1 /*******************************************************************************
2 
3   Intel 10 Gigabit PCI Express Linux driver
4   Copyright(c) 1999 - 2014 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   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 
27 *******************************************************************************/
28 
29 #include <linux/types.h>
30 #include <linux/module.h>
31 #include <linux/pci.h>
32 #include <linux/netdevice.h>
33 #include <linux/vmalloc.h>
34 #include <linux/string.h>
35 #include <linux/in.h>
36 #include <linux/interrupt.h>
37 #include <linux/ip.h>
38 #include <linux/tcp.h>
39 #include <linux/sctp.h>
40 #include <linux/pkt_sched.h>
41 #include <linux/ipv6.h>
42 #include <linux/slab.h>
43 #include <net/checksum.h>
44 #include <net/ip6_checksum.h>
45 #include <linux/ethtool.h>
46 #include <linux/if.h>
47 #include <linux/if_vlan.h>
48 #include <linux/if_macvlan.h>
49 #include <linux/if_bridge.h>
50 #include <linux/prefetch.h>
51 #include <scsi/fc/fc_fcoe.h>
52 
53 #include "ixgbe.h"
54 #include "ixgbe_common.h"
55 #include "ixgbe_dcb_82599.h"
56 #include "ixgbe_sriov.h"
57 
58 char ixgbe_driver_name[] = "ixgbe";
59 static const char ixgbe_driver_string[] =
60 			      "Intel(R) 10 Gigabit PCI Express Network Driver";
61 #ifdef IXGBE_FCOE
62 char ixgbe_default_device_descr[] =
63 			      "Intel(R) 10 Gigabit Network Connection";
64 #else
65 static char ixgbe_default_device_descr[] =
66 			      "Intel(R) 10 Gigabit Network Connection";
67 #endif
68 #define DRV_VERSION "3.19.1-k"
69 const char ixgbe_driver_version[] = DRV_VERSION;
70 static const char ixgbe_copyright[] =
71 				"Copyright (c) 1999-2014 Intel Corporation.";
72 
73 static const struct ixgbe_info *ixgbe_info_tbl[] = {
74 	[board_82598] = &ixgbe_82598_info,
75 	[board_82599] = &ixgbe_82599_info,
76 	[board_X540] = &ixgbe_X540_info,
77 };
78 
79 /* ixgbe_pci_tbl - PCI Device ID Table
80  *
81  * Wildcard entries (PCI_ANY_ID) should come last
82  * Last entry must be all 0s
83  *
84  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
85  *   Class, Class Mask, private data (not used) }
86  */
87 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
88 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
89 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
90 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
91 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
92 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
93 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
94 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
95 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
96 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
97 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
98 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
99 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
100 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
101 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
102 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
103 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
104 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
105 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
106 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
107 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
108 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
109 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
110 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
111 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
112 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
113 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
114 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_QSFP_SF_QP), board_82599 },
115 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
116 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
117 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
118 	/* required last entry */
119 	{0, }
120 };
121 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
122 
123 #ifdef CONFIG_IXGBE_DCA
124 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
125 			    void *p);
126 static struct notifier_block dca_notifier = {
127 	.notifier_call = ixgbe_notify_dca,
128 	.next          = NULL,
129 	.priority      = 0
130 };
131 #endif
132 
133 #ifdef CONFIG_PCI_IOV
134 static unsigned int max_vfs;
135 module_param(max_vfs, uint, 0);
136 MODULE_PARM_DESC(max_vfs,
137 		 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63. (Deprecated)");
138 #endif /* CONFIG_PCI_IOV */
139 
140 static unsigned int allow_unsupported_sfp;
141 module_param(allow_unsupported_sfp, uint, 0);
142 MODULE_PARM_DESC(allow_unsupported_sfp,
143 		 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
144 
145 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
146 static int debug = -1;
147 module_param(debug, int, 0);
148 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
149 
150 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
151 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
152 MODULE_LICENSE("GPL");
153 MODULE_VERSION(DRV_VERSION);
154 
155 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev);
156 
157 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
158 					  u32 reg, u16 *value)
159 {
160 	struct pci_dev *parent_dev;
161 	struct pci_bus *parent_bus;
162 
163 	parent_bus = adapter->pdev->bus->parent;
164 	if (!parent_bus)
165 		return -1;
166 
167 	parent_dev = parent_bus->self;
168 	if (!parent_dev)
169 		return -1;
170 
171 	if (!pci_is_pcie(parent_dev))
172 		return -1;
173 
174 	pcie_capability_read_word(parent_dev, reg, value);
175 	if (*value == IXGBE_FAILED_READ_CFG_WORD &&
176 	    ixgbe_check_cfg_remove(&adapter->hw, parent_dev))
177 		return -1;
178 	return 0;
179 }
180 
181 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
182 {
183 	struct ixgbe_hw *hw = &adapter->hw;
184 	u16 link_status = 0;
185 	int err;
186 
187 	hw->bus.type = ixgbe_bus_type_pci_express;
188 
189 	/* Get the negotiated link width and speed from PCI config space of the
190 	 * parent, as this device is behind a switch
191 	 */
192 	err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
193 
194 	/* assume caller will handle error case */
195 	if (err)
196 		return err;
197 
198 	hw->bus.width = ixgbe_convert_bus_width(link_status);
199 	hw->bus.speed = ixgbe_convert_bus_speed(link_status);
200 
201 	return 0;
202 }
203 
204 /**
205  * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
206  * @hw: hw specific details
207  *
208  * This function is used by probe to determine whether a device's PCI-Express
209  * bandwidth details should be gathered from the parent bus instead of from the
210  * device. Used to ensure that various locations all have the correct device ID
211  * checks.
212  */
213 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
214 {
215 	switch (hw->device_id) {
216 	case IXGBE_DEV_ID_82599_SFP_SF_QP:
217 	case IXGBE_DEV_ID_82599_QSFP_SF_QP:
218 		return true;
219 	default:
220 		return false;
221 	}
222 }
223 
224 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
225 				     int expected_gts)
226 {
227 	int max_gts = 0;
228 	enum pci_bus_speed speed = PCI_SPEED_UNKNOWN;
229 	enum pcie_link_width width = PCIE_LNK_WIDTH_UNKNOWN;
230 	struct pci_dev *pdev;
231 
232 	/* determine whether to use the the parent device
233 	 */
234 	if (ixgbe_pcie_from_parent(&adapter->hw))
235 		pdev = adapter->pdev->bus->parent->self;
236 	else
237 		pdev = adapter->pdev;
238 
239 	if (pcie_get_minimum_link(pdev, &speed, &width) ||
240 	    speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN) {
241 		e_dev_warn("Unable to determine PCI Express bandwidth.\n");
242 		return;
243 	}
244 
245 	switch (speed) {
246 	case PCIE_SPEED_2_5GT:
247 		/* 8b/10b encoding reduces max throughput by 20% */
248 		max_gts = 2 * width;
249 		break;
250 	case PCIE_SPEED_5_0GT:
251 		/* 8b/10b encoding reduces max throughput by 20% */
252 		max_gts = 4 * width;
253 		break;
254 	case PCIE_SPEED_8_0GT:
255 		/* 128b/130b encoding reduces throughput by less than 2% */
256 		max_gts = 8 * width;
257 		break;
258 	default:
259 		e_dev_warn("Unable to determine PCI Express bandwidth.\n");
260 		return;
261 	}
262 
263 	e_dev_info("PCI Express bandwidth of %dGT/s available\n",
264 		   max_gts);
265 	e_dev_info("(Speed:%s, Width: x%d, Encoding Loss:%s)\n",
266 		   (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" :
267 		    speed == PCIE_SPEED_5_0GT ? "5.0GT/s" :
268 		    speed == PCIE_SPEED_2_5GT ? "2.5GT/s" :
269 		    "Unknown"),
270 		   width,
271 		   (speed == PCIE_SPEED_2_5GT ? "20%" :
272 		    speed == PCIE_SPEED_5_0GT ? "20%" :
273 		    speed == PCIE_SPEED_8_0GT ? "<2%" :
274 		    "Unknown"));
275 
276 	if (max_gts < expected_gts) {
277 		e_dev_warn("This is not sufficient for optimal performance of this card.\n");
278 		e_dev_warn("For optimal performance, at least %dGT/s of bandwidth is required.\n",
279 			expected_gts);
280 		e_dev_warn("A slot with more lanes and/or higher speed is suggested.\n");
281 	}
282 }
283 
284 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
285 {
286 	if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
287 	    !test_bit(__IXGBE_REMOVING, &adapter->state) &&
288 	    !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
289 		schedule_work(&adapter->service_task);
290 }
291 
292 static void ixgbe_remove_adapter(struct ixgbe_hw *hw)
293 {
294 	struct ixgbe_adapter *adapter = hw->back;
295 
296 	if (!hw->hw_addr)
297 		return;
298 	hw->hw_addr = NULL;
299 	e_dev_err("Adapter removed\n");
300 	if (test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
301 		ixgbe_service_event_schedule(adapter);
302 }
303 
304 void ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg)
305 {
306 	u32 value;
307 
308 	/* The following check not only optimizes a bit by not
309 	 * performing a read on the status register when the
310 	 * register just read was a status register read that
311 	 * returned IXGBE_FAILED_READ_REG. It also blocks any
312 	 * potential recursion.
313 	 */
314 	if (reg == IXGBE_STATUS) {
315 		ixgbe_remove_adapter(hw);
316 		return;
317 	}
318 	value = ixgbe_read_reg(hw, IXGBE_STATUS);
319 	if (value == IXGBE_FAILED_READ_REG)
320 		ixgbe_remove_adapter(hw);
321 }
322 
323 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev)
324 {
325 	u16 value;
326 
327 	pci_read_config_word(pdev, PCI_VENDOR_ID, &value);
328 	if (value == IXGBE_FAILED_READ_CFG_WORD) {
329 		ixgbe_remove_adapter(hw);
330 		return true;
331 	}
332 	return false;
333 }
334 
335 u16 ixgbe_read_pci_cfg_word(struct ixgbe_hw *hw, u32 reg)
336 {
337 	struct ixgbe_adapter *adapter = hw->back;
338 	u16 value;
339 
340 	if (ixgbe_removed(hw->hw_addr))
341 		return IXGBE_FAILED_READ_CFG_WORD;
342 	pci_read_config_word(adapter->pdev, reg, &value);
343 	if (value == IXGBE_FAILED_READ_CFG_WORD &&
344 	    ixgbe_check_cfg_remove(hw, adapter->pdev))
345 		return IXGBE_FAILED_READ_CFG_WORD;
346 	return value;
347 }
348 
349 #ifdef CONFIG_PCI_IOV
350 static u32 ixgbe_read_pci_cfg_dword(struct ixgbe_hw *hw, u32 reg)
351 {
352 	struct ixgbe_adapter *adapter = hw->back;
353 	u32 value;
354 
355 	if (ixgbe_removed(hw->hw_addr))
356 		return IXGBE_FAILED_READ_CFG_DWORD;
357 	pci_read_config_dword(adapter->pdev, reg, &value);
358 	if (value == IXGBE_FAILED_READ_CFG_DWORD &&
359 	    ixgbe_check_cfg_remove(hw, adapter->pdev))
360 		return IXGBE_FAILED_READ_CFG_DWORD;
361 	return value;
362 }
363 #endif /* CONFIG_PCI_IOV */
364 
365 void ixgbe_write_pci_cfg_word(struct ixgbe_hw *hw, u32 reg, u16 value)
366 {
367 	struct ixgbe_adapter *adapter = hw->back;
368 
369 	if (ixgbe_removed(hw->hw_addr))
370 		return;
371 	pci_write_config_word(adapter->pdev, reg, value);
372 }
373 
374 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
375 {
376 	BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
377 
378 	/* flush memory to make sure state is correct before next watchdog */
379 	smp_mb__before_atomic();
380 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
381 }
382 
383 struct ixgbe_reg_info {
384 	u32 ofs;
385 	char *name;
386 };
387 
388 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
389 
390 	/* General Registers */
391 	{IXGBE_CTRL, "CTRL"},
392 	{IXGBE_STATUS, "STATUS"},
393 	{IXGBE_CTRL_EXT, "CTRL_EXT"},
394 
395 	/* Interrupt Registers */
396 	{IXGBE_EICR, "EICR"},
397 
398 	/* RX Registers */
399 	{IXGBE_SRRCTL(0), "SRRCTL"},
400 	{IXGBE_DCA_RXCTRL(0), "DRXCTL"},
401 	{IXGBE_RDLEN(0), "RDLEN"},
402 	{IXGBE_RDH(0), "RDH"},
403 	{IXGBE_RDT(0), "RDT"},
404 	{IXGBE_RXDCTL(0), "RXDCTL"},
405 	{IXGBE_RDBAL(0), "RDBAL"},
406 	{IXGBE_RDBAH(0), "RDBAH"},
407 
408 	/* TX Registers */
409 	{IXGBE_TDBAL(0), "TDBAL"},
410 	{IXGBE_TDBAH(0), "TDBAH"},
411 	{IXGBE_TDLEN(0), "TDLEN"},
412 	{IXGBE_TDH(0), "TDH"},
413 	{IXGBE_TDT(0), "TDT"},
414 	{IXGBE_TXDCTL(0), "TXDCTL"},
415 
416 	/* List Terminator */
417 	{}
418 };
419 
420 
421 /*
422  * ixgbe_regdump - register printout routine
423  */
424 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
425 {
426 	int i = 0, j = 0;
427 	char rname[16];
428 	u32 regs[64];
429 
430 	switch (reginfo->ofs) {
431 	case IXGBE_SRRCTL(0):
432 		for (i = 0; i < 64; i++)
433 			regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
434 		break;
435 	case IXGBE_DCA_RXCTRL(0):
436 		for (i = 0; i < 64; i++)
437 			regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
438 		break;
439 	case IXGBE_RDLEN(0):
440 		for (i = 0; i < 64; i++)
441 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
442 		break;
443 	case IXGBE_RDH(0):
444 		for (i = 0; i < 64; i++)
445 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
446 		break;
447 	case IXGBE_RDT(0):
448 		for (i = 0; i < 64; i++)
449 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
450 		break;
451 	case IXGBE_RXDCTL(0):
452 		for (i = 0; i < 64; i++)
453 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
454 		break;
455 	case IXGBE_RDBAL(0):
456 		for (i = 0; i < 64; i++)
457 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
458 		break;
459 	case IXGBE_RDBAH(0):
460 		for (i = 0; i < 64; i++)
461 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
462 		break;
463 	case IXGBE_TDBAL(0):
464 		for (i = 0; i < 64; i++)
465 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
466 		break;
467 	case IXGBE_TDBAH(0):
468 		for (i = 0; i < 64; i++)
469 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
470 		break;
471 	case IXGBE_TDLEN(0):
472 		for (i = 0; i < 64; i++)
473 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
474 		break;
475 	case IXGBE_TDH(0):
476 		for (i = 0; i < 64; i++)
477 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
478 		break;
479 	case IXGBE_TDT(0):
480 		for (i = 0; i < 64; i++)
481 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
482 		break;
483 	case IXGBE_TXDCTL(0):
484 		for (i = 0; i < 64; i++)
485 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
486 		break;
487 	default:
488 		pr_info("%-15s %08x\n", reginfo->name,
489 			IXGBE_READ_REG(hw, reginfo->ofs));
490 		return;
491 	}
492 
493 	for (i = 0; i < 8; i++) {
494 		snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
495 		pr_err("%-15s", rname);
496 		for (j = 0; j < 8; j++)
497 			pr_cont(" %08x", regs[i*8+j]);
498 		pr_cont("\n");
499 	}
500 
501 }
502 
503 /*
504  * ixgbe_dump - Print registers, tx-rings and rx-rings
505  */
506 static void ixgbe_dump(struct ixgbe_adapter *adapter)
507 {
508 	struct net_device *netdev = adapter->netdev;
509 	struct ixgbe_hw *hw = &adapter->hw;
510 	struct ixgbe_reg_info *reginfo;
511 	int n = 0;
512 	struct ixgbe_ring *tx_ring;
513 	struct ixgbe_tx_buffer *tx_buffer;
514 	union ixgbe_adv_tx_desc *tx_desc;
515 	struct my_u0 { u64 a; u64 b; } *u0;
516 	struct ixgbe_ring *rx_ring;
517 	union ixgbe_adv_rx_desc *rx_desc;
518 	struct ixgbe_rx_buffer *rx_buffer_info;
519 	u32 staterr;
520 	int i = 0;
521 
522 	if (!netif_msg_hw(adapter))
523 		return;
524 
525 	/* Print netdevice Info */
526 	if (netdev) {
527 		dev_info(&adapter->pdev->dev, "Net device Info\n");
528 		pr_info("Device Name     state            "
529 			"trans_start      last_rx\n");
530 		pr_info("%-15s %016lX %016lX %016lX\n",
531 			netdev->name,
532 			netdev->state,
533 			netdev->trans_start,
534 			netdev->last_rx);
535 	}
536 
537 	/* Print Registers */
538 	dev_info(&adapter->pdev->dev, "Register Dump\n");
539 	pr_info(" Register Name   Value\n");
540 	for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
541 	     reginfo->name; reginfo++) {
542 		ixgbe_regdump(hw, reginfo);
543 	}
544 
545 	/* Print TX Ring Summary */
546 	if (!netdev || !netif_running(netdev))
547 		goto exit;
548 
549 	dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
550 	pr_info(" %s     %s              %s        %s\n",
551 		"Queue [NTU] [NTC] [bi(ntc)->dma  ]",
552 		"leng", "ntw", "timestamp");
553 	for (n = 0; n < adapter->num_tx_queues; n++) {
554 		tx_ring = adapter->tx_ring[n];
555 		tx_buffer = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
556 		pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
557 			   n, tx_ring->next_to_use, tx_ring->next_to_clean,
558 			   (u64)dma_unmap_addr(tx_buffer, dma),
559 			   dma_unmap_len(tx_buffer, len),
560 			   tx_buffer->next_to_watch,
561 			   (u64)tx_buffer->time_stamp);
562 	}
563 
564 	/* Print TX Rings */
565 	if (!netif_msg_tx_done(adapter))
566 		goto rx_ring_summary;
567 
568 	dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
569 
570 	/* Transmit Descriptor Formats
571 	 *
572 	 * 82598 Advanced Transmit Descriptor
573 	 *   +--------------------------------------------------------------+
574 	 * 0 |         Buffer Address [63:0]                                |
575 	 *   +--------------------------------------------------------------+
576 	 * 8 |  PAYLEN  | POPTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
577 	 *   +--------------------------------------------------------------+
578 	 *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
579 	 *
580 	 * 82598 Advanced Transmit Descriptor (Write-Back Format)
581 	 *   +--------------------------------------------------------------+
582 	 * 0 |                          RSV [63:0]                          |
583 	 *   +--------------------------------------------------------------+
584 	 * 8 |            RSV           |  STA  |          NXTSEQ           |
585 	 *   +--------------------------------------------------------------+
586 	 *   63                       36 35   32 31                         0
587 	 *
588 	 * 82599+ Advanced Transmit Descriptor
589 	 *   +--------------------------------------------------------------+
590 	 * 0 |         Buffer Address [63:0]                                |
591 	 *   +--------------------------------------------------------------+
592 	 * 8 |PAYLEN  |POPTS|CC|IDX  |STA  |DCMD  |DTYP |MAC  |RSV  |DTALEN |
593 	 *   +--------------------------------------------------------------+
594 	 *   63     46 45 40 39 38 36 35 32 31  24 23 20 19 18 17 16 15     0
595 	 *
596 	 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
597 	 *   +--------------------------------------------------------------+
598 	 * 0 |                          RSV [63:0]                          |
599 	 *   +--------------------------------------------------------------+
600 	 * 8 |            RSV           |  STA  |           RSV             |
601 	 *   +--------------------------------------------------------------+
602 	 *   63                       36 35   32 31                         0
603 	 */
604 
605 	for (n = 0; n < adapter->num_tx_queues; n++) {
606 		tx_ring = adapter->tx_ring[n];
607 		pr_info("------------------------------------\n");
608 		pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
609 		pr_info("------------------------------------\n");
610 		pr_info("%s%s    %s              %s        %s          %s\n",
611 			"T [desc]     [address 63:0  ] ",
612 			"[PlPOIdStDDt Ln] [bi->dma       ] ",
613 			"leng", "ntw", "timestamp", "bi->skb");
614 
615 		for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
616 			tx_desc = IXGBE_TX_DESC(tx_ring, i);
617 			tx_buffer = &tx_ring->tx_buffer_info[i];
618 			u0 = (struct my_u0 *)tx_desc;
619 			if (dma_unmap_len(tx_buffer, len) > 0) {
620 				pr_info("T [0x%03X]    %016llX %016llX %016llX %08X %p %016llX %p",
621 					i,
622 					le64_to_cpu(u0->a),
623 					le64_to_cpu(u0->b),
624 					(u64)dma_unmap_addr(tx_buffer, dma),
625 					dma_unmap_len(tx_buffer, len),
626 					tx_buffer->next_to_watch,
627 					(u64)tx_buffer->time_stamp,
628 					tx_buffer->skb);
629 				if (i == tx_ring->next_to_use &&
630 					i == tx_ring->next_to_clean)
631 					pr_cont(" NTC/U\n");
632 				else if (i == tx_ring->next_to_use)
633 					pr_cont(" NTU\n");
634 				else if (i == tx_ring->next_to_clean)
635 					pr_cont(" NTC\n");
636 				else
637 					pr_cont("\n");
638 
639 				if (netif_msg_pktdata(adapter) &&
640 				    tx_buffer->skb)
641 					print_hex_dump(KERN_INFO, "",
642 						DUMP_PREFIX_ADDRESS, 16, 1,
643 						tx_buffer->skb->data,
644 						dma_unmap_len(tx_buffer, len),
645 						true);
646 			}
647 		}
648 	}
649 
650 	/* Print RX Rings Summary */
651 rx_ring_summary:
652 	dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
653 	pr_info("Queue [NTU] [NTC]\n");
654 	for (n = 0; n < adapter->num_rx_queues; n++) {
655 		rx_ring = adapter->rx_ring[n];
656 		pr_info("%5d %5X %5X\n",
657 			n, rx_ring->next_to_use, rx_ring->next_to_clean);
658 	}
659 
660 	/* Print RX Rings */
661 	if (!netif_msg_rx_status(adapter))
662 		goto exit;
663 
664 	dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
665 
666 	/* Receive Descriptor Formats
667 	 *
668 	 * 82598 Advanced Receive Descriptor (Read) Format
669 	 *    63                                           1        0
670 	 *    +-----------------------------------------------------+
671 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
672 	 *    +----------------------------------------------+------+
673 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
674 	 *    +-----------------------------------------------------+
675 	 *
676 	 *
677 	 * 82598 Advanced Receive Descriptor (Write-Back) Format
678 	 *
679 	 *   63       48 47    32 31  30      21 20 16 15   4 3     0
680 	 *   +------------------------------------------------------+
681 	 * 0 |       RSS Hash /  |SPH| HDR_LEN  | RSV |Packet|  RSS |
682 	 *   | Packet   | IP     |   |          |     | Type | Type |
683 	 *   | Checksum | Ident  |   |          |     |      |      |
684 	 *   +------------------------------------------------------+
685 	 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
686 	 *   +------------------------------------------------------+
687 	 *   63       48 47    32 31            20 19               0
688 	 *
689 	 * 82599+ Advanced Receive Descriptor (Read) Format
690 	 *    63                                           1        0
691 	 *    +-----------------------------------------------------+
692 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
693 	 *    +----------------------------------------------+------+
694 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
695 	 *    +-----------------------------------------------------+
696 	 *
697 	 *
698 	 * 82599+ Advanced Receive Descriptor (Write-Back) Format
699 	 *
700 	 *   63       48 47    32 31  30      21 20 17 16   4 3     0
701 	 *   +------------------------------------------------------+
702 	 * 0 |RSS / Frag Checksum|SPH| HDR_LEN  |RSC- |Packet|  RSS |
703 	 *   |/ RTT / PCoE_PARAM |   |          | CNT | Type | Type |
704 	 *   |/ Flow Dir Flt ID  |   |          |     |      |      |
705 	 *   +------------------------------------------------------+
706 	 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
707 	 *   +------------------------------------------------------+
708 	 *   63       48 47    32 31          20 19                 0
709 	 */
710 
711 	for (n = 0; n < adapter->num_rx_queues; n++) {
712 		rx_ring = adapter->rx_ring[n];
713 		pr_info("------------------------------------\n");
714 		pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
715 		pr_info("------------------------------------\n");
716 		pr_info("%s%s%s",
717 			"R  [desc]      [ PktBuf     A0] ",
718 			"[  HeadBuf   DD] [bi->dma       ] [bi->skb       ] ",
719 			"<-- Adv Rx Read format\n");
720 		pr_info("%s%s%s",
721 			"RWB[desc]      [PcsmIpSHl PtRs] ",
722 			"[vl er S cks ln] ---------------- [bi->skb       ] ",
723 			"<-- Adv Rx Write-Back format\n");
724 
725 		for (i = 0; i < rx_ring->count; i++) {
726 			rx_buffer_info = &rx_ring->rx_buffer_info[i];
727 			rx_desc = IXGBE_RX_DESC(rx_ring, i);
728 			u0 = (struct my_u0 *)rx_desc;
729 			staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
730 			if (staterr & IXGBE_RXD_STAT_DD) {
731 				/* Descriptor Done */
732 				pr_info("RWB[0x%03X]     %016llX "
733 					"%016llX ---------------- %p", i,
734 					le64_to_cpu(u0->a),
735 					le64_to_cpu(u0->b),
736 					rx_buffer_info->skb);
737 			} else {
738 				pr_info("R  [0x%03X]     %016llX "
739 					"%016llX %016llX %p", i,
740 					le64_to_cpu(u0->a),
741 					le64_to_cpu(u0->b),
742 					(u64)rx_buffer_info->dma,
743 					rx_buffer_info->skb);
744 
745 				if (netif_msg_pktdata(adapter) &&
746 				    rx_buffer_info->dma) {
747 					print_hex_dump(KERN_INFO, "",
748 					   DUMP_PREFIX_ADDRESS, 16, 1,
749 					   page_address(rx_buffer_info->page) +
750 						    rx_buffer_info->page_offset,
751 					   ixgbe_rx_bufsz(rx_ring), true);
752 				}
753 			}
754 
755 			if (i == rx_ring->next_to_use)
756 				pr_cont(" NTU\n");
757 			else if (i == rx_ring->next_to_clean)
758 				pr_cont(" NTC\n");
759 			else
760 				pr_cont("\n");
761 
762 		}
763 	}
764 
765 exit:
766 	return;
767 }
768 
769 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
770 {
771 	u32 ctrl_ext;
772 
773 	/* Let firmware take over control of h/w */
774 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
775 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
776 			ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
777 }
778 
779 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
780 {
781 	u32 ctrl_ext;
782 
783 	/* Let firmware know the driver has taken over */
784 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
785 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
786 			ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
787 }
788 
789 /**
790  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
791  * @adapter: pointer to adapter struct
792  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
793  * @queue: queue to map the corresponding interrupt to
794  * @msix_vector: the vector to map to the corresponding queue
795  *
796  */
797 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
798 			   u8 queue, u8 msix_vector)
799 {
800 	u32 ivar, index;
801 	struct ixgbe_hw *hw = &adapter->hw;
802 	switch (hw->mac.type) {
803 	case ixgbe_mac_82598EB:
804 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
805 		if (direction == -1)
806 			direction = 0;
807 		index = (((direction * 64) + queue) >> 2) & 0x1F;
808 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
809 		ivar &= ~(0xFF << (8 * (queue & 0x3)));
810 		ivar |= (msix_vector << (8 * (queue & 0x3)));
811 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
812 		break;
813 	case ixgbe_mac_82599EB:
814 	case ixgbe_mac_X540:
815 		if (direction == -1) {
816 			/* other causes */
817 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
818 			index = ((queue & 1) * 8);
819 			ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
820 			ivar &= ~(0xFF << index);
821 			ivar |= (msix_vector << index);
822 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
823 			break;
824 		} else {
825 			/* tx or rx causes */
826 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
827 			index = ((16 * (queue & 1)) + (8 * direction));
828 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
829 			ivar &= ~(0xFF << index);
830 			ivar |= (msix_vector << index);
831 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
832 			break;
833 		}
834 	default:
835 		break;
836 	}
837 }
838 
839 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
840 					  u64 qmask)
841 {
842 	u32 mask;
843 
844 	switch (adapter->hw.mac.type) {
845 	case ixgbe_mac_82598EB:
846 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
847 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
848 		break;
849 	case ixgbe_mac_82599EB:
850 	case ixgbe_mac_X540:
851 		mask = (qmask & 0xFFFFFFFF);
852 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
853 		mask = (qmask >> 32);
854 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
855 		break;
856 	default:
857 		break;
858 	}
859 }
860 
861 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *ring,
862 				      struct ixgbe_tx_buffer *tx_buffer)
863 {
864 	if (tx_buffer->skb) {
865 		dev_kfree_skb_any(tx_buffer->skb);
866 		if (dma_unmap_len(tx_buffer, len))
867 			dma_unmap_single(ring->dev,
868 					 dma_unmap_addr(tx_buffer, dma),
869 					 dma_unmap_len(tx_buffer, len),
870 					 DMA_TO_DEVICE);
871 	} else if (dma_unmap_len(tx_buffer, len)) {
872 		dma_unmap_page(ring->dev,
873 			       dma_unmap_addr(tx_buffer, dma),
874 			       dma_unmap_len(tx_buffer, len),
875 			       DMA_TO_DEVICE);
876 	}
877 	tx_buffer->next_to_watch = NULL;
878 	tx_buffer->skb = NULL;
879 	dma_unmap_len_set(tx_buffer, len, 0);
880 	/* tx_buffer must be completely set up in the transmit path */
881 }
882 
883 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
884 {
885 	struct ixgbe_hw *hw = &adapter->hw;
886 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
887 	int i;
888 	u32 data;
889 
890 	if ((hw->fc.current_mode != ixgbe_fc_full) &&
891 	    (hw->fc.current_mode != ixgbe_fc_rx_pause))
892 		return;
893 
894 	switch (hw->mac.type) {
895 	case ixgbe_mac_82598EB:
896 		data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
897 		break;
898 	default:
899 		data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
900 	}
901 	hwstats->lxoffrxc += data;
902 
903 	/* refill credits (no tx hang) if we received xoff */
904 	if (!data)
905 		return;
906 
907 	for (i = 0; i < adapter->num_tx_queues; i++)
908 		clear_bit(__IXGBE_HANG_CHECK_ARMED,
909 			  &adapter->tx_ring[i]->state);
910 }
911 
912 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
913 {
914 	struct ixgbe_hw *hw = &adapter->hw;
915 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
916 	u32 xoff[8] = {0};
917 	u8 tc;
918 	int i;
919 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
920 
921 	if (adapter->ixgbe_ieee_pfc)
922 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
923 
924 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
925 		ixgbe_update_xoff_rx_lfc(adapter);
926 		return;
927 	}
928 
929 	/* update stats for each tc, only valid with PFC enabled */
930 	for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
931 		u32 pxoffrxc;
932 
933 		switch (hw->mac.type) {
934 		case ixgbe_mac_82598EB:
935 			pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
936 			break;
937 		default:
938 			pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
939 		}
940 		hwstats->pxoffrxc[i] += pxoffrxc;
941 		/* Get the TC for given UP */
942 		tc = netdev_get_prio_tc_map(adapter->netdev, i);
943 		xoff[tc] += pxoffrxc;
944 	}
945 
946 	/* disarm tx queues that have received xoff frames */
947 	for (i = 0; i < adapter->num_tx_queues; i++) {
948 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
949 
950 		tc = tx_ring->dcb_tc;
951 		if (xoff[tc])
952 			clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
953 	}
954 }
955 
956 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
957 {
958 	return ring->stats.packets;
959 }
960 
961 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
962 {
963 	struct ixgbe_adapter *adapter;
964 	struct ixgbe_hw *hw;
965 	u32 head, tail;
966 
967 	if (ring->l2_accel_priv)
968 		adapter = ring->l2_accel_priv->real_adapter;
969 	else
970 		adapter = netdev_priv(ring->netdev);
971 
972 	hw = &adapter->hw;
973 	head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
974 	tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
975 
976 	if (head != tail)
977 		return (head < tail) ?
978 			tail - head : (tail + ring->count - head);
979 
980 	return 0;
981 }
982 
983 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
984 {
985 	u32 tx_done = ixgbe_get_tx_completed(tx_ring);
986 	u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
987 	u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
988 	bool ret = false;
989 
990 	clear_check_for_tx_hang(tx_ring);
991 
992 	/*
993 	 * Check for a hung queue, but be thorough. This verifies
994 	 * that a transmit has been completed since the previous
995 	 * check AND there is at least one packet pending. The
996 	 * ARMED bit is set to indicate a potential hang. The
997 	 * bit is cleared if a pause frame is received to remove
998 	 * false hang detection due to PFC or 802.3x frames. By
999 	 * requiring this to fail twice we avoid races with
1000 	 * pfc clearing the ARMED bit and conditions where we
1001 	 * run the check_tx_hang logic with a transmit completion
1002 	 * pending but without time to complete it yet.
1003 	 */
1004 	if ((tx_done_old == tx_done) && tx_pending) {
1005 		/* make sure it is true for two checks in a row */
1006 		ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
1007 				       &tx_ring->state);
1008 	} else {
1009 		/* update completed stats and continue */
1010 		tx_ring->tx_stats.tx_done_old = tx_done;
1011 		/* reset the countdown */
1012 		clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1013 	}
1014 
1015 	return ret;
1016 }
1017 
1018 /**
1019  * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1020  * @adapter: driver private struct
1021  **/
1022 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
1023 {
1024 
1025 	/* Do the reset outside of interrupt context */
1026 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1027 		adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
1028 		e_warn(drv, "initiating reset due to tx timeout\n");
1029 		ixgbe_service_event_schedule(adapter);
1030 	}
1031 }
1032 
1033 /**
1034  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1035  * @q_vector: structure containing interrupt and ring information
1036  * @tx_ring: tx ring to clean
1037  **/
1038 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
1039 			       struct ixgbe_ring *tx_ring)
1040 {
1041 	struct ixgbe_adapter *adapter = q_vector->adapter;
1042 	struct ixgbe_tx_buffer *tx_buffer;
1043 	union ixgbe_adv_tx_desc *tx_desc;
1044 	unsigned int total_bytes = 0, total_packets = 0;
1045 	unsigned int budget = q_vector->tx.work_limit;
1046 	unsigned int i = tx_ring->next_to_clean;
1047 
1048 	if (test_bit(__IXGBE_DOWN, &adapter->state))
1049 		return true;
1050 
1051 	tx_buffer = &tx_ring->tx_buffer_info[i];
1052 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
1053 	i -= tx_ring->count;
1054 
1055 	do {
1056 		union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1057 
1058 		/* if next_to_watch is not set then there is no work pending */
1059 		if (!eop_desc)
1060 			break;
1061 
1062 		/* prevent any other reads prior to eop_desc */
1063 		read_barrier_depends();
1064 
1065 		/* if DD is not set pending work has not been completed */
1066 		if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
1067 			break;
1068 
1069 		/* clear next_to_watch to prevent false hangs */
1070 		tx_buffer->next_to_watch = NULL;
1071 
1072 		/* update the statistics for this packet */
1073 		total_bytes += tx_buffer->bytecount;
1074 		total_packets += tx_buffer->gso_segs;
1075 
1076 		/* free the skb */
1077 		dev_kfree_skb_any(tx_buffer->skb);
1078 
1079 		/* unmap skb header data */
1080 		dma_unmap_single(tx_ring->dev,
1081 				 dma_unmap_addr(tx_buffer, dma),
1082 				 dma_unmap_len(tx_buffer, len),
1083 				 DMA_TO_DEVICE);
1084 
1085 		/* clear tx_buffer data */
1086 		tx_buffer->skb = NULL;
1087 		dma_unmap_len_set(tx_buffer, len, 0);
1088 
1089 		/* unmap remaining buffers */
1090 		while (tx_desc != eop_desc) {
1091 			tx_buffer++;
1092 			tx_desc++;
1093 			i++;
1094 			if (unlikely(!i)) {
1095 				i -= tx_ring->count;
1096 				tx_buffer = tx_ring->tx_buffer_info;
1097 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1098 			}
1099 
1100 			/* unmap any remaining paged data */
1101 			if (dma_unmap_len(tx_buffer, len)) {
1102 				dma_unmap_page(tx_ring->dev,
1103 					       dma_unmap_addr(tx_buffer, dma),
1104 					       dma_unmap_len(tx_buffer, len),
1105 					       DMA_TO_DEVICE);
1106 				dma_unmap_len_set(tx_buffer, len, 0);
1107 			}
1108 		}
1109 
1110 		/* move us one more past the eop_desc for start of next pkt */
1111 		tx_buffer++;
1112 		tx_desc++;
1113 		i++;
1114 		if (unlikely(!i)) {
1115 			i -= tx_ring->count;
1116 			tx_buffer = tx_ring->tx_buffer_info;
1117 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1118 		}
1119 
1120 		/* issue prefetch for next Tx descriptor */
1121 		prefetch(tx_desc);
1122 
1123 		/* update budget accounting */
1124 		budget--;
1125 	} while (likely(budget));
1126 
1127 	i += tx_ring->count;
1128 	tx_ring->next_to_clean = i;
1129 	u64_stats_update_begin(&tx_ring->syncp);
1130 	tx_ring->stats.bytes += total_bytes;
1131 	tx_ring->stats.packets += total_packets;
1132 	u64_stats_update_end(&tx_ring->syncp);
1133 	q_vector->tx.total_bytes += total_bytes;
1134 	q_vector->tx.total_packets += total_packets;
1135 
1136 	if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1137 		/* schedule immediate reset if we believe we hung */
1138 		struct ixgbe_hw *hw = &adapter->hw;
1139 		e_err(drv, "Detected Tx Unit Hang\n"
1140 			"  Tx Queue             <%d>\n"
1141 			"  TDH, TDT             <%x>, <%x>\n"
1142 			"  next_to_use          <%x>\n"
1143 			"  next_to_clean        <%x>\n"
1144 			"tx_buffer_info[next_to_clean]\n"
1145 			"  time_stamp           <%lx>\n"
1146 			"  jiffies              <%lx>\n",
1147 			tx_ring->queue_index,
1148 			IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1149 			IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1150 			tx_ring->next_to_use, i,
1151 			tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1152 
1153 		netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
1154 
1155 		e_info(probe,
1156 		       "tx hang %d detected on queue %d, resetting adapter\n",
1157 			adapter->tx_timeout_count + 1, tx_ring->queue_index);
1158 
1159 		/* schedule immediate reset if we believe we hung */
1160 		ixgbe_tx_timeout_reset(adapter);
1161 
1162 		/* the adapter is about to reset, no point in enabling stuff */
1163 		return true;
1164 	}
1165 
1166 	netdev_tx_completed_queue(txring_txq(tx_ring),
1167 				  total_packets, total_bytes);
1168 
1169 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1170 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1171 		     (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1172 		/* Make sure that anybody stopping the queue after this
1173 		 * sees the new next_to_clean.
1174 		 */
1175 		smp_mb();
1176 		if (__netif_subqueue_stopped(tx_ring->netdev,
1177 					     tx_ring->queue_index)
1178 		    && !test_bit(__IXGBE_DOWN, &adapter->state)) {
1179 			netif_wake_subqueue(tx_ring->netdev,
1180 					    tx_ring->queue_index);
1181 			++tx_ring->tx_stats.restart_queue;
1182 		}
1183 	}
1184 
1185 	return !!budget;
1186 }
1187 
1188 #ifdef CONFIG_IXGBE_DCA
1189 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1190 				struct ixgbe_ring *tx_ring,
1191 				int cpu)
1192 {
1193 	struct ixgbe_hw *hw = &adapter->hw;
1194 	u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
1195 	u16 reg_offset;
1196 
1197 	switch (hw->mac.type) {
1198 	case ixgbe_mac_82598EB:
1199 		reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1200 		break;
1201 	case ixgbe_mac_82599EB:
1202 	case ixgbe_mac_X540:
1203 		reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1204 		txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1205 		break;
1206 	default:
1207 		/* for unknown hardware do not write register */
1208 		return;
1209 	}
1210 
1211 	/*
1212 	 * We can enable relaxed ordering for reads, but not writes when
1213 	 * DCA is enabled.  This is due to a known issue in some chipsets
1214 	 * which will cause the DCA tag to be cleared.
1215 	 */
1216 	txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1217 		  IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1218 		  IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1219 
1220 	IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1221 }
1222 
1223 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1224 				struct ixgbe_ring *rx_ring,
1225 				int cpu)
1226 {
1227 	struct ixgbe_hw *hw = &adapter->hw;
1228 	u32 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1229 	u8 reg_idx = rx_ring->reg_idx;
1230 
1231 
1232 	switch (hw->mac.type) {
1233 	case ixgbe_mac_82599EB:
1234 	case ixgbe_mac_X540:
1235 		rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1236 		break;
1237 	default:
1238 		break;
1239 	}
1240 
1241 	/*
1242 	 * We can enable relaxed ordering for reads, but not writes when
1243 	 * DCA is enabled.  This is due to a known issue in some chipsets
1244 	 * which will cause the DCA tag to be cleared.
1245 	 */
1246 	rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1247 		  IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1248 
1249 	IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1250 }
1251 
1252 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1253 {
1254 	struct ixgbe_adapter *adapter = q_vector->adapter;
1255 	struct ixgbe_ring *ring;
1256 	int cpu = get_cpu();
1257 
1258 	if (q_vector->cpu == cpu)
1259 		goto out_no_update;
1260 
1261 	ixgbe_for_each_ring(ring, q_vector->tx)
1262 		ixgbe_update_tx_dca(adapter, ring, cpu);
1263 
1264 	ixgbe_for_each_ring(ring, q_vector->rx)
1265 		ixgbe_update_rx_dca(adapter, ring, cpu);
1266 
1267 	q_vector->cpu = cpu;
1268 out_no_update:
1269 	put_cpu();
1270 }
1271 
1272 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1273 {
1274 	int i;
1275 
1276 	if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1277 		return;
1278 
1279 	/* always use CB2 mode, difference is masked in the CB driver */
1280 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1281 
1282 	for (i = 0; i < adapter->num_q_vectors; i++) {
1283 		adapter->q_vector[i]->cpu = -1;
1284 		ixgbe_update_dca(adapter->q_vector[i]);
1285 	}
1286 }
1287 
1288 static int __ixgbe_notify_dca(struct device *dev, void *data)
1289 {
1290 	struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1291 	unsigned long event = *(unsigned long *)data;
1292 
1293 	if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1294 		return 0;
1295 
1296 	switch (event) {
1297 	case DCA_PROVIDER_ADD:
1298 		/* if we're already enabled, don't do it again */
1299 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1300 			break;
1301 		if (dca_add_requester(dev) == 0) {
1302 			adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1303 			ixgbe_setup_dca(adapter);
1304 			break;
1305 		}
1306 		/* Fall Through since DCA is disabled. */
1307 	case DCA_PROVIDER_REMOVE:
1308 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1309 			dca_remove_requester(dev);
1310 			adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1311 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1312 		}
1313 		break;
1314 	}
1315 
1316 	return 0;
1317 }
1318 
1319 #endif /* CONFIG_IXGBE_DCA */
1320 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1321 				 union ixgbe_adv_rx_desc *rx_desc,
1322 				 struct sk_buff *skb)
1323 {
1324 	if (ring->netdev->features & NETIF_F_RXHASH)
1325 		skb_set_hash(skb,
1326 			     le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1327 			     PKT_HASH_TYPE_L3);
1328 }
1329 
1330 #ifdef IXGBE_FCOE
1331 /**
1332  * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1333  * @ring: structure containing ring specific data
1334  * @rx_desc: advanced rx descriptor
1335  *
1336  * Returns : true if it is FCoE pkt
1337  */
1338 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1339 				    union ixgbe_adv_rx_desc *rx_desc)
1340 {
1341 	__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1342 
1343 	return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1344 	       ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1345 		(cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1346 			     IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1347 }
1348 
1349 #endif /* IXGBE_FCOE */
1350 /**
1351  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1352  * @ring: structure containing ring specific data
1353  * @rx_desc: current Rx descriptor being processed
1354  * @skb: skb currently being received and modified
1355  **/
1356 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1357 				     union ixgbe_adv_rx_desc *rx_desc,
1358 				     struct sk_buff *skb)
1359 {
1360 	skb_checksum_none_assert(skb);
1361 
1362 	/* Rx csum disabled */
1363 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
1364 		return;
1365 
1366 	/* if IP and error */
1367 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1368 	    ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1369 		ring->rx_stats.csum_err++;
1370 		return;
1371 	}
1372 
1373 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1374 		return;
1375 
1376 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1377 		__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1378 
1379 		/*
1380 		 * 82599 errata, UDP frames with a 0 checksum can be marked as
1381 		 * checksum errors.
1382 		 */
1383 		if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1384 		    test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1385 			return;
1386 
1387 		ring->rx_stats.csum_err++;
1388 		return;
1389 	}
1390 
1391 	/* It must be a TCP or UDP packet with a valid checksum */
1392 	skb->ip_summed = CHECKSUM_UNNECESSARY;
1393 }
1394 
1395 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1396 {
1397 	rx_ring->next_to_use = val;
1398 
1399 	/* update next to alloc since we have filled the ring */
1400 	rx_ring->next_to_alloc = val;
1401 	/*
1402 	 * Force memory writes to complete before letting h/w
1403 	 * know there are new descriptors to fetch.  (Only
1404 	 * applicable for weak-ordered memory model archs,
1405 	 * such as IA-64).
1406 	 */
1407 	wmb();
1408 	ixgbe_write_tail(rx_ring, val);
1409 }
1410 
1411 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1412 				    struct ixgbe_rx_buffer *bi)
1413 {
1414 	struct page *page = bi->page;
1415 	dma_addr_t dma = bi->dma;
1416 
1417 	/* since we are recycling buffers we should seldom need to alloc */
1418 	if (likely(dma))
1419 		return true;
1420 
1421 	/* alloc new page for storage */
1422 	if (likely(!page)) {
1423 		page = __skb_alloc_pages(GFP_ATOMIC | __GFP_COLD | __GFP_COMP,
1424 					 bi->skb, ixgbe_rx_pg_order(rx_ring));
1425 		if (unlikely(!page)) {
1426 			rx_ring->rx_stats.alloc_rx_page_failed++;
1427 			return false;
1428 		}
1429 		bi->page = page;
1430 	}
1431 
1432 	/* map page for use */
1433 	dma = dma_map_page(rx_ring->dev, page, 0,
1434 			   ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1435 
1436 	/*
1437 	 * if mapping failed free memory back to system since
1438 	 * there isn't much point in holding memory we can't use
1439 	 */
1440 	if (dma_mapping_error(rx_ring->dev, dma)) {
1441 		__free_pages(page, ixgbe_rx_pg_order(rx_ring));
1442 		bi->page = NULL;
1443 
1444 		rx_ring->rx_stats.alloc_rx_page_failed++;
1445 		return false;
1446 	}
1447 
1448 	bi->dma = dma;
1449 	bi->page_offset = 0;
1450 
1451 	return true;
1452 }
1453 
1454 /**
1455  * ixgbe_alloc_rx_buffers - Replace used receive buffers
1456  * @rx_ring: ring to place buffers on
1457  * @cleaned_count: number of buffers to replace
1458  **/
1459 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1460 {
1461 	union ixgbe_adv_rx_desc *rx_desc;
1462 	struct ixgbe_rx_buffer *bi;
1463 	u16 i = rx_ring->next_to_use;
1464 
1465 	/* nothing to do */
1466 	if (!cleaned_count)
1467 		return;
1468 
1469 	rx_desc = IXGBE_RX_DESC(rx_ring, i);
1470 	bi = &rx_ring->rx_buffer_info[i];
1471 	i -= rx_ring->count;
1472 
1473 	do {
1474 		if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1475 			break;
1476 
1477 		/*
1478 		 * Refresh the desc even if buffer_addrs didn't change
1479 		 * because each write-back erases this info.
1480 		 */
1481 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1482 
1483 		rx_desc++;
1484 		bi++;
1485 		i++;
1486 		if (unlikely(!i)) {
1487 			rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1488 			bi = rx_ring->rx_buffer_info;
1489 			i -= rx_ring->count;
1490 		}
1491 
1492 		/* clear the hdr_addr for the next_to_use descriptor */
1493 		rx_desc->read.hdr_addr = 0;
1494 
1495 		cleaned_count--;
1496 	} while (cleaned_count);
1497 
1498 	i += rx_ring->count;
1499 
1500 	if (rx_ring->next_to_use != i)
1501 		ixgbe_release_rx_desc(rx_ring, i);
1502 }
1503 
1504 /**
1505  * ixgbe_get_headlen - determine size of header for RSC/LRO/GRO/FCOE
1506  * @data: pointer to the start of the headers
1507  * @max_len: total length of section to find headers in
1508  *
1509  * This function is meant to determine the length of headers that will
1510  * be recognized by hardware for LRO, GRO, and RSC offloads.  The main
1511  * motivation of doing this is to only perform one pull for IPv4 TCP
1512  * packets so that we can do basic things like calculating the gso_size
1513  * based on the average data per packet.
1514  **/
1515 static unsigned int ixgbe_get_headlen(unsigned char *data,
1516 				      unsigned int max_len)
1517 {
1518 	union {
1519 		unsigned char *network;
1520 		/* l2 headers */
1521 		struct ethhdr *eth;
1522 		struct vlan_hdr *vlan;
1523 		/* l3 headers */
1524 		struct iphdr *ipv4;
1525 		struct ipv6hdr *ipv6;
1526 	} hdr;
1527 	__be16 protocol;
1528 	u8 nexthdr = 0;	/* default to not TCP */
1529 	u8 hlen;
1530 
1531 	/* this should never happen, but better safe than sorry */
1532 	if (max_len < ETH_HLEN)
1533 		return max_len;
1534 
1535 	/* initialize network frame pointer */
1536 	hdr.network = data;
1537 
1538 	/* set first protocol and move network header forward */
1539 	protocol = hdr.eth->h_proto;
1540 	hdr.network += ETH_HLEN;
1541 
1542 	/* handle any vlan tag if present */
1543 	if (protocol == htons(ETH_P_8021Q)) {
1544 		if ((hdr.network - data) > (max_len - VLAN_HLEN))
1545 			return max_len;
1546 
1547 		protocol = hdr.vlan->h_vlan_encapsulated_proto;
1548 		hdr.network += VLAN_HLEN;
1549 	}
1550 
1551 	/* handle L3 protocols */
1552 	if (protocol == htons(ETH_P_IP)) {
1553 		if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
1554 			return max_len;
1555 
1556 		/* access ihl as a u8 to avoid unaligned access on ia64 */
1557 		hlen = (hdr.network[0] & 0x0F) << 2;
1558 
1559 		/* verify hlen meets minimum size requirements */
1560 		if (hlen < sizeof(struct iphdr))
1561 			return hdr.network - data;
1562 
1563 		/* record next protocol if header is present */
1564 		if (!(hdr.ipv4->frag_off & htons(IP_OFFSET)))
1565 			nexthdr = hdr.ipv4->protocol;
1566 	} else if (protocol == htons(ETH_P_IPV6)) {
1567 		if ((hdr.network - data) > (max_len - sizeof(struct ipv6hdr)))
1568 			return max_len;
1569 
1570 		/* record next protocol */
1571 		nexthdr = hdr.ipv6->nexthdr;
1572 		hlen = sizeof(struct ipv6hdr);
1573 #ifdef IXGBE_FCOE
1574 	} else if (protocol == htons(ETH_P_FCOE)) {
1575 		if ((hdr.network - data) > (max_len - FCOE_HEADER_LEN))
1576 			return max_len;
1577 		hlen = FCOE_HEADER_LEN;
1578 #endif
1579 	} else {
1580 		return hdr.network - data;
1581 	}
1582 
1583 	/* relocate pointer to start of L4 header */
1584 	hdr.network += hlen;
1585 
1586 	/* finally sort out TCP/UDP */
1587 	if (nexthdr == IPPROTO_TCP) {
1588 		if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
1589 			return max_len;
1590 
1591 		/* access doff as a u8 to avoid unaligned access on ia64 */
1592 		hlen = (hdr.network[12] & 0xF0) >> 2;
1593 
1594 		/* verify hlen meets minimum size requirements */
1595 		if (hlen < sizeof(struct tcphdr))
1596 			return hdr.network - data;
1597 
1598 		hdr.network += hlen;
1599 	} else if (nexthdr == IPPROTO_UDP) {
1600 		if ((hdr.network - data) > (max_len - sizeof(struct udphdr)))
1601 			return max_len;
1602 
1603 		hdr.network += sizeof(struct udphdr);
1604 	}
1605 
1606 	/*
1607 	 * If everything has gone correctly hdr.network should be the
1608 	 * data section of the packet and will be the end of the header.
1609 	 * If not then it probably represents the end of the last recognized
1610 	 * header.
1611 	 */
1612 	if ((hdr.network - data) < max_len)
1613 		return hdr.network - data;
1614 	else
1615 		return max_len;
1616 }
1617 
1618 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1619 				   struct sk_buff *skb)
1620 {
1621 	u16 hdr_len = skb_headlen(skb);
1622 
1623 	/* set gso_size to avoid messing up TCP MSS */
1624 	skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1625 						 IXGBE_CB(skb)->append_cnt);
1626 	skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1627 }
1628 
1629 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1630 				   struct sk_buff *skb)
1631 {
1632 	/* if append_cnt is 0 then frame is not RSC */
1633 	if (!IXGBE_CB(skb)->append_cnt)
1634 		return;
1635 
1636 	rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1637 	rx_ring->rx_stats.rsc_flush++;
1638 
1639 	ixgbe_set_rsc_gso_size(rx_ring, skb);
1640 
1641 	/* gso_size is computed using append_cnt so always clear it last */
1642 	IXGBE_CB(skb)->append_cnt = 0;
1643 }
1644 
1645 /**
1646  * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1647  * @rx_ring: rx descriptor ring packet is being transacted on
1648  * @rx_desc: pointer to the EOP Rx descriptor
1649  * @skb: pointer to current skb being populated
1650  *
1651  * This function checks the ring, descriptor, and packet information in
1652  * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1653  * other fields within the skb.
1654  **/
1655 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1656 				     union ixgbe_adv_rx_desc *rx_desc,
1657 				     struct sk_buff *skb)
1658 {
1659 	struct net_device *dev = rx_ring->netdev;
1660 
1661 	ixgbe_update_rsc_stats(rx_ring, skb);
1662 
1663 	ixgbe_rx_hash(rx_ring, rx_desc, skb);
1664 
1665 	ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1666 
1667 	if (unlikely(ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_STAT_TS)))
1668 		ixgbe_ptp_rx_hwtstamp(rx_ring->q_vector->adapter, skb);
1669 
1670 	if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1671 	    ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1672 		u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1673 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1674 	}
1675 
1676 	skb_record_rx_queue(skb, rx_ring->queue_index);
1677 
1678 	skb->protocol = eth_type_trans(skb, dev);
1679 }
1680 
1681 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1682 			 struct sk_buff *skb)
1683 {
1684 	struct ixgbe_adapter *adapter = q_vector->adapter;
1685 
1686 	if (ixgbe_qv_busy_polling(q_vector))
1687 		netif_receive_skb(skb);
1688 	else if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1689 		napi_gro_receive(&q_vector->napi, skb);
1690 	else
1691 		netif_rx(skb);
1692 }
1693 
1694 /**
1695  * ixgbe_is_non_eop - process handling of non-EOP buffers
1696  * @rx_ring: Rx ring being processed
1697  * @rx_desc: Rx descriptor for current buffer
1698  * @skb: Current socket buffer containing buffer in progress
1699  *
1700  * This function updates next to clean.  If the buffer is an EOP buffer
1701  * this function exits returning false, otherwise it will place the
1702  * sk_buff in the next buffer to be chained and return true indicating
1703  * that this is in fact a non-EOP buffer.
1704  **/
1705 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1706 			     union ixgbe_adv_rx_desc *rx_desc,
1707 			     struct sk_buff *skb)
1708 {
1709 	u32 ntc = rx_ring->next_to_clean + 1;
1710 
1711 	/* fetch, update, and store next to clean */
1712 	ntc = (ntc < rx_ring->count) ? ntc : 0;
1713 	rx_ring->next_to_clean = ntc;
1714 
1715 	prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1716 
1717 	/* update RSC append count if present */
1718 	if (ring_is_rsc_enabled(rx_ring)) {
1719 		__le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1720 				     cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1721 
1722 		if (unlikely(rsc_enabled)) {
1723 			u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1724 
1725 			rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1726 			IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1727 
1728 			/* update ntc based on RSC value */
1729 			ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1730 			ntc &= IXGBE_RXDADV_NEXTP_MASK;
1731 			ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1732 		}
1733 	}
1734 
1735 	/* if we are the last buffer then there is nothing else to do */
1736 	if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1737 		return false;
1738 
1739 	/* place skb in next buffer to be received */
1740 	rx_ring->rx_buffer_info[ntc].skb = skb;
1741 	rx_ring->rx_stats.non_eop_descs++;
1742 
1743 	return true;
1744 }
1745 
1746 /**
1747  * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1748  * @rx_ring: rx descriptor ring packet is being transacted on
1749  * @skb: pointer to current skb being adjusted
1750  *
1751  * This function is an ixgbe specific version of __pskb_pull_tail.  The
1752  * main difference between this version and the original function is that
1753  * this function can make several assumptions about the state of things
1754  * that allow for significant optimizations versus the standard function.
1755  * As a result we can do things like drop a frag and maintain an accurate
1756  * truesize for the skb.
1757  */
1758 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1759 			    struct sk_buff *skb)
1760 {
1761 	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1762 	unsigned char *va;
1763 	unsigned int pull_len;
1764 
1765 	/*
1766 	 * it is valid to use page_address instead of kmap since we are
1767 	 * working with pages allocated out of the lomem pool per
1768 	 * alloc_page(GFP_ATOMIC)
1769 	 */
1770 	va = skb_frag_address(frag);
1771 
1772 	/*
1773 	 * we need the header to contain the greater of either ETH_HLEN or
1774 	 * 60 bytes if the skb->len is less than 60 for skb_pad.
1775 	 */
1776 	pull_len = ixgbe_get_headlen(va, IXGBE_RX_HDR_SIZE);
1777 
1778 	/* align pull length to size of long to optimize memcpy performance */
1779 	skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1780 
1781 	/* update all of the pointers */
1782 	skb_frag_size_sub(frag, pull_len);
1783 	frag->page_offset += pull_len;
1784 	skb->data_len -= pull_len;
1785 	skb->tail += pull_len;
1786 }
1787 
1788 /**
1789  * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1790  * @rx_ring: rx descriptor ring packet is being transacted on
1791  * @skb: pointer to current skb being updated
1792  *
1793  * This function provides a basic DMA sync up for the first fragment of an
1794  * skb.  The reason for doing this is that the first fragment cannot be
1795  * unmapped until we have reached the end of packet descriptor for a buffer
1796  * chain.
1797  */
1798 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1799 				struct sk_buff *skb)
1800 {
1801 	/* if the page was released unmap it, else just sync our portion */
1802 	if (unlikely(IXGBE_CB(skb)->page_released)) {
1803 		dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1804 			       ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1805 		IXGBE_CB(skb)->page_released = false;
1806 	} else {
1807 		struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1808 
1809 		dma_sync_single_range_for_cpu(rx_ring->dev,
1810 					      IXGBE_CB(skb)->dma,
1811 					      frag->page_offset,
1812 					      ixgbe_rx_bufsz(rx_ring),
1813 					      DMA_FROM_DEVICE);
1814 	}
1815 	IXGBE_CB(skb)->dma = 0;
1816 }
1817 
1818 /**
1819  * ixgbe_cleanup_headers - Correct corrupted or empty headers
1820  * @rx_ring: rx descriptor ring packet is being transacted on
1821  * @rx_desc: pointer to the EOP Rx descriptor
1822  * @skb: pointer to current skb being fixed
1823  *
1824  * Check for corrupted packet headers caused by senders on the local L2
1825  * embedded NIC switch not setting up their Tx Descriptors right.  These
1826  * should be very rare.
1827  *
1828  * Also address the case where we are pulling data in on pages only
1829  * and as such no data is present in the skb header.
1830  *
1831  * In addition if skb is not at least 60 bytes we need to pad it so that
1832  * it is large enough to qualify as a valid Ethernet frame.
1833  *
1834  * Returns true if an error was encountered and skb was freed.
1835  **/
1836 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1837 				  union ixgbe_adv_rx_desc *rx_desc,
1838 				  struct sk_buff *skb)
1839 {
1840 	struct net_device *netdev = rx_ring->netdev;
1841 
1842 	/* verify that the packet does not have any known errors */
1843 	if (unlikely(ixgbe_test_staterr(rx_desc,
1844 					IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1845 	    !(netdev->features & NETIF_F_RXALL))) {
1846 		dev_kfree_skb_any(skb);
1847 		return true;
1848 	}
1849 
1850 	/* place header in linear portion of buffer */
1851 	if (skb_is_nonlinear(skb))
1852 		ixgbe_pull_tail(rx_ring, skb);
1853 
1854 #ifdef IXGBE_FCOE
1855 	/* do not attempt to pad FCoE Frames as this will disrupt DDP */
1856 	if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1857 		return false;
1858 
1859 #endif
1860 	/* if skb_pad returns an error the skb was freed */
1861 	if (unlikely(skb->len < 60)) {
1862 		int pad_len = 60 - skb->len;
1863 
1864 		if (skb_pad(skb, pad_len))
1865 			return true;
1866 		__skb_put(skb, pad_len);
1867 	}
1868 
1869 	return false;
1870 }
1871 
1872 /**
1873  * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1874  * @rx_ring: rx descriptor ring to store buffers on
1875  * @old_buff: donor buffer to have page reused
1876  *
1877  * Synchronizes page for reuse by the adapter
1878  **/
1879 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1880 				struct ixgbe_rx_buffer *old_buff)
1881 {
1882 	struct ixgbe_rx_buffer *new_buff;
1883 	u16 nta = rx_ring->next_to_alloc;
1884 
1885 	new_buff = &rx_ring->rx_buffer_info[nta];
1886 
1887 	/* update, and store next to alloc */
1888 	nta++;
1889 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1890 
1891 	/* transfer page from old buffer to new buffer */
1892 	new_buff->page = old_buff->page;
1893 	new_buff->dma = old_buff->dma;
1894 	new_buff->page_offset = old_buff->page_offset;
1895 
1896 	/* sync the buffer for use by the device */
1897 	dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1898 					 new_buff->page_offset,
1899 					 ixgbe_rx_bufsz(rx_ring),
1900 					 DMA_FROM_DEVICE);
1901 }
1902 
1903 /**
1904  * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1905  * @rx_ring: rx descriptor ring to transact packets on
1906  * @rx_buffer: buffer containing page to add
1907  * @rx_desc: descriptor containing length of buffer written by hardware
1908  * @skb: sk_buff to place the data into
1909  *
1910  * This function will add the data contained in rx_buffer->page to the skb.
1911  * This is done either through a direct copy if the data in the buffer is
1912  * less than the skb header size, otherwise it will just attach the page as
1913  * a frag to the skb.
1914  *
1915  * The function will then update the page offset if necessary and return
1916  * true if the buffer can be reused by the adapter.
1917  **/
1918 static bool ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1919 			      struct ixgbe_rx_buffer *rx_buffer,
1920 			      union ixgbe_adv_rx_desc *rx_desc,
1921 			      struct sk_buff *skb)
1922 {
1923 	struct page *page = rx_buffer->page;
1924 	unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
1925 #if (PAGE_SIZE < 8192)
1926 	unsigned int truesize = ixgbe_rx_bufsz(rx_ring);
1927 #else
1928 	unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
1929 	unsigned int last_offset = ixgbe_rx_pg_size(rx_ring) -
1930 				   ixgbe_rx_bufsz(rx_ring);
1931 #endif
1932 
1933 	if ((size <= IXGBE_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
1934 		unsigned char *va = page_address(page) + rx_buffer->page_offset;
1935 
1936 		memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
1937 
1938 		/* we can reuse buffer as-is, just make sure it is local */
1939 		if (likely(page_to_nid(page) == numa_node_id()))
1940 			return true;
1941 
1942 		/* this page cannot be reused so discard it */
1943 		put_page(page);
1944 		return false;
1945 	}
1946 
1947 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
1948 			rx_buffer->page_offset, size, truesize);
1949 
1950 	/* avoid re-using remote pages */
1951 	if (unlikely(page_to_nid(page) != numa_node_id()))
1952 		return false;
1953 
1954 #if (PAGE_SIZE < 8192)
1955 	/* if we are only owner of page we can reuse it */
1956 	if (unlikely(page_count(page) != 1))
1957 		return false;
1958 
1959 	/* flip page offset to other buffer */
1960 	rx_buffer->page_offset ^= truesize;
1961 
1962 	/*
1963 	 * since we are the only owner of the page and we need to
1964 	 * increment it, just set the value to 2 in order to avoid
1965 	 * an unecessary locked operation
1966 	 */
1967 	atomic_set(&page->_count, 2);
1968 #else
1969 	/* move offset up to the next cache line */
1970 	rx_buffer->page_offset += truesize;
1971 
1972 	if (rx_buffer->page_offset > last_offset)
1973 		return false;
1974 
1975 	/* bump ref count on page before it is given to the stack */
1976 	get_page(page);
1977 #endif
1978 
1979 	return true;
1980 }
1981 
1982 static struct sk_buff *ixgbe_fetch_rx_buffer(struct ixgbe_ring *rx_ring,
1983 					     union ixgbe_adv_rx_desc *rx_desc)
1984 {
1985 	struct ixgbe_rx_buffer *rx_buffer;
1986 	struct sk_buff *skb;
1987 	struct page *page;
1988 
1989 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
1990 	page = rx_buffer->page;
1991 	prefetchw(page);
1992 
1993 	skb = rx_buffer->skb;
1994 
1995 	if (likely(!skb)) {
1996 		void *page_addr = page_address(page) +
1997 				  rx_buffer->page_offset;
1998 
1999 		/* prefetch first cache line of first page */
2000 		prefetch(page_addr);
2001 #if L1_CACHE_BYTES < 128
2002 		prefetch(page_addr + L1_CACHE_BYTES);
2003 #endif
2004 
2005 		/* allocate a skb to store the frags */
2006 		skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
2007 						IXGBE_RX_HDR_SIZE);
2008 		if (unlikely(!skb)) {
2009 			rx_ring->rx_stats.alloc_rx_buff_failed++;
2010 			return NULL;
2011 		}
2012 
2013 		/*
2014 		 * we will be copying header into skb->data in
2015 		 * pskb_may_pull so it is in our interest to prefetch
2016 		 * it now to avoid a possible cache miss
2017 		 */
2018 		prefetchw(skb->data);
2019 
2020 		/*
2021 		 * Delay unmapping of the first packet. It carries the
2022 		 * header information, HW may still access the header
2023 		 * after the writeback.  Only unmap it when EOP is
2024 		 * reached
2025 		 */
2026 		if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
2027 			goto dma_sync;
2028 
2029 		IXGBE_CB(skb)->dma = rx_buffer->dma;
2030 	} else {
2031 		if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2032 			ixgbe_dma_sync_frag(rx_ring, skb);
2033 
2034 dma_sync:
2035 		/* we are reusing so sync this buffer for CPU use */
2036 		dma_sync_single_range_for_cpu(rx_ring->dev,
2037 					      rx_buffer->dma,
2038 					      rx_buffer->page_offset,
2039 					      ixgbe_rx_bufsz(rx_ring),
2040 					      DMA_FROM_DEVICE);
2041 	}
2042 
2043 	/* pull page into skb */
2044 	if (ixgbe_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
2045 		/* hand second half of page back to the ring */
2046 		ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2047 	} else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
2048 		/* the page has been released from the ring */
2049 		IXGBE_CB(skb)->page_released = true;
2050 	} else {
2051 		/* we are not reusing the buffer so unmap it */
2052 		dma_unmap_page(rx_ring->dev, rx_buffer->dma,
2053 			       ixgbe_rx_pg_size(rx_ring),
2054 			       DMA_FROM_DEVICE);
2055 	}
2056 
2057 	/* clear contents of buffer_info */
2058 	rx_buffer->skb = NULL;
2059 	rx_buffer->dma = 0;
2060 	rx_buffer->page = NULL;
2061 
2062 	return skb;
2063 }
2064 
2065 /**
2066  * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2067  * @q_vector: structure containing interrupt and ring information
2068  * @rx_ring: rx descriptor ring to transact packets on
2069  * @budget: Total limit on number of packets to process
2070  *
2071  * This function provides a "bounce buffer" approach to Rx interrupt
2072  * processing.  The advantage to this is that on systems that have
2073  * expensive overhead for IOMMU access this provides a means of avoiding
2074  * it by maintaining the mapping of the page to the syste.
2075  *
2076  * Returns amount of work completed
2077  **/
2078 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2079 			       struct ixgbe_ring *rx_ring,
2080 			       const int budget)
2081 {
2082 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2083 #ifdef IXGBE_FCOE
2084 	struct ixgbe_adapter *adapter = q_vector->adapter;
2085 	int ddp_bytes;
2086 	unsigned int mss = 0;
2087 #endif /* IXGBE_FCOE */
2088 	u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2089 
2090 	while (likely(total_rx_packets < budget)) {
2091 		union ixgbe_adv_rx_desc *rx_desc;
2092 		struct sk_buff *skb;
2093 
2094 		/* return some buffers to hardware, one at a time is too slow */
2095 		if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2096 			ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2097 			cleaned_count = 0;
2098 		}
2099 
2100 		rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2101 
2102 		if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
2103 			break;
2104 
2105 		/*
2106 		 * This memory barrier is needed to keep us from reading
2107 		 * any other fields out of the rx_desc until we know the
2108 		 * RXD_STAT_DD bit is set
2109 		 */
2110 		rmb();
2111 
2112 		/* retrieve a buffer from the ring */
2113 		skb = ixgbe_fetch_rx_buffer(rx_ring, rx_desc);
2114 
2115 		/* exit if we failed to retrieve a buffer */
2116 		if (!skb)
2117 			break;
2118 
2119 		cleaned_count++;
2120 
2121 		/* place incomplete frames back on ring for completion */
2122 		if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2123 			continue;
2124 
2125 		/* verify the packet layout is correct */
2126 		if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2127 			continue;
2128 
2129 		/* probably a little skewed due to removing CRC */
2130 		total_rx_bytes += skb->len;
2131 
2132 		/* populate checksum, timestamp, VLAN, and protocol */
2133 		ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2134 
2135 #ifdef IXGBE_FCOE
2136 		/* if ddp, not passing to ULD unless for FCP_RSP or error */
2137 		if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2138 			ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2139 			/* include DDPed FCoE data */
2140 			if (ddp_bytes > 0) {
2141 				if (!mss) {
2142 					mss = rx_ring->netdev->mtu -
2143 						sizeof(struct fcoe_hdr) -
2144 						sizeof(struct fc_frame_header) -
2145 						sizeof(struct fcoe_crc_eof);
2146 					if (mss > 512)
2147 						mss &= ~511;
2148 				}
2149 				total_rx_bytes += ddp_bytes;
2150 				total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2151 								 mss);
2152 			}
2153 			if (!ddp_bytes) {
2154 				dev_kfree_skb_any(skb);
2155 				continue;
2156 			}
2157 		}
2158 
2159 #endif /* IXGBE_FCOE */
2160 		skb_mark_napi_id(skb, &q_vector->napi);
2161 		ixgbe_rx_skb(q_vector, skb);
2162 
2163 		/* update budget accounting */
2164 		total_rx_packets++;
2165 	}
2166 
2167 	u64_stats_update_begin(&rx_ring->syncp);
2168 	rx_ring->stats.packets += total_rx_packets;
2169 	rx_ring->stats.bytes += total_rx_bytes;
2170 	u64_stats_update_end(&rx_ring->syncp);
2171 	q_vector->rx.total_packets += total_rx_packets;
2172 	q_vector->rx.total_bytes += total_rx_bytes;
2173 
2174 	if (cleaned_count)
2175 		ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2176 
2177 	return total_rx_packets;
2178 }
2179 
2180 #ifdef CONFIG_NET_RX_BUSY_POLL
2181 /* must be called with local_bh_disable()d */
2182 static int ixgbe_low_latency_recv(struct napi_struct *napi)
2183 {
2184 	struct ixgbe_q_vector *q_vector =
2185 			container_of(napi, struct ixgbe_q_vector, napi);
2186 	struct ixgbe_adapter *adapter = q_vector->adapter;
2187 	struct ixgbe_ring  *ring;
2188 	int found = 0;
2189 
2190 	if (test_bit(__IXGBE_DOWN, &adapter->state))
2191 		return LL_FLUSH_FAILED;
2192 
2193 	if (!ixgbe_qv_lock_poll(q_vector))
2194 		return LL_FLUSH_BUSY;
2195 
2196 	ixgbe_for_each_ring(ring, q_vector->rx) {
2197 		found = ixgbe_clean_rx_irq(q_vector, ring, 4);
2198 #ifdef BP_EXTENDED_STATS
2199 		if (found)
2200 			ring->stats.cleaned += found;
2201 		else
2202 			ring->stats.misses++;
2203 #endif
2204 		if (found)
2205 			break;
2206 	}
2207 
2208 	ixgbe_qv_unlock_poll(q_vector);
2209 
2210 	return found;
2211 }
2212 #endif	/* CONFIG_NET_RX_BUSY_POLL */
2213 
2214 /**
2215  * ixgbe_configure_msix - Configure MSI-X hardware
2216  * @adapter: board private structure
2217  *
2218  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2219  * interrupts.
2220  **/
2221 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2222 {
2223 	struct ixgbe_q_vector *q_vector;
2224 	int v_idx;
2225 	u32 mask;
2226 
2227 	/* Populate MSIX to EITR Select */
2228 	if (adapter->num_vfs > 32) {
2229 		u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2230 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2231 	}
2232 
2233 	/*
2234 	 * Populate the IVAR table and set the ITR values to the
2235 	 * corresponding register.
2236 	 */
2237 	for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2238 		struct ixgbe_ring *ring;
2239 		q_vector = adapter->q_vector[v_idx];
2240 
2241 		ixgbe_for_each_ring(ring, q_vector->rx)
2242 			ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2243 
2244 		ixgbe_for_each_ring(ring, q_vector->tx)
2245 			ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2246 
2247 		ixgbe_write_eitr(q_vector);
2248 	}
2249 
2250 	switch (adapter->hw.mac.type) {
2251 	case ixgbe_mac_82598EB:
2252 		ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2253 			       v_idx);
2254 		break;
2255 	case ixgbe_mac_82599EB:
2256 	case ixgbe_mac_X540:
2257 		ixgbe_set_ivar(adapter, -1, 1, v_idx);
2258 		break;
2259 	default:
2260 		break;
2261 	}
2262 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2263 
2264 	/* set up to autoclear timer, and the vectors */
2265 	mask = IXGBE_EIMS_ENABLE_MASK;
2266 	mask &= ~(IXGBE_EIMS_OTHER |
2267 		  IXGBE_EIMS_MAILBOX |
2268 		  IXGBE_EIMS_LSC);
2269 
2270 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2271 }
2272 
2273 enum latency_range {
2274 	lowest_latency = 0,
2275 	low_latency = 1,
2276 	bulk_latency = 2,
2277 	latency_invalid = 255
2278 };
2279 
2280 /**
2281  * ixgbe_update_itr - update the dynamic ITR value based on statistics
2282  * @q_vector: structure containing interrupt and ring information
2283  * @ring_container: structure containing ring performance data
2284  *
2285  *      Stores a new ITR value based on packets and byte
2286  *      counts during the last interrupt.  The advantage of per interrupt
2287  *      computation is faster updates and more accurate ITR for the current
2288  *      traffic pattern.  Constants in this function were computed
2289  *      based on theoretical maximum wire speed and thresholds were set based
2290  *      on testing data as well as attempting to minimize response time
2291  *      while increasing bulk throughput.
2292  *      this functionality is controlled by the InterruptThrottleRate module
2293  *      parameter (see ixgbe_param.c)
2294  **/
2295 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2296 			     struct ixgbe_ring_container *ring_container)
2297 {
2298 	int bytes = ring_container->total_bytes;
2299 	int packets = ring_container->total_packets;
2300 	u32 timepassed_us;
2301 	u64 bytes_perint;
2302 	u8 itr_setting = ring_container->itr;
2303 
2304 	if (packets == 0)
2305 		return;
2306 
2307 	/* simple throttlerate management
2308 	 *   0-10MB/s   lowest (100000 ints/s)
2309 	 *  10-20MB/s   low    (20000 ints/s)
2310 	 *  20-1249MB/s bulk   (8000 ints/s)
2311 	 */
2312 	/* what was last interrupt timeslice? */
2313 	timepassed_us = q_vector->itr >> 2;
2314 	if (timepassed_us == 0)
2315 		return;
2316 
2317 	bytes_perint = bytes / timepassed_us; /* bytes/usec */
2318 
2319 	switch (itr_setting) {
2320 	case lowest_latency:
2321 		if (bytes_perint > 10)
2322 			itr_setting = low_latency;
2323 		break;
2324 	case low_latency:
2325 		if (bytes_perint > 20)
2326 			itr_setting = bulk_latency;
2327 		else if (bytes_perint <= 10)
2328 			itr_setting = lowest_latency;
2329 		break;
2330 	case bulk_latency:
2331 		if (bytes_perint <= 20)
2332 			itr_setting = low_latency;
2333 		break;
2334 	}
2335 
2336 	/* clear work counters since we have the values we need */
2337 	ring_container->total_bytes = 0;
2338 	ring_container->total_packets = 0;
2339 
2340 	/* write updated itr to ring container */
2341 	ring_container->itr = itr_setting;
2342 }
2343 
2344 /**
2345  * ixgbe_write_eitr - write EITR register in hardware specific way
2346  * @q_vector: structure containing interrupt and ring information
2347  *
2348  * This function is made to be called by ethtool and by the driver
2349  * when it needs to update EITR registers at runtime.  Hardware
2350  * specific quirks/differences are taken care of here.
2351  */
2352 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2353 {
2354 	struct ixgbe_adapter *adapter = q_vector->adapter;
2355 	struct ixgbe_hw *hw = &adapter->hw;
2356 	int v_idx = q_vector->v_idx;
2357 	u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2358 
2359 	switch (adapter->hw.mac.type) {
2360 	case ixgbe_mac_82598EB:
2361 		/* must write high and low 16 bits to reset counter */
2362 		itr_reg |= (itr_reg << 16);
2363 		break;
2364 	case ixgbe_mac_82599EB:
2365 	case ixgbe_mac_X540:
2366 		/*
2367 		 * set the WDIS bit to not clear the timer bits and cause an
2368 		 * immediate assertion of the interrupt
2369 		 */
2370 		itr_reg |= IXGBE_EITR_CNT_WDIS;
2371 		break;
2372 	default:
2373 		break;
2374 	}
2375 	IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2376 }
2377 
2378 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2379 {
2380 	u32 new_itr = q_vector->itr;
2381 	u8 current_itr;
2382 
2383 	ixgbe_update_itr(q_vector, &q_vector->tx);
2384 	ixgbe_update_itr(q_vector, &q_vector->rx);
2385 
2386 	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
2387 
2388 	switch (current_itr) {
2389 	/* counts and packets in update_itr are dependent on these numbers */
2390 	case lowest_latency:
2391 		new_itr = IXGBE_100K_ITR;
2392 		break;
2393 	case low_latency:
2394 		new_itr = IXGBE_20K_ITR;
2395 		break;
2396 	case bulk_latency:
2397 		new_itr = IXGBE_8K_ITR;
2398 		break;
2399 	default:
2400 		break;
2401 	}
2402 
2403 	if (new_itr != q_vector->itr) {
2404 		/* do an exponential smoothing */
2405 		new_itr = (10 * new_itr * q_vector->itr) /
2406 			  ((9 * new_itr) + q_vector->itr);
2407 
2408 		/* save the algorithm value here */
2409 		q_vector->itr = new_itr;
2410 
2411 		ixgbe_write_eitr(q_vector);
2412 	}
2413 }
2414 
2415 /**
2416  * ixgbe_check_overtemp_subtask - check for over temperature
2417  * @adapter: pointer to adapter
2418  **/
2419 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2420 {
2421 	struct ixgbe_hw *hw = &adapter->hw;
2422 	u32 eicr = adapter->interrupt_event;
2423 
2424 	if (test_bit(__IXGBE_DOWN, &adapter->state))
2425 		return;
2426 
2427 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2428 	    !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2429 		return;
2430 
2431 	adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2432 
2433 	switch (hw->device_id) {
2434 	case IXGBE_DEV_ID_82599_T3_LOM:
2435 		/*
2436 		 * Since the warning interrupt is for both ports
2437 		 * we don't have to check if:
2438 		 *  - This interrupt wasn't for our port.
2439 		 *  - We may have missed the interrupt so always have to
2440 		 *    check if we  got a LSC
2441 		 */
2442 		if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
2443 		    !(eicr & IXGBE_EICR_LSC))
2444 			return;
2445 
2446 		if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2447 			u32 speed;
2448 			bool link_up = false;
2449 
2450 			hw->mac.ops.check_link(hw, &speed, &link_up, false);
2451 
2452 			if (link_up)
2453 				return;
2454 		}
2455 
2456 		/* Check if this is not due to overtemp */
2457 		if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2458 			return;
2459 
2460 		break;
2461 	default:
2462 		if (!(eicr & IXGBE_EICR_GPI_SDP0))
2463 			return;
2464 		break;
2465 	}
2466 	e_crit(drv,
2467 	       "Network adapter has been stopped because it has over heated. "
2468 	       "Restart the computer. If the problem persists, "
2469 	       "power off the system and replace the adapter\n");
2470 
2471 	adapter->interrupt_event = 0;
2472 }
2473 
2474 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2475 {
2476 	struct ixgbe_hw *hw = &adapter->hw;
2477 
2478 	if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2479 	    (eicr & IXGBE_EICR_GPI_SDP1)) {
2480 		e_crit(probe, "Fan has stopped, replace the adapter\n");
2481 		/* write to clear the interrupt */
2482 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2483 	}
2484 }
2485 
2486 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2487 {
2488 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2489 		return;
2490 
2491 	switch (adapter->hw.mac.type) {
2492 	case ixgbe_mac_82599EB:
2493 		/*
2494 		 * Need to check link state so complete overtemp check
2495 		 * on service task
2496 		 */
2497 		if (((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)) &&
2498 		    (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2499 			adapter->interrupt_event = eicr;
2500 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2501 			ixgbe_service_event_schedule(adapter);
2502 			return;
2503 		}
2504 		return;
2505 	case ixgbe_mac_X540:
2506 		if (!(eicr & IXGBE_EICR_TS))
2507 			return;
2508 		break;
2509 	default:
2510 		return;
2511 	}
2512 
2513 	e_crit(drv,
2514 	       "Network adapter has been stopped because it has over heated. "
2515 	       "Restart the computer. If the problem persists, "
2516 	       "power off the system and replace the adapter\n");
2517 }
2518 
2519 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2520 {
2521 	struct ixgbe_hw *hw = &adapter->hw;
2522 
2523 	if (eicr & IXGBE_EICR_GPI_SDP2) {
2524 		/* Clear the interrupt */
2525 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
2526 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2527 			adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2528 			ixgbe_service_event_schedule(adapter);
2529 		}
2530 	}
2531 
2532 	if (eicr & IXGBE_EICR_GPI_SDP1) {
2533 		/* Clear the interrupt */
2534 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2535 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2536 			adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2537 			ixgbe_service_event_schedule(adapter);
2538 		}
2539 	}
2540 }
2541 
2542 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2543 {
2544 	struct ixgbe_hw *hw = &adapter->hw;
2545 
2546 	adapter->lsc_int++;
2547 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2548 	adapter->link_check_timeout = jiffies;
2549 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2550 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2551 		IXGBE_WRITE_FLUSH(hw);
2552 		ixgbe_service_event_schedule(adapter);
2553 	}
2554 }
2555 
2556 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2557 					   u64 qmask)
2558 {
2559 	u32 mask;
2560 	struct ixgbe_hw *hw = &adapter->hw;
2561 
2562 	switch (hw->mac.type) {
2563 	case ixgbe_mac_82598EB:
2564 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2565 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2566 		break;
2567 	case ixgbe_mac_82599EB:
2568 	case ixgbe_mac_X540:
2569 		mask = (qmask & 0xFFFFFFFF);
2570 		if (mask)
2571 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2572 		mask = (qmask >> 32);
2573 		if (mask)
2574 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2575 		break;
2576 	default:
2577 		break;
2578 	}
2579 	/* skip the flush */
2580 }
2581 
2582 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2583 					    u64 qmask)
2584 {
2585 	u32 mask;
2586 	struct ixgbe_hw *hw = &adapter->hw;
2587 
2588 	switch (hw->mac.type) {
2589 	case ixgbe_mac_82598EB:
2590 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2591 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2592 		break;
2593 	case ixgbe_mac_82599EB:
2594 	case ixgbe_mac_X540:
2595 		mask = (qmask & 0xFFFFFFFF);
2596 		if (mask)
2597 			IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2598 		mask = (qmask >> 32);
2599 		if (mask)
2600 			IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2601 		break;
2602 	default:
2603 		break;
2604 	}
2605 	/* skip the flush */
2606 }
2607 
2608 /**
2609  * ixgbe_irq_enable - Enable default interrupt generation settings
2610  * @adapter: board private structure
2611  **/
2612 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2613 				    bool flush)
2614 {
2615 	u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2616 
2617 	/* don't reenable LSC while waiting for link */
2618 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2619 		mask &= ~IXGBE_EIMS_LSC;
2620 
2621 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2622 		switch (adapter->hw.mac.type) {
2623 		case ixgbe_mac_82599EB:
2624 			mask |= IXGBE_EIMS_GPI_SDP0;
2625 			break;
2626 		case ixgbe_mac_X540:
2627 			mask |= IXGBE_EIMS_TS;
2628 			break;
2629 		default:
2630 			break;
2631 		}
2632 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2633 		mask |= IXGBE_EIMS_GPI_SDP1;
2634 	switch (adapter->hw.mac.type) {
2635 	case ixgbe_mac_82599EB:
2636 		mask |= IXGBE_EIMS_GPI_SDP1;
2637 		mask |= IXGBE_EIMS_GPI_SDP2;
2638 	case ixgbe_mac_X540:
2639 		mask |= IXGBE_EIMS_ECC;
2640 		mask |= IXGBE_EIMS_MAILBOX;
2641 		break;
2642 	default:
2643 		break;
2644 	}
2645 
2646 	if (adapter->hw.mac.type == ixgbe_mac_X540)
2647 		mask |= IXGBE_EIMS_TIMESYNC;
2648 
2649 	if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2650 	    !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2651 		mask |= IXGBE_EIMS_FLOW_DIR;
2652 
2653 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2654 	if (queues)
2655 		ixgbe_irq_enable_queues(adapter, ~0);
2656 	if (flush)
2657 		IXGBE_WRITE_FLUSH(&adapter->hw);
2658 }
2659 
2660 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2661 {
2662 	struct ixgbe_adapter *adapter = data;
2663 	struct ixgbe_hw *hw = &adapter->hw;
2664 	u32 eicr;
2665 
2666 	/*
2667 	 * Workaround for Silicon errata.  Use clear-by-write instead
2668 	 * of clear-by-read.  Reading with EICS will return the
2669 	 * interrupt causes without clearing, which later be done
2670 	 * with the write to EICR.
2671 	 */
2672 	eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2673 
2674 	/* The lower 16bits of the EICR register are for the queue interrupts
2675 	 * which should be masked here in order to not accidently clear them if
2676 	 * the bits are high when ixgbe_msix_other is called. There is a race
2677 	 * condition otherwise which results in possible performance loss
2678 	 * especially if the ixgbe_msix_other interrupt is triggering
2679 	 * consistently (as it would when PPS is turned on for the X540 device)
2680 	 */
2681 	eicr &= 0xFFFF0000;
2682 
2683 	IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2684 
2685 	if (eicr & IXGBE_EICR_LSC)
2686 		ixgbe_check_lsc(adapter);
2687 
2688 	if (eicr & IXGBE_EICR_MAILBOX)
2689 		ixgbe_msg_task(adapter);
2690 
2691 	switch (hw->mac.type) {
2692 	case ixgbe_mac_82599EB:
2693 	case ixgbe_mac_X540:
2694 		if (eicr & IXGBE_EICR_ECC) {
2695 			e_info(link, "Received ECC Err, initiating reset\n");
2696 			adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
2697 			ixgbe_service_event_schedule(adapter);
2698 			IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
2699 		}
2700 		/* Handle Flow Director Full threshold interrupt */
2701 		if (eicr & IXGBE_EICR_FLOW_DIR) {
2702 			int reinit_count = 0;
2703 			int i;
2704 			for (i = 0; i < adapter->num_tx_queues; i++) {
2705 				struct ixgbe_ring *ring = adapter->tx_ring[i];
2706 				if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2707 						       &ring->state))
2708 					reinit_count++;
2709 			}
2710 			if (reinit_count) {
2711 				/* no more flow director interrupts until after init */
2712 				IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2713 				adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2714 				ixgbe_service_event_schedule(adapter);
2715 			}
2716 		}
2717 		ixgbe_check_sfp_event(adapter, eicr);
2718 		ixgbe_check_overtemp_event(adapter, eicr);
2719 		break;
2720 	default:
2721 		break;
2722 	}
2723 
2724 	ixgbe_check_fan_failure(adapter, eicr);
2725 
2726 	if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2727 		ixgbe_ptp_check_pps_event(adapter, eicr);
2728 
2729 	/* re-enable the original interrupt state, no lsc, no queues */
2730 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
2731 		ixgbe_irq_enable(adapter, false, false);
2732 
2733 	return IRQ_HANDLED;
2734 }
2735 
2736 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2737 {
2738 	struct ixgbe_q_vector *q_vector = data;
2739 
2740 	/* EIAM disabled interrupts (on this vector) for us */
2741 
2742 	if (q_vector->rx.ring || q_vector->tx.ring)
2743 		napi_schedule(&q_vector->napi);
2744 
2745 	return IRQ_HANDLED;
2746 }
2747 
2748 /**
2749  * ixgbe_poll - NAPI Rx polling callback
2750  * @napi: structure for representing this polling device
2751  * @budget: how many packets driver is allowed to clean
2752  *
2753  * This function is used for legacy and MSI, NAPI mode
2754  **/
2755 int ixgbe_poll(struct napi_struct *napi, int budget)
2756 {
2757 	struct ixgbe_q_vector *q_vector =
2758 				container_of(napi, struct ixgbe_q_vector, napi);
2759 	struct ixgbe_adapter *adapter = q_vector->adapter;
2760 	struct ixgbe_ring *ring;
2761 	int per_ring_budget;
2762 	bool clean_complete = true;
2763 
2764 #ifdef CONFIG_IXGBE_DCA
2765 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2766 		ixgbe_update_dca(q_vector);
2767 #endif
2768 
2769 	ixgbe_for_each_ring(ring, q_vector->tx)
2770 		clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
2771 
2772 	if (!ixgbe_qv_lock_napi(q_vector))
2773 		return budget;
2774 
2775 	/* attempt to distribute budget to each queue fairly, but don't allow
2776 	 * the budget to go below 1 because we'll exit polling */
2777 	if (q_vector->rx.count > 1)
2778 		per_ring_budget = max(budget/q_vector->rx.count, 1);
2779 	else
2780 		per_ring_budget = budget;
2781 
2782 	ixgbe_for_each_ring(ring, q_vector->rx)
2783 		clean_complete &= (ixgbe_clean_rx_irq(q_vector, ring,
2784 				   per_ring_budget) < per_ring_budget);
2785 
2786 	ixgbe_qv_unlock_napi(q_vector);
2787 	/* If all work not completed, return budget and keep polling */
2788 	if (!clean_complete)
2789 		return budget;
2790 
2791 	/* all work done, exit the polling mode */
2792 	napi_complete(napi);
2793 	if (adapter->rx_itr_setting & 1)
2794 		ixgbe_set_itr(q_vector);
2795 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
2796 		ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2797 
2798 	return 0;
2799 }
2800 
2801 /**
2802  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2803  * @adapter: board private structure
2804  *
2805  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2806  * interrupts from the kernel.
2807  **/
2808 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2809 {
2810 	struct net_device *netdev = adapter->netdev;
2811 	int vector, err;
2812 	int ri = 0, ti = 0;
2813 
2814 	for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2815 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2816 		struct msix_entry *entry = &adapter->msix_entries[vector];
2817 
2818 		if (q_vector->tx.ring && q_vector->rx.ring) {
2819 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2820 				 "%s-%s-%d", netdev->name, "TxRx", ri++);
2821 			ti++;
2822 		} else if (q_vector->rx.ring) {
2823 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2824 				 "%s-%s-%d", netdev->name, "rx", ri++);
2825 		} else if (q_vector->tx.ring) {
2826 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2827 				 "%s-%s-%d", netdev->name, "tx", ti++);
2828 		} else {
2829 			/* skip this unused q_vector */
2830 			continue;
2831 		}
2832 		err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2833 				  q_vector->name, q_vector);
2834 		if (err) {
2835 			e_err(probe, "request_irq failed for MSIX interrupt "
2836 			      "Error: %d\n", err);
2837 			goto free_queue_irqs;
2838 		}
2839 		/* If Flow Director is enabled, set interrupt affinity */
2840 		if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2841 			/* assign the mask for this irq */
2842 			irq_set_affinity_hint(entry->vector,
2843 					      &q_vector->affinity_mask);
2844 		}
2845 	}
2846 
2847 	err = request_irq(adapter->msix_entries[vector].vector,
2848 			  ixgbe_msix_other, 0, netdev->name, adapter);
2849 	if (err) {
2850 		e_err(probe, "request_irq for msix_other failed: %d\n", err);
2851 		goto free_queue_irqs;
2852 	}
2853 
2854 	return 0;
2855 
2856 free_queue_irqs:
2857 	while (vector) {
2858 		vector--;
2859 		irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2860 				      NULL);
2861 		free_irq(adapter->msix_entries[vector].vector,
2862 			 adapter->q_vector[vector]);
2863 	}
2864 	adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2865 	pci_disable_msix(adapter->pdev);
2866 	kfree(adapter->msix_entries);
2867 	adapter->msix_entries = NULL;
2868 	return err;
2869 }
2870 
2871 /**
2872  * ixgbe_intr - legacy mode Interrupt Handler
2873  * @irq: interrupt number
2874  * @data: pointer to a network interface device structure
2875  **/
2876 static irqreturn_t ixgbe_intr(int irq, void *data)
2877 {
2878 	struct ixgbe_adapter *adapter = data;
2879 	struct ixgbe_hw *hw = &adapter->hw;
2880 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2881 	u32 eicr;
2882 
2883 	/*
2884 	 * Workaround for silicon errata #26 on 82598.  Mask the interrupt
2885 	 * before the read of EICR.
2886 	 */
2887 	IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2888 
2889 	/* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2890 	 * therefore no explicit interrupt disable is necessary */
2891 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2892 	if (!eicr) {
2893 		/*
2894 		 * shared interrupt alert!
2895 		 * make sure interrupts are enabled because the read will
2896 		 * have disabled interrupts due to EIAM
2897 		 * finish the workaround of silicon errata on 82598.  Unmask
2898 		 * the interrupt that we masked before the EICR read.
2899 		 */
2900 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
2901 			ixgbe_irq_enable(adapter, true, true);
2902 		return IRQ_NONE;	/* Not our interrupt */
2903 	}
2904 
2905 	if (eicr & IXGBE_EICR_LSC)
2906 		ixgbe_check_lsc(adapter);
2907 
2908 	switch (hw->mac.type) {
2909 	case ixgbe_mac_82599EB:
2910 		ixgbe_check_sfp_event(adapter, eicr);
2911 		/* Fall through */
2912 	case ixgbe_mac_X540:
2913 		if (eicr & IXGBE_EICR_ECC) {
2914 			e_info(link, "Received ECC Err, initiating reset\n");
2915 			adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
2916 			ixgbe_service_event_schedule(adapter);
2917 			IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
2918 		}
2919 		ixgbe_check_overtemp_event(adapter, eicr);
2920 		break;
2921 	default:
2922 		break;
2923 	}
2924 
2925 	ixgbe_check_fan_failure(adapter, eicr);
2926 	if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2927 		ixgbe_ptp_check_pps_event(adapter, eicr);
2928 
2929 	/* would disable interrupts here but EIAM disabled it */
2930 	napi_schedule(&q_vector->napi);
2931 
2932 	/*
2933 	 * re-enable link(maybe) and non-queue interrupts, no flush.
2934 	 * ixgbe_poll will re-enable the queue interrupts
2935 	 */
2936 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
2937 		ixgbe_irq_enable(adapter, false, false);
2938 
2939 	return IRQ_HANDLED;
2940 }
2941 
2942 /**
2943  * ixgbe_request_irq - initialize interrupts
2944  * @adapter: board private structure
2945  *
2946  * Attempts to configure interrupts using the best available
2947  * capabilities of the hardware and kernel.
2948  **/
2949 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2950 {
2951 	struct net_device *netdev = adapter->netdev;
2952 	int err;
2953 
2954 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2955 		err = ixgbe_request_msix_irqs(adapter);
2956 	else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2957 		err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2958 				  netdev->name, adapter);
2959 	else
2960 		err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2961 				  netdev->name, adapter);
2962 
2963 	if (err)
2964 		e_err(probe, "request_irq failed, Error %d\n", err);
2965 
2966 	return err;
2967 }
2968 
2969 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2970 {
2971 	int vector;
2972 
2973 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2974 		free_irq(adapter->pdev->irq, adapter);
2975 		return;
2976 	}
2977 
2978 	for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2979 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2980 		struct msix_entry *entry = &adapter->msix_entries[vector];
2981 
2982 		/* free only the irqs that were actually requested */
2983 		if (!q_vector->rx.ring && !q_vector->tx.ring)
2984 			continue;
2985 
2986 		/* clear the affinity_mask in the IRQ descriptor */
2987 		irq_set_affinity_hint(entry->vector, NULL);
2988 
2989 		free_irq(entry->vector, q_vector);
2990 	}
2991 
2992 	free_irq(adapter->msix_entries[vector++].vector, adapter);
2993 }
2994 
2995 /**
2996  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2997  * @adapter: board private structure
2998  **/
2999 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3000 {
3001 	switch (adapter->hw.mac.type) {
3002 	case ixgbe_mac_82598EB:
3003 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3004 		break;
3005 	case ixgbe_mac_82599EB:
3006 	case ixgbe_mac_X540:
3007 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3008 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3009 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3010 		break;
3011 	default:
3012 		break;
3013 	}
3014 	IXGBE_WRITE_FLUSH(&adapter->hw);
3015 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3016 		int vector;
3017 
3018 		for (vector = 0; vector < adapter->num_q_vectors; vector++)
3019 			synchronize_irq(adapter->msix_entries[vector].vector);
3020 
3021 		synchronize_irq(adapter->msix_entries[vector++].vector);
3022 	} else {
3023 		synchronize_irq(adapter->pdev->irq);
3024 	}
3025 }
3026 
3027 /**
3028  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3029  *
3030  **/
3031 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3032 {
3033 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3034 
3035 	ixgbe_write_eitr(q_vector);
3036 
3037 	ixgbe_set_ivar(adapter, 0, 0, 0);
3038 	ixgbe_set_ivar(adapter, 1, 0, 0);
3039 
3040 	e_info(hw, "Legacy interrupt IVAR setup done\n");
3041 }
3042 
3043 /**
3044  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3045  * @adapter: board private structure
3046  * @ring: structure containing ring specific data
3047  *
3048  * Configure the Tx descriptor ring after a reset.
3049  **/
3050 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3051 			     struct ixgbe_ring *ring)
3052 {
3053 	struct ixgbe_hw *hw = &adapter->hw;
3054 	u64 tdba = ring->dma;
3055 	int wait_loop = 10;
3056 	u32 txdctl = IXGBE_TXDCTL_ENABLE;
3057 	u8 reg_idx = ring->reg_idx;
3058 
3059 	/* disable queue to avoid issues while updating state */
3060 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3061 	IXGBE_WRITE_FLUSH(hw);
3062 
3063 	IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3064 			(tdba & DMA_BIT_MASK(32)));
3065 	IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3066 	IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3067 			ring->count * sizeof(union ixgbe_adv_tx_desc));
3068 	IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3069 	IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3070 	ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3071 
3072 	/*
3073 	 * set WTHRESH to encourage burst writeback, it should not be set
3074 	 * higher than 1 when:
3075 	 * - ITR is 0 as it could cause false TX hangs
3076 	 * - ITR is set to > 100k int/sec and BQL is enabled
3077 	 *
3078 	 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3079 	 * to or less than the number of on chip descriptors, which is
3080 	 * currently 40.
3081 	 */
3082 #if IS_ENABLED(CONFIG_BQL)
3083 	if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3084 #else
3085 	if (!ring->q_vector || (ring->q_vector->itr < 8))
3086 #endif
3087 		txdctl |= (1 << 16);	/* WTHRESH = 1 */
3088 	else
3089 		txdctl |= (8 << 16);	/* WTHRESH = 8 */
3090 
3091 	/*
3092 	 * Setting PTHRESH to 32 both improves performance
3093 	 * and avoids a TX hang with DFP enabled
3094 	 */
3095 	txdctl |= (1 << 8) |	/* HTHRESH = 1 */
3096 		   32;		/* PTHRESH = 32 */
3097 
3098 	/* reinitialize flowdirector state */
3099 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3100 		ring->atr_sample_rate = adapter->atr_sample_rate;
3101 		ring->atr_count = 0;
3102 		set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3103 	} else {
3104 		ring->atr_sample_rate = 0;
3105 	}
3106 
3107 	/* initialize XPS */
3108 	if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3109 		struct ixgbe_q_vector *q_vector = ring->q_vector;
3110 
3111 		if (q_vector)
3112 			netif_set_xps_queue(ring->netdev,
3113 					    &q_vector->affinity_mask,
3114 					    ring->queue_index);
3115 	}
3116 
3117 	clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3118 
3119 	/* enable queue */
3120 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3121 
3122 	/* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3123 	if (hw->mac.type == ixgbe_mac_82598EB &&
3124 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3125 		return;
3126 
3127 	/* poll to verify queue is enabled */
3128 	do {
3129 		usleep_range(1000, 2000);
3130 		txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3131 	} while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3132 	if (!wait_loop)
3133 		e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
3134 }
3135 
3136 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3137 {
3138 	struct ixgbe_hw *hw = &adapter->hw;
3139 	u32 rttdcs, mtqc;
3140 	u8 tcs = netdev_get_num_tc(adapter->netdev);
3141 
3142 	if (hw->mac.type == ixgbe_mac_82598EB)
3143 		return;
3144 
3145 	/* disable the arbiter while setting MTQC */
3146 	rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3147 	rttdcs |= IXGBE_RTTDCS_ARBDIS;
3148 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3149 
3150 	/* set transmit pool layout */
3151 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3152 		mtqc = IXGBE_MTQC_VT_ENA;
3153 		if (tcs > 4)
3154 			mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3155 		else if (tcs > 1)
3156 			mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3157 		else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3158 			mtqc |= IXGBE_MTQC_32VF;
3159 		else
3160 			mtqc |= IXGBE_MTQC_64VF;
3161 	} else {
3162 		if (tcs > 4)
3163 			mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3164 		else if (tcs > 1)
3165 			mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3166 		else
3167 			mtqc = IXGBE_MTQC_64Q_1PB;
3168 	}
3169 
3170 	IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3171 
3172 	/* Enable Security TX Buffer IFG for multiple pb */
3173 	if (tcs) {
3174 		u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3175 		sectx |= IXGBE_SECTX_DCB;
3176 		IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3177 	}
3178 
3179 	/* re-enable the arbiter */
3180 	rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3181 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3182 }
3183 
3184 /**
3185  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3186  * @adapter: board private structure
3187  *
3188  * Configure the Tx unit of the MAC after a reset.
3189  **/
3190 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3191 {
3192 	struct ixgbe_hw *hw = &adapter->hw;
3193 	u32 dmatxctl;
3194 	u32 i;
3195 
3196 	ixgbe_setup_mtqc(adapter);
3197 
3198 	if (hw->mac.type != ixgbe_mac_82598EB) {
3199 		/* DMATXCTL.EN must be before Tx queues are enabled */
3200 		dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3201 		dmatxctl |= IXGBE_DMATXCTL_TE;
3202 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3203 	}
3204 
3205 	/* Setup the HW Tx Head and Tail descriptor pointers */
3206 	for (i = 0; i < adapter->num_tx_queues; i++)
3207 		ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3208 }
3209 
3210 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3211 				 struct ixgbe_ring *ring)
3212 {
3213 	struct ixgbe_hw *hw = &adapter->hw;
3214 	u8 reg_idx = ring->reg_idx;
3215 	u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3216 
3217 	srrctl |= IXGBE_SRRCTL_DROP_EN;
3218 
3219 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3220 }
3221 
3222 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3223 				  struct ixgbe_ring *ring)
3224 {
3225 	struct ixgbe_hw *hw = &adapter->hw;
3226 	u8 reg_idx = ring->reg_idx;
3227 	u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3228 
3229 	srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3230 
3231 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3232 }
3233 
3234 #ifdef CONFIG_IXGBE_DCB
3235 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3236 #else
3237 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3238 #endif
3239 {
3240 	int i;
3241 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3242 
3243 	if (adapter->ixgbe_ieee_pfc)
3244 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3245 
3246 	/*
3247 	 * We should set the drop enable bit if:
3248 	 *  SR-IOV is enabled
3249 	 *   or
3250 	 *  Number of Rx queues > 1 and flow control is disabled
3251 	 *
3252 	 *  This allows us to avoid head of line blocking for security
3253 	 *  and performance reasons.
3254 	 */
3255 	if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3256 	    !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3257 		for (i = 0; i < adapter->num_rx_queues; i++)
3258 			ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3259 	} else {
3260 		for (i = 0; i < adapter->num_rx_queues; i++)
3261 			ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3262 	}
3263 }
3264 
3265 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3266 
3267 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3268 				   struct ixgbe_ring *rx_ring)
3269 {
3270 	struct ixgbe_hw *hw = &adapter->hw;
3271 	u32 srrctl;
3272 	u8 reg_idx = rx_ring->reg_idx;
3273 
3274 	if (hw->mac.type == ixgbe_mac_82598EB) {
3275 		u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3276 
3277 		/*
3278 		 * if VMDq is not active we must program one srrctl register
3279 		 * per RSS queue since we have enabled RDRXCTL.MVMEN
3280 		 */
3281 		reg_idx &= mask;
3282 	}
3283 
3284 	/* configure header buffer length, needed for RSC */
3285 	srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3286 
3287 	/* configure the packet buffer length */
3288 	srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3289 
3290 	/* configure descriptor type */
3291 	srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3292 
3293 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3294 }
3295 
3296 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3297 {
3298 	struct ixgbe_hw *hw = &adapter->hw;
3299 	static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
3300 			  0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
3301 			  0x6A3E67EA, 0x14364D17, 0x3BED200D};
3302 	u32 mrqc = 0, reta = 0;
3303 	u32 rxcsum;
3304 	int i, j;
3305 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3306 
3307 	/*
3308 	 * Program table for at least 2 queues w/ SR-IOV so that VFs can
3309 	 * make full use of any rings they may have.  We will use the
3310 	 * PSRTYPE register to control how many rings we use within the PF.
3311 	 */
3312 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 2))
3313 		rss_i = 2;
3314 
3315 	/* Fill out hash function seeds */
3316 	for (i = 0; i < 10; i++)
3317 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
3318 
3319 	/* Fill out redirection table */
3320 	for (i = 0, j = 0; i < 128; i++, j++) {
3321 		if (j == rss_i)
3322 			j = 0;
3323 		/* reta = 4-byte sliding window of
3324 		 * 0x00..(indices-1)(indices-1)00..etc. */
3325 		reta = (reta << 8) | (j * 0x11);
3326 		if ((i & 3) == 3)
3327 			IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3328 	}
3329 
3330 	/* Disable indicating checksum in descriptor, enables RSS hash */
3331 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3332 	rxcsum |= IXGBE_RXCSUM_PCSD;
3333 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3334 
3335 	if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3336 		if (adapter->ring_feature[RING_F_RSS].mask)
3337 			mrqc = IXGBE_MRQC_RSSEN;
3338 	} else {
3339 		u8 tcs = netdev_get_num_tc(adapter->netdev);
3340 
3341 		if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3342 			if (tcs > 4)
3343 				mrqc = IXGBE_MRQC_VMDQRT8TCEN;	/* 8 TCs */
3344 			else if (tcs > 1)
3345 				mrqc = IXGBE_MRQC_VMDQRT4TCEN;	/* 4 TCs */
3346 			else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3347 				mrqc = IXGBE_MRQC_VMDQRSS32EN;
3348 			else
3349 				mrqc = IXGBE_MRQC_VMDQRSS64EN;
3350 		} else {
3351 			if (tcs > 4)
3352 				mrqc = IXGBE_MRQC_RTRSS8TCEN;
3353 			else if (tcs > 1)
3354 				mrqc = IXGBE_MRQC_RTRSS4TCEN;
3355 			else
3356 				mrqc = IXGBE_MRQC_RSSEN;
3357 		}
3358 	}
3359 
3360 	/* Perform hash on these packet types */
3361 	mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3362 		IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3363 		IXGBE_MRQC_RSS_FIELD_IPV6 |
3364 		IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3365 
3366 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3367 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3368 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3369 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3370 
3371 	IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3372 }
3373 
3374 /**
3375  * ixgbe_configure_rscctl - enable RSC for the indicated ring
3376  * @adapter:    address of board private structure
3377  * @index:      index of ring to set
3378  **/
3379 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
3380 				   struct ixgbe_ring *ring)
3381 {
3382 	struct ixgbe_hw *hw = &adapter->hw;
3383 	u32 rscctrl;
3384 	u8 reg_idx = ring->reg_idx;
3385 
3386 	if (!ring_is_rsc_enabled(ring))
3387 		return;
3388 
3389 	rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
3390 	rscctrl |= IXGBE_RSCCTL_RSCEN;
3391 	/*
3392 	 * we must limit the number of descriptors so that the
3393 	 * total size of max desc * buf_len is not greater
3394 	 * than 65536
3395 	 */
3396 	rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
3397 	IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
3398 }
3399 
3400 #define IXGBE_MAX_RX_DESC_POLL 10
3401 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3402 				       struct ixgbe_ring *ring)
3403 {
3404 	struct ixgbe_hw *hw = &adapter->hw;
3405 	int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3406 	u32 rxdctl;
3407 	u8 reg_idx = ring->reg_idx;
3408 
3409 	if (ixgbe_removed(hw->hw_addr))
3410 		return;
3411 	/* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3412 	if (hw->mac.type == ixgbe_mac_82598EB &&
3413 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3414 		return;
3415 
3416 	do {
3417 		usleep_range(1000, 2000);
3418 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3419 	} while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3420 
3421 	if (!wait_loop) {
3422 		e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3423 		      "the polling period\n", reg_idx);
3424 	}
3425 }
3426 
3427 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3428 			    struct ixgbe_ring *ring)
3429 {
3430 	struct ixgbe_hw *hw = &adapter->hw;
3431 	int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3432 	u32 rxdctl;
3433 	u8 reg_idx = ring->reg_idx;
3434 
3435 	if (ixgbe_removed(hw->hw_addr))
3436 		return;
3437 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3438 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3439 
3440 	/* write value back with RXDCTL.ENABLE bit cleared */
3441 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3442 
3443 	if (hw->mac.type == ixgbe_mac_82598EB &&
3444 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3445 		return;
3446 
3447 	/* the hardware may take up to 100us to really disable the rx queue */
3448 	do {
3449 		udelay(10);
3450 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3451 	} while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3452 
3453 	if (!wait_loop) {
3454 		e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3455 		      "the polling period\n", reg_idx);
3456 	}
3457 }
3458 
3459 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3460 			     struct ixgbe_ring *ring)
3461 {
3462 	struct ixgbe_hw *hw = &adapter->hw;
3463 	u64 rdba = ring->dma;
3464 	u32 rxdctl;
3465 	u8 reg_idx = ring->reg_idx;
3466 
3467 	/* disable queue to avoid issues while updating state */
3468 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3469 	ixgbe_disable_rx_queue(adapter, ring);
3470 
3471 	IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3472 	IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3473 	IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3474 			ring->count * sizeof(union ixgbe_adv_rx_desc));
3475 	IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3476 	IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3477 	ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
3478 
3479 	ixgbe_configure_srrctl(adapter, ring);
3480 	ixgbe_configure_rscctl(adapter, ring);
3481 
3482 	if (hw->mac.type == ixgbe_mac_82598EB) {
3483 		/*
3484 		 * enable cache line friendly hardware writes:
3485 		 * PTHRESH=32 descriptors (half the internal cache),
3486 		 * this also removes ugly rx_no_buffer_count increment
3487 		 * HTHRESH=4 descriptors (to minimize latency on fetch)
3488 		 * WTHRESH=8 burst writeback up to two cache lines
3489 		 */
3490 		rxdctl &= ~0x3FFFFF;
3491 		rxdctl |=  0x080420;
3492 	}
3493 
3494 	/* enable receive descriptor ring */
3495 	rxdctl |= IXGBE_RXDCTL_ENABLE;
3496 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3497 
3498 	ixgbe_rx_desc_queue_enable(adapter, ring);
3499 	ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3500 }
3501 
3502 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3503 {
3504 	struct ixgbe_hw *hw = &adapter->hw;
3505 	int rss_i = adapter->ring_feature[RING_F_RSS].indices;
3506 	u16 pool;
3507 
3508 	/* PSRTYPE must be initialized in non 82598 adapters */
3509 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3510 		      IXGBE_PSRTYPE_UDPHDR |
3511 		      IXGBE_PSRTYPE_IPV4HDR |
3512 		      IXGBE_PSRTYPE_L2HDR |
3513 		      IXGBE_PSRTYPE_IPV6HDR;
3514 
3515 	if (hw->mac.type == ixgbe_mac_82598EB)
3516 		return;
3517 
3518 	if (rss_i > 3)
3519 		psrtype |= 2 << 29;
3520 	else if (rss_i > 1)
3521 		psrtype |= 1 << 29;
3522 
3523 	for_each_set_bit(pool, &adapter->fwd_bitmask, 32)
3524 		IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
3525 }
3526 
3527 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3528 {
3529 	struct ixgbe_hw *hw = &adapter->hw;
3530 	u32 reg_offset, vf_shift;
3531 	u32 gcr_ext, vmdctl;
3532 	int i;
3533 
3534 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3535 		return;
3536 
3537 	vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3538 	vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
3539 	vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
3540 	vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
3541 	vmdctl |= IXGBE_VT_CTL_REPLEN;
3542 	IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
3543 
3544 	vf_shift = VMDQ_P(0) % 32;
3545 	reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
3546 
3547 	/* Enable only the PF's pool for Tx/Rx */
3548 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (~0) << vf_shift);
3549 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
3550 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (~0) << vf_shift);
3551 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
3552 	if (adapter->flags2 & IXGBE_FLAG2_BRIDGE_MODE_VEB)
3553 		IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3554 
3555 	/* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3556 	hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
3557 
3558 	/*
3559 	 * Set up VF register offsets for selected VT Mode,
3560 	 * i.e. 32 or 64 VFs for SR-IOV
3561 	 */
3562 	switch (adapter->ring_feature[RING_F_VMDQ].mask) {
3563 	case IXGBE_82599_VMDQ_8Q_MASK:
3564 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
3565 		break;
3566 	case IXGBE_82599_VMDQ_4Q_MASK:
3567 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
3568 		break;
3569 	default:
3570 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
3571 		break;
3572 	}
3573 
3574 	IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3575 
3576 
3577 	/* Enable MAC Anti-Spoofing */
3578 	hw->mac.ops.set_mac_anti_spoofing(hw, (adapter->num_vfs != 0),
3579 					  adapter->num_vfs);
3580 	/* For VFs that have spoof checking turned off */
3581 	for (i = 0; i < adapter->num_vfs; i++) {
3582 		if (!adapter->vfinfo[i].spoofchk_enabled)
3583 			ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
3584 	}
3585 }
3586 
3587 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3588 {
3589 	struct ixgbe_hw *hw = &adapter->hw;
3590 	struct net_device *netdev = adapter->netdev;
3591 	int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3592 	struct ixgbe_ring *rx_ring;
3593 	int i;
3594 	u32 mhadd, hlreg0;
3595 
3596 #ifdef IXGBE_FCOE
3597 	/* adjust max frame to be able to do baby jumbo for FCoE */
3598 	if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3599 	    (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3600 		max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3601 
3602 #endif /* IXGBE_FCOE */
3603 
3604 	/* adjust max frame to be at least the size of a standard frame */
3605 	if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
3606 		max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
3607 
3608 	mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3609 	if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3610 		mhadd &= ~IXGBE_MHADD_MFS_MASK;
3611 		mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3612 
3613 		IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3614 	}
3615 
3616 	hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3617 	/* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3618 	hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3619 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3620 
3621 	/*
3622 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
3623 	 * the Base and Length of the Rx Descriptor Ring
3624 	 */
3625 	for (i = 0; i < adapter->num_rx_queues; i++) {
3626 		rx_ring = adapter->rx_ring[i];
3627 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3628 			set_ring_rsc_enabled(rx_ring);
3629 		else
3630 			clear_ring_rsc_enabled(rx_ring);
3631 	}
3632 }
3633 
3634 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3635 {
3636 	struct ixgbe_hw *hw = &adapter->hw;
3637 	u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3638 
3639 	switch (hw->mac.type) {
3640 	case ixgbe_mac_82598EB:
3641 		/*
3642 		 * For VMDq support of different descriptor types or
3643 		 * buffer sizes through the use of multiple SRRCTL
3644 		 * registers, RDRXCTL.MVMEN must be set to 1
3645 		 *
3646 		 * also, the manual doesn't mention it clearly but DCA hints
3647 		 * will only use queue 0's tags unless this bit is set.  Side
3648 		 * effects of setting this bit are only that SRRCTL must be
3649 		 * fully programmed [0..15]
3650 		 */
3651 		rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3652 		break;
3653 	case ixgbe_mac_82599EB:
3654 	case ixgbe_mac_X540:
3655 		/* Disable RSC for ACK packets */
3656 		IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3657 		   (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3658 		rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3659 		/* hardware requires some bits to be set by default */
3660 		rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3661 		rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3662 		break;
3663 	default:
3664 		/* We should do nothing since we don't know this hardware */
3665 		return;
3666 	}
3667 
3668 	IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3669 }
3670 
3671 /**
3672  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3673  * @adapter: board private structure
3674  *
3675  * Configure the Rx unit of the MAC after a reset.
3676  **/
3677 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3678 {
3679 	struct ixgbe_hw *hw = &adapter->hw;
3680 	int i;
3681 	u32 rxctrl, rfctl;
3682 
3683 	/* disable receives while setting up the descriptors */
3684 	rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3685 	IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3686 
3687 	ixgbe_setup_psrtype(adapter);
3688 	ixgbe_setup_rdrxctl(adapter);
3689 
3690 	/* RSC Setup */
3691 	rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
3692 	rfctl &= ~IXGBE_RFCTL_RSC_DIS;
3693 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
3694 		rfctl |= IXGBE_RFCTL_RSC_DIS;
3695 	IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
3696 
3697 	/* Program registers for the distribution of queues */
3698 	ixgbe_setup_mrqc(adapter);
3699 
3700 	/* set_rx_buffer_len must be called before ring initialization */
3701 	ixgbe_set_rx_buffer_len(adapter);
3702 
3703 	/*
3704 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
3705 	 * the Base and Length of the Rx Descriptor Ring
3706 	 */
3707 	for (i = 0; i < adapter->num_rx_queues; i++)
3708 		ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3709 
3710 	/* disable drop enable for 82598 parts */
3711 	if (hw->mac.type == ixgbe_mac_82598EB)
3712 		rxctrl |= IXGBE_RXCTRL_DMBYPS;
3713 
3714 	/* enable all receives */
3715 	rxctrl |= IXGBE_RXCTRL_RXEN;
3716 	hw->mac.ops.enable_rx_dma(hw, rxctrl);
3717 }
3718 
3719 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
3720 				 __be16 proto, u16 vid)
3721 {
3722 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3723 	struct ixgbe_hw *hw = &adapter->hw;
3724 
3725 	/* add VID to filter table */
3726 	hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true);
3727 	set_bit(vid, adapter->active_vlans);
3728 
3729 	return 0;
3730 }
3731 
3732 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
3733 				  __be16 proto, u16 vid)
3734 {
3735 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3736 	struct ixgbe_hw *hw = &adapter->hw;
3737 
3738 	/* remove VID from filter table */
3739 	hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), false);
3740 	clear_bit(vid, adapter->active_vlans);
3741 
3742 	return 0;
3743 }
3744 
3745 /**
3746  * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3747  * @adapter: driver data
3748  */
3749 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3750 {
3751 	struct ixgbe_hw *hw = &adapter->hw;
3752 	u32 vlnctrl;
3753 
3754 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3755 	vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3756 	IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3757 }
3758 
3759 /**
3760  * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3761  * @adapter: driver data
3762  */
3763 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3764 {
3765 	struct ixgbe_hw *hw = &adapter->hw;
3766 	u32 vlnctrl;
3767 
3768 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3769 	vlnctrl |= IXGBE_VLNCTRL_VFE;
3770 	vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3771 	IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3772 }
3773 
3774 /**
3775  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3776  * @adapter: driver data
3777  */
3778 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3779 {
3780 	struct ixgbe_hw *hw = &adapter->hw;
3781 	u32 vlnctrl;
3782 	int i, j;
3783 
3784 	switch (hw->mac.type) {
3785 	case ixgbe_mac_82598EB:
3786 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3787 		vlnctrl &= ~IXGBE_VLNCTRL_VME;
3788 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3789 		break;
3790 	case ixgbe_mac_82599EB:
3791 	case ixgbe_mac_X540:
3792 		for (i = 0; i < adapter->num_rx_queues; i++) {
3793 			struct ixgbe_ring *ring = adapter->rx_ring[i];
3794 
3795 			if (ring->l2_accel_priv)
3796 				continue;
3797 			j = ring->reg_idx;
3798 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3799 			vlnctrl &= ~IXGBE_RXDCTL_VME;
3800 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3801 		}
3802 		break;
3803 	default:
3804 		break;
3805 	}
3806 }
3807 
3808 /**
3809  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3810  * @adapter: driver data
3811  */
3812 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3813 {
3814 	struct ixgbe_hw *hw = &adapter->hw;
3815 	u32 vlnctrl;
3816 	int i, j;
3817 
3818 	switch (hw->mac.type) {
3819 	case ixgbe_mac_82598EB:
3820 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3821 		vlnctrl |= IXGBE_VLNCTRL_VME;
3822 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3823 		break;
3824 	case ixgbe_mac_82599EB:
3825 	case ixgbe_mac_X540:
3826 		for (i = 0; i < adapter->num_rx_queues; i++) {
3827 			struct ixgbe_ring *ring = adapter->rx_ring[i];
3828 
3829 			if (ring->l2_accel_priv)
3830 				continue;
3831 			j = ring->reg_idx;
3832 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3833 			vlnctrl |= IXGBE_RXDCTL_VME;
3834 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3835 		}
3836 		break;
3837 	default:
3838 		break;
3839 	}
3840 }
3841 
3842 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3843 {
3844 	u16 vid;
3845 
3846 	ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
3847 
3848 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3849 		ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
3850 }
3851 
3852 /**
3853  * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3854  * @netdev: network interface device structure
3855  *
3856  * Writes unicast address list to the RAR table.
3857  * Returns: -ENOMEM on failure/insufficient address space
3858  *                0 on no addresses written
3859  *                X on writing X addresses to the RAR table
3860  **/
3861 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3862 {
3863 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3864 	struct ixgbe_hw *hw = &adapter->hw;
3865 	unsigned int rar_entries = hw->mac.num_rar_entries - 1;
3866 	int count = 0;
3867 
3868 	/* In SR-IOV/VMDQ modes significantly less RAR entries are available */
3869 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3870 		rar_entries = IXGBE_MAX_PF_MACVLANS - 1;
3871 
3872 	/* return ENOMEM indicating insufficient memory for addresses */
3873 	if (netdev_uc_count(netdev) > rar_entries)
3874 		return -ENOMEM;
3875 
3876 	if (!netdev_uc_empty(netdev)) {
3877 		struct netdev_hw_addr *ha;
3878 		/* return error if we do not support writing to RAR table */
3879 		if (!hw->mac.ops.set_rar)
3880 			return -ENOMEM;
3881 
3882 		netdev_for_each_uc_addr(ha, netdev) {
3883 			if (!rar_entries)
3884 				break;
3885 			hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3886 					    VMDQ_P(0), IXGBE_RAH_AV);
3887 			count++;
3888 		}
3889 	}
3890 	/* write the addresses in reverse order to avoid write combining */
3891 	for (; rar_entries > 0 ; rar_entries--)
3892 		hw->mac.ops.clear_rar(hw, rar_entries);
3893 
3894 	return count;
3895 }
3896 
3897 /**
3898  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3899  * @netdev: network interface device structure
3900  *
3901  * The set_rx_method entry point is called whenever the unicast/multicast
3902  * address list or the network interface flags are updated.  This routine is
3903  * responsible for configuring the hardware for proper unicast, multicast and
3904  * promiscuous mode.
3905  **/
3906 void ixgbe_set_rx_mode(struct net_device *netdev)
3907 {
3908 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3909 	struct ixgbe_hw *hw = &adapter->hw;
3910 	u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3911 	int count;
3912 
3913 	/* Check for Promiscuous and All Multicast modes */
3914 
3915 	fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3916 
3917 	/* set all bits that we expect to always be set */
3918 	fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
3919 	fctrl |= IXGBE_FCTRL_BAM;
3920 	fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3921 	fctrl |= IXGBE_FCTRL_PMCF;
3922 
3923 	/* clear the bits we are changing the status of */
3924 	fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3925 
3926 	if (netdev->flags & IFF_PROMISC) {
3927 		hw->addr_ctrl.user_set_promisc = true;
3928 		fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3929 		vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3930 		/* Only disable hardware filter vlans in promiscuous mode
3931 		 * if SR-IOV and VMDQ are disabled - otherwise ensure
3932 		 * that hardware VLAN filters remain enabled.
3933 		 */
3934 		if (!(adapter->flags & (IXGBE_FLAG_VMDQ_ENABLED |
3935 					IXGBE_FLAG_SRIOV_ENABLED)))
3936 			ixgbe_vlan_filter_disable(adapter);
3937 		else
3938 			ixgbe_vlan_filter_enable(adapter);
3939 	} else {
3940 		if (netdev->flags & IFF_ALLMULTI) {
3941 			fctrl |= IXGBE_FCTRL_MPE;
3942 			vmolr |= IXGBE_VMOLR_MPE;
3943 		}
3944 		ixgbe_vlan_filter_enable(adapter);
3945 		hw->addr_ctrl.user_set_promisc = false;
3946 	}
3947 
3948 	/*
3949 	 * Write addresses to available RAR registers, if there is not
3950 	 * sufficient space to store all the addresses then enable
3951 	 * unicast promiscuous mode
3952 	 */
3953 	count = ixgbe_write_uc_addr_list(netdev);
3954 	if (count < 0) {
3955 		fctrl |= IXGBE_FCTRL_UPE;
3956 		vmolr |= IXGBE_VMOLR_ROPE;
3957 	}
3958 
3959 	/* Write addresses to the MTA, if the attempt fails
3960 	 * then we should just turn on promiscuous mode so
3961 	 * that we can at least receive multicast traffic
3962 	 */
3963 	hw->mac.ops.update_mc_addr_list(hw, netdev);
3964 	vmolr |= IXGBE_VMOLR_ROMPE;
3965 
3966 	if (adapter->num_vfs)
3967 		ixgbe_restore_vf_multicasts(adapter);
3968 
3969 	if (hw->mac.type != ixgbe_mac_82598EB) {
3970 		vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
3971 			 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3972 			   IXGBE_VMOLR_ROPE);
3973 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
3974 	}
3975 
3976 	/* This is useful for sniffing bad packets. */
3977 	if (adapter->netdev->features & NETIF_F_RXALL) {
3978 		/* UPE and MPE will be handled by normal PROMISC logic
3979 		 * in e1000e_set_rx_mode */
3980 		fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
3981 			  IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
3982 			  IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
3983 
3984 		fctrl &= ~(IXGBE_FCTRL_DPF);
3985 		/* NOTE:  VLAN filtering is disabled by setting PROMISC */
3986 	}
3987 
3988 	IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3989 
3990 	if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
3991 		ixgbe_vlan_strip_enable(adapter);
3992 	else
3993 		ixgbe_vlan_strip_disable(adapter);
3994 }
3995 
3996 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3997 {
3998 	int q_idx;
3999 
4000 	for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++) {
4001 		ixgbe_qv_init_lock(adapter->q_vector[q_idx]);
4002 		napi_enable(&adapter->q_vector[q_idx]->napi);
4003 	}
4004 }
4005 
4006 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
4007 {
4008 	int q_idx;
4009 
4010 	for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++) {
4011 		napi_disable(&adapter->q_vector[q_idx]->napi);
4012 		while (!ixgbe_qv_disable(adapter->q_vector[q_idx])) {
4013 			pr_info("QV %d locked\n", q_idx);
4014 			usleep_range(1000, 20000);
4015 		}
4016 	}
4017 }
4018 
4019 #ifdef CONFIG_IXGBE_DCB
4020 /**
4021  * ixgbe_configure_dcb - Configure DCB hardware
4022  * @adapter: ixgbe adapter struct
4023  *
4024  * This is called by the driver on open to configure the DCB hardware.
4025  * This is also called by the gennetlink interface when reconfiguring
4026  * the DCB state.
4027  */
4028 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
4029 {
4030 	struct ixgbe_hw *hw = &adapter->hw;
4031 	int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4032 
4033 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
4034 		if (hw->mac.type == ixgbe_mac_82598EB)
4035 			netif_set_gso_max_size(adapter->netdev, 65536);
4036 		return;
4037 	}
4038 
4039 	if (hw->mac.type == ixgbe_mac_82598EB)
4040 		netif_set_gso_max_size(adapter->netdev, 32768);
4041 
4042 #ifdef IXGBE_FCOE
4043 	if (adapter->netdev->features & NETIF_F_FCOE_MTU)
4044 		max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
4045 #endif
4046 
4047 	/* reconfigure the hardware */
4048 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
4049 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
4050 						DCB_TX_CONFIG);
4051 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
4052 						DCB_RX_CONFIG);
4053 		ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
4054 	} else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
4055 		ixgbe_dcb_hw_ets(&adapter->hw,
4056 				 adapter->ixgbe_ieee_ets,
4057 				 max_frame);
4058 		ixgbe_dcb_hw_pfc_config(&adapter->hw,
4059 					adapter->ixgbe_ieee_pfc->pfc_en,
4060 					adapter->ixgbe_ieee_ets->prio_tc);
4061 	}
4062 
4063 	/* Enable RSS Hash per TC */
4064 	if (hw->mac.type != ixgbe_mac_82598EB) {
4065 		u32 msb = 0;
4066 		u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
4067 
4068 		while (rss_i) {
4069 			msb++;
4070 			rss_i >>= 1;
4071 		}
4072 
4073 		/* write msb to all 8 TCs in one write */
4074 		IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
4075 	}
4076 }
4077 #endif
4078 
4079 /* Additional bittime to account for IXGBE framing */
4080 #define IXGBE_ETH_FRAMING 20
4081 
4082 /**
4083  * ixgbe_hpbthresh - calculate high water mark for flow control
4084  *
4085  * @adapter: board private structure to calculate for
4086  * @pb: packet buffer to calculate
4087  */
4088 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
4089 {
4090 	struct ixgbe_hw *hw = &adapter->hw;
4091 	struct net_device *dev = adapter->netdev;
4092 	int link, tc, kb, marker;
4093 	u32 dv_id, rx_pba;
4094 
4095 	/* Calculate max LAN frame size */
4096 	tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
4097 
4098 #ifdef IXGBE_FCOE
4099 	/* FCoE traffic class uses FCOE jumbo frames */
4100 	if ((dev->features & NETIF_F_FCOE_MTU) &&
4101 	    (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
4102 	    (pb == ixgbe_fcoe_get_tc(adapter)))
4103 		tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4104 
4105 #endif
4106 	/* Calculate delay value for device */
4107 	switch (hw->mac.type) {
4108 	case ixgbe_mac_X540:
4109 		dv_id = IXGBE_DV_X540(link, tc);
4110 		break;
4111 	default:
4112 		dv_id = IXGBE_DV(link, tc);
4113 		break;
4114 	}
4115 
4116 	/* Loopback switch introduces additional latency */
4117 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4118 		dv_id += IXGBE_B2BT(tc);
4119 
4120 	/* Delay value is calculated in bit times convert to KB */
4121 	kb = IXGBE_BT2KB(dv_id);
4122 	rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
4123 
4124 	marker = rx_pba - kb;
4125 
4126 	/* It is possible that the packet buffer is not large enough
4127 	 * to provide required headroom. In this case throw an error
4128 	 * to user and a do the best we can.
4129 	 */
4130 	if (marker < 0) {
4131 		e_warn(drv, "Packet Buffer(%i) can not provide enough"
4132 			    "headroom to support flow control."
4133 			    "Decrease MTU or number of traffic classes\n", pb);
4134 		marker = tc + 1;
4135 	}
4136 
4137 	return marker;
4138 }
4139 
4140 /**
4141  * ixgbe_lpbthresh - calculate low water mark for for flow control
4142  *
4143  * @adapter: board private structure to calculate for
4144  * @pb: packet buffer to calculate
4145  */
4146 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter)
4147 {
4148 	struct ixgbe_hw *hw = &adapter->hw;
4149 	struct net_device *dev = adapter->netdev;
4150 	int tc;
4151 	u32 dv_id;
4152 
4153 	/* Calculate max LAN frame size */
4154 	tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
4155 
4156 	/* Calculate delay value for device */
4157 	switch (hw->mac.type) {
4158 	case ixgbe_mac_X540:
4159 		dv_id = IXGBE_LOW_DV_X540(tc);
4160 		break;
4161 	default:
4162 		dv_id = IXGBE_LOW_DV(tc);
4163 		break;
4164 	}
4165 
4166 	/* Delay value is calculated in bit times convert to KB */
4167 	return IXGBE_BT2KB(dv_id);
4168 }
4169 
4170 /*
4171  * ixgbe_pbthresh_setup - calculate and setup high low water marks
4172  */
4173 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
4174 {
4175 	struct ixgbe_hw *hw = &adapter->hw;
4176 	int num_tc = netdev_get_num_tc(adapter->netdev);
4177 	int i;
4178 
4179 	if (!num_tc)
4180 		num_tc = 1;
4181 
4182 	hw->fc.low_water = ixgbe_lpbthresh(adapter);
4183 
4184 	for (i = 0; i < num_tc; i++) {
4185 		hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
4186 
4187 		/* Low water marks must not be larger than high water marks */
4188 		if (hw->fc.low_water > hw->fc.high_water[i])
4189 			hw->fc.low_water = 0;
4190 	}
4191 }
4192 
4193 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
4194 {
4195 	struct ixgbe_hw *hw = &adapter->hw;
4196 	int hdrm;
4197 	u8 tc = netdev_get_num_tc(adapter->netdev);
4198 
4199 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4200 	    adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
4201 		hdrm = 32 << adapter->fdir_pballoc;
4202 	else
4203 		hdrm = 0;
4204 
4205 	hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
4206 	ixgbe_pbthresh_setup(adapter);
4207 }
4208 
4209 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
4210 {
4211 	struct ixgbe_hw *hw = &adapter->hw;
4212 	struct hlist_node *node2;
4213 	struct ixgbe_fdir_filter *filter;
4214 
4215 	spin_lock(&adapter->fdir_perfect_lock);
4216 
4217 	if (!hlist_empty(&adapter->fdir_filter_list))
4218 		ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
4219 
4220 	hlist_for_each_entry_safe(filter, node2,
4221 				  &adapter->fdir_filter_list, fdir_node) {
4222 		ixgbe_fdir_write_perfect_filter_82599(hw,
4223 				&filter->filter,
4224 				filter->sw_idx,
4225 				(filter->action == IXGBE_FDIR_DROP_QUEUE) ?
4226 				IXGBE_FDIR_DROP_QUEUE :
4227 				adapter->rx_ring[filter->action]->reg_idx);
4228 	}
4229 
4230 	spin_unlock(&adapter->fdir_perfect_lock);
4231 }
4232 
4233 static void ixgbe_macvlan_set_rx_mode(struct net_device *dev, unsigned int pool,
4234 				      struct ixgbe_adapter *adapter)
4235 {
4236 	struct ixgbe_hw *hw = &adapter->hw;
4237 	u32 vmolr;
4238 
4239 	/* No unicast promiscuous support for VMDQ devices. */
4240 	vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(pool));
4241 	vmolr |= (IXGBE_VMOLR_ROMPE | IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE);
4242 
4243 	/* clear the affected bit */
4244 	vmolr &= ~IXGBE_VMOLR_MPE;
4245 
4246 	if (dev->flags & IFF_ALLMULTI) {
4247 		vmolr |= IXGBE_VMOLR_MPE;
4248 	} else {
4249 		vmolr |= IXGBE_VMOLR_ROMPE;
4250 		hw->mac.ops.update_mc_addr_list(hw, dev);
4251 	}
4252 	ixgbe_write_uc_addr_list(adapter->netdev);
4253 	IXGBE_WRITE_REG(hw, IXGBE_VMOLR(pool), vmolr);
4254 }
4255 
4256 static void ixgbe_add_mac_filter(struct ixgbe_adapter *adapter,
4257 				 u8 *addr, u16 pool)
4258 {
4259 	struct ixgbe_hw *hw = &adapter->hw;
4260 	unsigned int entry;
4261 
4262 	entry = hw->mac.num_rar_entries - pool;
4263 	hw->mac.ops.set_rar(hw, entry, addr, VMDQ_P(pool), IXGBE_RAH_AV);
4264 }
4265 
4266 static void ixgbe_fwd_psrtype(struct ixgbe_fwd_adapter *vadapter)
4267 {
4268 	struct ixgbe_adapter *adapter = vadapter->real_adapter;
4269 	int rss_i = adapter->num_rx_queues_per_pool;
4270 	struct ixgbe_hw *hw = &adapter->hw;
4271 	u16 pool = vadapter->pool;
4272 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4273 		      IXGBE_PSRTYPE_UDPHDR |
4274 		      IXGBE_PSRTYPE_IPV4HDR |
4275 		      IXGBE_PSRTYPE_L2HDR |
4276 		      IXGBE_PSRTYPE_IPV6HDR;
4277 
4278 	if (hw->mac.type == ixgbe_mac_82598EB)
4279 		return;
4280 
4281 	if (rss_i > 3)
4282 		psrtype |= 2 << 29;
4283 	else if (rss_i > 1)
4284 		psrtype |= 1 << 29;
4285 
4286 	IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4287 }
4288 
4289 /**
4290  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4291  * @rx_ring: ring to free buffers from
4292  **/
4293 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4294 {
4295 	struct device *dev = rx_ring->dev;
4296 	unsigned long size;
4297 	u16 i;
4298 
4299 	/* ring already cleared, nothing to do */
4300 	if (!rx_ring->rx_buffer_info)
4301 		return;
4302 
4303 	/* Free all the Rx ring sk_buffs */
4304 	for (i = 0; i < rx_ring->count; i++) {
4305 		struct ixgbe_rx_buffer *rx_buffer;
4306 
4307 		rx_buffer = &rx_ring->rx_buffer_info[i];
4308 		if (rx_buffer->skb) {
4309 			struct sk_buff *skb = rx_buffer->skb;
4310 			if (IXGBE_CB(skb)->page_released) {
4311 				dma_unmap_page(dev,
4312 					       IXGBE_CB(skb)->dma,
4313 					       ixgbe_rx_bufsz(rx_ring),
4314 					       DMA_FROM_DEVICE);
4315 				IXGBE_CB(skb)->page_released = false;
4316 			}
4317 			dev_kfree_skb(skb);
4318 		}
4319 		rx_buffer->skb = NULL;
4320 		if (rx_buffer->dma)
4321 			dma_unmap_page(dev, rx_buffer->dma,
4322 				       ixgbe_rx_pg_size(rx_ring),
4323 				       DMA_FROM_DEVICE);
4324 		rx_buffer->dma = 0;
4325 		if (rx_buffer->page)
4326 			__free_pages(rx_buffer->page,
4327 				     ixgbe_rx_pg_order(rx_ring));
4328 		rx_buffer->page = NULL;
4329 	}
4330 
4331 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4332 	memset(rx_ring->rx_buffer_info, 0, size);
4333 
4334 	/* Zero out the descriptor ring */
4335 	memset(rx_ring->desc, 0, rx_ring->size);
4336 
4337 	rx_ring->next_to_alloc = 0;
4338 	rx_ring->next_to_clean = 0;
4339 	rx_ring->next_to_use = 0;
4340 }
4341 
4342 static void ixgbe_disable_fwd_ring(struct ixgbe_fwd_adapter *vadapter,
4343 				   struct ixgbe_ring *rx_ring)
4344 {
4345 	struct ixgbe_adapter *adapter = vadapter->real_adapter;
4346 	int index = rx_ring->queue_index + vadapter->rx_base_queue;
4347 
4348 	/* shutdown specific queue receive and wait for dma to settle */
4349 	ixgbe_disable_rx_queue(adapter, rx_ring);
4350 	usleep_range(10000, 20000);
4351 	ixgbe_irq_disable_queues(adapter, ((u64)1 << index));
4352 	ixgbe_clean_rx_ring(rx_ring);
4353 	rx_ring->l2_accel_priv = NULL;
4354 }
4355 
4356 static int ixgbe_fwd_ring_down(struct net_device *vdev,
4357 			       struct ixgbe_fwd_adapter *accel)
4358 {
4359 	struct ixgbe_adapter *adapter = accel->real_adapter;
4360 	unsigned int rxbase = accel->rx_base_queue;
4361 	unsigned int txbase = accel->tx_base_queue;
4362 	int i;
4363 
4364 	netif_tx_stop_all_queues(vdev);
4365 
4366 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4367 		ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
4368 		adapter->rx_ring[rxbase + i]->netdev = adapter->netdev;
4369 	}
4370 
4371 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4372 		adapter->tx_ring[txbase + i]->l2_accel_priv = NULL;
4373 		adapter->tx_ring[txbase + i]->netdev = adapter->netdev;
4374 	}
4375 
4376 
4377 	return 0;
4378 }
4379 
4380 static int ixgbe_fwd_ring_up(struct net_device *vdev,
4381 			     struct ixgbe_fwd_adapter *accel)
4382 {
4383 	struct ixgbe_adapter *adapter = accel->real_adapter;
4384 	unsigned int rxbase, txbase, queues;
4385 	int i, baseq, err = 0;
4386 
4387 	if (!test_bit(accel->pool, &adapter->fwd_bitmask))
4388 		return 0;
4389 
4390 	baseq = accel->pool * adapter->num_rx_queues_per_pool;
4391 	netdev_dbg(vdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
4392 		   accel->pool, adapter->num_rx_pools,
4393 		   baseq, baseq + adapter->num_rx_queues_per_pool,
4394 		   adapter->fwd_bitmask);
4395 
4396 	accel->netdev = vdev;
4397 	accel->rx_base_queue = rxbase = baseq;
4398 	accel->tx_base_queue = txbase = baseq;
4399 
4400 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
4401 		ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
4402 
4403 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4404 		adapter->rx_ring[rxbase + i]->netdev = vdev;
4405 		adapter->rx_ring[rxbase + i]->l2_accel_priv = accel;
4406 		ixgbe_configure_rx_ring(adapter, adapter->rx_ring[rxbase + i]);
4407 	}
4408 
4409 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4410 		adapter->tx_ring[txbase + i]->netdev = vdev;
4411 		adapter->tx_ring[txbase + i]->l2_accel_priv = accel;
4412 	}
4413 
4414 	queues = min_t(unsigned int,
4415 		       adapter->num_rx_queues_per_pool, vdev->num_tx_queues);
4416 	err = netif_set_real_num_tx_queues(vdev, queues);
4417 	if (err)
4418 		goto fwd_queue_err;
4419 
4420 	err = netif_set_real_num_rx_queues(vdev, queues);
4421 	if (err)
4422 		goto fwd_queue_err;
4423 
4424 	if (is_valid_ether_addr(vdev->dev_addr))
4425 		ixgbe_add_mac_filter(adapter, vdev->dev_addr, accel->pool);
4426 
4427 	ixgbe_fwd_psrtype(accel);
4428 	ixgbe_macvlan_set_rx_mode(vdev, accel->pool, adapter);
4429 	return err;
4430 fwd_queue_err:
4431 	ixgbe_fwd_ring_down(vdev, accel);
4432 	return err;
4433 }
4434 
4435 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
4436 {
4437 	struct net_device *upper;
4438 	struct list_head *iter;
4439 	int err;
4440 
4441 	netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
4442 		if (netif_is_macvlan(upper)) {
4443 			struct macvlan_dev *dfwd = netdev_priv(upper);
4444 			struct ixgbe_fwd_adapter *vadapter = dfwd->fwd_priv;
4445 
4446 			if (dfwd->fwd_priv) {
4447 				err = ixgbe_fwd_ring_up(upper, vadapter);
4448 				if (err)
4449 					continue;
4450 			}
4451 		}
4452 	}
4453 }
4454 
4455 static void ixgbe_configure(struct ixgbe_adapter *adapter)
4456 {
4457 	struct ixgbe_hw *hw = &adapter->hw;
4458 
4459 	ixgbe_configure_pb(adapter);
4460 #ifdef CONFIG_IXGBE_DCB
4461 	ixgbe_configure_dcb(adapter);
4462 #endif
4463 	/*
4464 	 * We must restore virtualization before VLANs or else
4465 	 * the VLVF registers will not be populated
4466 	 */
4467 	ixgbe_configure_virtualization(adapter);
4468 
4469 	ixgbe_set_rx_mode(adapter->netdev);
4470 	ixgbe_restore_vlan(adapter);
4471 
4472 	switch (hw->mac.type) {
4473 	case ixgbe_mac_82599EB:
4474 	case ixgbe_mac_X540:
4475 		hw->mac.ops.disable_rx_buff(hw);
4476 		break;
4477 	default:
4478 		break;
4479 	}
4480 
4481 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
4482 		ixgbe_init_fdir_signature_82599(&adapter->hw,
4483 						adapter->fdir_pballoc);
4484 	} else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
4485 		ixgbe_init_fdir_perfect_82599(&adapter->hw,
4486 					      adapter->fdir_pballoc);
4487 		ixgbe_fdir_filter_restore(adapter);
4488 	}
4489 
4490 	switch (hw->mac.type) {
4491 	case ixgbe_mac_82599EB:
4492 	case ixgbe_mac_X540:
4493 		hw->mac.ops.enable_rx_buff(hw);
4494 		break;
4495 	default:
4496 		break;
4497 	}
4498 
4499 #ifdef IXGBE_FCOE
4500 	/* configure FCoE L2 filters, redirection table, and Rx control */
4501 	ixgbe_configure_fcoe(adapter);
4502 
4503 #endif /* IXGBE_FCOE */
4504 	ixgbe_configure_tx(adapter);
4505 	ixgbe_configure_rx(adapter);
4506 	ixgbe_configure_dfwd(adapter);
4507 }
4508 
4509 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
4510 {
4511 	switch (hw->phy.type) {
4512 	case ixgbe_phy_sfp_avago:
4513 	case ixgbe_phy_sfp_ftl:
4514 	case ixgbe_phy_sfp_intel:
4515 	case ixgbe_phy_sfp_unknown:
4516 	case ixgbe_phy_sfp_passive_tyco:
4517 	case ixgbe_phy_sfp_passive_unknown:
4518 	case ixgbe_phy_sfp_active_unknown:
4519 	case ixgbe_phy_sfp_ftl_active:
4520 	case ixgbe_phy_qsfp_passive_unknown:
4521 	case ixgbe_phy_qsfp_active_unknown:
4522 	case ixgbe_phy_qsfp_intel:
4523 	case ixgbe_phy_qsfp_unknown:
4524 		return true;
4525 	case ixgbe_phy_nl:
4526 		if (hw->mac.type == ixgbe_mac_82598EB)
4527 			return true;
4528 	default:
4529 		return false;
4530 	}
4531 }
4532 
4533 /**
4534  * ixgbe_sfp_link_config - set up SFP+ link
4535  * @adapter: pointer to private adapter struct
4536  **/
4537 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
4538 {
4539 	/*
4540 	 * We are assuming the worst case scenario here, and that
4541 	 * is that an SFP was inserted/removed after the reset
4542 	 * but before SFP detection was enabled.  As such the best
4543 	 * solution is to just start searching as soon as we start
4544 	 */
4545 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
4546 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
4547 
4548 	adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
4549 }
4550 
4551 /**
4552  * ixgbe_non_sfp_link_config - set up non-SFP+ link
4553  * @hw: pointer to private hardware struct
4554  *
4555  * Returns 0 on success, negative on failure
4556  **/
4557 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
4558 {
4559 	u32 speed;
4560 	bool autoneg, link_up = false;
4561 	u32 ret = IXGBE_ERR_LINK_SETUP;
4562 
4563 	if (hw->mac.ops.check_link)
4564 		ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
4565 
4566 	if (ret)
4567 		goto link_cfg_out;
4568 
4569 	speed = hw->phy.autoneg_advertised;
4570 	if ((!speed) && (hw->mac.ops.get_link_capabilities))
4571 		ret = hw->mac.ops.get_link_capabilities(hw, &speed,
4572 							&autoneg);
4573 	if (ret)
4574 		goto link_cfg_out;
4575 
4576 	if (hw->mac.ops.setup_link)
4577 		ret = hw->mac.ops.setup_link(hw, speed, link_up);
4578 link_cfg_out:
4579 	return ret;
4580 }
4581 
4582 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
4583 {
4584 	struct ixgbe_hw *hw = &adapter->hw;
4585 	u32 gpie = 0;
4586 
4587 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4588 		gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
4589 		       IXGBE_GPIE_OCD;
4590 		gpie |= IXGBE_GPIE_EIAME;
4591 		/*
4592 		 * use EIAM to auto-mask when MSI-X interrupt is asserted
4593 		 * this saves a register write for every interrupt
4594 		 */
4595 		switch (hw->mac.type) {
4596 		case ixgbe_mac_82598EB:
4597 			IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4598 			break;
4599 		case ixgbe_mac_82599EB:
4600 		case ixgbe_mac_X540:
4601 		default:
4602 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
4603 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
4604 			break;
4605 		}
4606 	} else {
4607 		/* legacy interrupts, use EIAM to auto-mask when reading EICR,
4608 		 * specifically only auto mask tx and rx interrupts */
4609 		IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4610 	}
4611 
4612 	/* XXX: to interrupt immediately for EICS writes, enable this */
4613 	/* gpie |= IXGBE_GPIE_EIMEN; */
4614 
4615 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
4616 		gpie &= ~IXGBE_GPIE_VTMODE_MASK;
4617 
4618 		switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4619 		case IXGBE_82599_VMDQ_8Q_MASK:
4620 			gpie |= IXGBE_GPIE_VTMODE_16;
4621 			break;
4622 		case IXGBE_82599_VMDQ_4Q_MASK:
4623 			gpie |= IXGBE_GPIE_VTMODE_32;
4624 			break;
4625 		default:
4626 			gpie |= IXGBE_GPIE_VTMODE_64;
4627 			break;
4628 		}
4629 	}
4630 
4631 	/* Enable Thermal over heat sensor interrupt */
4632 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
4633 		switch (adapter->hw.mac.type) {
4634 		case ixgbe_mac_82599EB:
4635 			gpie |= IXGBE_SDP0_GPIEN;
4636 			break;
4637 		case ixgbe_mac_X540:
4638 			gpie |= IXGBE_EIMS_TS;
4639 			break;
4640 		default:
4641 			break;
4642 		}
4643 	}
4644 
4645 	/* Enable fan failure interrupt */
4646 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
4647 		gpie |= IXGBE_SDP1_GPIEN;
4648 
4649 	if (hw->mac.type == ixgbe_mac_82599EB) {
4650 		gpie |= IXGBE_SDP1_GPIEN;
4651 		gpie |= IXGBE_SDP2_GPIEN;
4652 	}
4653 
4654 	IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
4655 }
4656 
4657 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
4658 {
4659 	struct ixgbe_hw *hw = &adapter->hw;
4660 	int err;
4661 	u32 ctrl_ext;
4662 
4663 	ixgbe_get_hw_control(adapter);
4664 	ixgbe_setup_gpie(adapter);
4665 
4666 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4667 		ixgbe_configure_msix(adapter);
4668 	else
4669 		ixgbe_configure_msi_and_legacy(adapter);
4670 
4671 	/* enable the optics for 82599 SFP+ fiber */
4672 	if (hw->mac.ops.enable_tx_laser)
4673 		hw->mac.ops.enable_tx_laser(hw);
4674 
4675 	smp_mb__before_atomic();
4676 	clear_bit(__IXGBE_DOWN, &adapter->state);
4677 	ixgbe_napi_enable_all(adapter);
4678 
4679 	if (ixgbe_is_sfp(hw)) {
4680 		ixgbe_sfp_link_config(adapter);
4681 	} else {
4682 		err = ixgbe_non_sfp_link_config(hw);
4683 		if (err)
4684 			e_err(probe, "link_config FAILED %d\n", err);
4685 	}
4686 
4687 	/* clear any pending interrupts, may auto mask */
4688 	IXGBE_READ_REG(hw, IXGBE_EICR);
4689 	ixgbe_irq_enable(adapter, true, true);
4690 
4691 	/*
4692 	 * If this adapter has a fan, check to see if we had a failure
4693 	 * before we enabled the interrupt.
4694 	 */
4695 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
4696 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
4697 		if (esdp & IXGBE_ESDP_SDP1)
4698 			e_crit(drv, "Fan has stopped, replace the adapter\n");
4699 	}
4700 
4701 	/* bring the link up in the watchdog, this could race with our first
4702 	 * link up interrupt but shouldn't be a problem */
4703 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4704 	adapter->link_check_timeout = jiffies;
4705 	mod_timer(&adapter->service_timer, jiffies);
4706 
4707 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
4708 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
4709 	ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
4710 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
4711 }
4712 
4713 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
4714 {
4715 	WARN_ON(in_interrupt());
4716 	/* put off any impending NetWatchDogTimeout */
4717 	adapter->netdev->trans_start = jiffies;
4718 
4719 	while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
4720 		usleep_range(1000, 2000);
4721 	ixgbe_down(adapter);
4722 	/*
4723 	 * If SR-IOV enabled then wait a bit before bringing the adapter
4724 	 * back up to give the VFs time to respond to the reset.  The
4725 	 * two second wait is based upon the watchdog timer cycle in
4726 	 * the VF driver.
4727 	 */
4728 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4729 		msleep(2000);
4730 	ixgbe_up(adapter);
4731 	clear_bit(__IXGBE_RESETTING, &adapter->state);
4732 }
4733 
4734 void ixgbe_up(struct ixgbe_adapter *adapter)
4735 {
4736 	/* hardware has been reset, we need to reload some things */
4737 	ixgbe_configure(adapter);
4738 
4739 	ixgbe_up_complete(adapter);
4740 }
4741 
4742 void ixgbe_reset(struct ixgbe_adapter *adapter)
4743 {
4744 	struct ixgbe_hw *hw = &adapter->hw;
4745 	int err;
4746 
4747 	if (ixgbe_removed(hw->hw_addr))
4748 		return;
4749 	/* lock SFP init bit to prevent race conditions with the watchdog */
4750 	while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
4751 		usleep_range(1000, 2000);
4752 
4753 	/* clear all SFP and link config related flags while holding SFP_INIT */
4754 	adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
4755 			     IXGBE_FLAG2_SFP_NEEDS_RESET);
4756 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
4757 
4758 	err = hw->mac.ops.init_hw(hw);
4759 	switch (err) {
4760 	case 0:
4761 	case IXGBE_ERR_SFP_NOT_PRESENT:
4762 	case IXGBE_ERR_SFP_NOT_SUPPORTED:
4763 		break;
4764 	case IXGBE_ERR_MASTER_REQUESTS_PENDING:
4765 		e_dev_err("master disable timed out\n");
4766 		break;
4767 	case IXGBE_ERR_EEPROM_VERSION:
4768 		/* We are running on a pre-production device, log a warning */
4769 		e_dev_warn("This device is a pre-production adapter/LOM. "
4770 			   "Please be aware there may be issues associated with "
4771 			   "your hardware.  If you are experiencing problems "
4772 			   "please contact your Intel or hardware "
4773 			   "representative who provided you with this "
4774 			   "hardware.\n");
4775 		break;
4776 	default:
4777 		e_dev_err("Hardware Error: %d\n", err);
4778 	}
4779 
4780 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
4781 
4782 	/* reprogram the RAR[0] in case user changed it. */
4783 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, VMDQ_P(0), IXGBE_RAH_AV);
4784 
4785 	/* update SAN MAC vmdq pool selection */
4786 	if (hw->mac.san_mac_rar_index)
4787 		hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
4788 
4789 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
4790 		ixgbe_ptp_reset(adapter);
4791 }
4792 
4793 /**
4794  * ixgbe_clean_tx_ring - Free Tx Buffers
4795  * @tx_ring: ring to be cleaned
4796  **/
4797 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4798 {
4799 	struct ixgbe_tx_buffer *tx_buffer_info;
4800 	unsigned long size;
4801 	u16 i;
4802 
4803 	/* ring already cleared, nothing to do */
4804 	if (!tx_ring->tx_buffer_info)
4805 		return;
4806 
4807 	/* Free all the Tx ring sk_buffs */
4808 	for (i = 0; i < tx_ring->count; i++) {
4809 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
4810 		ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4811 	}
4812 
4813 	netdev_tx_reset_queue(txring_txq(tx_ring));
4814 
4815 	size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4816 	memset(tx_ring->tx_buffer_info, 0, size);
4817 
4818 	/* Zero out the descriptor ring */
4819 	memset(tx_ring->desc, 0, tx_ring->size);
4820 
4821 	tx_ring->next_to_use = 0;
4822 	tx_ring->next_to_clean = 0;
4823 }
4824 
4825 /**
4826  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4827  * @adapter: board private structure
4828  **/
4829 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4830 {
4831 	int i;
4832 
4833 	for (i = 0; i < adapter->num_rx_queues; i++)
4834 		ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4835 }
4836 
4837 /**
4838  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4839  * @adapter: board private structure
4840  **/
4841 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4842 {
4843 	int i;
4844 
4845 	for (i = 0; i < adapter->num_tx_queues; i++)
4846 		ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4847 }
4848 
4849 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
4850 {
4851 	struct hlist_node *node2;
4852 	struct ixgbe_fdir_filter *filter;
4853 
4854 	spin_lock(&adapter->fdir_perfect_lock);
4855 
4856 	hlist_for_each_entry_safe(filter, node2,
4857 				  &adapter->fdir_filter_list, fdir_node) {
4858 		hlist_del(&filter->fdir_node);
4859 		kfree(filter);
4860 	}
4861 	adapter->fdir_filter_count = 0;
4862 
4863 	spin_unlock(&adapter->fdir_perfect_lock);
4864 }
4865 
4866 void ixgbe_down(struct ixgbe_adapter *adapter)
4867 {
4868 	struct net_device *netdev = adapter->netdev;
4869 	struct ixgbe_hw *hw = &adapter->hw;
4870 	struct net_device *upper;
4871 	struct list_head *iter;
4872 	u32 rxctrl;
4873 	int i;
4874 
4875 	/* signal that we are down to the interrupt handler */
4876 	if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
4877 		return; /* do nothing if already down */
4878 
4879 	/* disable receives */
4880 	rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4881 	IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4882 
4883 	/* disable all enabled rx queues */
4884 	for (i = 0; i < adapter->num_rx_queues; i++)
4885 		/* this call also flushes the previous write */
4886 		ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4887 
4888 	usleep_range(10000, 20000);
4889 
4890 	netif_tx_stop_all_queues(netdev);
4891 
4892 	/* call carrier off first to avoid false dev_watchdog timeouts */
4893 	netif_carrier_off(netdev);
4894 	netif_tx_disable(netdev);
4895 
4896 	/* disable any upper devices */
4897 	netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
4898 		if (netif_is_macvlan(upper)) {
4899 			struct macvlan_dev *vlan = netdev_priv(upper);
4900 
4901 			if (vlan->fwd_priv) {
4902 				netif_tx_stop_all_queues(upper);
4903 				netif_carrier_off(upper);
4904 				netif_tx_disable(upper);
4905 			}
4906 		}
4907 	}
4908 
4909 	ixgbe_irq_disable(adapter);
4910 
4911 	ixgbe_napi_disable_all(adapter);
4912 
4913 	adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
4914 			     IXGBE_FLAG2_RESET_REQUESTED);
4915 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4916 
4917 	del_timer_sync(&adapter->service_timer);
4918 
4919 	if (adapter->num_vfs) {
4920 		/* Clear EITR Select mapping */
4921 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
4922 
4923 		/* Mark all the VFs as inactive */
4924 		for (i = 0 ; i < adapter->num_vfs; i++)
4925 			adapter->vfinfo[i].clear_to_send = false;
4926 
4927 		/* ping all the active vfs to let them know we are going down */
4928 		ixgbe_ping_all_vfs(adapter);
4929 
4930 		/* Disable all VFTE/VFRE TX/RX */
4931 		ixgbe_disable_tx_rx(adapter);
4932 	}
4933 
4934 	/* disable transmits in the hardware now that interrupts are off */
4935 	for (i = 0; i < adapter->num_tx_queues; i++) {
4936 		u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4937 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
4938 	}
4939 
4940 	/* Disable the Tx DMA engine on 82599 and X540 */
4941 	switch (hw->mac.type) {
4942 	case ixgbe_mac_82599EB:
4943 	case ixgbe_mac_X540:
4944 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4945 				(IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4946 				 ~IXGBE_DMATXCTL_TE));
4947 		break;
4948 	default:
4949 		break;
4950 	}
4951 
4952 	if (!pci_channel_offline(adapter->pdev))
4953 		ixgbe_reset(adapter);
4954 
4955 	/* power down the optics for 82599 SFP+ fiber */
4956 	if (hw->mac.ops.disable_tx_laser)
4957 		hw->mac.ops.disable_tx_laser(hw);
4958 
4959 	ixgbe_clean_all_tx_rings(adapter);
4960 	ixgbe_clean_all_rx_rings(adapter);
4961 
4962 #ifdef CONFIG_IXGBE_DCA
4963 	/* since we reset the hardware DCA settings were cleared */
4964 	ixgbe_setup_dca(adapter);
4965 #endif
4966 }
4967 
4968 /**
4969  * ixgbe_tx_timeout - Respond to a Tx Hang
4970  * @netdev: network interface device structure
4971  **/
4972 static void ixgbe_tx_timeout(struct net_device *netdev)
4973 {
4974 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4975 
4976 	/* Do the reset outside of interrupt context */
4977 	ixgbe_tx_timeout_reset(adapter);
4978 }
4979 
4980 /**
4981  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4982  * @adapter: board private structure to initialize
4983  *
4984  * ixgbe_sw_init initializes the Adapter private data structure.
4985  * Fields are initialized based on PCI device information and
4986  * OS network device settings (MTU size).
4987  **/
4988 static int ixgbe_sw_init(struct ixgbe_adapter *adapter)
4989 {
4990 	struct ixgbe_hw *hw = &adapter->hw;
4991 	struct pci_dev *pdev = adapter->pdev;
4992 	unsigned int rss, fdir;
4993 	u32 fwsm;
4994 #ifdef CONFIG_IXGBE_DCB
4995 	int j;
4996 	struct tc_configuration *tc;
4997 #endif
4998 
4999 	/* PCI config space info */
5000 
5001 	hw->vendor_id = pdev->vendor;
5002 	hw->device_id = pdev->device;
5003 	hw->revision_id = pdev->revision;
5004 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
5005 	hw->subsystem_device_id = pdev->subsystem_device;
5006 
5007 	/* Set common capability flags and settings */
5008 	rss = min_t(int, IXGBE_MAX_RSS_INDICES, num_online_cpus());
5009 	adapter->ring_feature[RING_F_RSS].limit = rss;
5010 	adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
5011 	adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
5012 	adapter->max_q_vectors = MAX_Q_VECTORS_82599;
5013 	adapter->atr_sample_rate = 20;
5014 	fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
5015 	adapter->ring_feature[RING_F_FDIR].limit = fdir;
5016 	adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
5017 #ifdef CONFIG_IXGBE_DCA
5018 	adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
5019 #endif
5020 #ifdef IXGBE_FCOE
5021 	adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
5022 	adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5023 #ifdef CONFIG_IXGBE_DCB
5024 	/* Default traffic class to use for FCoE */
5025 	adapter->fcoe.up = IXGBE_FCOE_DEFTC;
5026 #endif /* CONFIG_IXGBE_DCB */
5027 #endif /* IXGBE_FCOE */
5028 
5029 	/* Set MAC specific capability flags and exceptions */
5030 	switch (hw->mac.type) {
5031 	case ixgbe_mac_82598EB:
5032 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
5033 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
5034 
5035 		if (hw->device_id == IXGBE_DEV_ID_82598AT)
5036 			adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
5037 
5038 		adapter->max_q_vectors = MAX_Q_VECTORS_82598;
5039 		adapter->ring_feature[RING_F_FDIR].limit = 0;
5040 		adapter->atr_sample_rate = 0;
5041 		adapter->fdir_pballoc = 0;
5042 #ifdef IXGBE_FCOE
5043 		adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
5044 		adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5045 #ifdef CONFIG_IXGBE_DCB
5046 		adapter->fcoe.up = 0;
5047 #endif /* IXGBE_DCB */
5048 #endif /* IXGBE_FCOE */
5049 		break;
5050 	case ixgbe_mac_82599EB:
5051 		if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
5052 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5053 		break;
5054 	case ixgbe_mac_X540:
5055 		fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM);
5056 		if (fwsm & IXGBE_FWSM_TS_ENABLED)
5057 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5058 		break;
5059 	default:
5060 		break;
5061 	}
5062 
5063 #ifdef IXGBE_FCOE
5064 	/* FCoE support exists, always init the FCoE lock */
5065 	spin_lock_init(&adapter->fcoe.lock);
5066 
5067 #endif
5068 	/* n-tuple support exists, always init our spinlock */
5069 	spin_lock_init(&adapter->fdir_perfect_lock);
5070 
5071 #ifdef CONFIG_IXGBE_DCB
5072 	switch (hw->mac.type) {
5073 	case ixgbe_mac_X540:
5074 		adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
5075 		adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
5076 		break;
5077 	default:
5078 		adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
5079 		adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
5080 		break;
5081 	}
5082 
5083 	/* Configure DCB traffic classes */
5084 	for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5085 		tc = &adapter->dcb_cfg.tc_config[j];
5086 		tc->path[DCB_TX_CONFIG].bwg_id = 0;
5087 		tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5088 		tc->path[DCB_RX_CONFIG].bwg_id = 0;
5089 		tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5090 		tc->dcb_pfc = pfc_disabled;
5091 	}
5092 
5093 	/* Initialize default user to priority mapping, UPx->TC0 */
5094 	tc = &adapter->dcb_cfg.tc_config[0];
5095 	tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
5096 	tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
5097 
5098 	adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5099 	adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5100 	adapter->dcb_cfg.pfc_mode_enable = false;
5101 	adapter->dcb_set_bitmap = 0x00;
5102 	adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
5103 	memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
5104 	       sizeof(adapter->temp_dcb_cfg));
5105 
5106 #endif
5107 
5108 	/* default flow control settings */
5109 	hw->fc.requested_mode = ixgbe_fc_full;
5110 	hw->fc.current_mode = ixgbe_fc_full;	/* init for ethtool output */
5111 	ixgbe_pbthresh_setup(adapter);
5112 	hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5113 	hw->fc.send_xon = true;
5114 	hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
5115 
5116 #ifdef CONFIG_PCI_IOV
5117 	if (max_vfs > 0)
5118 		e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
5119 
5120 	/* assign number of SR-IOV VFs */
5121 	if (hw->mac.type != ixgbe_mac_82598EB) {
5122 		if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
5123 			adapter->num_vfs = 0;
5124 			e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
5125 		} else {
5126 			adapter->num_vfs = max_vfs;
5127 		}
5128 	}
5129 #endif /* CONFIG_PCI_IOV */
5130 
5131 	/* enable itr by default in dynamic mode */
5132 	adapter->rx_itr_setting = 1;
5133 	adapter->tx_itr_setting = 1;
5134 
5135 	/* set default ring sizes */
5136 	adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5137 	adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5138 
5139 	/* set default work limits */
5140 	adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
5141 
5142 	/* initialize eeprom parameters */
5143 	if (ixgbe_init_eeprom_params_generic(hw)) {
5144 		e_dev_err("EEPROM initialization failed\n");
5145 		return -EIO;
5146 	}
5147 
5148 	/* PF holds first pool slot */
5149 	set_bit(0, &adapter->fwd_bitmask);
5150 	set_bit(__IXGBE_DOWN, &adapter->state);
5151 
5152 	return 0;
5153 }
5154 
5155 /**
5156  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5157  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
5158  *
5159  * Return 0 on success, negative on failure
5160  **/
5161 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5162 {
5163 	struct device *dev = tx_ring->dev;
5164 	int orig_node = dev_to_node(dev);
5165 	int numa_node = -1;
5166 	int size;
5167 
5168 	size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5169 
5170 	if (tx_ring->q_vector)
5171 		numa_node = tx_ring->q_vector->numa_node;
5172 
5173 	tx_ring->tx_buffer_info = vzalloc_node(size, numa_node);
5174 	if (!tx_ring->tx_buffer_info)
5175 		tx_ring->tx_buffer_info = vzalloc(size);
5176 	if (!tx_ring->tx_buffer_info)
5177 		goto err;
5178 
5179 	u64_stats_init(&tx_ring->syncp);
5180 
5181 	/* round up to nearest 4K */
5182 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5183 	tx_ring->size = ALIGN(tx_ring->size, 4096);
5184 
5185 	set_dev_node(dev, numa_node);
5186 	tx_ring->desc = dma_alloc_coherent(dev,
5187 					   tx_ring->size,
5188 					   &tx_ring->dma,
5189 					   GFP_KERNEL);
5190 	set_dev_node(dev, orig_node);
5191 	if (!tx_ring->desc)
5192 		tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5193 						   &tx_ring->dma, GFP_KERNEL);
5194 	if (!tx_ring->desc)
5195 		goto err;
5196 
5197 	tx_ring->next_to_use = 0;
5198 	tx_ring->next_to_clean = 0;
5199 	return 0;
5200 
5201 err:
5202 	vfree(tx_ring->tx_buffer_info);
5203 	tx_ring->tx_buffer_info = NULL;
5204 	dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5205 	return -ENOMEM;
5206 }
5207 
5208 /**
5209  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5210  * @adapter: board private structure
5211  *
5212  * If this function returns with an error, then it's possible one or
5213  * more of the rings is populated (while the rest are not).  It is the
5214  * callers duty to clean those orphaned rings.
5215  *
5216  * Return 0 on success, negative on failure
5217  **/
5218 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5219 {
5220 	int i, err = 0;
5221 
5222 	for (i = 0; i < adapter->num_tx_queues; i++) {
5223 		err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5224 		if (!err)
5225 			continue;
5226 
5227 		e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5228 		goto err_setup_tx;
5229 	}
5230 
5231 	return 0;
5232 err_setup_tx:
5233 	/* rewind the index freeing the rings as we go */
5234 	while (i--)
5235 		ixgbe_free_tx_resources(adapter->tx_ring[i]);
5236 	return err;
5237 }
5238 
5239 /**
5240  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5241  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
5242  *
5243  * Returns 0 on success, negative on failure
5244  **/
5245 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5246 {
5247 	struct device *dev = rx_ring->dev;
5248 	int orig_node = dev_to_node(dev);
5249 	int numa_node = -1;
5250 	int size;
5251 
5252 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5253 
5254 	if (rx_ring->q_vector)
5255 		numa_node = rx_ring->q_vector->numa_node;
5256 
5257 	rx_ring->rx_buffer_info = vzalloc_node(size, numa_node);
5258 	if (!rx_ring->rx_buffer_info)
5259 		rx_ring->rx_buffer_info = vzalloc(size);
5260 	if (!rx_ring->rx_buffer_info)
5261 		goto err;
5262 
5263 	u64_stats_init(&rx_ring->syncp);
5264 
5265 	/* Round up to nearest 4K */
5266 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5267 	rx_ring->size = ALIGN(rx_ring->size, 4096);
5268 
5269 	set_dev_node(dev, numa_node);
5270 	rx_ring->desc = dma_alloc_coherent(dev,
5271 					   rx_ring->size,
5272 					   &rx_ring->dma,
5273 					   GFP_KERNEL);
5274 	set_dev_node(dev, orig_node);
5275 	if (!rx_ring->desc)
5276 		rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5277 						   &rx_ring->dma, GFP_KERNEL);
5278 	if (!rx_ring->desc)
5279 		goto err;
5280 
5281 	rx_ring->next_to_clean = 0;
5282 	rx_ring->next_to_use = 0;
5283 
5284 	return 0;
5285 err:
5286 	vfree(rx_ring->rx_buffer_info);
5287 	rx_ring->rx_buffer_info = NULL;
5288 	dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5289 	return -ENOMEM;
5290 }
5291 
5292 /**
5293  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5294  * @adapter: board private structure
5295  *
5296  * If this function returns with an error, then it's possible one or
5297  * more of the rings is populated (while the rest are not).  It is the
5298  * callers duty to clean those orphaned rings.
5299  *
5300  * Return 0 on success, negative on failure
5301  **/
5302 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5303 {
5304 	int i, err = 0;
5305 
5306 	for (i = 0; i < adapter->num_rx_queues; i++) {
5307 		err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5308 		if (!err)
5309 			continue;
5310 
5311 		e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5312 		goto err_setup_rx;
5313 	}
5314 
5315 #ifdef IXGBE_FCOE
5316 	err = ixgbe_setup_fcoe_ddp_resources(adapter);
5317 	if (!err)
5318 #endif
5319 		return 0;
5320 err_setup_rx:
5321 	/* rewind the index freeing the rings as we go */
5322 	while (i--)
5323 		ixgbe_free_rx_resources(adapter->rx_ring[i]);
5324 	return err;
5325 }
5326 
5327 /**
5328  * ixgbe_free_tx_resources - Free Tx Resources per Queue
5329  * @tx_ring: Tx descriptor ring for a specific queue
5330  *
5331  * Free all transmit software resources
5332  **/
5333 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5334 {
5335 	ixgbe_clean_tx_ring(tx_ring);
5336 
5337 	vfree(tx_ring->tx_buffer_info);
5338 	tx_ring->tx_buffer_info = NULL;
5339 
5340 	/* if not set, then don't free */
5341 	if (!tx_ring->desc)
5342 		return;
5343 
5344 	dma_free_coherent(tx_ring->dev, tx_ring->size,
5345 			  tx_ring->desc, tx_ring->dma);
5346 
5347 	tx_ring->desc = NULL;
5348 }
5349 
5350 /**
5351  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5352  * @adapter: board private structure
5353  *
5354  * Free all transmit software resources
5355  **/
5356 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5357 {
5358 	int i;
5359 
5360 	for (i = 0; i < adapter->num_tx_queues; i++)
5361 		if (adapter->tx_ring[i]->desc)
5362 			ixgbe_free_tx_resources(adapter->tx_ring[i]);
5363 }
5364 
5365 /**
5366  * ixgbe_free_rx_resources - Free Rx Resources
5367  * @rx_ring: ring to clean the resources from
5368  *
5369  * Free all receive software resources
5370  **/
5371 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5372 {
5373 	ixgbe_clean_rx_ring(rx_ring);
5374 
5375 	vfree(rx_ring->rx_buffer_info);
5376 	rx_ring->rx_buffer_info = NULL;
5377 
5378 	/* if not set, then don't free */
5379 	if (!rx_ring->desc)
5380 		return;
5381 
5382 	dma_free_coherent(rx_ring->dev, rx_ring->size,
5383 			  rx_ring->desc, rx_ring->dma);
5384 
5385 	rx_ring->desc = NULL;
5386 }
5387 
5388 /**
5389  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5390  * @adapter: board private structure
5391  *
5392  * Free all receive software resources
5393  **/
5394 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5395 {
5396 	int i;
5397 
5398 #ifdef IXGBE_FCOE
5399 	ixgbe_free_fcoe_ddp_resources(adapter);
5400 
5401 #endif
5402 	for (i = 0; i < adapter->num_rx_queues; i++)
5403 		if (adapter->rx_ring[i]->desc)
5404 			ixgbe_free_rx_resources(adapter->rx_ring[i]);
5405 }
5406 
5407 /**
5408  * ixgbe_change_mtu - Change the Maximum Transfer Unit
5409  * @netdev: network interface device structure
5410  * @new_mtu: new value for maximum frame size
5411  *
5412  * Returns 0 on success, negative on failure
5413  **/
5414 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5415 {
5416 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
5417 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5418 
5419 	/* MTU < 68 is an error and causes problems on some kernels */
5420 	if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5421 		return -EINVAL;
5422 
5423 	/*
5424 	 * For 82599EB we cannot allow legacy VFs to enable their receive
5425 	 * paths when MTU greater than 1500 is configured.  So display a
5426 	 * warning that legacy VFs will be disabled.
5427 	 */
5428 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
5429 	    (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
5430 	    (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
5431 		e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
5432 
5433 	e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5434 
5435 	/* must set new MTU before calling down or up */
5436 	netdev->mtu = new_mtu;
5437 
5438 	if (netif_running(netdev))
5439 		ixgbe_reinit_locked(adapter);
5440 
5441 	return 0;
5442 }
5443 
5444 /**
5445  * ixgbe_open - Called when a network interface is made active
5446  * @netdev: network interface device structure
5447  *
5448  * Returns 0 on success, negative value on failure
5449  *
5450  * The open entry point is called when a network interface is made
5451  * active by the system (IFF_UP).  At this point all resources needed
5452  * for transmit and receive operations are allocated, the interrupt
5453  * handler is registered with the OS, the watchdog timer is started,
5454  * and the stack is notified that the interface is ready.
5455  **/
5456 static int ixgbe_open(struct net_device *netdev)
5457 {
5458 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
5459 	int err, queues;
5460 
5461 	/* disallow open during test */
5462 	if (test_bit(__IXGBE_TESTING, &adapter->state))
5463 		return -EBUSY;
5464 
5465 	netif_carrier_off(netdev);
5466 
5467 	/* allocate transmit descriptors */
5468 	err = ixgbe_setup_all_tx_resources(adapter);
5469 	if (err)
5470 		goto err_setup_tx;
5471 
5472 	/* allocate receive descriptors */
5473 	err = ixgbe_setup_all_rx_resources(adapter);
5474 	if (err)
5475 		goto err_setup_rx;
5476 
5477 	ixgbe_configure(adapter);
5478 
5479 	err = ixgbe_request_irq(adapter);
5480 	if (err)
5481 		goto err_req_irq;
5482 
5483 	/* Notify the stack of the actual queue counts. */
5484 	if (adapter->num_rx_pools > 1)
5485 		queues = adapter->num_rx_queues_per_pool;
5486 	else
5487 		queues = adapter->num_tx_queues;
5488 
5489 	err = netif_set_real_num_tx_queues(netdev, queues);
5490 	if (err)
5491 		goto err_set_queues;
5492 
5493 	if (adapter->num_rx_pools > 1 &&
5494 	    adapter->num_rx_queues > IXGBE_MAX_L2A_QUEUES)
5495 		queues = IXGBE_MAX_L2A_QUEUES;
5496 	else
5497 		queues = adapter->num_rx_queues;
5498 	err = netif_set_real_num_rx_queues(netdev, queues);
5499 	if (err)
5500 		goto err_set_queues;
5501 
5502 	ixgbe_ptp_init(adapter);
5503 
5504 	ixgbe_up_complete(adapter);
5505 
5506 	return 0;
5507 
5508 err_set_queues:
5509 	ixgbe_free_irq(adapter);
5510 err_req_irq:
5511 	ixgbe_free_all_rx_resources(adapter);
5512 err_setup_rx:
5513 	ixgbe_free_all_tx_resources(adapter);
5514 err_setup_tx:
5515 	ixgbe_reset(adapter);
5516 
5517 	return err;
5518 }
5519 
5520 /**
5521  * ixgbe_close - Disables a network interface
5522  * @netdev: network interface device structure
5523  *
5524  * Returns 0, this is not allowed to fail
5525  *
5526  * The close entry point is called when an interface is de-activated
5527  * by the OS.  The hardware is still under the drivers control, but
5528  * needs to be disabled.  A global MAC reset is issued to stop the
5529  * hardware, and all transmit and receive resources are freed.
5530  **/
5531 static int ixgbe_close(struct net_device *netdev)
5532 {
5533 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
5534 
5535 	ixgbe_ptp_stop(adapter);
5536 
5537 	ixgbe_down(adapter);
5538 	ixgbe_free_irq(adapter);
5539 
5540 	ixgbe_fdir_filter_exit(adapter);
5541 
5542 	ixgbe_free_all_tx_resources(adapter);
5543 	ixgbe_free_all_rx_resources(adapter);
5544 
5545 	ixgbe_release_hw_control(adapter);
5546 
5547 	return 0;
5548 }
5549 
5550 #ifdef CONFIG_PM
5551 static int ixgbe_resume(struct pci_dev *pdev)
5552 {
5553 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5554 	struct net_device *netdev = adapter->netdev;
5555 	u32 err;
5556 
5557 	adapter->hw.hw_addr = adapter->io_addr;
5558 	pci_set_power_state(pdev, PCI_D0);
5559 	pci_restore_state(pdev);
5560 	/*
5561 	 * pci_restore_state clears dev->state_saved so call
5562 	 * pci_save_state to restore it.
5563 	 */
5564 	pci_save_state(pdev);
5565 
5566 	err = pci_enable_device_mem(pdev);
5567 	if (err) {
5568 		e_dev_err("Cannot enable PCI device from suspend\n");
5569 		return err;
5570 	}
5571 	smp_mb__before_atomic();
5572 	clear_bit(__IXGBE_DISABLED, &adapter->state);
5573 	pci_set_master(pdev);
5574 
5575 	pci_wake_from_d3(pdev, false);
5576 
5577 	ixgbe_reset(adapter);
5578 
5579 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5580 
5581 	rtnl_lock();
5582 	err = ixgbe_init_interrupt_scheme(adapter);
5583 	if (!err && netif_running(netdev))
5584 		err = ixgbe_open(netdev);
5585 
5586 	rtnl_unlock();
5587 
5588 	if (err)
5589 		return err;
5590 
5591 	netif_device_attach(netdev);
5592 
5593 	return 0;
5594 }
5595 #endif /* CONFIG_PM */
5596 
5597 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5598 {
5599 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5600 	struct net_device *netdev = adapter->netdev;
5601 	struct ixgbe_hw *hw = &adapter->hw;
5602 	u32 ctrl, fctrl;
5603 	u32 wufc = adapter->wol;
5604 #ifdef CONFIG_PM
5605 	int retval = 0;
5606 #endif
5607 
5608 	netif_device_detach(netdev);
5609 
5610 	rtnl_lock();
5611 	if (netif_running(netdev)) {
5612 		ixgbe_down(adapter);
5613 		ixgbe_free_irq(adapter);
5614 		ixgbe_free_all_tx_resources(adapter);
5615 		ixgbe_free_all_rx_resources(adapter);
5616 	}
5617 	rtnl_unlock();
5618 
5619 	ixgbe_clear_interrupt_scheme(adapter);
5620 
5621 #ifdef CONFIG_PM
5622 	retval = pci_save_state(pdev);
5623 	if (retval)
5624 		return retval;
5625 
5626 #endif
5627 	if (hw->mac.ops.stop_link_on_d3)
5628 		hw->mac.ops.stop_link_on_d3(hw);
5629 
5630 	if (wufc) {
5631 		ixgbe_set_rx_mode(netdev);
5632 
5633 		/* enable the optics for 82599 SFP+ fiber as we can WoL */
5634 		if (hw->mac.ops.enable_tx_laser)
5635 			hw->mac.ops.enable_tx_laser(hw);
5636 
5637 		/* turn on all-multi mode if wake on multicast is enabled */
5638 		if (wufc & IXGBE_WUFC_MC) {
5639 			fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5640 			fctrl |= IXGBE_FCTRL_MPE;
5641 			IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5642 		}
5643 
5644 		ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5645 		ctrl |= IXGBE_CTRL_GIO_DIS;
5646 		IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5647 
5648 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5649 	} else {
5650 		IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5651 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5652 	}
5653 
5654 	switch (hw->mac.type) {
5655 	case ixgbe_mac_82598EB:
5656 		pci_wake_from_d3(pdev, false);
5657 		break;
5658 	case ixgbe_mac_82599EB:
5659 	case ixgbe_mac_X540:
5660 		pci_wake_from_d3(pdev, !!wufc);
5661 		break;
5662 	default:
5663 		break;
5664 	}
5665 
5666 	*enable_wake = !!wufc;
5667 
5668 	ixgbe_release_hw_control(adapter);
5669 
5670 	if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
5671 		pci_disable_device(pdev);
5672 
5673 	return 0;
5674 }
5675 
5676 #ifdef CONFIG_PM
5677 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5678 {
5679 	int retval;
5680 	bool wake;
5681 
5682 	retval = __ixgbe_shutdown(pdev, &wake);
5683 	if (retval)
5684 		return retval;
5685 
5686 	if (wake) {
5687 		pci_prepare_to_sleep(pdev);
5688 	} else {
5689 		pci_wake_from_d3(pdev, false);
5690 		pci_set_power_state(pdev, PCI_D3hot);
5691 	}
5692 
5693 	return 0;
5694 }
5695 #endif /* CONFIG_PM */
5696 
5697 static void ixgbe_shutdown(struct pci_dev *pdev)
5698 {
5699 	bool wake;
5700 
5701 	__ixgbe_shutdown(pdev, &wake);
5702 
5703 	if (system_state == SYSTEM_POWER_OFF) {
5704 		pci_wake_from_d3(pdev, wake);
5705 		pci_set_power_state(pdev, PCI_D3hot);
5706 	}
5707 }
5708 
5709 /**
5710  * ixgbe_update_stats - Update the board statistics counters.
5711  * @adapter: board private structure
5712  **/
5713 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5714 {
5715 	struct net_device *netdev = adapter->netdev;
5716 	struct ixgbe_hw *hw = &adapter->hw;
5717 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
5718 	u64 total_mpc = 0;
5719 	u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5720 	u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5721 	u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5722 	u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
5723 
5724 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5725 	    test_bit(__IXGBE_RESETTING, &adapter->state))
5726 		return;
5727 
5728 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5729 		u64 rsc_count = 0;
5730 		u64 rsc_flush = 0;
5731 		for (i = 0; i < adapter->num_rx_queues; i++) {
5732 			rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5733 			rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5734 		}
5735 		adapter->rsc_total_count = rsc_count;
5736 		adapter->rsc_total_flush = rsc_flush;
5737 	}
5738 
5739 	for (i = 0; i < adapter->num_rx_queues; i++) {
5740 		struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5741 		non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5742 		alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5743 		alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5744 		hw_csum_rx_error += rx_ring->rx_stats.csum_err;
5745 		bytes += rx_ring->stats.bytes;
5746 		packets += rx_ring->stats.packets;
5747 	}
5748 	adapter->non_eop_descs = non_eop_descs;
5749 	adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5750 	adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5751 	adapter->hw_csum_rx_error = hw_csum_rx_error;
5752 	netdev->stats.rx_bytes = bytes;
5753 	netdev->stats.rx_packets = packets;
5754 
5755 	bytes = 0;
5756 	packets = 0;
5757 	/* gather some stats to the adapter struct that are per queue */
5758 	for (i = 0; i < adapter->num_tx_queues; i++) {
5759 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5760 		restart_queue += tx_ring->tx_stats.restart_queue;
5761 		tx_busy += tx_ring->tx_stats.tx_busy;
5762 		bytes += tx_ring->stats.bytes;
5763 		packets += tx_ring->stats.packets;
5764 	}
5765 	adapter->restart_queue = restart_queue;
5766 	adapter->tx_busy = tx_busy;
5767 	netdev->stats.tx_bytes = bytes;
5768 	netdev->stats.tx_packets = packets;
5769 
5770 	hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5771 
5772 	/* 8 register reads */
5773 	for (i = 0; i < 8; i++) {
5774 		/* for packet buffers not used, the register should read 0 */
5775 		mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5776 		missed_rx += mpc;
5777 		hwstats->mpc[i] += mpc;
5778 		total_mpc += hwstats->mpc[i];
5779 		hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5780 		hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5781 		switch (hw->mac.type) {
5782 		case ixgbe_mac_82598EB:
5783 			hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5784 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5785 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5786 			hwstats->pxonrxc[i] +=
5787 				IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5788 			break;
5789 		case ixgbe_mac_82599EB:
5790 		case ixgbe_mac_X540:
5791 			hwstats->pxonrxc[i] +=
5792 				IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5793 			break;
5794 		default:
5795 			break;
5796 		}
5797 	}
5798 
5799 	/*16 register reads */
5800 	for (i = 0; i < 16; i++) {
5801 		hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5802 		hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5803 		if ((hw->mac.type == ixgbe_mac_82599EB) ||
5804 		    (hw->mac.type == ixgbe_mac_X540)) {
5805 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
5806 			IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
5807 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
5808 			IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
5809 		}
5810 	}
5811 
5812 	hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5813 	/* work around hardware counting issue */
5814 	hwstats->gprc -= missed_rx;
5815 
5816 	ixgbe_update_xoff_received(adapter);
5817 
5818 	/* 82598 hardware only has a 32 bit counter in the high register */
5819 	switch (hw->mac.type) {
5820 	case ixgbe_mac_82598EB:
5821 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5822 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5823 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5824 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5825 		break;
5826 	case ixgbe_mac_X540:
5827 		/* OS2BMC stats are X540 only*/
5828 		hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5829 		hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5830 		hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5831 		hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5832 	case ixgbe_mac_82599EB:
5833 		for (i = 0; i < 16; i++)
5834 			adapter->hw_rx_no_dma_resources +=
5835 					     IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5836 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5837 		IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5838 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5839 		IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5840 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5841 		IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5842 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5843 		hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5844 		hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5845 #ifdef IXGBE_FCOE
5846 		hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5847 		hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5848 		hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5849 		hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5850 		hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5851 		hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5852 		/* Add up per cpu counters for total ddp aloc fail */
5853 		if (adapter->fcoe.ddp_pool) {
5854 			struct ixgbe_fcoe *fcoe = &adapter->fcoe;
5855 			struct ixgbe_fcoe_ddp_pool *ddp_pool;
5856 			unsigned int cpu;
5857 			u64 noddp = 0, noddp_ext_buff = 0;
5858 			for_each_possible_cpu(cpu) {
5859 				ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
5860 				noddp += ddp_pool->noddp;
5861 				noddp_ext_buff += ddp_pool->noddp_ext_buff;
5862 			}
5863 			hwstats->fcoe_noddp = noddp;
5864 			hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
5865 		}
5866 #endif /* IXGBE_FCOE */
5867 		break;
5868 	default:
5869 		break;
5870 	}
5871 	bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5872 	hwstats->bprc += bprc;
5873 	hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5874 	if (hw->mac.type == ixgbe_mac_82598EB)
5875 		hwstats->mprc -= bprc;
5876 	hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5877 	hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5878 	hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5879 	hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5880 	hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5881 	hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5882 	hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5883 	hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5884 	lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5885 	hwstats->lxontxc += lxon;
5886 	lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5887 	hwstats->lxofftxc += lxoff;
5888 	hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5889 	hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5890 	/*
5891 	 * 82598 errata - tx of flow control packets is included in tx counters
5892 	 */
5893 	xon_off_tot = lxon + lxoff;
5894 	hwstats->gptc -= xon_off_tot;
5895 	hwstats->mptc -= xon_off_tot;
5896 	hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5897 	hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5898 	hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5899 	hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5900 	hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5901 	hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5902 	hwstats->ptc64 -= xon_off_tot;
5903 	hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5904 	hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5905 	hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5906 	hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5907 	hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5908 	hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5909 
5910 	/* Fill out the OS statistics structure */
5911 	netdev->stats.multicast = hwstats->mprc;
5912 
5913 	/* Rx Errors */
5914 	netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5915 	netdev->stats.rx_dropped = 0;
5916 	netdev->stats.rx_length_errors = hwstats->rlec;
5917 	netdev->stats.rx_crc_errors = hwstats->crcerrs;
5918 	netdev->stats.rx_missed_errors = total_mpc;
5919 }
5920 
5921 /**
5922  * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5923  * @adapter: pointer to the device adapter structure
5924  **/
5925 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5926 {
5927 	struct ixgbe_hw *hw = &adapter->hw;
5928 	int i;
5929 
5930 	if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5931 		return;
5932 
5933 	adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5934 
5935 	/* if interface is down do nothing */
5936 	if (test_bit(__IXGBE_DOWN, &adapter->state))
5937 		return;
5938 
5939 	/* do nothing if we are not using signature filters */
5940 	if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5941 		return;
5942 
5943 	adapter->fdir_overflow++;
5944 
5945 	if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5946 		for (i = 0; i < adapter->num_tx_queues; i++)
5947 			set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5948 			        &(adapter->tx_ring[i]->state));
5949 		/* re-enable flow director interrupts */
5950 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5951 	} else {
5952 		e_err(probe, "failed to finish FDIR re-initialization, "
5953 		      "ignored adding FDIR ATR filters\n");
5954 	}
5955 }
5956 
5957 /**
5958  * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5959  * @adapter: pointer to the device adapter structure
5960  *
5961  * This function serves two purposes.  First it strobes the interrupt lines
5962  * in order to make certain interrupts are occurring.  Secondly it sets the
5963  * bits needed to check for TX hangs.  As a result we should immediately
5964  * determine if a hang has occurred.
5965  */
5966 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5967 {
5968 	struct ixgbe_hw *hw = &adapter->hw;
5969 	u64 eics = 0;
5970 	int i;
5971 
5972 	/* If we're down, removing or resetting, just bail */
5973 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5974 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
5975 	    test_bit(__IXGBE_RESETTING, &adapter->state))
5976 		return;
5977 
5978 	/* Force detection of hung controller */
5979 	if (netif_carrier_ok(adapter->netdev)) {
5980 		for (i = 0; i < adapter->num_tx_queues; i++)
5981 			set_check_for_tx_hang(adapter->tx_ring[i]);
5982 	}
5983 
5984 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5985 		/*
5986 		 * for legacy and MSI interrupts don't set any bits
5987 		 * that are enabled for EIAM, because this operation
5988 		 * would set *both* EIMS and EICS for any bit in EIAM
5989 		 */
5990 		IXGBE_WRITE_REG(hw, IXGBE_EICS,
5991 			(IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5992 	} else {
5993 		/* get one bit for every active tx/rx interrupt vector */
5994 		for (i = 0; i < adapter->num_q_vectors; i++) {
5995 			struct ixgbe_q_vector *qv = adapter->q_vector[i];
5996 			if (qv->rx.ring || qv->tx.ring)
5997 				eics |= ((u64)1 << i);
5998 		}
5999 	}
6000 
6001 	/* Cause software interrupt to ensure rings are cleaned */
6002 	ixgbe_irq_rearm_queues(adapter, eics);
6003 
6004 }
6005 
6006 /**
6007  * ixgbe_watchdog_update_link - update the link status
6008  * @adapter: pointer to the device adapter structure
6009  * @link_speed: pointer to a u32 to store the link_speed
6010  **/
6011 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
6012 {
6013 	struct ixgbe_hw *hw = &adapter->hw;
6014 	u32 link_speed = adapter->link_speed;
6015 	bool link_up = adapter->link_up;
6016 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
6017 
6018 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
6019 		return;
6020 
6021 	if (hw->mac.ops.check_link) {
6022 		hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
6023 	} else {
6024 		/* always assume link is up, if no check link function */
6025 		link_speed = IXGBE_LINK_SPEED_10GB_FULL;
6026 		link_up = true;
6027 	}
6028 
6029 	if (adapter->ixgbe_ieee_pfc)
6030 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
6031 
6032 	if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
6033 		hw->mac.ops.fc_enable(hw);
6034 		ixgbe_set_rx_drop_en(adapter);
6035 	}
6036 
6037 	if (link_up ||
6038 	    time_after(jiffies, (adapter->link_check_timeout +
6039 				 IXGBE_TRY_LINK_TIMEOUT))) {
6040 		adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6041 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
6042 		IXGBE_WRITE_FLUSH(hw);
6043 	}
6044 
6045 	adapter->link_up = link_up;
6046 	adapter->link_speed = link_speed;
6047 }
6048 
6049 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
6050 {
6051 #ifdef CONFIG_IXGBE_DCB
6052 	struct net_device *netdev = adapter->netdev;
6053 	struct dcb_app app = {
6054 			      .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
6055 			      .protocol = 0,
6056 			     };
6057 	u8 up = 0;
6058 
6059 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
6060 		up = dcb_ieee_getapp_mask(netdev, &app);
6061 
6062 	adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
6063 #endif
6064 }
6065 
6066 /**
6067  * ixgbe_watchdog_link_is_up - update netif_carrier status and
6068  *                             print link up message
6069  * @adapter: pointer to the device adapter structure
6070  **/
6071 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
6072 {
6073 	struct net_device *netdev = adapter->netdev;
6074 	struct ixgbe_hw *hw = &adapter->hw;
6075 	struct net_device *upper;
6076 	struct list_head *iter;
6077 	u32 link_speed = adapter->link_speed;
6078 	bool flow_rx, flow_tx;
6079 
6080 	/* only continue if link was previously down */
6081 	if (netif_carrier_ok(netdev))
6082 		return;
6083 
6084 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
6085 
6086 	switch (hw->mac.type) {
6087 	case ixgbe_mac_82598EB: {
6088 		u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6089 		u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
6090 		flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
6091 		flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
6092 	}
6093 		break;
6094 	case ixgbe_mac_X540:
6095 	case ixgbe_mac_82599EB: {
6096 		u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
6097 		u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
6098 		flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
6099 		flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
6100 	}
6101 		break;
6102 	default:
6103 		flow_tx = false;
6104 		flow_rx = false;
6105 		break;
6106 	}
6107 
6108 	adapter->last_rx_ptp_check = jiffies;
6109 
6110 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
6111 		ixgbe_ptp_start_cyclecounter(adapter);
6112 
6113 	e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
6114 	       (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
6115 	       "10 Gbps" :
6116 	       (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
6117 	       "1 Gbps" :
6118 	       (link_speed == IXGBE_LINK_SPEED_100_FULL ?
6119 	       "100 Mbps" :
6120 	       "unknown speed"))),
6121 	       ((flow_rx && flow_tx) ? "RX/TX" :
6122 	       (flow_rx ? "RX" :
6123 	       (flow_tx ? "TX" : "None"))));
6124 
6125 	netif_carrier_on(netdev);
6126 	ixgbe_check_vf_rate_limit(adapter);
6127 
6128 	/* enable transmits */
6129 	netif_tx_wake_all_queues(adapter->netdev);
6130 
6131 	/* enable any upper devices */
6132 	rtnl_lock();
6133 	netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
6134 		if (netif_is_macvlan(upper)) {
6135 			struct macvlan_dev *vlan = netdev_priv(upper);
6136 
6137 			if (vlan->fwd_priv)
6138 				netif_tx_wake_all_queues(upper);
6139 		}
6140 	}
6141 	rtnl_unlock();
6142 
6143 	/* update the default user priority for VFs */
6144 	ixgbe_update_default_up(adapter);
6145 
6146 	/* ping all the active vfs to let them know link has changed */
6147 	ixgbe_ping_all_vfs(adapter);
6148 }
6149 
6150 /**
6151  * ixgbe_watchdog_link_is_down - update netif_carrier status and
6152  *                               print link down message
6153  * @adapter: pointer to the adapter structure
6154  **/
6155 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
6156 {
6157 	struct net_device *netdev = adapter->netdev;
6158 	struct ixgbe_hw *hw = &adapter->hw;
6159 
6160 	adapter->link_up = false;
6161 	adapter->link_speed = 0;
6162 
6163 	/* only continue if link was up previously */
6164 	if (!netif_carrier_ok(netdev))
6165 		return;
6166 
6167 	/* poll for SFP+ cable when link is down */
6168 	if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
6169 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
6170 
6171 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
6172 		ixgbe_ptp_start_cyclecounter(adapter);
6173 
6174 	e_info(drv, "NIC Link is Down\n");
6175 	netif_carrier_off(netdev);
6176 
6177 	/* ping all the active vfs to let them know link has changed */
6178 	ixgbe_ping_all_vfs(adapter);
6179 }
6180 
6181 /**
6182  * ixgbe_watchdog_flush_tx - flush queues on link down
6183  * @adapter: pointer to the device adapter structure
6184  **/
6185 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
6186 {
6187 	int i;
6188 	int some_tx_pending = 0;
6189 
6190 	if (!netif_carrier_ok(adapter->netdev)) {
6191 		for (i = 0; i < adapter->num_tx_queues; i++) {
6192 			struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
6193 			if (tx_ring->next_to_use != tx_ring->next_to_clean) {
6194 				some_tx_pending = 1;
6195 				break;
6196 			}
6197 		}
6198 
6199 		if (some_tx_pending) {
6200 			/* We've lost link, so the controller stops DMA,
6201 			 * but we've got queued Tx work that's never going
6202 			 * to get done, so reset controller to flush Tx.
6203 			 * (Do the reset outside of interrupt context).
6204 			 */
6205 			e_warn(drv, "initiating reset to clear Tx work after link loss\n");
6206 			adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
6207 		}
6208 	}
6209 }
6210 
6211 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
6212 {
6213 	u32 ssvpc;
6214 
6215 	/* Do not perform spoof check for 82598 or if not in IOV mode */
6216 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
6217 	    adapter->num_vfs == 0)
6218 		return;
6219 
6220 	ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
6221 
6222 	/*
6223 	 * ssvpc register is cleared on read, if zero then no
6224 	 * spoofed packets in the last interval.
6225 	 */
6226 	if (!ssvpc)
6227 		return;
6228 
6229 	e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
6230 }
6231 
6232 /**
6233  * ixgbe_watchdog_subtask - check and bring link up
6234  * @adapter: pointer to the device adapter structure
6235  **/
6236 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
6237 {
6238 	/* if interface is down, removing or resetting, do nothing */
6239 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6240 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
6241 	    test_bit(__IXGBE_RESETTING, &adapter->state))
6242 		return;
6243 
6244 	ixgbe_watchdog_update_link(adapter);
6245 
6246 	if (adapter->link_up)
6247 		ixgbe_watchdog_link_is_up(adapter);
6248 	else
6249 		ixgbe_watchdog_link_is_down(adapter);
6250 
6251 	ixgbe_spoof_check(adapter);
6252 	ixgbe_update_stats(adapter);
6253 
6254 	ixgbe_watchdog_flush_tx(adapter);
6255 }
6256 
6257 /**
6258  * ixgbe_sfp_detection_subtask - poll for SFP+ cable
6259  * @adapter: the ixgbe adapter structure
6260  **/
6261 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
6262 {
6263 	struct ixgbe_hw *hw = &adapter->hw;
6264 	s32 err;
6265 
6266 	/* not searching for SFP so there is nothing to do here */
6267 	if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
6268 	    !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6269 		return;
6270 
6271 	/* someone else is in init, wait until next service event */
6272 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6273 		return;
6274 
6275 	err = hw->phy.ops.identify_sfp(hw);
6276 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6277 		goto sfp_out;
6278 
6279 	if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
6280 		/* If no cable is present, then we need to reset
6281 		 * the next time we find a good cable. */
6282 		adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
6283 	}
6284 
6285 	/* exit on error */
6286 	if (err)
6287 		goto sfp_out;
6288 
6289 	/* exit if reset not needed */
6290 	if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6291 		goto sfp_out;
6292 
6293 	adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
6294 
6295 	/*
6296 	 * A module may be identified correctly, but the EEPROM may not have
6297 	 * support for that module.  setup_sfp() will fail in that case, so
6298 	 * we should not allow that module to load.
6299 	 */
6300 	if (hw->mac.type == ixgbe_mac_82598EB)
6301 		err = hw->phy.ops.reset(hw);
6302 	else
6303 		err = hw->mac.ops.setup_sfp(hw);
6304 
6305 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6306 		goto sfp_out;
6307 
6308 	adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
6309 	e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
6310 
6311 sfp_out:
6312 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6313 
6314 	if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
6315 	    (adapter->netdev->reg_state == NETREG_REGISTERED)) {
6316 		e_dev_err("failed to initialize because an unsupported "
6317 			  "SFP+ module type was detected.\n");
6318 		e_dev_err("Reload the driver after installing a "
6319 			  "supported module.\n");
6320 		unregister_netdev(adapter->netdev);
6321 	}
6322 }
6323 
6324 /**
6325  * ixgbe_sfp_link_config_subtask - set up link SFP after module install
6326  * @adapter: the ixgbe adapter structure
6327  **/
6328 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
6329 {
6330 	struct ixgbe_hw *hw = &adapter->hw;
6331 	u32 speed;
6332 	bool autoneg = false;
6333 
6334 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
6335 		return;
6336 
6337 	/* someone else is in init, wait until next service event */
6338 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6339 		return;
6340 
6341 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
6342 
6343 	speed = hw->phy.autoneg_advertised;
6344 	if ((!speed) && (hw->mac.ops.get_link_capabilities)) {
6345 		hw->mac.ops.get_link_capabilities(hw, &speed, &autoneg);
6346 
6347 		/* setup the highest link when no autoneg */
6348 		if (!autoneg) {
6349 			if (speed & IXGBE_LINK_SPEED_10GB_FULL)
6350 				speed = IXGBE_LINK_SPEED_10GB_FULL;
6351 		}
6352 	}
6353 
6354 	if (hw->mac.ops.setup_link)
6355 		hw->mac.ops.setup_link(hw, speed, true);
6356 
6357 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
6358 	adapter->link_check_timeout = jiffies;
6359 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6360 }
6361 
6362 #ifdef CONFIG_PCI_IOV
6363 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
6364 {
6365 	int vf;
6366 	struct ixgbe_hw *hw = &adapter->hw;
6367 	struct net_device *netdev = adapter->netdev;
6368 	u32 gpc;
6369 	u32 ciaa, ciad;
6370 
6371 	gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
6372 	if (gpc) /* If incrementing then no need for the check below */
6373 		return;
6374 	/*
6375 	 * Check to see if a bad DMA write target from an errant or
6376 	 * malicious VF has caused a PCIe error.  If so then we can
6377 	 * issue a VFLR to the offending VF(s) and then resume without
6378 	 * requesting a full slot reset.
6379 	 */
6380 
6381 	for (vf = 0; vf < adapter->num_vfs; vf++) {
6382 		ciaa = (vf << 16) | 0x80000000;
6383 		/* 32 bit read so align, we really want status at offset 6 */
6384 		ciaa |= PCI_COMMAND;
6385 		IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6386 		ciad = IXGBE_READ_REG(hw, IXGBE_CIAD_82599);
6387 		ciaa &= 0x7FFFFFFF;
6388 		/* disable debug mode asap after reading data */
6389 		IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6390 		/* Get the upper 16 bits which will be the PCI status reg */
6391 		ciad >>= 16;
6392 		if (ciad & PCI_STATUS_REC_MASTER_ABORT) {
6393 			netdev_err(netdev, "VF %d Hung DMA\n", vf);
6394 			/* Issue VFLR */
6395 			ciaa = (vf << 16) | 0x80000000;
6396 			ciaa |= 0xA8;
6397 			IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6398 			ciad = 0x00008000;  /* VFLR */
6399 			IXGBE_WRITE_REG(hw, IXGBE_CIAD_82599, ciad);
6400 			ciaa &= 0x7FFFFFFF;
6401 			IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6402 		}
6403 	}
6404 }
6405 
6406 #endif
6407 /**
6408  * ixgbe_service_timer - Timer Call-back
6409  * @data: pointer to adapter cast into an unsigned long
6410  **/
6411 static void ixgbe_service_timer(unsigned long data)
6412 {
6413 	struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
6414 	unsigned long next_event_offset;
6415 	bool ready = true;
6416 
6417 	/* poll faster when waiting for link */
6418 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
6419 		next_event_offset = HZ / 10;
6420 	else
6421 		next_event_offset = HZ * 2;
6422 
6423 #ifdef CONFIG_PCI_IOV
6424 	/*
6425 	 * don't bother with SR-IOV VF DMA hang check if there are
6426 	 * no VFs or the link is down
6427 	 */
6428 	if (!adapter->num_vfs ||
6429 	    (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
6430 		goto normal_timer_service;
6431 
6432 	/* If we have VFs allocated then we must check for DMA hangs */
6433 	ixgbe_check_for_bad_vf(adapter);
6434 	next_event_offset = HZ / 50;
6435 	adapter->timer_event_accumulator++;
6436 
6437 	if (adapter->timer_event_accumulator >= 100)
6438 		adapter->timer_event_accumulator = 0;
6439 	else
6440 		ready = false;
6441 
6442 normal_timer_service:
6443 #endif
6444 	/* Reset the timer */
6445 	mod_timer(&adapter->service_timer, next_event_offset + jiffies);
6446 
6447 	if (ready)
6448 		ixgbe_service_event_schedule(adapter);
6449 }
6450 
6451 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
6452 {
6453 	if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
6454 		return;
6455 
6456 	adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
6457 
6458 	/* If we're already down, removing or resetting, just bail */
6459 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6460 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
6461 	    test_bit(__IXGBE_RESETTING, &adapter->state))
6462 		return;
6463 
6464 	ixgbe_dump(adapter);
6465 	netdev_err(adapter->netdev, "Reset adapter\n");
6466 	adapter->tx_timeout_count++;
6467 
6468 	rtnl_lock();
6469 	ixgbe_reinit_locked(adapter);
6470 	rtnl_unlock();
6471 }
6472 
6473 /**
6474  * ixgbe_service_task - manages and runs subtasks
6475  * @work: pointer to work_struct containing our data
6476  **/
6477 static void ixgbe_service_task(struct work_struct *work)
6478 {
6479 	struct ixgbe_adapter *adapter = container_of(work,
6480 						     struct ixgbe_adapter,
6481 						     service_task);
6482 	if (ixgbe_removed(adapter->hw.hw_addr)) {
6483 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
6484 			rtnl_lock();
6485 			ixgbe_down(adapter);
6486 			rtnl_unlock();
6487 		}
6488 		ixgbe_service_event_complete(adapter);
6489 		return;
6490 	}
6491 	ixgbe_reset_subtask(adapter);
6492 	ixgbe_sfp_detection_subtask(adapter);
6493 	ixgbe_sfp_link_config_subtask(adapter);
6494 	ixgbe_check_overtemp_subtask(adapter);
6495 	ixgbe_watchdog_subtask(adapter);
6496 	ixgbe_fdir_reinit_subtask(adapter);
6497 	ixgbe_check_hang_subtask(adapter);
6498 
6499 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
6500 		ixgbe_ptp_overflow_check(adapter);
6501 		ixgbe_ptp_rx_hang(adapter);
6502 	}
6503 
6504 	ixgbe_service_event_complete(adapter);
6505 }
6506 
6507 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
6508 		     struct ixgbe_tx_buffer *first,
6509 		     u8 *hdr_len)
6510 {
6511 	struct sk_buff *skb = first->skb;
6512 	u32 vlan_macip_lens, type_tucmd;
6513 	u32 mss_l4len_idx, l4len;
6514 	int err;
6515 
6516 	if (skb->ip_summed != CHECKSUM_PARTIAL)
6517 		return 0;
6518 
6519 	if (!skb_is_gso(skb))
6520 		return 0;
6521 
6522 	err = skb_cow_head(skb, 0);
6523 	if (err < 0)
6524 		return err;
6525 
6526 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6527 	type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
6528 
6529 	if (first->protocol == htons(ETH_P_IP)) {
6530 		struct iphdr *iph = ip_hdr(skb);
6531 		iph->tot_len = 0;
6532 		iph->check = 0;
6533 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6534 							 iph->daddr, 0,
6535 							 IPPROTO_TCP,
6536 							 0);
6537 		type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6538 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
6539 				   IXGBE_TX_FLAGS_CSUM |
6540 				   IXGBE_TX_FLAGS_IPV4;
6541 	} else if (skb_is_gso_v6(skb)) {
6542 		ipv6_hdr(skb)->payload_len = 0;
6543 		tcp_hdr(skb)->check =
6544 		    ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6545 				     &ipv6_hdr(skb)->daddr,
6546 				     0, IPPROTO_TCP, 0);
6547 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
6548 				   IXGBE_TX_FLAGS_CSUM;
6549 	}
6550 
6551 	/* compute header lengths */
6552 	l4len = tcp_hdrlen(skb);
6553 	*hdr_len = skb_transport_offset(skb) + l4len;
6554 
6555 	/* update gso size and bytecount with header size */
6556 	first->gso_segs = skb_shinfo(skb)->gso_segs;
6557 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
6558 
6559 	/* mss_l4len_id: use 0 as index for TSO */
6560 	mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
6561 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
6562 
6563 	/* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
6564 	vlan_macip_lens = skb_network_header_len(skb);
6565 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6566 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6567 
6568 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
6569 			  mss_l4len_idx);
6570 
6571 	return 1;
6572 }
6573 
6574 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
6575 			  struct ixgbe_tx_buffer *first)
6576 {
6577 	struct sk_buff *skb = first->skb;
6578 	u32 vlan_macip_lens = 0;
6579 	u32 mss_l4len_idx = 0;
6580 	u32 type_tucmd = 0;
6581 
6582 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
6583 		if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
6584 		    !(first->tx_flags & IXGBE_TX_FLAGS_CC))
6585 			return;
6586 	} else {
6587 		u8 l4_hdr = 0;
6588 		switch (first->protocol) {
6589 		case htons(ETH_P_IP):
6590 			vlan_macip_lens |= skb_network_header_len(skb);
6591 			type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6592 			l4_hdr = ip_hdr(skb)->protocol;
6593 			break;
6594 		case htons(ETH_P_IPV6):
6595 			vlan_macip_lens |= skb_network_header_len(skb);
6596 			l4_hdr = ipv6_hdr(skb)->nexthdr;
6597 			break;
6598 		default:
6599 			if (unlikely(net_ratelimit())) {
6600 				dev_warn(tx_ring->dev,
6601 				 "partial checksum but proto=%x!\n",
6602 				 first->protocol);
6603 			}
6604 			break;
6605 		}
6606 
6607 		switch (l4_hdr) {
6608 		case IPPROTO_TCP:
6609 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6610 			mss_l4len_idx = tcp_hdrlen(skb) <<
6611 					IXGBE_ADVTXD_L4LEN_SHIFT;
6612 			break;
6613 		case IPPROTO_SCTP:
6614 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6615 			mss_l4len_idx = sizeof(struct sctphdr) <<
6616 					IXGBE_ADVTXD_L4LEN_SHIFT;
6617 			break;
6618 		case IPPROTO_UDP:
6619 			mss_l4len_idx = sizeof(struct udphdr) <<
6620 					IXGBE_ADVTXD_L4LEN_SHIFT;
6621 			break;
6622 		default:
6623 			if (unlikely(net_ratelimit())) {
6624 				dev_warn(tx_ring->dev,
6625 				 "partial checksum but l4 proto=%x!\n",
6626 				 l4_hdr);
6627 			}
6628 			break;
6629 		}
6630 
6631 		/* update TX checksum flag */
6632 		first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
6633 	}
6634 
6635 	/* vlan_macip_lens: MACLEN, VLAN tag */
6636 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6637 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6638 
6639 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
6640 			  type_tucmd, mss_l4len_idx);
6641 }
6642 
6643 #define IXGBE_SET_FLAG(_input, _flag, _result) \
6644 	((_flag <= _result) ? \
6645 	 ((u32)(_input & _flag) * (_result / _flag)) : \
6646 	 ((u32)(_input & _flag) / (_flag / _result)))
6647 
6648 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
6649 {
6650 	/* set type for advanced descriptor with frame checksum insertion */
6651 	u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
6652 		       IXGBE_ADVTXD_DCMD_DEXT |
6653 		       IXGBE_ADVTXD_DCMD_IFCS;
6654 
6655 	/* set HW vlan bit if vlan is present */
6656 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
6657 				   IXGBE_ADVTXD_DCMD_VLE);
6658 
6659 	/* set segmentation enable bits for TSO/FSO */
6660 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
6661 				   IXGBE_ADVTXD_DCMD_TSE);
6662 
6663 	/* set timestamp bit if present */
6664 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
6665 				   IXGBE_ADVTXD_MAC_TSTAMP);
6666 
6667 	/* insert frame checksum */
6668 	cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
6669 
6670 	return cmd_type;
6671 }
6672 
6673 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
6674 				   u32 tx_flags, unsigned int paylen)
6675 {
6676 	u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
6677 
6678 	/* enable L4 checksum for TSO and TX checksum offload */
6679 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
6680 					IXGBE_TX_FLAGS_CSUM,
6681 					IXGBE_ADVTXD_POPTS_TXSM);
6682 
6683 	/* enble IPv4 checksum for TSO */
6684 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
6685 					IXGBE_TX_FLAGS_IPV4,
6686 					IXGBE_ADVTXD_POPTS_IXSM);
6687 
6688 	/*
6689 	 * Check Context must be set if Tx switch is enabled, which it
6690 	 * always is for case where virtual functions are running
6691 	 */
6692 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
6693 					IXGBE_TX_FLAGS_CC,
6694 					IXGBE_ADVTXD_CC);
6695 
6696 	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6697 }
6698 
6699 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
6700 		       IXGBE_TXD_CMD_RS)
6701 
6702 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
6703 			 struct ixgbe_tx_buffer *first,
6704 			 const u8 hdr_len)
6705 {
6706 	struct sk_buff *skb = first->skb;
6707 	struct ixgbe_tx_buffer *tx_buffer;
6708 	union ixgbe_adv_tx_desc *tx_desc;
6709 	struct skb_frag_struct *frag;
6710 	dma_addr_t dma;
6711 	unsigned int data_len, size;
6712 	u32 tx_flags = first->tx_flags;
6713 	u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
6714 	u16 i = tx_ring->next_to_use;
6715 
6716 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
6717 
6718 	ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
6719 
6720 	size = skb_headlen(skb);
6721 	data_len = skb->data_len;
6722 
6723 #ifdef IXGBE_FCOE
6724 	if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6725 		if (data_len < sizeof(struct fcoe_crc_eof)) {
6726 			size -= sizeof(struct fcoe_crc_eof) - data_len;
6727 			data_len = 0;
6728 		} else {
6729 			data_len -= sizeof(struct fcoe_crc_eof);
6730 		}
6731 	}
6732 
6733 #endif
6734 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
6735 
6736 	tx_buffer = first;
6737 
6738 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
6739 		if (dma_mapping_error(tx_ring->dev, dma))
6740 			goto dma_error;
6741 
6742 		/* record length, and DMA address */
6743 		dma_unmap_len_set(tx_buffer, len, size);
6744 		dma_unmap_addr_set(tx_buffer, dma, dma);
6745 
6746 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
6747 
6748 		while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
6749 			tx_desc->read.cmd_type_len =
6750 				cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
6751 
6752 			i++;
6753 			tx_desc++;
6754 			if (i == tx_ring->count) {
6755 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6756 				i = 0;
6757 			}
6758 			tx_desc->read.olinfo_status = 0;
6759 
6760 			dma += IXGBE_MAX_DATA_PER_TXD;
6761 			size -= IXGBE_MAX_DATA_PER_TXD;
6762 
6763 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
6764 		}
6765 
6766 		if (likely(!data_len))
6767 			break;
6768 
6769 		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
6770 
6771 		i++;
6772 		tx_desc++;
6773 		if (i == tx_ring->count) {
6774 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6775 			i = 0;
6776 		}
6777 		tx_desc->read.olinfo_status = 0;
6778 
6779 #ifdef IXGBE_FCOE
6780 		size = min_t(unsigned int, data_len, skb_frag_size(frag));
6781 #else
6782 		size = skb_frag_size(frag);
6783 #endif
6784 		data_len -= size;
6785 
6786 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
6787 				       DMA_TO_DEVICE);
6788 
6789 		tx_buffer = &tx_ring->tx_buffer_info[i];
6790 	}
6791 
6792 	/* write last descriptor with RS and EOP bits */
6793 	cmd_type |= size | IXGBE_TXD_CMD;
6794 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
6795 
6796 	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
6797 
6798 	/* set the timestamp */
6799 	first->time_stamp = jiffies;
6800 
6801 	/*
6802 	 * Force memory writes to complete before letting h/w know there
6803 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
6804 	 * memory model archs, such as IA-64).
6805 	 *
6806 	 * We also need this memory barrier to make certain all of the
6807 	 * status bits have been updated before next_to_watch is written.
6808 	 */
6809 	wmb();
6810 
6811 	/* set next_to_watch value indicating a packet is present */
6812 	first->next_to_watch = tx_desc;
6813 
6814 	i++;
6815 	if (i == tx_ring->count)
6816 		i = 0;
6817 
6818 	tx_ring->next_to_use = i;
6819 
6820 	/* notify HW of packet */
6821 	ixgbe_write_tail(tx_ring, i);
6822 
6823 	return;
6824 dma_error:
6825 	dev_err(tx_ring->dev, "TX DMA map failed\n");
6826 
6827 	/* clear dma mappings for failed tx_buffer_info map */
6828 	for (;;) {
6829 		tx_buffer = &tx_ring->tx_buffer_info[i];
6830 		ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
6831 		if (tx_buffer == first)
6832 			break;
6833 		if (i == 0)
6834 			i = tx_ring->count;
6835 		i--;
6836 	}
6837 
6838 	tx_ring->next_to_use = i;
6839 }
6840 
6841 static void ixgbe_atr(struct ixgbe_ring *ring,
6842 		      struct ixgbe_tx_buffer *first)
6843 {
6844 	struct ixgbe_q_vector *q_vector = ring->q_vector;
6845 	union ixgbe_atr_hash_dword input = { .dword = 0 };
6846 	union ixgbe_atr_hash_dword common = { .dword = 0 };
6847 	union {
6848 		unsigned char *network;
6849 		struct iphdr *ipv4;
6850 		struct ipv6hdr *ipv6;
6851 	} hdr;
6852 	struct tcphdr *th;
6853 	__be16 vlan_id;
6854 
6855 	/* if ring doesn't have a interrupt vector, cannot perform ATR */
6856 	if (!q_vector)
6857 		return;
6858 
6859 	/* do nothing if sampling is disabled */
6860 	if (!ring->atr_sample_rate)
6861 		return;
6862 
6863 	ring->atr_count++;
6864 
6865 	/* snag network header to get L4 type and address */
6866 	hdr.network = skb_network_header(first->skb);
6867 
6868 	/* Currently only IPv4/IPv6 with TCP is supported */
6869 	if ((first->protocol != htons(ETH_P_IPV6) ||
6870 	     hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6871 	    (first->protocol != htons(ETH_P_IP) ||
6872 	     hdr.ipv4->protocol != IPPROTO_TCP))
6873 		return;
6874 
6875 	th = tcp_hdr(first->skb);
6876 
6877 	/* skip this packet since it is invalid or the socket is closing */
6878 	if (!th || th->fin)
6879 		return;
6880 
6881 	/* sample on all syn packets or once every atr sample count */
6882 	if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6883 		return;
6884 
6885 	/* reset sample count */
6886 	ring->atr_count = 0;
6887 
6888 	vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6889 
6890 	/*
6891 	 * src and dst are inverted, think how the receiver sees them
6892 	 *
6893 	 * The input is broken into two sections, a non-compressed section
6894 	 * containing vm_pool, vlan_id, and flow_type.  The rest of the data
6895 	 * is XORed together and stored in the compressed dword.
6896 	 */
6897 	input.formatted.vlan_id = vlan_id;
6898 
6899 	/*
6900 	 * since src port and flex bytes occupy the same word XOR them together
6901 	 * and write the value to source port portion of compressed dword
6902 	 */
6903 	if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6904 		common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
6905 	else
6906 		common.port.src ^= th->dest ^ first->protocol;
6907 	common.port.dst ^= th->source;
6908 
6909 	if (first->protocol == htons(ETH_P_IP)) {
6910 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6911 		common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6912 	} else {
6913 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6914 		common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6915 			     hdr.ipv6->saddr.s6_addr32[1] ^
6916 			     hdr.ipv6->saddr.s6_addr32[2] ^
6917 			     hdr.ipv6->saddr.s6_addr32[3] ^
6918 			     hdr.ipv6->daddr.s6_addr32[0] ^
6919 			     hdr.ipv6->daddr.s6_addr32[1] ^
6920 			     hdr.ipv6->daddr.s6_addr32[2] ^
6921 			     hdr.ipv6->daddr.s6_addr32[3];
6922 	}
6923 
6924 	/* This assumes the Rx queue and Tx queue are bound to the same CPU */
6925 	ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6926 					      input, common, ring->queue_index);
6927 }
6928 
6929 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6930 {
6931 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6932 	/* Herbert's original patch had:
6933 	 *  smp_mb__after_netif_stop_queue();
6934 	 * but since that doesn't exist yet, just open code it. */
6935 	smp_mb();
6936 
6937 	/* We need to check again in a case another CPU has just
6938 	 * made room available. */
6939 	if (likely(ixgbe_desc_unused(tx_ring) < size))
6940 		return -EBUSY;
6941 
6942 	/* A reprieve! - use start_queue because it doesn't call schedule */
6943 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6944 	++tx_ring->tx_stats.restart_queue;
6945 	return 0;
6946 }
6947 
6948 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6949 {
6950 	if (likely(ixgbe_desc_unused(tx_ring) >= size))
6951 		return 0;
6952 	return __ixgbe_maybe_stop_tx(tx_ring, size);
6953 }
6954 
6955 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
6956 			      void *accel_priv, select_queue_fallback_t fallback)
6957 {
6958 	struct ixgbe_fwd_adapter *fwd_adapter = accel_priv;
6959 #ifdef IXGBE_FCOE
6960 	struct ixgbe_adapter *adapter;
6961 	struct ixgbe_ring_feature *f;
6962 	int txq;
6963 #endif
6964 
6965 	if (fwd_adapter)
6966 		return skb->queue_mapping + fwd_adapter->tx_base_queue;
6967 
6968 #ifdef IXGBE_FCOE
6969 
6970 	/*
6971 	 * only execute the code below if protocol is FCoE
6972 	 * or FIP and we have FCoE enabled on the adapter
6973 	 */
6974 	switch (vlan_get_protocol(skb)) {
6975 	case htons(ETH_P_FCOE):
6976 	case htons(ETH_P_FIP):
6977 		adapter = netdev_priv(dev);
6978 
6979 		if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
6980 			break;
6981 	default:
6982 		return fallback(dev, skb);
6983 	}
6984 
6985 	f = &adapter->ring_feature[RING_F_FCOE];
6986 
6987 	txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6988 					   smp_processor_id();
6989 
6990 	while (txq >= f->indices)
6991 		txq -= f->indices;
6992 
6993 	return txq + f->offset;
6994 #else
6995 	return fallback(dev, skb);
6996 #endif
6997 }
6998 
6999 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
7000 			  struct ixgbe_adapter *adapter,
7001 			  struct ixgbe_ring *tx_ring)
7002 {
7003 	struct ixgbe_tx_buffer *first;
7004 	int tso;
7005 	u32 tx_flags = 0;
7006 	unsigned short f;
7007 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
7008 	__be16 protocol = skb->protocol;
7009 	u8 hdr_len = 0;
7010 
7011 	/*
7012 	 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
7013 	 *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
7014 	 *       + 2 desc gap to keep tail from touching head,
7015 	 *       + 1 desc for context descriptor,
7016 	 * otherwise try next time
7017 	 */
7018 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
7019 		count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
7020 
7021 	if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
7022 		tx_ring->tx_stats.tx_busy++;
7023 		return NETDEV_TX_BUSY;
7024 	}
7025 
7026 	/* record the location of the first descriptor for this packet */
7027 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
7028 	first->skb = skb;
7029 	first->bytecount = skb->len;
7030 	first->gso_segs = 1;
7031 
7032 	/* if we have a HW VLAN tag being added default to the HW one */
7033 	if (vlan_tx_tag_present(skb)) {
7034 		tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
7035 		tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
7036 	/* else if it is a SW VLAN check the next protocol and store the tag */
7037 	} else if (protocol == htons(ETH_P_8021Q)) {
7038 		struct vlan_hdr *vhdr, _vhdr;
7039 		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
7040 		if (!vhdr)
7041 			goto out_drop;
7042 
7043 		protocol = vhdr->h_vlan_encapsulated_proto;
7044 		tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
7045 				  IXGBE_TX_FLAGS_VLAN_SHIFT;
7046 		tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
7047 	}
7048 
7049 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP &&
7050 		     !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
7051 					    &adapter->state))) {
7052 		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
7053 		tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
7054 
7055 		/* schedule check for Tx timestamp */
7056 		adapter->ptp_tx_skb = skb_get(skb);
7057 		adapter->ptp_tx_start = jiffies;
7058 		schedule_work(&adapter->ptp_tx_work);
7059 	}
7060 
7061 	skb_tx_timestamp(skb);
7062 
7063 #ifdef CONFIG_PCI_IOV
7064 	/*
7065 	 * Use the l2switch_enable flag - would be false if the DMA
7066 	 * Tx switch had been disabled.
7067 	 */
7068 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7069 		tx_flags |= IXGBE_TX_FLAGS_CC;
7070 
7071 #endif
7072 	/* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
7073 	if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
7074 	    ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
7075 	     (skb->priority != TC_PRIO_CONTROL))) {
7076 		tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
7077 		tx_flags |= (skb->priority & 0x7) <<
7078 					IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
7079 		if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
7080 			struct vlan_ethhdr *vhdr;
7081 
7082 			if (skb_cow_head(skb, 0))
7083 				goto out_drop;
7084 			vhdr = (struct vlan_ethhdr *)skb->data;
7085 			vhdr->h_vlan_TCI = htons(tx_flags >>
7086 						 IXGBE_TX_FLAGS_VLAN_SHIFT);
7087 		} else {
7088 			tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
7089 		}
7090 	}
7091 
7092 	/* record initial flags and protocol */
7093 	first->tx_flags = tx_flags;
7094 	first->protocol = protocol;
7095 
7096 #ifdef IXGBE_FCOE
7097 	/* setup tx offload for FCoE */
7098 	if ((protocol == htons(ETH_P_FCOE)) &&
7099 	    (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
7100 		tso = ixgbe_fso(tx_ring, first, &hdr_len);
7101 		if (tso < 0)
7102 			goto out_drop;
7103 
7104 		goto xmit_fcoe;
7105 	}
7106 
7107 #endif /* IXGBE_FCOE */
7108 	tso = ixgbe_tso(tx_ring, first, &hdr_len);
7109 	if (tso < 0)
7110 		goto out_drop;
7111 	else if (!tso)
7112 		ixgbe_tx_csum(tx_ring, first);
7113 
7114 	/* add the ATR filter if ATR is on */
7115 	if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
7116 		ixgbe_atr(tx_ring, first);
7117 
7118 #ifdef IXGBE_FCOE
7119 xmit_fcoe:
7120 #endif /* IXGBE_FCOE */
7121 	ixgbe_tx_map(tx_ring, first, hdr_len);
7122 
7123 	ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
7124 
7125 	return NETDEV_TX_OK;
7126 
7127 out_drop:
7128 	dev_kfree_skb_any(first->skb);
7129 	first->skb = NULL;
7130 
7131 	return NETDEV_TX_OK;
7132 }
7133 
7134 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
7135 				      struct net_device *netdev,
7136 				      struct ixgbe_ring *ring)
7137 {
7138 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7139 	struct ixgbe_ring *tx_ring;
7140 
7141 	/*
7142 	 * The minimum packet size for olinfo paylen is 17 so pad the skb
7143 	 * in order to meet this minimum size requirement.
7144 	 */
7145 	if (unlikely(skb->len < 17)) {
7146 		if (skb_pad(skb, 17 - skb->len))
7147 			return NETDEV_TX_OK;
7148 		skb->len = 17;
7149 		skb_set_tail_pointer(skb, 17);
7150 	}
7151 
7152 	tx_ring = ring ? ring : adapter->tx_ring[skb->queue_mapping];
7153 
7154 	return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
7155 }
7156 
7157 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
7158 				    struct net_device *netdev)
7159 {
7160 	return __ixgbe_xmit_frame(skb, netdev, NULL);
7161 }
7162 
7163 /**
7164  * ixgbe_set_mac - Change the Ethernet Address of the NIC
7165  * @netdev: network interface device structure
7166  * @p: pointer to an address structure
7167  *
7168  * Returns 0 on success, negative on failure
7169  **/
7170 static int ixgbe_set_mac(struct net_device *netdev, void *p)
7171 {
7172 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7173 	struct ixgbe_hw *hw = &adapter->hw;
7174 	struct sockaddr *addr = p;
7175 
7176 	if (!is_valid_ether_addr(addr->sa_data))
7177 		return -EADDRNOTAVAIL;
7178 
7179 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
7180 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
7181 
7182 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, VMDQ_P(0), IXGBE_RAH_AV);
7183 
7184 	return 0;
7185 }
7186 
7187 static int
7188 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
7189 {
7190 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7191 	struct ixgbe_hw *hw = &adapter->hw;
7192 	u16 value;
7193 	int rc;
7194 
7195 	if (prtad != hw->phy.mdio.prtad)
7196 		return -EINVAL;
7197 	rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
7198 	if (!rc)
7199 		rc = value;
7200 	return rc;
7201 }
7202 
7203 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
7204 			    u16 addr, u16 value)
7205 {
7206 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7207 	struct ixgbe_hw *hw = &adapter->hw;
7208 
7209 	if (prtad != hw->phy.mdio.prtad)
7210 		return -EINVAL;
7211 	return hw->phy.ops.write_reg(hw, addr, devad, value);
7212 }
7213 
7214 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
7215 {
7216 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7217 
7218 	switch (cmd) {
7219 	case SIOCSHWTSTAMP:
7220 		return ixgbe_ptp_set_ts_config(adapter, req);
7221 	case SIOCGHWTSTAMP:
7222 		return ixgbe_ptp_get_ts_config(adapter, req);
7223 	default:
7224 		return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
7225 	}
7226 }
7227 
7228 /**
7229  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
7230  * netdev->dev_addrs
7231  * @netdev: network interface device structure
7232  *
7233  * Returns non-zero on failure
7234  **/
7235 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
7236 {
7237 	int err = 0;
7238 	struct ixgbe_adapter *adapter = netdev_priv(dev);
7239 	struct ixgbe_hw *hw = &adapter->hw;
7240 
7241 	if (is_valid_ether_addr(hw->mac.san_addr)) {
7242 		rtnl_lock();
7243 		err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
7244 		rtnl_unlock();
7245 
7246 		/* update SAN MAC vmdq pool selection */
7247 		hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
7248 	}
7249 	return err;
7250 }
7251 
7252 /**
7253  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
7254  * netdev->dev_addrs
7255  * @netdev: network interface device structure
7256  *
7257  * Returns non-zero on failure
7258  **/
7259 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
7260 {
7261 	int err = 0;
7262 	struct ixgbe_adapter *adapter = netdev_priv(dev);
7263 	struct ixgbe_mac_info *mac = &adapter->hw.mac;
7264 
7265 	if (is_valid_ether_addr(mac->san_addr)) {
7266 		rtnl_lock();
7267 		err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
7268 		rtnl_unlock();
7269 	}
7270 	return err;
7271 }
7272 
7273 #ifdef CONFIG_NET_POLL_CONTROLLER
7274 /*
7275  * Polling 'interrupt' - used by things like netconsole to send skbs
7276  * without having to re-enable interrupts. It's not called while
7277  * the interrupt routine is executing.
7278  */
7279 static void ixgbe_netpoll(struct net_device *netdev)
7280 {
7281 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7282 	int i;
7283 
7284 	/* if interface is down do nothing */
7285 	if (test_bit(__IXGBE_DOWN, &adapter->state))
7286 		return;
7287 
7288 	adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
7289 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
7290 		for (i = 0; i < adapter->num_q_vectors; i++)
7291 			ixgbe_msix_clean_rings(0, adapter->q_vector[i]);
7292 	} else {
7293 		ixgbe_intr(adapter->pdev->irq, netdev);
7294 	}
7295 	adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
7296 }
7297 
7298 #endif
7299 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
7300 						   struct rtnl_link_stats64 *stats)
7301 {
7302 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7303 	int i;
7304 
7305 	rcu_read_lock();
7306 	for (i = 0; i < adapter->num_rx_queues; i++) {
7307 		struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
7308 		u64 bytes, packets;
7309 		unsigned int start;
7310 
7311 		if (ring) {
7312 			do {
7313 				start = u64_stats_fetch_begin_irq(&ring->syncp);
7314 				packets = ring->stats.packets;
7315 				bytes   = ring->stats.bytes;
7316 			} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
7317 			stats->rx_packets += packets;
7318 			stats->rx_bytes   += bytes;
7319 		}
7320 	}
7321 
7322 	for (i = 0; i < adapter->num_tx_queues; i++) {
7323 		struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
7324 		u64 bytes, packets;
7325 		unsigned int start;
7326 
7327 		if (ring) {
7328 			do {
7329 				start = u64_stats_fetch_begin_irq(&ring->syncp);
7330 				packets = ring->stats.packets;
7331 				bytes   = ring->stats.bytes;
7332 			} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
7333 			stats->tx_packets += packets;
7334 			stats->tx_bytes   += bytes;
7335 		}
7336 	}
7337 	rcu_read_unlock();
7338 	/* following stats updated by ixgbe_watchdog_task() */
7339 	stats->multicast	= netdev->stats.multicast;
7340 	stats->rx_errors	= netdev->stats.rx_errors;
7341 	stats->rx_length_errors	= netdev->stats.rx_length_errors;
7342 	stats->rx_crc_errors	= netdev->stats.rx_crc_errors;
7343 	stats->rx_missed_errors	= netdev->stats.rx_missed_errors;
7344 	return stats;
7345 }
7346 
7347 #ifdef CONFIG_IXGBE_DCB
7348 /**
7349  * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
7350  * @adapter: pointer to ixgbe_adapter
7351  * @tc: number of traffic classes currently enabled
7352  *
7353  * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
7354  * 802.1Q priority maps to a packet buffer that exists.
7355  */
7356 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
7357 {
7358 	struct ixgbe_hw *hw = &adapter->hw;
7359 	u32 reg, rsave;
7360 	int i;
7361 
7362 	/* 82598 have a static priority to TC mapping that can not
7363 	 * be changed so no validation is needed.
7364 	 */
7365 	if (hw->mac.type == ixgbe_mac_82598EB)
7366 		return;
7367 
7368 	reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
7369 	rsave = reg;
7370 
7371 	for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
7372 		u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
7373 
7374 		/* If up2tc is out of bounds default to zero */
7375 		if (up2tc > tc)
7376 			reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
7377 	}
7378 
7379 	if (reg != rsave)
7380 		IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
7381 
7382 	return;
7383 }
7384 
7385 /**
7386  * ixgbe_set_prio_tc_map - Configure netdev prio tc map
7387  * @adapter: Pointer to adapter struct
7388  *
7389  * Populate the netdev user priority to tc map
7390  */
7391 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
7392 {
7393 	struct net_device *dev = adapter->netdev;
7394 	struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
7395 	struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
7396 	u8 prio;
7397 
7398 	for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
7399 		u8 tc = 0;
7400 
7401 		if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
7402 			tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
7403 		else if (ets)
7404 			tc = ets->prio_tc[prio];
7405 
7406 		netdev_set_prio_tc_map(dev, prio, tc);
7407 	}
7408 }
7409 
7410 #endif /* CONFIG_IXGBE_DCB */
7411 /**
7412  * ixgbe_setup_tc - configure net_device for multiple traffic classes
7413  *
7414  * @netdev: net device to configure
7415  * @tc: number of traffic classes to enable
7416  */
7417 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
7418 {
7419 	struct ixgbe_adapter *adapter = netdev_priv(dev);
7420 	struct ixgbe_hw *hw = &adapter->hw;
7421 	bool pools;
7422 
7423 	/* Hardware supports up to 8 traffic classes */
7424 	if (tc > adapter->dcb_cfg.num_tcs.pg_tcs ||
7425 	    (hw->mac.type == ixgbe_mac_82598EB &&
7426 	     tc < MAX_TRAFFIC_CLASS))
7427 		return -EINVAL;
7428 
7429 	pools = (find_first_zero_bit(&adapter->fwd_bitmask, 32) > 1);
7430 	if (tc && pools && adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS)
7431 		return -EBUSY;
7432 
7433 	/* Hardware has to reinitialize queues and interrupts to
7434 	 * match packet buffer alignment. Unfortunately, the
7435 	 * hardware is not flexible enough to do this dynamically.
7436 	 */
7437 	if (netif_running(dev))
7438 		ixgbe_close(dev);
7439 	ixgbe_clear_interrupt_scheme(adapter);
7440 
7441 #ifdef CONFIG_IXGBE_DCB
7442 	if (tc) {
7443 		netdev_set_num_tc(dev, tc);
7444 		ixgbe_set_prio_tc_map(adapter);
7445 
7446 		adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
7447 
7448 		if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
7449 			adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
7450 			adapter->hw.fc.requested_mode = ixgbe_fc_none;
7451 		}
7452 	} else {
7453 		netdev_reset_tc(dev);
7454 
7455 		if (adapter->hw.mac.type == ixgbe_mac_82598EB)
7456 			adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
7457 
7458 		adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
7459 
7460 		adapter->temp_dcb_cfg.pfc_mode_enable = false;
7461 		adapter->dcb_cfg.pfc_mode_enable = false;
7462 	}
7463 
7464 	ixgbe_validate_rtr(adapter, tc);
7465 
7466 #endif /* CONFIG_IXGBE_DCB */
7467 	ixgbe_init_interrupt_scheme(adapter);
7468 
7469 	if (netif_running(dev))
7470 		return ixgbe_open(dev);
7471 
7472 	return 0;
7473 }
7474 
7475 #ifdef CONFIG_PCI_IOV
7476 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
7477 {
7478 	struct net_device *netdev = adapter->netdev;
7479 
7480 	rtnl_lock();
7481 	ixgbe_setup_tc(netdev, netdev_get_num_tc(netdev));
7482 	rtnl_unlock();
7483 }
7484 
7485 #endif
7486 void ixgbe_do_reset(struct net_device *netdev)
7487 {
7488 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7489 
7490 	if (netif_running(netdev))
7491 		ixgbe_reinit_locked(adapter);
7492 	else
7493 		ixgbe_reset(adapter);
7494 }
7495 
7496 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
7497 					    netdev_features_t features)
7498 {
7499 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7500 
7501 	/* If Rx checksum is disabled, then RSC/LRO should also be disabled */
7502 	if (!(features & NETIF_F_RXCSUM))
7503 		features &= ~NETIF_F_LRO;
7504 
7505 	/* Turn off LRO if not RSC capable */
7506 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
7507 		features &= ~NETIF_F_LRO;
7508 
7509 	return features;
7510 }
7511 
7512 static int ixgbe_set_features(struct net_device *netdev,
7513 			      netdev_features_t features)
7514 {
7515 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7516 	netdev_features_t changed = netdev->features ^ features;
7517 	bool need_reset = false;
7518 
7519 	/* Make sure RSC matches LRO, reset if change */
7520 	if (!(features & NETIF_F_LRO)) {
7521 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7522 			need_reset = true;
7523 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
7524 	} else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
7525 		   !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
7526 		if (adapter->rx_itr_setting == 1 ||
7527 		    adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
7528 			adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
7529 			need_reset = true;
7530 		} else if ((changed ^ features) & NETIF_F_LRO) {
7531 			e_info(probe, "rx-usecs set too low, "
7532 			       "disabling RSC\n");
7533 		}
7534 	}
7535 
7536 	/*
7537 	 * Check if Flow Director n-tuple support was enabled or disabled.  If
7538 	 * the state changed, we need to reset.
7539 	 */
7540 	switch (features & NETIF_F_NTUPLE) {
7541 	case NETIF_F_NTUPLE:
7542 		/* turn off ATR, enable perfect filters and reset */
7543 		if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
7544 			need_reset = true;
7545 
7546 		adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
7547 		adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7548 		break;
7549 	default:
7550 		/* turn off perfect filters, enable ATR and reset */
7551 		if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7552 			need_reset = true;
7553 
7554 		adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7555 
7556 		/* We cannot enable ATR if SR-IOV is enabled */
7557 		if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7558 			break;
7559 
7560 		/* We cannot enable ATR if we have 2 or more traffic classes */
7561 		if (netdev_get_num_tc(netdev) > 1)
7562 			break;
7563 
7564 		/* We cannot enable ATR if RSS is disabled */
7565 		if (adapter->ring_feature[RING_F_RSS].limit <= 1)
7566 			break;
7567 
7568 		/* A sample rate of 0 indicates ATR disabled */
7569 		if (!adapter->atr_sample_rate)
7570 			break;
7571 
7572 		adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
7573 		break;
7574 	}
7575 
7576 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
7577 		ixgbe_vlan_strip_enable(adapter);
7578 	else
7579 		ixgbe_vlan_strip_disable(adapter);
7580 
7581 	if (changed & NETIF_F_RXALL)
7582 		need_reset = true;
7583 
7584 	netdev->features = features;
7585 	if (need_reset)
7586 		ixgbe_do_reset(netdev);
7587 
7588 	return 0;
7589 }
7590 
7591 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
7592 			     struct net_device *dev,
7593 			     const unsigned char *addr,
7594 			     u16 flags)
7595 {
7596 	struct ixgbe_adapter *adapter = netdev_priv(dev);
7597 	int err;
7598 
7599 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
7600 		return ndo_dflt_fdb_add(ndm, tb, dev, addr, flags);
7601 
7602 	/* Hardware does not support aging addresses so if a
7603 	 * ndm_state is given only allow permanent addresses
7604 	 */
7605 	if (ndm->ndm_state && !(ndm->ndm_state & NUD_PERMANENT)) {
7606 		pr_info("%s: FDB only supports static addresses\n",
7607 			ixgbe_driver_name);
7608 		return -EINVAL;
7609 	}
7610 
7611 	if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
7612 		u32 rar_uc_entries = IXGBE_MAX_PF_MACVLANS;
7613 
7614 		if (netdev_uc_count(dev) < rar_uc_entries)
7615 			err = dev_uc_add_excl(dev, addr);
7616 		else
7617 			err = -ENOMEM;
7618 	} else if (is_multicast_ether_addr(addr)) {
7619 		err = dev_mc_add_excl(dev, addr);
7620 	} else {
7621 		err = -EINVAL;
7622 	}
7623 
7624 	/* Only return duplicate errors if NLM_F_EXCL is set */
7625 	if (err == -EEXIST && !(flags & NLM_F_EXCL))
7626 		err = 0;
7627 
7628 	return err;
7629 }
7630 
7631 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
7632 				    struct nlmsghdr *nlh)
7633 {
7634 	struct ixgbe_adapter *adapter = netdev_priv(dev);
7635 	struct nlattr *attr, *br_spec;
7636 	int rem;
7637 
7638 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
7639 		return -EOPNOTSUPP;
7640 
7641 	br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
7642 
7643 	nla_for_each_nested(attr, br_spec, rem) {
7644 		__u16 mode;
7645 		u32 reg = 0;
7646 
7647 		if (nla_type(attr) != IFLA_BRIDGE_MODE)
7648 			continue;
7649 
7650 		mode = nla_get_u16(attr);
7651 		if (mode == BRIDGE_MODE_VEPA) {
7652 			reg = 0;
7653 			adapter->flags2 &= ~IXGBE_FLAG2_BRIDGE_MODE_VEB;
7654 		} else if (mode == BRIDGE_MODE_VEB) {
7655 			reg = IXGBE_PFDTXGSWC_VT_LBEN;
7656 			adapter->flags2 |= IXGBE_FLAG2_BRIDGE_MODE_VEB;
7657 		} else
7658 			return -EINVAL;
7659 
7660 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, reg);
7661 
7662 		e_info(drv, "enabling bridge mode: %s\n",
7663 			mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
7664 	}
7665 
7666 	return 0;
7667 }
7668 
7669 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
7670 				    struct net_device *dev,
7671 				    u32 filter_mask)
7672 {
7673 	struct ixgbe_adapter *adapter = netdev_priv(dev);
7674 	u16 mode;
7675 
7676 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
7677 		return 0;
7678 
7679 	if (adapter->flags2 & IXGBE_FLAG2_BRIDGE_MODE_VEB)
7680 		mode = BRIDGE_MODE_VEB;
7681 	else
7682 		mode = BRIDGE_MODE_VEPA;
7683 
7684 	return ndo_dflt_bridge_getlink(skb, pid, seq, dev, mode);
7685 }
7686 
7687 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
7688 {
7689 	struct ixgbe_fwd_adapter *fwd_adapter = NULL;
7690 	struct ixgbe_adapter *adapter = netdev_priv(pdev);
7691 	unsigned int limit;
7692 	int pool, err;
7693 
7694 #ifdef CONFIG_RPS
7695 	if (vdev->num_rx_queues != vdev->num_tx_queues) {
7696 		netdev_info(pdev, "%s: Only supports a single queue count for TX and RX\n",
7697 			    vdev->name);
7698 		return ERR_PTR(-EINVAL);
7699 	}
7700 #endif
7701 	/* Check for hardware restriction on number of rx/tx queues */
7702 	if (vdev->num_tx_queues > IXGBE_MAX_L2A_QUEUES ||
7703 	    vdev->num_tx_queues == IXGBE_BAD_L2A_QUEUE) {
7704 		netdev_info(pdev,
7705 			    "%s: Supports RX/TX Queue counts 1,2, and 4\n",
7706 			    pdev->name);
7707 		return ERR_PTR(-EINVAL);
7708 	}
7709 
7710 	if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
7711 	      adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS - 1) ||
7712 	    (adapter->num_rx_pools > IXGBE_MAX_MACVLANS))
7713 		return ERR_PTR(-EBUSY);
7714 
7715 	fwd_adapter = kcalloc(1, sizeof(struct ixgbe_fwd_adapter), GFP_KERNEL);
7716 	if (!fwd_adapter)
7717 		return ERR_PTR(-ENOMEM);
7718 
7719 	pool = find_first_zero_bit(&adapter->fwd_bitmask, 32);
7720 	adapter->num_rx_pools++;
7721 	set_bit(pool, &adapter->fwd_bitmask);
7722 	limit = find_last_bit(&adapter->fwd_bitmask, 32);
7723 
7724 	/* Enable VMDq flag so device will be set in VM mode */
7725 	adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED | IXGBE_FLAG_SRIOV_ENABLED;
7726 	adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
7727 	adapter->ring_feature[RING_F_RSS].limit = vdev->num_tx_queues;
7728 
7729 	/* Force reinit of ring allocation with VMDQ enabled */
7730 	err = ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev));
7731 	if (err)
7732 		goto fwd_add_err;
7733 	fwd_adapter->pool = pool;
7734 	fwd_adapter->real_adapter = adapter;
7735 	err = ixgbe_fwd_ring_up(vdev, fwd_adapter);
7736 	if (err)
7737 		goto fwd_add_err;
7738 	netif_tx_start_all_queues(vdev);
7739 	return fwd_adapter;
7740 fwd_add_err:
7741 	/* unwind counter and free adapter struct */
7742 	netdev_info(pdev,
7743 		    "%s: dfwd hardware acceleration failed\n", vdev->name);
7744 	clear_bit(pool, &adapter->fwd_bitmask);
7745 	adapter->num_rx_pools--;
7746 	kfree(fwd_adapter);
7747 	return ERR_PTR(err);
7748 }
7749 
7750 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
7751 {
7752 	struct ixgbe_fwd_adapter *fwd_adapter = priv;
7753 	struct ixgbe_adapter *adapter = fwd_adapter->real_adapter;
7754 	unsigned int limit;
7755 
7756 	clear_bit(fwd_adapter->pool, &adapter->fwd_bitmask);
7757 	adapter->num_rx_pools--;
7758 
7759 	limit = find_last_bit(&adapter->fwd_bitmask, 32);
7760 	adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
7761 	ixgbe_fwd_ring_down(fwd_adapter->netdev, fwd_adapter);
7762 	ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev));
7763 	netdev_dbg(pdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
7764 		   fwd_adapter->pool, adapter->num_rx_pools,
7765 		   fwd_adapter->rx_base_queue,
7766 		   fwd_adapter->rx_base_queue + adapter->num_rx_queues_per_pool,
7767 		   adapter->fwd_bitmask);
7768 	kfree(fwd_adapter);
7769 }
7770 
7771 static const struct net_device_ops ixgbe_netdev_ops = {
7772 	.ndo_open		= ixgbe_open,
7773 	.ndo_stop		= ixgbe_close,
7774 	.ndo_start_xmit		= ixgbe_xmit_frame,
7775 	.ndo_select_queue	= ixgbe_select_queue,
7776 	.ndo_set_rx_mode	= ixgbe_set_rx_mode,
7777 	.ndo_validate_addr	= eth_validate_addr,
7778 	.ndo_set_mac_address	= ixgbe_set_mac,
7779 	.ndo_change_mtu		= ixgbe_change_mtu,
7780 	.ndo_tx_timeout		= ixgbe_tx_timeout,
7781 	.ndo_vlan_rx_add_vid	= ixgbe_vlan_rx_add_vid,
7782 	.ndo_vlan_rx_kill_vid	= ixgbe_vlan_rx_kill_vid,
7783 	.ndo_do_ioctl		= ixgbe_ioctl,
7784 	.ndo_set_vf_mac		= ixgbe_ndo_set_vf_mac,
7785 	.ndo_set_vf_vlan	= ixgbe_ndo_set_vf_vlan,
7786 	.ndo_set_vf_tx_rate	= ixgbe_ndo_set_vf_bw,
7787 	.ndo_set_vf_spoofchk	= ixgbe_ndo_set_vf_spoofchk,
7788 	.ndo_get_vf_config	= ixgbe_ndo_get_vf_config,
7789 	.ndo_get_stats64	= ixgbe_get_stats64,
7790 #ifdef CONFIG_IXGBE_DCB
7791 	.ndo_setup_tc		= ixgbe_setup_tc,
7792 #endif
7793 #ifdef CONFIG_NET_POLL_CONTROLLER
7794 	.ndo_poll_controller	= ixgbe_netpoll,
7795 #endif
7796 #ifdef CONFIG_NET_RX_BUSY_POLL
7797 	.ndo_busy_poll		= ixgbe_low_latency_recv,
7798 #endif
7799 #ifdef IXGBE_FCOE
7800 	.ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
7801 	.ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
7802 	.ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
7803 	.ndo_fcoe_enable = ixgbe_fcoe_enable,
7804 	.ndo_fcoe_disable = ixgbe_fcoe_disable,
7805 	.ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
7806 	.ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
7807 #endif /* IXGBE_FCOE */
7808 	.ndo_set_features = ixgbe_set_features,
7809 	.ndo_fix_features = ixgbe_fix_features,
7810 	.ndo_fdb_add		= ixgbe_ndo_fdb_add,
7811 	.ndo_bridge_setlink	= ixgbe_ndo_bridge_setlink,
7812 	.ndo_bridge_getlink	= ixgbe_ndo_bridge_getlink,
7813 	.ndo_dfwd_add_station	= ixgbe_fwd_add,
7814 	.ndo_dfwd_del_station	= ixgbe_fwd_del,
7815 };
7816 
7817 /**
7818  * ixgbe_enumerate_functions - Get the number of ports this device has
7819  * @adapter: adapter structure
7820  *
7821  * This function enumerates the phsyical functions co-located on a single slot,
7822  * in order to determine how many ports a device has. This is most useful in
7823  * determining the required GT/s of PCIe bandwidth necessary for optimal
7824  * performance.
7825  **/
7826 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
7827 {
7828 	struct list_head *entry;
7829 	int physfns = 0;
7830 
7831 	/* Some cards can not use the generic count PCIe functions method,
7832 	 * because they are behind a parent switch, so we hardcode these with
7833 	 * the correct number of functions.
7834 	 */
7835 	if (ixgbe_pcie_from_parent(&adapter->hw)) {
7836 		physfns = 4;
7837 	} else {
7838 		list_for_each(entry, &adapter->pdev->bus_list) {
7839 			struct pci_dev *pdev =
7840 				list_entry(entry, struct pci_dev, bus_list);
7841 			/* don't count virtual functions */
7842 			if (!pdev->is_virtfn)
7843 				physfns++;
7844 		}
7845 	}
7846 
7847 	return physfns;
7848 }
7849 
7850 /**
7851  * ixgbe_wol_supported - Check whether device supports WoL
7852  * @hw: hw specific details
7853  * @device_id: the device ID
7854  * @subdev_id: the subsystem device ID
7855  *
7856  * This function is used by probe and ethtool to determine
7857  * which devices have WoL support
7858  *
7859  **/
7860 int ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
7861 			u16 subdevice_id)
7862 {
7863 	struct ixgbe_hw *hw = &adapter->hw;
7864 	u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
7865 	int is_wol_supported = 0;
7866 
7867 	switch (device_id) {
7868 	case IXGBE_DEV_ID_82599_SFP:
7869 		/* Only these subdevices could supports WOL */
7870 		switch (subdevice_id) {
7871 		case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
7872 		case IXGBE_SUBDEV_ID_82599_560FLR:
7873 			/* only support first port */
7874 			if (hw->bus.func != 0)
7875 				break;
7876 		case IXGBE_SUBDEV_ID_82599_SP_560FLR:
7877 		case IXGBE_SUBDEV_ID_82599_SFP:
7878 		case IXGBE_SUBDEV_ID_82599_RNDC:
7879 		case IXGBE_SUBDEV_ID_82599_ECNA_DP:
7880 		case IXGBE_SUBDEV_ID_82599_LOM_SFP:
7881 			is_wol_supported = 1;
7882 			break;
7883 		}
7884 		break;
7885 	case IXGBE_DEV_ID_82599EN_SFP:
7886 		/* Only this subdevice supports WOL */
7887 		switch (subdevice_id) {
7888 		case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
7889 			is_wol_supported = 1;
7890 			break;
7891 		}
7892 		break;
7893 	case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7894 		/* All except this subdevice support WOL */
7895 		if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7896 			is_wol_supported = 1;
7897 		break;
7898 	case IXGBE_DEV_ID_82599_KX4:
7899 		is_wol_supported = 1;
7900 		break;
7901 	case IXGBE_DEV_ID_X540T:
7902 	case IXGBE_DEV_ID_X540T1:
7903 		/* check eeprom to see if enabled wol */
7904 		if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
7905 		    ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
7906 		     (hw->bus.func == 0))) {
7907 			is_wol_supported = 1;
7908 		}
7909 		break;
7910 	}
7911 
7912 	return is_wol_supported;
7913 }
7914 
7915 /**
7916  * ixgbe_probe - Device Initialization Routine
7917  * @pdev: PCI device information struct
7918  * @ent: entry in ixgbe_pci_tbl
7919  *
7920  * Returns 0 on success, negative on failure
7921  *
7922  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7923  * The OS initialization, configuring of the adapter private structure,
7924  * and a hardware reset occur.
7925  **/
7926 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
7927 {
7928 	struct net_device *netdev;
7929 	struct ixgbe_adapter *adapter = NULL;
7930 	struct ixgbe_hw *hw;
7931 	const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7932 	static int cards_found;
7933 	int i, err, pci_using_dac, expected_gts;
7934 	unsigned int indices = MAX_TX_QUEUES;
7935 	u8 part_str[IXGBE_PBANUM_LENGTH];
7936 #ifdef IXGBE_FCOE
7937 	u16 device_caps;
7938 #endif
7939 	u32 eec;
7940 
7941 	/* Catch broken hardware that put the wrong VF device ID in
7942 	 * the PCIe SR-IOV capability.
7943 	 */
7944 	if (pdev->is_virtfn) {
7945 		WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7946 		     pci_name(pdev), pdev->vendor, pdev->device);
7947 		return -EINVAL;
7948 	}
7949 
7950 	err = pci_enable_device_mem(pdev);
7951 	if (err)
7952 		return err;
7953 
7954 	if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
7955 		pci_using_dac = 1;
7956 	} else {
7957 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
7958 		if (err) {
7959 			dev_err(&pdev->dev,
7960 				"No usable DMA configuration, aborting\n");
7961 			goto err_dma;
7962 		}
7963 		pci_using_dac = 0;
7964 	}
7965 
7966 	err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7967 					   IORESOURCE_MEM), ixgbe_driver_name);
7968 	if (err) {
7969 		dev_err(&pdev->dev,
7970 			"pci_request_selected_regions failed 0x%x\n", err);
7971 		goto err_pci_reg;
7972 	}
7973 
7974 	pci_enable_pcie_error_reporting(pdev);
7975 
7976 	pci_set_master(pdev);
7977 	pci_save_state(pdev);
7978 
7979 	if (ii->mac == ixgbe_mac_82598EB) {
7980 #ifdef CONFIG_IXGBE_DCB
7981 		/* 8 TC w/ 4 queues per TC */
7982 		indices = 4 * MAX_TRAFFIC_CLASS;
7983 #else
7984 		indices = IXGBE_MAX_RSS_INDICES;
7985 #endif
7986 	}
7987 
7988 	netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7989 	if (!netdev) {
7990 		err = -ENOMEM;
7991 		goto err_alloc_etherdev;
7992 	}
7993 
7994 	SET_NETDEV_DEV(netdev, &pdev->dev);
7995 
7996 	adapter = netdev_priv(netdev);
7997 	pci_set_drvdata(pdev, adapter);
7998 
7999 	adapter->netdev = netdev;
8000 	adapter->pdev = pdev;
8001 	hw = &adapter->hw;
8002 	hw->back = adapter;
8003 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
8004 
8005 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
8006 			      pci_resource_len(pdev, 0));
8007 	adapter->io_addr = hw->hw_addr;
8008 	if (!hw->hw_addr) {
8009 		err = -EIO;
8010 		goto err_ioremap;
8011 	}
8012 
8013 	netdev->netdev_ops = &ixgbe_netdev_ops;
8014 	ixgbe_set_ethtool_ops(netdev);
8015 	netdev->watchdog_timeo = 5 * HZ;
8016 	strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
8017 
8018 	adapter->bd_number = cards_found;
8019 
8020 	/* Setup hw api */
8021 	memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
8022 	hw->mac.type  = ii->mac;
8023 
8024 	/* EEPROM */
8025 	memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
8026 	eec = IXGBE_READ_REG(hw, IXGBE_EEC);
8027 	if (ixgbe_removed(hw->hw_addr)) {
8028 		err = -EIO;
8029 		goto err_ioremap;
8030 	}
8031 	/* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
8032 	if (!(eec & (1 << 8)))
8033 		hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
8034 
8035 	/* PHY */
8036 	memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
8037 	hw->phy.sfp_type = ixgbe_sfp_type_unknown;
8038 	/* ixgbe_identify_phy_generic will set prtad and mmds properly */
8039 	hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
8040 	hw->phy.mdio.mmds = 0;
8041 	hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
8042 	hw->phy.mdio.dev = netdev;
8043 	hw->phy.mdio.mdio_read = ixgbe_mdio_read;
8044 	hw->phy.mdio.mdio_write = ixgbe_mdio_write;
8045 
8046 	ii->get_invariants(hw);
8047 
8048 	/* setup the private structure */
8049 	err = ixgbe_sw_init(adapter);
8050 	if (err)
8051 		goto err_sw_init;
8052 
8053 	/* Make it possible the adapter to be woken up via WOL */
8054 	switch (adapter->hw.mac.type) {
8055 	case ixgbe_mac_82599EB:
8056 	case ixgbe_mac_X540:
8057 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
8058 		break;
8059 	default:
8060 		break;
8061 	}
8062 
8063 	/*
8064 	 * If there is a fan on this device and it has failed log the
8065 	 * failure.
8066 	 */
8067 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
8068 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
8069 		if (esdp & IXGBE_ESDP_SDP1)
8070 			e_crit(probe, "Fan has stopped, replace the adapter\n");
8071 	}
8072 
8073 	if (allow_unsupported_sfp)
8074 		hw->allow_unsupported_sfp = allow_unsupported_sfp;
8075 
8076 	/* reset_hw fills in the perm_addr as well */
8077 	hw->phy.reset_if_overtemp = true;
8078 	err = hw->mac.ops.reset_hw(hw);
8079 	hw->phy.reset_if_overtemp = false;
8080 	if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
8081 	    hw->mac.type == ixgbe_mac_82598EB) {
8082 		err = 0;
8083 	} else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
8084 		e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
8085 		e_dev_err("Reload the driver after installing a supported module.\n");
8086 		goto err_sw_init;
8087 	} else if (err) {
8088 		e_dev_err("HW Init failed: %d\n", err);
8089 		goto err_sw_init;
8090 	}
8091 
8092 #ifdef CONFIG_PCI_IOV
8093 	/* SR-IOV not supported on the 82598 */
8094 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
8095 		goto skip_sriov;
8096 	/* Mailbox */
8097 	ixgbe_init_mbx_params_pf(hw);
8098 	memcpy(&hw->mbx.ops, ii->mbx_ops, sizeof(hw->mbx.ops));
8099 	pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
8100 	ixgbe_enable_sriov(adapter);
8101 skip_sriov:
8102 
8103 #endif
8104 	netdev->features = NETIF_F_SG |
8105 			   NETIF_F_IP_CSUM |
8106 			   NETIF_F_IPV6_CSUM |
8107 			   NETIF_F_HW_VLAN_CTAG_TX |
8108 			   NETIF_F_HW_VLAN_CTAG_RX |
8109 			   NETIF_F_HW_VLAN_CTAG_FILTER |
8110 			   NETIF_F_TSO |
8111 			   NETIF_F_TSO6 |
8112 			   NETIF_F_RXHASH |
8113 			   NETIF_F_RXCSUM;
8114 
8115 	netdev->hw_features = netdev->features | NETIF_F_HW_L2FW_DOFFLOAD;
8116 
8117 	switch (adapter->hw.mac.type) {
8118 	case ixgbe_mac_82599EB:
8119 	case ixgbe_mac_X540:
8120 		netdev->features |= NETIF_F_SCTP_CSUM;
8121 		netdev->hw_features |= NETIF_F_SCTP_CSUM |
8122 				       NETIF_F_NTUPLE;
8123 		break;
8124 	default:
8125 		break;
8126 	}
8127 
8128 	netdev->hw_features |= NETIF_F_RXALL;
8129 
8130 	netdev->vlan_features |= NETIF_F_TSO;
8131 	netdev->vlan_features |= NETIF_F_TSO6;
8132 	netdev->vlan_features |= NETIF_F_IP_CSUM;
8133 	netdev->vlan_features |= NETIF_F_IPV6_CSUM;
8134 	netdev->vlan_features |= NETIF_F_SG;
8135 
8136 	netdev->priv_flags |= IFF_UNICAST_FLT;
8137 	netdev->priv_flags |= IFF_SUPP_NOFCS;
8138 
8139 #ifdef CONFIG_IXGBE_DCB
8140 	netdev->dcbnl_ops = &dcbnl_ops;
8141 #endif
8142 
8143 #ifdef IXGBE_FCOE
8144 	if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
8145 		unsigned int fcoe_l;
8146 
8147 		if (hw->mac.ops.get_device_caps) {
8148 			hw->mac.ops.get_device_caps(hw, &device_caps);
8149 			if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
8150 				adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
8151 		}
8152 
8153 
8154 		fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
8155 		adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
8156 
8157 		netdev->features |= NETIF_F_FSO |
8158 				    NETIF_F_FCOE_CRC;
8159 
8160 		netdev->vlan_features |= NETIF_F_FSO |
8161 					 NETIF_F_FCOE_CRC |
8162 					 NETIF_F_FCOE_MTU;
8163 	}
8164 #endif /* IXGBE_FCOE */
8165 	if (pci_using_dac) {
8166 		netdev->features |= NETIF_F_HIGHDMA;
8167 		netdev->vlan_features |= NETIF_F_HIGHDMA;
8168 	}
8169 
8170 	if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
8171 		netdev->hw_features |= NETIF_F_LRO;
8172 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
8173 		netdev->features |= NETIF_F_LRO;
8174 
8175 	/* make sure the EEPROM is good */
8176 	if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
8177 		e_dev_err("The EEPROM Checksum Is Not Valid\n");
8178 		err = -EIO;
8179 		goto err_sw_init;
8180 	}
8181 
8182 	memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
8183 
8184 	if (!is_valid_ether_addr(netdev->dev_addr)) {
8185 		e_dev_err("invalid MAC address\n");
8186 		err = -EIO;
8187 		goto err_sw_init;
8188 	}
8189 
8190 	setup_timer(&adapter->service_timer, &ixgbe_service_timer,
8191 		    (unsigned long) adapter);
8192 
8193 	if (ixgbe_removed(hw->hw_addr)) {
8194 		err = -EIO;
8195 		goto err_sw_init;
8196 	}
8197 	INIT_WORK(&adapter->service_task, ixgbe_service_task);
8198 	set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
8199 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
8200 
8201 	err = ixgbe_init_interrupt_scheme(adapter);
8202 	if (err)
8203 		goto err_sw_init;
8204 
8205 	/* WOL not supported for all devices */
8206 	adapter->wol = 0;
8207 	hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
8208 	hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
8209 						pdev->subsystem_device);
8210 	if (hw->wol_enabled)
8211 		adapter->wol = IXGBE_WUFC_MAG;
8212 
8213 	device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
8214 
8215 	/* save off EEPROM version number */
8216 	hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
8217 	hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
8218 
8219 	/* pick up the PCI bus settings for reporting later */
8220 	hw->mac.ops.get_bus_info(hw);
8221 	if (ixgbe_pcie_from_parent(hw))
8222 		ixgbe_get_parent_bus_info(adapter);
8223 
8224 	/* calculate the expected PCIe bandwidth required for optimal
8225 	 * performance. Note that some older parts will never have enough
8226 	 * bandwidth due to being older generation PCIe parts. We clamp these
8227 	 * parts to ensure no warning is displayed if it can't be fixed.
8228 	 */
8229 	switch (hw->mac.type) {
8230 	case ixgbe_mac_82598EB:
8231 		expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
8232 		break;
8233 	default:
8234 		expected_gts = ixgbe_enumerate_functions(adapter) * 10;
8235 		break;
8236 	}
8237 	ixgbe_check_minimum_link(adapter, expected_gts);
8238 
8239 	err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
8240 	if (err)
8241 		strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
8242 	if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
8243 		e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
8244 			   hw->mac.type, hw->phy.type, hw->phy.sfp_type,
8245 		           part_str);
8246 	else
8247 		e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
8248 			   hw->mac.type, hw->phy.type, part_str);
8249 
8250 	e_dev_info("%pM\n", netdev->dev_addr);
8251 
8252 	/* reset the hardware with the new settings */
8253 	err = hw->mac.ops.start_hw(hw);
8254 	if (err == IXGBE_ERR_EEPROM_VERSION) {
8255 		/* We are running on a pre-production device, log a warning */
8256 		e_dev_warn("This device is a pre-production adapter/LOM. "
8257 			   "Please be aware there may be issues associated "
8258 			   "with your hardware.  If you are experiencing "
8259 			   "problems please contact your Intel or hardware "
8260 			   "representative who provided you with this "
8261 			   "hardware.\n");
8262 	}
8263 	strcpy(netdev->name, "eth%d");
8264 	err = register_netdev(netdev);
8265 	if (err)
8266 		goto err_register;
8267 
8268 	/* power down the optics for 82599 SFP+ fiber */
8269 	if (hw->mac.ops.disable_tx_laser)
8270 		hw->mac.ops.disable_tx_laser(hw);
8271 
8272 	/* carrier off reporting is important to ethtool even BEFORE open */
8273 	netif_carrier_off(netdev);
8274 
8275 #ifdef CONFIG_IXGBE_DCA
8276 	if (dca_add_requester(&pdev->dev) == 0) {
8277 		adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
8278 		ixgbe_setup_dca(adapter);
8279 	}
8280 #endif
8281 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
8282 		e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
8283 		for (i = 0; i < adapter->num_vfs; i++)
8284 			ixgbe_vf_configuration(pdev, (i | 0x10000000));
8285 	}
8286 
8287 	/* firmware requires driver version to be 0xFFFFFFFF
8288 	 * since os does not support feature
8289 	 */
8290 	if (hw->mac.ops.set_fw_drv_ver)
8291 		hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
8292 					   0xFF);
8293 
8294 	/* add san mac addr to netdev */
8295 	ixgbe_add_sanmac_netdev(netdev);
8296 
8297 	e_dev_info("%s\n", ixgbe_default_device_descr);
8298 	cards_found++;
8299 
8300 #ifdef CONFIG_IXGBE_HWMON
8301 	if (ixgbe_sysfs_init(adapter))
8302 		e_err(probe, "failed to allocate sysfs resources\n");
8303 #endif /* CONFIG_IXGBE_HWMON */
8304 
8305 	ixgbe_dbg_adapter_init(adapter);
8306 
8307 	/* Need link setup for MNG FW, else wait for IXGBE_UP */
8308 	if (ixgbe_mng_enabled(hw) && hw->mac.ops.setup_link)
8309 		hw->mac.ops.setup_link(hw,
8310 			IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
8311 			true);
8312 
8313 	return 0;
8314 
8315 err_register:
8316 	ixgbe_release_hw_control(adapter);
8317 	ixgbe_clear_interrupt_scheme(adapter);
8318 err_sw_init:
8319 	ixgbe_disable_sriov(adapter);
8320 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
8321 	iounmap(adapter->io_addr);
8322 err_ioremap:
8323 	free_netdev(netdev);
8324 err_alloc_etherdev:
8325 	pci_release_selected_regions(pdev,
8326 				     pci_select_bars(pdev, IORESOURCE_MEM));
8327 err_pci_reg:
8328 err_dma:
8329 	if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
8330 		pci_disable_device(pdev);
8331 	return err;
8332 }
8333 
8334 /**
8335  * ixgbe_remove - Device Removal Routine
8336  * @pdev: PCI device information struct
8337  *
8338  * ixgbe_remove is called by the PCI subsystem to alert the driver
8339  * that it should release a PCI device.  The could be caused by a
8340  * Hot-Plug event, or because the driver is going to be removed from
8341  * memory.
8342  **/
8343 static void ixgbe_remove(struct pci_dev *pdev)
8344 {
8345 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8346 	struct net_device *netdev = adapter->netdev;
8347 
8348 	ixgbe_dbg_adapter_exit(adapter);
8349 
8350 	set_bit(__IXGBE_REMOVING, &adapter->state);
8351 	cancel_work_sync(&adapter->service_task);
8352 
8353 
8354 #ifdef CONFIG_IXGBE_DCA
8355 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
8356 		adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
8357 		dca_remove_requester(&pdev->dev);
8358 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
8359 	}
8360 
8361 #endif
8362 #ifdef CONFIG_IXGBE_HWMON
8363 	ixgbe_sysfs_exit(adapter);
8364 #endif /* CONFIG_IXGBE_HWMON */
8365 
8366 	/* remove the added san mac */
8367 	ixgbe_del_sanmac_netdev(netdev);
8368 
8369 	if (netdev->reg_state == NETREG_REGISTERED)
8370 		unregister_netdev(netdev);
8371 
8372 #ifdef CONFIG_PCI_IOV
8373 	/*
8374 	 * Only disable SR-IOV on unload if the user specified the now
8375 	 * deprecated max_vfs module parameter.
8376 	 */
8377 	if (max_vfs)
8378 		ixgbe_disable_sriov(adapter);
8379 #endif
8380 	ixgbe_clear_interrupt_scheme(adapter);
8381 
8382 	ixgbe_release_hw_control(adapter);
8383 
8384 #ifdef CONFIG_DCB
8385 	kfree(adapter->ixgbe_ieee_pfc);
8386 	kfree(adapter->ixgbe_ieee_ets);
8387 
8388 #endif
8389 	iounmap(adapter->io_addr);
8390 	pci_release_selected_regions(pdev, pci_select_bars(pdev,
8391 				     IORESOURCE_MEM));
8392 
8393 	e_dev_info("complete\n");
8394 
8395 	free_netdev(netdev);
8396 
8397 	pci_disable_pcie_error_reporting(pdev);
8398 
8399 	if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
8400 		pci_disable_device(pdev);
8401 }
8402 
8403 /**
8404  * ixgbe_io_error_detected - called when PCI error is detected
8405  * @pdev: Pointer to PCI device
8406  * @state: The current pci connection state
8407  *
8408  * This function is called after a PCI bus error affecting
8409  * this device has been detected.
8410  */
8411 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
8412 						pci_channel_state_t state)
8413 {
8414 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8415 	struct net_device *netdev = adapter->netdev;
8416 
8417 #ifdef CONFIG_PCI_IOV
8418 	struct ixgbe_hw *hw = &adapter->hw;
8419 	struct pci_dev *bdev, *vfdev;
8420 	u32 dw0, dw1, dw2, dw3;
8421 	int vf, pos;
8422 	u16 req_id, pf_func;
8423 
8424 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
8425 	    adapter->num_vfs == 0)
8426 		goto skip_bad_vf_detection;
8427 
8428 	bdev = pdev->bus->self;
8429 	while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
8430 		bdev = bdev->bus->self;
8431 
8432 	if (!bdev)
8433 		goto skip_bad_vf_detection;
8434 
8435 	pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
8436 	if (!pos)
8437 		goto skip_bad_vf_detection;
8438 
8439 	dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
8440 	dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
8441 	dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
8442 	dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
8443 	if (ixgbe_removed(hw->hw_addr))
8444 		goto skip_bad_vf_detection;
8445 
8446 	req_id = dw1 >> 16;
8447 	/* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
8448 	if (!(req_id & 0x0080))
8449 		goto skip_bad_vf_detection;
8450 
8451 	pf_func = req_id & 0x01;
8452 	if ((pf_func & 1) == (pdev->devfn & 1)) {
8453 		unsigned int device_id;
8454 
8455 		vf = (req_id & 0x7F) >> 1;
8456 		e_dev_err("VF %d has caused a PCIe error\n", vf);
8457 		e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
8458 				"%8.8x\tdw3: %8.8x\n",
8459 		dw0, dw1, dw2, dw3);
8460 		switch (adapter->hw.mac.type) {
8461 		case ixgbe_mac_82599EB:
8462 			device_id = IXGBE_82599_VF_DEVICE_ID;
8463 			break;
8464 		case ixgbe_mac_X540:
8465 			device_id = IXGBE_X540_VF_DEVICE_ID;
8466 			break;
8467 		default:
8468 			device_id = 0;
8469 			break;
8470 		}
8471 
8472 		/* Find the pci device of the offending VF */
8473 		vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
8474 		while (vfdev) {
8475 			if (vfdev->devfn == (req_id & 0xFF))
8476 				break;
8477 			vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
8478 					       device_id, vfdev);
8479 		}
8480 		/*
8481 		 * There's a slim chance the VF could have been hot plugged,
8482 		 * so if it is no longer present we don't need to issue the
8483 		 * VFLR.  Just clean up the AER in that case.
8484 		 */
8485 		if (vfdev) {
8486 			e_dev_err("Issuing VFLR to VF %d\n", vf);
8487 			pci_write_config_dword(vfdev, 0xA8, 0x00008000);
8488 			/* Free device reference count */
8489 			pci_dev_put(vfdev);
8490 		}
8491 
8492 		pci_cleanup_aer_uncorrect_error_status(pdev);
8493 	}
8494 
8495 	/*
8496 	 * Even though the error may have occurred on the other port
8497 	 * we still need to increment the vf error reference count for
8498 	 * both ports because the I/O resume function will be called
8499 	 * for both of them.
8500 	 */
8501 	adapter->vferr_refcount++;
8502 
8503 	return PCI_ERS_RESULT_RECOVERED;
8504 
8505 skip_bad_vf_detection:
8506 #endif /* CONFIG_PCI_IOV */
8507 	if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
8508 		return PCI_ERS_RESULT_DISCONNECT;
8509 
8510 	rtnl_lock();
8511 	netif_device_detach(netdev);
8512 
8513 	if (state == pci_channel_io_perm_failure) {
8514 		rtnl_unlock();
8515 		return PCI_ERS_RESULT_DISCONNECT;
8516 	}
8517 
8518 	if (netif_running(netdev))
8519 		ixgbe_down(adapter);
8520 
8521 	if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
8522 		pci_disable_device(pdev);
8523 	rtnl_unlock();
8524 
8525 	/* Request a slot reset. */
8526 	return PCI_ERS_RESULT_NEED_RESET;
8527 }
8528 
8529 /**
8530  * ixgbe_io_slot_reset - called after the pci bus has been reset.
8531  * @pdev: Pointer to PCI device
8532  *
8533  * Restart the card from scratch, as if from a cold-boot.
8534  */
8535 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
8536 {
8537 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8538 	pci_ers_result_t result;
8539 	int err;
8540 
8541 	if (pci_enable_device_mem(pdev)) {
8542 		e_err(probe, "Cannot re-enable PCI device after reset.\n");
8543 		result = PCI_ERS_RESULT_DISCONNECT;
8544 	} else {
8545 		smp_mb__before_atomic();
8546 		clear_bit(__IXGBE_DISABLED, &adapter->state);
8547 		adapter->hw.hw_addr = adapter->io_addr;
8548 		pci_set_master(pdev);
8549 		pci_restore_state(pdev);
8550 		pci_save_state(pdev);
8551 
8552 		pci_wake_from_d3(pdev, false);
8553 
8554 		ixgbe_reset(adapter);
8555 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
8556 		result = PCI_ERS_RESULT_RECOVERED;
8557 	}
8558 
8559 	err = pci_cleanup_aer_uncorrect_error_status(pdev);
8560 	if (err) {
8561 		e_dev_err("pci_cleanup_aer_uncorrect_error_status "
8562 			  "failed 0x%0x\n", err);
8563 		/* non-fatal, continue */
8564 	}
8565 
8566 	return result;
8567 }
8568 
8569 /**
8570  * ixgbe_io_resume - called when traffic can start flowing again.
8571  * @pdev: Pointer to PCI device
8572  *
8573  * This callback is called when the error recovery driver tells us that
8574  * its OK to resume normal operation.
8575  */
8576 static void ixgbe_io_resume(struct pci_dev *pdev)
8577 {
8578 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8579 	struct net_device *netdev = adapter->netdev;
8580 
8581 #ifdef CONFIG_PCI_IOV
8582 	if (adapter->vferr_refcount) {
8583 		e_info(drv, "Resuming after VF err\n");
8584 		adapter->vferr_refcount--;
8585 		return;
8586 	}
8587 
8588 #endif
8589 	if (netif_running(netdev))
8590 		ixgbe_up(adapter);
8591 
8592 	netif_device_attach(netdev);
8593 }
8594 
8595 static const struct pci_error_handlers ixgbe_err_handler = {
8596 	.error_detected = ixgbe_io_error_detected,
8597 	.slot_reset = ixgbe_io_slot_reset,
8598 	.resume = ixgbe_io_resume,
8599 };
8600 
8601 static struct pci_driver ixgbe_driver = {
8602 	.name     = ixgbe_driver_name,
8603 	.id_table = ixgbe_pci_tbl,
8604 	.probe    = ixgbe_probe,
8605 	.remove   = ixgbe_remove,
8606 #ifdef CONFIG_PM
8607 	.suspend  = ixgbe_suspend,
8608 	.resume   = ixgbe_resume,
8609 #endif
8610 	.shutdown = ixgbe_shutdown,
8611 	.sriov_configure = ixgbe_pci_sriov_configure,
8612 	.err_handler = &ixgbe_err_handler
8613 };
8614 
8615 /**
8616  * ixgbe_init_module - Driver Registration Routine
8617  *
8618  * ixgbe_init_module is the first routine called when the driver is
8619  * loaded. All it does is register with the PCI subsystem.
8620  **/
8621 static int __init ixgbe_init_module(void)
8622 {
8623 	int ret;
8624 	pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
8625 	pr_info("%s\n", ixgbe_copyright);
8626 
8627 	ixgbe_dbg_init();
8628 
8629 	ret = pci_register_driver(&ixgbe_driver);
8630 	if (ret) {
8631 		ixgbe_dbg_exit();
8632 		return ret;
8633 	}
8634 
8635 #ifdef CONFIG_IXGBE_DCA
8636 	dca_register_notify(&dca_notifier);
8637 #endif
8638 
8639 	return 0;
8640 }
8641 
8642 module_init(ixgbe_init_module);
8643 
8644 /**
8645  * ixgbe_exit_module - Driver Exit Cleanup Routine
8646  *
8647  * ixgbe_exit_module is called just before the driver is removed
8648  * from memory.
8649  **/
8650 static void __exit ixgbe_exit_module(void)
8651 {
8652 #ifdef CONFIG_IXGBE_DCA
8653 	dca_unregister_notify(&dca_notifier);
8654 #endif
8655 	pci_unregister_driver(&ixgbe_driver);
8656 
8657 	ixgbe_dbg_exit();
8658 
8659 	rcu_barrier(); /* Wait for completion of call_rcu()'s */
8660 }
8661 
8662 #ifdef CONFIG_IXGBE_DCA
8663 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
8664 			    void *p)
8665 {
8666 	int ret_val;
8667 
8668 	ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
8669 					 __ixgbe_notify_dca);
8670 
8671 	return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
8672 }
8673 
8674 #endif /* CONFIG_IXGBE_DCA */
8675 
8676 module_exit(ixgbe_exit_module);
8677 
8678 /* ixgbe_main.c */
8679