192)
1865 	unsigned int truesize = ixgbe_rx_bufsz(rx_ring);
1866 #else
1867 	unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
1868 	unsigned int last_offset = ixgbe_rx_pg_size(rx_ring) -
1869 				   ixgbe_rx_bufsz(rx_ring);
1870 #endif
1871 
1872 	if ((size <= IXGBE_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
1873 		unsigned char *va = page_address(page) + rx_buffer->page_offset;
1874 
1875 		memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
1876 
1877 		/* we can reuse buffer as-is, just make sure it is local */
1878 		if (likely(page_to_nid(page) == numa_node_id()))
1879 			return true;
1880 
1881 		/* this page cannot be reused so discard it */
1882 		put_page(page);
1883 		return false;
1884 	}
1885 
1886 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
1887 			rx_buffer->page_offset, size, truesize);
1888 
1889 	/* avoid re-using remote pages */
1890 	if (unlikely(page_to_nid(page) != numa_node_id()))
1891 		return false;
1892 
1893 #if (PAGE_SIZE < 8192)
1894 	/* if we are only owner of page we can reuse it */
1895 	if (unlikely(page_count(page) != 1))
1896 		return false;
1897 
1898 	/* flip page offset to other buffer */
1899 	rx_buffer->page_offset ^= truesize;
1900 
1901 	/*
1902 	 * since we are the only owner of the page and we need to
1903 	 * increment it, just set the value to 2 in order to avoid
1904 	 * an unecessary locked operation
1905 	 */
1906 	atomic_set(&page->_count, 2);
1907 #else
1908 	/* move offset up to the next cache line */
1909 	rx_buffer->page_offset += truesize;
1910 
1911 	if (rx_buffer->page_offset > last_offset)
1912 		return false;
1913 
1914 	/* bump ref count on page before it is given to the stack */
1915 	get_page(page);
1916 #endif
1917 
1918 	return true;
1919 }
1920 
1921 static struct sk_buff *ixgbe_fetch_rx_buffer(struct ixgbe_ring *rx_ring,
1922 					     union ixgbe_adv_rx_desc *rx_desc)
1923 {
1924 	struct ixgbe_rx_buffer *rx_buffer;
1925 	struct sk_buff *skb;
1926 	struct page *page;
1927 
1928 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
1929 	page = rx_buffer->page;
1930 	prefetchw(page);
1931 
1932 	skb = rx_buffer->skb;
1933 
1934 	if (likely(!skb)) {
1935 		void *page_addr = page_address(page) +
1936 				  rx_buffer->page_offset;
1937 
1938 		/* prefetch first cache line of first page */
1939 		prefetch(page_addr);
1940 #if L1_CACHE_BYTES < 128
1941 		prefetch(page_addr + L1_CACHE_BYTES);
1942 #endif
1943 
1944 		/* allocate a skb to store the frags */
1945 		skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1946 						IXGBE_RX_HDR_SIZE);
1947 		if (unlikely(!skb)) {
1948 			rx_ring->rx_stats.alloc_rx_buff_failed++;
1949 			return NULL;
1950 		}
1951 
1952 		/*
1953 		 * we will be copying header into skb->data in
1954 		 * pskb_may_pull so it is in our interest to prefetch
1955 		 * it now to avoid a possible cache miss
1956 		 */
1957 		prefetchw(skb->data);
1958 
1959 		/*
1960 		 * Delay unmapping of the first packet. It carries the
1961 		 * header information, HW may still access the header
1962 		 * after the writeback.  Only unmap it when EOP is
1963 		 * reached
1964 		 */
1965 		if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1966 			goto dma_sync;
1967 
1968 		IXGBE_CB(skb)->dma = rx_buffer->dma;
1969 	} else {
1970 		if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
1971 			ixgbe_dma_sync_frag(rx_ring, skb);
1972 
1973 dma_sync:
1974 		/* we are reusing so sync this buffer for CPU use */
1975 		dma_sync_single_range_for_cpu(rx_ring->dev,
1976 					      rx_buffer->dma,
1977 					      rx_buffer->page_offset,
1978 					      ixgbe_rx_bufsz(rx_ring),
1979 					      DMA_FROM_DEVICE);
1980 	}
1981 
1982 	/* pull page into skb */
1983 	if (ixgbe_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
1984 		/* hand second half of page back to the ring */
1985 		ixgbe_reuse_rx_page(rx_ring, rx_buffer);
1986 	} else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
1987 		/* the page has been released from the ring */
1988 		IXGBE_CB(skb)->page_released = true;
1989 	} else {
1990 		/* we are not reusing the buffer so unmap it */
1991 		dma_unmap_page(rx_ring->dev, rx_buffer->dma,
1992 			       ixgbe_rx_pg_size(rx_ring),
1993 			       DMA_FROM_DEVICE);
1994 	}
1995 
1996 	/* clear contents of buffer_info */
1997 	rx_buffer->skb = NULL;
1998 	rx_buffer->dma = 0;
1999 	rx_buffer->page = NULL;
2000 
2001 	return skb;
2002 }
2003 
2004 /**
2005  * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2006  * @q_vector: structure containing interrupt and ring information
2007  * @rx_ring: rx descriptor ring to transact packets on
2008  * @budget: Total limit on number of packets to process
2009  *
2010  * This function provides a "bounce buffer" approach to Rx interrupt
2011  * processing.  The advantage to this is that on systems that have
2012  * expensive overhead for IOMMU access this provides a means of avoiding
2013  * it by maintaining the mapping of the page to the syste.
2014  *
2015  * Returns amount of work completed
2016  **/
2017 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2018 			       struct ixgbe_ring *rx_ring,
2019 			       const int budget)
2020 {
2021 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2022 #ifdef IXGBE_FCOE
2023 	struct ixgbe_adapter *adapter = q_vector->adapter;
2024 	int ddp_bytes;
2025 	unsigned int mss = 0;
2026 #endif /* IXGBE_FCOE */
2027 	u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2028 
2029 	do {
2030 		union ixgbe_adv_rx_desc *rx_desc;
2031 		struct sk_buff *skb;
2032 
2033 		/* return some buffers to hardware, one at a time is too slow */
2034 		if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2035 			ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2036 			cleaned_count = 0;
2037 		}
2038 
2039 		rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2040 
2041 		if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
2042 			break;
2043 
2044 		/*
2045 		 * This memory barrier is needed to keep us from reading
2046 		 * any other fields out of the rx_desc until we know the
2047 		 * RXD_STAT_DD bit is set
2048 		 */
2049 		rmb();
2050 
2051 		/* retrieve a buffer from the ring */
2052 		skb = ixgbe_fetch_rx_buffer(rx_ring, rx_desc);
2053 
2054 		/* exit if we failed to retrieve a buffer */
2055 		if (!skb)
2056 			break;
2057 
2058 		cleaned_count++;
2059 
2060 		/* place incomplete frames back on ring for completion */
2061 		if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2062 			continue;
2063 
2064 		/* verify the packet layout is correct */
2065 		if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2066 			continue;
2067 
2068 		/* probably a little skewed due to removing CRC */
2069 		total_rx_bytes += skb->len;
2070 
2071 		/* populate checksum, timestamp, VLAN, and protocol */
2072 		ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2073 
2074 #ifdef IXGBE_FCOE
2075 		/* if ddp, not passing to ULD unless for FCP_RSP or error */
2076 		if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2077 			ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2078 			/* include DDPed FCoE data */
2079 			if (ddp_bytes > 0) {
2080 				if (!mss) {
2081 					mss = rx_ring->netdev->mtu -
2082 						sizeof(struct fcoe_hdr) -
2083 						sizeof(struct fc_frame_header) -
2084 						sizeof(struct fcoe_crc_eof);
2085 					if (mss > 512)
2086 						mss &= ~511;
2087 				}
2088 				total_rx_bytes += ddp_bytes;
2089 				total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2090 								 mss);
2091 			}
2092 			if (!ddp_bytes) {
2093 				dev_kfree_skb_any(skb);
2094 				continue;
2095 			}
2096 		}
2097 
2098 #endif /* IXGBE_FCOE */
2099 		skb_mark_napi_id(skb, &q_vector->napi);
2100 		ixgbe_rx_skb(q_vector, skb);
2101 
2102 		/* update budget accounting */
2103 		total_rx_packets++;
2104 	} while (likely(total_rx_packets < budget));
2105 
2106 	u64_stats_update_begin(&rx_ring->syncp);
2107 	rx_ring->stats.packets += total_rx_packets;
2108 	rx_ring->stats.bytes += total_rx_bytes;
2109 	u64_stats_update_end(&rx_ring->syncp);
2110 	q_vector->rx.total_packets += total_rx_packets;
2111 	q_vector->rx.total_bytes += total_rx_bytes;
2112 
2113 	if (cleaned_count)
2114 		ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2115 
2116 	return total_rx_packets;
2117 }
2118 
2119 #ifdef CONFIG_NET_RX_BUSY_POLL
2120 /* must be called with local_bh_disable()d */
2121 static int ixgbe_low_latency_recv(struct napi_struct *napi)
2122 {
2123 	struct ixgbe_q_vector *q_vector =
2124 			container_of(napi, struct ixgbe_q_vector, napi);
2125 	struct ixgbe_adapter *adapter = q_vector->adapter;
2126 	struct ixgbe_ring  *ring;
2127 	int found = 0;
2128 
2129 	if (test_bit(__IXGBE_DOWN, &adapter->state))
2130 		return LL_FLUSH_FAILED;
2131 
2132 	if (!ixgbe_qv_lock_poll(q_vector))
2133 		return LL_FLUSH_BUSY;
2134 
2135 	ixgbe_for_each_ring(ring, q_vector->rx) {
2136 		found = ixgbe_clean_rx_irq(q_vector, ring, 4);
2137 #ifdef BP_EXTENDED_STATS
2138 		if (found)
2139 			ring->stats.cleaned += found;
2140 		else
2141 			ring->stats.misses++;
2142 #endif
2143 		if (found)
2144 			break;
2145 	}
2146 
2147 	ixgbe_qv_unlock_poll(q_vector);
2148 
2149 	return found;
2150 }
2151 #endif	/* CONFIG_NET_RX_BUSY_POLL */
2152 
2153 /**
2154  * ixgbe_configure_msix - Configure MSI-X hardware
2155  * @adapter: board private structure
2156  *
2157  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2158  * interrupts.
2159  **/
2160 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2161 {
2162 	struct ixgbe_q_vector *q_vector;
2163 	int v_idx;
2164 	u32 mask;
2165 
2166 	/* Populate MSIX to EITR Select */
2167 	if (adapter->num_vfs > 32) {
2168 		u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2169 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2170 	}
2171 
2172 	/*
2173 	 * Populate the IVAR table and set the ITR values to the
2174 	 * corresponding register.
2175 	 */
2176 	for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2177 		struct ixgbe_ring *ring;
2178 		q_vector = adapter->q_vector[v_idx];
2179 
2180 		ixgbe_for_each_ring(ring, q_vector->rx)
2181 			ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2182 
2183 		ixgbe_for_each_ring(ring, q_vector->tx)
2184 			ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2185 
2186 		ixgbe_write_eitr(q_vector);
2187 	}
2188 
2189 	switch (adapter->hw.mac.type) {
2190 	case ixgbe_mac_82598EB:
2191 		ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2192 			       v_idx);
2193 		break;
2194 	case ixgbe_mac_82599EB:
2195 	case ixgbe_mac_X540:
2196 		ixgbe_set_ivar(adapter, -1, 1, v_idx);
2197 		break;
2198 	default:
2199 		break;
2200 	}
2201 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2202 
2203 	/* set up to autoclear timer, and the vectors */
2204 	mask = IXGBE_EIMS_ENABLE_MASK;
2205 	mask &= ~(IXGBE_EIMS_OTHER |
2206 		  IXGBE_EIMS_MAILBOX |
2207 		  IXGBE_EIMS_LSC);
2208 
2209 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2210 }
2211 
2212 enum latency_range {
2213 	lowest_latency = 0,
2214 	low_latency = 1,
2215 	bulk_latency = 2,
2216 	latency_invalid = 255
2217 };
2218 
2219 /**
2220  * ixgbe_update_itr - update the dynamic ITR value based on statistics
2221  * @q_vector: structure containing interrupt and ring information
2222  * @ring_container: structure containing ring performance data
2223  *
2224  *      Stores a new ITR value based on packets and byte
2225  *      counts during the last interrupt.  The advantage of per interrupt
2226  *      computation is faster updates and more accurate ITR for the current
2227  *      traffic pattern.  Constants in this function were computed
2228  *      based on theoretical maximum wire speed and thresholds were set based
2229  *      on testing data as well as attempting to minimize response time
2230  *      while increasing bulk throughput.
2231  *      this functionality is controlled by the InterruptThrottleRate module
2232  *      parameter (see ixgbe_param.c)
2233  **/
2234 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2235 			     struct ixgbe_ring_container *ring_container)
2236 {
2237 	int bytes = ring_container->total_bytes;
2238 	int packets = ring_container->total_packets;
2239 	u32 timepassed_us;
2240 	u64 bytes_perint;
2241 	u8 itr_setting = ring_container->itr;
2242 
2243 	if (packets == 0)
2244 		return;
2245 
2246 	/* simple throttlerate management
2247 	 *   0-10MB/s   lowest (100000 ints/s)
2248 	 *  10-20MB/s   low    (20000 ints/s)
2249 	 *  20-1249MB/s bulk   (8000 ints/s)
2250 	 */
2251 	/* what was last interrupt timeslice? */
2252 	timepassed_us = q_vector->itr >> 2;
2253 	if (timepassed_us == 0)
2254 		return;
2255 
2256 	bytes_perint = bytes / timepassed_us; /* bytes/usec */
2257 
2258 	switch (itr_setting) {
2259 	case lowest_latency:
2260 		if (bytes_perint > 10)
2261 			itr_setting = low_latency;
2262 		break;
2263 	case low_latency:
2264 		if (bytes_perint > 20)
2265 			itr_setting = bulk_latency;
2266 		else if (bytes_perint <= 10)
2267 			itr_setting = lowest_latency;
2268 		break;
2269 	case bulk_latency:
2270 		if (bytes_perint <= 20)
2271 			itr_setting = low_latency;
2272 		break;
2273 	}
2274 
2275 	/* clear work counters since we have the values we need */
2276 	ring_container->total_bytes = 0;
2277 	ring_container->total_packets = 0;
2278 
2279 	/* write updated itr to ring container */
2280 	ring_container->itr = itr_setting;
2281 }
2282 
2283 /**
2284  * ixgbe_write_eitr - write EITR register in hardware specific way
2285  * @q_vector: structure containing interrupt and ring information
2286  *
2287  * This function is made to be called by ethtool and by the driver
2288  * when it needs to update EITR registers at runtime.  Hardware
2289  * specific quirks/differences are taken care of here.
2290  */
2291 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2292 {
2293 	struct ixgbe_adapter *adapter = q_vector->adapter;
2294 	struct ixgbe_hw *hw = &adapter->hw;
2295 	int v_idx = q_vector->v_idx;
2296 	u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2297 
2298 	switch (adapter->hw.mac.type) {
2299 	case ixgbe_mac_82598EB:
2300 		/* must write high and low 16 bits to reset counter */
2301 		itr_reg |= (itr_reg << 16);
2302 		break;
2303 	case ixgbe_mac_82599EB:
2304 	case ixgbe_mac_X540:
2305 		/*
2306 		 * set the WDIS bit to not clear the timer bits and cause an
2307 		 * immediate assertion of the interrupt
2308 		 */
2309 		itr_reg |= IXGBE_EITR_CNT_WDIS;
2310 		break;
2311 	default:
2312 		break;
2313 	}
2314 	IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2315 }
2316 
2317 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2318 {
2319 	u32 new_itr = q_vector->itr;
2320 	u8 current_itr;
2321 
2322 	ixgbe_update_itr(q_vector, &q_vector->tx);
2323 	ixgbe_update_itr(q_vector, &q_vector->rx);
2324 
2325 	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
2326 
2327 	switch (current_itr) {
2328 	/* counts and packets in update_itr are dependent on these numbers */
2329 	case lowest_latency:
2330 		new_itr = IXGBE_100K_ITR;
2331 		break;
2332 	case low_latency:
2333 		new_itr = IXGBE_20K_ITR;
2334 		break;
2335 	case bulk_latency:
2336 		new_itr = IXGBE_8K_ITR;
2337 		break;
2338 	default:
2339 		break;
2340 	}
2341 
2342 	if (new_itr != q_vector->itr) {
2343 		/* do an exponential smoothing */
2344 		new_itr = (10 * new_itr * q_vector->itr) /
2345 			  ((9 * new_itr) + q_vector->itr);
2346 
2347 		/* save the algorithm value here */
2348 		q_vector->itr = new_itr;
2349 
2350 		ixgbe_write_eitr(q_vector);
2351 	}
2352 }
2353 
2354 /**
2355  * ixgbe_check_overtemp_subtask - check for over temperature
2356  * @adapter: pointer to adapter
2357  **/
2358 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2359 {
2360 	struct ixgbe_hw *hw = &adapter->hw;
2361 	u32 eicr = adapter->interrupt_event;
2362 
2363 	if (test_bit(__IXGBE_DOWN, &adapter->state))
2364 		return;
2365 
2366 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2367 	    !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2368 		return;
2369 
2370 	adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2371 
2372 	switch (hw->device_id) {
2373 	case IXGBE_DEV_ID_82599_T3_LOM:
2374 		/*
2375 		 * Since the warning interrupt is for both ports
2376 		 * we don't have to check if:
2377 		 *  - This interrupt wasn't for our port.
2378 		 *  - We may have missed the interrupt so always have to
2379 		 *    check if we  got a LSC
2380 		 */
2381 		if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
2382 		    !(eicr & IXGBE_EICR_LSC))
2383 			return;
2384 
2385 		if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2386 			u32 speed;
2387 			bool link_up = false;
2388 
2389 			hw->mac.ops.check_link(hw, &speed, &link_up, false);
2390 
2391 			if (link_up)
2392 				return;
2393 		}
2394 
2395 		/* Check if this is not due to overtemp */
2396 		if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2397 			return;
2398 
2399 		break;
2400 	default:
2401 		if (!(eicr & IXGBE_EICR_GPI_SDP0))
2402 			return;
2403 		break;
2404 	}
2405 	e_crit(drv,
2406 	       "Network adapter has been stopped because it has over heated. "
2407 	       "Restart the computer. If the problem persists, "
2408 	       "power off the system and replace the adapter\n");
2409 
2410 	adapter->interrupt_event = 0;
2411 }
2412 
2413 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2414 {
2415 	struct ixgbe_hw *hw = &adapter->hw;
2416 
2417 	if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2418 	    (eicr & IXGBE_EICR_GPI_SDP1)) {
2419 		e_crit(probe, "Fan has stopped, replace the adapter\n");
2420 		/* write to clear the interrupt */
2421 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2422 	}
2423 }
2424 
2425 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2426 {
2427 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2428 		return;
2429 
2430 	switch (adapter->hw.mac.type) {
2431 	case ixgbe_mac_82599EB:
2432 		/*
2433 		 * Need to check link state so complete overtemp check
2434 		 * on service task
2435 		 */
2436 		if (((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)) &&
2437 		    (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2438 			adapter->interrupt_event = eicr;
2439 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2440 			ixgbe_service_event_schedule(adapter);
2441 			return;
2442 		}
2443 		return;
2444 	case ixgbe_mac_X540:
2445 		if (!(eicr & IXGBE_EICR_TS))
2446 			return;
2447 		break;
2448 	default:
2449 		return;
2450 	}
2451 
2452 	e_crit(drv,
2453 	       "Network adapter has been stopped because it has over heated. "
2454 	       "Restart the computer. If the problem persists, "
2455 	       "power off the system and replace the adapter\n");
2456 }
2457 
2458 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2459 {
2460 	struct ixgbe_hw *hw = &adapter->hw;
2461 
2462 	if (eicr & IXGBE_EICR_GPI_SDP2) {
2463 		/* Clear the interrupt */
2464 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
2465 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2466 			adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2467 			ixgbe_service_event_schedule(adapter);
2468 		}
2469 	}
2470 
2471 	if (eicr & IXGBE_EICR_GPI_SDP1) {
2472 		/* Clear the interrupt */
2473 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2474 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2475 			adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2476 			ixgbe_service_event_schedule(adapter);
2477 		}
2478 	}
2479 }
2480 
2481 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2482 {
2483 	struct ixgbe_hw *hw = &adapter->hw;
2484 
2485 	adapter->lsc_int++;
2486 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2487 	adapter->link_check_timeout = jiffies;
2488 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2489 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2490 		IXGBE_WRITE_FLUSH(hw);
2491 		ixgbe_service_event_schedule(adapter);
2492 	}
2493 }
2494 
2495 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2496 					   u64 qmask)
2497 {
2498 	u32 mask;
2499 	struct ixgbe_hw *hw = &adapter->hw;
2500 
2501 	switch (hw->mac.type) {
2502 	case ixgbe_mac_82598EB:
2503 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2504 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2505 		break;
2506 	case ixgbe_mac_82599EB:
2507 	case ixgbe_mac_X540:
2508 		mask = (qmask & 0xFFFFFFFF);
2509 		if (mask)
2510 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2511 		mask = (qmask >> 32);
2512 		if (mask)
2513 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2514 		break;
2515 	default:
2516 		break;
2517 	}
2518 	/* skip the flush */
2519 }
2520 
2521 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2522 					    u64 qmask)
2523 {
2524 	u32 mask;
2525 	struct ixgbe_hw *hw = &adapter->hw;
2526 
2527 	switch (hw->mac.type) {
2528 	case ixgbe_mac_82598EB:
2529 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2530 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2531 		break;
2532 	case ixgbe_mac_82599EB:
2533 	case ixgbe_mac_X540:
2534 		mask = (qmask & 0xFFFFFFFF);
2535 		if (mask)
2536 			IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2537 		mask = (qmask >> 32);
2538 		if (mask)
2539 			IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2540 		break;
2541 	default:
2542 		break;
2543 	}
2544 	/* skip the flush */
2545 }
2546 
2547 /**
2548  * ixgbe_irq_enable - Enable default interrupt generation settings
2549  * @adapter: board private structure
2550  **/
2551 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2552 				    bool flush)
2553 {
2554 	u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2555 
2556 	/* don't reenable LSC while waiting for link */
2557 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2558 		mask &= ~IXGBE_EIMS_LSC;
2559 
2560 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2561 		switch (adapter->hw.mac.type) {
2562 		case ixgbe_mac_82599EB:
2563 			mask |= IXGBE_EIMS_GPI_SDP0;
2564 			break;
2565 		case ixgbe_mac_X540:
2566 			mask |= IXGBE_EIMS_TS;
2567 			break;
2568 		default:
2569 			break;
2570 		}
2571 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2572 		mask |= IXGBE_EIMS_GPI_SDP1;
2573 	switch (adapter->hw.mac.type) {
2574 	case ixgbe_mac_82599EB:
2575 		mask |= IXGBE_EIMS_GPI_SDP1;
2576 		mask |= IXGBE_EIMS_GPI_SDP2;
2577 	case ixgbe_mac_X540:
2578 		mask |= IXGBE_EIMS_ECC;
2579 		mask |= IXGBE_EIMS_MAILBOX;
2580 		break;
2581 	default:
2582 		break;
2583 	}
2584 
2585 	if (adapter->hw.mac.type == ixgbe_mac_X540)
2586 		mask |= IXGBE_EIMS_TIMESYNC;
2587 
2588 	if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2589 	    !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2590 		mask |= IXGBE_EIMS_FLOW_DIR;
2591 
2592 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2593 	if (queues)
2594 		ixgbe_irq_enable_queues(adapter, ~0);
2595 	if (flush)
2596 		IXGBE_WRITE_FLUSH(&adapter->hw);
2597 }
2598 
2599 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2600 {
2601 	struct ixgbe_adapter *adapter = data;
2602 	struct ixgbe_hw *hw = &adapter->hw;
2603 	u32 eicr;
2604 
2605 	/*
2606 	 * Workaround for Silicon errata.  Use clear-by-write instead
2607 	 * of clear-by-read.  Reading with EICS will return the
2608 	 * interrupt causes without clearing, which later be done
2609 	 * with the write to EICR.
2610 	 */
2611 	eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2612 
2613 	/* The lower 16bits of the EICR register are for the queue interrupts
2614 	 * which should be masked here in order to not accidently clear them if
2615 	 * the bits are high when ixgbe_msix_other is called. There is a race
2616 	 * condition otherwise which results in possible performance loss
2617 	 * especially if the ixgbe_msix_other interrupt is triggering
2618 	 * consistently (as it would when PPS is turned on for the X540 device)
2619 	 */
2620 	eicr &= 0xFFFF0000;
2621 
2622 	IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2623 
2624 	if (eicr & IXGBE_EICR_LSC)
2625 		ixgbe_check_lsc(adapter);
2626 
2627 	if (eicr & IXGBE_EICR_MAILBOX)
2628 		ixgbe_msg_task(adapter);
2629 
2630 	switch (hw->mac.type) {
2631 	case ixgbe_mac_82599EB:
2632 	case ixgbe_mac_X540:
2633 		if (eicr & IXGBE_EICR_ECC)
2634 			e_info(link, "Received unrecoverable ECC Err, please "
2635 			       "reboot\n");
2636 		/* Handle Flow Director Full threshold interrupt */
2637 		if (eicr & IXGBE_EICR_FLOW_DIR) {
2638 			int reinit_count = 0;
2639 			int i;
2640 			for (i = 0; i < adapter->num_tx_queues; i++) {
2641 				struct ixgbe_ring *ring = adapter->tx_ring[i];
2642 				if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2643 						       &ring->state))
2644 					reinit_count++;
2645 			}
2646 			if (reinit_count) {
2647 				/* no more flow director interrupts until after init */
2648 				IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2649 				adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2650 				ixgbe_service_event_schedule(adapter);
2651 			}
2652 		}
2653 		ixgbe_check_sfp_event(adapter, eicr);
2654 		ixgbe_check_overtemp_event(adapter, eicr);
2655 		break;
2656 	default:
2657 		break;
2658 	}
2659 
2660 	ixgbe_check_fan_failure(adapter, eicr);
2661 
2662 	if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2663 		ixgbe_ptp_check_pps_event(adapter, eicr);
2664 
2665 	/* re-enable the original interrupt state, no lsc, no queues */
2666 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
2667 		ixgbe_irq_enable(adapter, false, false);
2668 
2669 	return IRQ_HANDLED;
2670 }
2671 
2672 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2673 {
2674 	struct ixgbe_q_vector *q_vector = data;
2675 
2676 	/* EIAM disabled interrupts (on this vector) for us */
2677 
2678 	if (q_vector->rx.ring || q_vector->tx.ring)
2679 		napi_schedule(&q_vector->napi);
2680 
2681 	return IRQ_HANDLED;
2682 }
2683 
2684 /**
2685  * ixgbe_poll - NAPI Rx polling callback
2686  * @napi: structure for representing this polling device
2687  * @budget: how many packets driver is allowed to clean
2688  *
2689  * This function is used for legacy and MSI, NAPI mode
2690  **/
2691 int ixgbe_poll(struct napi_struct *napi, int budget)
2692 {
2693 	struct ixgbe_q_vector *q_vector =
2694 				container_of(napi, struct ixgbe_q_vector, napi);
2695 	struct ixgbe_adapter *adapter = q_vector->adapter;
2696 	struct ixgbe_ring *ring;
2697 	int per_ring_budget;
2698 	bool clean_complete = true;
2699 
2700 #ifdef CONFIG_IXGBE_DCA
2701 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2702 		ixgbe_update_dca(q_vector);
2703 #endif
2704 
2705 	ixgbe_for_each_ring(ring, q_vector->tx)
2706 		clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
2707 
2708 	if (!ixgbe_qv_lock_napi(q_vector))
2709 		return budget;
2710 
2711 	/* attempt to distribute budget to each queue fairly, but don't allow
2712 	 * the budget to go below 1 because we'll exit polling */
2713 	if (q_vector->rx.count > 1)
2714 		per_ring_budget = max(budget/q_vector->rx.count, 1);
2715 	else
2716 		per_ring_budget = budget;
2717 
2718 	ixgbe_for_each_ring(ring, q_vector->rx)
2719 		clean_complete &= (ixgbe_clean_rx_irq(q_vector, ring,
2720 				   per_ring_budget) < per_ring_budget);
2721 
2722 	ixgbe_qv_unlock_napi(q_vector);
2723 	/* If all work not completed, return budget and keep polling */
2724 	if (!clean_complete)
2725 		return budget;
2726 
2727 	/* all work done, exit the polling mode */
2728 	napi_complete(napi);
2729 	if (adapter->rx_itr_setting & 1)
2730 		ixgbe_set_itr(q_vector);
2731 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
2732 		ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2733 
2734 	return 0;
2735 }
2736 
2737 /**
2738  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2739  * @adapter: board private structure
2740  *
2741  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2742  * interrupts from the kernel.
2743  **/
2744 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2745 {
2746 	struct net_device *netdev = adapter->netdev;
2747 	int vector, err;
2748 	int ri = 0, ti = 0;
2749 
2750 	for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2751 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2752 		struct msix_entry *entry = &adapter->msix_entries[vector];
2753 
2754 		if (q_vector->tx.ring && q_vector->rx.ring) {
2755 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2756 				 "%s-%s-%d", netdev->name, "TxRx", ri++);
2757 			ti++;
2758 		} else if (q_vector->rx.ring) {
2759 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2760 				 "%s-%s-%d", netdev->name, "rx", ri++);
2761 		} else if (q_vector->tx.ring) {
2762 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2763 				 "%s-%s-%d", netdev->name, "tx", ti++);
2764 		} else {
2765 			/* skip this unused q_vector */
2766 			continue;
2767 		}
2768 		err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2769 				  q_vector->name, q_vector);
2770 		if (err) {
2771 			e_err(probe, "request_irq failed for MSIX interrupt "
2772 			      "Error: %d\n", err);
2773 			goto free_queue_irqs;
2774 		}
2775 		/* If Flow Director is enabled, set interrupt affinity */
2776 		if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2777 			/* assign the mask for this irq */
2778 			irq_set_affinity_hint(entry->vector,
2779 					      &q_vector->affinity_mask);
2780 		}
2781 	}
2782 
2783 	err = request_irq(adapter->msix_entries[vector].vector,
2784 			  ixgbe_msix_other, 0, netdev->name, adapter);
2785 	if (err) {
2786 		e_err(probe, "request_irq for msix_other failed: %d\n", err);
2787 		goto free_queue_irqs;
2788 	}
2789 
2790 	return 0;
2791 
2792 free_queue_irqs:
2793 	while (vector) {
2794 		vector--;
2795 		irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2796 				      NULL);
2797 		free_irq(adapter->msix_entries[vector].vector,
2798 			 adapter->q_vector[vector]);
2799 	}
2800 	adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2801 	pci_disable_msix(adapter->pdev);
2802 	kfree(adapter->msix_entries);
2803 	adapter->msix_entries = NULL;
2804 	return err;
2805 }
2806 
2807 /**
2808  * ixgbe_intr - legacy mode Interrupt Handler
2809  * @irq: interrupt number
2810  * @data: pointer to a network interface device structure
2811  **/
2812 static irqreturn_t ixgbe_intr(int irq, void *data)
2813 {
2814 	struct ixgbe_adapter *adapter = data;
2815 	struct ixgbe_hw *hw = &adapter->hw;
2816 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2817 	u32 eicr;
2818 
2819 	/*
2820 	 * Workaround for silicon errata #26 on 82598.  Mask the interrupt
2821 	 * before the read of EICR.
2822 	 */
2823 	IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2824 
2825 	/* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2826 	 * therefore no explicit interrupt disable is necessary */
2827 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2828 	if (!eicr) {
2829 		/*
2830 		 * shared interrupt alert!
2831 		 * make sure interrupts are enabled because the read will
2832 		 * have disabled interrupts due to EIAM
2833 		 * finish the workaround of silicon errata on 82598.  Unmask
2834 		 * the interrupt that we masked before the EICR read.
2835 		 */
2836 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
2837 			ixgbe_irq_enable(adapter, true, true);
2838 		return IRQ_NONE;	/* Not our interrupt */
2839 	}
2840 
2841 	if (eicr & IXGBE_EICR_LSC)
2842 		ixgbe_check_lsc(adapter);
2843 
2844 	switch (hw->mac.type) {
2845 	case ixgbe_mac_82599EB:
2846 		ixgbe_check_sfp_event(adapter, eicr);
2847 		/* Fall through */
2848 	case ixgbe_mac_X540:
2849 		if (eicr & IXGBE_EICR_ECC)
2850 			e_info(link, "Received unrecoverable ECC err, please "
2851 				     "reboot\n");
2852 		ixgbe_check_overtemp_event(adapter, eicr);
2853 		break;
2854 	default:
2855 		break;
2856 	}
2857 
2858 	ixgbe_check_fan_failure(adapter, eicr);
2859 	if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2860 		ixgbe_ptp_check_pps_event(adapter, eicr);
2861 
2862 	/* would disable interrupts here but EIAM disabled it */
2863 	napi_schedule(&q_vector->napi);
2864 
2865 	/*
2866 	 * re-enable link(maybe) and non-queue interrupts, no flush.
2867 	 * ixgbe_poll will re-enable the queue interrupts
2868 	 */
2869 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
2870 		ixgbe_irq_enable(adapter, false, false);
2871 
2872 	return IRQ_HANDLED;
2873 }
2874 
2875 /**
2876  * ixgbe_request_irq - initialize interrupts
2877  * @adapter: board private structure
2878  *
2879  * Attempts to configure interrupts using the best available
2880  * capabilities of the hardware and kernel.
2881  **/
2882 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2883 {
2884 	struct net_device *netdev = adapter->netdev;
2885 	int err;
2886 
2887 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2888 		err = ixgbe_request_msix_irqs(adapter);
2889 	else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2890 		err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2891 				  netdev->name, adapter);
2892 	else
2893 		err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2894 				  netdev->name, adapter);
2895 
2896 	if (err)
2897 		e_err(probe, "request_irq failed, Error %d\n", err);
2898 
2899 	return err;
2900 }
2901 
2902 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2903 {
2904 	int vector;
2905 
2906 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2907 		free_irq(adapter->pdev->irq, adapter);
2908 		return;
2909 	}
2910 
2911 	for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2912 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2913 		struct msix_entry *entry = &adapter->msix_entries[vector];
2914 
2915 		/* free only the irqs that were actually requested */
2916 		if (!q_vector->rx.ring && !q_vector->tx.ring)
2917 			continue;
2918 
2919 		/* clear the affinity_mask in the IRQ descriptor */
2920 		irq_set_affinity_hint(entry->vector, NULL);
2921 
2922 		free_irq(entry->vector, q_vector);
2923 	}
2924 
2925 	free_irq(adapter->msix_entries[vector++].vector, adapter);
2926 }
2927 
2928 /**
2929  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2930  * @adapter: board private structure
2931  **/
2932 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2933 {
2934 	switch (adapter->hw.mac.type) {
2935 	case ixgbe_mac_82598EB:
2936 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2937 		break;
2938 	case ixgbe_mac_82599EB:
2939 	case ixgbe_mac_X540:
2940 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2941 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2942 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2943 		break;
2944 	default:
2945 		break;
2946 	}
2947 	IXGBE_WRITE_FLUSH(&adapter->hw);
2948 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2949 		int vector;
2950 
2951 		for (vector = 0; vector < adapter->num_q_vectors; vector++)
2952 			synchronize_irq(adapter->msix_entries[vector].vector);
2953 
2954 		synchronize_irq(adapter->msix_entries[vector++].vector);
2955 	} else {
2956 		synchronize_irq(adapter->pdev->irq);
2957 	}
2958 }
2959 
2960 /**
2961  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2962  *
2963  **/
2964 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2965 {
2966 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2967 
2968 	ixgbe_write_eitr(q_vector);
2969 
2970 	ixgbe_set_ivar(adapter, 0, 0, 0);
2971 	ixgbe_set_ivar(adapter, 1, 0, 0);
2972 
2973 	e_info(hw, "Legacy interrupt IVAR setup done\n");
2974 }
2975 
2976 /**
2977  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2978  * @adapter: board private structure
2979  * @ring: structure containing ring specific data
2980  *
2981  * Configure the Tx descriptor ring after a reset.
2982  **/
2983 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2984 			     struct ixgbe_ring *ring)
2985 {
2986 	struct ixgbe_hw *hw = &adapter->hw;
2987 	u64 tdba = ring->dma;
2988 	int wait_loop = 10;
2989 	u32 txdctl = IXGBE_TXDCTL_ENABLE;
2990 	u8 reg_idx = ring->reg_idx;
2991 
2992 	/* disable queue to avoid issues while updating state */
2993 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
2994 	IXGBE_WRITE_FLUSH(hw);
2995 
2996 	IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2997 			(tdba & DMA_BIT_MASK(32)));
2998 	IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2999 	IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3000 			ring->count * sizeof(union ixgbe_adv_tx_desc));
3001 	IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3002 	IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3003 	ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3004 
3005 	/*
3006 	 * set WTHRESH to encourage burst writeback, it should not be set
3007 	 * higher than 1 when:
3008 	 * - ITR is 0 as it could cause false TX hangs
3009 	 * - ITR is set to > 100k int/sec and BQL is enabled
3010 	 *
3011 	 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3012 	 * to or less than the number of on chip descriptors, which is
3013 	 * currently 40.
3014 	 */
3015 #if IS_ENABLED(CONFIG_BQL)
3016 	if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3017 #else
3018 	if (!ring->q_vector || (ring->q_vector->itr < 8))
3019 #endif
3020 		txdctl |= (1 << 16);	/* WTHRESH = 1 */
3021 	else
3022 		txdctl |= (8 << 16);	/* WTHRESH = 8 */
3023 
3024 	/*
3025 	 * Setting PTHRESH to 32 both improves performance
3026 	 * and avoids a TX hang with DFP enabled
3027 	 */
3028 	txdctl |= (1 << 8) |	/* HTHRESH = 1 */
3029 		   32;		/* PTHRESH = 32 */
3030 
3031 	/* reinitialize flowdirector state */
3032 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3033 		ring->atr_sample_rate = adapter->atr_sample_rate;
3034 		ring->atr_count = 0;
3035 		set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3036 	} else {
3037 		ring->atr_sample_rate = 0;
3038 	}
3039 
3040 	/* initialize XPS */
3041 	if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3042 		struct ixgbe_q_vector *q_vector = ring->q_vector;
3043 
3044 		if (q_vector)
3045 			netif_set_xps_queue(ring->netdev,
3046 					    &q_vector->affinity_mask,
3047 					    ring->queue_index);
3048 	}
3049 
3050 	clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3051 
3052 	/* enable queue */
3053 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3054 
3055 	/* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3056 	if (hw->mac.type == ixgbe_mac_82598EB &&
3057 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3058 		return;
3059 
3060 	/* poll to verify queue is enabled */
3061 	do {
3062 		usleep_range(1000, 2000);
3063 		txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3064 	} while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3065 	if (!wait_loop)
3066 		e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
3067 }
3068 
3069 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3070 {
3071 	struct ixgbe_hw *hw = &adapter->hw;
3072 	u32 rttdcs, mtqc;
3073 	u8 tcs = netdev_get_num_tc(adapter->netdev);
3074 
3075 	if (hw->mac.type == ixgbe_mac_82598EB)
3076 		return;
3077 
3078 	/* disable the arbiter while setting MTQC */
3079 	rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3080 	rttdcs |= IXGBE_RTTDCS_ARBDIS;
3081 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3082 
3083 	/* set transmit pool layout */
3084 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3085 		mtqc = IXGBE_MTQC_VT_ENA;
3086 		if (tcs > 4)
3087 			mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3088 		else if (tcs > 1)
3089 			mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3090 		else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3091 			mtqc |= IXGBE_MTQC_32VF;
3092 		else
3093 			mtqc |= IXGBE_MTQC_64VF;
3094 	} else {
3095 		if (tcs > 4)
3096 			mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3097 		else if (tcs > 1)
3098 			mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3099 		else
3100 			mtqc = IXGBE_MTQC_64Q_1PB;
3101 	}
3102 
3103 	IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3104 
3105 	/* Enable Security TX Buffer IFG for multiple pb */
3106 	if (tcs) {
3107 		u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3108 		sectx |= IXGBE_SECTX_DCB;
3109 		IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3110 	}
3111 
3112 	/* re-enable the arbiter */
3113 	rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3114 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3115 }
3116 
3117 /**
3118  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3119  * @adapter: board private structure
3120  *
3121  * Configure the Tx unit of the MAC after a reset.
3122  **/
3123 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3124 {
3125 	struct ixgbe_hw *hw = &adapter->hw;
3126 	u32 dmatxctl;
3127 	u32 i;
3128 
3129 	ixgbe_setup_mtqc(adapter);
3130 
3131 	if (hw->mac.type != ixgbe_mac_82598EB) {
3132 		/* DMATXCTL.EN must be before Tx queues are enabled */
3133 		dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3134 		dmatxctl |= IXGBE_DMATXCTL_TE;
3135 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3136 	}
3137 
3138 	/* Setup the HW Tx Head and Tail descriptor pointers */
3139 	for (i = 0; i < adapter->num_tx_queues; i++)
3140 		ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3141 }
3142 
3143 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3144 				 struct ixgbe_ring *ring)
3145 {
3146 	struct ixgbe_hw *hw = &adapter->hw;
3147 	u8 reg_idx = ring->reg_idx;
3148 	u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3149 
3150 	srrctl |= IXGBE_SRRCTL_DROP_EN;
3151 
3152 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3153 }
3154 
3155 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3156 				  struct ixgbe_ring *ring)
3157 {
3158 	struct ixgbe_hw *hw = &adapter->hw;
3159 	u8 reg_idx = ring->reg_idx;
3160 	u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3161 
3162 	srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3163 
3164 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3165 }
3166 
3167 #ifdef CONFIG_IXGBE_DCB
3168 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3169 #else
3170 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3171 #endif
3172 {
3173 	int i;
3174 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3175 
3176 	if (adapter->ixgbe_ieee_pfc)
3177 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3178 
3179 	/*
3180 	 * We should set the drop enable bit if:
3181 	 *  SR-IOV is enabled
3182 	 *   or
3183 	 *  Number of Rx queues > 1 and flow control is disabled
3184 	 *
3185 	 *  This allows us to avoid head of line blocking for security
3186 	 *  and performance reasons.
3187 	 */
3188 	if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3189 	    !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3190 		for (i = 0; i < adapter->num_rx_queues; i++)
3191 			ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3192 	} else {
3193 		for (i = 0; i < adapter->num_rx_queues; i++)
3194 			ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3195 	}
3196 }
3197 
3198 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3199 
3200 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3201 				   struct ixgbe_ring *rx_ring)
3202 {
3203 	struct ixgbe_hw *hw = &adapter->hw;
3204 	u32 srrctl;
3205 	u8 reg_idx = rx_ring->reg_idx;
3206 
3207 	if (hw->mac.type == ixgbe_mac_82598EB) {
3208 		u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3209 
3210 		/*
3211 		 * if VMDq is not active we must program one srrctl register
3212 		 * per RSS queue since we have enabled RDRXCTL.MVMEN
3213 		 */
3214 		reg_idx &= mask;
3215 	}
3216 
3217 	/* configure header buffer length, needed for RSC */
3218 	srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3219 
3220 	/* configure the packet buffer length */
3221 	srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3222 
3223 	/* configure descriptor type */
3224 	srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3225 
3226 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3227 }
3228 
3229 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3230 {
3231 	struct ixgbe_hw *hw = &adapter->hw;
3232 	static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
3233 			  0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
3234 			  0x6A3E67EA, 0x14364D17, 0x3BED200D};
3235 	u32 mrqc = 0, reta = 0;
3236 	u32 rxcsum;
3237 	int i, j;
3238 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3239 
3240 	/*
3241 	 * Program table for at least 2 queues w/ SR-IOV so that VFs can
3242 	 * make full use of any rings they may have.  We will use the
3243 	 * PSRTYPE register to control how many rings we use within the PF.
3244 	 */
3245 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 2))
3246 		rss_i = 2;
3247 
3248 	/* Fill out hash function seeds */
3249 	for (i = 0; i < 10; i++)
3250 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
3251 
3252 	/* Fill out redirection table */
3253 	for (i = 0, j = 0; i < 128; i++, j++) {
3254 		if (j == rss_i)
3255 			j = 0;
3256 		/* reta = 4-byte sliding window of
3257 		 * 0x00..(indices-1)(indices-1)00..etc. */
3258 		reta = (reta << 8) | (j * 0x11);
3259 		if ((i & 3) == 3)
3260 			IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3261 	}
3262 
3263 	/* Disable indicating checksum in descriptor, enables RSS hash */
3264 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3265 	rxcsum |= IXGBE_RXCSUM_PCSD;
3266 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3267 
3268 	if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3269 		if (adapter->ring_feature[RING_F_RSS].mask)
3270 			mrqc = IXGBE_MRQC_RSSEN;
3271 	} else {
3272 		u8 tcs = netdev_get_num_tc(adapter->netdev);
3273 
3274 		if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3275 			if (tcs > 4)
3276 				mrqc = IXGBE_MRQC_VMDQRT8TCEN;	/* 8 TCs */
3277 			else if (tcs > 1)
3278 				mrqc = IXGBE_MRQC_VMDQRT4TCEN;	/* 4 TCs */
3279 			else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3280 				mrqc = IXGBE_MRQC_VMDQRSS32EN;
3281 			else
3282 				mrqc = IXGBE_MRQC_VMDQRSS64EN;
3283 		} else {
3284 			if (tcs > 4)
3285 				mrqc = IXGBE_MRQC_RTRSS8TCEN;
3286 			else if (tcs > 1)
3287 				mrqc = IXGBE_MRQC_RTRSS4TCEN;
3288 			else
3289 				mrqc = IXGBE_MRQC_RSSEN;
3290 		}
3291 	}
3292 
3293 	/* Perform hash on these packet types */
3294 	mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3295 		IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3296 		IXGBE_MRQC_RSS_FIELD_IPV6 |
3297 		IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3298 
3299 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3300 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3301 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3302 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3303 
3304 	IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3305 }
3306 
3307 /**
3308  * ixgbe_configure_rscctl - enable RSC for the indicated ring
3309  * @adapter:    address of board private structure
3310  * @index:      index of ring to set
3311  **/
3312 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
3313 				   struct ixgbe_ring *ring)
3314 {
3315 	struct ixgbe_hw *hw = &adapter->hw;
3316 	u32 rscctrl;
3317 	u8 reg_idx = ring->reg_idx;
3318 
3319 	if (!ring_is_rsc_enabled(ring))
3320 		return;
3321 
3322 	rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
3323 	rscctrl |= IXGBE_RSCCTL_RSCEN;
3324 	/*
3325 	 * we must limit the number of descriptors so that the
3326 	 * total size of max desc * buf_len is not greater
3327 	 * than 65536
3328 	 */
3329 	rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
3330 	IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
3331 }
3332 
3333 #define IXGBE_MAX_RX_DESC_POLL 10
3334 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3335 				       struct ixgbe_ring *ring)
3336 {
3337 	struct ixgbe_hw *hw = &adapter->hw;
3338 	int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3339 	u32 rxdctl;
3340 	u8 reg_idx = ring->reg_idx;
3341 
3342 	if (ixgbe_removed(hw->hw_addr))
3343 		return;
3344 	/* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3345 	if (hw->mac.type == ixgbe_mac_82598EB &&
3346 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3347 		return;
3348 
3349 	do {
3350 		usleep_range(1000, 2000);
3351 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3352 	} while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3353 
3354 	if (!wait_loop) {
3355 		e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3356 		      "the polling period\n", reg_idx);
3357 	}
3358 }
3359 
3360 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3361 			    struct ixgbe_ring *ring)
3362 {
3363 	struct ixgbe_hw *hw = &adapter->hw;
3364 	int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3365 	u32 rxdctl;
3366 	u8 reg_idx = ring->reg_idx;
3367 
3368 	if (ixgbe_removed(hw->hw_addr))
3369 		return;
3370 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3371 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3372 
3373 	/* write value back with RXDCTL.ENABLE bit cleared */
3374 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3375 
3376 	if (hw->mac.type == ixgbe_mac_82598EB &&
3377 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3378 		return;
3379 
3380 	/* the hardware may take up to 100us to really disable the rx queue */
3381 	do {
3382 		udelay(10);
3383 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3384 	} while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3385 
3386 	if (!wait_loop) {
3387 		e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3388 		      "the polling period\n", reg_idx);
3389 	}
3390 }
3391 
3392 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3393 			     struct ixgbe_ring *ring)
3394 {
3395 	struct ixgbe_hw *hw = &adapter->hw;
3396 	u64 rdba = ring->dma;
3397 	u32 rxdctl;
3398 	u8 reg_idx = ring->reg_idx;
3399 
3400 	/* disable queue to avoid issues while updating state */
3401 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3402 	ixgbe_disable_rx_queue(adapter, ring);
3403 
3404 	IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3405 	IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3406 	IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3407 			ring->count * sizeof(union ixgbe_adv_rx_desc));
3408 	IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3409 	IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3410 	ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
3411 
3412 	ixgbe_configure_srrctl(adapter, ring);
3413 	ixgbe_configure_rscctl(adapter, ring);
3414 
3415 	if (hw->mac.type == ixgbe_mac_82598EB) {
3416 		/*
3417 		 * enable cache line friendly hardware writes:
3418 		 * PTHRESH=32 descriptors (half the internal cache),
3419 		 * this also removes ugly rx_no_buffer_count increment
3420 		 * HTHRESH=4 descriptors (to minimize latency on fetch)
3421 		 * WTHRESH=8 burst writeback up to two cache lines
3422 		 */
3423 		rxdctl &= ~0x3FFFFF;
3424 		rxdctl |=  0x080420;
3425 	}
3426 
3427 	/* enable receive descriptor ring */
3428 	rxdctl |= IXGBE_RXDCTL_ENABLE;
3429 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3430 
3431 	ixgbe_rx_desc_queue_enable(adapter, ring);
3432 	ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3433 }
3434 
3435 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3436 {
3437 	struct ixgbe_hw *hw = &adapter->hw;
3438 	int rss_i = adapter->ring_feature[RING_F_RSS].indices;
3439 	u16 pool;
3440 
3441 	/* PSRTYPE must be initialized in non 82598 adapters */
3442 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3443 		      IXGBE_PSRTYPE_UDPHDR |
3444 		      IXGBE_PSRTYPE_IPV4HDR |
3445 		      IXGBE_PSRTYPE_L2HDR |
3446 		      IXGBE_PSRTYPE_IPV6HDR;
3447 
3448 	if (hw->mac.type == ixgbe_mac_82598EB)
3449 		return;
3450 
3451 	if (rss_i > 3)
3452 		psrtype |= 2 << 29;
3453 	else if (rss_i > 1)
3454 		psrtype |= 1 << 29;
3455 
3456 	for_each_set_bit(pool, &adapter->fwd_bitmask, 32)
3457 		IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
3458 }
3459 
3460 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3461 {
3462 	struct ixgbe_hw *hw = &adapter->hw;
3463 	u32 reg_offset, vf_shift;
3464 	u32 gcr_ext, vmdctl;
3465 	int i;
3466 
3467 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3468 		return;
3469 
3470 	vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3471 	vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
3472 	vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
3473 	vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
3474 	vmdctl |= IXGBE_VT_CTL_REPLEN;
3475 	IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
3476 
3477 	vf_shift = VMDQ_P(0) % 32;
3478 	reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
3479 
3480 	/* Enable only the PF's pool for Tx/Rx */
3481 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (~0) << vf_shift);
3482 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
3483 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (~0) << vf_shift);
3484 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
3485 	if (adapter->flags2 & IXGBE_FLAG2_BRIDGE_MODE_VEB)
3486 		IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3487 
3488 	/* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3489 	hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
3490 
3491 	/*
3492 	 * Set up VF register offsets for selected VT Mode,
3493 	 * i.e. 32 or 64 VFs for SR-IOV
3494 	 */
3495 	switch (adapter->ring_feature[RING_F_VMDQ].mask) {
3496 	case IXGBE_82599_VMDQ_8Q_MASK:
3497 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
3498 		break;
3499 	case IXGBE_82599_VMDQ_4Q_MASK:
3500 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
3501 		break;
3502 	default:
3503 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
3504 		break;
3505 	}
3506 
3507 	IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3508 
3509 
3510 	/* Enable MAC Anti-Spoofing */
3511 	hw->mac.ops.set_mac_anti_spoofing(hw, (adapter->num_vfs != 0),
3512 					  adapter->num_vfs);
3513 	/* For VFs that have spoof checking turned off */
3514 	for (i = 0; i < adapter->num_vfs; i++) {
3515 		if (!adapter->vfinfo[i].spoofchk_enabled)
3516 			ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
3517 	}
3518 }
3519 
3520 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3521 {
3522 	struct ixgbe_hw *hw = &adapter->hw;
3523 	struct net_device *netdev = adapter->netdev;
3524 	int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3525 	struct ixgbe_ring *rx_ring;
3526 	int i;
3527 	u32 mhadd, hlreg0;
3528 
3529 #ifdef IXGBE_FCOE
3530 	/* adjust max frame to be able to do baby jumbo for FCoE */
3531 	if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3532 	    (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3533 		max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3534 
3535 #endif /* IXGBE_FCOE */
3536 
3537 	/* adjust max frame to be at least the size of a standard frame */
3538 	if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
3539 		max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
3540 
3541 	mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3542 	if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3543 		mhadd &= ~IXGBE_MHADD_MFS_MASK;
3544 		mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3545 
3546 		IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3547 	}
3548 
3549 	hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3550 	/* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3551 	hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3552 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3553 
3554 	/*
3555 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
3556 	 * the Base and Length of the Rx Descriptor Ring
3557 	 */
3558 	for (i = 0; i < adapter->num_rx_queues; i++) {
3559 		rx_ring = adapter->rx_ring[i];
3560 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3561 			set_ring_rsc_enabled(rx_ring);
3562 		else
3563 			clear_ring_rsc_enabled(rx_ring);
3564 	}
3565 }
3566 
3567 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3568 {
3569 	struct ixgbe_hw *hw = &adapter->hw;
3570 	u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3571 
3572 	switch (hw->mac.type) {
3573 	case ixgbe_mac_82598EB:
3574 		/*
3575 		 * For VMDq support of different descriptor types or
3576 		 * buffer sizes through the use of multiple SRRCTL
3577 		 * registers, RDRXCTL.MVMEN must be set to 1
3578 		 *
3579 		 * also, the manual doesn't mention it clearly but DCA hints
3580 		 * will only use queue 0's tags unless this bit is set.  Side
3581 		 * effects of setting this bit are only that SRRCTL must be
3582 		 * fully programmed [0..15]
3583 		 */
3584 		rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3585 		break;
3586 	case ixgbe_mac_82599EB:
3587 	case ixgbe_mac_X540:
3588 		/* Disable RSC for ACK packets */
3589 		IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3590 		   (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3591 		rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3592 		/* hardware requires some bits to be set by default */
3593 		rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3594 		rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3595 		break;
3596 	default:
3597 		/* We should do nothing since we don't know this hardware */
3598 		return;
3599 	}
3600 
3601 	IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3602 }
3603 
3604 /**
3605  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3606  * @adapter: board private structure
3607  *
3608  * Configure the Rx unit of the MAC after a reset.
3609  **/
3610 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3611 {
3612 	struct ixgbe_hw *hw = &adapter->hw;
3613 	int i;
3614 	u32 rxctrl, rfctl;
3615 
3616 	/* disable receives while setting up the descriptors */
3617 	rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3618 	IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3619 
3620 	ixgbe_setup_psrtype(adapter);
3621 	ixgbe_setup_rdrxctl(adapter);
3622 
3623 	/* RSC Setup */
3624 	rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
3625 	rfctl &= ~IXGBE_RFCTL_RSC_DIS;
3626 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
3627 		rfctl |= IXGBE_RFCTL_RSC_DIS;
3628 	IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
3629 
3630 	/* Program registers for the distribution of queues */
3631 	ixgbe_setup_mrqc(adapter);
3632 
3633 	/* set_rx_buffer_len must be called before ring initialization */
3634 	ixgbe_set_rx_buffer_len(adapter);
3635 
3636 	/*
3637 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
3638 	 * the Base and Length of the Rx Descriptor Ring
3639 	 */
3640 	for (i = 0; i < adapter->num_rx_queues; i++)
3641 		ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3642 
3643 	/* disable drop enable for 82598 parts */
3644 	if (hw->mac.type == ixgbe_mac_82598EB)
3645 		rxctrl |= IXGBE_RXCTRL_DMBYPS;
3646 
3647 	/* enable all receives */
3648 	rxctrl |= IXGBE_RXCTRL_RXEN;
3649 	hw->mac.ops.enable_rx_dma(hw, rxctrl);
3650 }
3651 
3652 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
3653 				 __be16 proto, u16 vid)
3654 {
3655 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3656 	struct ixgbe_hw *hw = &adapter->hw;
3657 
3658 	/* add VID to filter table */
3659 	hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true);
3660 	set_bit(vid, adapter->active_vlans);
3661 
3662 	return 0;
3663 }
3664 
3665 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
3666 				  __be16 proto, u16 vid)
3667 {
3668 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3669 	struct ixgbe_hw *hw = &adapter->hw;
3670 
3671 	/* remove VID from filter table */
3672 	hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), false);
3673 	clear_bit(vid, adapter->active_vlans);
3674 
3675 	return 0;
3676 }
3677 
3678 /**
3679  * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3680  * @adapter: driver data
3681  */
3682 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3683 {
3684 	struct ixgbe_hw *hw = &adapter->hw;
3685 	u32 vlnctrl;
3686 
3687 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3688 	vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3689 	IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3690 }
3691 
3692 /**
3693  * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3694  * @adapter: driver data
3695  */
3696 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3697 {
3698 	struct ixgbe_hw *hw = &adapter->hw;
3699 	u32 vlnctrl;
3700 
3701 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3702 	vlnctrl |= IXGBE_VLNCTRL_VFE;
3703 	vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3704 	IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3705 }
3706 
3707 /**
3708  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3709  * @adapter: driver data
3710  */
3711 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3712 {
3713 	struct ixgbe_hw *hw = &adapter->hw;
3714 	u32 vlnctrl;
3715 	int i, j;
3716 
3717 	switch (hw->mac.type) {
3718 	case ixgbe_mac_82598EB:
3719 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3720 		vlnctrl &= ~IXGBE_VLNCTRL_VME;
3721 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3722 		break;
3723 	case ixgbe_mac_82599EB:
3724 	case ixgbe_mac_X540:
3725 		for (i = 0; i < adapter->num_rx_queues; i++) {
3726 			struct ixgbe_ring *ring = adapter->rx_ring[i];
3727 
3728 			if (ring->l2_accel_priv)
3729 				continue;
3730 			j = ring->reg_idx;
3731 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3732 			vlnctrl &= ~IXGBE_RXDCTL_VME;
3733 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3734 		}
3735 		break;
3736 	default:
3737 		break;
3738 	}
3739 }
3740 
3741 /**
3742  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3743  * @adapter: driver data
3744  */
3745 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3746 {
3747 	struct ixgbe_hw *hw = &adapter->hw;
3748 	u32 vlnctrl;
3749 	int i, j;
3750 
3751 	switch (hw->mac.type) {
3752 	case ixgbe_mac_82598EB:
3753 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3754 		vlnctrl |= IXGBE_VLNCTRL_VME;
3755 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3756 		break;
3757 	case ixgbe_mac_82599EB:
3758 	case ixgbe_mac_X540:
3759 		for (i = 0; i < adapter->num_rx_queues; i++) {
3760 			struct ixgbe_ring *ring = adapter->rx_ring[i];
3761 
3762 			if (ring->l2_accel_priv)
3763 				continue;
3764 			j = ring->reg_idx;
3765 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3766 			vlnctrl |= IXGBE_RXDCTL_VME;
3767 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3768 		}
3769 		break;
3770 	default:
3771 		break;
3772 	}
3773 }
3774 
3775 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3776 {
3777 	u16 vid;
3778 
3779 	ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
3780 
3781 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3782 		ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
3783 }
3784 
3785 /**
3786  * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3787  * @netdev: network interface device structure
3788  *
3789  * Writes unicast address list to the RAR table.
3790  * Returns: -ENOMEM on failure/insufficient address space
3791  *                0 on no addresses written
3792  *                X on writing X addresses to the RAR table
3793  **/
3794 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3795 {
3796 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3797 	struct ixgbe_hw *hw = &adapter->hw;
3798 	unsigned int rar_entries = hw->mac.num_rar_entries - 1;
3799 	int count = 0;
3800 
3801 	/* In SR-IOV/VMDQ modes significantly less RAR entries are available */
3802 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3803 		rar_entries = IXGBE_MAX_PF_MACVLANS - 1;
3804 
3805 	/* return ENOMEM indicating insufficient memory for addresses */
3806 	if (netdev_uc_count(netdev) > rar_entries)
3807 		return -ENOMEM;
3808 
3809 	if (!netdev_uc_empty(netdev)) {
3810 		struct netdev_hw_addr *ha;
3811 		/* return error if we do not support writing to RAR table */
3812 		if (!hw->mac.ops.set_rar)
3813 			return -ENOMEM;
3814 
3815 		netdev_for_each_uc_addr(ha, netdev) {
3816 			if (!rar_entries)
3817 				break;
3818 			hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3819 					    VMDQ_P(0), IXGBE_RAH_AV);
3820 			count++;
3821 		}
3822 	}
3823 	/* write the addresses in reverse order to avoid write combining */
3824 	for (; rar_entries > 0 ; rar_entries--)
3825 		hw->mac.ops.clear_rar(hw, rar_entries);
3826 
3827 	return count;
3828 }
3829 
3830 /**
3831  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3832  * @netdev: network interface device structure
3833  *
3834  * The set_rx_method entry point is called whenever the unicast/multicast
3835  * address list or the network interface flags are updated.  This routine is
3836  * responsible for configuring the hardware for proper unicast, multicast and
3837  * promiscuous mode.
3838  **/
3839 void ixgbe_set_rx_mode(struct net_device *netdev)
3840 {
3841 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3842 	struct ixgbe_hw *hw = &adapter->hw;
3843 	u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3844 	int count;
3845 
3846 	/* Check for Promiscuous and All Multicast modes */
3847 
3848 	fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3849 
3850 	/* set all bits that we expect to always be set */
3851 	fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
3852 	fctrl |= IXGBE_FCTRL_BAM;
3853 	fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3854 	fctrl |= IXGBE_FCTRL_PMCF;
3855 
3856 	/* clear the bits we are changing the status of */
3857 	fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3858 
3859 	if (netdev->flags & IFF_PROMISC) {
3860 		hw->addr_ctrl.user_set_promisc = true;
3861 		fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3862 		vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3863 		/* Only disable hardware filter vlans in promiscuous mode
3864 		 * if SR-IOV and VMDQ are disabled - otherwise ensure
3865 		 * that hardware VLAN filters remain enabled.
3866 		 */
3867 		if (!(adapter->flags & (IXGBE_FLAG_VMDQ_ENABLED |
3868 					IXGBE_FLAG_SRIOV_ENABLED)))
3869 			ixgbe_vlan_filter_disable(adapter);
3870 		else
3871 			ixgbe_vlan_filter_enable(adapter);
3872 	} else {
3873 		if (netdev->flags & IFF_ALLMULTI) {
3874 			fctrl |= IXGBE_FCTRL_MPE;
3875 			vmolr |= IXGBE_VMOLR_MPE;
3876 		}
3877 		ixgbe_vlan_filter_enable(adapter);
3878 		hw->addr_ctrl.user_set_promisc = false;
3879 	}
3880 
3881 	/*
3882 	 * Write addresses to available RAR registers, if there is not
3883 	 * sufficient space to store all the addresses then enable
3884 	 * unicast promiscuous mode
3885 	 */
3886 	count = ixgbe_write_uc_addr_list(netdev);
3887 	if (count < 0) {
3888 		fctrl |= IXGBE_FCTRL_UPE;
3889 		vmolr |= IXGBE_VMOLR_ROPE;
3890 	}
3891 
3892 	/* Write addresses to the MTA, if the attempt fails
3893 	 * then we should just turn on promiscuous mode so
3894 	 * that we can at least receive multicast traffic
3895 	 */
3896 	hw->mac.ops.update_mc_addr_list(hw, netdev);
3897 	vmolr |= IXGBE_VMOLR_ROMPE;
3898 
3899 	if (adapter->num_vfs)
3900 		ixgbe_restore_vf_multicasts(adapter);
3901 
3902 	if (hw->mac.type != ixgbe_mac_82598EB) {
3903 		vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
3904 			 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3905 			   IXGBE_VMOLR_ROPE);
3906 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
3907 	}
3908 
3909 	/* This is useful for sniffing bad packets. */
3910 	if (adapter->netdev->features & NETIF_F_RXALL) {
3911 		/* UPE and MPE will be handled by normal PROMISC logic
3912 		 * in e1000e_set_rx_mode */
3913 		fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
3914 			  IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
3915 			  IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
3916 
3917 		fctrl &= ~(IXGBE_FCTRL_DPF);
3918 		/* NOTE:  VLAN filtering is disabled by setting PROMISC */
3919 	}
3920 
3921 	IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3922 
3923 	if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
3924 		ixgbe_vlan_strip_enable(adapter);
3925 	else
3926 		ixgbe_vlan_strip_disable(adapter);
3927 }
3928 
3929 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3930 {
3931 	int q_idx;
3932 
3933 	for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++) {
3934 		ixgbe_qv_init_lock(adapter->q_vector[q_idx]);
3935 		napi_enable(&adapter->q_vector[q_idx]->napi);
3936 	}
3937 }
3938 
3939 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3940 {
3941 	int q_idx;
3942 
3943 	for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++) {
3944 		napi_disable(&adapter->q_vector[q_idx]->napi);
3945 		while (!ixgbe_qv_disable(adapter->q_vector[q_idx])) {
3946 			pr_info("QV %d locked\n", q_idx);
3947 			usleep_range(1000, 20000);
3948 		}
3949 	}
3950 }
3951 
3952 #ifdef CONFIG_IXGBE_DCB
3953 /**
3954  * ixgbe_configure_dcb - Configure DCB hardware
3955  * @adapter: ixgbe adapter struct
3956  *
3957  * This is called by the driver on open to configure the DCB hardware.
3958  * This is also called by the gennetlink interface when reconfiguring
3959  * the DCB state.
3960  */
3961 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3962 {
3963 	struct ixgbe_hw *hw = &adapter->hw;
3964 	int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3965 
3966 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3967 		if (hw->mac.type == ixgbe_mac_82598EB)
3968 			netif_set_gso_max_size(adapter->netdev, 65536);
3969 		return;
3970 	}
3971 
3972 	if (hw->mac.type == ixgbe_mac_82598EB)
3973 		netif_set_gso_max_size(adapter->netdev, 32768);
3974 
3975 #ifdef IXGBE_FCOE
3976 	if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3977 		max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3978 #endif
3979 
3980 	/* reconfigure the hardware */
3981 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
3982 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3983 						DCB_TX_CONFIG);
3984 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3985 						DCB_RX_CONFIG);
3986 		ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3987 	} else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
3988 		ixgbe_dcb_hw_ets(&adapter->hw,
3989 				 adapter->ixgbe_ieee_ets,
3990 				 max_frame);
3991 		ixgbe_dcb_hw_pfc_config(&adapter->hw,
3992 					adapter->ixgbe_ieee_pfc->pfc_en,
3993 					adapter->ixgbe_ieee_ets->prio_tc);
3994 	}
3995 
3996 	/* Enable RSS Hash per TC */
3997 	if (hw->mac.type != ixgbe_mac_82598EB) {
3998 		u32 msb = 0;
3999 		u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
4000 
4001 		while (rss_i) {
4002 			msb++;
4003 			rss_i >>= 1;
4004 		}
4005 
4006 		/* write msb to all 8 TCs in one write */
4007 		IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
4008 	}
4009 }
4010 #endif
4011 
4012 /* Additional bittime to account for IXGBE framing */
4013 #define IXGBE_ETH_FRAMING 20
4014 
4015 /**
4016  * ixgbe_hpbthresh - calculate high water mark for flow control
4017  *
4018  * @adapter: board private structure to calculate for
4019  * @pb: packet buffer to calculate
4020  */
4021 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
4022 {
4023 	struct ixgbe_hw *hw = &adapter->hw;
4024 	struct net_device *dev = adapter->netdev;
4025 	int link, tc, kb, marker;
4026 	u32 dv_id, rx_pba;
4027 
4028 	/* Calculate max LAN frame size */
4029 	tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
4030 
4031 #ifdef IXGBE_FCOE
4032 	/* FCoE traffic class uses FCOE jumbo frames */
4033 	if ((dev->features & NETIF_F_FCOE_MTU) &&
4034 	    (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
4035 	    (pb == ixgbe_fcoe_get_tc(adapter)))
4036 		tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4037 
4038 #endif
4039 	/* Calculate delay value for device */
4040 	switch (hw->mac.type) {
4041 	case ixgbe_mac_X540:
4042 		dv_id = IXGBE_DV_X540(link, tc);
4043 		break;
4044 	default:
4045 		dv_id = IXGBE_DV(link, tc);
4046 		break;
4047 	}
4048 
4049 	/* Loopback switch introduces additional latency */
4050 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4051 		dv_id += IXGBE_B2BT(tc);
4052 
4053 	/* Delay value is calculated in bit times convert to KB */
4054 	kb = IXGBE_BT2KB(dv_id);
4055 	rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
4056 
4057 	marker = rx_pba - kb;
4058 
4059 	/* It is possible that the packet buffer is not large enough
4060 	 * to provide required headroom. In this case throw an error
4061 	 * to user and a do the best we can.
4062 	 */
4063 	if (marker < 0) {
4064 		e_warn(drv, "Packet Buffer(%i) can not provide enough"
4065 			    "headroom to support flow control."
4066 			    "Decrease MTU or number of traffic classes\n", pb);
4067 		marker = tc + 1;
4068 	}
4069 
4070 	return marker;
4071 }
4072 
4073 /**
4074  * ixgbe_lpbthresh - calculate low water mark for for flow control
4075  *
4076  * @adapter: board private structure to calculate for
4077  * @pb: packet buffer to calculate
4078  */
4079 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter)
4080 {
4081 	struct ixgbe_hw *hw = &adapter->hw;
4082 	struct net_device *dev = adapter->netdev;
4083 	int tc;
4084 	u32 dv_id;
4085 
4086 	/* Calculate max LAN frame size */
4087 	tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
4088 
4089 	/* Calculate delay value for device */
4090 	switch (hw->mac.type) {
4091 	case ixgbe_mac_X540:
4092 		dv_id = IXGBE_LOW_DV_X540(tc);
4093 		break;
4094 	default:
4095 		dv_id = IXGBE_LOW_DV(tc);
4096 		break;
4097 	}
4098 
4099 	/* Delay value is calculated in bit times convert to KB */
4100 	return IXGBE_BT2KB(dv_id);
4101 }
4102 
4103 /*
4104  * ixgbe_pbthresh_setup - calculate and setup high low water marks
4105  */
4106 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
4107 {
4108 	struct ixgbe_hw *hw = &adapter->hw;
4109 	int num_tc = netdev_get_num_tc(adapter->netdev);
4110 	int i;
4111 
4112 	if (!num_tc)
4113 		num_tc = 1;
4114 
4115 	hw->fc.low_water = ixgbe_lpbthresh(adapter);
4116 
4117 	for (i = 0; i < num_tc; i++) {
4118 		hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
4119 
4120 		/* Low water marks must not be larger than high water marks */
4121 		if (hw->fc.low_water > hw->fc.high_water[i])
4122 			hw->fc.low_water = 0;
4123 	}
4124 }
4125 
4126 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
4127 {
4128 	struct ixgbe_hw *hw = &adapter->hw;
4129 	int hdrm;
4130 	u8 tc = netdev_get_num_tc(adapter->netdev);
4131 
4132 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4133 	    adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
4134 		hdrm = 32 << adapter->fdir_pballoc;
4135 	else
4136 		hdrm = 0;
4137 
4138 	hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
4139 	ixgbe_pbthresh_setup(adapter);
4140 }
4141 
4142 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
4143 {
4144 	struct ixgbe_hw *hw = &adapter->hw;
4145 	struct hlist_node *node2;
4146 	struct ixgbe_fdir_filter *filter;
4147 
4148 	spin_lock(&adapter->fdir_perfect_lock);
4149 
4150 	if (!hlist_empty(&adapter->fdir_filter_list))
4151 		ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
4152 
4153 	hlist_for_each_entry_safe(filter, node2,
4154 				  &adapter->fdir_filter_list, fdir_node) {
4155 		ixgbe_fdir_write_perfect_filter_82599(hw,
4156 				&filter->filter,
4157 				filter->sw_idx,
4158 				(filter->action == IXGBE_FDIR_DROP_QUEUE) ?
4159 				IXGBE_FDIR_DROP_QUEUE :
4160 				adapter->rx_ring[filter->action]->reg_idx);
4161 	}
4162 
4163 	spin_unlock(&adapter->fdir_perfect_lock);
4164 }
4165 
4166 static void ixgbe_macvlan_set_rx_mode(struct net_device *dev, unsigned int pool,
4167 				      struct ixgbe_adapter *adapter)
4168 {
4169 	struct ixgbe_hw *hw = &adapter->hw;
4170 	u32 vmolr;
4171 
4172 	/* No unicast promiscuous support for VMDQ devices. */
4173 	vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(pool));
4174 	vmolr |= (IXGBE_VMOLR_ROMPE | IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE);
4175 
4176 	/* clear the affected bit */
4177 	vmolr &= ~IXGBE_VMOLR_MPE;
4178 
4179 	if (dev->flags & IFF_ALLMULTI) {
4180 		vmolr |= IXGBE_VMOLR_MPE;
4181 	} else {
4182 		vmolr |= IXGBE_VMOLR_ROMPE;
4183 		hw->mac.ops.update_mc_addr_list(hw, dev);
4184 	}
4185 	ixgbe_write_uc_addr_list(adapter->netdev);
4186 	IXGBE_WRITE_REG(hw, IXGBE_VMOLR(pool), vmolr);
4187 }
4188 
4189 static void ixgbe_add_mac_filter(struct ixgbe_adapter *adapter,
4190 				 u8 *addr, u16 pool)
4191 {
4192 	struct ixgbe_hw *hw = &adapter->hw;
4193 	unsigned int entry;
4194 
4195 	entry = hw->mac.num_rar_entries - pool;
4196 	hw->mac.ops.set_rar(hw, entry, addr, VMDQ_P(pool), IXGBE_RAH_AV);
4197 }
4198 
4199 static void ixgbe_fwd_psrtype(struct ixgbe_fwd_adapter *vadapter)
4200 {
4201 	struct ixgbe_adapter *adapter = vadapter->real_adapter;
4202 	int rss_i = adapter->num_rx_queues_per_pool;
4203 	struct ixgbe_hw *hw = &adapter->hw;
4204 	u16 pool = vadapter->pool;
4205 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4206 		      IXGBE_PSRTYPE_UDPHDR |
4207 		      IXGBE_PSRTYPE_IPV4HDR |
4208 		      IXGBE_PSRTYPE_L2HDR |
4209 		      IXGBE_PSRTYPE_IPV6HDR;
4210 
4211 	if (hw->mac.type == ixgbe_mac_82598EB)
4212 		return;
4213 
4214 	if (rss_i > 3)
4215 		psrtype |= 2 << 29;
4216 	else if (rss_i > 1)
4217 		psrtype |= 1 << 29;
4218 
4219 	IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4220 }
4221 
4222 /**
4223  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4224  * @rx_ring: ring to free buffers from
4225  **/
4226 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4227 {
4228 	struct device *dev = rx_ring->dev;
4229 	unsigned long size;
4230 	u16 i;
4231 
4232 	/* ring already cleared, nothing to do */
4233 	if (!rx_ring->rx_buffer_info)
4234 		return;
4235 
4236 	/* Free all the Rx ring sk_buffs */
4237 	for (i = 0; i < rx_ring->count; i++) {
4238 		struct ixgbe_rx_buffer *rx_buffer;
4239 
4240 		rx_buffer = &rx_ring->rx_buffer_info[i];
4241 		if (rx_buffer->skb) {
4242 			struct sk_buff *skb = rx_buffer->skb;
4243 			if (IXGBE_CB(skb)->page_released) {
4244 				dma_unmap_page(dev,
4245 					       IXGBE_CB(skb)->dma,
4246 					       ixgbe_rx_bufsz(rx_ring),
4247 					       DMA_FROM_DEVICE);
4248 				IXGBE_CB(skb)->page_released = false;
4249 			}
4250 			dev_kfree_skb(skb);
4251 		}
4252 		rx_buffer->skb = NULL;
4253 		if (rx_buffer->dma)
4254 			dma_unmap_page(dev, rx_buffer->dma,
4255 				       ixgbe_rx_pg_size(rx_ring),
4256 				       DMA_FROM_DEVICE);
4257 		rx_buffer->dma = 0;
4258 		if (rx_buffer->page)
4259 			__free_pages(rx_buffer->page,
4260 				     ixgbe_rx_pg_order(rx_ring));
4261 		rx_buffer->page = NULL;
4262 	}
4263 
4264 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4265 	memset(rx_ring->rx_buffer_info, 0, size);
4266 
4267 	/* Zero out the descriptor ring */
4268 	memset(rx_ring->desc, 0, rx_ring->size);
4269 
4270 	rx_ring->next_to_alloc = 0;
4271 	rx_ring->next_to_clean = 0;
4272 	rx_ring->next_to_use = 0;
4273 }
4274 
4275 static void ixgbe_disable_fwd_ring(struct ixgbe_fwd_adapter *vadapter,
4276 				   struct ixgbe_ring *rx_ring)
4277 {
4278 	struct ixgbe_adapter *adapter = vadapter->real_adapter;
4279 	int index = rx_ring->queue_index + vadapter->rx_base_queue;
4280 
4281 	/* shutdown specific queue receive and wait for dma to settle */
4282 	ixgbe_disable_rx_queue(adapter, rx_ring);
4283 	usleep_range(10000, 20000);
4284 	ixgbe_irq_disable_queues(adapter, ((u64)1 << index));
4285 	ixgbe_clean_rx_ring(rx_ring);
4286 	rx_ring->l2_accel_priv = NULL;
4287 }
4288 
4289 static int ixgbe_fwd_ring_down(struct net_device *vdev,
4290 			       struct ixgbe_fwd_adapter *accel)
4291 {
4292 	struct ixgbe_adapter *adapter = accel->real_adapter;
4293 	unsigned int rxbase = accel->rx_base_queue;
4294 	unsigned int txbase = accel->tx_base_queue;
4295 	int i;
4296 
4297 	netif_tx_stop_all_queues(vdev);
4298 
4299 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4300 		ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
4301 		adapter->rx_ring[rxbase + i]->netdev = adapter->netdev;
4302 	}
4303 
4304 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4305 		adapter->tx_ring[txbase + i]->l2_accel_priv = NULL;
4306 		adapter->tx_ring[txbase + i]->netdev = adapter->netdev;
4307 	}
4308 
4309 
4310 	return 0;
4311 }
4312 
4313 static int ixgbe_fwd_ring_up(struct net_device *vdev,
4314 			     struct ixgbe_fwd_adapter *accel)
4315 {
4316 	struct ixgbe_adapter *adapter = accel->real_adapter;
4317 	unsigned int rxbase, txbase, queues;
4318 	int i, baseq, err = 0;
4319 
4320 	if (!test_bit(accel->pool, &adapter->fwd_bitmask))
4321 		return 0;
4322 
4323 	baseq = accel->pool * adapter->num_rx_queues_per_pool;
4324 	netdev_dbg(vdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
4325 		   accel->pool, adapter->num_rx_pools,
4326 		   baseq, baseq + adapter->num_rx_queues_per_pool,
4327 		   adapter->fwd_bitmask);
4328 
4329 	accel->netdev = vdev;
4330 	accel->rx_base_queue = rxbase = baseq;
4331 	accel->tx_base_queue = txbase = baseq;
4332 
4333 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
4334 		ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
4335 
4336 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4337 		adapter->rx_ring[rxbase + i]->netdev = vdev;
4338 		adapter->rx_ring[rxbase + i]->l2_accel_priv = accel;
4339 		ixgbe_configure_rx_ring(adapter, adapter->rx_ring[rxbase + i]);
4340 	}
4341 
4342 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4343 		adapter->tx_ring[txbase + i]->netdev = vdev;
4344 		adapter->tx_ring[txbase + i]->l2_accel_priv = accel;
4345 	}
4346 
4347 	queues = min_t(unsigned int,
4348 		       adapter->num_rx_queues_per_pool, vdev->num_tx_queues);
4349 	err = netif_set_real_num_tx_queues(vdev, queues);
4350 	if (err)
4351 		goto fwd_queue_err;
4352 
4353 	err = netif_set_real_num_rx_queues(vdev, queues);
4354 	if (err)
4355 		goto fwd_queue_err;
4356 
4357 	if (is_valid_ether_addr(vdev->dev_addr))
4358 		ixgbe_add_mac_filter(adapter, vdev->dev_addr, accel->pool);
4359 
4360 	ixgbe_fwd_psrtype(accel);
4361 	ixgbe_macvlan_set_rx_mode(vdev, accel->pool, adapter);
4362 	return err;
4363 fwd_queue_err:
4364 	ixgbe_fwd_ring_down(vdev, accel);
4365 	return err;
4366 }
4367 
4368 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
4369 {
4370 	struct net_device *upper;
4371 	struct list_head *iter;
4372 	int err;
4373 
4374 	netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
4375 		if (netif_is_macvlan(upper)) {
4376 			struct macvlan_dev *dfwd = netdev_priv(upper);
4377 			struct ixgbe_fwd_adapter *vadapter = dfwd->fwd_priv;
4378 
4379 			if (dfwd->fwd_priv) {
4380 				err = ixgbe_fwd_ring_up(upper, vadapter);
4381 				if (err)
4382 					continue;
4383 			}
4384 		}
4385 	}
4386 }
4387 
4388 static void ixgbe_configure(struct ixgbe_adapter *adapter)
4389 {
4390 	struct ixgbe_hw *hw = &adapter->hw;
4391 
4392 	ixgbe_configure_pb(adapter);
4393 #ifdef CONFIG_IXGBE_DCB
4394 	ixgbe_configure_dcb(adapter);
4395 #endif
4396 	/*
4397 	 * We must restore virtualization before VLANs or else
4398 	 * the VLVF registers will not be populated
4399 	 */
4400 	ixgbe_configure_virtualization(adapter);
4401 
4402 	ixgbe_set_rx_mode(adapter->netdev);
4403 	ixgbe_restore_vlan(adapter);
4404 
4405 	switch (hw->mac.type) {
4406 	case ixgbe_mac_82599EB:
4407 	case ixgbe_mac_X540:
4408 		hw->mac.ops.disable_rx_buff(hw);
4409 		break;
4410 	default:
4411 		break;
4412 	}
4413 
4414 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
4415 		ixgbe_init_fdir_signature_82599(&adapter->hw,
4416 						adapter->fdir_pballoc);
4417 	} else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
4418 		ixgbe_init_fdir_perfect_82599(&adapter->hw,
4419 					      adapter->fdir_pballoc);
4420 		ixgbe_fdir_filter_restore(adapter);
4421 	}
4422 
4423 	switch (hw->mac.type) {
4424 	case ixgbe_mac_82599EB:
4425 	case ixgbe_mac_X540:
4426 		hw->mac.ops.enable_rx_buff(hw);
4427 		break;
4428 	default:
4429 		break;
4430 	}
4431 
4432 #ifdef IXGBE_FCOE
4433 	/* configure FCoE L2 filters, redirection table, and Rx control */
4434 	ixgbe_configure_fcoe(adapter);
4435 
4436 #endif /* IXGBE_FCOE */
4437 	ixgbe_configure_tx(adapter);
4438 	ixgbe_configure_rx(adapter);
4439 	ixgbe_configure_dfwd(adapter);
4440 }
4441 
4442 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
4443 {
4444 	switch (hw->phy.type) {
4445 	case ixgbe_phy_sfp_avago:
4446 	case ixgbe_phy_sfp_ftl:
4447 	case ixgbe_phy_sfp_intel:
4448 	case ixgbe_phy_sfp_unknown:
4449 	case ixgbe_phy_sfp_passive_tyco:
4450 	case ixgbe_phy_sfp_passive_unknown:
4451 	case ixgbe_phy_sfp_active_unknown:
4452 	case ixgbe_phy_sfp_ftl_active:
4453 	case ixgbe_phy_qsfp_passive_unknown:
4454 	case ixgbe_phy_qsfp_active_unknown:
4455 	case ixgbe_phy_qsfp_intel:
4456 	case ixgbe_phy_qsfp_unknown:
4457 		return true;
4458 	case ixgbe_phy_nl:
4459 		if (hw->mac.type == ixgbe_mac_82598EB)
4460 			return true;
4461 	default:
4462 		return false;
4463 	}
4464 }
4465 
4466 /**
4467  * ixgbe_sfp_link_config - set up SFP+ link
4468  * @adapter: pointer to private adapter struct
4469  **/
4470 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
4471 {
4472 	/*
4473 	 * We are assuming the worst case scenario here, and that
4474 	 * is that an SFP was inserted/removed after the reset
4475 	 * but before SFP detection was enabled.  As such the best
4476 	 * solution is to just start searching as soon as we start
4477 	 */
4478 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
4479 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
4480 
4481 	adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
4482 }
4483 
4484 /**
4485  * ixgbe_non_sfp_link_config - set up non-SFP+ link
4486  * @hw: pointer to private hardware struct
4487  *
4488  * Returns 0 on success, negative on failure
4489  **/
4490 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
4491 {
4492 	u32 speed;
4493 	bool autoneg, link_up = false;
4494 	u32 ret = IXGBE_ERR_LINK_SETUP;
4495 
4496 	if (hw->mac.ops.check_link)
4497 		ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
4498 
4499 	if (ret)
4500 		goto link_cfg_out;
4501 
4502 	speed = hw->phy.autoneg_advertised;
4503 	if ((!speed) && (hw->mac.ops.get_link_capabilities))
4504 		ret = hw->mac.ops.get_link_capabilities(hw, &speed,
4505 							&autoneg);
4506 	if (ret)
4507 		goto link_cfg_out;
4508 
4509 	if (hw->mac.ops.setup_link)
4510 		ret = hw->mac.ops.setup_link(hw, speed, link_up);
4511 link_cfg_out:
4512 	return ret;
4513 }
4514 
4515 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
4516 {
4517 	struct ixgbe_hw *hw = &adapter->hw;
4518 	u32 gpie = 0;
4519 
4520 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4521 		gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
4522 		       IXGBE_GPIE_OCD;
4523 		gpie |= IXGBE_GPIE_EIAME;
4524 		/*
4525 		 * use EIAM to auto-mask when MSI-X interrupt is asserted
4526 		 * this saves a register write for every interrupt
4527 		 */
4528 		switch (hw->mac.type) {
4529 		case ixgbe_mac_82598EB:
4530 			IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4531 			break;
4532 		case ixgbe_mac_82599EB:
4533 		case ixgbe_mac_X540:
4534 		default:
4535 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
4536 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
4537 			break;
4538 		}
4539 	} else {
4540 		/* legacy interrupts, use EIAM to auto-mask when reading EICR,
4541 		 * specifically only auto mask tx and rx interrupts */
4542 		IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4543 	}
4544 
4545 	/* XXX: to interrupt immediately for EICS writes, enable this */
4546 	/* gpie |= IXGBE_GPIE_EIMEN; */
4547 
4548 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
4549 		gpie &= ~IXGBE_GPIE_VTMODE_MASK;
4550 
4551 		switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4552 		case IXGBE_82599_VMDQ_8Q_MASK:
4553 			gpie |= IXGBE_GPIE_VTMODE_16;
4554 			break;
4555 		case IXGBE_82599_VMDQ_4Q_MASK:
4556 			gpie |= IXGBE_GPIE_VTMODE_32;
4557 			break;
4558 		default:
4559 			gpie |= IXGBE_GPIE_VTMODE_64;
4560 			break;
4561 		}
4562 	}
4563 
4564 	/* Enable Thermal over heat sensor interrupt */
4565 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
4566 		switch (adapter->hw.mac.type) {
4567 		case ixgbe_mac_82599EB:
4568 			gpie |= IXGBE_SDP0_GPIEN;
4569 			break;
4570 		case ixgbe_mac_X540:
4571 			gpie |= IXGBE_EIMS_TS;
4572 			break;
4573 		default:
4574 			break;
4575 		}
4576 	}
4577 
4578 	/* Enable fan failure interrupt */
4579 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
4580 		gpie |= IXGBE_SDP1_GPIEN;
4581 
4582 	if (hw->mac.type == ixgbe_mac_82599EB) {
4583 		gpie |= IXGBE_SDP1_GPIEN;
4584 		gpie |= IXGBE_SDP2_GPIEN;
4585 	}
4586 
4587 	IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
4588 }
4589 
4590 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
4591 {
4592 	struct ixgbe_hw *hw = &adapter->hw;
4593 	struct net_device *upper;
4594 	struct list_head *iter;
4595 	int err;
4596 	u32 ctrl_ext;
4597 
4598 	ixgbe_get_hw_control(adapter);
4599 	ixgbe_setup_gpie(adapter);
4600 
4601 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4602 		ixgbe_configure_msix(adapter);
4603 	else
4604 		ixgbe_configure_msi_and_legacy(adapter);
4605 
4606 	/* enable the optics for 82599 SFP+ fiber */
4607 	if (hw->mac.ops.enable_tx_laser)
4608 		hw->mac.ops.enable_tx_laser(hw);
4609 
4610 	smp_mb__before_clear_bit();
4611 	clear_bit(__IXGBE_DOWN, &adapter->state);
4612 	ixgbe_napi_enable_all(adapter);
4613 
4614 	if (ixgbe_is_sfp(hw)) {
4615 		ixgbe_sfp_link_config(adapter);
4616 	} else {
4617 		err = ixgbe_non_sfp_link_config(hw);
4618 		if (err)
4619 			e_err(probe, "link_config FAILED %d\n", err);
4620 	}
4621 
4622 	/* clear any pending interrupts, may auto mask */
4623 	IXGBE_READ_REG(hw, IXGBE_EICR);
4624 	ixgbe_irq_enable(adapter, true, true);
4625 
4626 	/*
4627 	 * If this adapter has a fan, check to see if we had a failure
4628 	 * before we enabled the interrupt.
4629 	 */
4630 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
4631 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
4632 		if (esdp & IXGBE_ESDP_SDP1)
4633 			e_crit(drv, "Fan has stopped, replace the adapter\n");
4634 	}
4635 
4636 	/* enable transmits */
4637 	netif_tx_start_all_queues(adapter->netdev);
4638 
4639 	/* enable any upper devices */
4640 	netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
4641 		if (netif_is_macvlan(upper)) {
4642 			struct macvlan_dev *vlan = netdev_priv(upper);
4643 
4644 			if (vlan->fwd_priv)
4645 				netif_tx_start_all_queues(upper);
4646 		}
4647 	}
4648 
4649 	/* bring the link up in the watchdog, this could race with our first
4650 	 * link up interrupt but shouldn't be a problem */
4651 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4652 	adapter->link_check_timeout = jiffies;
4653 	mod_timer(&adapter->service_timer, jiffies);
4654 
4655 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
4656 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
4657 	ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
4658 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
4659 }
4660 
4661 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
4662 {
4663 	WARN_ON(in_interrupt());
4664 	/* put off any impending NetWatchDogTimeout */
4665 	adapter->netdev->trans_start = jiffies;
4666 
4667 	while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
4668 		usleep_range(1000, 2000);
4669 	ixgbe_down(adapter);
4670 	/*
4671 	 * If SR-IOV enabled then wait a bit before bringing the adapter
4672 	 * back up to give the VFs time to respond to the reset.  The
4673 	 * two second wait is based upon the watchdog timer cycle in
4674 	 * the VF driver.
4675 	 */
4676 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4677 		msleep(2000);
4678 	ixgbe_up(adapter);
4679 	clear_bit(__IXGBE_RESETTING, &adapter->state);
4680 }
4681 
4682 void ixgbe_up(struct ixgbe_adapter *adapter)
4683 {
4684 	/* hardware has been reset, we need to reload some things */
4685 	ixgbe_configure(adapter);
4686 
4687 	ixgbe_up_complete(adapter);
4688 }
4689 
4690 void ixgbe_reset(struct ixgbe_adapter *adapter)
4691 {
4692 	struct ixgbe_hw *hw = &adapter->hw;
4693 	int err;
4694 
4695 	if (ixgbe_removed(hw->hw_addr))
4696 		return;
4697 	/* lock SFP init bit to prevent race conditions with the watchdog */
4698 	while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
4699 		usleep_range(1000, 2000);
4700 
4701 	/* clear all SFP and link config related flags while holding SFP_INIT */
4702 	adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
4703 			     IXGBE_FLAG2_SFP_NEEDS_RESET);
4704 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
4705 
4706 	err = hw->mac.ops.init_hw(hw);
4707 	switch (err) {
4708 	case 0:
4709 	case IXGBE_ERR_SFP_NOT_PRESENT:
4710 	case IXGBE_ERR_SFP_NOT_SUPPORTED:
4711 		break;
4712 	case IXGBE_ERR_MASTER_REQUESTS_PENDING:
4713 		e_dev_err("master disable timed out\n");
4714 		break;
4715 	case IXGBE_ERR_EEPROM_VERSION:
4716 		/* We are running on a pre-production device, log a warning */
4717 		e_dev_warn("This device is a pre-production adapter/LOM. "
4718 			   "Please be aware there may be issues associated with "
4719 			   "your hardware.  If you are experiencing problems "
4720 			   "please contact your Intel or hardware "
4721 			   "representative who provided you with this "
4722 			   "hardware.\n");
4723 		break;
4724 	default:
4725 		e_dev_err("Hardware Error: %d\n", err);
4726 	}
4727 
4728 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
4729 
4730 	/* reprogram the RAR[0] in case user changed it. */
4731 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, VMDQ_P(0), IXGBE_RAH_AV);
4732 
4733 	/* update SAN MAC vmdq pool selection */
4734 	if (hw->mac.san_mac_rar_index)
4735 		hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
4736 
4737 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
4738 		ixgbe_ptp_reset(adapter);
4739 }
4740 
4741 /**
4742  * ixgbe_clean_tx_ring - Free Tx Buffers
4743  * @tx_ring: ring to be cleaned
4744  **/
4745 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4746 {
4747 	struct ixgbe_tx_buffer *tx_buffer_info;
4748 	unsigned long size;
4749 	u16 i;
4750 
4751 	/* ring already cleared, nothing to do */
4752 	if (!tx_ring->tx_buffer_info)
4753 		return;
4754 
4755 	/* Free all the Tx ring sk_buffs */
4756 	for (i = 0; i < tx_ring->count; i++) {
4757 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
4758 		ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4759 	}
4760 
4761 	netdev_tx_reset_queue(txring_txq(tx_ring));
4762 
4763 	size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4764 	memset(tx_ring->tx_buffer_info, 0, size);
4765 
4766 	/* Zero out the descriptor ring */
4767 	memset(tx_ring->desc, 0, tx_ring->size);
4768 
4769 	tx_ring->next_to_use = 0;
4770 	tx_ring->next_to_clean = 0;
4771 }
4772 
4773 /**
4774  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4775  * @adapter: board private structure
4776  **/
4777 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4778 {
4779 	int i;
4780 
4781 	for (i = 0; i < adapter->num_rx_queues; i++)
4782 		ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4783 }
4784 
4785 /**
4786  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4787  * @adapter: board private structure
4788  **/
4789 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4790 {
4791 	int i;
4792 
4793 	for (i = 0; i < adapter->num_tx_queues; i++)
4794 		ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4795 }
4796 
4797 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
4798 {
4799 	struct hlist_node *node2;
4800 	struct ixgbe_fdir_filter *filter;
4801 
4802 	spin_lock(&adapter->fdir_perfect_lock);
4803 
4804 	hlist_for_each_entry_safe(filter, node2,
4805 				  &adapter->fdir_filter_list, fdir_node) {
4806 		hlist_del(&filter->fdir_node);
4807 		kfree(filter);
4808 	}
4809 	adapter->fdir_filter_count = 0;
4810 
4811 	spin_unlock(&adapter->fdir_perfect_lock);
4812 }
4813 
4814 void ixgbe_down(struct ixgbe_adapter *adapter)
4815 {
4816 	struct net_device *netdev = adapter->netdev;
4817 	struct ixgbe_hw *hw = &adapter->hw;
4818 	struct net_device *upper;
4819 	struct list_head *iter;
4820 	u32 rxctrl;
4821 	int i;
4822 
4823 	/* signal that we are down to the interrupt handler */
4824 	if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
4825 		return; /* do nothing if already down */
4826 
4827 	/* disable receives */
4828 	rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4829 	IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4830 
4831 	/* disable all enabled rx queues */
4832 	for (i = 0; i < adapter->num_rx_queues; i++)
4833 		/* this call also flushes the previous write */
4834 		ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4835 
4836 	usleep_range(10000, 20000);
4837 
4838 	netif_tx_stop_all_queues(netdev);
4839 
4840 	/* call carrier off first to avoid false dev_watchdog timeouts */
4841 	netif_carrier_off(netdev);
4842 	netif_tx_disable(netdev);
4843 
4844 	/* disable any upper devices */
4845 	netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
4846 		if (netif_is_macvlan(upper)) {
4847 			struct macvlan_dev *vlan = netdev_priv(upper);
4848 
4849 			if (vlan->fwd_priv) {
4850 				netif_tx_stop_all_queues(upper);
4851 				netif_carrier_off(upper);
4852 				netif_tx_disable(upper);
4853 			}
4854 		}
4855 	}
4856 
4857 	ixgbe_irq_disable(adapter);
4858 
4859 	ixgbe_napi_disable_all(adapter);
4860 
4861 	adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
4862 			     IXGBE_FLAG2_RESET_REQUESTED);
4863 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4864 
4865 	del_timer_sync(&adapter->service_timer);
4866 
4867 	if (adapter->num_vfs) {
4868 		/* Clear EITR Select mapping */
4869 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
4870 
4871 		/* Mark all the VFs as inactive */
4872 		for (i = 0 ; i < adapter->num_vfs; i++)
4873 			adapter->vfinfo[i].clear_to_send = false;
4874 
4875 		/* ping all the active vfs to let them know we are going down */
4876 		ixgbe_ping_all_vfs(adapter);
4877 
4878 		/* Disable all VFTE/VFRE TX/RX */
4879 		ixgbe_disable_tx_rx(adapter);
4880 	}
4881 
4882 	/* disable transmits in the hardware now that interrupts are off */
4883 	for (i = 0; i < adapter->num_tx_queues; i++) {
4884 		u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4885 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
4886 	}
4887 
4888 	/* Disable the Tx DMA engine on 82599 and X540 */
4889 	switch (hw->mac.type) {
4890 	case ixgbe_mac_82599EB:
4891 	case ixgbe_mac_X540:
4892 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4893 				(IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4894 				 ~IXGBE_DMATXCTL_TE));
4895 		break;
4896 	default:
4897 		break;
4898 	}
4899 
4900 	if (!pci_channel_offline(adapter->pdev))
4901 		ixgbe_reset(adapter);
4902 
4903 	/* power down the optics for 82599 SFP+ fiber */
4904 	if (hw->mac.ops.disable_tx_laser)
4905 		hw->mac.ops.disable_tx_laser(hw);
4906 
4907 	ixgbe_clean_all_tx_rings(adapter);
4908 	ixgbe_clean_all_rx_rings(adapter);
4909 
4910 #ifdef CONFIG_IXGBE_DCA
4911 	/* since we reset the hardware DCA settings were cleared */
4912 	ixgbe_setup_dca(adapter);
4913 #endif
4914 }
4915 
4916 /**
4917  * ixgbe_tx_timeout - Respond to a Tx Hang
4918  * @netdev: network interface device structure
4919  **/
4920 static void ixgbe_tx_timeout(struct net_device *netdev)
4921 {
4922 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4923 
4924 	/* Do the reset outside of interrupt context */
4925 	ixgbe_tx_timeout_reset(adapter);
4926 }
4927 
4928 /**
4929  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4930  * @adapter: board private structure to initialize
4931  *
4932  * ixgbe_sw_init initializes the Adapter private data structure.
4933  * Fields are initialized based on PCI device information and
4934  * OS network device settings (MTU size).
4935  **/
4936 static int ixgbe_sw_init(struct ixgbe_adapter *adapter)
4937 {
4938 	struct ixgbe_hw *hw = &adapter->hw;
4939 	struct pci_dev *pdev = adapter->pdev;
4940 	unsigned int rss, fdir;
4941 	u32 fwsm;
4942 #ifdef CONFIG_IXGBE_DCB
4943 	int j;
4944 	struct tc_configuration *tc;
4945 #endif
4946 
4947 	/* PCI config space info */
4948 
4949 	hw->vendor_id = pdev->vendor;
4950 	hw->device_id = pdev->device;
4951 	hw->revision_id = pdev->revision;
4952 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
4953 	hw->subsystem_device_id = pdev->subsystem_device;
4954 
4955 	/* Set common capability flags and settings */
4956 	rss = min_t(int, IXGBE_MAX_RSS_INDICES, num_online_cpus());
4957 	adapter->ring_feature[RING_F_RSS].limit = rss;
4958 	adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4959 	adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4960 	adapter->max_q_vectors = MAX_Q_VECTORS_82599;
4961 	adapter->atr_sample_rate = 20;
4962 	fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
4963 	adapter->ring_feature[RING_F_FDIR].limit = fdir;
4964 	adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
4965 #ifdef CONFIG_IXGBE_DCA
4966 	adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
4967 #endif
4968 #ifdef IXGBE_FCOE
4969 	adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4970 	adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4971 #ifdef CONFIG_IXGBE_DCB
4972 	/* Default traffic class to use for FCoE */
4973 	adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4974 #endif /* CONFIG_IXGBE_DCB */
4975 #endif /* IXGBE_FCOE */
4976 
4977 	/* Set MAC specific capability flags and exceptions */
4978 	switch (hw->mac.type) {
4979 	case ixgbe_mac_82598EB:
4980 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
4981 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
4982 
4983 		if (hw->device_id == IXGBE_DEV_ID_82598AT)
4984 			adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4985 
4986 		adapter->max_q_vectors = MAX_Q_VECTORS_82598;
4987 		adapter->ring_feature[RING_F_FDIR].limit = 0;
4988 		adapter->atr_sample_rate = 0;
4989 		adapter->fdir_pballoc = 0;
4990 #ifdef IXGBE_FCOE
4991 		adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
4992 		adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4993 #ifdef CONFIG_IXGBE_DCB
4994 		adapter->fcoe.up = 0;
4995 #endif /* IXGBE_DCB */
4996 #endif /* IXGBE_FCOE */
4997 		break;
4998 	case ixgbe_mac_82599EB:
4999 		if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
5000 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5001 		break;
5002 	case ixgbe_mac_X540:
5003 		fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM);
5004 		if (fwsm & IXGBE_FWSM_TS_ENABLED)
5005 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5006 		break;
5007 	default:
5008 		break;
5009 	}
5010 
5011 #ifdef IXGBE_FCOE
5012 	/* FCoE support exists, always init the FCoE lock */
5013 	spin_lock_init(&adapter->fcoe.lock);
5014 
5015 #endif
5016 	/* n-tuple support exists, always init our spinlock */
5017 	spin_lock_init(&adapter->fdir_perfect_lock);
5018 
5019 #ifdef CONFIG_IXGBE_DCB
5020 	switch (hw->mac.type) {
5021 	case ixgbe_mac_X540:
5022 		adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
5023 		adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
5024 		break;
5025 	default:
5026 		adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
5027 		adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
5028 		break;
5029 	}
5030 
5031 	/* Configure DCB traffic classes */
5032 	for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5033 		tc = &adapter->dcb_cfg.tc_config[j];
5034 		tc->path[DCB_TX_CONFIG].bwg_id = 0;
5035 		tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5036 		tc->path[DCB_RX_CONFIG].bwg_id = 0;
5037 		tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5038 		tc->dcb_pfc = pfc_disabled;
5039 	}
5040 
5041 	/* Initialize default user to priority mapping, UPx->TC0 */
5042 	tc = &adapter->dcb_cfg.tc_config[0];
5043 	tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
5044 	tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
5045 
5046 	adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5047 	adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5048 	adapter->dcb_cfg.pfc_mode_enable = false;
5049 	adapter->dcb_set_bitmap = 0x00;
5050 	adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
5051 	memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
5052 	       sizeof(adapter->temp_dcb_cfg));
5053 
5054 #endif
5055 
5056 	/* default flow control settings */
5057 	hw->fc.requested_mode = ixgbe_fc_full;
5058 	hw->fc.current_mode = ixgbe_fc_full;	/* init for ethtool output */
5059 	ixgbe_pbthresh_setup(adapter);
5060 	hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5061 	hw->fc.send_xon = true;
5062 	hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
5063 
5064 #ifdef CONFIG_PCI_IOV
5065 	if (max_vfs > 0)
5066 		e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
5067 
5068 	/* assign number of SR-IOV VFs */
5069 	if (hw->mac.type != ixgbe_mac_82598EB) {
5070 		if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
5071 			adapter->num_vfs = 0;
5072 			e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
5073 		} else {
5074 			adapter->num_vfs = max_vfs;
5075 		}
5076 	}
5077 #endif /* CONFIG_PCI_IOV */
5078 
5079 	/* enable itr by default in dynamic mode */
5080 	adapter->rx_itr_setting = 1;
5081 	adapter->tx_itr_setting = 1;
5082 
5083 	/* set default ring sizes */
5084 	adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5085 	adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5086 
5087 	/* set default work limits */
5088 	adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
5089 
5090 	/* initialize eeprom parameters */
5091 	if (ixgbe_init_eeprom_params_generic(hw)) {
5092 		e_dev_err("EEPROM initialization failed\n");
5093 		return -EIO;
5094 	}
5095 
5096 	/* PF holds first pool slot */
5097 	set_bit(0, &adapter->fwd_bitmask);
5098 	set_bit(__IXGBE_DOWN, &adapter->state);
5099 
5100 	return 0;
5101 }
5102 
5103 /**
5104  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5105  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
5106  *
5107  * Return 0 on success, negative on failure
5108  **/
5109 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5110 {
5111 	struct device *dev = tx_ring->dev;
5112 	int orig_node = dev_to_node(dev);
5113 	int numa_node = -1;
5114 	int size;
5115 
5116 	size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5117 
5118 	if (tx_ring->q_vector)
5119 		numa_node = tx_ring->q_vector->numa_node;
5120 
5121 	tx_ring->tx_buffer_info = vzalloc_node(size, numa_node);
5122 	if (!tx_ring->tx_buffer_info)
5123 		tx_ring->tx_buffer_info = vzalloc(size);
5124 	if (!tx_ring->tx_buffer_info)
5125 		goto err;
5126 
5127 	u64_stats_init(&tx_ring->syncp);
5128 
5129 	/* round up to nearest 4K */
5130 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5131 	tx_ring->size = ALIGN(tx_ring->size, 4096);
5132 
5133 	set_dev_node(dev, numa_node);
5134 	tx_ring->desc = dma_alloc_coherent(dev,
5135 					   tx_ring->size,
5136 					   &tx_ring->dma,
5137 					   GFP_KERNEL);
5138 	set_dev_node(dev, orig_node);
5139 	if (!tx_ring->desc)
5140 		tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5141 						   &tx_ring->dma, GFP_KERNEL);
5142 	if (!tx_ring->desc)
5143 		goto err;
5144 
5145 	tx_ring->next_to_use = 0;
5146 	tx_ring->next_to_clean = 0;
5147 	return 0;
5148 
5149 err:
5150 	vfree(tx_ring->tx_buffer_info);
5151 	tx_ring->tx_buffer_info = NULL;
5152 	dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5153 	return -ENOMEM;
5154 }
5155 
5156 /**
5157  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5158  * @adapter: board private structure
5159  *
5160  * If this function returns with an error, then it's possible one or
5161  * more of the rings is populated (while the rest are not).  It is the
5162  * callers duty to clean those orphaned rings.
5163  *
5164  * Return 0 on success, negative on failure
5165  **/
5166 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5167 {
5168 	int i, err = 0;
5169 
5170 	for (i = 0; i < adapter->num_tx_queues; i++) {
5171 		err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5172 		if (!err)
5173 			continue;
5174 
5175 		e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5176 		goto err_setup_tx;
5177 	}
5178 
5179 	return 0;
5180 err_setup_tx:
5181 	/* rewind the index freeing the rings as we go */
5182 	while (i--)
5183 		ixgbe_free_tx_resources(adapter->tx_ring[i]);
5184 	return err;
5185 }
5186 
5187 /**
5188  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5189  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
5190  *
5191  * Returns 0 on success, negative on failure
5192  **/
5193 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5194 {
5195 	struct device *dev = rx_ring->dev;
5196 	int orig_node = dev_to_node(dev);
5197 	int numa_node = -1;
5198 	int size;
5199 
5200 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5201 
5202 	if (rx_ring->q_vector)
5203 		numa_node = rx_ring->q_vector->numa_node;
5204 
5205 	rx_ring->rx_buffer_info = vzalloc_node(size, numa_node);
5206 	if (!rx_ring->rx_buffer_info)
5207 		rx_ring->rx_buffer_info = vzalloc(size);
5208 	if (!rx_ring->rx_buffer_info)
5209 		goto err;
5210 
5211 	u64_stats_init(&rx_ring->syncp);
5212 
5213 	/* Round up to nearest 4K */
5214 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5215 	rx_ring->size = ALIGN(rx_ring->size, 4096);
5216 
5217 	set_dev_node(dev, numa_node);
5218 	rx_ring->desc = dma_alloc_coherent(dev,
5219 					   rx_ring->size,
5220 					   &rx_ring->dma,
5221 					   GFP_KERNEL);
5222 	set_dev_node(dev, orig_node);
5223 	if (!rx_ring->desc)
5224 		rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5225 						   &rx_ring->dma, GFP_KERNEL);
5226 	if (!rx_ring->desc)
5227 		goto err;
5228 
5229 	rx_ring->next_to_clean = 0;
5230 	rx_ring->next_to_use = 0;
5231 
5232 	return 0;
5233 err:
5234 	vfree(rx_ring->rx_buffer_info);
5235 	rx_ring->rx_buffer_info = NULL;
5236 	dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5237 	return -ENOMEM;
5238 }
5239 
5240 /**
5241  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5242  * @adapter: board private structure
5243  *
5244  * If this function returns with an error, then it's possible one or
5245  * more of the rings is populated (while the rest are not).  It is the
5246  * callers duty to clean those orphaned rings.
5247  *
5248  * Return 0 on success, negative on failure
5249  **/
5250 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5251 {
5252 	int i, err = 0;
5253 
5254 	for (i = 0; i < adapter->num_rx_queues; i++) {
5255 		err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5256 		if (!err)
5257 			continue;
5258 
5259 		e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5260 		goto err_setup_rx;
5261 	}
5262 
5263 #ifdef IXGBE_FCOE
5264 	err = ixgbe_setup_fcoe_ddp_resources(adapter);
5265 	if (!err)
5266 #endif
5267 		return 0;
5268 err_setup_rx:
5269 	/* rewind the index freeing the rings as we go */
5270 	while (i--)
5271 		ixgbe_free_rx_resources(adapter->rx_ring[i]);
5272 	return err;
5273 }
5274 
5275 /**
5276  * ixgbe_free_tx_resources - Free Tx Resources per Queue
5277  * @tx_ring: Tx descriptor ring for a specific queue
5278  *
5279  * Free all transmit software resources
5280  **/
5281 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5282 {
5283 	ixgbe_clean_tx_ring(tx_ring);
5284 
5285 	vfree(tx_ring->tx_buffer_info);
5286 	tx_ring->tx_buffer_info = NULL;
5287 
5288 	/* if not set, then don't free */
5289 	if (!tx_ring->desc)
5290 		return;
5291 
5292 	dma_free_coherent(tx_ring->dev, tx_ring->size,
5293 			  tx_ring->desc, tx_ring->dma);
5294 
5295 	tx_ring->desc = NULL;
5296 }
5297 
5298 /**
5299  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5300  * @adapter: board private structure
5301  *
5302  * Free all transmit software resources
5303  **/
5304 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5305 {
5306 	int i;
5307 
5308 	for (i = 0; i < adapter->num_tx_queues; i++)
5309 		if (adapter->tx_ring[i]->desc)
5310 			ixgbe_free_tx_resources(adapter->tx_ring[i]);
5311 }
5312 
5313 /**
5314  * ixgbe_free_rx_resources - Free Rx Resources
5315  * @rx_ring: ring to clean the resources from
5316  *
5317  * Free all receive software resources
5318  **/
5319 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5320 {
5321 	ixgbe_clean_rx_ring(rx_ring);
5322 
5323 	vfree(rx_ring->rx_buffer_info);
5324 	rx_ring->rx_buffer_info = NULL;
5325 
5326 	/* if not set, then don't free */
5327 	if (!rx_ring->desc)
5328 		return;
5329 
5330 	dma_free_coherent(rx_ring->dev, rx_ring->size,
5331 			  rx_ring->desc, rx_ring->dma);
5332 
5333 	rx_ring->desc = NULL;
5334 }
5335 
5336 /**
5337  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5338  * @adapter: board private structure
5339  *
5340  * Free all receive software resources
5341  **/
5342 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5343 {
5344 	int i;
5345 
5346 #ifdef IXGBE_FCOE
5347 	ixgbe_free_fcoe_ddp_resources(adapter);
5348 
5349 #endif
5350 	for (i = 0; i < adapter->num_rx_queues; i++)
5351 		if (adapter->rx_ring[i]->desc)
5352 			ixgbe_free_rx_resources(adapter->rx_ring[i]);
5353 }
5354 
5355 /**
5356  * ixgbe_change_mtu - Change the Maximum Transfer Unit
5357  * @netdev: network interface device structure
5358  * @new_mtu: new value for maximum frame size
5359  *
5360  * Returns 0 on success, negative on failure
5361  **/
5362 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5363 {
5364 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
5365 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5366 
5367 	/* MTU < 68 is an error and causes problems on some kernels */
5368 	if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5369 		return -EINVAL;
5370 
5371 	/*
5372 	 * For 82599EB we cannot allow legacy VFs to enable their receive
5373 	 * paths when MTU greater than 1500 is configured.  So display a
5374 	 * warning that legacy VFs will be disabled.
5375 	 */
5376 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
5377 	    (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
5378 	    (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
5379 		e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
5380 
5381 	e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5382 
5383 	/* must set new MTU before calling down or up */
5384 	netdev->mtu = new_mtu;
5385 
5386 	if (netif_running(netdev))
5387 		ixgbe_reinit_locked(adapter);
5388 
5389 	return 0;
5390 }
5391 
5392 /**
5393  * ixgbe_open - Called when a network interface is made active
5394  * @netdev: network interface device structure
5395  *
5396  * Returns 0 on success, negative value on failure
5397  *
5398  * The open entry point is called when a network interface is made
5399  * active by the system (IFF_UP).  At this point all resources needed
5400  * for transmit and receive operations are allocated, the interrupt
5401  * handler is registered with the OS, the watchdog timer is started,
5402  * and the stack is notified that the interface is ready.
5403  **/
5404 static int ixgbe_open(struct net_device *netdev)
5405 {
5406 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
5407 	int err, queues;
5408 
5409 	/* disallow open during test */
5410 	if (test_bit(__IXGBE_TESTING, &adapter->state))
5411 		return -EBUSY;
5412 
5413 	netif_carrier_off(netdev);
5414 
5415 	/* allocate transmit descriptors */
5416 	err = ixgbe_setup_all_tx_resources(adapter);
5417 	if (err)
5418 		goto err_setup_tx;
5419 
5420 	/* allocate receive descriptors */
5421 	err = ixgbe_setup_all_rx_resources(adapter);
5422 	if (err)
5423 		goto err_setup_rx;
5424 
5425 	ixgbe_configure(adapter);
5426 
5427 	err = ixgbe_request_irq(adapter);
5428 	if (err)
5429 		goto err_req_irq;
5430 
5431 	/* Notify the stack of the actual queue counts. */
5432 	if (adapter->num_rx_pools > 1)
5433 		queues = adapter->num_rx_queues_per_pool;
5434 	else
5435 		queues = adapter->num_tx_queues;
5436 
5437 	err = netif_set_real_num_tx_queues(netdev, queues);
5438 	if (err)
5439 		goto err_set_queues;
5440 
5441 	if (adapter->num_rx_pools > 1 &&
5442 	    adapter->num_rx_queues > IXGBE_MAX_L2A_QUEUES)
5443 		queues = IXGBE_MAX_L2A_QUEUES;
5444 	else
5445 		queues = adapter->num_rx_queues;
5446 	err = netif_set_real_num_rx_queues(netdev, queues);
5447 	if (err)
5448 		goto err_set_queues;
5449 
5450 	ixgbe_ptp_init(adapter);
5451 
5452 	ixgbe_up_complete(adapter);
5453 
5454 	return 0;
5455 
5456 err_set_queues:
5457 	ixgbe_free_irq(adapter);
5458 err_req_irq:
5459 	ixgbe_free_all_rx_resources(adapter);
5460 err_setup_rx:
5461 	ixgbe_free_all_tx_resources(adapter);
5462 err_setup_tx:
5463 	ixgbe_reset(adapter);
5464 
5465 	return err;
5466 }
5467 
5468 /**
5469  * ixgbe_close - Disables a network interface
5470  * @netdev: network interface device structure
5471  *
5472  * Returns 0, this is not allowed to fail
5473  *
5474  * The close entry point is called when an interface is de-activated
5475  * by the OS.  The hardware is still under the drivers control, but
5476  * needs to be disabled.  A global MAC reset is issued to stop the
5477  * hardware, and all transmit and receive resources are freed.
5478  **/
5479 static int ixgbe_close(struct net_device *netdev)
5480 {
5481 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
5482 
5483 	ixgbe_ptp_stop(adapter);
5484 
5485 	ixgbe_down(adapter);
5486 	ixgbe_free_irq(adapter);
5487 
5488 	ixgbe_fdir_filter_exit(adapter);
5489 
5490 	ixgbe_free_all_tx_resources(adapter);
5491 	ixgbe_free_all_rx_resources(adapter);
5492 
5493 	ixgbe_release_hw_control(adapter);
5494 
5495 	return 0;
5496 }
5497 
5498 #ifdef CONFIG_PM
5499 static int ixgbe_resume(struct pci_dev *pdev)
5500 {
5501 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5502 	struct net_device *netdev = adapter->netdev;
5503 	u32 err;
5504 
5505 	pci_set_power_state(pdev, PCI_D0);
5506 	pci_restore_state(pdev);
5507 	/*
5508 	 * pci_restore_state clears dev->state_saved so call
5509 	 * pci_save_state to restore it.
5510 	 */
5511 	pci_save_state(pdev);
5512 
5513 	err = pci_enable_device_mem(pdev);
5514 	if (err) {
5515 		e_dev_err("Cannot enable PCI device from suspend\n");
5516 		return err;
5517 	}
5518 	pci_set_master(pdev);
5519 
5520 	pci_wake_from_d3(pdev, false);
5521 
5522 	ixgbe_reset(adapter);
5523 
5524 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5525 
5526 	rtnl_lock();
5527 	err = ixgbe_init_interrupt_scheme(adapter);
5528 	if (!err && netif_running(netdev))
5529 		err = ixgbe_open(netdev);
5530 
5531 	rtnl_unlock();
5532 
5533 	if (err)
5534 		return err;
5535 
5536 	netif_device_attach(netdev);
5537 
5538 	return 0;
5539 }
5540 #endif /* CONFIG_PM */
5541 
5542 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5543 {
5544 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5545 	struct net_device *netdev = adapter->netdev;
5546 	struct ixgbe_hw *hw = &adapter->hw;
5547 	u32 ctrl, fctrl;
5548 	u32 wufc = adapter->wol;
5549 #ifdef CONFIG_PM
5550 	int retval = 0;
5551 #endif
5552 
5553 	netif_device_detach(netdev);
5554 
5555 	rtnl_lock();
5556 	if (netif_running(netdev)) {
5557 		ixgbe_down(adapter);
5558 		ixgbe_free_irq(adapter);
5559 		ixgbe_free_all_tx_resources(adapter);
5560 		ixgbe_free_all_rx_resources(adapter);
5561 	}
5562 	rtnl_unlock();
5563 
5564 	ixgbe_clear_interrupt_scheme(adapter);
5565 
5566 #ifdef CONFIG_PM
5567 	retval = pci_save_state(pdev);
5568 	if (retval)
5569 		return retval;
5570 
5571 #endif
5572 	if (hw->mac.ops.stop_link_on_d3)
5573 		hw->mac.ops.stop_link_on_d3(hw);
5574 
5575 	if (wufc) {
5576 		ixgbe_set_rx_mode(netdev);
5577 
5578 		/* enable the optics for 82599 SFP+ fiber as we can WoL */
5579 		if (hw->mac.ops.enable_tx_laser)
5580 			hw->mac.ops.enable_tx_laser(hw);
5581 
5582 		/* turn on all-multi mode if wake on multicast is enabled */
5583 		if (wufc & IXGBE_WUFC_MC) {
5584 			fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5585 			fctrl |= IXGBE_FCTRL_MPE;
5586 			IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5587 		}
5588 
5589 		ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5590 		ctrl |= IXGBE_CTRL_GIO_DIS;
5591 		IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5592 
5593 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5594 	} else {
5595 		IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5596 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5597 	}
5598 
5599 	switch (hw->mac.type) {
5600 	case ixgbe_mac_82598EB:
5601 		pci_wake_from_d3(pdev, false);
5602 		break;
5603 	case ixgbe_mac_82599EB:
5604 	case ixgbe_mac_X540:
5605 		pci_wake_from_d3(pdev, !!wufc);
5606 		break;
5607 	default:
5608 		break;
5609 	}
5610 
5611 	*enable_wake = !!wufc;
5612 
5613 	ixgbe_release_hw_control(adapter);
5614 
5615 	pci_disable_device(pdev);
5616 
5617 	return 0;
5618 }
5619 
5620 #ifdef CONFIG_PM
5621 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5622 {
5623 	int retval;
5624 	bool wake;
5625 
5626 	retval = __ixgbe_shutdown(pdev, &wake);
5627 	if (retval)
5628 		return retval;
5629 
5630 	if (wake) {
5631 		pci_prepare_to_sleep(pdev);
5632 	} else {
5633 		pci_wake_from_d3(pdev, false);
5634 		pci_set_power_state(pdev, PCI_D3hot);
5635 	}
5636 
5637 	return 0;
5638 }
5639 #endif /* CONFIG_PM */
5640 
5641 static void ixgbe_shutdown(struct pci_dev *pdev)
5642 {
5643 	bool wake;
5644 
5645 	__ixgbe_shutdown(pdev, &wake);
5646 
5647 	if (system_state == SYSTEM_POWER_OFF) {
5648 		pci_wake_from_d3(pdev, wake);
5649 		pci_set_power_state(pdev, PCI_D3hot);
5650 	}
5651 }
5652 
5653 /**
5654  * ixgbe_update_stats - Update the board statistics counters.
5655  * @adapter: board private structure
5656  **/
5657 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5658 {
5659 	struct net_device *netdev = adapter->netdev;
5660 	struct ixgbe_hw *hw = &adapter->hw;
5661 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
5662 	u64 total_mpc = 0;
5663 	u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5664 	u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5665 	u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5666 	u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
5667 
5668 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5669 	    test_bit(__IXGBE_RESETTING, &adapter->state))
5670 		return;
5671 
5672 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5673 		u64 rsc_count = 0;
5674 		u64 rsc_flush = 0;
5675 		for (i = 0; i < adapter->num_rx_queues; i++) {
5676 			rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5677 			rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5678 		}
5679 		adapter->rsc_total_count = rsc_count;
5680 		adapter->rsc_total_flush = rsc_flush;
5681 	}
5682 
5683 	for (i = 0; i < adapter->num_rx_queues; i++) {
5684 		struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5685 		non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5686 		alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5687 		alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5688 		hw_csum_rx_error += rx_ring->rx_stats.csum_err;
5689 		bytes += rx_ring->stats.bytes;
5690 		packets += rx_ring->stats.packets;
5691 	}
5692 	adapter->non_eop_descs = non_eop_descs;
5693 	adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5694 	adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5695 	adapter->hw_csum_rx_error = hw_csum_rx_error;
5696 	netdev->stats.rx_bytes = bytes;
5697 	netdev->stats.rx_packets = packets;
5698 
5699 	bytes = 0;
5700 	packets = 0;
5701 	/* gather some stats to the adapter struct that are per queue */
5702 	for (i = 0; i < adapter->num_tx_queues; i++) {
5703 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5704 		restart_queue += tx_ring->tx_stats.restart_queue;
5705 		tx_busy += tx_ring->tx_stats.tx_busy;
5706 		bytes += tx_ring->stats.bytes;
5707 		packets += tx_ring->stats.packets;
5708 	}
5709 	adapter->restart_queue = restart_queue;
5710 	adapter->tx_busy = tx_busy;
5711 	netdev->stats.tx_bytes = bytes;
5712 	netdev->stats.tx_packets = packets;
5713 
5714 	hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5715 
5716 	/* 8 register reads */
5717 	for (i = 0; i < 8; i++) {
5718 		/* for packet buffers not used, the register should read 0 */
5719 		mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5720 		missed_rx += mpc;
5721 		hwstats->mpc[i] += mpc;
5722 		total_mpc += hwstats->mpc[i];
5723 		hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5724 		hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5725 		switch (hw->mac.type) {
5726 		case ixgbe_mac_82598EB:
5727 			hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5728 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5729 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5730 			hwstats->pxonrxc[i] +=
5731 				IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5732 			break;
5733 		case ixgbe_mac_82599EB:
5734 		case ixgbe_mac_X540:
5735 			hwstats->pxonrxc[i] +=
5736 				IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5737 			break;
5738 		default:
5739 			break;
5740 		}
5741 	}
5742 
5743 	/*16 register reads */
5744 	for (i = 0; i < 16; i++) {
5745 		hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5746 		hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5747 		if ((hw->mac.type == ixgbe_mac_82599EB) ||
5748 		    (hw->mac.type == ixgbe_mac_X540)) {
5749 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
5750 			IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
5751 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
5752 			IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
5753 		}
5754 	}
5755 
5756 	hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5757 	/* work around hardware counting issue */
5758 	hwstats->gprc -= missed_rx;
5759 
5760 	ixgbe_update_xoff_received(adapter);
5761 
5762 	/* 82598 hardware only has a 32 bit counter in the high register */
5763 	switch (hw->mac.type) {
5764 	case ixgbe_mac_82598EB:
5765 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5766 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5767 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5768 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5769 		break;
5770 	case ixgbe_mac_X540:
5771 		/* OS2BMC stats are X540 only*/
5772 		hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5773 		hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5774 		hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5775 		hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5776 	case ixgbe_mac_82599EB:
5777 		for (i = 0; i < 16; i++)
5778 			adapter->hw_rx_no_dma_resources +=
5779 					     IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5780 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5781 		IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5782 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5783 		IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5784 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5785 		IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5786 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5787 		hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5788 		hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5789 #ifdef IXGBE_FCOE
5790 		hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5791 		hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5792 		hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5793 		hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5794 		hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5795 		hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5796 		/* Add up per cpu counters for total ddp aloc fail */
5797 		if (adapter->fcoe.ddp_pool) {
5798 			struct ixgbe_fcoe *fcoe = &adapter->fcoe;
5799 			struct ixgbe_fcoe_ddp_pool *ddp_pool;
5800 			unsigned int cpu;
5801 			u64 noddp = 0, noddp_ext_buff = 0;
5802 			for_each_possible_cpu(cpu) {
5803 				ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
5804 				noddp += ddp_pool->noddp;
5805 				noddp_ext_buff += ddp_pool->noddp_ext_buff;
5806 			}
5807 			hwstats->fcoe_noddp = noddp;
5808 			hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
5809 		}
5810 #endif /* IXGBE_FCOE */
5811 		break;
5812 	default:
5813 		break;
5814 	}
5815 	bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5816 	hwstats->bprc += bprc;
5817 	hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5818 	if (hw->mac.type == ixgbe_mac_82598EB)
5819 		hwstats->mprc -= bprc;
5820 	hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5821 	hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5822 	hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5823 	hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5824 	hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5825 	hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5826 	hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5827 	hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5828 	lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5829 	hwstats->lxontxc += lxon;
5830 	lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5831 	hwstats->lxofftxc += lxoff;
5832 	hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5833 	hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5834 	/*
5835 	 * 82598 errata - tx of flow control packets is included in tx counters
5836 	 */
5837 	xon_off_tot = lxon + lxoff;
5838 	hwstats->gptc -= xon_off_tot;
5839 	hwstats->mptc -= xon_off_tot;
5840 	hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5841 	hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5842 	hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5843 	hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5844 	hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5845 	hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5846 	hwstats->ptc64 -= xon_off_tot;
5847 	hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5848 	hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5849 	hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5850 	hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5851 	hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5852 	hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5853 
5854 	/* Fill out the OS statistics structure */
5855 	netdev->stats.multicast = hwstats->mprc;
5856 
5857 	/* Rx Errors */
5858 	netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5859 	netdev->stats.rx_dropped = 0;
5860 	netdev->stats.rx_length_errors = hwstats->rlec;
5861 	netdev->stats.rx_crc_errors = hwstats->crcerrs;
5862 	netdev->stats.rx_missed_errors = total_mpc;
5863 }
5864 
5865 /**
5866  * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5867  * @adapter: pointer to the device adapter structure
5868  **/
5869 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5870 {
5871 	struct ixgbe_hw *hw = &adapter->hw;
5872 	int i;
5873 
5874 	if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5875 		return;
5876 
5877 	adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5878 
5879 	/* if interface is down do nothing */
5880 	if (test_bit(__IXGBE_DOWN, &adapter->state))
5881 		return;
5882 
5883 	/* do nothing if we are not using signature filters */
5884 	if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5885 		return;
5886 
5887 	adapter->fdir_overflow++;
5888 
5889 	if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5890 		for (i = 0; i < adapter->num_tx_queues; i++)
5891 			set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5892 			        &(adapter->tx_ring[i]->state));
5893 		/* re-enable flow director interrupts */
5894 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5895 	} else {
5896 		e_err(probe, "failed to finish FDIR re-initialization, "
5897 		      "ignored adding FDIR ATR filters\n");
5898 	}
5899 }
5900 
5901 /**
5902  * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5903  * @adapter: pointer to the device adapter structure
5904  *
5905  * This function serves two purposes.  First it strobes the interrupt lines
5906  * in order to make certain interrupts are occurring.  Secondly it sets the
5907  * bits needed to check for TX hangs.  As a result we should immediately
5908  * determine if a hang has occurred.
5909  */
5910 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5911 {
5912 	struct ixgbe_hw *hw = &adapter->hw;
5913 	u64 eics = 0;
5914 	int i;
5915 
5916 	/* If we're down, removing or resetting, just bail */
5917 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5918 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
5919 	    test_bit(__IXGBE_RESETTING, &adapter->state))
5920 		return;
5921 
5922 	/* Force detection of hung controller */
5923 	if (netif_carrier_ok(adapter->netdev)) {
5924 		for (i = 0; i < adapter->num_tx_queues; i++)
5925 			set_check_for_tx_hang(adapter->tx_ring[i]);
5926 	}
5927 
5928 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5929 		/*
5930 		 * for legacy and MSI interrupts don't set any bits
5931 		 * that are enabled for EIAM, because this operation
5932 		 * would set *both* EIMS and EICS for any bit in EIAM
5933 		 */
5934 		IXGBE_WRITE_REG(hw, IXGBE_EICS,
5935 			(IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5936 	} else {
5937 		/* get one bit for every active tx/rx interrupt vector */
5938 		for (i = 0; i < adapter->num_q_vectors; i++) {
5939 			struct ixgbe_q_vector *qv = adapter->q_vector[i];
5940 			if (qv->rx.ring || qv->tx.ring)
5941 				eics |= ((u64)1 << i);
5942 		}
5943 	}
5944 
5945 	/* Cause software interrupt to ensure rings are cleaned */
5946 	ixgbe_irq_rearm_queues(adapter, eics);
5947 
5948 }
5949 
5950 /**
5951  * ixgbe_watchdog_update_link - update the link status
5952  * @adapter: pointer to the device adapter structure
5953  * @link_speed: pointer to a u32 to store the link_speed
5954  **/
5955 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
5956 {
5957 	struct ixgbe_hw *hw = &adapter->hw;
5958 	u32 link_speed = adapter->link_speed;
5959 	bool link_up = adapter->link_up;
5960 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
5961 
5962 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5963 		return;
5964 
5965 	if (hw->mac.ops.check_link) {
5966 		hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5967 	} else {
5968 		/* always assume link is up, if no check link function */
5969 		link_speed = IXGBE_LINK_SPEED_10GB_FULL;
5970 		link_up = true;
5971 	}
5972 
5973 	if (adapter->ixgbe_ieee_pfc)
5974 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
5975 
5976 	if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
5977 		hw->mac.ops.fc_enable(hw);
5978 		ixgbe_set_rx_drop_en(adapter);
5979 	}
5980 
5981 	if (link_up ||
5982 	    time_after(jiffies, (adapter->link_check_timeout +
5983 				 IXGBE_TRY_LINK_TIMEOUT))) {
5984 		adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5985 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5986 		IXGBE_WRITE_FLUSH(hw);
5987 	}
5988 
5989 	adapter->link_up = link_up;
5990 	adapter->link_speed = link_speed;
5991 }
5992 
5993 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
5994 {
5995 #ifdef CONFIG_IXGBE_DCB
5996 	struct net_device *netdev = adapter->netdev;
5997 	struct dcb_app app = {
5998 			      .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
5999 			      .protocol = 0,
6000 			     };
6001 	u8 up = 0;
6002 
6003 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
6004 		up = dcb_ieee_getapp_mask(netdev, &app);
6005 
6006 	adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
6007 #endif
6008 }
6009 
6010 /**
6011  * ixgbe_watchdog_link_is_up - update netif_carrier status and
6012  *                             print link up message
6013  * @adapter: pointer to the device adapter structure
6014  **/
6015 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
6016 {
6017 	struct net_device *netdev = adapter->netdev;
6018 	struct ixgbe_hw *hw = &adapter->hw;
6019 	u32 link_speed = adapter->link_speed;
6020 	bool flow_rx, flow_tx;
6021 
6022 	/* only continue if link was previously down */
6023 	if (netif_carrier_ok(netdev))
6024 		return;
6025 
6026 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
6027 
6028 	switch (hw->mac.type) {
6029 	case ixgbe_mac_82598EB: {
6030 		u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6031 		u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
6032 		flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
6033 		flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
6034 	}
6035 		break;
6036 	case ixgbe_mac_X540:
6037 	case ixgbe_mac_82599EB: {
6038 		u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
6039 		u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
6040 		flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
6041 		flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
6042 	}
6043 		break;
6044 	default:
6045 		flow_tx = false;
6046 		flow_rx = false;
6047 		break;
6048 	}
6049 
6050 	adapter->last_rx_ptp_check = jiffies;
6051 
6052 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
6053 		ixgbe_ptp_start_cyclecounter(adapter);
6054 
6055 	e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
6056 	       (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
6057 	       "10 Gbps" :
6058 	       (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
6059 	       "1 Gbps" :
6060 	       (link_speed == IXGBE_LINK_SPEED_100_FULL ?
6061 	       "100 Mbps" :
6062 	       "unknown speed"))),
6063 	       ((flow_rx && flow_tx) ? "RX/TX" :
6064 	       (flow_rx ? "RX" :
6065 	       (flow_tx ? "TX" : "None"))));
6066 
6067 	netif_carrier_on(netdev);
6068 	ixgbe_check_vf_rate_limit(adapter);
6069 
6070 	/* update the default user priority for VFs */
6071 	ixgbe_update_default_up(adapter);
6072 
6073 	/* ping all the active vfs to let them know link has changed */
6074 	ixgbe_ping_all_vfs(adapter);
6075 }
6076 
6077 /**
6078  * ixgbe_watchdog_link_is_down - update netif_carrier status and
6079  *                               print link down message
6080  * @adapter: pointer to the adapter structure
6081  **/
6082 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
6083 {
6084 	struct net_device *netdev = adapter->netdev;
6085 	struct ixgbe_hw *hw = &adapter->hw;
6086 
6087 	adapter->link_up = false;
6088 	adapter->link_speed = 0;
6089 
6090 	/* only continue if link was up previously */
6091 	if (!netif_carrier_ok(netdev))
6092 		return;
6093 
6094 	/* poll for SFP+ cable when link is down */
6095 	if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
6096 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
6097 
6098 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
6099 		ixgbe_ptp_start_cyclecounter(adapter);
6100 
6101 	e_info(drv, "NIC Link is Down\n");
6102 	netif_carrier_off(netdev);
6103 
6104 	/* ping all the active vfs to let them know link has changed */
6105 	ixgbe_ping_all_vfs(adapter);
6106 }
6107 
6108 /**
6109  * ixgbe_watchdog_flush_tx - flush queues on link down
6110  * @adapter: pointer to the device adapter structure
6111  **/
6112 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
6113 {
6114 	int i;
6115 	int some_tx_pending = 0;
6116 
6117 	if (!netif_carrier_ok(adapter->netdev)) {
6118 		for (i = 0; i < adapter->num_tx_queues; i++) {
6119 			struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
6120 			if (tx_ring->next_to_use != tx_ring->next_to_clean) {
6121 				some_tx_pending = 1;
6122 				break;
6123 			}
6124 		}
6125 
6126 		if (some_tx_pending) {
6127 			/* We've lost link, so the controller stops DMA,
6128 			 * but we've got queued Tx work that's never going
6129 			 * to get done, so reset controller to flush Tx.
6130 			 * (Do the reset outside of interrupt context).
6131 			 */
6132 			e_warn(drv, "initiating reset to clear Tx work after link loss\n");
6133 			adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
6134 		}
6135 	}
6136 }
6137 
6138 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
6139 {
6140 	u32 ssvpc;
6141 
6142 	/* Do not perform spoof check for 82598 or if not in IOV mode */
6143 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
6144 	    adapter->num_vfs == 0)
6145 		return;
6146 
6147 	ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
6148 
6149 	/*
6150 	 * ssvpc register is cleared on read, if zero then no
6151 	 * spoofed packets in the last interval.
6152 	 */
6153 	if (!ssvpc)
6154 		return;
6155 
6156 	e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
6157 }
6158 
6159 /**
6160  * ixgbe_watchdog_subtask - check and bring link up
6161  * @adapter: pointer to the device adapter structure
6162  **/
6163 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
6164 {
6165 	/* if interface is down, removing or resetting, do nothing */
6166 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6167 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
6168 	    test_bit(__IXGBE_RESETTING, &adapter->state))
6169 		return;
6170 
6171 	ixgbe_watchdog_update_link(adapter);
6172 
6173 	if (adapter->link_up)
6174 		ixgbe_watchdog_link_is_up(adapter);
6175 	else
6176 		ixgbe_watchdog_link_is_down(adapter);
6177 
6178 	ixgbe_spoof_check(adapter);
6179 	ixgbe_update_stats(adapter);
6180 
6181 	ixgbe_watchdog_flush_tx(adapter);
6182 }
6183 
6184 /**
6185  * ixgbe_sfp_detection_subtask - poll for SFP+ cable
6186  * @adapter: the ixgbe adapter structure
6187  **/
6188 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
6189 {
6190 	struct ixgbe_hw *hw = &adapter->hw;
6191 	s32 err;
6192 
6193 	/* not searching for SFP so there is nothing to do here */
6194 	if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
6195 	    !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6196 		return;
6197 
6198 	/* someone else is in init, wait until next service event */
6199 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6200 		return;
6201 
6202 	err = hw->phy.ops.identify_sfp(hw);
6203 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6204 		goto sfp_out;
6205 
6206 	if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
6207 		/* If no cable is present, then we need to reset
6208 		 * the next time we find a good cable. */
6209 		adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
6210 	}
6211 
6212 	/* exit on error */
6213 	if (err)
6214 		goto sfp_out;
6215 
6216 	/* exit if reset not needed */
6217 	if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6218 		goto sfp_out;
6219 
6220 	adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
6221 
6222 	/*
6223 	 * A module may be identified correctly, but the EEPROM may not have
6224 	 * support for that module.  setup_sfp() will fail in that case, so
6225 	 * we should not allow that module to load.
6226 	 */
6227 	if (hw->mac.type == ixgbe_mac_82598EB)
6228 		err = hw->phy.ops.reset(hw);
6229 	else
6230 		err = hw->mac.ops.setup_sfp(hw);
6231 
6232 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6233 		goto sfp_out;
6234 
6235 	adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
6236 	e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
6237 
6238 sfp_out:
6239 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6240 
6241 	if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
6242 	    (adapter->netdev->reg_state == NETREG_REGISTERED)) {
6243 		e_dev_err("failed to initialize because an unsupported "
6244 			  "SFP+ module type was detected.\n");
6245 		e_dev_err("Reload the driver after installing a "
6246 			  "supported module.\n");
6247 		unregister_netdev(adapter->netdev);
6248 	}
6249 }
6250 
6251 /**
6252  * ixgbe_sfp_link_config_subtask - set up link SFP after module install
6253  * @adapter: the ixgbe adapter structure
6254  **/
6255 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
6256 {
6257 	struct ixgbe_hw *hw = &adapter->hw;
6258 	u32 speed;
6259 	bool autoneg = false;
6260 
6261 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
6262 		return;
6263 
6264 	/* someone else is in init, wait until next service event */
6265 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6266 		return;
6267 
6268 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
6269 
6270 	speed = hw->phy.autoneg_advertised;
6271 	if ((!speed) && (hw->mac.ops.get_link_capabilities)) {
6272 		hw->mac.ops.get_link_capabilities(hw, &speed, &autoneg);
6273 
6274 		/* setup the highest link when no autoneg */
6275 		if (!autoneg) {
6276 			if (speed & IXGBE_LINK_SPEED_10GB_FULL)
6277 				speed = IXGBE_LINK_SPEED_10GB_FULL;
6278 		}
6279 	}
6280 
6281 	if (hw->mac.ops.setup_link)
6282 		hw->mac.ops.setup_link(hw, speed, true);
6283 
6284 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
6285 	adapter->link_check_timeout = jiffies;
6286 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6287 }
6288 
6289 #ifdef CONFIG_PCI_IOV
6290 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
6291 {
6292 	int vf;
6293 	struct ixgbe_hw *hw = &adapter->hw;
6294 	struct net_device *netdev = adapter->netdev;
6295 	u32 gpc;
6296 	u32 ciaa, ciad;
6297 
6298 	gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
6299 	if (gpc) /* If incrementing then no need for the check below */
6300 		return;
6301 	/*
6302 	 * Check to see if a bad DMA write target from an errant or
6303 	 * malicious VF has caused a PCIe error.  If so then we can
6304 	 * issue a VFLR to the offending VF(s) and then resume without
6305 	 * requesting a full slot reset.
6306 	 */
6307 
6308 	for (vf = 0; vf < adapter->num_vfs; vf++) {
6309 		ciaa = (vf << 16) | 0x80000000;
6310 		/* 32 bit read so align, we really want status at offset 6 */
6311 		ciaa |= PCI_COMMAND;
6312 		IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6313 		ciad = IXGBE_READ_REG(hw, IXGBE_CIAD_82599);
6314 		ciaa &= 0x7FFFFFFF;
6315 		/* disable debug mode asap after reading data */
6316 		IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6317 		/* Get the upper 16 bits which will be the PCI status reg */
6318 		ciad >>= 16;
6319 		if (ciad & PCI_STATUS_REC_MASTER_ABORT) {
6320 			netdev_err(netdev, "VF %d Hung DMA\n", vf);
6321 			/* Issue VFLR */
6322 			ciaa = (vf << 16) | 0x80000000;
6323 			ciaa |= 0xA8;
6324 			IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6325 			ciad = 0x00008000;  /* VFLR */
6326 			IXGBE_WRITE_REG(hw, IXGBE_CIAD_82599, ciad);
6327 			ciaa &= 0x7FFFFFFF;
6328 			IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6329 		}
6330 	}
6331 }
6332 
6333 #endif
6334 /**
6335  * ixgbe_service_timer - Timer Call-back
6336  * @data: pointer to adapter cast into an unsigned long
6337  **/
6338 static void ixgbe_service_timer(unsigned long data)
6339 {
6340 	struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
6341 	unsigned long next_event_offset;
6342 	bool ready = true;
6343 
6344 	/* poll faster when waiting for link */
6345 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
6346 		next_event_offset = HZ / 10;
6347 	else
6348 		next_event_offset = HZ * 2;
6349 
6350 #ifdef CONFIG_PCI_IOV
6351 	/*
6352 	 * don't bother with SR-IOV VF DMA hang check if there are
6353 	 * no VFs or the link is down
6354 	 */
6355 	if (!adapter->num_vfs ||
6356 	    (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
6357 		goto normal_timer_service;
6358 
6359 	/* If we have VFs allocated then we must check for DMA hangs */
6360 	ixgbe_check_for_bad_vf(adapter);
6361 	next_event_offset = HZ / 50;
6362 	adapter->timer_event_accumulator++;
6363 
6364 	if (adapter->timer_event_accumulator >= 100)
6365 		adapter->timer_event_accumulator = 0;
6366 	else
6367 		ready = false;
6368 
6369 normal_timer_service:
6370 #endif
6371 	/* Reset the timer */
6372 	mod_timer(&adapter->service_timer, next_event_offset + jiffies);
6373 
6374 	if (ready)
6375 		ixgbe_service_event_schedule(adapter);
6376 }
6377 
6378 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
6379 {
6380 	if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
6381 		return;
6382 
6383 	adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
6384 
6385 	/* If we're already down, removing or resetting, just bail */
6386 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6387 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
6388 	    test_bit(__IXGBE_RESETTING, &adapter->state))
6389 		return;
6390 
6391 	ixgbe_dump(adapter);
6392 	netdev_err(adapter->netdev, "Reset adapter\n");
6393 	adapter->tx_timeout_count++;
6394 
6395 	rtnl_lock();
6396 	ixgbe_reinit_locked(adapter);
6397 	rtnl_unlock();
6398 }
6399 
6400 /**
6401  * ixgbe_service_task - manages and runs subtasks
6402  * @work: pointer to work_struct containing our data
6403  **/
6404 static void ixgbe_service_task(struct work_struct *work)
6405 {
6406 	struct ixgbe_adapter *adapter = container_of(work,
6407 						     struct ixgbe_adapter,
6408 						     service_task);
6409 	if (ixgbe_removed(adapter->hw.hw_addr)) {
6410 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
6411 			rtnl_lock();
6412 			ixgbe_down(adapter);
6413 			rtnl_unlock();
6414 		}
6415 		ixgbe_service_event_complete(adapter);
6416 		return;
6417 	}
6418 	ixgbe_reset_subtask(adapter);
6419 	ixgbe_sfp_detection_subtask(adapter);
6420 	ixgbe_sfp_link_config_subtask(adapter);
6421 	ixgbe_check_overtemp_subtask(adapter);
6422 	ixgbe_watchdog_subtask(adapter);
6423 	ixgbe_fdir_reinit_subtask(adapter);
6424 	ixgbe_check_hang_subtask(adapter);
6425 
6426 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
6427 		ixgbe_ptp_overflow_check(adapter);
6428 		ixgbe_ptp_rx_hang(adapter);
6429 	}
6430 
6431 	ixgbe_service_event_complete(adapter);
6432 }
6433 
6434 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
6435 		     struct ixgbe_tx_buffer *first,
6436 		     u8 *hdr_len)
6437 {
6438 	struct sk_buff *skb = first->skb;
6439 	u32 vlan_macip_lens, type_tucmd;
6440 	u32 mss_l4len_idx, l4len;
6441 
6442 	if (skb->ip_summed != CHECKSUM_PARTIAL)
6443 		return 0;
6444 
6445 	if (!skb_is_gso(skb))
6446 		return 0;
6447 
6448 	if (skb_header_cloned(skb)) {
6449 		int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
6450 		if (err)
6451 			return err;
6452 	}
6453 
6454 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6455 	type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
6456 
6457 	if (first->protocol == __constant_htons(ETH_P_IP)) {
6458 		struct iphdr *iph = ip_hdr(skb);
6459 		iph->tot_len = 0;
6460 		iph->check = 0;
6461 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6462 							 iph->daddr, 0,
6463 							 IPPROTO_TCP,
6464 							 0);
6465 		type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6466 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
6467 				   IXGBE_TX_FLAGS_CSUM |
6468 				   IXGBE_TX_FLAGS_IPV4;
6469 	} else if (skb_is_gso_v6(skb)) {
6470 		ipv6_hdr(skb)->payload_len = 0;
6471 		tcp_hdr(skb)->check =
6472 		    ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6473 				     &ipv6_hdr(skb)->daddr,
6474 				     0, IPPROTO_TCP, 0);
6475 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
6476 				   IXGBE_TX_FLAGS_CSUM;
6477 	}
6478 
6479 	/* compute header lengths */
6480 	l4len = tcp_hdrlen(skb);
6481 	*hdr_len = skb_transport_offset(skb) + l4len;
6482 
6483 	/* update gso size and bytecount with header size */
6484 	first->gso_segs = skb_shinfo(skb)->gso_segs;
6485 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
6486 
6487 	/* mss_l4len_id: use 0 as index for TSO */
6488 	mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
6489 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
6490 
6491 	/* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
6492 	vlan_macip_lens = skb_network_header_len(skb);
6493 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6494 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6495 
6496 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
6497 			  mss_l4len_idx);
6498 
6499 	return 1;
6500 }
6501 
6502 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
6503 			  struct ixgbe_tx_buffer *first)
6504 {
6505 	struct sk_buff *skb = first->skb;
6506 	u32 vlan_macip_lens = 0;
6507 	u32 mss_l4len_idx = 0;
6508 	u32 type_tucmd = 0;
6509 
6510 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
6511 		if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
6512 		    !(first->tx_flags & IXGBE_TX_FLAGS_CC))
6513 			return;
6514 	} else {
6515 		u8 l4_hdr = 0;
6516 		switch (first->protocol) {
6517 		case __constant_htons(ETH_P_IP):
6518 			vlan_macip_lens |= skb_network_header_len(skb);
6519 			type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6520 			l4_hdr = ip_hdr(skb)->protocol;
6521 			break;
6522 		case __constant_htons(ETH_P_IPV6):
6523 			vlan_macip_lens |= skb_network_header_len(skb);
6524 			l4_hdr = ipv6_hdr(skb)->nexthdr;
6525 			break;
6526 		default:
6527 			if (unlikely(net_ratelimit())) {
6528 				dev_warn(tx_ring->dev,
6529 				 "partial checksum but proto=%x!\n",
6530 				 first->protocol);
6531 			}
6532 			break;
6533 		}
6534 
6535 		switch (l4_hdr) {
6536 		case IPPROTO_TCP:
6537 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6538 			mss_l4len_idx = tcp_hdrlen(skb) <<
6539 					IXGBE_ADVTXD_L4LEN_SHIFT;
6540 			break;
6541 		case IPPROTO_SCTP:
6542 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6543 			mss_l4len_idx = sizeof(struct sctphdr) <<
6544 					IXGBE_ADVTXD_L4LEN_SHIFT;
6545 			break;
6546 		case IPPROTO_UDP:
6547 			mss_l4len_idx = sizeof(struct udphdr) <<
6548 					IXGBE_ADVTXD_L4LEN_SHIFT;
6549 			break;
6550 		default:
6551 			if (unlikely(net_ratelimit())) {
6552 				dev_warn(tx_ring->dev,
6553 				 "partial checksum but l4 proto=%x!\n",
6554 				 l4_hdr);
6555 			}
6556 			break;
6557 		}
6558 
6559 		/* update TX checksum flag */
6560 		first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
6561 	}
6562 
6563 	/* vlan_macip_lens: MACLEN, VLAN tag */
6564 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6565 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6566 
6567 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
6568 			  type_tucmd, mss_l4len_idx);
6569 }
6570 
6571 #define IXGBE_SET_FLAG(_input, _flag, _result) \
6572 	((_flag <= _result) ? \
6573 	 ((u32)(_input & _flag) * (_result / _flag)) : \
6574 	 ((u32)(_input & _flag) / (_flag / _result)))
6575 
6576 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
6577 {
6578 	/* set type for advanced descriptor with frame checksum insertion */
6579 	u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
6580 		       IXGBE_ADVTXD_DCMD_DEXT |
6581 		       IXGBE_ADVTXD_DCMD_IFCS;
6582 
6583 	/* set HW vlan bit if vlan is present */
6584 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
6585 				   IXGBE_ADVTXD_DCMD_VLE);
6586 
6587 	/* set segmentation enable bits for TSO/FSO */
6588 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
6589 				   IXGBE_ADVTXD_DCMD_TSE);
6590 
6591 	/* set timestamp bit if present */
6592 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
6593 				   IXGBE_ADVTXD_MAC_TSTAMP);
6594 
6595 	/* insert frame checksum */
6596 	cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
6597 
6598 	return cmd_type;
6599 }
6600 
6601 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
6602 				   u32 tx_flags, unsigned int paylen)
6603 {
6604 	u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
6605 
6606 	/* enable L4 checksum for TSO and TX checksum offload */
6607 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
6608 					IXGBE_TX_FLAGS_CSUM,
6609 					IXGBE_ADVTXD_POPTS_TXSM);
6610 
6611 	/* enble IPv4 checksum for TSO */
6612 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
6613 					IXGBE_TX_FLAGS_IPV4,
6614 					IXGBE_ADVTXD_POPTS_IXSM);
6615 
6616 	/*
6617 	 * Check Context must be set if Tx switch is enabled, which it
6618 	 * always is for case where virtual functions are running
6619 	 */
6620 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
6621 					IXGBE_TX_FLAGS_CC,
6622 					IXGBE_ADVTXD_CC);
6623 
6624 	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6625 }
6626 
6627 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
6628 		       IXGBE_TXD_CMD_RS)
6629 
6630 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
6631 			 struct ixgbe_tx_buffer *first,
6632 			 const u8 hdr_len)
6633 {
6634 	struct sk_buff *skb = first->skb;
6635 	struct ixgbe_tx_buffer *tx_buffer;
6636 	union ixgbe_adv_tx_desc *tx_desc;
6637 	struct skb_frag_struct *frag;
6638 	dma_addr_t dma;
6639 	unsigned int data_len, size;
6640 	u32 tx_flags = first->tx_flags;
6641 	u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
6642 	u16 i = tx_ring->next_to_use;
6643 
6644 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
6645 
6646 	ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
6647 
6648 	size = skb_headlen(skb);
6649 	data_len = skb->data_len;
6650 
6651 #ifdef IXGBE_FCOE
6652 	if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6653 		if (data_len < sizeof(struct fcoe_crc_eof)) {
6654 			size -= sizeof(struct fcoe_crc_eof) - data_len;
6655 			data_len = 0;
6656 		} else {
6657 			data_len -= sizeof(struct fcoe_crc_eof);
6658 		}
6659 	}
6660 
6661 #endif
6662 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
6663 
6664 	tx_buffer = first;
6665 
6666 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
6667 		if (dma_mapping_error(tx_ring->dev, dma))
6668 			goto dma_error;
6669 
6670 		/* record length, and DMA address */
6671 		dma_unmap_len_set(tx_buffer, len, size);
6672 		dma_unmap_addr_set(tx_buffer, dma, dma);
6673 
6674 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
6675 
6676 		while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
6677 			tx_desc->read.cmd_type_len =
6678 				cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
6679 
6680 			i++;
6681 			tx_desc++;
6682 			if (i == tx_ring->count) {
6683 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6684 				i = 0;
6685 			}
6686 			tx_desc->read.olinfo_status = 0;
6687 
6688 			dma += IXGBE_MAX_DATA_PER_TXD;
6689 			size -= IXGBE_MAX_DATA_PER_TXD;
6690 
6691 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
6692 		}
6693 
6694 		if (likely(!data_len))
6695 			break;
6696 
6697 		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
6698 
6699 		i++;
6700 		tx_desc++;
6701 		if (i == tx_ring->count) {
6702 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6703 			i = 0;
6704 		}
6705 		tx_desc->read.olinfo_status = 0;
6706 
6707 #ifdef IXGBE_FCOE
6708 		size = min_t(unsigned int, data_len, skb_frag_size(frag));
6709 #else
6710 		size = skb_frag_size(frag);
6711 #endif
6712 		data_len -= size;
6713 
6714 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
6715 				       DMA_TO_DEVICE);
6716 
6717 		tx_buffer = &tx_ring->tx_buffer_info[i];
6718 	}
6719 
6720 	/* write last descriptor with RS and EOP bits */
6721 	cmd_type |= size | IXGBE_TXD_CMD;
6722 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
6723 
6724 	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
6725 
6726 	/* set the timestamp */
6727 	first->time_stamp = jiffies;
6728 
6729 	/*
6730 	 * Force memory writes to complete before letting h/w know there
6731 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
6732 	 * memory model archs, such as IA-64).
6733 	 *
6734 	 * We also need this memory barrier to make certain all of the
6735 	 * status bits have been updated before next_to_watch is written.
6736 	 */
6737 	wmb();
6738 
6739 	/* set next_to_watch value indicating a packet is present */
6740 	first->next_to_watch = tx_desc;
6741 
6742 	i++;
6743 	if (i == tx_ring->count)
6744 		i = 0;
6745 
6746 	tx_ring->next_to_use = i;
6747 
6748 	/* notify HW of packet */
6749 	ixgbe_write_tail(tx_ring, i);
6750 
6751 	return;
6752 dma_error:
6753 	dev_err(tx_ring->dev, "TX DMA map failed\n");
6754 
6755 	/* clear dma mappings for failed tx_buffer_info map */
6756 	for (;;) {
6757 		tx_buffer = &tx_ring->tx_buffer_info[i];
6758 		ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
6759 		if (tx_buffer == first)
6760 			break;
6761 		if (i == 0)
6762 			i = tx_ring->count;
6763 		i--;
6764 	}
6765 
6766 	tx_ring->next_to_use = i;
6767 }
6768 
6769 static void ixgbe_atr(struct ixgbe_ring *ring,
6770 		      struct ixgbe_tx_buffer *first)
6771 {
6772 	struct ixgbe_q_vector *q_vector = ring->q_vector;
6773 	union ixgbe_atr_hash_dword input = { .dword = 0 };
6774 	union ixgbe_atr_hash_dword common = { .dword = 0 };
6775 	union {
6776 		unsigned char *network;
6777 		struct iphdr *ipv4;
6778 		struct ipv6hdr *ipv6;
6779 	} hdr;
6780 	struct tcphdr *th;
6781 	__be16 vlan_id;
6782 
6783 	/* if ring doesn't have a interrupt vector, cannot perform ATR */
6784 	if (!q_vector)
6785 		return;
6786 
6787 	/* do nothing if sampling is disabled */
6788 	if (!ring->atr_sample_rate)
6789 		return;
6790 
6791 	ring->atr_count++;
6792 
6793 	/* snag network header to get L4 type and address */
6794 	hdr.network = skb_network_header(first->skb);
6795 
6796 	/* Currently only IPv4/IPv6 with TCP is supported */
6797 	if ((first->protocol != __constant_htons(ETH_P_IPV6) ||
6798 	     hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6799 	    (first->protocol != __constant_htons(ETH_P_IP) ||
6800 	     hdr.ipv4->protocol != IPPROTO_TCP))
6801 		return;
6802 
6803 	th = tcp_hdr(first->skb);
6804 
6805 	/* skip this packet since it is invalid or the socket is closing */
6806 	if (!th || th->fin)
6807 		return;
6808 
6809 	/* sample on all syn packets or once every atr sample count */
6810 	if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6811 		return;
6812 
6813 	/* reset sample count */
6814 	ring->atr_count = 0;
6815 
6816 	vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6817 
6818 	/*
6819 	 * src and dst are inverted, think how the receiver sees them
6820 	 *
6821 	 * The input is broken into two sections, a non-compressed section
6822 	 * containing vm_pool, vlan_id, and flow_type.  The rest of the data
6823 	 * is XORed together and stored in the compressed dword.
6824 	 */
6825 	input.formatted.vlan_id = vlan_id;
6826 
6827 	/*
6828 	 * since src port and flex bytes occupy the same word XOR them together
6829 	 * and write the value to source port portion of compressed dword
6830 	 */
6831 	if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6832 		common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6833 	else
6834 		common.port.src ^= th->dest ^ first->protocol;
6835 	common.port.dst ^= th->source;
6836 
6837 	if (first->protocol == __constant_htons(ETH_P_IP)) {
6838 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6839 		common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6840 	} else {
6841 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6842 		common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6843 			     hdr.ipv6->saddr.s6_addr32[1] ^
6844 			     hdr.ipv6->saddr.s6_addr32[2] ^
6845 			     hdr.ipv6->saddr.s6_addr32[3] ^
6846 			     hdr.ipv6->daddr.s6_addr32[0] ^
6847 			     hdr.ipv6->daddr.s6_addr32[1] ^
6848 			     hdr.ipv6->daddr.s6_addr32[2] ^
6849 			     hdr.ipv6->daddr.s6_addr32[3];
6850 	}
6851 
6852 	/* This assumes the Rx queue and Tx queue are bound to the same CPU */
6853 	ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6854 					      input, common, ring->queue_index);
6855 }
6856 
6857 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6858 {
6859 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6860 	/* Herbert's original patch had:
6861 	 *  smp_mb__after_netif_stop_queue();
6862 	 * but since that doesn't exist yet, just open code it. */
6863 	smp_mb();
6864 
6865 	/* We need to check again in a case another CPU has just
6866 	 * made room available. */
6867 	if (likely(ixgbe_desc_unused(tx_ring) < size))
6868 		return -EBUSY;
6869 
6870 	/* A reprieve! - use start_queue because it doesn't call schedule */
6871 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6872 	++tx_ring->tx_stats.restart_queue;
6873 	return 0;
6874 }
6875 
6876 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6877 {
6878 	if (likely(ixgbe_desc_unused(tx_ring) >= size))
6879 		return 0;
6880 	return __ixgbe_maybe_stop_tx(tx_ring, size);
6881 }
6882 
6883 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
6884 			      void *accel_priv)
6885 {
6886 	struct ixgbe_fwd_adapter *fwd_adapter = accel_priv;
6887 #ifdef IXGBE_FCOE
6888 	struct ixgbe_adapter *adapter;
6889 	struct ixgbe_ring_feature *f;
6890 	int txq;
6891 #endif
6892 
6893 	if (fwd_adapter)
6894 		return skb->queue_mapping + fwd_adapter->tx_base_queue;
6895 
6896 #ifdef IXGBE_FCOE
6897 
6898 	/*
6899 	 * only execute the code below if protocol is FCoE
6900 	 * or FIP and we have FCoE enabled on the adapter
6901 	 */
6902 	switch (vlan_get_protocol(skb)) {
6903 	case __constant_htons(ETH_P_FCOE):
6904 	case __constant_htons(ETH_P_FIP):
6905 		adapter = netdev_priv(dev);
6906 
6907 		if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
6908 			break;
6909 	default:
6910 		return __netdev_pick_tx(dev, skb);
6911 	}
6912 
6913 	f = &adapter->ring_feature[RING_F_FCOE];
6914 
6915 	txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6916 					   smp_processor_id();
6917 
6918 	while (txq >= f->indices)
6919 		txq -= f->indices;
6920 
6921 	return txq + f->offset;
6922 #else
6923 	return __netdev_pick_tx(dev, skb);
6924 #endif
6925 }
6926 
6927 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6928 			  struct ixgbe_adapter *adapter,
6929 			  struct ixgbe_ring *tx_ring)
6930 {
6931 	struct ixgbe_tx_buffer *first;
6932 	int tso;
6933 	u32 tx_flags = 0;
6934 	unsigned short f;
6935 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
6936 	__be16 protocol = skb->protocol;
6937 	u8 hdr_len = 0;
6938 
6939 	/*
6940 	 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6941 	 *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
6942 	 *       + 2 desc gap to keep tail from touching head,
6943 	 *       + 1 desc for context descriptor,
6944 	 * otherwise try next time
6945 	 */
6946 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6947 		count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6948 
6949 	if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
6950 		tx_ring->tx_stats.tx_busy++;
6951 		return NETDEV_TX_BUSY;
6952 	}
6953 
6954 	/* record the location of the first descriptor for this packet */
6955 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6956 	first->skb = skb;
6957 	first->bytecount = skb->len;
6958 	first->gso_segs = 1;
6959 
6960 	/* if we have a HW VLAN tag being added default to the HW one */
6961 	if (vlan_tx_tag_present(skb)) {
6962 		tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6963 		tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6964 	/* else if it is a SW VLAN check the next protocol and store the tag */
6965 	} else if (protocol == __constant_htons(ETH_P_8021Q)) {
6966 		struct vlan_hdr *vhdr, _vhdr;
6967 		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
6968 		if (!vhdr)
6969 			goto out_drop;
6970 
6971 		protocol = vhdr->h_vlan_encapsulated_proto;
6972 		tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
6973 				  IXGBE_TX_FLAGS_VLAN_SHIFT;
6974 		tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
6975 	}
6976 
6977 	skb_tx_timestamp(skb);
6978 
6979 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
6980 		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
6981 		tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
6982 
6983 		/* schedule check for Tx timestamp */
6984 		adapter->ptp_tx_skb = skb_get(skb);
6985 		adapter->ptp_tx_start = jiffies;
6986 		schedule_work(&adapter->ptp_tx_work);
6987 	}
6988 
6989 #ifdef CONFIG_PCI_IOV
6990 	/*
6991 	 * Use the l2switch_enable flag - would be false if the DMA
6992 	 * Tx switch had been disabled.
6993 	 */
6994 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6995 		tx_flags |= IXGBE_TX_FLAGS_CC;
6996 
6997 #endif
6998 	/* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
6999 	if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
7000 	    ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
7001 	     (skb->priority != TC_PRIO_CONTROL))) {
7002 		tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
7003 		tx_flags |= (skb->priority & 0x7) <<
7004 					IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
7005 		if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
7006 			struct vlan_ethhdr *vhdr;
7007 			if (skb_header_cloned(skb) &&
7008 			    pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
7009 				goto out_drop;
7010 			vhdr = (struct vlan_ethhdr *)skb->data;
7011 			vhdr->h_vlan_TCI = htons(tx_flags >>
7012 						 IXGBE_TX_FLAGS_VLAN_SHIFT);
7013 		} else {
7014 			tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
7015 		}
7016 	}
7017 
7018 	/* record initial flags and protocol */
7019 	first->tx_flags = tx_flags;
7020 	first->protocol = protocol;
7021 
7022 #ifdef IXGBE_FCOE
7023 	/* setup tx offload for FCoE */
7024 	if ((protocol == __constant_htons(ETH_P_FCOE)) &&
7025 	    (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
7026 		tso = ixgbe_fso(tx_ring, first, &hdr_len);
7027 		if (tso < 0)
7028 			goto out_drop;
7029 
7030 		goto xmit_fcoe;
7031 	}
7032 
7033 #endif /* IXGBE_FCOE */
7034 	tso = ixgbe_tso(tx_ring, first, &hdr_len);
7035 	if (tso < 0)
7036 		goto out_drop;
7037 	else if (!tso)
7038 		ixgbe_tx_csum(tx_ring, first);
7039 
7040 	/* add the ATR filter if ATR is on */
7041 	if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
7042 		ixgbe_atr(tx_ring, first);
7043 
7044 #ifdef IXGBE_FCOE
7045 xmit_fcoe:
7046 #endif /* IXGBE_FCOE */
7047 	ixgbe_tx_map(tx_ring, first, hdr_len);
7048 
7049 	ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
7050 
7051 	return NETDEV_TX_OK;
7052 
7053 out_drop:
7054 	dev_kfree_skb_any(first->skb);
7055 	first->skb = NULL;
7056 
7057 	return NETDEV_TX_OK;
7058 }
7059 
7060 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
7061 				      struct net_device *netdev,
7062 				      struct ixgbe_ring *ring)
7063 {
7064 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7065 	struct ixgbe_ring *tx_ring;
7066 
7067 	/*
7068 	 * The minimum packet size for olinfo paylen is 17 so pad the skb
7069 	 * in order to meet this minimum size requirement.
7070 	 */
7071 	if (unlikely(skb->len < 17)) {
7072 		if (skb_pad(skb, 17 - skb->len))
7073 			return NETDEV_TX_OK;
7074 		skb->len = 17;
7075 		skb_set_tail_pointer(skb, 17);
7076 	}
7077 
7078 	tx_ring = ring ? ring : adapter->tx_ring[skb->queue_mapping];
7079 
7080 	return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
7081 }
7082 
7083 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
7084 				    struct net_device *netdev)
7085 {
7086 	return __ixgbe_xmit_frame(skb, netdev, NULL);
7087 }
7088 
7089 /**
7090  * ixgbe_set_mac - Change the Ethernet Address of the NIC
7091  * @netdev: network interface device structure
7092  * @p: pointer to an address structure
7093  *
7094  * Returns 0 on success, negative on failure
7095  **/
7096 static int ixgbe_set_mac(struct net_device *netdev, void *p)
7097 {
7098 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7099 	struct ixgbe_hw *hw = &adapter->hw;
7100 	struct sockaddr *addr = p;
7101 
7102 	if (!is_valid_ether_addr(addr->sa_data))
7103 		return -EADDRNOTAVAIL;
7104 
7105 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
7106 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
7107 
7108 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, VMDQ_P(0), IXGBE_RAH_AV);
7109 
7110 	return 0;
7111 }
7112 
7113 static int
7114 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
7115 {
7116 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7117 	struct ixgbe_hw *hw = &adapter->hw;
7118 	u16 value;
7119 	int rc;
7120 
7121 	if (prtad != hw->phy.mdio.prtad)
7122 		return -EINVAL;
7123 	rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
7124 	if (!rc)
7125 		rc = value;
7126 	return rc;
7127 }
7128 
7129 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
7130 			    u16 addr, u16 value)
7131 {
7132 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7133 	struct ixgbe_hw *hw = &adapter->hw;
7134 
7135 	if (prtad != hw->phy.mdio.prtad)
7136 		return -EINVAL;
7137 	return hw->phy.ops.write_reg(hw, addr, devad, value);
7138 }
7139 
7140 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
7141 {
7142 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7143 
7144 	switch (cmd) {
7145 	case SIOCSHWTSTAMP:
7146 		return ixgbe_ptp_hwtstamp_ioctl(adapter, req, cmd);
7147 	default:
7148 		return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
7149 	}
7150 }
7151 
7152 /**
7153  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
7154  * netdev->dev_addrs
7155  * @netdev: network interface device structure
7156  *
7157  * Returns non-zero on failure
7158  **/
7159 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
7160 {
7161 	int err = 0;
7162 	struct ixgbe_adapter *adapter = netdev_priv(dev);
7163 	struct ixgbe_hw *hw = &adapter->hw;
7164 
7165 	if (is_valid_ether_addr(hw->mac.san_addr)) {
7166 		rtnl_lock();
7167 		err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
7168 		rtnl_unlock();
7169 
7170 		/* update SAN MAC vmdq pool selection */
7171 		hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
7172 	}
7173 	return err;
7174 }
7175 
7176 /**
7177  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
7178  * netdev->dev_addrs
7179  * @netdev: network interface device structure
7180  *
7181  * Returns non-zero on failure
7182  **/
7183 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
7184 {
7185 	int err = 0;
7186 	struct ixgbe_adapter *adapter = netdev_priv(dev);
7187 	struct ixgbe_mac_info *mac = &adapter->hw.mac;
7188 
7189 	if (is_valid_ether_addr(mac->san_addr)) {
7190 		rtnl_lock();
7191 		err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
7192 		rtnl_unlock();
7193 	}
7194 	return err;
7195 }
7196 
7197 #ifdef CONFIG_NET_POLL_CONTROLLER
7198 /*
7199  * Polling 'interrupt' - used by things like netconsole to send skbs
7200  * without having to re-enable interrupts. It's not called while
7201  * the interrupt routine is executing.
7202  */
7203 static void ixgbe_netpoll(struct net_device *netdev)
7204 {
7205 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7206 	int i;
7207 
7208 	/* if interface is down do nothing */
7209 	if (test_bit(__IXGBE_DOWN, &adapter->state))
7210 		return;
7211 
7212 	adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
7213 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
7214 		for (i = 0; i < adapter->num_q_vectors; i++)
7215 			ixgbe_msix_clean_rings(0, adapter->q_vector[i]);
7216 	} else {
7217 		ixgbe_intr(adapter->pdev->irq, netdev);
7218 	}
7219 	adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
7220 }
7221 
7222 #endif
7223 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
7224 						   struct rtnl_link_stats64 *stats)
7225 {
7226 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7227 	int i;
7228 
7229 	rcu_read_lock();
7230 	for (i = 0; i < adapter->num_rx_queues; i++) {
7231 		struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
7232 		u64 bytes, packets;
7233 		unsigned int start;
7234 
7235 		if (ring) {
7236 			do {
7237 				start = u64_stats_fetch_begin_bh(&ring->syncp);
7238 				packets = ring->stats.packets;
7239 				bytes   = ring->stats.bytes;
7240 			} while (u64_stats_fetch_retry_bh(&ring->syncp, start));
7241 			stats->rx_packets += packets;
7242 			stats->rx_bytes   += bytes;
7243 		}
7244 	}
7245 
7246 	for (i = 0; i < adapter->num_tx_queues; i++) {
7247 		struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
7248 		u64 bytes, packets;
7249 		unsigned int start;
7250 
7251 		if (ring) {
7252 			do {
7253 				start = u64_stats_fetch_begin_bh(&ring->syncp);
7254 				packets = ring->stats.packets;
7255 				bytes   = ring->stats.bytes;
7256 			} while (u64_stats_fetch_retry_bh(&ring->syncp, start));
7257 			stats->tx_packets += packets;
7258 			stats->tx_bytes   += bytes;
7259 		}
7260 	}
7261 	rcu_read_unlock();
7262 	/* following stats updated by ixgbe_watchdog_task() */
7263 	stats->multicast	= netdev->stats.multicast;
7264 	stats->rx_errors	= netdev->stats.rx_errors;
7265 	stats->rx_length_errors	= netdev->stats.rx_length_errors;
7266 	stats->rx_crc_errors	= netdev->stats.rx_crc_errors;
7267 	stats->rx_missed_errors	= netdev->stats.rx_missed_errors;
7268 	return stats;
7269 }
7270 
7271 #ifdef CONFIG_IXGBE_DCB
7272 /**
7273  * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
7274  * @adapter: pointer to ixgbe_adapter
7275  * @tc: number of traffic classes currently enabled
7276  *
7277  * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
7278  * 802.1Q priority maps to a packet buffer that exists.
7279  */
7280 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
7281 {
7282 	struct ixgbe_hw *hw = &adapter->hw;
7283 	u32 reg, rsave;
7284 	int i;
7285 
7286 	/* 82598 have a static priority to TC mapping that can not
7287 	 * be changed so no validation is needed.
7288 	 */
7289 	if (hw->mac.type == ixgbe_mac_82598EB)
7290 		return;
7291 
7292 	reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
7293 	rsave = reg;
7294 
7295 	for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
7296 		u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
7297 
7298 		/* If up2tc is out of bounds default to zero */
7299 		if (up2tc > tc)
7300 			reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
7301 	}
7302 
7303 	if (reg != rsave)
7304 		IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
7305 
7306 	return;
7307 }
7308 
7309 /**
7310  * ixgbe_set_prio_tc_map - Configure netdev prio tc map
7311  * @adapter: Pointer to adapter struct
7312  *
7313  * Populate the netdev user priority to tc map
7314  */
7315 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
7316 {
7317 	struct net_device *dev = adapter->netdev;
7318 	struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
7319 	struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
7320 	u8 prio;
7321 
7322 	for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
7323 		u8 tc = 0;
7324 
7325 		if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
7326 			tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
7327 		else if (ets)
7328 			tc = ets->prio_tc[prio];
7329 
7330 		netdev_set_prio_tc_map(dev, prio, tc);
7331 	}
7332 }
7333 
7334 #endif /* CONFIG_IXGBE_DCB */
7335 /**
7336  * ixgbe_setup_tc - configure net_device for multiple traffic classes
7337  *
7338  * @netdev: net device to configure
7339  * @tc: number of traffic classes to enable
7340  */
7341 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
7342 {
7343 	struct ixgbe_adapter *adapter = netdev_priv(dev);
7344 	struct ixgbe_hw *hw = &adapter->hw;
7345 	bool pools;
7346 
7347 	/* Hardware supports up to 8 traffic classes */
7348 	if (tc > adapter->dcb_cfg.num_tcs.pg_tcs ||
7349 	    (hw->mac.type == ixgbe_mac_82598EB &&
7350 	     tc < MAX_TRAFFIC_CLASS))
7351 		return -EINVAL;
7352 
7353 	pools = (find_first_zero_bit(&adapter->fwd_bitmask, 32) > 1);
7354 	if (tc && pools && adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS)
7355 		return -EBUSY;
7356 
7357 	/* Hardware has to reinitialize queues and interrupts to
7358 	 * match packet buffer alignment. Unfortunately, the
7359 	 * hardware is not flexible enough to do this dynamically.
7360 	 */
7361 	if (netif_running(dev))
7362 		ixgbe_close(dev);
7363 	ixgbe_clear_interrupt_scheme(adapter);
7364 
7365 #ifdef CONFIG_IXGBE_DCB
7366 	if (tc) {
7367 		netdev_set_num_tc(dev, tc);
7368 		ixgbe_set_prio_tc_map(adapter);
7369 
7370 		adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
7371 
7372 		if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
7373 			adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
7374 			adapter->hw.fc.requested_mode = ixgbe_fc_none;
7375 		}
7376 	} else {
7377 		netdev_reset_tc(dev);
7378 
7379 		if (adapter->hw.mac.type == ixgbe_mac_82598EB)
7380 			adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
7381 
7382 		adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
7383 
7384 		adapter->temp_dcb_cfg.pfc_mode_enable = false;
7385 		adapter->dcb_cfg.pfc_mode_enable = false;
7386 	}
7387 
7388 	ixgbe_validate_rtr(adapter, tc);
7389 
7390 #endif /* CONFIG_IXGBE_DCB */
7391 	ixgbe_init_interrupt_scheme(adapter);
7392 
7393 	if (netif_running(dev))
7394 		return ixgbe_open(dev);
7395 
7396 	return 0;
7397 }
7398 
7399 #ifdef CONFIG_PCI_IOV
7400 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
7401 {
7402 	struct net_device *netdev = adapter->netdev;
7403 
7404 	rtnl_lock();
7405 	ixgbe_setup_tc(netdev, netdev_get_num_tc(netdev));
7406 	rtnl_unlock();
7407 }
7408 
7409 #endif
7410 void ixgbe_do_reset(struct net_device *netdev)
7411 {
7412 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7413 
7414 	if (netif_running(netdev))
7415 		ixgbe_reinit_locked(adapter);
7416 	else
7417 		ixgbe_reset(adapter);
7418 }
7419 
7420 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
7421 					    netdev_features_t features)
7422 {
7423 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7424 
7425 	/* If Rx checksum is disabled, then RSC/LRO should also be disabled */
7426 	if (!(features & NETIF_F_RXCSUM))
7427 		features &= ~NETIF_F_LRO;
7428 
7429 	/* Turn off LRO if not RSC capable */
7430 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
7431 		features &= ~NETIF_F_LRO;
7432 
7433 	return features;
7434 }
7435 
7436 static int ixgbe_set_features(struct net_device *netdev,
7437 			      netdev_features_t features)
7438 {
7439 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7440 	netdev_features_t changed = netdev->features ^ features;
7441 	bool need_reset = false;
7442 
7443 	/* Make sure RSC matches LRO, reset if change */
7444 	if (!(features & NETIF_F_LRO)) {
7445 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7446 			need_reset = true;
7447 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
7448 	} else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
7449 		   !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
7450 		if (adapter->rx_itr_setting == 1 ||
7451 		    adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
7452 			adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
7453 			need_reset = true;
7454 		} else if ((changed ^ features) & NETIF_F_LRO) {
7455 			e_info(probe, "rx-usecs set too low, "
7456 			       "disabling RSC\n");
7457 		}
7458 	}
7459 
7460 	/*
7461 	 * Check if Flow Director n-tuple support was enabled or disabled.  If
7462 	 * the state changed, we need to reset.
7463 	 */
7464 	switch (features & NETIF_F_NTUPLE) {
7465 	case NETIF_F_NTUPLE:
7466 		/* turn off ATR, enable perfect filters and reset */
7467 		if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
7468 			need_reset = true;
7469 
7470 		adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
7471 		adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7472 		break;
7473 	default:
7474 		/* turn off perfect filters, enable ATR and reset */
7475 		if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7476 			need_reset = true;
7477 
7478 		adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7479 
7480 		/* We cannot enable ATR if SR-IOV is enabled */
7481 		if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7482 			break;
7483 
7484 		/* We cannot enable ATR if we have 2 or more traffic classes */
7485 		if (netdev_get_num_tc(netdev) > 1)
7486 			break;
7487 
7488 		/* We cannot enable ATR if RSS is disabled */
7489 		if (adapter->ring_feature[RING_F_RSS].limit <= 1)
7490 			break;
7491 
7492 		/* A sample rate of 0 indicates ATR disabled */
7493 		if (!adapter->atr_sample_rate)
7494 			break;
7495 
7496 		adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
7497 		break;
7498 	}
7499 
7500 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
7501 		ixgbe_vlan_strip_enable(adapter);
7502 	else
7503 		ixgbe_vlan_strip_disable(adapter);
7504 
7505 	if (changed & NETIF_F_RXALL)
7506 		need_reset = true;
7507 
7508 	netdev->features = features;
7509 	if (need_reset)
7510 		ixgbe_do_reset(netdev);
7511 
7512 	return 0;
7513 }
7514 
7515 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
7516 			     struct net_device *dev,
7517 			     const unsigned char *addr,
7518 			     u16 flags)
7519 {
7520 	struct ixgbe_adapter *adapter = netdev_priv(dev);
7521 	int err;
7522 
7523 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
7524 		return ndo_dflt_fdb_add(ndm, tb, dev, addr, flags);
7525 
7526 	/* Hardware does not support aging addresses so if a
7527 	 * ndm_state is given only allow permanent addresses
7528 	 */
7529 	if (ndm->ndm_state && !(ndm->ndm_state & NUD_PERMANENT)) {
7530 		pr_info("%s: FDB only supports static addresses\n",
7531 			ixgbe_driver_name);
7532 		return -EINVAL;
7533 	}
7534 
7535 	if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
7536 		u32 rar_uc_entries = IXGBE_MAX_PF_MACVLANS;
7537 
7538 		if (netdev_uc_count(dev) < rar_uc_entries)
7539 			err = dev_uc_add_excl(dev, addr);
7540 		else
7541 			err = -ENOMEM;
7542 	} else if (is_multicast_ether_addr(addr)) {
7543 		err = dev_mc_add_excl(dev, addr);
7544 	} else {
7545 		err = -EINVAL;
7546 	}
7547 
7548 	/* Only return duplicate errors if NLM_F_EXCL is set */
7549 	if (err == -EEXIST && !(flags & NLM_F_EXCL))
7550 		err = 0;
7551 
7552 	return err;
7553 }
7554 
7555 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
7556 				    struct nlmsghdr *nlh)
7557 {
7558 	struct ixgbe_adapter *adapter = netdev_priv(dev);
7559 	struct nlattr *attr, *br_spec;
7560 	int rem;
7561 
7562 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
7563 		return -EOPNOTSUPP;
7564 
7565 	br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
7566 
7567 	nla_for_each_nested(attr, br_spec, rem) {
7568 		__u16 mode;
7569 		u32 reg = 0;
7570 
7571 		if (nla_type(attr) != IFLA_BRIDGE_MODE)
7572 			continue;
7573 
7574 		mode = nla_get_u16(attr);
7575 		if (mode == BRIDGE_MODE_VEPA) {
7576 			reg = 0;
7577 			adapter->flags2 &= ~IXGBE_FLAG2_BRIDGE_MODE_VEB;
7578 		} else if (mode == BRIDGE_MODE_VEB) {
7579 			reg = IXGBE_PFDTXGSWC_VT_LBEN;
7580 			adapter->flags2 |= IXGBE_FLAG2_BRIDGE_MODE_VEB;
7581 		} else
7582 			return -EINVAL;
7583 
7584 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, reg);
7585 
7586 		e_info(drv, "enabling bridge mode: %s\n",
7587 			mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
7588 	}
7589 
7590 	return 0;
7591 }
7592 
7593 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
7594 				    struct net_device *dev,
7595 				    u32 filter_mask)
7596 {
7597 	struct ixgbe_adapter *adapter = netdev_priv(dev);
7598 	u16 mode;
7599 
7600 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
7601 		return 0;
7602 
7603 	if (adapter->flags2 & IXGBE_FLAG2_BRIDGE_MODE_VEB)
7604 		mode = BRIDGE_MODE_VEB;
7605 	else
7606 		mode = BRIDGE_MODE_VEPA;
7607 
7608 	return ndo_dflt_bridge_getlink(skb, pid, seq, dev, mode);
7609 }
7610 
7611 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
7612 {
7613 	struct ixgbe_fwd_adapter *fwd_adapter = NULL;
7614 	struct ixgbe_adapter *adapter = netdev_priv(pdev);
7615 	unsigned int limit;
7616 	int pool, err;
7617 
7618 #ifdef CONFIG_RPS
7619 	if (vdev->num_rx_queues != vdev->num_tx_queues) {
7620 		netdev_info(pdev, "%s: Only supports a single queue count for TX and RX\n",
7621 			    vdev->name);
7622 		return ERR_PTR(-EINVAL);
7623 	}
7624 #endif
7625 	/* Check for hardware restriction on number of rx/tx queues */
7626 	if (vdev->num_tx_queues > IXGBE_MAX_L2A_QUEUES ||
7627 	    vdev->num_tx_queues == IXGBE_BAD_L2A_QUEUE) {
7628 		netdev_info(pdev,
7629 			    "%s: Supports RX/TX Queue counts 1,2, and 4\n",
7630 			    pdev->name);
7631 		return ERR_PTR(-EINVAL);
7632 	}
7633 
7634 	if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
7635 	      adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS - 1) ||
7636 	    (adapter->num_rx_pools > IXGBE_MAX_MACVLANS))
7637 		return ERR_PTR(-EBUSY);
7638 
7639 	fwd_adapter = kcalloc(1, sizeof(struct ixgbe_fwd_adapter), GFP_KERNEL);
7640 	if (!fwd_adapter)
7641 		return ERR_PTR(-ENOMEM);
7642 
7643 	pool = find_first_zero_bit(&adapter->fwd_bitmask, 32);
7644 	adapter->num_rx_pools++;
7645 	set_bit(pool, &adapter->fwd_bitmask);
7646 	limit = find_last_bit(&adapter->fwd_bitmask, 32);
7647 
7648 	/* Enable VMDq flag so device will be set in VM mode */
7649 	adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED | IXGBE_FLAG_SRIOV_ENABLED;
7650 	adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
7651 	adapter->ring_feature[RING_F_RSS].limit = vdev->num_tx_queues;
7652 
7653 	/* Force reinit of ring allocation with VMDQ enabled */
7654 	err = ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev));
7655 	if (err)
7656 		goto fwd_add_err;
7657 	fwd_adapter->pool = pool;
7658 	fwd_adapter->real_adapter = adapter;
7659 	err = ixgbe_fwd_ring_up(vdev, fwd_adapter);
7660 	if (err)
7661 		goto fwd_add_err;
7662 	netif_tx_start_all_queues(vdev);
7663 	return fwd_adapter;
7664 fwd_add_err:
7665 	/* unwind counter and free adapter struct */
7666 	netdev_info(pdev,
7667 		    "%s: dfwd hardware acceleration failed\n", vdev->name);
7668 	clear_bit(pool, &adapter->fwd_bitmask);
7669 	adapter->num_rx_pools--;
7670 	kfree(fwd_adapter);
7671 	return ERR_PTR(err);
7672 }
7673 
7674 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
7675 {
7676 	struct ixgbe_fwd_adapter *fwd_adapter = priv;
7677 	struct ixgbe_adapter *adapter = fwd_adapter->real_adapter;
7678 	unsigned int limit;
7679 
7680 	clear_bit(fwd_adapter->pool, &adapter->fwd_bitmask);
7681 	adapter->num_rx_pools--;
7682 
7683 	limit = find_last_bit(&adapter->fwd_bitmask, 32);
7684 	adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
7685 	ixgbe_fwd_ring_down(fwd_adapter->netdev, fwd_adapter);
7686 	ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev));
7687 	netdev_dbg(pdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
7688 		   fwd_adapter->pool, adapter->num_rx_pools,
7689 		   fwd_adapter->rx_base_queue,
7690 		   fwd_adapter->rx_base_queue + adapter->num_rx_queues_per_pool,
7691 		   adapter->fwd_bitmask);
7692 	kfree(fwd_adapter);
7693 }
7694 
7695 static const struct net_device_ops ixgbe_netdev_ops = {
7696 	.ndo_open		= ixgbe_open,
7697 	.ndo_stop		= ixgbe_close,
7698 	.ndo_start_xmit		= ixgbe_xmit_frame,
7699 	.ndo_select_queue	= ixgbe_select_queue,
7700 	.ndo_set_rx_mode	= ixgbe_set_rx_mode,
7701 	.ndo_validate_addr	= eth_validate_addr,
7702 	.ndo_set_mac_address	= ixgbe_set_mac,
7703 	.ndo_change_mtu		= ixgbe_change_mtu,
7704 	.ndo_tx_timeout		= ixgbe_tx_timeout,
7705 	.ndo_vlan_rx_add_vid	= ixgbe_vlan_rx_add_vid,
7706 	.ndo_vlan_rx_kill_vid	= ixgbe_vlan_rx_kill_vid,
7707 	.ndo_do_ioctl		= ixgbe_ioctl,
7708 	.ndo_set_vf_mac		= ixgbe_ndo_set_vf_mac,
7709 	.ndo_set_vf_vlan	= ixgbe_ndo_set_vf_vlan,
7710 	.ndo_set_vf_tx_rate	= ixgbe_ndo_set_vf_bw,
7711 	.ndo_set_vf_spoofchk	= ixgbe_ndo_set_vf_spoofchk,
7712 	.ndo_get_vf_config	= ixgbe_ndo_get_vf_config,
7713 	.ndo_get_stats64	= ixgbe_get_stats64,
7714 #ifdef CONFIG_IXGBE_DCB
7715 	.ndo_setup_tc		= ixgbe_setup_tc,
7716 #endif
7717 #ifdef CONFIG_NET_POLL_CONTROLLER
7718 	.ndo_poll_controller	= ixgbe_netpoll,
7719 #endif
7720 #ifdef CONFIG_NET_RX_BUSY_POLL
7721 	.ndo_busy_poll		= ixgbe_low_latency_recv,
7722 #endif
7723 #ifdef IXGBE_FCOE
7724 	.ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
7725 	.ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
7726 	.ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
7727 	.ndo_fcoe_enable = ixgbe_fcoe_enable,
7728 	.ndo_fcoe_disable = ixgbe_fcoe_disable,
7729 	.ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
7730 	.ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
7731 #endif /* IXGBE_FCOE */
7732 	.ndo_set_features = ixgbe_set_features,
7733 	.ndo_fix_features = ixgbe_fix_features,
7734 	.ndo_fdb_add		= ixgbe_ndo_fdb_add,
7735 	.ndo_bridge_setlink	= ixgbe_ndo_bridge_setlink,
7736 	.ndo_bridge_getlink	= ixgbe_ndo_bridge_getlink,
7737 	.ndo_dfwd_add_station	= ixgbe_fwd_add,
7738 	.ndo_dfwd_del_station	= ixgbe_fwd_del,
7739 };
7740 
7741 /**
7742  * ixgbe_enumerate_functions - Get the number of ports this device has
7743  * @adapter: adapter structure
7744  *
7745  * This function enumerates the phsyical functions co-located on a single slot,
7746  * in order to determine how many ports a device has. This is most useful in
7747  * determining the required GT/s of PCIe bandwidth necessary for optimal
7748  * performance.
7749  **/
7750 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
7751 {
7752 	struct list_head *entry;
7753 	int physfns = 0;
7754 
7755 	/* Some cards can not use the generic count PCIe functions method,
7756 	 * because they are behind a parent switch, so we hardcode these with
7757 	 * the correct number of functions.
7758 	 */
7759 	if (ixgbe_pcie_from_parent(&adapter->hw)) {
7760 		physfns = 4;
7761 	} else {
7762 		list_for_each(entry, &adapter->pdev->bus_list) {
7763 			struct pci_dev *pdev =
7764 				list_entry(entry, struct pci_dev, bus_list);
7765 			/* don't count virtual functions */
7766 			if (!pdev->is_virtfn)
7767 				physfns++;
7768 		}
7769 	}
7770 
7771 	return physfns;
7772 }
7773 
7774 /**
7775  * ixgbe_wol_supported - Check whether device supports WoL
7776  * @hw: hw specific details
7777  * @device_id: the device ID
7778  * @subdev_id: the subsystem device ID
7779  *
7780  * This function is used by probe and ethtool to determine
7781  * which devices have WoL support
7782  *
7783  **/
7784 int ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
7785 			u16 subdevice_id)
7786 {
7787 	struct ixgbe_hw *hw = &adapter->hw;
7788 	u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
7789 	int is_wol_supported = 0;
7790 
7791 	switch (device_id) {
7792 	case IXGBE_DEV_ID_82599_SFP:
7793 		/* Only these subdevices could supports WOL */
7794 		switch (subdevice_id) {
7795 		case IXGBE_SUBDEV_ID_82599_560FLR:
7796 			/* only support first port */
7797 			if (hw->bus.func != 0)
7798 				break;
7799 		case IXGBE_SUBDEV_ID_82599_SP_560FLR:
7800 		case IXGBE_SUBDEV_ID_82599_SFP:
7801 		case IXGBE_SUBDEV_ID_82599_RNDC:
7802 		case IXGBE_SUBDEV_ID_82599_ECNA_DP:
7803 		case IXGBE_SUBDEV_ID_82599_LOM_SFP:
7804 			is_wol_supported = 1;
7805 			break;
7806 		}
7807 		break;
7808 	case IXGBE_DEV_ID_82599EN_SFP:
7809 		/* Only this subdevice supports WOL */
7810 		switch (subdevice_id) {
7811 		case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
7812 			is_wol_supported = 1;
7813 			break;
7814 		}
7815 		break;
7816 	case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7817 		/* All except this subdevice support WOL */
7818 		if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7819 			is_wol_supported = 1;
7820 		break;
7821 	case IXGBE_DEV_ID_82599_KX4:
7822 		is_wol_supported = 1;
7823 		break;
7824 	case IXGBE_DEV_ID_X540T:
7825 	case IXGBE_DEV_ID_X540T1:
7826 		/* check eeprom to see if enabled wol */
7827 		if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
7828 		    ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
7829 		     (hw->bus.func == 0))) {
7830 			is_wol_supported = 1;
7831 		}
7832 		break;
7833 	}
7834 
7835 	return is_wol_supported;
7836 }
7837 
7838 /**
7839  * ixgbe_probe - Device Initialization Routine
7840  * @pdev: PCI device information struct
7841  * @ent: entry in ixgbe_pci_tbl
7842  *
7843  * Returns 0 on success, negative on failure
7844  *
7845  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7846  * The OS initialization, configuring of the adapter private structure,
7847  * and a hardware reset occur.
7848  **/
7849 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
7850 {
7851 	struct net_device *netdev;
7852 	struct ixgbe_adapter *adapter = NULL;
7853 	struct ixgbe_hw *hw;
7854 	const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7855 	static int cards_found;
7856 	int i, err, pci_using_dac, expected_gts;
7857 	unsigned int indices = MAX_TX_QUEUES;
7858 	u8 part_str[IXGBE_PBANUM_LENGTH];
7859 #ifdef IXGBE_FCOE
7860 	u16 device_caps;
7861 #endif
7862 	u32 eec;
7863 
7864 	/* Catch broken hardware that put the wrong VF device ID in
7865 	 * the PCIe SR-IOV capability.
7866 	 */
7867 	if (pdev->is_virtfn) {
7868 		WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7869 		     pci_name(pdev), pdev->vendor, pdev->device);
7870 		return -EINVAL;
7871 	}
7872 
7873 	err = pci_enable_device_mem(pdev);
7874 	if (err)
7875 		return err;
7876 
7877 	if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
7878 		pci_using_dac = 1;
7879 	} else {
7880 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
7881 		if (err) {
7882 			dev_err(&pdev->dev,
7883 				"No usable DMA configuration, aborting\n");
7884 			goto err_dma;
7885 		}
7886 		pci_using_dac = 0;
7887 	}
7888 
7889 	err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7890 					   IORESOURCE_MEM), ixgbe_driver_name);
7891 	if (err) {
7892 		dev_err(&pdev->dev,
7893 			"pci_request_selected_regions failed 0x%x\n", err);
7894 		goto err_pci_reg;
7895 	}
7896 
7897 	pci_enable_pcie_error_reporting(pdev);
7898 
7899 	pci_set_master(pdev);
7900 	pci_save_state(pdev);
7901 
7902 	if (ii->mac == ixgbe_mac_82598EB) {
7903 #ifdef CONFIG_IXGBE_DCB
7904 		/* 8 TC w/ 4 queues per TC */
7905 		indices = 4 * MAX_TRAFFIC_CLASS;
7906 #else
7907 		indices = IXGBE_MAX_RSS_INDICES;
7908 #endif
7909 	}
7910 
7911 	netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7912 	if (!netdev) {
7913 		err = -ENOMEM;
7914 		goto err_alloc_etherdev;
7915 	}
7916 
7917 	SET_NETDEV_DEV(netdev, &pdev->dev);
7918 
7919 	adapter = netdev_priv(netdev);
7920 	pci_set_drvdata(pdev, adapter);
7921 
7922 	adapter->netdev = netdev;
7923 	adapter->pdev = pdev;
7924 	hw = &adapter->hw;
7925 	hw->back = adapter;
7926 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
7927 
7928 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7929 			      pci_resource_len(pdev, 0));
7930 	adapter->io_addr = hw->hw_addr;
7931 	if (!hw->hw_addr) {
7932 		err = -EIO;
7933 		goto err_ioremap;
7934 	}
7935 
7936 	netdev->netdev_ops = &ixgbe_netdev_ops;
7937 	ixgbe_set_ethtool_ops(netdev);
7938 	netdev->watchdog_timeo = 5 * HZ;
7939 	strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7940 
7941 	adapter->bd_number = cards_found;
7942 
7943 	/* Setup hw api */
7944 	memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7945 	hw->mac.type  = ii->mac;
7946 
7947 	/* EEPROM */
7948 	memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7949 	eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7950 	/* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7951 	if (!(eec & (1 << 8)))
7952 		hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7953 
7954 	/* PHY */
7955 	memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7956 	hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7957 	/* ixgbe_identify_phy_generic will set prtad and mmds properly */
7958 	hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7959 	hw->phy.mdio.mmds = 0;
7960 	hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7961 	hw->phy.mdio.dev = netdev;
7962 	hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7963 	hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7964 
7965 	ii->get_invariants(hw);
7966 
7967 	/* setup the private structure */
7968 	err = ixgbe_sw_init(adapter);
7969 	if (err)
7970 		goto err_sw_init;
7971 
7972 	/* Cache if MNG FW is up so we don't have to read the REG later */
7973 	if (hw->mac.ops.mng_fw_enabled)
7974 		hw->mng_fw_enabled = hw->mac.ops.mng_fw_enabled(hw);
7975 
7976 	/* Make it possible the adapter to be woken up via WOL */
7977 	switch (adapter->hw.mac.type) {
7978 	case ixgbe_mac_82599EB:
7979 	case ixgbe_mac_X540:
7980 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7981 		break;
7982 	default:
7983 		break;
7984 	}
7985 
7986 	/*
7987 	 * If there is a fan on this device and it has failed log the
7988 	 * failure.
7989 	 */
7990 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7991 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7992 		if (esdp & IXGBE_ESDP_SDP1)
7993 			e_crit(probe, "Fan has stopped, replace the adapter\n");
7994 	}
7995 
7996 	if (allow_unsupported_sfp)
7997 		hw->allow_unsupported_sfp = allow_unsupported_sfp;
7998 
7999 	/* reset_hw fills in the perm_addr as well */
8000 	hw->phy.reset_if_overtemp = true;
8001 	err = hw->mac.ops.reset_hw(hw);
8002 	hw->phy.reset_if_overtemp = false;
8003 	if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
8004 	    hw->mac.type == ixgbe_mac_82598EB) {
8005 		err = 0;
8006 	} else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
8007 		e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
8008 		e_dev_err("Reload the driver after installing a supported module.\n");
8009 		goto err_sw_init;
8010 	} else if (err) {
8011 		e_dev_err("HW Init failed: %d\n", err);
8012 		goto err_sw_init;
8013 	}
8014 
8015 #ifdef CONFIG_PCI_IOV
8016 	/* SR-IOV not supported on the 82598 */
8017 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
8018 		goto skip_sriov;
8019 	/* Mailbox */
8020 	ixgbe_init_mbx_params_pf(hw);
8021 	memcpy(&hw->mbx.ops, ii->mbx_ops, sizeof(hw->mbx.ops));
8022 	pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
8023 	ixgbe_enable_sriov(adapter);
8024 skip_sriov:
8025 
8026 #endif
8027 	netdev->features = NETIF_F_SG |
8028 			   NETIF_F_IP_CSUM |
8029 			   NETIF_F_IPV6_CSUM |
8030 			   NETIF_F_HW_VLAN_CTAG_TX |
8031 			   NETIF_F_HW_VLAN_CTAG_RX |
8032 			   NETIF_F_HW_VLAN_CTAG_FILTER |
8033 			   NETIF_F_TSO |
8034 			   NETIF_F_TSO6 |
8035 			   NETIF_F_RXHASH |
8036 			   NETIF_F_RXCSUM;
8037 
8038 	netdev->hw_features = netdev->features | NETIF_F_HW_L2FW_DOFFLOAD;
8039 
8040 	switch (adapter->hw.mac.type) {
8041 	case ixgbe_mac_82599EB:
8042 	case ixgbe_mac_X540:
8043 		netdev->features |= NETIF_F_SCTP_CSUM;
8044 		netdev->hw_features |= NETIF_F_SCTP_CSUM |
8045 				       NETIF_F_NTUPLE;
8046 		break;
8047 	default:
8048 		break;
8049 	}
8050 
8051 	netdev->hw_features |= NETIF_F_RXALL;
8052 
8053 	netdev->vlan_features |= NETIF_F_TSO;
8054 	netdev->vlan_features |= NETIF_F_TSO6;
8055 	netdev->vlan_features |= NETIF_F_IP_CSUM;
8056 	netdev->vlan_features |= NETIF_F_IPV6_CSUM;
8057 	netdev->vlan_features |= NETIF_F_SG;
8058 
8059 	netdev->priv_flags |= IFF_UNICAST_FLT;
8060 	netdev->priv_flags |= IFF_SUPP_NOFCS;
8061 
8062 #ifdef CONFIG_IXGBE_DCB
8063 	netdev->dcbnl_ops = &dcbnl_ops;
8064 #endif
8065 
8066 #ifdef IXGBE_FCOE
8067 	if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
8068 		unsigned int fcoe_l;
8069 
8070 		if (hw->mac.ops.get_device_caps) {
8071 			hw->mac.ops.get_device_caps(hw, &device_caps);
8072 			if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
8073 				adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
8074 		}
8075 
8076 
8077 		fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
8078 		adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
8079 
8080 		netdev->features |= NETIF_F_FSO |
8081 				    NETIF_F_FCOE_CRC;
8082 
8083 		netdev->vlan_features |= NETIF_F_FSO |
8084 					 NETIF_F_FCOE_CRC |
8085 					 NETIF_F_FCOE_MTU;
8086 	}
8087 #endif /* IXGBE_FCOE */
8088 	if (pci_using_dac) {
8089 		netdev->features |= NETIF_F_HIGHDMA;
8090 		netdev->vlan_features |= NETIF_F_HIGHDMA;
8091 	}
8092 
8093 	if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
8094 		netdev->hw_features |= NETIF_F_LRO;
8095 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
8096 		netdev->features |= NETIF_F_LRO;
8097 
8098 	/* make sure the EEPROM is good */
8099 	if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
8100 		e_dev_err("The EEPROM Checksum Is Not Valid\n");
8101 		err = -EIO;
8102 		goto err_sw_init;
8103 	}
8104 
8105 	memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
8106 
8107 	if (!is_valid_ether_addr(netdev->dev_addr)) {
8108 		e_dev_err("invalid MAC address\n");
8109 		err = -EIO;
8110 		goto err_sw_init;
8111 	}
8112 
8113 	setup_timer(&adapter->service_timer, &ixgbe_service_timer,
8114 		    (unsigned long) adapter);
8115 
8116 	INIT_WORK(&adapter->service_task, ixgbe_service_task);
8117 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
8118 
8119 	err = ixgbe_init_interrupt_scheme(adapter);
8120 	if (err)
8121 		goto err_sw_init;
8122 
8123 	/* WOL not supported for all devices */
8124 	adapter->wol = 0;
8125 	hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
8126 	hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
8127 						pdev->subsystem_device);
8128 	if (hw->wol_enabled)
8129 		adapter->wol = IXGBE_WUFC_MAG;
8130 
8131 	device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
8132 
8133 	/* save off EEPROM version number */
8134 	hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
8135 	hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
8136 
8137 	/* pick up the PCI bus settings for reporting later */
8138 	hw->mac.ops.get_bus_info(hw);
8139 	if (ixgbe_pcie_from_parent(hw))
8140 		ixgbe_get_parent_bus_info(adapter);
8141 
8142 	/* calculate the expected PCIe bandwidth required for optimal
8143 	 * performance. Note that some older parts will never have enough
8144 	 * bandwidth due to being older generation PCIe parts. We clamp these
8145 	 * parts to ensure no warning is displayed if it can't be fixed.
8146 	 */
8147 	switch (hw->mac.type) {
8148 	case ixgbe_mac_82598EB:
8149 		expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
8150 		break;
8151 	default:
8152 		expected_gts = ixgbe_enumerate_functions(adapter) * 10;
8153 		break;
8154 	}
8155 	ixgbe_check_minimum_link(adapter, expected_gts);
8156 
8157 	err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
8158 	if (err)
8159 		strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
8160 	if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
8161 		e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
8162 			   hw->mac.type, hw->phy.type, hw->phy.sfp_type,
8163 		           part_str);
8164 	else
8165 		e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
8166 			   hw->mac.type, hw->phy.type, part_str);
8167 
8168 	e_dev_info("%pM\n", netdev->dev_addr);
8169 
8170 	/* reset the hardware with the new settings */
8171 	err = hw->mac.ops.start_hw(hw);
8172 	if (err == IXGBE_ERR_EEPROM_VERSION) {
8173 		/* We are running on a pre-production device, log a warning */
8174 		e_dev_warn("This device is a pre-production adapter/LOM. "
8175 			   "Please be aware there may be issues associated "
8176 			   "with your hardware.  If you are experiencing "
8177 			   "problems please contact your Intel or hardware "
8178 			   "representative who provided you with this "
8179 			   "hardware.\n");
8180 	}
8181 	strcpy(netdev->name, "eth%d");
8182 	err = register_netdev(netdev);
8183 	if (err)
8184 		goto err_register;
8185 
8186 	/* power down the optics for 82599 SFP+ fiber */
8187 	if (hw->mac.ops.disable_tx_laser)
8188 		hw->mac.ops.disable_tx_laser(hw);
8189 
8190 	/* carrier off reporting is important to ethtool even BEFORE open */
8191 	netif_carrier_off(netdev);
8192 
8193 #ifdef CONFIG_IXGBE_DCA
8194 	if (dca_add_requester(&pdev->dev) == 0) {
8195 		adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
8196 		ixgbe_setup_dca(adapter);
8197 	}
8198 #endif
8199 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
8200 		e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
8201 		for (i = 0; i < adapter->num_vfs; i++)
8202 			ixgbe_vf_configuration(pdev, (i | 0x10000000));
8203 	}
8204 
8205 	/* firmware requires driver version to be 0xFFFFFFFF
8206 	 * since os does not support feature
8207 	 */
8208 	if (hw->mac.ops.set_fw_drv_ver)
8209 		hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
8210 					   0xFF);
8211 
8212 	/* add san mac addr to netdev */
8213 	ixgbe_add_sanmac_netdev(netdev);
8214 
8215 	e_dev_info("%s\n", ixgbe_default_device_descr);
8216 	cards_found++;
8217 
8218 #ifdef CONFIG_IXGBE_HWMON
8219 	if (ixgbe_sysfs_init(adapter))
8220 		e_err(probe, "failed to allocate sysfs resources\n");
8221 #endif /* CONFIG_IXGBE_HWMON */
8222 
8223 	ixgbe_dbg_adapter_init(adapter);
8224 
8225 	/* Need link setup for MNG FW, else wait for IXGBE_UP */
8226 	if (hw->mng_fw_enabled && hw->mac.ops.setup_link)
8227 		hw->mac.ops.setup_link(hw,
8228 			IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
8229 			true);
8230 
8231 	return 0;
8232 
8233 err_register:
8234 	ixgbe_release_hw_control(adapter);
8235 	ixgbe_clear_interrupt_scheme(adapter);
8236 err_sw_init:
8237 	ixgbe_disable_sriov(adapter);
8238 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
8239 	iounmap(adapter->io_addr);
8240 err_ioremap:
8241 	free_netdev(netdev);
8242 err_alloc_etherdev:
8243 	pci_release_selected_regions(pdev,
8244 				     pci_select_bars(pdev, IORESOURCE_MEM));
8245 err_pci_reg:
8246 err_dma:
8247 	pci_disable_device(pdev);
8248 	return err;
8249 }
8250 
8251 /**
8252  * ixgbe_remove - Device Removal Routine
8253  * @pdev: PCI device information struct
8254  *
8255  * ixgbe_remove is called by the PCI subsystem to alert the driver
8256  * that it should release a PCI device.  The could be caused by a
8257  * Hot-Plug event, or because the driver is going to be removed from
8258  * memory.
8259  **/
8260 static void ixgbe_remove(struct pci_dev *pdev)
8261 {
8262 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8263 	struct net_device *netdev = adapter->netdev;
8264 
8265 	ixgbe_dbg_adapter_exit(adapter);
8266 
8267 	set_bit(__IXGBE_REMOVING, &adapter->state);
8268 	cancel_work_sync(&adapter->service_task);
8269 
8270 
8271 #ifdef CONFIG_IXGBE_DCA
8272 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
8273 		adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
8274 		dca_remove_requester(&pdev->dev);
8275 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
8276 	}
8277 
8278 #endif
8279 #ifdef CONFIG_IXGBE_HWMON
8280 	ixgbe_sysfs_exit(adapter);
8281 #endif /* CONFIG_IXGBE_HWMON */
8282 
8283 	/* remove the added san mac */
8284 	ixgbe_del_sanmac_netdev(netdev);
8285 
8286 	if (netdev->reg_state == NETREG_REGISTERED)
8287 		unregister_netdev(netdev);
8288 
8289 #ifdef CONFIG_PCI_IOV
8290 	/*
8291 	 * Only disable SR-IOV on unload if the user specified the now
8292 	 * deprecated max_vfs module parameter.
8293 	 */
8294 	if (max_vfs)
8295 		ixgbe_disable_sriov(adapter);
8296 #endif
8297 	ixgbe_clear_interrupt_scheme(adapter);
8298 
8299 	ixgbe_release_hw_control(adapter);
8300 
8301 #ifdef CONFIG_DCB
8302 	kfree(adapter->ixgbe_ieee_pfc);
8303 	kfree(adapter->ixgbe_ieee_ets);
8304 
8305 #endif
8306 	iounmap(adapter->io_addr);
8307 	pci_release_selected_regions(pdev, pci_select_bars(pdev,
8308 				     IORESOURCE_MEM));
8309 
8310 	e_dev_info("complete\n");
8311 
8312 	free_netdev(netdev);
8313 
8314 	pci_disable_pcie_error_reporting(pdev);
8315 
8316 	pci_disable_device(pdev);
8317 }
8318 
8319 /**
8320  * ixgbe_io_error_detected - called when PCI error is detected
8321  * @pdev: Pointer to PCI device
8322  * @state: The current pci connection state
8323  *
8324  * This function is called after a PCI bus error affecting
8325  * this device has been detected.
8326  */
8327 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
8328 						pci_channel_state_t state)
8329 {
8330 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8331 	struct net_device *netdev = adapter->netdev;
8332 
8333 #ifdef CONFIG_PCI_IOV
8334 	struct pci_dev *bdev, *vfdev;
8335 	u32 dw0, dw1, dw2, dw3;
8336 	int vf, pos;
8337 	u16 req_id, pf_func;
8338 
8339 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
8340 	    adapter->num_vfs == 0)
8341 		goto skip_bad_vf_detection;
8342 
8343 	bdev = pdev->bus->self;
8344 	while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
8345 		bdev = bdev->bus->self;
8346 
8347 	if (!bdev)
8348 		goto skip_bad_vf_detection;
8349 
8350 	pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
8351 	if (!pos)
8352 		goto skip_bad_vf_detection;
8353 
8354 	pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG, &dw0);
8355 	pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 4, &dw1);
8356 	pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 8, &dw2);
8357 	pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 12, &dw3);
8358 
8359 	req_id = dw1 >> 16;
8360 	/* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
8361 	if (!(req_id & 0x0080))
8362 		goto skip_bad_vf_detection;
8363 
8364 	pf_func = req_id & 0x01;
8365 	if ((pf_func & 1) == (pdev->devfn & 1)) {
8366 		unsigned int device_id;
8367 
8368 		vf = (req_id & 0x7F) >> 1;
8369 		e_dev_err("VF %d has caused a PCIe error\n", vf);
8370 		e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
8371 				"%8.8x\tdw3: %8.8x\n",
8372 		dw0, dw1, dw2, dw3);
8373 		switch (adapter->hw.mac.type) {
8374 		case ixgbe_mac_82599EB:
8375 			device_id = IXGBE_82599_VF_DEVICE_ID;
8376 			break;
8377 		case ixgbe_mac_X540:
8378 			device_id = IXGBE_X540_VF_DEVICE_ID;
8379 			break;
8380 		default:
8381 			device_id = 0;
8382 			break;
8383 		}
8384 
8385 		/* Find the pci device of the offending VF */
8386 		vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
8387 		while (vfdev) {
8388 			if (vfdev->devfn == (req_id & 0xFF))
8389 				break;
8390 			vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
8391 					       device_id, vfdev);
8392 		}
8393 		/*
8394 		 * There's a slim chance the VF could have been hot plugged,
8395 		 * so if it is no longer present we don't need to issue the
8396 		 * VFLR.  Just clean up the AER in that case.
8397 		 */
8398 		if (vfdev) {
8399 			e_dev_err("Issuing VFLR to VF %d\n", vf);
8400 			pci_write_config_dword(vfdev, 0xA8, 0x00008000);
8401 			/* Free device reference count */
8402 			pci_dev_put(vfdev);
8403 		}
8404 
8405 		pci_cleanup_aer_uncorrect_error_status(pdev);
8406 	}
8407 
8408 	/*
8409 	 * Even though the error may have occurred on the other port
8410 	 * we still need to increment the vf error reference count for
8411 	 * both ports because the I/O resume function will be called
8412 	 * for both of them.
8413 	 */
8414 	adapter->vferr_refcount++;
8415 
8416 	return PCI_ERS_RESULT_RECOVERED;
8417 
8418 skip_bad_vf_detection:
8419 #endif /* CONFIG_PCI_IOV */
8420 	netif_device_detach(netdev);
8421 
8422 	if (state == pci_channel_io_perm_failure)
8423 		return PCI_ERS_RESULT_DISCONNECT;
8424 
8425 	if (netif_running(netdev))
8426 		ixgbe_down(adapter);
8427 	pci_disable_device(pdev);
8428 
8429 	/* Request a slot reset. */
8430 	return PCI_ERS_RESULT_NEED_RESET;
8431 }
8432 
8433 /**
8434  * ixgbe_io_slot_reset - called after the pci bus has been reset.
8435  * @pdev: Pointer to PCI device
8436  *
8437  * Restart the card from scratch, as if from a cold-boot.
8438  */
8439 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
8440 {
8441 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8442 	pci_ers_result_t result;
8443 	int err;
8444 
8445 	if (pci_enable_device_mem(pdev)) {
8446 		e_err(probe, "Cannot re-enable PCI device after reset.\n");
8447 		result = PCI_ERS_RESULT_DISCONNECT;
8448 	} else {
8449 		pci_set_master(pdev);
8450 		pci_restore_state(pdev);
8451 		pci_save_state(pdev);
8452 
8453 		pci_wake_from_d3(pdev, false);
8454 
8455 		ixgbe_reset(adapter);
8456 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
8457 		result = PCI_ERS_RESULT_RECOVERED;
8458 	}
8459 
8460 	err = pci_cleanup_aer_uncorrect_error_status(pdev);
8461 	if (err) {
8462 		e_dev_err("pci_cleanup_aer_uncorrect_error_status "
8463 			  "failed 0x%0x\n", err);
8464 		/* non-fatal, continue */
8465 	}
8466 
8467 	return result;
8468 }
8469 
8470 /**
8471  * ixgbe_io_resume - called when traffic can start flowing again.
8472  * @pdev: Pointer to PCI device
8473  *
8474  * This callback is called when the error recovery driver tells us that
8475  * its OK to resume normal operation.
8476  */
8477 static void ixgbe_io_resume(struct pci_dev *pdev)
8478 {
8479 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8480 	struct net_device *netdev = adapter->netdev;
8481 
8482 #ifdef CONFIG_PCI_IOV
8483 	if (adapter->vferr_refcount) {
8484 		e_info(drv, "Resuming after VF err\n");
8485 		adapter->vferr_refcount--;
8486 		return;
8487 	}
8488 
8489 #endif
8490 	if (netif_running(netdev))
8491 		ixgbe_up(adapter);
8492 
8493 	netif_device_attach(netdev);
8494 }
8495 
8496 static const struct pci_error_handlers ixgbe_err_handler = {
8497 	.error_detected = ixgbe_io_error_detected,
8498 	.slot_reset = ixgbe_io_slot_reset,
8499 	.resume = ixgbe_io_resume,
8500 };
8501 
8502 static struct pci_driver ixgbe_driver = {
8503 	.name     = ixgbe_driver_name,
8504 	.id_table = ixgbe_pci_tbl,
8505 	.probe    = ixgbe_probe,
8506 	.remove   = ixgbe_remove,
8507 #ifdef CONFIG_PM
8508 	.suspend  = ixgbe_suspend,
8509 	.resume   = ixgbe_resume,
8510 #endif
8511 	.shutdown = ixgbe_shutdown,
8512 	.sriov_configure = ixgbe_pci_sriov_configure,
8513 	.err_handler = &ixgbe_err_handler
8514 };
8515 
8516 /**
8517  * ixgbe_init_module - Driver Registration Routine
8518  *
8519  * ixgbe_init_module is the first routine called when the driver is
8520  * loaded. All it does is register with the PCI subsystem.
8521  **/
8522 static int __init ixgbe_init_module(void)
8523 {
8524 	int ret;
8525 	pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
8526 	pr_info("%s\n", ixgbe_copyright);
8527 
8528 	ixgbe_dbg_init();
8529 
8530 	ret = pci_register_driver(&ixgbe_driver);
8531 	if (ret) {
8532 		ixgbe_dbg_exit();
8533 		return ret;
8534 	}
8535 
8536 #ifdef CONFIG_IXGBE_DCA
8537 	dca_register_notify(&dca_notifier);
8538 #endif
8539 
8540 	return 0;
8541 }
8542 
8543 module_init(ixgbe_init_module);
8544 
8545 /**
8546  * ixgbe_exit_module - Driver Exit Cleanup Routine
8547  *
8548  * ixgbe_exit_module is called just before the driver is removed
8549  * from memory.
8550  **/
8551 static void __exit ixgbe_exit_module(void)
8552 {
8553 #ifdef CONFIG_IXGBE_DCA
8554 	dca_unregister_notify(&dca_notifier);
8555 #endif
8556 	pci_unregister_driver(&ixgbe_driver);
8557 
8558 	ixgbe_dbg_exit();
8559 
8560 	rcu_barrier(); /* Wait for completion of call_rcu()'s */
8561 }
8562 
8563 #ifdef CONFIG_IXGBE_DCA
8564 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
8565 			    void *p)
8566 {
8567 	int ret_val;
8568 
8569 	ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
8570 					 __ixgbe_notify_dca);
8571 
8572 	return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
8573 }
8574 
8575 #endif /* CONFIG_IXGBE_DCA */
8576 
8577 module_exit(ixgbe_exit_module);
8578 
8579 /* ixgbe_main.c */
8580