xref: /linux/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c (revision 957e3facd147510f2cf8780e38606f1d707f0e33)
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/etherdevice.h>
46 #include <linux/ethtool.h>
47 #include <linux/if.h>
48 #include <linux/if_vlan.h>
49 #include <linux/if_macvlan.h>
50 #include <linux/if_bridge.h>
51 #include <linux/prefetch.h>
52 #include <scsi/fc/fc_fcoe.h>
53 
54 #ifdef CONFIG_OF
55 #include <linux/of_net.h>
56 #endif
57 
58 #ifdef CONFIG_SPARC
59 #include <asm/idprom.h>
60 #include <asm/prom.h>
61 #endif
62 
63 #include "ixgbe.h"
64 #include "ixgbe_common.h"
65 #include "ixgbe_dcb_82599.h"
66 #include "ixgbe_sriov.h"
67 
68 char ixgbe_driver_name[] = "ixgbe";
69 static const char ixgbe_driver_string[] =
70 			      "Intel(R) 10 Gigabit PCI Express Network Driver";
71 #ifdef IXGBE_FCOE
72 char ixgbe_default_device_descr[] =
73 			      "Intel(R) 10 Gigabit Network Connection";
74 #else
75 static char ixgbe_default_device_descr[] =
76 			      "Intel(R) 10 Gigabit Network Connection";
77 #endif
78 #define DRV_VERSION "4.0.1-k"
79 const char ixgbe_driver_version[] = DRV_VERSION;
80 static const char ixgbe_copyright[] =
81 				"Copyright (c) 1999-2014 Intel Corporation.";
82 
83 static const struct ixgbe_info *ixgbe_info_tbl[] = {
84 	[board_82598]		= &ixgbe_82598_info,
85 	[board_82599]		= &ixgbe_82599_info,
86 	[board_X540]		= &ixgbe_X540_info,
87 	[board_X550]		= &ixgbe_X550_info,
88 	[board_X550EM_x]	= &ixgbe_X550EM_x_info,
89 };
90 
91 /* ixgbe_pci_tbl - PCI Device ID Table
92  *
93  * Wildcard entries (PCI_ANY_ID) should come last
94  * Last entry must be all 0s
95  *
96  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
97  *   Class, Class Mask, private data (not used) }
98  */
99 static const struct pci_device_id ixgbe_pci_tbl[] = {
100 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
101 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
102 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
103 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
104 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
105 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
106 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
107 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
108 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
109 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
110 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
111 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
112 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
113 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
114 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
115 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
116 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
117 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
118 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
119 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
120 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
121 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
122 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
123 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
124 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
125 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
126 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_QSFP_SF_QP), board_82599 },
127 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
128 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
129 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
130 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T), board_X550},
131 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KX4), board_X550EM_x},
132 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KR), board_X550EM_x},
133 	/* required last entry */
134 	{0, }
135 };
136 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
137 
138 #ifdef CONFIG_IXGBE_DCA
139 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
140 			    void *p);
141 static struct notifier_block dca_notifier = {
142 	.notifier_call = ixgbe_notify_dca,
143 	.next          = NULL,
144 	.priority      = 0
145 };
146 #endif
147 
148 #ifdef CONFIG_PCI_IOV
149 static unsigned int max_vfs;
150 module_param(max_vfs, uint, 0);
151 MODULE_PARM_DESC(max_vfs,
152 		 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63. (Deprecated)");
153 #endif /* CONFIG_PCI_IOV */
154 
155 static unsigned int allow_unsupported_sfp;
156 module_param(allow_unsupported_sfp, uint, 0);
157 MODULE_PARM_DESC(allow_unsupported_sfp,
158 		 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
159 
160 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
161 static int debug = -1;
162 module_param(debug, int, 0);
163 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
164 
165 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
166 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
167 MODULE_LICENSE("GPL");
168 MODULE_VERSION(DRV_VERSION);
169 
170 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev);
171 
172 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
173 					  u32 reg, u16 *value)
174 {
175 	struct pci_dev *parent_dev;
176 	struct pci_bus *parent_bus;
177 
178 	parent_bus = adapter->pdev->bus->parent;
179 	if (!parent_bus)
180 		return -1;
181 
182 	parent_dev = parent_bus->self;
183 	if (!parent_dev)
184 		return -1;
185 
186 	if (!pci_is_pcie(parent_dev))
187 		return -1;
188 
189 	pcie_capability_read_word(parent_dev, reg, value);
190 	if (*value == IXGBE_FAILED_READ_CFG_WORD &&
191 	    ixgbe_check_cfg_remove(&adapter->hw, parent_dev))
192 		return -1;
193 	return 0;
194 }
195 
196 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
197 {
198 	struct ixgbe_hw *hw = &adapter->hw;
199 	u16 link_status = 0;
200 	int err;
201 
202 	hw->bus.type = ixgbe_bus_type_pci_express;
203 
204 	/* Get the negotiated link width and speed from PCI config space of the
205 	 * parent, as this device is behind a switch
206 	 */
207 	err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
208 
209 	/* assume caller will handle error case */
210 	if (err)
211 		return err;
212 
213 	hw->bus.width = ixgbe_convert_bus_width(link_status);
214 	hw->bus.speed = ixgbe_convert_bus_speed(link_status);
215 
216 	return 0;
217 }
218 
219 /**
220  * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
221  * @hw: hw specific details
222  *
223  * This function is used by probe to determine whether a device's PCI-Express
224  * bandwidth details should be gathered from the parent bus instead of from the
225  * device. Used to ensure that various locations all have the correct device ID
226  * checks.
227  */
228 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
229 {
230 	switch (hw->device_id) {
231 	case IXGBE_DEV_ID_82599_SFP_SF_QP:
232 	case IXGBE_DEV_ID_82599_QSFP_SF_QP:
233 		return true;
234 	default:
235 		return false;
236 	}
237 }
238 
239 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
240 				     int expected_gts)
241 {
242 	int max_gts = 0;
243 	enum pci_bus_speed speed = PCI_SPEED_UNKNOWN;
244 	enum pcie_link_width width = PCIE_LNK_WIDTH_UNKNOWN;
245 	struct pci_dev *pdev;
246 
247 	/* determine whether to use the the parent device
248 	 */
249 	if (ixgbe_pcie_from_parent(&adapter->hw))
250 		pdev = adapter->pdev->bus->parent->self;
251 	else
252 		pdev = adapter->pdev;
253 
254 	if (pcie_get_minimum_link(pdev, &speed, &width) ||
255 	    speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN) {
256 		e_dev_warn("Unable to determine PCI Express bandwidth.\n");
257 		return;
258 	}
259 
260 	switch (speed) {
261 	case PCIE_SPEED_2_5GT:
262 		/* 8b/10b encoding reduces max throughput by 20% */
263 		max_gts = 2 * width;
264 		break;
265 	case PCIE_SPEED_5_0GT:
266 		/* 8b/10b encoding reduces max throughput by 20% */
267 		max_gts = 4 * width;
268 		break;
269 	case PCIE_SPEED_8_0GT:
270 		/* 128b/130b encoding reduces throughput by less than 2% */
271 		max_gts = 8 * width;
272 		break;
273 	default:
274 		e_dev_warn("Unable to determine PCI Express bandwidth.\n");
275 		return;
276 	}
277 
278 	e_dev_info("PCI Express bandwidth of %dGT/s available\n",
279 		   max_gts);
280 	e_dev_info("(Speed:%s, Width: x%d, Encoding Loss:%s)\n",
281 		   (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" :
282 		    speed == PCIE_SPEED_5_0GT ? "5.0GT/s" :
283 		    speed == PCIE_SPEED_2_5GT ? "2.5GT/s" :
284 		    "Unknown"),
285 		   width,
286 		   (speed == PCIE_SPEED_2_5GT ? "20%" :
287 		    speed == PCIE_SPEED_5_0GT ? "20%" :
288 		    speed == PCIE_SPEED_8_0GT ? "<2%" :
289 		    "Unknown"));
290 
291 	if (max_gts < expected_gts) {
292 		e_dev_warn("This is not sufficient for optimal performance of this card.\n");
293 		e_dev_warn("For optimal performance, at least %dGT/s of bandwidth is required.\n",
294 			expected_gts);
295 		e_dev_warn("A slot with more lanes and/or higher speed is suggested.\n");
296 	}
297 }
298 
299 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
300 {
301 	if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
302 	    !test_bit(__IXGBE_REMOVING, &adapter->state) &&
303 	    !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
304 		schedule_work(&adapter->service_task);
305 }
306 
307 static void ixgbe_remove_adapter(struct ixgbe_hw *hw)
308 {
309 	struct ixgbe_adapter *adapter = hw->back;
310 
311 	if (!hw->hw_addr)
312 		return;
313 	hw->hw_addr = NULL;
314 	e_dev_err("Adapter removed\n");
315 	if (test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
316 		ixgbe_service_event_schedule(adapter);
317 }
318 
319 static void ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg)
320 {
321 	u32 value;
322 
323 	/* The following check not only optimizes a bit by not
324 	 * performing a read on the status register when the
325 	 * register just read was a status register read that
326 	 * returned IXGBE_FAILED_READ_REG. It also blocks any
327 	 * potential recursion.
328 	 */
329 	if (reg == IXGBE_STATUS) {
330 		ixgbe_remove_adapter(hw);
331 		return;
332 	}
333 	value = ixgbe_read_reg(hw, IXGBE_STATUS);
334 	if (value == IXGBE_FAILED_READ_REG)
335 		ixgbe_remove_adapter(hw);
336 }
337 
338 /**
339  * ixgbe_read_reg - Read from device register
340  * @hw: hw specific details
341  * @reg: offset of register to read
342  *
343  * Returns : value read or IXGBE_FAILED_READ_REG if removed
344  *
345  * This function is used to read device registers. It checks for device
346  * removal by confirming any read that returns all ones by checking the
347  * status register value for all ones. This function avoids reading from
348  * the hardware if a removal was previously detected in which case it
349  * returns IXGBE_FAILED_READ_REG (all ones).
350  */
351 u32 ixgbe_read_reg(struct ixgbe_hw *hw, u32 reg)
352 {
353 	u8 __iomem *reg_addr = ACCESS_ONCE(hw->hw_addr);
354 	u32 value;
355 
356 	if (ixgbe_removed(reg_addr))
357 		return IXGBE_FAILED_READ_REG;
358 	value = readl(reg_addr + reg);
359 	if (unlikely(value == IXGBE_FAILED_READ_REG))
360 		ixgbe_check_remove(hw, reg);
361 	return value;
362 }
363 
364 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev)
365 {
366 	u16 value;
367 
368 	pci_read_config_word(pdev, PCI_VENDOR_ID, &value);
369 	if (value == IXGBE_FAILED_READ_CFG_WORD) {
370 		ixgbe_remove_adapter(hw);
371 		return true;
372 	}
373 	return false;
374 }
375 
376 u16 ixgbe_read_pci_cfg_word(struct ixgbe_hw *hw, u32 reg)
377 {
378 	struct ixgbe_adapter *adapter = hw->back;
379 	u16 value;
380 
381 	if (ixgbe_removed(hw->hw_addr))
382 		return IXGBE_FAILED_READ_CFG_WORD;
383 	pci_read_config_word(adapter->pdev, reg, &value);
384 	if (value == IXGBE_FAILED_READ_CFG_WORD &&
385 	    ixgbe_check_cfg_remove(hw, adapter->pdev))
386 		return IXGBE_FAILED_READ_CFG_WORD;
387 	return value;
388 }
389 
390 #ifdef CONFIG_PCI_IOV
391 static u32 ixgbe_read_pci_cfg_dword(struct ixgbe_hw *hw, u32 reg)
392 {
393 	struct ixgbe_adapter *adapter = hw->back;
394 	u32 value;
395 
396 	if (ixgbe_removed(hw->hw_addr))
397 		return IXGBE_FAILED_READ_CFG_DWORD;
398 	pci_read_config_dword(adapter->pdev, reg, &value);
399 	if (value == IXGBE_FAILED_READ_CFG_DWORD &&
400 	    ixgbe_check_cfg_remove(hw, adapter->pdev))
401 		return IXGBE_FAILED_READ_CFG_DWORD;
402 	return value;
403 }
404 #endif /* CONFIG_PCI_IOV */
405 
406 void ixgbe_write_pci_cfg_word(struct ixgbe_hw *hw, u32 reg, u16 value)
407 {
408 	struct ixgbe_adapter *adapter = hw->back;
409 
410 	if (ixgbe_removed(hw->hw_addr))
411 		return;
412 	pci_write_config_word(adapter->pdev, reg, value);
413 }
414 
415 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
416 {
417 	BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
418 
419 	/* flush memory to make sure state is correct before next watchdog */
420 	smp_mb__before_atomic();
421 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
422 }
423 
424 struct ixgbe_reg_info {
425 	u32 ofs;
426 	char *name;
427 };
428 
429 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
430 
431 	/* General Registers */
432 	{IXGBE_CTRL, "CTRL"},
433 	{IXGBE_STATUS, "STATUS"},
434 	{IXGBE_CTRL_EXT, "CTRL_EXT"},
435 
436 	/* Interrupt Registers */
437 	{IXGBE_EICR, "EICR"},
438 
439 	/* RX Registers */
440 	{IXGBE_SRRCTL(0), "SRRCTL"},
441 	{IXGBE_DCA_RXCTRL(0), "DRXCTL"},
442 	{IXGBE_RDLEN(0), "RDLEN"},
443 	{IXGBE_RDH(0), "RDH"},
444 	{IXGBE_RDT(0), "RDT"},
445 	{IXGBE_RXDCTL(0), "RXDCTL"},
446 	{IXGBE_RDBAL(0), "RDBAL"},
447 	{IXGBE_RDBAH(0), "RDBAH"},
448 
449 	/* TX Registers */
450 	{IXGBE_TDBAL(0), "TDBAL"},
451 	{IXGBE_TDBAH(0), "TDBAH"},
452 	{IXGBE_TDLEN(0), "TDLEN"},
453 	{IXGBE_TDH(0), "TDH"},
454 	{IXGBE_TDT(0), "TDT"},
455 	{IXGBE_TXDCTL(0), "TXDCTL"},
456 
457 	/* List Terminator */
458 	{ .name = NULL }
459 };
460 
461 
462 /*
463  * ixgbe_regdump - register printout routine
464  */
465 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
466 {
467 	int i = 0, j = 0;
468 	char rname[16];
469 	u32 regs[64];
470 
471 	switch (reginfo->ofs) {
472 	case IXGBE_SRRCTL(0):
473 		for (i = 0; i < 64; i++)
474 			regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
475 		break;
476 	case IXGBE_DCA_RXCTRL(0):
477 		for (i = 0; i < 64; i++)
478 			regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
479 		break;
480 	case IXGBE_RDLEN(0):
481 		for (i = 0; i < 64; i++)
482 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
483 		break;
484 	case IXGBE_RDH(0):
485 		for (i = 0; i < 64; i++)
486 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
487 		break;
488 	case IXGBE_RDT(0):
489 		for (i = 0; i < 64; i++)
490 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
491 		break;
492 	case IXGBE_RXDCTL(0):
493 		for (i = 0; i < 64; i++)
494 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
495 		break;
496 	case IXGBE_RDBAL(0):
497 		for (i = 0; i < 64; i++)
498 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
499 		break;
500 	case IXGBE_RDBAH(0):
501 		for (i = 0; i < 64; i++)
502 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
503 		break;
504 	case IXGBE_TDBAL(0):
505 		for (i = 0; i < 64; i++)
506 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
507 		break;
508 	case IXGBE_TDBAH(0):
509 		for (i = 0; i < 64; i++)
510 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
511 		break;
512 	case IXGBE_TDLEN(0):
513 		for (i = 0; i < 64; i++)
514 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
515 		break;
516 	case IXGBE_TDH(0):
517 		for (i = 0; i < 64; i++)
518 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
519 		break;
520 	case IXGBE_TDT(0):
521 		for (i = 0; i < 64; i++)
522 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
523 		break;
524 	case IXGBE_TXDCTL(0):
525 		for (i = 0; i < 64; i++)
526 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
527 		break;
528 	default:
529 		pr_info("%-15s %08x\n", reginfo->name,
530 			IXGBE_READ_REG(hw, reginfo->ofs));
531 		return;
532 	}
533 
534 	for (i = 0; i < 8; i++) {
535 		snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
536 		pr_err("%-15s", rname);
537 		for (j = 0; j < 8; j++)
538 			pr_cont(" %08x", regs[i*8+j]);
539 		pr_cont("\n");
540 	}
541 
542 }
543 
544 /*
545  * ixgbe_dump - Print registers, tx-rings and rx-rings
546  */
547 static void ixgbe_dump(struct ixgbe_adapter *adapter)
548 {
549 	struct net_device *netdev = adapter->netdev;
550 	struct ixgbe_hw *hw = &adapter->hw;
551 	struct ixgbe_reg_info *reginfo;
552 	int n = 0;
553 	struct ixgbe_ring *tx_ring;
554 	struct ixgbe_tx_buffer *tx_buffer;
555 	union ixgbe_adv_tx_desc *tx_desc;
556 	struct my_u0 { u64 a; u64 b; } *u0;
557 	struct ixgbe_ring *rx_ring;
558 	union ixgbe_adv_rx_desc *rx_desc;
559 	struct ixgbe_rx_buffer *rx_buffer_info;
560 	u32 staterr;
561 	int i = 0;
562 
563 	if (!netif_msg_hw(adapter))
564 		return;
565 
566 	/* Print netdevice Info */
567 	if (netdev) {
568 		dev_info(&adapter->pdev->dev, "Net device Info\n");
569 		pr_info("Device Name     state            "
570 			"trans_start      last_rx\n");
571 		pr_info("%-15s %016lX %016lX %016lX\n",
572 			netdev->name,
573 			netdev->state,
574 			netdev->trans_start,
575 			netdev->last_rx);
576 	}
577 
578 	/* Print Registers */
579 	dev_info(&adapter->pdev->dev, "Register Dump\n");
580 	pr_info(" Register Name   Value\n");
581 	for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
582 	     reginfo->name; reginfo++) {
583 		ixgbe_regdump(hw, reginfo);
584 	}
585 
586 	/* Print TX Ring Summary */
587 	if (!netdev || !netif_running(netdev))
588 		return;
589 
590 	dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
591 	pr_info(" %s     %s              %s        %s\n",
592 		"Queue [NTU] [NTC] [bi(ntc)->dma  ]",
593 		"leng", "ntw", "timestamp");
594 	for (n = 0; n < adapter->num_tx_queues; n++) {
595 		tx_ring = adapter->tx_ring[n];
596 		tx_buffer = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
597 		pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
598 			   n, tx_ring->next_to_use, tx_ring->next_to_clean,
599 			   (u64)dma_unmap_addr(tx_buffer, dma),
600 			   dma_unmap_len(tx_buffer, len),
601 			   tx_buffer->next_to_watch,
602 			   (u64)tx_buffer->time_stamp);
603 	}
604 
605 	/* Print TX Rings */
606 	if (!netif_msg_tx_done(adapter))
607 		goto rx_ring_summary;
608 
609 	dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
610 
611 	/* Transmit Descriptor Formats
612 	 *
613 	 * 82598 Advanced Transmit Descriptor
614 	 *   +--------------------------------------------------------------+
615 	 * 0 |         Buffer Address [63:0]                                |
616 	 *   +--------------------------------------------------------------+
617 	 * 8 |  PAYLEN  | POPTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
618 	 *   +--------------------------------------------------------------+
619 	 *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
620 	 *
621 	 * 82598 Advanced Transmit Descriptor (Write-Back Format)
622 	 *   +--------------------------------------------------------------+
623 	 * 0 |                          RSV [63:0]                          |
624 	 *   +--------------------------------------------------------------+
625 	 * 8 |            RSV           |  STA  |          NXTSEQ           |
626 	 *   +--------------------------------------------------------------+
627 	 *   63                       36 35   32 31                         0
628 	 *
629 	 * 82599+ Advanced Transmit Descriptor
630 	 *   +--------------------------------------------------------------+
631 	 * 0 |         Buffer Address [63:0]                                |
632 	 *   +--------------------------------------------------------------+
633 	 * 8 |PAYLEN  |POPTS|CC|IDX  |STA  |DCMD  |DTYP |MAC  |RSV  |DTALEN |
634 	 *   +--------------------------------------------------------------+
635 	 *   63     46 45 40 39 38 36 35 32 31  24 23 20 19 18 17 16 15     0
636 	 *
637 	 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
638 	 *   +--------------------------------------------------------------+
639 	 * 0 |                          RSV [63:0]                          |
640 	 *   +--------------------------------------------------------------+
641 	 * 8 |            RSV           |  STA  |           RSV             |
642 	 *   +--------------------------------------------------------------+
643 	 *   63                       36 35   32 31                         0
644 	 */
645 
646 	for (n = 0; n < adapter->num_tx_queues; n++) {
647 		tx_ring = adapter->tx_ring[n];
648 		pr_info("------------------------------------\n");
649 		pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
650 		pr_info("------------------------------------\n");
651 		pr_info("%s%s    %s              %s        %s          %s\n",
652 			"T [desc]     [address 63:0  ] ",
653 			"[PlPOIdStDDt Ln] [bi->dma       ] ",
654 			"leng", "ntw", "timestamp", "bi->skb");
655 
656 		for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
657 			tx_desc = IXGBE_TX_DESC(tx_ring, i);
658 			tx_buffer = &tx_ring->tx_buffer_info[i];
659 			u0 = (struct my_u0 *)tx_desc;
660 			if (dma_unmap_len(tx_buffer, len) > 0) {
661 				pr_info("T [0x%03X]    %016llX %016llX %016llX %08X %p %016llX %p",
662 					i,
663 					le64_to_cpu(u0->a),
664 					le64_to_cpu(u0->b),
665 					(u64)dma_unmap_addr(tx_buffer, dma),
666 					dma_unmap_len(tx_buffer, len),
667 					tx_buffer->next_to_watch,
668 					(u64)tx_buffer->time_stamp,
669 					tx_buffer->skb);
670 				if (i == tx_ring->next_to_use &&
671 					i == tx_ring->next_to_clean)
672 					pr_cont(" NTC/U\n");
673 				else if (i == tx_ring->next_to_use)
674 					pr_cont(" NTU\n");
675 				else if (i == tx_ring->next_to_clean)
676 					pr_cont(" NTC\n");
677 				else
678 					pr_cont("\n");
679 
680 				if (netif_msg_pktdata(adapter) &&
681 				    tx_buffer->skb)
682 					print_hex_dump(KERN_INFO, "",
683 						DUMP_PREFIX_ADDRESS, 16, 1,
684 						tx_buffer->skb->data,
685 						dma_unmap_len(tx_buffer, len),
686 						true);
687 			}
688 		}
689 	}
690 
691 	/* Print RX Rings Summary */
692 rx_ring_summary:
693 	dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
694 	pr_info("Queue [NTU] [NTC]\n");
695 	for (n = 0; n < adapter->num_rx_queues; n++) {
696 		rx_ring = adapter->rx_ring[n];
697 		pr_info("%5d %5X %5X\n",
698 			n, rx_ring->next_to_use, rx_ring->next_to_clean);
699 	}
700 
701 	/* Print RX Rings */
702 	if (!netif_msg_rx_status(adapter))
703 		return;
704 
705 	dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
706 
707 	/* Receive Descriptor Formats
708 	 *
709 	 * 82598 Advanced Receive Descriptor (Read) Format
710 	 *    63                                           1        0
711 	 *    +-----------------------------------------------------+
712 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
713 	 *    +----------------------------------------------+------+
714 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
715 	 *    +-----------------------------------------------------+
716 	 *
717 	 *
718 	 * 82598 Advanced Receive Descriptor (Write-Back) Format
719 	 *
720 	 *   63       48 47    32 31  30      21 20 16 15   4 3     0
721 	 *   +------------------------------------------------------+
722 	 * 0 |       RSS Hash /  |SPH| HDR_LEN  | RSV |Packet|  RSS |
723 	 *   | Packet   | IP     |   |          |     | Type | Type |
724 	 *   | Checksum | Ident  |   |          |     |      |      |
725 	 *   +------------------------------------------------------+
726 	 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
727 	 *   +------------------------------------------------------+
728 	 *   63       48 47    32 31            20 19               0
729 	 *
730 	 * 82599+ Advanced Receive Descriptor (Read) Format
731 	 *    63                                           1        0
732 	 *    +-----------------------------------------------------+
733 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
734 	 *    +----------------------------------------------+------+
735 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
736 	 *    +-----------------------------------------------------+
737 	 *
738 	 *
739 	 * 82599+ Advanced Receive Descriptor (Write-Back) Format
740 	 *
741 	 *   63       48 47    32 31  30      21 20 17 16   4 3     0
742 	 *   +------------------------------------------------------+
743 	 * 0 |RSS / Frag Checksum|SPH| HDR_LEN  |RSC- |Packet|  RSS |
744 	 *   |/ RTT / PCoE_PARAM |   |          | CNT | Type | Type |
745 	 *   |/ Flow Dir Flt ID  |   |          |     |      |      |
746 	 *   +------------------------------------------------------+
747 	 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
748 	 *   +------------------------------------------------------+
749 	 *   63       48 47    32 31          20 19                 0
750 	 */
751 
752 	for (n = 0; n < adapter->num_rx_queues; n++) {
753 		rx_ring = adapter->rx_ring[n];
754 		pr_info("------------------------------------\n");
755 		pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
756 		pr_info("------------------------------------\n");
757 		pr_info("%s%s%s",
758 			"R  [desc]      [ PktBuf     A0] ",
759 			"[  HeadBuf   DD] [bi->dma       ] [bi->skb       ] ",
760 			"<-- Adv Rx Read format\n");
761 		pr_info("%s%s%s",
762 			"RWB[desc]      [PcsmIpSHl PtRs] ",
763 			"[vl er S cks ln] ---------------- [bi->skb       ] ",
764 			"<-- Adv Rx Write-Back format\n");
765 
766 		for (i = 0; i < rx_ring->count; i++) {
767 			rx_buffer_info = &rx_ring->rx_buffer_info[i];
768 			rx_desc = IXGBE_RX_DESC(rx_ring, i);
769 			u0 = (struct my_u0 *)rx_desc;
770 			staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
771 			if (staterr & IXGBE_RXD_STAT_DD) {
772 				/* Descriptor Done */
773 				pr_info("RWB[0x%03X]     %016llX "
774 					"%016llX ---------------- %p", i,
775 					le64_to_cpu(u0->a),
776 					le64_to_cpu(u0->b),
777 					rx_buffer_info->skb);
778 			} else {
779 				pr_info("R  [0x%03X]     %016llX "
780 					"%016llX %016llX %p", i,
781 					le64_to_cpu(u0->a),
782 					le64_to_cpu(u0->b),
783 					(u64)rx_buffer_info->dma,
784 					rx_buffer_info->skb);
785 
786 				if (netif_msg_pktdata(adapter) &&
787 				    rx_buffer_info->dma) {
788 					print_hex_dump(KERN_INFO, "",
789 					   DUMP_PREFIX_ADDRESS, 16, 1,
790 					   page_address(rx_buffer_info->page) +
791 						    rx_buffer_info->page_offset,
792 					   ixgbe_rx_bufsz(rx_ring), true);
793 				}
794 			}
795 
796 			if (i == rx_ring->next_to_use)
797 				pr_cont(" NTU\n");
798 			else if (i == rx_ring->next_to_clean)
799 				pr_cont(" NTC\n");
800 			else
801 				pr_cont("\n");
802 
803 		}
804 	}
805 }
806 
807 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
808 {
809 	u32 ctrl_ext;
810 
811 	/* Let firmware take over control of h/w */
812 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
813 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
814 			ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
815 }
816 
817 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
818 {
819 	u32 ctrl_ext;
820 
821 	/* Let firmware know the driver has taken over */
822 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
823 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
824 			ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
825 }
826 
827 /**
828  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
829  * @adapter: pointer to adapter struct
830  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
831  * @queue: queue to map the corresponding interrupt to
832  * @msix_vector: the vector to map to the corresponding queue
833  *
834  */
835 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
836 			   u8 queue, u8 msix_vector)
837 {
838 	u32 ivar, index;
839 	struct ixgbe_hw *hw = &adapter->hw;
840 	switch (hw->mac.type) {
841 	case ixgbe_mac_82598EB:
842 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
843 		if (direction == -1)
844 			direction = 0;
845 		index = (((direction * 64) + queue) >> 2) & 0x1F;
846 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
847 		ivar &= ~(0xFF << (8 * (queue & 0x3)));
848 		ivar |= (msix_vector << (8 * (queue & 0x3)));
849 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
850 		break;
851 	case ixgbe_mac_82599EB:
852 	case ixgbe_mac_X540:
853 	case ixgbe_mac_X550:
854 	case ixgbe_mac_X550EM_x:
855 		if (direction == -1) {
856 			/* other causes */
857 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
858 			index = ((queue & 1) * 8);
859 			ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
860 			ivar &= ~(0xFF << index);
861 			ivar |= (msix_vector << index);
862 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
863 			break;
864 		} else {
865 			/* tx or rx causes */
866 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
867 			index = ((16 * (queue & 1)) + (8 * direction));
868 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
869 			ivar &= ~(0xFF << index);
870 			ivar |= (msix_vector << index);
871 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
872 			break;
873 		}
874 	default:
875 		break;
876 	}
877 }
878 
879 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
880 					  u64 qmask)
881 {
882 	u32 mask;
883 
884 	switch (adapter->hw.mac.type) {
885 	case ixgbe_mac_82598EB:
886 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
887 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
888 		break;
889 	case ixgbe_mac_82599EB:
890 	case ixgbe_mac_X540:
891 	case ixgbe_mac_X550:
892 	case ixgbe_mac_X550EM_x:
893 		mask = (qmask & 0xFFFFFFFF);
894 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
895 		mask = (qmask >> 32);
896 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
897 		break;
898 	default:
899 		break;
900 	}
901 }
902 
903 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *ring,
904 				      struct ixgbe_tx_buffer *tx_buffer)
905 {
906 	if (tx_buffer->skb) {
907 		dev_kfree_skb_any(tx_buffer->skb);
908 		if (dma_unmap_len(tx_buffer, len))
909 			dma_unmap_single(ring->dev,
910 					 dma_unmap_addr(tx_buffer, dma),
911 					 dma_unmap_len(tx_buffer, len),
912 					 DMA_TO_DEVICE);
913 	} else if (dma_unmap_len(tx_buffer, len)) {
914 		dma_unmap_page(ring->dev,
915 			       dma_unmap_addr(tx_buffer, dma),
916 			       dma_unmap_len(tx_buffer, len),
917 			       DMA_TO_DEVICE);
918 	}
919 	tx_buffer->next_to_watch = NULL;
920 	tx_buffer->skb = NULL;
921 	dma_unmap_len_set(tx_buffer, len, 0);
922 	/* tx_buffer must be completely set up in the transmit path */
923 }
924 
925 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
926 {
927 	struct ixgbe_hw *hw = &adapter->hw;
928 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
929 	int i;
930 	u32 data;
931 
932 	if ((hw->fc.current_mode != ixgbe_fc_full) &&
933 	    (hw->fc.current_mode != ixgbe_fc_rx_pause))
934 		return;
935 
936 	switch (hw->mac.type) {
937 	case ixgbe_mac_82598EB:
938 		data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
939 		break;
940 	default:
941 		data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
942 	}
943 	hwstats->lxoffrxc += data;
944 
945 	/* refill credits (no tx hang) if we received xoff */
946 	if (!data)
947 		return;
948 
949 	for (i = 0; i < adapter->num_tx_queues; i++)
950 		clear_bit(__IXGBE_HANG_CHECK_ARMED,
951 			  &adapter->tx_ring[i]->state);
952 }
953 
954 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
955 {
956 	struct ixgbe_hw *hw = &adapter->hw;
957 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
958 	u32 xoff[8] = {0};
959 	u8 tc;
960 	int i;
961 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
962 
963 	if (adapter->ixgbe_ieee_pfc)
964 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
965 
966 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
967 		ixgbe_update_xoff_rx_lfc(adapter);
968 		return;
969 	}
970 
971 	/* update stats for each tc, only valid with PFC enabled */
972 	for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
973 		u32 pxoffrxc;
974 
975 		switch (hw->mac.type) {
976 		case ixgbe_mac_82598EB:
977 			pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
978 			break;
979 		default:
980 			pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
981 		}
982 		hwstats->pxoffrxc[i] += pxoffrxc;
983 		/* Get the TC for given UP */
984 		tc = netdev_get_prio_tc_map(adapter->netdev, i);
985 		xoff[tc] += pxoffrxc;
986 	}
987 
988 	/* disarm tx queues that have received xoff frames */
989 	for (i = 0; i < adapter->num_tx_queues; i++) {
990 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
991 
992 		tc = tx_ring->dcb_tc;
993 		if (xoff[tc])
994 			clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
995 	}
996 }
997 
998 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
999 {
1000 	return ring->stats.packets;
1001 }
1002 
1003 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
1004 {
1005 	struct ixgbe_adapter *adapter;
1006 	struct ixgbe_hw *hw;
1007 	u32 head, tail;
1008 
1009 	if (ring->l2_accel_priv)
1010 		adapter = ring->l2_accel_priv->real_adapter;
1011 	else
1012 		adapter = netdev_priv(ring->netdev);
1013 
1014 	hw = &adapter->hw;
1015 	head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
1016 	tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
1017 
1018 	if (head != tail)
1019 		return (head < tail) ?
1020 			tail - head : (tail + ring->count - head);
1021 
1022 	return 0;
1023 }
1024 
1025 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
1026 {
1027 	u32 tx_done = ixgbe_get_tx_completed(tx_ring);
1028 	u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
1029 	u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
1030 
1031 	clear_check_for_tx_hang(tx_ring);
1032 
1033 	/*
1034 	 * Check for a hung queue, but be thorough. This verifies
1035 	 * that a transmit has been completed since the previous
1036 	 * check AND there is at least one packet pending. The
1037 	 * ARMED bit is set to indicate a potential hang. The
1038 	 * bit is cleared if a pause frame is received to remove
1039 	 * false hang detection due to PFC or 802.3x frames. By
1040 	 * requiring this to fail twice we avoid races with
1041 	 * pfc clearing the ARMED bit and conditions where we
1042 	 * run the check_tx_hang logic with a transmit completion
1043 	 * pending but without time to complete it yet.
1044 	 */
1045 	if (tx_done_old == tx_done && tx_pending)
1046 		/* make sure it is true for two checks in a row */
1047 		return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
1048 					&tx_ring->state);
1049 	/* update completed stats and continue */
1050 	tx_ring->tx_stats.tx_done_old = tx_done;
1051 	/* reset the countdown */
1052 	clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1053 
1054 	return false;
1055 }
1056 
1057 /**
1058  * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1059  * @adapter: driver private struct
1060  **/
1061 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
1062 {
1063 
1064 	/* Do the reset outside of interrupt context */
1065 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1066 		adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
1067 		e_warn(drv, "initiating reset due to tx timeout\n");
1068 		ixgbe_service_event_schedule(adapter);
1069 	}
1070 }
1071 
1072 /**
1073  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1074  * @q_vector: structure containing interrupt and ring information
1075  * @tx_ring: tx ring to clean
1076  **/
1077 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
1078 			       struct ixgbe_ring *tx_ring)
1079 {
1080 	struct ixgbe_adapter *adapter = q_vector->adapter;
1081 	struct ixgbe_tx_buffer *tx_buffer;
1082 	union ixgbe_adv_tx_desc *tx_desc;
1083 	unsigned int total_bytes = 0, total_packets = 0;
1084 	unsigned int budget = q_vector->tx.work_limit;
1085 	unsigned int i = tx_ring->next_to_clean;
1086 
1087 	if (test_bit(__IXGBE_DOWN, &adapter->state))
1088 		return true;
1089 
1090 	tx_buffer = &tx_ring->tx_buffer_info[i];
1091 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
1092 	i -= tx_ring->count;
1093 
1094 	do {
1095 		union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1096 
1097 		/* if next_to_watch is not set then there is no work pending */
1098 		if (!eop_desc)
1099 			break;
1100 
1101 		/* prevent any other reads prior to eop_desc */
1102 		read_barrier_depends();
1103 
1104 		/* if DD is not set pending work has not been completed */
1105 		if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
1106 			break;
1107 
1108 		/* clear next_to_watch to prevent false hangs */
1109 		tx_buffer->next_to_watch = NULL;
1110 
1111 		/* update the statistics for this packet */
1112 		total_bytes += tx_buffer->bytecount;
1113 		total_packets += tx_buffer->gso_segs;
1114 
1115 		/* free the skb */
1116 		dev_consume_skb_any(tx_buffer->skb);
1117 
1118 		/* unmap skb header data */
1119 		dma_unmap_single(tx_ring->dev,
1120 				 dma_unmap_addr(tx_buffer, dma),
1121 				 dma_unmap_len(tx_buffer, len),
1122 				 DMA_TO_DEVICE);
1123 
1124 		/* clear tx_buffer data */
1125 		tx_buffer->skb = NULL;
1126 		dma_unmap_len_set(tx_buffer, len, 0);
1127 
1128 		/* unmap remaining buffers */
1129 		while (tx_desc != eop_desc) {
1130 			tx_buffer++;
1131 			tx_desc++;
1132 			i++;
1133 			if (unlikely(!i)) {
1134 				i -= tx_ring->count;
1135 				tx_buffer = tx_ring->tx_buffer_info;
1136 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1137 			}
1138 
1139 			/* unmap any remaining paged data */
1140 			if (dma_unmap_len(tx_buffer, len)) {
1141 				dma_unmap_page(tx_ring->dev,
1142 					       dma_unmap_addr(tx_buffer, dma),
1143 					       dma_unmap_len(tx_buffer, len),
1144 					       DMA_TO_DEVICE);
1145 				dma_unmap_len_set(tx_buffer, len, 0);
1146 			}
1147 		}
1148 
1149 		/* move us one more past the eop_desc for start of next pkt */
1150 		tx_buffer++;
1151 		tx_desc++;
1152 		i++;
1153 		if (unlikely(!i)) {
1154 			i -= tx_ring->count;
1155 			tx_buffer = tx_ring->tx_buffer_info;
1156 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1157 		}
1158 
1159 		/* issue prefetch for next Tx descriptor */
1160 		prefetch(tx_desc);
1161 
1162 		/* update budget accounting */
1163 		budget--;
1164 	} while (likely(budget));
1165 
1166 	i += tx_ring->count;
1167 	tx_ring->next_to_clean = i;
1168 	u64_stats_update_begin(&tx_ring->syncp);
1169 	tx_ring->stats.bytes += total_bytes;
1170 	tx_ring->stats.packets += total_packets;
1171 	u64_stats_update_end(&tx_ring->syncp);
1172 	q_vector->tx.total_bytes += total_bytes;
1173 	q_vector->tx.total_packets += total_packets;
1174 
1175 	if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1176 		/* schedule immediate reset if we believe we hung */
1177 		struct ixgbe_hw *hw = &adapter->hw;
1178 		e_err(drv, "Detected Tx Unit Hang\n"
1179 			"  Tx Queue             <%d>\n"
1180 			"  TDH, TDT             <%x>, <%x>\n"
1181 			"  next_to_use          <%x>\n"
1182 			"  next_to_clean        <%x>\n"
1183 			"tx_buffer_info[next_to_clean]\n"
1184 			"  time_stamp           <%lx>\n"
1185 			"  jiffies              <%lx>\n",
1186 			tx_ring->queue_index,
1187 			IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1188 			IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1189 			tx_ring->next_to_use, i,
1190 			tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1191 
1192 		netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
1193 
1194 		e_info(probe,
1195 		       "tx hang %d detected on queue %d, resetting adapter\n",
1196 			adapter->tx_timeout_count + 1, tx_ring->queue_index);
1197 
1198 		/* schedule immediate reset if we believe we hung */
1199 		ixgbe_tx_timeout_reset(adapter);
1200 
1201 		/* the adapter is about to reset, no point in enabling stuff */
1202 		return true;
1203 	}
1204 
1205 	netdev_tx_completed_queue(txring_txq(tx_ring),
1206 				  total_packets, total_bytes);
1207 
1208 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1209 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1210 		     (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1211 		/* Make sure that anybody stopping the queue after this
1212 		 * sees the new next_to_clean.
1213 		 */
1214 		smp_mb();
1215 		if (__netif_subqueue_stopped(tx_ring->netdev,
1216 					     tx_ring->queue_index)
1217 		    && !test_bit(__IXGBE_DOWN, &adapter->state)) {
1218 			netif_wake_subqueue(tx_ring->netdev,
1219 					    tx_ring->queue_index);
1220 			++tx_ring->tx_stats.restart_queue;
1221 		}
1222 	}
1223 
1224 	return !!budget;
1225 }
1226 
1227 #ifdef CONFIG_IXGBE_DCA
1228 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1229 				struct ixgbe_ring *tx_ring,
1230 				int cpu)
1231 {
1232 	struct ixgbe_hw *hw = &adapter->hw;
1233 	u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
1234 	u16 reg_offset;
1235 
1236 	switch (hw->mac.type) {
1237 	case ixgbe_mac_82598EB:
1238 		reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1239 		break;
1240 	case ixgbe_mac_82599EB:
1241 	case ixgbe_mac_X540:
1242 		reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1243 		txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1244 		break;
1245 	default:
1246 		/* for unknown hardware do not write register */
1247 		return;
1248 	}
1249 
1250 	/*
1251 	 * We can enable relaxed ordering for reads, but not writes when
1252 	 * DCA is enabled.  This is due to a known issue in some chipsets
1253 	 * which will cause the DCA tag to be cleared.
1254 	 */
1255 	txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1256 		  IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1257 		  IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1258 
1259 	IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1260 }
1261 
1262 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1263 				struct ixgbe_ring *rx_ring,
1264 				int cpu)
1265 {
1266 	struct ixgbe_hw *hw = &adapter->hw;
1267 	u32 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1268 	u8 reg_idx = rx_ring->reg_idx;
1269 
1270 
1271 	switch (hw->mac.type) {
1272 	case ixgbe_mac_82599EB:
1273 	case ixgbe_mac_X540:
1274 		rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1275 		break;
1276 	default:
1277 		break;
1278 	}
1279 
1280 	/*
1281 	 * We can enable relaxed ordering for reads, but not writes when
1282 	 * DCA is enabled.  This is due to a known issue in some chipsets
1283 	 * which will cause the DCA tag to be cleared.
1284 	 */
1285 	rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1286 		  IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1287 
1288 	IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1289 }
1290 
1291 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1292 {
1293 	struct ixgbe_adapter *adapter = q_vector->adapter;
1294 	struct ixgbe_ring *ring;
1295 	int cpu = get_cpu();
1296 
1297 	if (q_vector->cpu == cpu)
1298 		goto out_no_update;
1299 
1300 	ixgbe_for_each_ring(ring, q_vector->tx)
1301 		ixgbe_update_tx_dca(adapter, ring, cpu);
1302 
1303 	ixgbe_for_each_ring(ring, q_vector->rx)
1304 		ixgbe_update_rx_dca(adapter, ring, cpu);
1305 
1306 	q_vector->cpu = cpu;
1307 out_no_update:
1308 	put_cpu();
1309 }
1310 
1311 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1312 {
1313 	int i;
1314 
1315 	if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1316 		return;
1317 
1318 	/* always use CB2 mode, difference is masked in the CB driver */
1319 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1320 
1321 	for (i = 0; i < adapter->num_q_vectors; i++) {
1322 		adapter->q_vector[i]->cpu = -1;
1323 		ixgbe_update_dca(adapter->q_vector[i]);
1324 	}
1325 }
1326 
1327 static int __ixgbe_notify_dca(struct device *dev, void *data)
1328 {
1329 	struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1330 	unsigned long event = *(unsigned long *)data;
1331 
1332 	if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1333 		return 0;
1334 
1335 	switch (event) {
1336 	case DCA_PROVIDER_ADD:
1337 		/* if we're already enabled, don't do it again */
1338 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1339 			break;
1340 		if (dca_add_requester(dev) == 0) {
1341 			adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1342 			ixgbe_setup_dca(adapter);
1343 			break;
1344 		}
1345 		/* Fall Through since DCA is disabled. */
1346 	case DCA_PROVIDER_REMOVE:
1347 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1348 			dca_remove_requester(dev);
1349 			adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1350 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1351 		}
1352 		break;
1353 	}
1354 
1355 	return 0;
1356 }
1357 
1358 #endif /* CONFIG_IXGBE_DCA */
1359 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1360 				 union ixgbe_adv_rx_desc *rx_desc,
1361 				 struct sk_buff *skb)
1362 {
1363 	if (ring->netdev->features & NETIF_F_RXHASH)
1364 		skb_set_hash(skb,
1365 			     le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1366 			     PKT_HASH_TYPE_L3);
1367 }
1368 
1369 #ifdef IXGBE_FCOE
1370 /**
1371  * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1372  * @ring: structure containing ring specific data
1373  * @rx_desc: advanced rx descriptor
1374  *
1375  * Returns : true if it is FCoE pkt
1376  */
1377 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1378 				    union ixgbe_adv_rx_desc *rx_desc)
1379 {
1380 	__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1381 
1382 	return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1383 	       ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1384 		(cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1385 			     IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1386 }
1387 
1388 #endif /* IXGBE_FCOE */
1389 /**
1390  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1391  * @ring: structure containing ring specific data
1392  * @rx_desc: current Rx descriptor being processed
1393  * @skb: skb currently being received and modified
1394  **/
1395 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1396 				     union ixgbe_adv_rx_desc *rx_desc,
1397 				     struct sk_buff *skb)
1398 {
1399 	skb_checksum_none_assert(skb);
1400 
1401 	/* Rx csum disabled */
1402 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
1403 		return;
1404 
1405 	/* if IP and error */
1406 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1407 	    ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1408 		ring->rx_stats.csum_err++;
1409 		return;
1410 	}
1411 
1412 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1413 		return;
1414 
1415 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1416 		__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1417 
1418 		/*
1419 		 * 82599 errata, UDP frames with a 0 checksum can be marked as
1420 		 * checksum errors.
1421 		 */
1422 		if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1423 		    test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1424 			return;
1425 
1426 		ring->rx_stats.csum_err++;
1427 		return;
1428 	}
1429 
1430 	/* It must be a TCP or UDP packet with a valid checksum */
1431 	skb->ip_summed = CHECKSUM_UNNECESSARY;
1432 }
1433 
1434 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1435 				    struct ixgbe_rx_buffer *bi)
1436 {
1437 	struct page *page = bi->page;
1438 	dma_addr_t dma;
1439 
1440 	/* since we are recycling buffers we should seldom need to alloc */
1441 	if (likely(page))
1442 		return true;
1443 
1444 	/* alloc new page for storage */
1445 	page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
1446 	if (unlikely(!page)) {
1447 		rx_ring->rx_stats.alloc_rx_page_failed++;
1448 		return false;
1449 	}
1450 
1451 	/* map page for use */
1452 	dma = dma_map_page(rx_ring->dev, page, 0,
1453 			   ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1454 
1455 	/*
1456 	 * if mapping failed free memory back to system since
1457 	 * there isn't much point in holding memory we can't use
1458 	 */
1459 	if (dma_mapping_error(rx_ring->dev, dma)) {
1460 		__free_pages(page, ixgbe_rx_pg_order(rx_ring));
1461 
1462 		rx_ring->rx_stats.alloc_rx_page_failed++;
1463 		return false;
1464 	}
1465 
1466 	bi->dma = dma;
1467 	bi->page = page;
1468 	bi->page_offset = 0;
1469 
1470 	return true;
1471 }
1472 
1473 /**
1474  * ixgbe_alloc_rx_buffers - Replace used receive buffers
1475  * @rx_ring: ring to place buffers on
1476  * @cleaned_count: number of buffers to replace
1477  **/
1478 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1479 {
1480 	union ixgbe_adv_rx_desc *rx_desc;
1481 	struct ixgbe_rx_buffer *bi;
1482 	u16 i = rx_ring->next_to_use;
1483 
1484 	/* nothing to do */
1485 	if (!cleaned_count)
1486 		return;
1487 
1488 	rx_desc = IXGBE_RX_DESC(rx_ring, i);
1489 	bi = &rx_ring->rx_buffer_info[i];
1490 	i -= rx_ring->count;
1491 
1492 	do {
1493 		if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1494 			break;
1495 
1496 		/*
1497 		 * Refresh the desc even if buffer_addrs didn't change
1498 		 * because each write-back erases this info.
1499 		 */
1500 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1501 
1502 		rx_desc++;
1503 		bi++;
1504 		i++;
1505 		if (unlikely(!i)) {
1506 			rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1507 			bi = rx_ring->rx_buffer_info;
1508 			i -= rx_ring->count;
1509 		}
1510 
1511 		/* clear the status bits for the next_to_use descriptor */
1512 		rx_desc->wb.upper.status_error = 0;
1513 
1514 		cleaned_count--;
1515 	} while (cleaned_count);
1516 
1517 	i += rx_ring->count;
1518 
1519 	if (rx_ring->next_to_use != i) {
1520 		rx_ring->next_to_use = i;
1521 
1522 		/* update next to alloc since we have filled the ring */
1523 		rx_ring->next_to_alloc = i;
1524 
1525 		/* Force memory writes to complete before letting h/w
1526 		 * know there are new descriptors to fetch.  (Only
1527 		 * applicable for weak-ordered memory model archs,
1528 		 * such as IA-64).
1529 		 */
1530 		wmb();
1531 		writel(i, rx_ring->tail);
1532 	}
1533 }
1534 
1535 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1536 				   struct sk_buff *skb)
1537 {
1538 	u16 hdr_len = skb_headlen(skb);
1539 
1540 	/* set gso_size to avoid messing up TCP MSS */
1541 	skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1542 						 IXGBE_CB(skb)->append_cnt);
1543 	skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1544 }
1545 
1546 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1547 				   struct sk_buff *skb)
1548 {
1549 	/* if append_cnt is 0 then frame is not RSC */
1550 	if (!IXGBE_CB(skb)->append_cnt)
1551 		return;
1552 
1553 	rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1554 	rx_ring->rx_stats.rsc_flush++;
1555 
1556 	ixgbe_set_rsc_gso_size(rx_ring, skb);
1557 
1558 	/* gso_size is computed using append_cnt so always clear it last */
1559 	IXGBE_CB(skb)->append_cnt = 0;
1560 }
1561 
1562 /**
1563  * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1564  * @rx_ring: rx descriptor ring packet is being transacted on
1565  * @rx_desc: pointer to the EOP Rx descriptor
1566  * @skb: pointer to current skb being populated
1567  *
1568  * This function checks the ring, descriptor, and packet information in
1569  * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1570  * other fields within the skb.
1571  **/
1572 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1573 				     union ixgbe_adv_rx_desc *rx_desc,
1574 				     struct sk_buff *skb)
1575 {
1576 	struct net_device *dev = rx_ring->netdev;
1577 
1578 	ixgbe_update_rsc_stats(rx_ring, skb);
1579 
1580 	ixgbe_rx_hash(rx_ring, rx_desc, skb);
1581 
1582 	ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1583 
1584 	if (unlikely(ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_STAT_TS)))
1585 		ixgbe_ptp_rx_hwtstamp(rx_ring->q_vector->adapter, skb);
1586 
1587 	if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1588 	    ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1589 		u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1590 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1591 	}
1592 
1593 	skb_record_rx_queue(skb, rx_ring->queue_index);
1594 
1595 	skb->protocol = eth_type_trans(skb, dev);
1596 }
1597 
1598 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1599 			 struct sk_buff *skb)
1600 {
1601 	struct ixgbe_adapter *adapter = q_vector->adapter;
1602 
1603 	if (ixgbe_qv_busy_polling(q_vector))
1604 		netif_receive_skb(skb);
1605 	else if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1606 		napi_gro_receive(&q_vector->napi, skb);
1607 	else
1608 		netif_rx(skb);
1609 }
1610 
1611 /**
1612  * ixgbe_is_non_eop - process handling of non-EOP buffers
1613  * @rx_ring: Rx ring being processed
1614  * @rx_desc: Rx descriptor for current buffer
1615  * @skb: Current socket buffer containing buffer in progress
1616  *
1617  * This function updates next to clean.  If the buffer is an EOP buffer
1618  * this function exits returning false, otherwise it will place the
1619  * sk_buff in the next buffer to be chained and return true indicating
1620  * that this is in fact a non-EOP buffer.
1621  **/
1622 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1623 			     union ixgbe_adv_rx_desc *rx_desc,
1624 			     struct sk_buff *skb)
1625 {
1626 	u32 ntc = rx_ring->next_to_clean + 1;
1627 
1628 	/* fetch, update, and store next to clean */
1629 	ntc = (ntc < rx_ring->count) ? ntc : 0;
1630 	rx_ring->next_to_clean = ntc;
1631 
1632 	prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1633 
1634 	/* update RSC append count if present */
1635 	if (ring_is_rsc_enabled(rx_ring)) {
1636 		__le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1637 				     cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1638 
1639 		if (unlikely(rsc_enabled)) {
1640 			u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1641 
1642 			rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1643 			IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1644 
1645 			/* update ntc based on RSC value */
1646 			ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1647 			ntc &= IXGBE_RXDADV_NEXTP_MASK;
1648 			ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1649 		}
1650 	}
1651 
1652 	/* if we are the last buffer then there is nothing else to do */
1653 	if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1654 		return false;
1655 
1656 	/* place skb in next buffer to be received */
1657 	rx_ring->rx_buffer_info[ntc].skb = skb;
1658 	rx_ring->rx_stats.non_eop_descs++;
1659 
1660 	return true;
1661 }
1662 
1663 /**
1664  * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1665  * @rx_ring: rx descriptor ring packet is being transacted on
1666  * @skb: pointer to current skb being adjusted
1667  *
1668  * This function is an ixgbe specific version of __pskb_pull_tail.  The
1669  * main difference between this version and the original function is that
1670  * this function can make several assumptions about the state of things
1671  * that allow for significant optimizations versus the standard function.
1672  * As a result we can do things like drop a frag and maintain an accurate
1673  * truesize for the skb.
1674  */
1675 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1676 			    struct sk_buff *skb)
1677 {
1678 	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1679 	unsigned char *va;
1680 	unsigned int pull_len;
1681 
1682 	/*
1683 	 * it is valid to use page_address instead of kmap since we are
1684 	 * working with pages allocated out of the lomem pool per
1685 	 * alloc_page(GFP_ATOMIC)
1686 	 */
1687 	va = skb_frag_address(frag);
1688 
1689 	/*
1690 	 * we need the header to contain the greater of either ETH_HLEN or
1691 	 * 60 bytes if the skb->len is less than 60 for skb_pad.
1692 	 */
1693 	pull_len = eth_get_headlen(va, IXGBE_RX_HDR_SIZE);
1694 
1695 	/* align pull length to size of long to optimize memcpy performance */
1696 	skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1697 
1698 	/* update all of the pointers */
1699 	skb_frag_size_sub(frag, pull_len);
1700 	frag->page_offset += pull_len;
1701 	skb->data_len -= pull_len;
1702 	skb->tail += pull_len;
1703 }
1704 
1705 /**
1706  * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1707  * @rx_ring: rx descriptor ring packet is being transacted on
1708  * @skb: pointer to current skb being updated
1709  *
1710  * This function provides a basic DMA sync up for the first fragment of an
1711  * skb.  The reason for doing this is that the first fragment cannot be
1712  * unmapped until we have reached the end of packet descriptor for a buffer
1713  * chain.
1714  */
1715 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1716 				struct sk_buff *skb)
1717 {
1718 	/* if the page was released unmap it, else just sync our portion */
1719 	if (unlikely(IXGBE_CB(skb)->page_released)) {
1720 		dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1721 			       ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1722 		IXGBE_CB(skb)->page_released = false;
1723 	} else {
1724 		struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1725 
1726 		dma_sync_single_range_for_cpu(rx_ring->dev,
1727 					      IXGBE_CB(skb)->dma,
1728 					      frag->page_offset,
1729 					      ixgbe_rx_bufsz(rx_ring),
1730 					      DMA_FROM_DEVICE);
1731 	}
1732 	IXGBE_CB(skb)->dma = 0;
1733 }
1734 
1735 /**
1736  * ixgbe_cleanup_headers - Correct corrupted or empty headers
1737  * @rx_ring: rx descriptor ring packet is being transacted on
1738  * @rx_desc: pointer to the EOP Rx descriptor
1739  * @skb: pointer to current skb being fixed
1740  *
1741  * Check for corrupted packet headers caused by senders on the local L2
1742  * embedded NIC switch not setting up their Tx Descriptors right.  These
1743  * should be very rare.
1744  *
1745  * Also address the case where we are pulling data in on pages only
1746  * and as such no data is present in the skb header.
1747  *
1748  * In addition if skb is not at least 60 bytes we need to pad it so that
1749  * it is large enough to qualify as a valid Ethernet frame.
1750  *
1751  * Returns true if an error was encountered and skb was freed.
1752  **/
1753 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1754 				  union ixgbe_adv_rx_desc *rx_desc,
1755 				  struct sk_buff *skb)
1756 {
1757 	struct net_device *netdev = rx_ring->netdev;
1758 
1759 	/* verify that the packet does not have any known errors */
1760 	if (unlikely(ixgbe_test_staterr(rx_desc,
1761 					IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1762 	    !(netdev->features & NETIF_F_RXALL))) {
1763 		dev_kfree_skb_any(skb);
1764 		return true;
1765 	}
1766 
1767 	/* place header in linear portion of buffer */
1768 	if (skb_is_nonlinear(skb))
1769 		ixgbe_pull_tail(rx_ring, skb);
1770 
1771 #ifdef IXGBE_FCOE
1772 	/* do not attempt to pad FCoE Frames as this will disrupt DDP */
1773 	if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1774 		return false;
1775 
1776 #endif
1777 	/* if eth_skb_pad returns an error the skb was freed */
1778 	if (eth_skb_pad(skb))
1779 		return true;
1780 
1781 	return false;
1782 }
1783 
1784 /**
1785  * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1786  * @rx_ring: rx descriptor ring to store buffers on
1787  * @old_buff: donor buffer to have page reused
1788  *
1789  * Synchronizes page for reuse by the adapter
1790  **/
1791 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1792 				struct ixgbe_rx_buffer *old_buff)
1793 {
1794 	struct ixgbe_rx_buffer *new_buff;
1795 	u16 nta = rx_ring->next_to_alloc;
1796 
1797 	new_buff = &rx_ring->rx_buffer_info[nta];
1798 
1799 	/* update, and store next to alloc */
1800 	nta++;
1801 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1802 
1803 	/* transfer page from old buffer to new buffer */
1804 	*new_buff = *old_buff;
1805 
1806 	/* sync the buffer for use by the device */
1807 	dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1808 					 new_buff->page_offset,
1809 					 ixgbe_rx_bufsz(rx_ring),
1810 					 DMA_FROM_DEVICE);
1811 }
1812 
1813 static inline bool ixgbe_page_is_reserved(struct page *page)
1814 {
1815 	return (page_to_nid(page) != numa_mem_id()) || page->pfmemalloc;
1816 }
1817 
1818 /**
1819  * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1820  * @rx_ring: rx descriptor ring to transact packets on
1821  * @rx_buffer: buffer containing page to add
1822  * @rx_desc: descriptor containing length of buffer written by hardware
1823  * @skb: sk_buff to place the data into
1824  *
1825  * This function will add the data contained in rx_buffer->page to the skb.
1826  * This is done either through a direct copy if the data in the buffer is
1827  * less than the skb header size, otherwise it will just attach the page as
1828  * a frag to the skb.
1829  *
1830  * The function will then update the page offset if necessary and return
1831  * true if the buffer can be reused by the adapter.
1832  **/
1833 static bool ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1834 			      struct ixgbe_rx_buffer *rx_buffer,
1835 			      union ixgbe_adv_rx_desc *rx_desc,
1836 			      struct sk_buff *skb)
1837 {
1838 	struct page *page = rx_buffer->page;
1839 	unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
1840 #if (PAGE_SIZE < 8192)
1841 	unsigned int truesize = ixgbe_rx_bufsz(rx_ring);
1842 #else
1843 	unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
1844 	unsigned int last_offset = ixgbe_rx_pg_size(rx_ring) -
1845 				   ixgbe_rx_bufsz(rx_ring);
1846 #endif
1847 
1848 	if ((size <= IXGBE_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
1849 		unsigned char *va = page_address(page) + rx_buffer->page_offset;
1850 
1851 		memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
1852 
1853 		/* page is not reserved, we can reuse buffer as-is */
1854 		if (likely(!ixgbe_page_is_reserved(page)))
1855 			return true;
1856 
1857 		/* this page cannot be reused so discard it */
1858 		__free_pages(page, ixgbe_rx_pg_order(rx_ring));
1859 		return false;
1860 	}
1861 
1862 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
1863 			rx_buffer->page_offset, size, truesize);
1864 
1865 	/* avoid re-using remote pages */
1866 	if (unlikely(ixgbe_page_is_reserved(page)))
1867 		return false;
1868 
1869 #if (PAGE_SIZE < 8192)
1870 	/* if we are only owner of page we can reuse it */
1871 	if (unlikely(page_count(page) != 1))
1872 		return false;
1873 
1874 	/* flip page offset to other buffer */
1875 	rx_buffer->page_offset ^= truesize;
1876 #else
1877 	/* move offset up to the next cache line */
1878 	rx_buffer->page_offset += truesize;
1879 
1880 	if (rx_buffer->page_offset > last_offset)
1881 		return false;
1882 #endif
1883 
1884 	/* Even if we own the page, we are not allowed to use atomic_set()
1885 	 * This would break get_page_unless_zero() users.
1886 	 */
1887 	atomic_inc(&page->_count);
1888 
1889 	return true;
1890 }
1891 
1892 static struct sk_buff *ixgbe_fetch_rx_buffer(struct ixgbe_ring *rx_ring,
1893 					     union ixgbe_adv_rx_desc *rx_desc)
1894 {
1895 	struct ixgbe_rx_buffer *rx_buffer;
1896 	struct sk_buff *skb;
1897 	struct page *page;
1898 
1899 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
1900 	page = rx_buffer->page;
1901 	prefetchw(page);
1902 
1903 	skb = rx_buffer->skb;
1904 
1905 	if (likely(!skb)) {
1906 		void *page_addr = page_address(page) +
1907 				  rx_buffer->page_offset;
1908 
1909 		/* prefetch first cache line of first page */
1910 		prefetch(page_addr);
1911 #if L1_CACHE_BYTES < 128
1912 		prefetch(page_addr + L1_CACHE_BYTES);
1913 #endif
1914 
1915 		/* allocate a skb to store the frags */
1916 		skb = napi_alloc_skb(&rx_ring->q_vector->napi,
1917 				     IXGBE_RX_HDR_SIZE);
1918 		if (unlikely(!skb)) {
1919 			rx_ring->rx_stats.alloc_rx_buff_failed++;
1920 			return NULL;
1921 		}
1922 
1923 		/*
1924 		 * we will be copying header into skb->data in
1925 		 * pskb_may_pull so it is in our interest to prefetch
1926 		 * it now to avoid a possible cache miss
1927 		 */
1928 		prefetchw(skb->data);
1929 
1930 		/*
1931 		 * Delay unmapping of the first packet. It carries the
1932 		 * header information, HW may still access the header
1933 		 * after the writeback.  Only unmap it when EOP is
1934 		 * reached
1935 		 */
1936 		if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1937 			goto dma_sync;
1938 
1939 		IXGBE_CB(skb)->dma = rx_buffer->dma;
1940 	} else {
1941 		if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
1942 			ixgbe_dma_sync_frag(rx_ring, skb);
1943 
1944 dma_sync:
1945 		/* we are reusing so sync this buffer for CPU use */
1946 		dma_sync_single_range_for_cpu(rx_ring->dev,
1947 					      rx_buffer->dma,
1948 					      rx_buffer->page_offset,
1949 					      ixgbe_rx_bufsz(rx_ring),
1950 					      DMA_FROM_DEVICE);
1951 
1952 		rx_buffer->skb = NULL;
1953 	}
1954 
1955 	/* pull page into skb */
1956 	if (ixgbe_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
1957 		/* hand second half of page back to the ring */
1958 		ixgbe_reuse_rx_page(rx_ring, rx_buffer);
1959 	} else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
1960 		/* the page has been released from the ring */
1961 		IXGBE_CB(skb)->page_released = true;
1962 	} else {
1963 		/* we are not reusing the buffer so unmap it */
1964 		dma_unmap_page(rx_ring->dev, rx_buffer->dma,
1965 			       ixgbe_rx_pg_size(rx_ring),
1966 			       DMA_FROM_DEVICE);
1967 	}
1968 
1969 	/* clear contents of buffer_info */
1970 	rx_buffer->page = NULL;
1971 
1972 	return skb;
1973 }
1974 
1975 /**
1976  * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1977  * @q_vector: structure containing interrupt and ring information
1978  * @rx_ring: rx descriptor ring to transact packets on
1979  * @budget: Total limit on number of packets to process
1980  *
1981  * This function provides a "bounce buffer" approach to Rx interrupt
1982  * processing.  The advantage to this is that on systems that have
1983  * expensive overhead for IOMMU access this provides a means of avoiding
1984  * it by maintaining the mapping of the page to the syste.
1985  *
1986  * Returns amount of work completed
1987  **/
1988 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1989 			       struct ixgbe_ring *rx_ring,
1990 			       const int budget)
1991 {
1992 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1993 #ifdef IXGBE_FCOE
1994 	struct ixgbe_adapter *adapter = q_vector->adapter;
1995 	int ddp_bytes;
1996 	unsigned int mss = 0;
1997 #endif /* IXGBE_FCOE */
1998 	u16 cleaned_count = ixgbe_desc_unused(rx_ring);
1999 
2000 	while (likely(total_rx_packets < budget)) {
2001 		union ixgbe_adv_rx_desc *rx_desc;
2002 		struct sk_buff *skb;
2003 
2004 		/* return some buffers to hardware, one at a time is too slow */
2005 		if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2006 			ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2007 			cleaned_count = 0;
2008 		}
2009 
2010 		rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2011 
2012 		if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
2013 			break;
2014 
2015 		/*
2016 		 * This memory barrier is needed to keep us from reading
2017 		 * any other fields out of the rx_desc until we know the
2018 		 * RXD_STAT_DD bit is set
2019 		 */
2020 		rmb();
2021 
2022 		/* retrieve a buffer from the ring */
2023 		skb = ixgbe_fetch_rx_buffer(rx_ring, rx_desc);
2024 
2025 		/* exit if we failed to retrieve a buffer */
2026 		if (!skb)
2027 			break;
2028 
2029 		cleaned_count++;
2030 
2031 		/* place incomplete frames back on ring for completion */
2032 		if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2033 			continue;
2034 
2035 		/* verify the packet layout is correct */
2036 		if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2037 			continue;
2038 
2039 		/* probably a little skewed due to removing CRC */
2040 		total_rx_bytes += skb->len;
2041 
2042 		/* populate checksum, timestamp, VLAN, and protocol */
2043 		ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2044 
2045 #ifdef IXGBE_FCOE
2046 		/* if ddp, not passing to ULD unless for FCP_RSP or error */
2047 		if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2048 			ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2049 			/* include DDPed FCoE data */
2050 			if (ddp_bytes > 0) {
2051 				if (!mss) {
2052 					mss = rx_ring->netdev->mtu -
2053 						sizeof(struct fcoe_hdr) -
2054 						sizeof(struct fc_frame_header) -
2055 						sizeof(struct fcoe_crc_eof);
2056 					if (mss > 512)
2057 						mss &= ~511;
2058 				}
2059 				total_rx_bytes += ddp_bytes;
2060 				total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2061 								 mss);
2062 			}
2063 			if (!ddp_bytes) {
2064 				dev_kfree_skb_any(skb);
2065 				continue;
2066 			}
2067 		}
2068 
2069 #endif /* IXGBE_FCOE */
2070 		skb_mark_napi_id(skb, &q_vector->napi);
2071 		ixgbe_rx_skb(q_vector, skb);
2072 
2073 		/* update budget accounting */
2074 		total_rx_packets++;
2075 	}
2076 
2077 	u64_stats_update_begin(&rx_ring->syncp);
2078 	rx_ring->stats.packets += total_rx_packets;
2079 	rx_ring->stats.bytes += total_rx_bytes;
2080 	u64_stats_update_end(&rx_ring->syncp);
2081 	q_vector->rx.total_packets += total_rx_packets;
2082 	q_vector->rx.total_bytes += total_rx_bytes;
2083 
2084 	return total_rx_packets;
2085 }
2086 
2087 #ifdef CONFIG_NET_RX_BUSY_POLL
2088 /* must be called with local_bh_disable()d */
2089 static int ixgbe_low_latency_recv(struct napi_struct *napi)
2090 {
2091 	struct ixgbe_q_vector *q_vector =
2092 			container_of(napi, struct ixgbe_q_vector, napi);
2093 	struct ixgbe_adapter *adapter = q_vector->adapter;
2094 	struct ixgbe_ring  *ring;
2095 	int found = 0;
2096 
2097 	if (test_bit(__IXGBE_DOWN, &adapter->state))
2098 		return LL_FLUSH_FAILED;
2099 
2100 	if (!ixgbe_qv_lock_poll(q_vector))
2101 		return LL_FLUSH_BUSY;
2102 
2103 	ixgbe_for_each_ring(ring, q_vector->rx) {
2104 		found = ixgbe_clean_rx_irq(q_vector, ring, 4);
2105 #ifdef BP_EXTENDED_STATS
2106 		if (found)
2107 			ring->stats.cleaned += found;
2108 		else
2109 			ring->stats.misses++;
2110 #endif
2111 		if (found)
2112 			break;
2113 	}
2114 
2115 	ixgbe_qv_unlock_poll(q_vector);
2116 
2117 	return found;
2118 }
2119 #endif	/* CONFIG_NET_RX_BUSY_POLL */
2120 
2121 /**
2122  * ixgbe_configure_msix - Configure MSI-X hardware
2123  * @adapter: board private structure
2124  *
2125  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2126  * interrupts.
2127  **/
2128 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2129 {
2130 	struct ixgbe_q_vector *q_vector;
2131 	int v_idx;
2132 	u32 mask;
2133 
2134 	/* Populate MSIX to EITR Select */
2135 	if (adapter->num_vfs > 32) {
2136 		u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2137 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2138 	}
2139 
2140 	/*
2141 	 * Populate the IVAR table and set the ITR values to the
2142 	 * corresponding register.
2143 	 */
2144 	for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2145 		struct ixgbe_ring *ring;
2146 		q_vector = adapter->q_vector[v_idx];
2147 
2148 		ixgbe_for_each_ring(ring, q_vector->rx)
2149 			ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2150 
2151 		ixgbe_for_each_ring(ring, q_vector->tx)
2152 			ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2153 
2154 		ixgbe_write_eitr(q_vector);
2155 	}
2156 
2157 	switch (adapter->hw.mac.type) {
2158 	case ixgbe_mac_82598EB:
2159 		ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2160 			       v_idx);
2161 		break;
2162 	case ixgbe_mac_82599EB:
2163 	case ixgbe_mac_X540:
2164 	case ixgbe_mac_X550:
2165 	case ixgbe_mac_X550EM_x:
2166 		ixgbe_set_ivar(adapter, -1, 1, v_idx);
2167 		break;
2168 	default:
2169 		break;
2170 	}
2171 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2172 
2173 	/* set up to autoclear timer, and the vectors */
2174 	mask = IXGBE_EIMS_ENABLE_MASK;
2175 	mask &= ~(IXGBE_EIMS_OTHER |
2176 		  IXGBE_EIMS_MAILBOX |
2177 		  IXGBE_EIMS_LSC);
2178 
2179 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2180 }
2181 
2182 enum latency_range {
2183 	lowest_latency = 0,
2184 	low_latency = 1,
2185 	bulk_latency = 2,
2186 	latency_invalid = 255
2187 };
2188 
2189 /**
2190  * ixgbe_update_itr - update the dynamic ITR value based on statistics
2191  * @q_vector: structure containing interrupt and ring information
2192  * @ring_container: structure containing ring performance data
2193  *
2194  *      Stores a new ITR value based on packets and byte
2195  *      counts during the last interrupt.  The advantage of per interrupt
2196  *      computation is faster updates and more accurate ITR for the current
2197  *      traffic pattern.  Constants in this function were computed
2198  *      based on theoretical maximum wire speed and thresholds were set based
2199  *      on testing data as well as attempting to minimize response time
2200  *      while increasing bulk throughput.
2201  *      this functionality is controlled by the InterruptThrottleRate module
2202  *      parameter (see ixgbe_param.c)
2203  **/
2204 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2205 			     struct ixgbe_ring_container *ring_container)
2206 {
2207 	int bytes = ring_container->total_bytes;
2208 	int packets = ring_container->total_packets;
2209 	u32 timepassed_us;
2210 	u64 bytes_perint;
2211 	u8 itr_setting = ring_container->itr;
2212 
2213 	if (packets == 0)
2214 		return;
2215 
2216 	/* simple throttlerate management
2217 	 *   0-10MB/s   lowest (100000 ints/s)
2218 	 *  10-20MB/s   low    (20000 ints/s)
2219 	 *  20-1249MB/s bulk   (8000 ints/s)
2220 	 */
2221 	/* what was last interrupt timeslice? */
2222 	timepassed_us = q_vector->itr >> 2;
2223 	if (timepassed_us == 0)
2224 		return;
2225 
2226 	bytes_perint = bytes / timepassed_us; /* bytes/usec */
2227 
2228 	switch (itr_setting) {
2229 	case lowest_latency:
2230 		if (bytes_perint > 10)
2231 			itr_setting = low_latency;
2232 		break;
2233 	case low_latency:
2234 		if (bytes_perint > 20)
2235 			itr_setting = bulk_latency;
2236 		else if (bytes_perint <= 10)
2237 			itr_setting = lowest_latency;
2238 		break;
2239 	case bulk_latency:
2240 		if (bytes_perint <= 20)
2241 			itr_setting = low_latency;
2242 		break;
2243 	}
2244 
2245 	/* clear work counters since we have the values we need */
2246 	ring_container->total_bytes = 0;
2247 	ring_container->total_packets = 0;
2248 
2249 	/* write updated itr to ring container */
2250 	ring_container->itr = itr_setting;
2251 }
2252 
2253 /**
2254  * ixgbe_write_eitr - write EITR register in hardware specific way
2255  * @q_vector: structure containing interrupt and ring information
2256  *
2257  * This function is made to be called by ethtool and by the driver
2258  * when it needs to update EITR registers at runtime.  Hardware
2259  * specific quirks/differences are taken care of here.
2260  */
2261 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2262 {
2263 	struct ixgbe_adapter *adapter = q_vector->adapter;
2264 	struct ixgbe_hw *hw = &adapter->hw;
2265 	int v_idx = q_vector->v_idx;
2266 	u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2267 
2268 	switch (adapter->hw.mac.type) {
2269 	case ixgbe_mac_82598EB:
2270 		/* must write high and low 16 bits to reset counter */
2271 		itr_reg |= (itr_reg << 16);
2272 		break;
2273 	case ixgbe_mac_82599EB:
2274 	case ixgbe_mac_X540:
2275 	case ixgbe_mac_X550:
2276 	case ixgbe_mac_X550EM_x:
2277 		/*
2278 		 * set the WDIS bit to not clear the timer bits and cause an
2279 		 * immediate assertion of the interrupt
2280 		 */
2281 		itr_reg |= IXGBE_EITR_CNT_WDIS;
2282 		break;
2283 	default:
2284 		break;
2285 	}
2286 	IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2287 }
2288 
2289 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2290 {
2291 	u32 new_itr = q_vector->itr;
2292 	u8 current_itr;
2293 
2294 	ixgbe_update_itr(q_vector, &q_vector->tx);
2295 	ixgbe_update_itr(q_vector, &q_vector->rx);
2296 
2297 	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
2298 
2299 	switch (current_itr) {
2300 	/* counts and packets in update_itr are dependent on these numbers */
2301 	case lowest_latency:
2302 		new_itr = IXGBE_100K_ITR;
2303 		break;
2304 	case low_latency:
2305 		new_itr = IXGBE_20K_ITR;
2306 		break;
2307 	case bulk_latency:
2308 		new_itr = IXGBE_8K_ITR;
2309 		break;
2310 	default:
2311 		break;
2312 	}
2313 
2314 	if (new_itr != q_vector->itr) {
2315 		/* do an exponential smoothing */
2316 		new_itr = (10 * new_itr * q_vector->itr) /
2317 			  ((9 * new_itr) + q_vector->itr);
2318 
2319 		/* save the algorithm value here */
2320 		q_vector->itr = new_itr;
2321 
2322 		ixgbe_write_eitr(q_vector);
2323 	}
2324 }
2325 
2326 /**
2327  * ixgbe_check_overtemp_subtask - check for over temperature
2328  * @adapter: pointer to adapter
2329  **/
2330 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2331 {
2332 	struct ixgbe_hw *hw = &adapter->hw;
2333 	u32 eicr = adapter->interrupt_event;
2334 
2335 	if (test_bit(__IXGBE_DOWN, &adapter->state))
2336 		return;
2337 
2338 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2339 	    !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2340 		return;
2341 
2342 	adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2343 
2344 	switch (hw->device_id) {
2345 	case IXGBE_DEV_ID_82599_T3_LOM:
2346 		/*
2347 		 * Since the warning interrupt is for both ports
2348 		 * we don't have to check if:
2349 		 *  - This interrupt wasn't for our port.
2350 		 *  - We may have missed the interrupt so always have to
2351 		 *    check if we  got a LSC
2352 		 */
2353 		if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
2354 		    !(eicr & IXGBE_EICR_LSC))
2355 			return;
2356 
2357 		if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2358 			u32 speed;
2359 			bool link_up = false;
2360 
2361 			hw->mac.ops.check_link(hw, &speed, &link_up, false);
2362 
2363 			if (link_up)
2364 				return;
2365 		}
2366 
2367 		/* Check if this is not due to overtemp */
2368 		if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2369 			return;
2370 
2371 		break;
2372 	default:
2373 		if (!(eicr & IXGBE_EICR_GPI_SDP0))
2374 			return;
2375 		break;
2376 	}
2377 	e_crit(drv,
2378 	       "Network adapter has been stopped because it has over heated. "
2379 	       "Restart the computer. If the problem persists, "
2380 	       "power off the system and replace the adapter\n");
2381 
2382 	adapter->interrupt_event = 0;
2383 }
2384 
2385 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2386 {
2387 	struct ixgbe_hw *hw = &adapter->hw;
2388 
2389 	if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2390 	    (eicr & IXGBE_EICR_GPI_SDP1)) {
2391 		e_crit(probe, "Fan has stopped, replace the adapter\n");
2392 		/* write to clear the interrupt */
2393 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2394 	}
2395 }
2396 
2397 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2398 {
2399 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2400 		return;
2401 
2402 	switch (adapter->hw.mac.type) {
2403 	case ixgbe_mac_82599EB:
2404 		/*
2405 		 * Need to check link state so complete overtemp check
2406 		 * on service task
2407 		 */
2408 		if (((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)) &&
2409 		    (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2410 			adapter->interrupt_event = eicr;
2411 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2412 			ixgbe_service_event_schedule(adapter);
2413 			return;
2414 		}
2415 		return;
2416 	case ixgbe_mac_X540:
2417 		if (!(eicr & IXGBE_EICR_TS))
2418 			return;
2419 		break;
2420 	default:
2421 		return;
2422 	}
2423 
2424 	e_crit(drv,
2425 	       "Network adapter has been stopped because it has over heated. "
2426 	       "Restart the computer. If the problem persists, "
2427 	       "power off the system and replace the adapter\n");
2428 }
2429 
2430 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2431 {
2432 	struct ixgbe_hw *hw = &adapter->hw;
2433 
2434 	if (eicr & IXGBE_EICR_GPI_SDP2) {
2435 		/* Clear the interrupt */
2436 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
2437 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2438 			adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2439 			ixgbe_service_event_schedule(adapter);
2440 		}
2441 	}
2442 
2443 	if (eicr & IXGBE_EICR_GPI_SDP1) {
2444 		/* Clear the interrupt */
2445 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2446 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2447 			adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2448 			ixgbe_service_event_schedule(adapter);
2449 		}
2450 	}
2451 }
2452 
2453 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2454 {
2455 	struct ixgbe_hw *hw = &adapter->hw;
2456 
2457 	adapter->lsc_int++;
2458 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2459 	adapter->link_check_timeout = jiffies;
2460 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2461 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2462 		IXGBE_WRITE_FLUSH(hw);
2463 		ixgbe_service_event_schedule(adapter);
2464 	}
2465 }
2466 
2467 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2468 					   u64 qmask)
2469 {
2470 	u32 mask;
2471 	struct ixgbe_hw *hw = &adapter->hw;
2472 
2473 	switch (hw->mac.type) {
2474 	case ixgbe_mac_82598EB:
2475 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2476 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2477 		break;
2478 	case ixgbe_mac_82599EB:
2479 	case ixgbe_mac_X540:
2480 	case ixgbe_mac_X550:
2481 	case ixgbe_mac_X550EM_x:
2482 		mask = (qmask & 0xFFFFFFFF);
2483 		if (mask)
2484 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2485 		mask = (qmask >> 32);
2486 		if (mask)
2487 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2488 		break;
2489 	default:
2490 		break;
2491 	}
2492 	/* skip the flush */
2493 }
2494 
2495 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2496 					    u64 qmask)
2497 {
2498 	u32 mask;
2499 	struct ixgbe_hw *hw = &adapter->hw;
2500 
2501 	switch (hw->mac.type) {
2502 	case ixgbe_mac_82598EB:
2503 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2504 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2505 		break;
2506 	case ixgbe_mac_82599EB:
2507 	case ixgbe_mac_X540:
2508 	case ixgbe_mac_X550:
2509 	case ixgbe_mac_X550EM_x:
2510 		mask = (qmask & 0xFFFFFFFF);
2511 		if (mask)
2512 			IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2513 		mask = (qmask >> 32);
2514 		if (mask)
2515 			IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2516 		break;
2517 	default:
2518 		break;
2519 	}
2520 	/* skip the flush */
2521 }
2522 
2523 /**
2524  * ixgbe_irq_enable - Enable default interrupt generation settings
2525  * @adapter: board private structure
2526  **/
2527 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2528 				    bool flush)
2529 {
2530 	u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2531 
2532 	/* don't reenable LSC while waiting for link */
2533 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2534 		mask &= ~IXGBE_EIMS_LSC;
2535 
2536 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2537 		switch (adapter->hw.mac.type) {
2538 		case ixgbe_mac_82599EB:
2539 			mask |= IXGBE_EIMS_GPI_SDP0;
2540 			break;
2541 		case ixgbe_mac_X540:
2542 		case ixgbe_mac_X550:
2543 		case ixgbe_mac_X550EM_x:
2544 			mask |= IXGBE_EIMS_TS;
2545 			break;
2546 		default:
2547 			break;
2548 		}
2549 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2550 		mask |= IXGBE_EIMS_GPI_SDP1;
2551 	switch (adapter->hw.mac.type) {
2552 	case ixgbe_mac_82599EB:
2553 		mask |= IXGBE_EIMS_GPI_SDP1;
2554 		mask |= IXGBE_EIMS_GPI_SDP2;
2555 		/* fall through */
2556 	case ixgbe_mac_X540:
2557 	case ixgbe_mac_X550:
2558 	case ixgbe_mac_X550EM_x:
2559 		mask |= IXGBE_EIMS_ECC;
2560 		mask |= IXGBE_EIMS_MAILBOX;
2561 		break;
2562 	default:
2563 		break;
2564 	}
2565 
2566 	if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2567 	    !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2568 		mask |= IXGBE_EIMS_FLOW_DIR;
2569 
2570 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2571 	if (queues)
2572 		ixgbe_irq_enable_queues(adapter, ~0);
2573 	if (flush)
2574 		IXGBE_WRITE_FLUSH(&adapter->hw);
2575 }
2576 
2577 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2578 {
2579 	struct ixgbe_adapter *adapter = data;
2580 	struct ixgbe_hw *hw = &adapter->hw;
2581 	u32 eicr;
2582 
2583 	/*
2584 	 * Workaround for Silicon errata.  Use clear-by-write instead
2585 	 * of clear-by-read.  Reading with EICS will return the
2586 	 * interrupt causes without clearing, which later be done
2587 	 * with the write to EICR.
2588 	 */
2589 	eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2590 
2591 	/* The lower 16bits of the EICR register are for the queue interrupts
2592 	 * which should be masked here in order to not accidently clear them if
2593 	 * the bits are high when ixgbe_msix_other is called. There is a race
2594 	 * condition otherwise which results in possible performance loss
2595 	 * especially if the ixgbe_msix_other interrupt is triggering
2596 	 * consistently (as it would when PPS is turned on for the X540 device)
2597 	 */
2598 	eicr &= 0xFFFF0000;
2599 
2600 	IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2601 
2602 	if (eicr & IXGBE_EICR_LSC)
2603 		ixgbe_check_lsc(adapter);
2604 
2605 	if (eicr & IXGBE_EICR_MAILBOX)
2606 		ixgbe_msg_task(adapter);
2607 
2608 	switch (hw->mac.type) {
2609 	case ixgbe_mac_82599EB:
2610 	case ixgbe_mac_X540:
2611 	case ixgbe_mac_X550:
2612 	case ixgbe_mac_X550EM_x:
2613 		if (eicr & IXGBE_EICR_ECC) {
2614 			e_info(link, "Received ECC Err, initiating reset\n");
2615 			adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
2616 			ixgbe_service_event_schedule(adapter);
2617 			IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
2618 		}
2619 		/* Handle Flow Director Full threshold interrupt */
2620 		if (eicr & IXGBE_EICR_FLOW_DIR) {
2621 			int reinit_count = 0;
2622 			int i;
2623 			for (i = 0; i < adapter->num_tx_queues; i++) {
2624 				struct ixgbe_ring *ring = adapter->tx_ring[i];
2625 				if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2626 						       &ring->state))
2627 					reinit_count++;
2628 			}
2629 			if (reinit_count) {
2630 				/* no more flow director interrupts until after init */
2631 				IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2632 				adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2633 				ixgbe_service_event_schedule(adapter);
2634 			}
2635 		}
2636 		ixgbe_check_sfp_event(adapter, eicr);
2637 		ixgbe_check_overtemp_event(adapter, eicr);
2638 		break;
2639 	default:
2640 		break;
2641 	}
2642 
2643 	ixgbe_check_fan_failure(adapter, eicr);
2644 
2645 	if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2646 		ixgbe_ptp_check_pps_event(adapter, eicr);
2647 
2648 	/* re-enable the original interrupt state, no lsc, no queues */
2649 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
2650 		ixgbe_irq_enable(adapter, false, false);
2651 
2652 	return IRQ_HANDLED;
2653 }
2654 
2655 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2656 {
2657 	struct ixgbe_q_vector *q_vector = data;
2658 
2659 	/* EIAM disabled interrupts (on this vector) for us */
2660 
2661 	if (q_vector->rx.ring || q_vector->tx.ring)
2662 		napi_schedule(&q_vector->napi);
2663 
2664 	return IRQ_HANDLED;
2665 }
2666 
2667 /**
2668  * ixgbe_poll - NAPI Rx polling callback
2669  * @napi: structure for representing this polling device
2670  * @budget: how many packets driver is allowed to clean
2671  *
2672  * This function is used for legacy and MSI, NAPI mode
2673  **/
2674 int ixgbe_poll(struct napi_struct *napi, int budget)
2675 {
2676 	struct ixgbe_q_vector *q_vector =
2677 				container_of(napi, struct ixgbe_q_vector, napi);
2678 	struct ixgbe_adapter *adapter = q_vector->adapter;
2679 	struct ixgbe_ring *ring;
2680 	int per_ring_budget;
2681 	bool clean_complete = true;
2682 
2683 #ifdef CONFIG_IXGBE_DCA
2684 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2685 		ixgbe_update_dca(q_vector);
2686 #endif
2687 
2688 	ixgbe_for_each_ring(ring, q_vector->tx)
2689 		clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
2690 
2691 	if (!ixgbe_qv_lock_napi(q_vector))
2692 		return budget;
2693 
2694 	/* attempt to distribute budget to each queue fairly, but don't allow
2695 	 * the budget to go below 1 because we'll exit polling */
2696 	if (q_vector->rx.count > 1)
2697 		per_ring_budget = max(budget/q_vector->rx.count, 1);
2698 	else
2699 		per_ring_budget = budget;
2700 
2701 	ixgbe_for_each_ring(ring, q_vector->rx)
2702 		clean_complete &= (ixgbe_clean_rx_irq(q_vector, ring,
2703 				   per_ring_budget) < per_ring_budget);
2704 
2705 	ixgbe_qv_unlock_napi(q_vector);
2706 	/* If all work not completed, return budget and keep polling */
2707 	if (!clean_complete)
2708 		return budget;
2709 
2710 	/* all work done, exit the polling mode */
2711 	napi_complete(napi);
2712 	if (adapter->rx_itr_setting & 1)
2713 		ixgbe_set_itr(q_vector);
2714 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
2715 		ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2716 
2717 	return 0;
2718 }
2719 
2720 /**
2721  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2722  * @adapter: board private structure
2723  *
2724  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2725  * interrupts from the kernel.
2726  **/
2727 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2728 {
2729 	struct net_device *netdev = adapter->netdev;
2730 	int vector, err;
2731 	int ri = 0, ti = 0;
2732 
2733 	for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2734 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2735 		struct msix_entry *entry = &adapter->msix_entries[vector];
2736 
2737 		if (q_vector->tx.ring && q_vector->rx.ring) {
2738 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2739 				 "%s-%s-%d", netdev->name, "TxRx", ri++);
2740 			ti++;
2741 		} else if (q_vector->rx.ring) {
2742 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2743 				 "%s-%s-%d", netdev->name, "rx", ri++);
2744 		} else if (q_vector->tx.ring) {
2745 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2746 				 "%s-%s-%d", netdev->name, "tx", ti++);
2747 		} else {
2748 			/* skip this unused q_vector */
2749 			continue;
2750 		}
2751 		err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2752 				  q_vector->name, q_vector);
2753 		if (err) {
2754 			e_err(probe, "request_irq failed for MSIX interrupt "
2755 			      "Error: %d\n", err);
2756 			goto free_queue_irqs;
2757 		}
2758 		/* If Flow Director is enabled, set interrupt affinity */
2759 		if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2760 			/* assign the mask for this irq */
2761 			irq_set_affinity_hint(entry->vector,
2762 					      &q_vector->affinity_mask);
2763 		}
2764 	}
2765 
2766 	err = request_irq(adapter->msix_entries[vector].vector,
2767 			  ixgbe_msix_other, 0, netdev->name, adapter);
2768 	if (err) {
2769 		e_err(probe, "request_irq for msix_other failed: %d\n", err);
2770 		goto free_queue_irqs;
2771 	}
2772 
2773 	return 0;
2774 
2775 free_queue_irqs:
2776 	while (vector) {
2777 		vector--;
2778 		irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2779 				      NULL);
2780 		free_irq(adapter->msix_entries[vector].vector,
2781 			 adapter->q_vector[vector]);
2782 	}
2783 	adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2784 	pci_disable_msix(adapter->pdev);
2785 	kfree(adapter->msix_entries);
2786 	adapter->msix_entries = NULL;
2787 	return err;
2788 }
2789 
2790 /**
2791  * ixgbe_intr - legacy mode Interrupt Handler
2792  * @irq: interrupt number
2793  * @data: pointer to a network interface device structure
2794  **/
2795 static irqreturn_t ixgbe_intr(int irq, void *data)
2796 {
2797 	struct ixgbe_adapter *adapter = data;
2798 	struct ixgbe_hw *hw = &adapter->hw;
2799 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2800 	u32 eicr;
2801 
2802 	/*
2803 	 * Workaround for silicon errata #26 on 82598.  Mask the interrupt
2804 	 * before the read of EICR.
2805 	 */
2806 	IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2807 
2808 	/* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2809 	 * therefore no explicit interrupt disable is necessary */
2810 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2811 	if (!eicr) {
2812 		/*
2813 		 * shared interrupt alert!
2814 		 * make sure interrupts are enabled because the read will
2815 		 * have disabled interrupts due to EIAM
2816 		 * finish the workaround of silicon errata on 82598.  Unmask
2817 		 * the interrupt that we masked before the EICR read.
2818 		 */
2819 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
2820 			ixgbe_irq_enable(adapter, true, true);
2821 		return IRQ_NONE;	/* Not our interrupt */
2822 	}
2823 
2824 	if (eicr & IXGBE_EICR_LSC)
2825 		ixgbe_check_lsc(adapter);
2826 
2827 	switch (hw->mac.type) {
2828 	case ixgbe_mac_82599EB:
2829 		ixgbe_check_sfp_event(adapter, eicr);
2830 		/* Fall through */
2831 	case ixgbe_mac_X540:
2832 	case ixgbe_mac_X550:
2833 	case ixgbe_mac_X550EM_x:
2834 		if (eicr & IXGBE_EICR_ECC) {
2835 			e_info(link, "Received ECC Err, initiating reset\n");
2836 			adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
2837 			ixgbe_service_event_schedule(adapter);
2838 			IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
2839 		}
2840 		ixgbe_check_overtemp_event(adapter, eicr);
2841 		break;
2842 	default:
2843 		break;
2844 	}
2845 
2846 	ixgbe_check_fan_failure(adapter, eicr);
2847 	if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2848 		ixgbe_ptp_check_pps_event(adapter, eicr);
2849 
2850 	/* would disable interrupts here but EIAM disabled it */
2851 	napi_schedule(&q_vector->napi);
2852 
2853 	/*
2854 	 * re-enable link(maybe) and non-queue interrupts, no flush.
2855 	 * ixgbe_poll will re-enable the queue interrupts
2856 	 */
2857 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
2858 		ixgbe_irq_enable(adapter, false, false);
2859 
2860 	return IRQ_HANDLED;
2861 }
2862 
2863 /**
2864  * ixgbe_request_irq - initialize interrupts
2865  * @adapter: board private structure
2866  *
2867  * Attempts to configure interrupts using the best available
2868  * capabilities of the hardware and kernel.
2869  **/
2870 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2871 {
2872 	struct net_device *netdev = adapter->netdev;
2873 	int err;
2874 
2875 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2876 		err = ixgbe_request_msix_irqs(adapter);
2877 	else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2878 		err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2879 				  netdev->name, adapter);
2880 	else
2881 		err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2882 				  netdev->name, adapter);
2883 
2884 	if (err)
2885 		e_err(probe, "request_irq failed, Error %d\n", err);
2886 
2887 	return err;
2888 }
2889 
2890 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2891 {
2892 	int vector;
2893 
2894 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2895 		free_irq(adapter->pdev->irq, adapter);
2896 		return;
2897 	}
2898 
2899 	for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2900 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2901 		struct msix_entry *entry = &adapter->msix_entries[vector];
2902 
2903 		/* free only the irqs that were actually requested */
2904 		if (!q_vector->rx.ring && !q_vector->tx.ring)
2905 			continue;
2906 
2907 		/* clear the affinity_mask in the IRQ descriptor */
2908 		irq_set_affinity_hint(entry->vector, NULL);
2909 
2910 		free_irq(entry->vector, q_vector);
2911 	}
2912 
2913 	free_irq(adapter->msix_entries[vector++].vector, adapter);
2914 }
2915 
2916 /**
2917  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2918  * @adapter: board private structure
2919  **/
2920 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2921 {
2922 	switch (adapter->hw.mac.type) {
2923 	case ixgbe_mac_82598EB:
2924 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2925 		break;
2926 	case ixgbe_mac_82599EB:
2927 	case ixgbe_mac_X540:
2928 	case ixgbe_mac_X550:
2929 	case ixgbe_mac_X550EM_x:
2930 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2931 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2932 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2933 		break;
2934 	default:
2935 		break;
2936 	}
2937 	IXGBE_WRITE_FLUSH(&adapter->hw);
2938 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2939 		int vector;
2940 
2941 		for (vector = 0; vector < adapter->num_q_vectors; vector++)
2942 			synchronize_irq(adapter->msix_entries[vector].vector);
2943 
2944 		synchronize_irq(adapter->msix_entries[vector++].vector);
2945 	} else {
2946 		synchronize_irq(adapter->pdev->irq);
2947 	}
2948 }
2949 
2950 /**
2951  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2952  *
2953  **/
2954 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2955 {
2956 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2957 
2958 	ixgbe_write_eitr(q_vector);
2959 
2960 	ixgbe_set_ivar(adapter, 0, 0, 0);
2961 	ixgbe_set_ivar(adapter, 1, 0, 0);
2962 
2963 	e_info(hw, "Legacy interrupt IVAR setup done\n");
2964 }
2965 
2966 /**
2967  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2968  * @adapter: board private structure
2969  * @ring: structure containing ring specific data
2970  *
2971  * Configure the Tx descriptor ring after a reset.
2972  **/
2973 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2974 			     struct ixgbe_ring *ring)
2975 {
2976 	struct ixgbe_hw *hw = &adapter->hw;
2977 	u64 tdba = ring->dma;
2978 	int wait_loop = 10;
2979 	u32 txdctl = IXGBE_TXDCTL_ENABLE;
2980 	u8 reg_idx = ring->reg_idx;
2981 
2982 	/* disable queue to avoid issues while updating state */
2983 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
2984 	IXGBE_WRITE_FLUSH(hw);
2985 
2986 	IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2987 			(tdba & DMA_BIT_MASK(32)));
2988 	IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2989 	IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2990 			ring->count * sizeof(union ixgbe_adv_tx_desc));
2991 	IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2992 	IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2993 	ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
2994 
2995 	/*
2996 	 * set WTHRESH to encourage burst writeback, it should not be set
2997 	 * higher than 1 when:
2998 	 * - ITR is 0 as it could cause false TX hangs
2999 	 * - ITR is set to > 100k int/sec and BQL is enabled
3000 	 *
3001 	 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3002 	 * to or less than the number of on chip descriptors, which is
3003 	 * currently 40.
3004 	 */
3005 	if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3006 		txdctl |= (1 << 16);	/* WTHRESH = 1 */
3007 	else
3008 		txdctl |= (8 << 16);	/* WTHRESH = 8 */
3009 
3010 	/*
3011 	 * Setting PTHRESH to 32 both improves performance
3012 	 * and avoids a TX hang with DFP enabled
3013 	 */
3014 	txdctl |= (1 << 8) |	/* HTHRESH = 1 */
3015 		   32;		/* PTHRESH = 32 */
3016 
3017 	/* reinitialize flowdirector state */
3018 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3019 		ring->atr_sample_rate = adapter->atr_sample_rate;
3020 		ring->atr_count = 0;
3021 		set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3022 	} else {
3023 		ring->atr_sample_rate = 0;
3024 	}
3025 
3026 	/* initialize XPS */
3027 	if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3028 		struct ixgbe_q_vector *q_vector = ring->q_vector;
3029 
3030 		if (q_vector)
3031 			netif_set_xps_queue(ring->netdev,
3032 					    &q_vector->affinity_mask,
3033 					    ring->queue_index);
3034 	}
3035 
3036 	clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3037 
3038 	/* enable queue */
3039 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3040 
3041 	/* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3042 	if (hw->mac.type == ixgbe_mac_82598EB &&
3043 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3044 		return;
3045 
3046 	/* poll to verify queue is enabled */
3047 	do {
3048 		usleep_range(1000, 2000);
3049 		txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3050 	} while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3051 	if (!wait_loop)
3052 		e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
3053 }
3054 
3055 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3056 {
3057 	struct ixgbe_hw *hw = &adapter->hw;
3058 	u32 rttdcs, mtqc;
3059 	u8 tcs = netdev_get_num_tc(adapter->netdev);
3060 
3061 	if (hw->mac.type == ixgbe_mac_82598EB)
3062 		return;
3063 
3064 	/* disable the arbiter while setting MTQC */
3065 	rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3066 	rttdcs |= IXGBE_RTTDCS_ARBDIS;
3067 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3068 
3069 	/* set transmit pool layout */
3070 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3071 		mtqc = IXGBE_MTQC_VT_ENA;
3072 		if (tcs > 4)
3073 			mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3074 		else if (tcs > 1)
3075 			mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3076 		else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3077 			mtqc |= IXGBE_MTQC_32VF;
3078 		else
3079 			mtqc |= IXGBE_MTQC_64VF;
3080 	} else {
3081 		if (tcs > 4)
3082 			mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3083 		else if (tcs > 1)
3084 			mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3085 		else
3086 			mtqc = IXGBE_MTQC_64Q_1PB;
3087 	}
3088 
3089 	IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3090 
3091 	/* Enable Security TX Buffer IFG for multiple pb */
3092 	if (tcs) {
3093 		u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3094 		sectx |= IXGBE_SECTX_DCB;
3095 		IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3096 	}
3097 
3098 	/* re-enable the arbiter */
3099 	rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3100 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3101 }
3102 
3103 /**
3104  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3105  * @adapter: board private structure
3106  *
3107  * Configure the Tx unit of the MAC after a reset.
3108  **/
3109 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3110 {
3111 	struct ixgbe_hw *hw = &adapter->hw;
3112 	u32 dmatxctl;
3113 	u32 i;
3114 
3115 	ixgbe_setup_mtqc(adapter);
3116 
3117 	if (hw->mac.type != ixgbe_mac_82598EB) {
3118 		/* DMATXCTL.EN must be before Tx queues are enabled */
3119 		dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3120 		dmatxctl |= IXGBE_DMATXCTL_TE;
3121 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3122 	}
3123 
3124 	/* Setup the HW Tx Head and Tail descriptor pointers */
3125 	for (i = 0; i < adapter->num_tx_queues; i++)
3126 		ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3127 }
3128 
3129 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3130 				 struct ixgbe_ring *ring)
3131 {
3132 	struct ixgbe_hw *hw = &adapter->hw;
3133 	u8 reg_idx = ring->reg_idx;
3134 	u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3135 
3136 	srrctl |= IXGBE_SRRCTL_DROP_EN;
3137 
3138 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3139 }
3140 
3141 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3142 				  struct ixgbe_ring *ring)
3143 {
3144 	struct ixgbe_hw *hw = &adapter->hw;
3145 	u8 reg_idx = ring->reg_idx;
3146 	u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3147 
3148 	srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3149 
3150 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3151 }
3152 
3153 #ifdef CONFIG_IXGBE_DCB
3154 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3155 #else
3156 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3157 #endif
3158 {
3159 	int i;
3160 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3161 
3162 	if (adapter->ixgbe_ieee_pfc)
3163 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3164 
3165 	/*
3166 	 * We should set the drop enable bit if:
3167 	 *  SR-IOV is enabled
3168 	 *   or
3169 	 *  Number of Rx queues > 1 and flow control is disabled
3170 	 *
3171 	 *  This allows us to avoid head of line blocking for security
3172 	 *  and performance reasons.
3173 	 */
3174 	if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3175 	    !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3176 		for (i = 0; i < adapter->num_rx_queues; i++)
3177 			ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3178 	} else {
3179 		for (i = 0; i < adapter->num_rx_queues; i++)
3180 			ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3181 	}
3182 }
3183 
3184 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3185 
3186 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3187 				   struct ixgbe_ring *rx_ring)
3188 {
3189 	struct ixgbe_hw *hw = &adapter->hw;
3190 	u32 srrctl;
3191 	u8 reg_idx = rx_ring->reg_idx;
3192 
3193 	if (hw->mac.type == ixgbe_mac_82598EB) {
3194 		u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3195 
3196 		/*
3197 		 * if VMDq is not active we must program one srrctl register
3198 		 * per RSS queue since we have enabled RDRXCTL.MVMEN
3199 		 */
3200 		reg_idx &= mask;
3201 	}
3202 
3203 	/* configure header buffer length, needed for RSC */
3204 	srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3205 
3206 	/* configure the packet buffer length */
3207 	srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3208 
3209 	/* configure descriptor type */
3210 	srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3211 
3212 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3213 }
3214 
3215 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter, const u32 *seed)
3216 {
3217 	struct ixgbe_hw *hw = &adapter->hw;
3218 	u32 reta = 0;
3219 	int i, j;
3220 	int reta_entries = 128;
3221 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3222 	int indices_multi;
3223 
3224 	/*
3225 	 * Program table for at least 2 queues w/ SR-IOV so that VFs can
3226 	 * make full use of any rings they may have.  We will use the
3227 	 * PSRTYPE register to control how many rings we use within the PF.
3228 	 */
3229 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 2))
3230 		rss_i = 2;
3231 
3232 	/* Fill out hash function seeds */
3233 	for (i = 0; i < 10; i++)
3234 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
3235 
3236 	/* Fill out the redirection table as follows:
3237 	 * 82598: 128 (8 bit wide) entries containing pair of 4 bit RSS indices
3238 	 * 82599/X540: 128 (8 bit wide) entries containing 4 bit RSS index
3239 	 * X550: 512 (8 bit wide) entries containing 6 bit RSS index
3240 	 */
3241 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3242 		indices_multi = 0x11;
3243 	else
3244 		indices_multi = 0x1;
3245 
3246 	switch (adapter->hw.mac.type) {
3247 	case ixgbe_mac_X550:
3248 	case ixgbe_mac_X550EM_x:
3249 		if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3250 			reta_entries = 512;
3251 	default:
3252 		break;
3253 	}
3254 
3255 	/* Fill out redirection table */
3256 	for (i = 0, j = 0; i < reta_entries; i++, j++) {
3257 		if (j == rss_i)
3258 			j = 0;
3259 		reta = (reta << 8) | (j * indices_multi);
3260 		if ((i & 3) == 3) {
3261 			if (i < 128)
3262 				IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3263 			else
3264 				IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3265 						reta);
3266 		}
3267 	}
3268 }
3269 
3270 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter, const u32 *seed)
3271 {
3272 	struct ixgbe_hw *hw = &adapter->hw;
3273 	u32 vfreta = 0;
3274 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3275 	unsigned int pf_pool = adapter->num_vfs;
3276 	int i, j;
3277 
3278 	/* Fill out hash function seeds */
3279 	for (i = 0; i < 10; i++)
3280 		IXGBE_WRITE_REG(hw, IXGBE_PFVFRSSRK(i, pf_pool), seed[i]);
3281 
3282 	/* Fill out the redirection table */
3283 	for (i = 0, j = 0; i < 64; i++, j++) {
3284 		if (j == rss_i)
3285 			j = 0;
3286 		vfreta = (vfreta << 8) | j;
3287 		if ((i & 3) == 3)
3288 			IXGBE_WRITE_REG(hw, IXGBE_PFVFRETA(i >> 2, pf_pool),
3289 					vfreta);
3290 	}
3291 }
3292 
3293 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3294 {
3295 	struct ixgbe_hw *hw = &adapter->hw;
3296 	u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3297 	u32 rss_key[10];
3298 	u32 rxcsum;
3299 
3300 	/* Disable indicating checksum in descriptor, enables RSS hash */
3301 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3302 	rxcsum |= IXGBE_RXCSUM_PCSD;
3303 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3304 
3305 	if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3306 		if (adapter->ring_feature[RING_F_RSS].mask)
3307 			mrqc = IXGBE_MRQC_RSSEN;
3308 	} else {
3309 		u8 tcs = netdev_get_num_tc(adapter->netdev);
3310 
3311 		if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3312 			if (tcs > 4)
3313 				mrqc = IXGBE_MRQC_VMDQRT8TCEN;	/* 8 TCs */
3314 			else if (tcs > 1)
3315 				mrqc = IXGBE_MRQC_VMDQRT4TCEN;	/* 4 TCs */
3316 			else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3317 				mrqc = IXGBE_MRQC_VMDQRSS32EN;
3318 			else
3319 				mrqc = IXGBE_MRQC_VMDQRSS64EN;
3320 		} else {
3321 			if (tcs > 4)
3322 				mrqc = IXGBE_MRQC_RTRSS8TCEN;
3323 			else if (tcs > 1)
3324 				mrqc = IXGBE_MRQC_RTRSS4TCEN;
3325 			else
3326 				mrqc = IXGBE_MRQC_RSSEN;
3327 		}
3328 	}
3329 
3330 	/* Perform hash on these packet types */
3331 	rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3332 		     IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3333 		     IXGBE_MRQC_RSS_FIELD_IPV6 |
3334 		     IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3335 
3336 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3337 		rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3338 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3339 		rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3340 
3341 	netdev_rss_key_fill(rss_key, sizeof(rss_key));
3342 	if ((hw->mac.type >= ixgbe_mac_X550) &&
3343 	    (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
3344 		unsigned int pf_pool = adapter->num_vfs;
3345 
3346 		/* Enable VF RSS mode */
3347 		mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
3348 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3349 
3350 		/* Setup RSS through the VF registers */
3351 		ixgbe_setup_vfreta(adapter, rss_key);
3352 		vfmrqc = IXGBE_MRQC_RSSEN;
3353 		vfmrqc |= rss_field;
3354 		IXGBE_WRITE_REG(hw, IXGBE_PFVFMRQC(pf_pool), vfmrqc);
3355 	} else {
3356 		ixgbe_setup_reta(adapter, rss_key);
3357 		mrqc |= rss_field;
3358 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3359 	}
3360 }
3361 
3362 /**
3363  * ixgbe_configure_rscctl - enable RSC for the indicated ring
3364  * @adapter:    address of board private structure
3365  * @index:      index of ring to set
3366  **/
3367 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
3368 				   struct ixgbe_ring *ring)
3369 {
3370 	struct ixgbe_hw *hw = &adapter->hw;
3371 	u32 rscctrl;
3372 	u8 reg_idx = ring->reg_idx;
3373 
3374 	if (!ring_is_rsc_enabled(ring))
3375 		return;
3376 
3377 	rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
3378 	rscctrl |= IXGBE_RSCCTL_RSCEN;
3379 	/*
3380 	 * we must limit the number of descriptors so that the
3381 	 * total size of max desc * buf_len is not greater
3382 	 * than 65536
3383 	 */
3384 	rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
3385 	IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
3386 }
3387 
3388 #define IXGBE_MAX_RX_DESC_POLL 10
3389 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3390 				       struct ixgbe_ring *ring)
3391 {
3392 	struct ixgbe_hw *hw = &adapter->hw;
3393 	int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3394 	u32 rxdctl;
3395 	u8 reg_idx = ring->reg_idx;
3396 
3397 	if (ixgbe_removed(hw->hw_addr))
3398 		return;
3399 	/* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3400 	if (hw->mac.type == ixgbe_mac_82598EB &&
3401 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3402 		return;
3403 
3404 	do {
3405 		usleep_range(1000, 2000);
3406 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3407 	} while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3408 
3409 	if (!wait_loop) {
3410 		e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3411 		      "the polling period\n", reg_idx);
3412 	}
3413 }
3414 
3415 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3416 			    struct ixgbe_ring *ring)
3417 {
3418 	struct ixgbe_hw *hw = &adapter->hw;
3419 	int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3420 	u32 rxdctl;
3421 	u8 reg_idx = ring->reg_idx;
3422 
3423 	if (ixgbe_removed(hw->hw_addr))
3424 		return;
3425 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3426 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3427 
3428 	/* write value back with RXDCTL.ENABLE bit cleared */
3429 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3430 
3431 	if (hw->mac.type == ixgbe_mac_82598EB &&
3432 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3433 		return;
3434 
3435 	/* the hardware may take up to 100us to really disable the rx queue */
3436 	do {
3437 		udelay(10);
3438 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3439 	} while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3440 
3441 	if (!wait_loop) {
3442 		e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3443 		      "the polling period\n", reg_idx);
3444 	}
3445 }
3446 
3447 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3448 			     struct ixgbe_ring *ring)
3449 {
3450 	struct ixgbe_hw *hw = &adapter->hw;
3451 	u64 rdba = ring->dma;
3452 	u32 rxdctl;
3453 	u8 reg_idx = ring->reg_idx;
3454 
3455 	/* disable queue to avoid issues while updating state */
3456 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3457 	ixgbe_disable_rx_queue(adapter, ring);
3458 
3459 	IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3460 	IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3461 	IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3462 			ring->count * sizeof(union ixgbe_adv_rx_desc));
3463 	IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3464 	IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3465 	ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
3466 
3467 	ixgbe_configure_srrctl(adapter, ring);
3468 	ixgbe_configure_rscctl(adapter, ring);
3469 
3470 	if (hw->mac.type == ixgbe_mac_82598EB) {
3471 		/*
3472 		 * enable cache line friendly hardware writes:
3473 		 * PTHRESH=32 descriptors (half the internal cache),
3474 		 * this also removes ugly rx_no_buffer_count increment
3475 		 * HTHRESH=4 descriptors (to minimize latency on fetch)
3476 		 * WTHRESH=8 burst writeback up to two cache lines
3477 		 */
3478 		rxdctl &= ~0x3FFFFF;
3479 		rxdctl |=  0x080420;
3480 	}
3481 
3482 	/* enable receive descriptor ring */
3483 	rxdctl |= IXGBE_RXDCTL_ENABLE;
3484 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3485 
3486 	ixgbe_rx_desc_queue_enable(adapter, ring);
3487 	ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3488 }
3489 
3490 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3491 {
3492 	struct ixgbe_hw *hw = &adapter->hw;
3493 	int rss_i = adapter->ring_feature[RING_F_RSS].indices;
3494 	u16 pool;
3495 
3496 	/* PSRTYPE must be initialized in non 82598 adapters */
3497 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3498 		      IXGBE_PSRTYPE_UDPHDR |
3499 		      IXGBE_PSRTYPE_IPV4HDR |
3500 		      IXGBE_PSRTYPE_L2HDR |
3501 		      IXGBE_PSRTYPE_IPV6HDR;
3502 
3503 	if (hw->mac.type == ixgbe_mac_82598EB)
3504 		return;
3505 
3506 	if (rss_i > 3)
3507 		psrtype |= 2 << 29;
3508 	else if (rss_i > 1)
3509 		psrtype |= 1 << 29;
3510 
3511 	for_each_set_bit(pool, &adapter->fwd_bitmask, 32)
3512 		IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
3513 }
3514 
3515 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3516 {
3517 	struct ixgbe_hw *hw = &adapter->hw;
3518 	u32 reg_offset, vf_shift;
3519 	u32 gcr_ext, vmdctl;
3520 	int i;
3521 
3522 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3523 		return;
3524 
3525 	vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3526 	vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
3527 	vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
3528 	vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
3529 	vmdctl |= IXGBE_VT_CTL_REPLEN;
3530 	IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
3531 
3532 	vf_shift = VMDQ_P(0) % 32;
3533 	reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
3534 
3535 	/* Enable only the PF's pool for Tx/Rx */
3536 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (~0) << vf_shift);
3537 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
3538 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (~0) << vf_shift);
3539 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
3540 	if (adapter->flags2 & IXGBE_FLAG2_BRIDGE_MODE_VEB)
3541 		IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3542 
3543 	/* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3544 	hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
3545 
3546 	/*
3547 	 * Set up VF register offsets for selected VT Mode,
3548 	 * i.e. 32 or 64 VFs for SR-IOV
3549 	 */
3550 	switch (adapter->ring_feature[RING_F_VMDQ].mask) {
3551 	case IXGBE_82599_VMDQ_8Q_MASK:
3552 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
3553 		break;
3554 	case IXGBE_82599_VMDQ_4Q_MASK:
3555 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
3556 		break;
3557 	default:
3558 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
3559 		break;
3560 	}
3561 
3562 	IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3563 
3564 
3565 	/* Enable MAC Anti-Spoofing */
3566 	hw->mac.ops.set_mac_anti_spoofing(hw, (adapter->num_vfs != 0),
3567 					  adapter->num_vfs);
3568 	/* For VFs that have spoof checking turned off */
3569 	for (i = 0; i < adapter->num_vfs; i++) {
3570 		if (!adapter->vfinfo[i].spoofchk_enabled)
3571 			ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
3572 	}
3573 }
3574 
3575 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3576 {
3577 	struct ixgbe_hw *hw = &adapter->hw;
3578 	struct net_device *netdev = adapter->netdev;
3579 	int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3580 	struct ixgbe_ring *rx_ring;
3581 	int i;
3582 	u32 mhadd, hlreg0;
3583 
3584 #ifdef IXGBE_FCOE
3585 	/* adjust max frame to be able to do baby jumbo for FCoE */
3586 	if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3587 	    (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3588 		max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3589 
3590 #endif /* IXGBE_FCOE */
3591 
3592 	/* adjust max frame to be at least the size of a standard frame */
3593 	if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
3594 		max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
3595 
3596 	mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3597 	if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3598 		mhadd &= ~IXGBE_MHADD_MFS_MASK;
3599 		mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3600 
3601 		IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3602 	}
3603 
3604 	hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3605 	/* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3606 	hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3607 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3608 
3609 	/*
3610 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
3611 	 * the Base and Length of the Rx Descriptor Ring
3612 	 */
3613 	for (i = 0; i < adapter->num_rx_queues; i++) {
3614 		rx_ring = adapter->rx_ring[i];
3615 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3616 			set_ring_rsc_enabled(rx_ring);
3617 		else
3618 			clear_ring_rsc_enabled(rx_ring);
3619 	}
3620 }
3621 
3622 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3623 {
3624 	struct ixgbe_hw *hw = &adapter->hw;
3625 	u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3626 
3627 	switch (hw->mac.type) {
3628 	case ixgbe_mac_X550:
3629 	case ixgbe_mac_X550EM_x:
3630 	case ixgbe_mac_82598EB:
3631 		/*
3632 		 * For VMDq support of different descriptor types or
3633 		 * buffer sizes through the use of multiple SRRCTL
3634 		 * registers, RDRXCTL.MVMEN must be set to 1
3635 		 *
3636 		 * also, the manual doesn't mention it clearly but DCA hints
3637 		 * will only use queue 0's tags unless this bit is set.  Side
3638 		 * effects of setting this bit are only that SRRCTL must be
3639 		 * fully programmed [0..15]
3640 		 */
3641 		rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3642 		break;
3643 	case ixgbe_mac_82599EB:
3644 	case ixgbe_mac_X540:
3645 		/* Disable RSC for ACK packets */
3646 		IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3647 		   (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3648 		rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3649 		/* hardware requires some bits to be set by default */
3650 		rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3651 		rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3652 		break;
3653 	default:
3654 		/* We should do nothing since we don't know this hardware */
3655 		return;
3656 	}
3657 
3658 	IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3659 }
3660 
3661 /**
3662  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3663  * @adapter: board private structure
3664  *
3665  * Configure the Rx unit of the MAC after a reset.
3666  **/
3667 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3668 {
3669 	struct ixgbe_hw *hw = &adapter->hw;
3670 	int i;
3671 	u32 rxctrl, rfctl;
3672 
3673 	/* disable receives while setting up the descriptors */
3674 	rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3675 	IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3676 
3677 	ixgbe_setup_psrtype(adapter);
3678 	ixgbe_setup_rdrxctl(adapter);
3679 
3680 	/* RSC Setup */
3681 	rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
3682 	rfctl &= ~IXGBE_RFCTL_RSC_DIS;
3683 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
3684 		rfctl |= IXGBE_RFCTL_RSC_DIS;
3685 	IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
3686 
3687 	/* Program registers for the distribution of queues */
3688 	ixgbe_setup_mrqc(adapter);
3689 
3690 	/* set_rx_buffer_len must be called before ring initialization */
3691 	ixgbe_set_rx_buffer_len(adapter);
3692 
3693 	/*
3694 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
3695 	 * the Base and Length of the Rx Descriptor Ring
3696 	 */
3697 	for (i = 0; i < adapter->num_rx_queues; i++)
3698 		ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3699 
3700 	/* disable drop enable for 82598 parts */
3701 	if (hw->mac.type == ixgbe_mac_82598EB)
3702 		rxctrl |= IXGBE_RXCTRL_DMBYPS;
3703 
3704 	/* enable all receives */
3705 	rxctrl |= IXGBE_RXCTRL_RXEN;
3706 	hw->mac.ops.enable_rx_dma(hw, rxctrl);
3707 }
3708 
3709 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
3710 				 __be16 proto, u16 vid)
3711 {
3712 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3713 	struct ixgbe_hw *hw = &adapter->hw;
3714 
3715 	/* add VID to filter table */
3716 	hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true);
3717 	set_bit(vid, adapter->active_vlans);
3718 
3719 	return 0;
3720 }
3721 
3722 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
3723 				  __be16 proto, u16 vid)
3724 {
3725 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3726 	struct ixgbe_hw *hw = &adapter->hw;
3727 
3728 	/* remove VID from filter table */
3729 	hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), false);
3730 	clear_bit(vid, adapter->active_vlans);
3731 
3732 	return 0;
3733 }
3734 
3735 /**
3736  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3737  * @adapter: driver data
3738  */
3739 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3740 {
3741 	struct ixgbe_hw *hw = &adapter->hw;
3742 	u32 vlnctrl;
3743 	int i, j;
3744 
3745 	switch (hw->mac.type) {
3746 	case ixgbe_mac_82598EB:
3747 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3748 		vlnctrl &= ~IXGBE_VLNCTRL_VME;
3749 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3750 		break;
3751 	case ixgbe_mac_82599EB:
3752 	case ixgbe_mac_X540:
3753 	case ixgbe_mac_X550:
3754 	case ixgbe_mac_X550EM_x:
3755 		for (i = 0; i < adapter->num_rx_queues; i++) {
3756 			struct ixgbe_ring *ring = adapter->rx_ring[i];
3757 
3758 			if (ring->l2_accel_priv)
3759 				continue;
3760 			j = ring->reg_idx;
3761 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3762 			vlnctrl &= ~IXGBE_RXDCTL_VME;
3763 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3764 		}
3765 		break;
3766 	default:
3767 		break;
3768 	}
3769 }
3770 
3771 /**
3772  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3773  * @adapter: driver data
3774  */
3775 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3776 {
3777 	struct ixgbe_hw *hw = &adapter->hw;
3778 	u32 vlnctrl;
3779 	int i, j;
3780 
3781 	switch (hw->mac.type) {
3782 	case ixgbe_mac_82598EB:
3783 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3784 		vlnctrl |= IXGBE_VLNCTRL_VME;
3785 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3786 		break;
3787 	case ixgbe_mac_82599EB:
3788 	case ixgbe_mac_X540:
3789 	case ixgbe_mac_X550:
3790 	case ixgbe_mac_X550EM_x:
3791 		for (i = 0; i < adapter->num_rx_queues; i++) {
3792 			struct ixgbe_ring *ring = adapter->rx_ring[i];
3793 
3794 			if (ring->l2_accel_priv)
3795 				continue;
3796 			j = ring->reg_idx;
3797 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3798 			vlnctrl |= IXGBE_RXDCTL_VME;
3799 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3800 		}
3801 		break;
3802 	default:
3803 		break;
3804 	}
3805 }
3806 
3807 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3808 {
3809 	u16 vid;
3810 
3811 	ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
3812 
3813 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3814 		ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
3815 }
3816 
3817 /**
3818  * ixgbe_write_mc_addr_list - write multicast addresses to MTA
3819  * @netdev: network interface device structure
3820  *
3821  * Writes multicast address list to the MTA hash table.
3822  * Returns: -ENOMEM on failure
3823  *                0 on no addresses written
3824  *                X on writing X addresses to MTA
3825  **/
3826 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
3827 {
3828 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3829 	struct ixgbe_hw *hw = &adapter->hw;
3830 
3831 	if (!netif_running(netdev))
3832 		return 0;
3833 
3834 	if (hw->mac.ops.update_mc_addr_list)
3835 		hw->mac.ops.update_mc_addr_list(hw, netdev);
3836 	else
3837 		return -ENOMEM;
3838 
3839 #ifdef CONFIG_PCI_IOV
3840 	ixgbe_restore_vf_multicasts(adapter);
3841 #endif
3842 
3843 	return netdev_mc_count(netdev);
3844 }
3845 
3846 #ifdef CONFIG_PCI_IOV
3847 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
3848 {
3849 	struct ixgbe_hw *hw = &adapter->hw;
3850 	int i;
3851 	for (i = 0; i < hw->mac.num_rar_entries; i++) {
3852 		if (adapter->mac_table[i].state & IXGBE_MAC_STATE_IN_USE)
3853 			hw->mac.ops.set_rar(hw, i, adapter->mac_table[i].addr,
3854 					    adapter->mac_table[i].queue,
3855 					    IXGBE_RAH_AV);
3856 		else
3857 			hw->mac.ops.clear_rar(hw, i);
3858 
3859 		adapter->mac_table[i].state &= ~(IXGBE_MAC_STATE_MODIFIED);
3860 	}
3861 }
3862 #endif
3863 
3864 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
3865 {
3866 	struct ixgbe_hw *hw = &adapter->hw;
3867 	int i;
3868 	for (i = 0; i < hw->mac.num_rar_entries; i++) {
3869 		if (adapter->mac_table[i].state & IXGBE_MAC_STATE_MODIFIED) {
3870 			if (adapter->mac_table[i].state &
3871 			    IXGBE_MAC_STATE_IN_USE)
3872 				hw->mac.ops.set_rar(hw, i,
3873 						adapter->mac_table[i].addr,
3874 						adapter->mac_table[i].queue,
3875 						IXGBE_RAH_AV);
3876 			else
3877 				hw->mac.ops.clear_rar(hw, i);
3878 
3879 			adapter->mac_table[i].state &=
3880 						~(IXGBE_MAC_STATE_MODIFIED);
3881 		}
3882 	}
3883 }
3884 
3885 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
3886 {
3887 	int i;
3888 	struct ixgbe_hw *hw = &adapter->hw;
3889 
3890 	for (i = 0; i < hw->mac.num_rar_entries; i++) {
3891 		adapter->mac_table[i].state |= IXGBE_MAC_STATE_MODIFIED;
3892 		adapter->mac_table[i].state &= ~IXGBE_MAC_STATE_IN_USE;
3893 		memset(adapter->mac_table[i].addr, 0, ETH_ALEN);
3894 		adapter->mac_table[i].queue = 0;
3895 	}
3896 	ixgbe_sync_mac_table(adapter);
3897 }
3898 
3899 static int ixgbe_available_rars(struct ixgbe_adapter *adapter)
3900 {
3901 	struct ixgbe_hw *hw = &adapter->hw;
3902 	int i, count = 0;
3903 
3904 	for (i = 0; i < hw->mac.num_rar_entries; i++) {
3905 		if (adapter->mac_table[i].state == 0)
3906 			count++;
3907 	}
3908 	return count;
3909 }
3910 
3911 /* this function destroys the first RAR entry */
3912 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter,
3913 					 u8 *addr)
3914 {
3915 	struct ixgbe_hw *hw = &adapter->hw;
3916 
3917 	memcpy(&adapter->mac_table[0].addr, addr, ETH_ALEN);
3918 	adapter->mac_table[0].queue = VMDQ_P(0);
3919 	adapter->mac_table[0].state = (IXGBE_MAC_STATE_DEFAULT |
3920 				       IXGBE_MAC_STATE_IN_USE);
3921 	hw->mac.ops.set_rar(hw, 0, adapter->mac_table[0].addr,
3922 			    adapter->mac_table[0].queue,
3923 			    IXGBE_RAH_AV);
3924 }
3925 
3926 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter, u8 *addr, u16 queue)
3927 {
3928 	struct ixgbe_hw *hw = &adapter->hw;
3929 	int i;
3930 
3931 	if (is_zero_ether_addr(addr))
3932 		return -EINVAL;
3933 
3934 	for (i = 0; i < hw->mac.num_rar_entries; i++) {
3935 		if (adapter->mac_table[i].state & IXGBE_MAC_STATE_IN_USE)
3936 			continue;
3937 		adapter->mac_table[i].state |= (IXGBE_MAC_STATE_MODIFIED |
3938 						IXGBE_MAC_STATE_IN_USE);
3939 		ether_addr_copy(adapter->mac_table[i].addr, addr);
3940 		adapter->mac_table[i].queue = queue;
3941 		ixgbe_sync_mac_table(adapter);
3942 		return i;
3943 	}
3944 	return -ENOMEM;
3945 }
3946 
3947 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter, u8 *addr, u16 queue)
3948 {
3949 	/* search table for addr, if found, set to 0 and sync */
3950 	int i;
3951 	struct ixgbe_hw *hw = &adapter->hw;
3952 
3953 	if (is_zero_ether_addr(addr))
3954 		return -EINVAL;
3955 
3956 	for (i = 0; i < hw->mac.num_rar_entries; i++) {
3957 		if (ether_addr_equal(addr, adapter->mac_table[i].addr) &&
3958 		    adapter->mac_table[i].queue == queue) {
3959 			adapter->mac_table[i].state |= IXGBE_MAC_STATE_MODIFIED;
3960 			adapter->mac_table[i].state &= ~IXGBE_MAC_STATE_IN_USE;
3961 			memset(adapter->mac_table[i].addr, 0, ETH_ALEN);
3962 			adapter->mac_table[i].queue = 0;
3963 			ixgbe_sync_mac_table(adapter);
3964 			return 0;
3965 		}
3966 	}
3967 	return -ENOMEM;
3968 }
3969 /**
3970  * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3971  * @netdev: network interface device structure
3972  *
3973  * Writes unicast address list to the RAR table.
3974  * Returns: -ENOMEM on failure/insufficient address space
3975  *                0 on no addresses written
3976  *                X on writing X addresses to the RAR table
3977  **/
3978 static int ixgbe_write_uc_addr_list(struct net_device *netdev, int vfn)
3979 {
3980 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3981 	int count = 0;
3982 
3983 	/* return ENOMEM indicating insufficient memory for addresses */
3984 	if (netdev_uc_count(netdev) > ixgbe_available_rars(adapter))
3985 		return -ENOMEM;
3986 
3987 	if (!netdev_uc_empty(netdev)) {
3988 		struct netdev_hw_addr *ha;
3989 		netdev_for_each_uc_addr(ha, netdev) {
3990 			ixgbe_del_mac_filter(adapter, ha->addr, vfn);
3991 			ixgbe_add_mac_filter(adapter, ha->addr, vfn);
3992 			count++;
3993 		}
3994 	}
3995 	return count;
3996 }
3997 
3998 /**
3999  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4000  * @netdev: network interface device structure
4001  *
4002  * The set_rx_method entry point is called whenever the unicast/multicast
4003  * address list or the network interface flags are updated.  This routine is
4004  * responsible for configuring the hardware for proper unicast, multicast and
4005  * promiscuous mode.
4006  **/
4007 void ixgbe_set_rx_mode(struct net_device *netdev)
4008 {
4009 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4010 	struct ixgbe_hw *hw = &adapter->hw;
4011 	u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4012 	u32 vlnctrl;
4013 	int count;
4014 
4015 	/* Check for Promiscuous and All Multicast modes */
4016 	fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4017 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4018 
4019 	/* set all bits that we expect to always be set */
4020 	fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4021 	fctrl |= IXGBE_FCTRL_BAM;
4022 	fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4023 	fctrl |= IXGBE_FCTRL_PMCF;
4024 
4025 	/* clear the bits we are changing the status of */
4026 	fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4027 	vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
4028 	if (netdev->flags & IFF_PROMISC) {
4029 		hw->addr_ctrl.user_set_promisc = true;
4030 		fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4031 		vmolr |= IXGBE_VMOLR_MPE;
4032 		/* Only disable hardware filter vlans in promiscuous mode
4033 		 * if SR-IOV and VMDQ are disabled - otherwise ensure
4034 		 * that hardware VLAN filters remain enabled.
4035 		 */
4036 		if (adapter->flags & (IXGBE_FLAG_VMDQ_ENABLED |
4037 				      IXGBE_FLAG_SRIOV_ENABLED))
4038 			vlnctrl |= (IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
4039 	} else {
4040 		if (netdev->flags & IFF_ALLMULTI) {
4041 			fctrl |= IXGBE_FCTRL_MPE;
4042 			vmolr |= IXGBE_VMOLR_MPE;
4043 		}
4044 		vlnctrl |= IXGBE_VLNCTRL_VFE;
4045 		hw->addr_ctrl.user_set_promisc = false;
4046 	}
4047 
4048 	/*
4049 	 * Write addresses to available RAR registers, if there is not
4050 	 * sufficient space to store all the addresses then enable
4051 	 * unicast promiscuous mode
4052 	 */
4053 	count = ixgbe_write_uc_addr_list(netdev, VMDQ_P(0));
4054 	if (count < 0) {
4055 		fctrl |= IXGBE_FCTRL_UPE;
4056 		vmolr |= IXGBE_VMOLR_ROPE;
4057 	}
4058 
4059 	/* Write addresses to the MTA, if the attempt fails
4060 	 * then we should just turn on promiscuous mode so
4061 	 * that we can at least receive multicast traffic
4062 	 */
4063 	count = ixgbe_write_mc_addr_list(netdev);
4064 	if (count < 0) {
4065 		fctrl |= IXGBE_FCTRL_MPE;
4066 		vmolr |= IXGBE_VMOLR_MPE;
4067 	} else if (count) {
4068 		vmolr |= IXGBE_VMOLR_ROMPE;
4069 	}
4070 
4071 	if (hw->mac.type != ixgbe_mac_82598EB) {
4072 		vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4073 			 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4074 			   IXGBE_VMOLR_ROPE);
4075 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4076 	}
4077 
4078 	/* This is useful for sniffing bad packets. */
4079 	if (adapter->netdev->features & NETIF_F_RXALL) {
4080 		/* UPE and MPE will be handled by normal PROMISC logic
4081 		 * in e1000e_set_rx_mode */
4082 		fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4083 			  IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4084 			  IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4085 
4086 		fctrl &= ~(IXGBE_FCTRL_DPF);
4087 		/* NOTE:  VLAN filtering is disabled by setting PROMISC */
4088 	}
4089 
4090 	IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4091 	IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4092 
4093 	if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
4094 		ixgbe_vlan_strip_enable(adapter);
4095 	else
4096 		ixgbe_vlan_strip_disable(adapter);
4097 }
4098 
4099 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
4100 {
4101 	int q_idx;
4102 
4103 	for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++) {
4104 		ixgbe_qv_init_lock(adapter->q_vector[q_idx]);
4105 		napi_enable(&adapter->q_vector[q_idx]->napi);
4106 	}
4107 }
4108 
4109 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
4110 {
4111 	int q_idx;
4112 
4113 	for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++) {
4114 		napi_disable(&adapter->q_vector[q_idx]->napi);
4115 		while (!ixgbe_qv_disable(adapter->q_vector[q_idx])) {
4116 			pr_info("QV %d locked\n", q_idx);
4117 			usleep_range(1000, 20000);
4118 		}
4119 	}
4120 }
4121 
4122 #ifdef CONFIG_IXGBE_DCB
4123 /**
4124  * ixgbe_configure_dcb - Configure DCB hardware
4125  * @adapter: ixgbe adapter struct
4126  *
4127  * This is called by the driver on open to configure the DCB hardware.
4128  * This is also called by the gennetlink interface when reconfiguring
4129  * the DCB state.
4130  */
4131 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
4132 {
4133 	struct ixgbe_hw *hw = &adapter->hw;
4134 	int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4135 
4136 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
4137 		if (hw->mac.type == ixgbe_mac_82598EB)
4138 			netif_set_gso_max_size(adapter->netdev, 65536);
4139 		return;
4140 	}
4141 
4142 	if (hw->mac.type == ixgbe_mac_82598EB)
4143 		netif_set_gso_max_size(adapter->netdev, 32768);
4144 
4145 #ifdef IXGBE_FCOE
4146 	if (adapter->netdev->features & NETIF_F_FCOE_MTU)
4147 		max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
4148 #endif
4149 
4150 	/* reconfigure the hardware */
4151 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
4152 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
4153 						DCB_TX_CONFIG);
4154 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
4155 						DCB_RX_CONFIG);
4156 		ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
4157 	} else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
4158 		ixgbe_dcb_hw_ets(&adapter->hw,
4159 				 adapter->ixgbe_ieee_ets,
4160 				 max_frame);
4161 		ixgbe_dcb_hw_pfc_config(&adapter->hw,
4162 					adapter->ixgbe_ieee_pfc->pfc_en,
4163 					adapter->ixgbe_ieee_ets->prio_tc);
4164 	}
4165 
4166 	/* Enable RSS Hash per TC */
4167 	if (hw->mac.type != ixgbe_mac_82598EB) {
4168 		u32 msb = 0;
4169 		u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
4170 
4171 		while (rss_i) {
4172 			msb++;
4173 			rss_i >>= 1;
4174 		}
4175 
4176 		/* write msb to all 8 TCs in one write */
4177 		IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
4178 	}
4179 }
4180 #endif
4181 
4182 /* Additional bittime to account for IXGBE framing */
4183 #define IXGBE_ETH_FRAMING 20
4184 
4185 /**
4186  * ixgbe_hpbthresh - calculate high water mark for flow control
4187  *
4188  * @adapter: board private structure to calculate for
4189  * @pb: packet buffer to calculate
4190  */
4191 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
4192 {
4193 	struct ixgbe_hw *hw = &adapter->hw;
4194 	struct net_device *dev = adapter->netdev;
4195 	int link, tc, kb, marker;
4196 	u32 dv_id, rx_pba;
4197 
4198 	/* Calculate max LAN frame size */
4199 	tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
4200 
4201 #ifdef IXGBE_FCOE
4202 	/* FCoE traffic class uses FCOE jumbo frames */
4203 	if ((dev->features & NETIF_F_FCOE_MTU) &&
4204 	    (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
4205 	    (pb == ixgbe_fcoe_get_tc(adapter)))
4206 		tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4207 #endif
4208 
4209 	/* Calculate delay value for device */
4210 	switch (hw->mac.type) {
4211 	case ixgbe_mac_X540:
4212 	case ixgbe_mac_X550:
4213 	case ixgbe_mac_X550EM_x:
4214 		dv_id = IXGBE_DV_X540(link, tc);
4215 		break;
4216 	default:
4217 		dv_id = IXGBE_DV(link, tc);
4218 		break;
4219 	}
4220 
4221 	/* Loopback switch introduces additional latency */
4222 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4223 		dv_id += IXGBE_B2BT(tc);
4224 
4225 	/* Delay value is calculated in bit times convert to KB */
4226 	kb = IXGBE_BT2KB(dv_id);
4227 	rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
4228 
4229 	marker = rx_pba - kb;
4230 
4231 	/* It is possible that the packet buffer is not large enough
4232 	 * to provide required headroom. In this case throw an error
4233 	 * to user and a do the best we can.
4234 	 */
4235 	if (marker < 0) {
4236 		e_warn(drv, "Packet Buffer(%i) can not provide enough"
4237 			    "headroom to support flow control."
4238 			    "Decrease MTU or number of traffic classes\n", pb);
4239 		marker = tc + 1;
4240 	}
4241 
4242 	return marker;
4243 }
4244 
4245 /**
4246  * ixgbe_lpbthresh - calculate low water mark for for flow control
4247  *
4248  * @adapter: board private structure to calculate for
4249  * @pb: packet buffer to calculate
4250  */
4251 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
4252 {
4253 	struct ixgbe_hw *hw = &adapter->hw;
4254 	struct net_device *dev = adapter->netdev;
4255 	int tc;
4256 	u32 dv_id;
4257 
4258 	/* Calculate max LAN frame size */
4259 	tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
4260 
4261 #ifdef IXGBE_FCOE
4262 	/* FCoE traffic class uses FCOE jumbo frames */
4263 	if ((dev->features & NETIF_F_FCOE_MTU) &&
4264 	    (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
4265 	    (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
4266 		tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4267 #endif
4268 
4269 	/* Calculate delay value for device */
4270 	switch (hw->mac.type) {
4271 	case ixgbe_mac_X540:
4272 	case ixgbe_mac_X550:
4273 	case ixgbe_mac_X550EM_x:
4274 		dv_id = IXGBE_LOW_DV_X540(tc);
4275 		break;
4276 	default:
4277 		dv_id = IXGBE_LOW_DV(tc);
4278 		break;
4279 	}
4280 
4281 	/* Delay value is calculated in bit times convert to KB */
4282 	return IXGBE_BT2KB(dv_id);
4283 }
4284 
4285 /*
4286  * ixgbe_pbthresh_setup - calculate and setup high low water marks
4287  */
4288 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
4289 {
4290 	struct ixgbe_hw *hw = &adapter->hw;
4291 	int num_tc = netdev_get_num_tc(adapter->netdev);
4292 	int i;
4293 
4294 	if (!num_tc)
4295 		num_tc = 1;
4296 
4297 	for (i = 0; i < num_tc; i++) {
4298 		hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
4299 		hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
4300 
4301 		/* Low water marks must not be larger than high water marks */
4302 		if (hw->fc.low_water[i] > hw->fc.high_water[i])
4303 			hw->fc.low_water[i] = 0;
4304 	}
4305 
4306 	for (; i < MAX_TRAFFIC_CLASS; i++)
4307 		hw->fc.high_water[i] = 0;
4308 }
4309 
4310 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
4311 {
4312 	struct ixgbe_hw *hw = &adapter->hw;
4313 	int hdrm;
4314 	u8 tc = netdev_get_num_tc(adapter->netdev);
4315 
4316 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4317 	    adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
4318 		hdrm = 32 << adapter->fdir_pballoc;
4319 	else
4320 		hdrm = 0;
4321 
4322 	hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
4323 	ixgbe_pbthresh_setup(adapter);
4324 }
4325 
4326 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
4327 {
4328 	struct ixgbe_hw *hw = &adapter->hw;
4329 	struct hlist_node *node2;
4330 	struct ixgbe_fdir_filter *filter;
4331 
4332 	spin_lock(&adapter->fdir_perfect_lock);
4333 
4334 	if (!hlist_empty(&adapter->fdir_filter_list))
4335 		ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
4336 
4337 	hlist_for_each_entry_safe(filter, node2,
4338 				  &adapter->fdir_filter_list, fdir_node) {
4339 		ixgbe_fdir_write_perfect_filter_82599(hw,
4340 				&filter->filter,
4341 				filter->sw_idx,
4342 				(filter->action == IXGBE_FDIR_DROP_QUEUE) ?
4343 				IXGBE_FDIR_DROP_QUEUE :
4344 				adapter->rx_ring[filter->action]->reg_idx);
4345 	}
4346 
4347 	spin_unlock(&adapter->fdir_perfect_lock);
4348 }
4349 
4350 static void ixgbe_macvlan_set_rx_mode(struct net_device *dev, unsigned int pool,
4351 				      struct ixgbe_adapter *adapter)
4352 {
4353 	struct ixgbe_hw *hw = &adapter->hw;
4354 	u32 vmolr;
4355 
4356 	/* No unicast promiscuous support for VMDQ devices. */
4357 	vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(pool));
4358 	vmolr |= (IXGBE_VMOLR_ROMPE | IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE);
4359 
4360 	/* clear the affected bit */
4361 	vmolr &= ~IXGBE_VMOLR_MPE;
4362 
4363 	if (dev->flags & IFF_ALLMULTI) {
4364 		vmolr |= IXGBE_VMOLR_MPE;
4365 	} else {
4366 		vmolr |= IXGBE_VMOLR_ROMPE;
4367 		hw->mac.ops.update_mc_addr_list(hw, dev);
4368 	}
4369 	ixgbe_write_uc_addr_list(adapter->netdev, pool);
4370 	IXGBE_WRITE_REG(hw, IXGBE_VMOLR(pool), vmolr);
4371 }
4372 
4373 static void ixgbe_fwd_psrtype(struct ixgbe_fwd_adapter *vadapter)
4374 {
4375 	struct ixgbe_adapter *adapter = vadapter->real_adapter;
4376 	int rss_i = adapter->num_rx_queues_per_pool;
4377 	struct ixgbe_hw *hw = &adapter->hw;
4378 	u16 pool = vadapter->pool;
4379 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4380 		      IXGBE_PSRTYPE_UDPHDR |
4381 		      IXGBE_PSRTYPE_IPV4HDR |
4382 		      IXGBE_PSRTYPE_L2HDR |
4383 		      IXGBE_PSRTYPE_IPV6HDR;
4384 
4385 	if (hw->mac.type == ixgbe_mac_82598EB)
4386 		return;
4387 
4388 	if (rss_i > 3)
4389 		psrtype |= 2 << 29;
4390 	else if (rss_i > 1)
4391 		psrtype |= 1 << 29;
4392 
4393 	IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4394 }
4395 
4396 /**
4397  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4398  * @rx_ring: ring to free buffers from
4399  **/
4400 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4401 {
4402 	struct device *dev = rx_ring->dev;
4403 	unsigned long size;
4404 	u16 i;
4405 
4406 	/* ring already cleared, nothing to do */
4407 	if (!rx_ring->rx_buffer_info)
4408 		return;
4409 
4410 	/* Free all the Rx ring sk_buffs */
4411 	for (i = 0; i < rx_ring->count; i++) {
4412 		struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
4413 
4414 		if (rx_buffer->skb) {
4415 			struct sk_buff *skb = rx_buffer->skb;
4416 			if (IXGBE_CB(skb)->page_released)
4417 				dma_unmap_page(dev,
4418 					       IXGBE_CB(skb)->dma,
4419 					       ixgbe_rx_bufsz(rx_ring),
4420 					       DMA_FROM_DEVICE);
4421 			dev_kfree_skb(skb);
4422 			rx_buffer->skb = NULL;
4423 		}
4424 
4425 		if (!rx_buffer->page)
4426 			continue;
4427 
4428 		dma_unmap_page(dev, rx_buffer->dma,
4429 			       ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
4430 		__free_pages(rx_buffer->page, ixgbe_rx_pg_order(rx_ring));
4431 
4432 		rx_buffer->page = NULL;
4433 	}
4434 
4435 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4436 	memset(rx_ring->rx_buffer_info, 0, size);
4437 
4438 	/* Zero out the descriptor ring */
4439 	memset(rx_ring->desc, 0, rx_ring->size);
4440 
4441 	rx_ring->next_to_alloc = 0;
4442 	rx_ring->next_to_clean = 0;
4443 	rx_ring->next_to_use = 0;
4444 }
4445 
4446 static void ixgbe_disable_fwd_ring(struct ixgbe_fwd_adapter *vadapter,
4447 				   struct ixgbe_ring *rx_ring)
4448 {
4449 	struct ixgbe_adapter *adapter = vadapter->real_adapter;
4450 	int index = rx_ring->queue_index + vadapter->rx_base_queue;
4451 
4452 	/* shutdown specific queue receive and wait for dma to settle */
4453 	ixgbe_disable_rx_queue(adapter, rx_ring);
4454 	usleep_range(10000, 20000);
4455 	ixgbe_irq_disable_queues(adapter, ((u64)1 << index));
4456 	ixgbe_clean_rx_ring(rx_ring);
4457 	rx_ring->l2_accel_priv = NULL;
4458 }
4459 
4460 static int ixgbe_fwd_ring_down(struct net_device *vdev,
4461 			       struct ixgbe_fwd_adapter *accel)
4462 {
4463 	struct ixgbe_adapter *adapter = accel->real_adapter;
4464 	unsigned int rxbase = accel->rx_base_queue;
4465 	unsigned int txbase = accel->tx_base_queue;
4466 	int i;
4467 
4468 	netif_tx_stop_all_queues(vdev);
4469 
4470 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4471 		ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
4472 		adapter->rx_ring[rxbase + i]->netdev = adapter->netdev;
4473 	}
4474 
4475 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4476 		adapter->tx_ring[txbase + i]->l2_accel_priv = NULL;
4477 		adapter->tx_ring[txbase + i]->netdev = adapter->netdev;
4478 	}
4479 
4480 
4481 	return 0;
4482 }
4483 
4484 static int ixgbe_fwd_ring_up(struct net_device *vdev,
4485 			     struct ixgbe_fwd_adapter *accel)
4486 {
4487 	struct ixgbe_adapter *adapter = accel->real_adapter;
4488 	unsigned int rxbase, txbase, queues;
4489 	int i, baseq, err = 0;
4490 
4491 	if (!test_bit(accel->pool, &adapter->fwd_bitmask))
4492 		return 0;
4493 
4494 	baseq = accel->pool * adapter->num_rx_queues_per_pool;
4495 	netdev_dbg(vdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
4496 		   accel->pool, adapter->num_rx_pools,
4497 		   baseq, baseq + adapter->num_rx_queues_per_pool,
4498 		   adapter->fwd_bitmask);
4499 
4500 	accel->netdev = vdev;
4501 	accel->rx_base_queue = rxbase = baseq;
4502 	accel->tx_base_queue = txbase = baseq;
4503 
4504 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
4505 		ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
4506 
4507 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4508 		adapter->rx_ring[rxbase + i]->netdev = vdev;
4509 		adapter->rx_ring[rxbase + i]->l2_accel_priv = accel;
4510 		ixgbe_configure_rx_ring(adapter, adapter->rx_ring[rxbase + i]);
4511 	}
4512 
4513 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4514 		adapter->tx_ring[txbase + i]->netdev = vdev;
4515 		adapter->tx_ring[txbase + i]->l2_accel_priv = accel;
4516 	}
4517 
4518 	queues = min_t(unsigned int,
4519 		       adapter->num_rx_queues_per_pool, vdev->num_tx_queues);
4520 	err = netif_set_real_num_tx_queues(vdev, queues);
4521 	if (err)
4522 		goto fwd_queue_err;
4523 
4524 	err = netif_set_real_num_rx_queues(vdev, queues);
4525 	if (err)
4526 		goto fwd_queue_err;
4527 
4528 	if (is_valid_ether_addr(vdev->dev_addr))
4529 		ixgbe_add_mac_filter(adapter, vdev->dev_addr, accel->pool);
4530 
4531 	ixgbe_fwd_psrtype(accel);
4532 	ixgbe_macvlan_set_rx_mode(vdev, accel->pool, adapter);
4533 	return err;
4534 fwd_queue_err:
4535 	ixgbe_fwd_ring_down(vdev, accel);
4536 	return err;
4537 }
4538 
4539 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
4540 {
4541 	struct net_device *upper;
4542 	struct list_head *iter;
4543 	int err;
4544 
4545 	netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
4546 		if (netif_is_macvlan(upper)) {
4547 			struct macvlan_dev *dfwd = netdev_priv(upper);
4548 			struct ixgbe_fwd_adapter *vadapter = dfwd->fwd_priv;
4549 
4550 			if (dfwd->fwd_priv) {
4551 				err = ixgbe_fwd_ring_up(upper, vadapter);
4552 				if (err)
4553 					continue;
4554 			}
4555 		}
4556 	}
4557 }
4558 
4559 static void ixgbe_configure(struct ixgbe_adapter *adapter)
4560 {
4561 	struct ixgbe_hw *hw = &adapter->hw;
4562 
4563 	ixgbe_configure_pb(adapter);
4564 #ifdef CONFIG_IXGBE_DCB
4565 	ixgbe_configure_dcb(adapter);
4566 #endif
4567 	/*
4568 	 * We must restore virtualization before VLANs or else
4569 	 * the VLVF registers will not be populated
4570 	 */
4571 	ixgbe_configure_virtualization(adapter);
4572 
4573 	ixgbe_set_rx_mode(adapter->netdev);
4574 	ixgbe_restore_vlan(adapter);
4575 
4576 	switch (hw->mac.type) {
4577 	case ixgbe_mac_82599EB:
4578 	case ixgbe_mac_X540:
4579 		hw->mac.ops.disable_rx_buff(hw);
4580 		break;
4581 	default:
4582 		break;
4583 	}
4584 
4585 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
4586 		ixgbe_init_fdir_signature_82599(&adapter->hw,
4587 						adapter->fdir_pballoc);
4588 	} else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
4589 		ixgbe_init_fdir_perfect_82599(&adapter->hw,
4590 					      adapter->fdir_pballoc);
4591 		ixgbe_fdir_filter_restore(adapter);
4592 	}
4593 
4594 	switch (hw->mac.type) {
4595 	case ixgbe_mac_82599EB:
4596 	case ixgbe_mac_X540:
4597 		hw->mac.ops.enable_rx_buff(hw);
4598 		break;
4599 	default:
4600 		break;
4601 	}
4602 
4603 #ifdef IXGBE_FCOE
4604 	/* configure FCoE L2 filters, redirection table, and Rx control */
4605 	ixgbe_configure_fcoe(adapter);
4606 
4607 #endif /* IXGBE_FCOE */
4608 	ixgbe_configure_tx(adapter);
4609 	ixgbe_configure_rx(adapter);
4610 	ixgbe_configure_dfwd(adapter);
4611 }
4612 
4613 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
4614 {
4615 	switch (hw->phy.type) {
4616 	case ixgbe_phy_sfp_avago:
4617 	case ixgbe_phy_sfp_ftl:
4618 	case ixgbe_phy_sfp_intel:
4619 	case ixgbe_phy_sfp_unknown:
4620 	case ixgbe_phy_sfp_passive_tyco:
4621 	case ixgbe_phy_sfp_passive_unknown:
4622 	case ixgbe_phy_sfp_active_unknown:
4623 	case ixgbe_phy_sfp_ftl_active:
4624 	case ixgbe_phy_qsfp_passive_unknown:
4625 	case ixgbe_phy_qsfp_active_unknown:
4626 	case ixgbe_phy_qsfp_intel:
4627 	case ixgbe_phy_qsfp_unknown:
4628 	/* ixgbe_phy_none is set when no SFP module is present */
4629 	case ixgbe_phy_none:
4630 		return true;
4631 	case ixgbe_phy_nl:
4632 		if (hw->mac.type == ixgbe_mac_82598EB)
4633 			return true;
4634 	default:
4635 		return false;
4636 	}
4637 }
4638 
4639 /**
4640  * ixgbe_sfp_link_config - set up SFP+ link
4641  * @adapter: pointer to private adapter struct
4642  **/
4643 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
4644 {
4645 	/*
4646 	 * We are assuming the worst case scenario here, and that
4647 	 * is that an SFP was inserted/removed after the reset
4648 	 * but before SFP detection was enabled.  As such the best
4649 	 * solution is to just start searching as soon as we start
4650 	 */
4651 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
4652 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
4653 
4654 	adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
4655 }
4656 
4657 /**
4658  * ixgbe_non_sfp_link_config - set up non-SFP+ link
4659  * @hw: pointer to private hardware struct
4660  *
4661  * Returns 0 on success, negative on failure
4662  **/
4663 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
4664 {
4665 	u32 speed;
4666 	bool autoneg, link_up = false;
4667 	u32 ret = IXGBE_ERR_LINK_SETUP;
4668 
4669 	if (hw->mac.ops.check_link)
4670 		ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
4671 
4672 	if (ret)
4673 		return ret;
4674 
4675 	speed = hw->phy.autoneg_advertised;
4676 	if ((!speed) && (hw->mac.ops.get_link_capabilities))
4677 		ret = hw->mac.ops.get_link_capabilities(hw, &speed,
4678 							&autoneg);
4679 	if (ret)
4680 		return ret;
4681 
4682 	if (hw->mac.ops.setup_link)
4683 		ret = hw->mac.ops.setup_link(hw, speed, link_up);
4684 
4685 	return ret;
4686 }
4687 
4688 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
4689 {
4690 	struct ixgbe_hw *hw = &adapter->hw;
4691 	u32 gpie = 0;
4692 
4693 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4694 		gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
4695 		       IXGBE_GPIE_OCD;
4696 		gpie |= IXGBE_GPIE_EIAME;
4697 		/*
4698 		 * use EIAM to auto-mask when MSI-X interrupt is asserted
4699 		 * this saves a register write for every interrupt
4700 		 */
4701 		switch (hw->mac.type) {
4702 		case ixgbe_mac_82598EB:
4703 			IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4704 			break;
4705 		case ixgbe_mac_82599EB:
4706 		case ixgbe_mac_X540:
4707 		case ixgbe_mac_X550:
4708 		case ixgbe_mac_X550EM_x:
4709 		default:
4710 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
4711 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
4712 			break;
4713 		}
4714 	} else {
4715 		/* legacy interrupts, use EIAM to auto-mask when reading EICR,
4716 		 * specifically only auto mask tx and rx interrupts */
4717 		IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4718 	}
4719 
4720 	/* XXX: to interrupt immediately for EICS writes, enable this */
4721 	/* gpie |= IXGBE_GPIE_EIMEN; */
4722 
4723 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
4724 		gpie &= ~IXGBE_GPIE_VTMODE_MASK;
4725 
4726 		switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4727 		case IXGBE_82599_VMDQ_8Q_MASK:
4728 			gpie |= IXGBE_GPIE_VTMODE_16;
4729 			break;
4730 		case IXGBE_82599_VMDQ_4Q_MASK:
4731 			gpie |= IXGBE_GPIE_VTMODE_32;
4732 			break;
4733 		default:
4734 			gpie |= IXGBE_GPIE_VTMODE_64;
4735 			break;
4736 		}
4737 	}
4738 
4739 	/* Enable Thermal over heat sensor interrupt */
4740 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
4741 		switch (adapter->hw.mac.type) {
4742 		case ixgbe_mac_82599EB:
4743 			gpie |= IXGBE_SDP0_GPIEN;
4744 			break;
4745 		case ixgbe_mac_X540:
4746 			gpie |= IXGBE_EIMS_TS;
4747 			break;
4748 		default:
4749 			break;
4750 		}
4751 	}
4752 
4753 	/* Enable fan failure interrupt */
4754 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
4755 		gpie |= IXGBE_SDP1_GPIEN;
4756 
4757 	if (hw->mac.type == ixgbe_mac_82599EB) {
4758 		gpie |= IXGBE_SDP1_GPIEN;
4759 		gpie |= IXGBE_SDP2_GPIEN;
4760 	}
4761 
4762 	IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
4763 }
4764 
4765 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
4766 {
4767 	struct ixgbe_hw *hw = &adapter->hw;
4768 	int err;
4769 	u32 ctrl_ext;
4770 
4771 	ixgbe_get_hw_control(adapter);
4772 	ixgbe_setup_gpie(adapter);
4773 
4774 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4775 		ixgbe_configure_msix(adapter);
4776 	else
4777 		ixgbe_configure_msi_and_legacy(adapter);
4778 
4779 	/* enable the optics for 82599 SFP+ fiber */
4780 	if (hw->mac.ops.enable_tx_laser)
4781 		hw->mac.ops.enable_tx_laser(hw);
4782 
4783 	smp_mb__before_atomic();
4784 	clear_bit(__IXGBE_DOWN, &adapter->state);
4785 	ixgbe_napi_enable_all(adapter);
4786 
4787 	if (ixgbe_is_sfp(hw)) {
4788 		ixgbe_sfp_link_config(adapter);
4789 	} else {
4790 		err = ixgbe_non_sfp_link_config(hw);
4791 		if (err)
4792 			e_err(probe, "link_config FAILED %d\n", err);
4793 	}
4794 
4795 	/* clear any pending interrupts, may auto mask */
4796 	IXGBE_READ_REG(hw, IXGBE_EICR);
4797 	ixgbe_irq_enable(adapter, true, true);
4798 
4799 	/*
4800 	 * If this adapter has a fan, check to see if we had a failure
4801 	 * before we enabled the interrupt.
4802 	 */
4803 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
4804 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
4805 		if (esdp & IXGBE_ESDP_SDP1)
4806 			e_crit(drv, "Fan has stopped, replace the adapter\n");
4807 	}
4808 
4809 	/* bring the link up in the watchdog, this could race with our first
4810 	 * link up interrupt but shouldn't be a problem */
4811 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4812 	adapter->link_check_timeout = jiffies;
4813 	mod_timer(&adapter->service_timer, jiffies);
4814 
4815 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
4816 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
4817 	ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
4818 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
4819 }
4820 
4821 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
4822 {
4823 	WARN_ON(in_interrupt());
4824 	/* put off any impending NetWatchDogTimeout */
4825 	adapter->netdev->trans_start = jiffies;
4826 
4827 	while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
4828 		usleep_range(1000, 2000);
4829 	ixgbe_down(adapter);
4830 	/*
4831 	 * If SR-IOV enabled then wait a bit before bringing the adapter
4832 	 * back up to give the VFs time to respond to the reset.  The
4833 	 * two second wait is based upon the watchdog timer cycle in
4834 	 * the VF driver.
4835 	 */
4836 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4837 		msleep(2000);
4838 	ixgbe_up(adapter);
4839 	clear_bit(__IXGBE_RESETTING, &adapter->state);
4840 }
4841 
4842 void ixgbe_up(struct ixgbe_adapter *adapter)
4843 {
4844 	/* hardware has been reset, we need to reload some things */
4845 	ixgbe_configure(adapter);
4846 
4847 	ixgbe_up_complete(adapter);
4848 }
4849 
4850 void ixgbe_reset(struct ixgbe_adapter *adapter)
4851 {
4852 	struct ixgbe_hw *hw = &adapter->hw;
4853 	struct net_device *netdev = adapter->netdev;
4854 	int err;
4855 	u8 old_addr[ETH_ALEN];
4856 
4857 	if (ixgbe_removed(hw->hw_addr))
4858 		return;
4859 	/* lock SFP init bit to prevent race conditions with the watchdog */
4860 	while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
4861 		usleep_range(1000, 2000);
4862 
4863 	/* clear all SFP and link config related flags while holding SFP_INIT */
4864 	adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
4865 			     IXGBE_FLAG2_SFP_NEEDS_RESET);
4866 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
4867 
4868 	err = hw->mac.ops.init_hw(hw);
4869 	switch (err) {
4870 	case 0:
4871 	case IXGBE_ERR_SFP_NOT_PRESENT:
4872 	case IXGBE_ERR_SFP_NOT_SUPPORTED:
4873 		break;
4874 	case IXGBE_ERR_MASTER_REQUESTS_PENDING:
4875 		e_dev_err("master disable timed out\n");
4876 		break;
4877 	case IXGBE_ERR_EEPROM_VERSION:
4878 		/* We are running on a pre-production device, log a warning */
4879 		e_dev_warn("This device is a pre-production adapter/LOM. "
4880 			   "Please be aware there may be issues associated with "
4881 			   "your hardware.  If you are experiencing problems "
4882 			   "please contact your Intel or hardware "
4883 			   "representative who provided you with this "
4884 			   "hardware.\n");
4885 		break;
4886 	default:
4887 		e_dev_err("Hardware Error: %d\n", err);
4888 	}
4889 
4890 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
4891 	/* do not flush user set addresses */
4892 	memcpy(old_addr, &adapter->mac_table[0].addr, netdev->addr_len);
4893 	ixgbe_flush_sw_mac_table(adapter);
4894 	ixgbe_mac_set_default_filter(adapter, old_addr);
4895 
4896 	/* update SAN MAC vmdq pool selection */
4897 	if (hw->mac.san_mac_rar_index)
4898 		hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
4899 
4900 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
4901 		ixgbe_ptp_reset(adapter);
4902 }
4903 
4904 /**
4905  * ixgbe_clean_tx_ring - Free Tx Buffers
4906  * @tx_ring: ring to be cleaned
4907  **/
4908 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4909 {
4910 	struct ixgbe_tx_buffer *tx_buffer_info;
4911 	unsigned long size;
4912 	u16 i;
4913 
4914 	/* ring already cleared, nothing to do */
4915 	if (!tx_ring->tx_buffer_info)
4916 		return;
4917 
4918 	/* Free all the Tx ring sk_buffs */
4919 	for (i = 0; i < tx_ring->count; i++) {
4920 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
4921 		ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4922 	}
4923 
4924 	netdev_tx_reset_queue(txring_txq(tx_ring));
4925 
4926 	size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4927 	memset(tx_ring->tx_buffer_info, 0, size);
4928 
4929 	/* Zero out the descriptor ring */
4930 	memset(tx_ring->desc, 0, tx_ring->size);
4931 
4932 	tx_ring->next_to_use = 0;
4933 	tx_ring->next_to_clean = 0;
4934 }
4935 
4936 /**
4937  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4938  * @adapter: board private structure
4939  **/
4940 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4941 {
4942 	int i;
4943 
4944 	for (i = 0; i < adapter->num_rx_queues; i++)
4945 		ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4946 }
4947 
4948 /**
4949  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4950  * @adapter: board private structure
4951  **/
4952 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4953 {
4954 	int i;
4955 
4956 	for (i = 0; i < adapter->num_tx_queues; i++)
4957 		ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4958 }
4959 
4960 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
4961 {
4962 	struct hlist_node *node2;
4963 	struct ixgbe_fdir_filter *filter;
4964 
4965 	spin_lock(&adapter->fdir_perfect_lock);
4966 
4967 	hlist_for_each_entry_safe(filter, node2,
4968 				  &adapter->fdir_filter_list, fdir_node) {
4969 		hlist_del(&filter->fdir_node);
4970 		kfree(filter);
4971 	}
4972 	adapter->fdir_filter_count = 0;
4973 
4974 	spin_unlock(&adapter->fdir_perfect_lock);
4975 }
4976 
4977 void ixgbe_down(struct ixgbe_adapter *adapter)
4978 {
4979 	struct net_device *netdev = adapter->netdev;
4980 	struct ixgbe_hw *hw = &adapter->hw;
4981 	struct net_device *upper;
4982 	struct list_head *iter;
4983 	u32 rxctrl;
4984 	int i;
4985 
4986 	/* signal that we are down to the interrupt handler */
4987 	if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
4988 		return; /* do nothing if already down */
4989 
4990 	/* disable receives */
4991 	rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4992 	IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4993 
4994 	/* disable all enabled rx queues */
4995 	for (i = 0; i < adapter->num_rx_queues; i++)
4996 		/* this call also flushes the previous write */
4997 		ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4998 
4999 	usleep_range(10000, 20000);
5000 
5001 	netif_tx_stop_all_queues(netdev);
5002 
5003 	/* call carrier off first to avoid false dev_watchdog timeouts */
5004 	netif_carrier_off(netdev);
5005 	netif_tx_disable(netdev);
5006 
5007 	/* disable any upper devices */
5008 	netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
5009 		if (netif_is_macvlan(upper)) {
5010 			struct macvlan_dev *vlan = netdev_priv(upper);
5011 
5012 			if (vlan->fwd_priv) {
5013 				netif_tx_stop_all_queues(upper);
5014 				netif_carrier_off(upper);
5015 				netif_tx_disable(upper);
5016 			}
5017 		}
5018 	}
5019 
5020 	ixgbe_irq_disable(adapter);
5021 
5022 	ixgbe_napi_disable_all(adapter);
5023 
5024 	adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
5025 			     IXGBE_FLAG2_RESET_REQUESTED);
5026 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5027 
5028 	del_timer_sync(&adapter->service_timer);
5029 
5030 	if (adapter->num_vfs) {
5031 		/* Clear EITR Select mapping */
5032 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
5033 
5034 		/* Mark all the VFs as inactive */
5035 		for (i = 0 ; i < adapter->num_vfs; i++)
5036 			adapter->vfinfo[i].clear_to_send = false;
5037 
5038 		/* ping all the active vfs to let them know we are going down */
5039 		ixgbe_ping_all_vfs(adapter);
5040 
5041 		/* Disable all VFTE/VFRE TX/RX */
5042 		ixgbe_disable_tx_rx(adapter);
5043 	}
5044 
5045 	/* disable transmits in the hardware now that interrupts are off */
5046 	for (i = 0; i < adapter->num_tx_queues; i++) {
5047 		u8 reg_idx = adapter->tx_ring[i]->reg_idx;
5048 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5049 	}
5050 
5051 	/* Disable the Tx DMA engine on 82599 and later MAC */
5052 	switch (hw->mac.type) {
5053 	case ixgbe_mac_82599EB:
5054 	case ixgbe_mac_X540:
5055 	case ixgbe_mac_X550:
5056 	case ixgbe_mac_X550EM_x:
5057 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5058 				(IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5059 				 ~IXGBE_DMATXCTL_TE));
5060 		break;
5061 	default:
5062 		break;
5063 	}
5064 
5065 	if (!pci_channel_offline(adapter->pdev))
5066 		ixgbe_reset(adapter);
5067 
5068 	/* power down the optics for 82599 SFP+ fiber */
5069 	if (hw->mac.ops.disable_tx_laser)
5070 		hw->mac.ops.disable_tx_laser(hw);
5071 
5072 	ixgbe_clean_all_tx_rings(adapter);
5073 	ixgbe_clean_all_rx_rings(adapter);
5074 
5075 #ifdef CONFIG_IXGBE_DCA
5076 	/* since we reset the hardware DCA settings were cleared */
5077 	ixgbe_setup_dca(adapter);
5078 #endif
5079 }
5080 
5081 /**
5082  * ixgbe_tx_timeout - Respond to a Tx Hang
5083  * @netdev: network interface device structure
5084  **/
5085 static void ixgbe_tx_timeout(struct net_device *netdev)
5086 {
5087 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
5088 
5089 	/* Do the reset outside of interrupt context */
5090 	ixgbe_tx_timeout_reset(adapter);
5091 }
5092 
5093 /**
5094  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
5095  * @adapter: board private structure to initialize
5096  *
5097  * ixgbe_sw_init initializes the Adapter private data structure.
5098  * Fields are initialized based on PCI device information and
5099  * OS network device settings (MTU size).
5100  **/
5101 static int ixgbe_sw_init(struct ixgbe_adapter *adapter)
5102 {
5103 	struct ixgbe_hw *hw = &adapter->hw;
5104 	struct pci_dev *pdev = adapter->pdev;
5105 	unsigned int rss, fdir;
5106 	u32 fwsm;
5107 #ifdef CONFIG_IXGBE_DCB
5108 	int j;
5109 	struct tc_configuration *tc;
5110 #endif
5111 
5112 	/* PCI config space info */
5113 
5114 	hw->vendor_id = pdev->vendor;
5115 	hw->device_id = pdev->device;
5116 	hw->revision_id = pdev->revision;
5117 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
5118 	hw->subsystem_device_id = pdev->subsystem_device;
5119 
5120 	/* Set common capability flags and settings */
5121 	rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
5122 	adapter->ring_feature[RING_F_RSS].limit = rss;
5123 	adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
5124 	adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
5125 	adapter->max_q_vectors = MAX_Q_VECTORS_82599;
5126 	adapter->atr_sample_rate = 20;
5127 	fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
5128 	adapter->ring_feature[RING_F_FDIR].limit = fdir;
5129 	adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
5130 #ifdef CONFIG_IXGBE_DCA
5131 	adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
5132 #endif
5133 #ifdef IXGBE_FCOE
5134 	adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
5135 	adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5136 #ifdef CONFIG_IXGBE_DCB
5137 	/* Default traffic class to use for FCoE */
5138 	adapter->fcoe.up = IXGBE_FCOE_DEFTC;
5139 #endif /* CONFIG_IXGBE_DCB */
5140 #endif /* IXGBE_FCOE */
5141 
5142 	adapter->mac_table = kzalloc(sizeof(struct ixgbe_mac_addr) *
5143 				     hw->mac.num_rar_entries,
5144 				     GFP_ATOMIC);
5145 
5146 	/* Set MAC specific capability flags and exceptions */
5147 	switch (hw->mac.type) {
5148 	case ixgbe_mac_82598EB:
5149 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
5150 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
5151 
5152 		if (hw->device_id == IXGBE_DEV_ID_82598AT)
5153 			adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
5154 
5155 		adapter->max_q_vectors = MAX_Q_VECTORS_82598;
5156 		adapter->ring_feature[RING_F_FDIR].limit = 0;
5157 		adapter->atr_sample_rate = 0;
5158 		adapter->fdir_pballoc = 0;
5159 #ifdef IXGBE_FCOE
5160 		adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
5161 		adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5162 #ifdef CONFIG_IXGBE_DCB
5163 		adapter->fcoe.up = 0;
5164 #endif /* IXGBE_DCB */
5165 #endif /* IXGBE_FCOE */
5166 		break;
5167 	case ixgbe_mac_82599EB:
5168 		if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
5169 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5170 		break;
5171 	case ixgbe_mac_X540:
5172 		fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM);
5173 		if (fwsm & IXGBE_FWSM_TS_ENABLED)
5174 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5175 		break;
5176 	case ixgbe_mac_X550EM_x:
5177 	case ixgbe_mac_X550:
5178 #ifdef CONFIG_IXGBE_DCA
5179 		adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
5180 #endif
5181 		break;
5182 	default:
5183 		break;
5184 	}
5185 
5186 #ifdef IXGBE_FCOE
5187 	/* FCoE support exists, always init the FCoE lock */
5188 	spin_lock_init(&adapter->fcoe.lock);
5189 
5190 #endif
5191 	/* n-tuple support exists, always init our spinlock */
5192 	spin_lock_init(&adapter->fdir_perfect_lock);
5193 
5194 #ifdef CONFIG_IXGBE_DCB
5195 	switch (hw->mac.type) {
5196 	case ixgbe_mac_X540:
5197 	case ixgbe_mac_X550:
5198 	case ixgbe_mac_X550EM_x:
5199 		adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
5200 		adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
5201 		break;
5202 	default:
5203 		adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
5204 		adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
5205 		break;
5206 	}
5207 
5208 	/* Configure DCB traffic classes */
5209 	for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5210 		tc = &adapter->dcb_cfg.tc_config[j];
5211 		tc->path[DCB_TX_CONFIG].bwg_id = 0;
5212 		tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5213 		tc->path[DCB_RX_CONFIG].bwg_id = 0;
5214 		tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5215 		tc->dcb_pfc = pfc_disabled;
5216 	}
5217 
5218 	/* Initialize default user to priority mapping, UPx->TC0 */
5219 	tc = &adapter->dcb_cfg.tc_config[0];
5220 	tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
5221 	tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
5222 
5223 	adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5224 	adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5225 	adapter->dcb_cfg.pfc_mode_enable = false;
5226 	adapter->dcb_set_bitmap = 0x00;
5227 	adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
5228 	memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
5229 	       sizeof(adapter->temp_dcb_cfg));
5230 
5231 #endif
5232 
5233 	/* default flow control settings */
5234 	hw->fc.requested_mode = ixgbe_fc_full;
5235 	hw->fc.current_mode = ixgbe_fc_full;	/* init for ethtool output */
5236 	ixgbe_pbthresh_setup(adapter);
5237 	hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5238 	hw->fc.send_xon = true;
5239 	hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
5240 
5241 #ifdef CONFIG_PCI_IOV
5242 	if (max_vfs > 0)
5243 		e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
5244 
5245 	/* assign number of SR-IOV VFs */
5246 	if (hw->mac.type != ixgbe_mac_82598EB) {
5247 		if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
5248 			adapter->num_vfs = 0;
5249 			e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
5250 		} else {
5251 			adapter->num_vfs = max_vfs;
5252 		}
5253 	}
5254 #endif /* CONFIG_PCI_IOV */
5255 
5256 	/* enable itr by default in dynamic mode */
5257 	adapter->rx_itr_setting = 1;
5258 	adapter->tx_itr_setting = 1;
5259 
5260 	/* set default ring sizes */
5261 	adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5262 	adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5263 
5264 	/* set default work limits */
5265 	adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
5266 
5267 	/* initialize eeprom parameters */
5268 	if (ixgbe_init_eeprom_params_generic(hw)) {
5269 		e_dev_err("EEPROM initialization failed\n");
5270 		return -EIO;
5271 	}
5272 
5273 	/* PF holds first pool slot */
5274 	set_bit(0, &adapter->fwd_bitmask);
5275 	set_bit(__IXGBE_DOWN, &adapter->state);
5276 
5277 	return 0;
5278 }
5279 
5280 /**
5281  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5282  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
5283  *
5284  * Return 0 on success, negative on failure
5285  **/
5286 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5287 {
5288 	struct device *dev = tx_ring->dev;
5289 	int orig_node = dev_to_node(dev);
5290 	int ring_node = -1;
5291 	int size;
5292 
5293 	size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5294 
5295 	if (tx_ring->q_vector)
5296 		ring_node = tx_ring->q_vector->numa_node;
5297 
5298 	tx_ring->tx_buffer_info = vzalloc_node(size, ring_node);
5299 	if (!tx_ring->tx_buffer_info)
5300 		tx_ring->tx_buffer_info = vzalloc(size);
5301 	if (!tx_ring->tx_buffer_info)
5302 		goto err;
5303 
5304 	u64_stats_init(&tx_ring->syncp);
5305 
5306 	/* round up to nearest 4K */
5307 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5308 	tx_ring->size = ALIGN(tx_ring->size, 4096);
5309 
5310 	set_dev_node(dev, ring_node);
5311 	tx_ring->desc = dma_alloc_coherent(dev,
5312 					   tx_ring->size,
5313 					   &tx_ring->dma,
5314 					   GFP_KERNEL);
5315 	set_dev_node(dev, orig_node);
5316 	if (!tx_ring->desc)
5317 		tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5318 						   &tx_ring->dma, GFP_KERNEL);
5319 	if (!tx_ring->desc)
5320 		goto err;
5321 
5322 	tx_ring->next_to_use = 0;
5323 	tx_ring->next_to_clean = 0;
5324 	return 0;
5325 
5326 err:
5327 	vfree(tx_ring->tx_buffer_info);
5328 	tx_ring->tx_buffer_info = NULL;
5329 	dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5330 	return -ENOMEM;
5331 }
5332 
5333 /**
5334  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5335  * @adapter: board private structure
5336  *
5337  * If this function returns with an error, then it's possible one or
5338  * more of the rings is populated (while the rest are not).  It is the
5339  * callers duty to clean those orphaned rings.
5340  *
5341  * Return 0 on success, negative on failure
5342  **/
5343 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5344 {
5345 	int i, err = 0;
5346 
5347 	for (i = 0; i < adapter->num_tx_queues; i++) {
5348 		err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5349 		if (!err)
5350 			continue;
5351 
5352 		e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5353 		goto err_setup_tx;
5354 	}
5355 
5356 	return 0;
5357 err_setup_tx:
5358 	/* rewind the index freeing the rings as we go */
5359 	while (i--)
5360 		ixgbe_free_tx_resources(adapter->tx_ring[i]);
5361 	return err;
5362 }
5363 
5364 /**
5365  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5366  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
5367  *
5368  * Returns 0 on success, negative on failure
5369  **/
5370 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5371 {
5372 	struct device *dev = rx_ring->dev;
5373 	int orig_node = dev_to_node(dev);
5374 	int ring_node = -1;
5375 	int size;
5376 
5377 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5378 
5379 	if (rx_ring->q_vector)
5380 		ring_node = rx_ring->q_vector->numa_node;
5381 
5382 	rx_ring->rx_buffer_info = vzalloc_node(size, ring_node);
5383 	if (!rx_ring->rx_buffer_info)
5384 		rx_ring->rx_buffer_info = vzalloc(size);
5385 	if (!rx_ring->rx_buffer_info)
5386 		goto err;
5387 
5388 	u64_stats_init(&rx_ring->syncp);
5389 
5390 	/* Round up to nearest 4K */
5391 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5392 	rx_ring->size = ALIGN(rx_ring->size, 4096);
5393 
5394 	set_dev_node(dev, ring_node);
5395 	rx_ring->desc = dma_alloc_coherent(dev,
5396 					   rx_ring->size,
5397 					   &rx_ring->dma,
5398 					   GFP_KERNEL);
5399 	set_dev_node(dev, orig_node);
5400 	if (!rx_ring->desc)
5401 		rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5402 						   &rx_ring->dma, GFP_KERNEL);
5403 	if (!rx_ring->desc)
5404 		goto err;
5405 
5406 	rx_ring->next_to_clean = 0;
5407 	rx_ring->next_to_use = 0;
5408 
5409 	return 0;
5410 err:
5411 	vfree(rx_ring->rx_buffer_info);
5412 	rx_ring->rx_buffer_info = NULL;
5413 	dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5414 	return -ENOMEM;
5415 }
5416 
5417 /**
5418  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5419  * @adapter: board private structure
5420  *
5421  * If this function returns with an error, then it's possible one or
5422  * more of the rings is populated (while the rest are not).  It is the
5423  * callers duty to clean those orphaned rings.
5424  *
5425  * Return 0 on success, negative on failure
5426  **/
5427 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5428 {
5429 	int i, err = 0;
5430 
5431 	for (i = 0; i < adapter->num_rx_queues; i++) {
5432 		err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5433 		if (!err)
5434 			continue;
5435 
5436 		e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5437 		goto err_setup_rx;
5438 	}
5439 
5440 #ifdef IXGBE_FCOE
5441 	err = ixgbe_setup_fcoe_ddp_resources(adapter);
5442 	if (!err)
5443 #endif
5444 		return 0;
5445 err_setup_rx:
5446 	/* rewind the index freeing the rings as we go */
5447 	while (i--)
5448 		ixgbe_free_rx_resources(adapter->rx_ring[i]);
5449 	return err;
5450 }
5451 
5452 /**
5453  * ixgbe_free_tx_resources - Free Tx Resources per Queue
5454  * @tx_ring: Tx descriptor ring for a specific queue
5455  *
5456  * Free all transmit software resources
5457  **/
5458 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5459 {
5460 	ixgbe_clean_tx_ring(tx_ring);
5461 
5462 	vfree(tx_ring->tx_buffer_info);
5463 	tx_ring->tx_buffer_info = NULL;
5464 
5465 	/* if not set, then don't free */
5466 	if (!tx_ring->desc)
5467 		return;
5468 
5469 	dma_free_coherent(tx_ring->dev, tx_ring->size,
5470 			  tx_ring->desc, tx_ring->dma);
5471 
5472 	tx_ring->desc = NULL;
5473 }
5474 
5475 /**
5476  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5477  * @adapter: board private structure
5478  *
5479  * Free all transmit software resources
5480  **/
5481 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5482 {
5483 	int i;
5484 
5485 	for (i = 0; i < adapter->num_tx_queues; i++)
5486 		if (adapter->tx_ring[i]->desc)
5487 			ixgbe_free_tx_resources(adapter->tx_ring[i]);
5488 }
5489 
5490 /**
5491  * ixgbe_free_rx_resources - Free Rx Resources
5492  * @rx_ring: ring to clean the resources from
5493  *
5494  * Free all receive software resources
5495  **/
5496 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5497 {
5498 	ixgbe_clean_rx_ring(rx_ring);
5499 
5500 	vfree(rx_ring->rx_buffer_info);
5501 	rx_ring->rx_buffer_info = NULL;
5502 
5503 	/* if not set, then don't free */
5504 	if (!rx_ring->desc)
5505 		return;
5506 
5507 	dma_free_coherent(rx_ring->dev, rx_ring->size,
5508 			  rx_ring->desc, rx_ring->dma);
5509 
5510 	rx_ring->desc = NULL;
5511 }
5512 
5513 /**
5514  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5515  * @adapter: board private structure
5516  *
5517  * Free all receive software resources
5518  **/
5519 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5520 {
5521 	int i;
5522 
5523 #ifdef IXGBE_FCOE
5524 	ixgbe_free_fcoe_ddp_resources(adapter);
5525 
5526 #endif
5527 	for (i = 0; i < adapter->num_rx_queues; i++)
5528 		if (adapter->rx_ring[i]->desc)
5529 			ixgbe_free_rx_resources(adapter->rx_ring[i]);
5530 }
5531 
5532 /**
5533  * ixgbe_change_mtu - Change the Maximum Transfer Unit
5534  * @netdev: network interface device structure
5535  * @new_mtu: new value for maximum frame size
5536  *
5537  * Returns 0 on success, negative on failure
5538  **/
5539 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5540 {
5541 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
5542 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5543 
5544 	/* MTU < 68 is an error and causes problems on some kernels */
5545 	if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5546 		return -EINVAL;
5547 
5548 	/*
5549 	 * For 82599EB we cannot allow legacy VFs to enable their receive
5550 	 * paths when MTU greater than 1500 is configured.  So display a
5551 	 * warning that legacy VFs will be disabled.
5552 	 */
5553 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
5554 	    (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
5555 	    (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
5556 		e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
5557 
5558 	e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5559 
5560 	/* must set new MTU before calling down or up */
5561 	netdev->mtu = new_mtu;
5562 
5563 	if (netif_running(netdev))
5564 		ixgbe_reinit_locked(adapter);
5565 
5566 	return 0;
5567 }
5568 
5569 /**
5570  * ixgbe_open - Called when a network interface is made active
5571  * @netdev: network interface device structure
5572  *
5573  * Returns 0 on success, negative value on failure
5574  *
5575  * The open entry point is called when a network interface is made
5576  * active by the system (IFF_UP).  At this point all resources needed
5577  * for transmit and receive operations are allocated, the interrupt
5578  * handler is registered with the OS, the watchdog timer is started,
5579  * and the stack is notified that the interface is ready.
5580  **/
5581 static int ixgbe_open(struct net_device *netdev)
5582 {
5583 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
5584 	int err, queues;
5585 
5586 	/* disallow open during test */
5587 	if (test_bit(__IXGBE_TESTING, &adapter->state))
5588 		return -EBUSY;
5589 
5590 	netif_carrier_off(netdev);
5591 
5592 	/* allocate transmit descriptors */
5593 	err = ixgbe_setup_all_tx_resources(adapter);
5594 	if (err)
5595 		goto err_setup_tx;
5596 
5597 	/* allocate receive descriptors */
5598 	err = ixgbe_setup_all_rx_resources(adapter);
5599 	if (err)
5600 		goto err_setup_rx;
5601 
5602 	ixgbe_configure(adapter);
5603 
5604 	err = ixgbe_request_irq(adapter);
5605 	if (err)
5606 		goto err_req_irq;
5607 
5608 	/* Notify the stack of the actual queue counts. */
5609 	if (adapter->num_rx_pools > 1)
5610 		queues = adapter->num_rx_queues_per_pool;
5611 	else
5612 		queues = adapter->num_tx_queues;
5613 
5614 	err = netif_set_real_num_tx_queues(netdev, queues);
5615 	if (err)
5616 		goto err_set_queues;
5617 
5618 	if (adapter->num_rx_pools > 1 &&
5619 	    adapter->num_rx_queues > IXGBE_MAX_L2A_QUEUES)
5620 		queues = IXGBE_MAX_L2A_QUEUES;
5621 	else
5622 		queues = adapter->num_rx_queues;
5623 	err = netif_set_real_num_rx_queues(netdev, queues);
5624 	if (err)
5625 		goto err_set_queues;
5626 
5627 	ixgbe_ptp_init(adapter);
5628 
5629 	ixgbe_up_complete(adapter);
5630 
5631 	return 0;
5632 
5633 err_set_queues:
5634 	ixgbe_free_irq(adapter);
5635 err_req_irq:
5636 	ixgbe_free_all_rx_resources(adapter);
5637 err_setup_rx:
5638 	ixgbe_free_all_tx_resources(adapter);
5639 err_setup_tx:
5640 	ixgbe_reset(adapter);
5641 
5642 	return err;
5643 }
5644 
5645 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
5646 {
5647 	ixgbe_ptp_suspend(adapter);
5648 
5649 	ixgbe_down(adapter);
5650 	ixgbe_free_irq(adapter);
5651 
5652 	ixgbe_free_all_tx_resources(adapter);
5653 	ixgbe_free_all_rx_resources(adapter);
5654 }
5655 
5656 /**
5657  * ixgbe_close - Disables a network interface
5658  * @netdev: network interface device structure
5659  *
5660  * Returns 0, this is not allowed to fail
5661  *
5662  * The close entry point is called when an interface is de-activated
5663  * by the OS.  The hardware is still under the drivers control, but
5664  * needs to be disabled.  A global MAC reset is issued to stop the
5665  * hardware, and all transmit and receive resources are freed.
5666  **/
5667 static int ixgbe_close(struct net_device *netdev)
5668 {
5669 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
5670 
5671 	ixgbe_ptp_stop(adapter);
5672 
5673 	ixgbe_close_suspend(adapter);
5674 
5675 	ixgbe_fdir_filter_exit(adapter);
5676 
5677 	ixgbe_release_hw_control(adapter);
5678 
5679 	return 0;
5680 }
5681 
5682 #ifdef CONFIG_PM
5683 static int ixgbe_resume(struct pci_dev *pdev)
5684 {
5685 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5686 	struct net_device *netdev = adapter->netdev;
5687 	u32 err;
5688 
5689 	adapter->hw.hw_addr = adapter->io_addr;
5690 	pci_set_power_state(pdev, PCI_D0);
5691 	pci_restore_state(pdev);
5692 	/*
5693 	 * pci_restore_state clears dev->state_saved so call
5694 	 * pci_save_state to restore it.
5695 	 */
5696 	pci_save_state(pdev);
5697 
5698 	err = pci_enable_device_mem(pdev);
5699 	if (err) {
5700 		e_dev_err("Cannot enable PCI device from suspend\n");
5701 		return err;
5702 	}
5703 	smp_mb__before_atomic();
5704 	clear_bit(__IXGBE_DISABLED, &adapter->state);
5705 	pci_set_master(pdev);
5706 
5707 	pci_wake_from_d3(pdev, false);
5708 
5709 	ixgbe_reset(adapter);
5710 
5711 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5712 
5713 	rtnl_lock();
5714 	err = ixgbe_init_interrupt_scheme(adapter);
5715 	if (!err && netif_running(netdev))
5716 		err = ixgbe_open(netdev);
5717 
5718 	rtnl_unlock();
5719 
5720 	if (err)
5721 		return err;
5722 
5723 	netif_device_attach(netdev);
5724 
5725 	return 0;
5726 }
5727 #endif /* CONFIG_PM */
5728 
5729 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5730 {
5731 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5732 	struct net_device *netdev = adapter->netdev;
5733 	struct ixgbe_hw *hw = &adapter->hw;
5734 	u32 ctrl, fctrl;
5735 	u32 wufc = adapter->wol;
5736 #ifdef CONFIG_PM
5737 	int retval = 0;
5738 #endif
5739 
5740 	netif_device_detach(netdev);
5741 
5742 	rtnl_lock();
5743 	if (netif_running(netdev))
5744 		ixgbe_close_suspend(adapter);
5745 	rtnl_unlock();
5746 
5747 	ixgbe_clear_interrupt_scheme(adapter);
5748 
5749 #ifdef CONFIG_PM
5750 	retval = pci_save_state(pdev);
5751 	if (retval)
5752 		return retval;
5753 
5754 #endif
5755 	if (hw->mac.ops.stop_link_on_d3)
5756 		hw->mac.ops.stop_link_on_d3(hw);
5757 
5758 	if (wufc) {
5759 		ixgbe_set_rx_mode(netdev);
5760 
5761 		/* enable the optics for 82599 SFP+ fiber as we can WoL */
5762 		if (hw->mac.ops.enable_tx_laser)
5763 			hw->mac.ops.enable_tx_laser(hw);
5764 
5765 		/* turn on all-multi mode if wake on multicast is enabled */
5766 		if (wufc & IXGBE_WUFC_MC) {
5767 			fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5768 			fctrl |= IXGBE_FCTRL_MPE;
5769 			IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5770 		}
5771 
5772 		ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5773 		ctrl |= IXGBE_CTRL_GIO_DIS;
5774 		IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5775 
5776 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5777 	} else {
5778 		IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5779 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5780 	}
5781 
5782 	switch (hw->mac.type) {
5783 	case ixgbe_mac_82598EB:
5784 		pci_wake_from_d3(pdev, false);
5785 		break;
5786 	case ixgbe_mac_82599EB:
5787 	case ixgbe_mac_X540:
5788 	case ixgbe_mac_X550:
5789 	case ixgbe_mac_X550EM_x:
5790 		pci_wake_from_d3(pdev, !!wufc);
5791 		break;
5792 	default:
5793 		break;
5794 	}
5795 
5796 	*enable_wake = !!wufc;
5797 
5798 	ixgbe_release_hw_control(adapter);
5799 
5800 	if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
5801 		pci_disable_device(pdev);
5802 
5803 	return 0;
5804 }
5805 
5806 #ifdef CONFIG_PM
5807 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5808 {
5809 	int retval;
5810 	bool wake;
5811 
5812 	retval = __ixgbe_shutdown(pdev, &wake);
5813 	if (retval)
5814 		return retval;
5815 
5816 	if (wake) {
5817 		pci_prepare_to_sleep(pdev);
5818 	} else {
5819 		pci_wake_from_d3(pdev, false);
5820 		pci_set_power_state(pdev, PCI_D3hot);
5821 	}
5822 
5823 	return 0;
5824 }
5825 #endif /* CONFIG_PM */
5826 
5827 static void ixgbe_shutdown(struct pci_dev *pdev)
5828 {
5829 	bool wake;
5830 
5831 	__ixgbe_shutdown(pdev, &wake);
5832 
5833 	if (system_state == SYSTEM_POWER_OFF) {
5834 		pci_wake_from_d3(pdev, wake);
5835 		pci_set_power_state(pdev, PCI_D3hot);
5836 	}
5837 }
5838 
5839 /**
5840  * ixgbe_update_stats - Update the board statistics counters.
5841  * @adapter: board private structure
5842  **/
5843 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5844 {
5845 	struct net_device *netdev = adapter->netdev;
5846 	struct ixgbe_hw *hw = &adapter->hw;
5847 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
5848 	u64 total_mpc = 0;
5849 	u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5850 	u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5851 	u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5852 	u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
5853 
5854 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5855 	    test_bit(__IXGBE_RESETTING, &adapter->state))
5856 		return;
5857 
5858 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5859 		u64 rsc_count = 0;
5860 		u64 rsc_flush = 0;
5861 		for (i = 0; i < adapter->num_rx_queues; i++) {
5862 			rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5863 			rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5864 		}
5865 		adapter->rsc_total_count = rsc_count;
5866 		adapter->rsc_total_flush = rsc_flush;
5867 	}
5868 
5869 	for (i = 0; i < adapter->num_rx_queues; i++) {
5870 		struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5871 		non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5872 		alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5873 		alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5874 		hw_csum_rx_error += rx_ring->rx_stats.csum_err;
5875 		bytes += rx_ring->stats.bytes;
5876 		packets += rx_ring->stats.packets;
5877 	}
5878 	adapter->non_eop_descs = non_eop_descs;
5879 	adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5880 	adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5881 	adapter->hw_csum_rx_error = hw_csum_rx_error;
5882 	netdev->stats.rx_bytes = bytes;
5883 	netdev->stats.rx_packets = packets;
5884 
5885 	bytes = 0;
5886 	packets = 0;
5887 	/* gather some stats to the adapter struct that are per queue */
5888 	for (i = 0; i < adapter->num_tx_queues; i++) {
5889 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5890 		restart_queue += tx_ring->tx_stats.restart_queue;
5891 		tx_busy += tx_ring->tx_stats.tx_busy;
5892 		bytes += tx_ring->stats.bytes;
5893 		packets += tx_ring->stats.packets;
5894 	}
5895 	adapter->restart_queue = restart_queue;
5896 	adapter->tx_busy = tx_busy;
5897 	netdev->stats.tx_bytes = bytes;
5898 	netdev->stats.tx_packets = packets;
5899 
5900 	hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5901 
5902 	/* 8 register reads */
5903 	for (i = 0; i < 8; i++) {
5904 		/* for packet buffers not used, the register should read 0 */
5905 		mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5906 		missed_rx += mpc;
5907 		hwstats->mpc[i] += mpc;
5908 		total_mpc += hwstats->mpc[i];
5909 		hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5910 		hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5911 		switch (hw->mac.type) {
5912 		case ixgbe_mac_82598EB:
5913 			hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5914 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5915 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5916 			hwstats->pxonrxc[i] +=
5917 				IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5918 			break;
5919 		case ixgbe_mac_82599EB:
5920 		case ixgbe_mac_X540:
5921 		case ixgbe_mac_X550:
5922 		case ixgbe_mac_X550EM_x:
5923 			hwstats->pxonrxc[i] +=
5924 				IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5925 			break;
5926 		default:
5927 			break;
5928 		}
5929 	}
5930 
5931 	/*16 register reads */
5932 	for (i = 0; i < 16; i++) {
5933 		hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5934 		hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5935 		if ((hw->mac.type == ixgbe_mac_82599EB) ||
5936 		    (hw->mac.type == ixgbe_mac_X540) ||
5937 		    (hw->mac.type == ixgbe_mac_X550) ||
5938 		    (hw->mac.type == ixgbe_mac_X550EM_x)) {
5939 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
5940 			IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
5941 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
5942 			IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
5943 		}
5944 	}
5945 
5946 	hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5947 	/* work around hardware counting issue */
5948 	hwstats->gprc -= missed_rx;
5949 
5950 	ixgbe_update_xoff_received(adapter);
5951 
5952 	/* 82598 hardware only has a 32 bit counter in the high register */
5953 	switch (hw->mac.type) {
5954 	case ixgbe_mac_82598EB:
5955 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5956 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5957 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5958 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5959 		break;
5960 	case ixgbe_mac_X540:
5961 	case ixgbe_mac_X550:
5962 	case ixgbe_mac_X550EM_x:
5963 		/* OS2BMC stats are X540 and later */
5964 		hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5965 		hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5966 		hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5967 		hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5968 	case ixgbe_mac_82599EB:
5969 		for (i = 0; i < 16; i++)
5970 			adapter->hw_rx_no_dma_resources +=
5971 					     IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5972 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5973 		IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5974 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5975 		IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5976 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5977 		IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5978 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5979 		hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5980 		hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5981 #ifdef IXGBE_FCOE
5982 		hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5983 		hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5984 		hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5985 		hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5986 		hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5987 		hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5988 		/* Add up per cpu counters for total ddp aloc fail */
5989 		if (adapter->fcoe.ddp_pool) {
5990 			struct ixgbe_fcoe *fcoe = &adapter->fcoe;
5991 			struct ixgbe_fcoe_ddp_pool *ddp_pool;
5992 			unsigned int cpu;
5993 			u64 noddp = 0, noddp_ext_buff = 0;
5994 			for_each_possible_cpu(cpu) {
5995 				ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
5996 				noddp += ddp_pool->noddp;
5997 				noddp_ext_buff += ddp_pool->noddp_ext_buff;
5998 			}
5999 			hwstats->fcoe_noddp = noddp;
6000 			hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
6001 		}
6002 #endif /* IXGBE_FCOE */
6003 		break;
6004 	default:
6005 		break;
6006 	}
6007 	bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
6008 	hwstats->bprc += bprc;
6009 	hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
6010 	if (hw->mac.type == ixgbe_mac_82598EB)
6011 		hwstats->mprc -= bprc;
6012 	hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
6013 	hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
6014 	hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
6015 	hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
6016 	hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
6017 	hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
6018 	hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
6019 	hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
6020 	lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
6021 	hwstats->lxontxc += lxon;
6022 	lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
6023 	hwstats->lxofftxc += lxoff;
6024 	hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
6025 	hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
6026 	/*
6027 	 * 82598 errata - tx of flow control packets is included in tx counters
6028 	 */
6029 	xon_off_tot = lxon + lxoff;
6030 	hwstats->gptc -= xon_off_tot;
6031 	hwstats->mptc -= xon_off_tot;
6032 	hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
6033 	hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
6034 	hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
6035 	hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
6036 	hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
6037 	hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
6038 	hwstats->ptc64 -= xon_off_tot;
6039 	hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
6040 	hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
6041 	hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
6042 	hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
6043 	hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
6044 	hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
6045 
6046 	/* Fill out the OS statistics structure */
6047 	netdev->stats.multicast = hwstats->mprc;
6048 
6049 	/* Rx Errors */
6050 	netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
6051 	netdev->stats.rx_dropped = 0;
6052 	netdev->stats.rx_length_errors = hwstats->rlec;
6053 	netdev->stats.rx_crc_errors = hwstats->crcerrs;
6054 	netdev->stats.rx_missed_errors = total_mpc;
6055 }
6056 
6057 /**
6058  * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
6059  * @adapter: pointer to the device adapter structure
6060  **/
6061 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
6062 {
6063 	struct ixgbe_hw *hw = &adapter->hw;
6064 	int i;
6065 
6066 	if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
6067 		return;
6068 
6069 	adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6070 
6071 	/* if interface is down do nothing */
6072 	if (test_bit(__IXGBE_DOWN, &adapter->state))
6073 		return;
6074 
6075 	/* do nothing if we are not using signature filters */
6076 	if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
6077 		return;
6078 
6079 	adapter->fdir_overflow++;
6080 
6081 	if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
6082 		for (i = 0; i < adapter->num_tx_queues; i++)
6083 			set_bit(__IXGBE_TX_FDIR_INIT_DONE,
6084 				&(adapter->tx_ring[i]->state));
6085 		/* re-enable flow director interrupts */
6086 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
6087 	} else {
6088 		e_err(probe, "failed to finish FDIR re-initialization, "
6089 		      "ignored adding FDIR ATR filters\n");
6090 	}
6091 }
6092 
6093 /**
6094  * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
6095  * @adapter: pointer to the device adapter structure
6096  *
6097  * This function serves two purposes.  First it strobes the interrupt lines
6098  * in order to make certain interrupts are occurring.  Secondly it sets the
6099  * bits needed to check for TX hangs.  As a result we should immediately
6100  * determine if a hang has occurred.
6101  */
6102 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
6103 {
6104 	struct ixgbe_hw *hw = &adapter->hw;
6105 	u64 eics = 0;
6106 	int i;
6107 
6108 	/* If we're down, removing or resetting, just bail */
6109 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6110 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
6111 	    test_bit(__IXGBE_RESETTING, &adapter->state))
6112 		return;
6113 
6114 	/* Force detection of hung controller */
6115 	if (netif_carrier_ok(adapter->netdev)) {
6116 		for (i = 0; i < adapter->num_tx_queues; i++)
6117 			set_check_for_tx_hang(adapter->tx_ring[i]);
6118 	}
6119 
6120 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
6121 		/*
6122 		 * for legacy and MSI interrupts don't set any bits
6123 		 * that are enabled for EIAM, because this operation
6124 		 * would set *both* EIMS and EICS for any bit in EIAM
6125 		 */
6126 		IXGBE_WRITE_REG(hw, IXGBE_EICS,
6127 			(IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
6128 	} else {
6129 		/* get one bit for every active tx/rx interrupt vector */
6130 		for (i = 0; i < adapter->num_q_vectors; i++) {
6131 			struct ixgbe_q_vector *qv = adapter->q_vector[i];
6132 			if (qv->rx.ring || qv->tx.ring)
6133 				eics |= ((u64)1 << i);
6134 		}
6135 	}
6136 
6137 	/* Cause software interrupt to ensure rings are cleaned */
6138 	ixgbe_irq_rearm_queues(adapter, eics);
6139 
6140 }
6141 
6142 /**
6143  * ixgbe_watchdog_update_link - update the link status
6144  * @adapter: pointer to the device adapter structure
6145  * @link_speed: pointer to a u32 to store the link_speed
6146  **/
6147 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
6148 {
6149 	struct ixgbe_hw *hw = &adapter->hw;
6150 	u32 link_speed = adapter->link_speed;
6151 	bool link_up = adapter->link_up;
6152 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
6153 
6154 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
6155 		return;
6156 
6157 	if (hw->mac.ops.check_link) {
6158 		hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
6159 	} else {
6160 		/* always assume link is up, if no check link function */
6161 		link_speed = IXGBE_LINK_SPEED_10GB_FULL;
6162 		link_up = true;
6163 	}
6164 
6165 	if (adapter->ixgbe_ieee_pfc)
6166 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
6167 
6168 	if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
6169 		hw->mac.ops.fc_enable(hw);
6170 		ixgbe_set_rx_drop_en(adapter);
6171 	}
6172 
6173 	if (link_up ||
6174 	    time_after(jiffies, (adapter->link_check_timeout +
6175 				 IXGBE_TRY_LINK_TIMEOUT))) {
6176 		adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6177 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
6178 		IXGBE_WRITE_FLUSH(hw);
6179 	}
6180 
6181 	adapter->link_up = link_up;
6182 	adapter->link_speed = link_speed;
6183 }
6184 
6185 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
6186 {
6187 #ifdef CONFIG_IXGBE_DCB
6188 	struct net_device *netdev = adapter->netdev;
6189 	struct dcb_app app = {
6190 			      .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
6191 			      .protocol = 0,
6192 			     };
6193 	u8 up = 0;
6194 
6195 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
6196 		up = dcb_ieee_getapp_mask(netdev, &app);
6197 
6198 	adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
6199 #endif
6200 }
6201 
6202 /**
6203  * ixgbe_watchdog_link_is_up - update netif_carrier status and
6204  *                             print link up message
6205  * @adapter: pointer to the device adapter structure
6206  **/
6207 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
6208 {
6209 	struct net_device *netdev = adapter->netdev;
6210 	struct ixgbe_hw *hw = &adapter->hw;
6211 	struct net_device *upper;
6212 	struct list_head *iter;
6213 	u32 link_speed = adapter->link_speed;
6214 	bool flow_rx, flow_tx;
6215 
6216 	/* only continue if link was previously down */
6217 	if (netif_carrier_ok(netdev))
6218 		return;
6219 
6220 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
6221 
6222 	switch (hw->mac.type) {
6223 	case ixgbe_mac_82598EB: {
6224 		u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6225 		u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
6226 		flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
6227 		flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
6228 	}
6229 		break;
6230 	case ixgbe_mac_X540:
6231 	case ixgbe_mac_X550:
6232 	case ixgbe_mac_X550EM_x:
6233 	case ixgbe_mac_82599EB: {
6234 		u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
6235 		u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
6236 		flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
6237 		flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
6238 	}
6239 		break;
6240 	default:
6241 		flow_tx = false;
6242 		flow_rx = false;
6243 		break;
6244 	}
6245 
6246 	adapter->last_rx_ptp_check = jiffies;
6247 
6248 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
6249 		ixgbe_ptp_start_cyclecounter(adapter);
6250 
6251 	e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
6252 	       (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
6253 	       "10 Gbps" :
6254 	       (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
6255 	       "1 Gbps" :
6256 	       (link_speed == IXGBE_LINK_SPEED_100_FULL ?
6257 	       "100 Mbps" :
6258 	       "unknown speed"))),
6259 	       ((flow_rx && flow_tx) ? "RX/TX" :
6260 	       (flow_rx ? "RX" :
6261 	       (flow_tx ? "TX" : "None"))));
6262 
6263 	netif_carrier_on(netdev);
6264 	ixgbe_check_vf_rate_limit(adapter);
6265 
6266 	/* enable transmits */
6267 	netif_tx_wake_all_queues(adapter->netdev);
6268 
6269 	/* enable any upper devices */
6270 	rtnl_lock();
6271 	netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
6272 		if (netif_is_macvlan(upper)) {
6273 			struct macvlan_dev *vlan = netdev_priv(upper);
6274 
6275 			if (vlan->fwd_priv)
6276 				netif_tx_wake_all_queues(upper);
6277 		}
6278 	}
6279 	rtnl_unlock();
6280 
6281 	/* update the default user priority for VFs */
6282 	ixgbe_update_default_up(adapter);
6283 
6284 	/* ping all the active vfs to let them know link has changed */
6285 	ixgbe_ping_all_vfs(adapter);
6286 }
6287 
6288 /**
6289  * ixgbe_watchdog_link_is_down - update netif_carrier status and
6290  *                               print link down message
6291  * @adapter: pointer to the adapter structure
6292  **/
6293 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
6294 {
6295 	struct net_device *netdev = adapter->netdev;
6296 	struct ixgbe_hw *hw = &adapter->hw;
6297 
6298 	adapter->link_up = false;
6299 	adapter->link_speed = 0;
6300 
6301 	/* only continue if link was up previously */
6302 	if (!netif_carrier_ok(netdev))
6303 		return;
6304 
6305 	/* poll for SFP+ cable when link is down */
6306 	if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
6307 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
6308 
6309 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
6310 		ixgbe_ptp_start_cyclecounter(adapter);
6311 
6312 	e_info(drv, "NIC Link is Down\n");
6313 	netif_carrier_off(netdev);
6314 
6315 	/* ping all the active vfs to let them know link has changed */
6316 	ixgbe_ping_all_vfs(adapter);
6317 }
6318 
6319 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
6320 {
6321 	int i;
6322 
6323 	for (i = 0; i < adapter->num_tx_queues; i++) {
6324 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
6325 
6326 		if (tx_ring->next_to_use != tx_ring->next_to_clean)
6327 			return true;
6328 	}
6329 
6330 	return false;
6331 }
6332 
6333 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
6334 {
6335 	struct ixgbe_hw *hw = &adapter->hw;
6336 	struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
6337 	u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
6338 
6339 	int i, j;
6340 
6341 	if (!adapter->num_vfs)
6342 		return false;
6343 
6344 	/* resetting the PF is only needed for MAC before X550 */
6345 	if (hw->mac.type >= ixgbe_mac_X550)
6346 		return false;
6347 
6348 	for (i = 0; i < adapter->num_vfs; i++) {
6349 		for (j = 0; j < q_per_pool; j++) {
6350 			u32 h, t;
6351 
6352 			h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
6353 			t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
6354 
6355 			if (h != t)
6356 				return true;
6357 		}
6358 	}
6359 
6360 	return false;
6361 }
6362 
6363 /**
6364  * ixgbe_watchdog_flush_tx - flush queues on link down
6365  * @adapter: pointer to the device adapter structure
6366  **/
6367 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
6368 {
6369 	if (!netif_carrier_ok(adapter->netdev)) {
6370 		if (ixgbe_ring_tx_pending(adapter) ||
6371 		    ixgbe_vf_tx_pending(adapter)) {
6372 			/* We've lost link, so the controller stops DMA,
6373 			 * but we've got queued Tx work that's never going
6374 			 * to get done, so reset controller to flush Tx.
6375 			 * (Do the reset outside of interrupt context).
6376 			 */
6377 			e_warn(drv, "initiating reset to clear Tx work after link loss\n");
6378 			adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
6379 		}
6380 	}
6381 }
6382 
6383 #ifdef CONFIG_PCI_IOV
6384 static inline void ixgbe_issue_vf_flr(struct ixgbe_adapter *adapter,
6385 				      struct pci_dev *vfdev)
6386 {
6387 	if (!pci_wait_for_pending_transaction(vfdev))
6388 		e_dev_warn("Issuing VFLR with pending transactions\n");
6389 
6390 	e_dev_err("Issuing VFLR for VF %s\n", pci_name(vfdev));
6391 	pcie_capability_set_word(vfdev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
6392 
6393 	msleep(100);
6394 }
6395 
6396 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
6397 {
6398 	struct ixgbe_hw *hw = &adapter->hw;
6399 	struct pci_dev *pdev = adapter->pdev;
6400 	struct pci_dev *vfdev;
6401 	u32 gpc;
6402 	int pos;
6403 	unsigned short vf_id;
6404 
6405 	if (!(netif_carrier_ok(adapter->netdev)))
6406 		return;
6407 
6408 	gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
6409 	if (gpc) /* If incrementing then no need for the check below */
6410 		return;
6411 	/* Check to see if a bad DMA write target from an errant or
6412 	 * malicious VF has caused a PCIe error.  If so then we can
6413 	 * issue a VFLR to the offending VF(s) and then resume without
6414 	 * requesting a full slot reset.
6415 	 */
6416 
6417 	if (!pdev)
6418 		return;
6419 
6420 	pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
6421 	if (!pos)
6422 		return;
6423 
6424 	/* get the device ID for the VF */
6425 	pci_read_config_word(pdev, pos + PCI_SRIOV_VF_DID, &vf_id);
6426 
6427 	/* check status reg for all VFs owned by this PF */
6428 	vfdev = pci_get_device(pdev->vendor, vf_id, NULL);
6429 	while (vfdev) {
6430 		if (vfdev->is_virtfn && (vfdev->physfn == pdev)) {
6431 			u16 status_reg;
6432 
6433 			pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
6434 			if (status_reg & PCI_STATUS_REC_MASTER_ABORT)
6435 				/* issue VFLR */
6436 				ixgbe_issue_vf_flr(adapter, vfdev);
6437 		}
6438 
6439 		vfdev = pci_get_device(pdev->vendor, vf_id, vfdev);
6440 	}
6441 }
6442 
6443 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
6444 {
6445 	u32 ssvpc;
6446 
6447 	/* Do not perform spoof check for 82598 or if not in IOV mode */
6448 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
6449 	    adapter->num_vfs == 0)
6450 		return;
6451 
6452 	ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
6453 
6454 	/*
6455 	 * ssvpc register is cleared on read, if zero then no
6456 	 * spoofed packets in the last interval.
6457 	 */
6458 	if (!ssvpc)
6459 		return;
6460 
6461 	e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
6462 }
6463 #else
6464 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
6465 {
6466 }
6467 
6468 static void
6469 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
6470 {
6471 }
6472 #endif /* CONFIG_PCI_IOV */
6473 
6474 
6475 /**
6476  * ixgbe_watchdog_subtask - check and bring link up
6477  * @adapter: pointer to the device adapter structure
6478  **/
6479 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
6480 {
6481 	/* if interface is down, removing or resetting, do nothing */
6482 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6483 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
6484 	    test_bit(__IXGBE_RESETTING, &adapter->state))
6485 		return;
6486 
6487 	ixgbe_watchdog_update_link(adapter);
6488 
6489 	if (adapter->link_up)
6490 		ixgbe_watchdog_link_is_up(adapter);
6491 	else
6492 		ixgbe_watchdog_link_is_down(adapter);
6493 
6494 	ixgbe_check_for_bad_vf(adapter);
6495 	ixgbe_spoof_check(adapter);
6496 	ixgbe_update_stats(adapter);
6497 
6498 	ixgbe_watchdog_flush_tx(adapter);
6499 }
6500 
6501 /**
6502  * ixgbe_sfp_detection_subtask - poll for SFP+ cable
6503  * @adapter: the ixgbe adapter structure
6504  **/
6505 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
6506 {
6507 	struct ixgbe_hw *hw = &adapter->hw;
6508 	s32 err;
6509 
6510 	/* not searching for SFP so there is nothing to do here */
6511 	if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
6512 	    !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6513 		return;
6514 
6515 	/* someone else is in init, wait until next service event */
6516 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6517 		return;
6518 
6519 	err = hw->phy.ops.identify_sfp(hw);
6520 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6521 		goto sfp_out;
6522 
6523 	if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
6524 		/* If no cable is present, then we need to reset
6525 		 * the next time we find a good cable. */
6526 		adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
6527 	}
6528 
6529 	/* exit on error */
6530 	if (err)
6531 		goto sfp_out;
6532 
6533 	/* exit if reset not needed */
6534 	if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6535 		goto sfp_out;
6536 
6537 	adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
6538 
6539 	/*
6540 	 * A module may be identified correctly, but the EEPROM may not have
6541 	 * support for that module.  setup_sfp() will fail in that case, so
6542 	 * we should not allow that module to load.
6543 	 */
6544 	if (hw->mac.type == ixgbe_mac_82598EB)
6545 		err = hw->phy.ops.reset(hw);
6546 	else
6547 		err = hw->mac.ops.setup_sfp(hw);
6548 
6549 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6550 		goto sfp_out;
6551 
6552 	adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
6553 	e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
6554 
6555 sfp_out:
6556 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6557 
6558 	if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
6559 	    (adapter->netdev->reg_state == NETREG_REGISTERED)) {
6560 		e_dev_err("failed to initialize because an unsupported "
6561 			  "SFP+ module type was detected.\n");
6562 		e_dev_err("Reload the driver after installing a "
6563 			  "supported module.\n");
6564 		unregister_netdev(adapter->netdev);
6565 	}
6566 }
6567 
6568 /**
6569  * ixgbe_sfp_link_config_subtask - set up link SFP after module install
6570  * @adapter: the ixgbe adapter structure
6571  **/
6572 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
6573 {
6574 	struct ixgbe_hw *hw = &adapter->hw;
6575 	u32 speed;
6576 	bool autoneg = false;
6577 
6578 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
6579 		return;
6580 
6581 	/* someone else is in init, wait until next service event */
6582 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6583 		return;
6584 
6585 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
6586 
6587 	speed = hw->phy.autoneg_advertised;
6588 	if ((!speed) && (hw->mac.ops.get_link_capabilities)) {
6589 		hw->mac.ops.get_link_capabilities(hw, &speed, &autoneg);
6590 
6591 		/* setup the highest link when no autoneg */
6592 		if (!autoneg) {
6593 			if (speed & IXGBE_LINK_SPEED_10GB_FULL)
6594 				speed = IXGBE_LINK_SPEED_10GB_FULL;
6595 		}
6596 	}
6597 
6598 	if (hw->mac.ops.setup_link)
6599 		hw->mac.ops.setup_link(hw, speed, true);
6600 
6601 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
6602 	adapter->link_check_timeout = jiffies;
6603 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6604 }
6605 
6606 /**
6607  * ixgbe_service_timer - Timer Call-back
6608  * @data: pointer to adapter cast into an unsigned long
6609  **/
6610 static void ixgbe_service_timer(unsigned long data)
6611 {
6612 	struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
6613 	unsigned long next_event_offset;
6614 
6615 	/* poll faster when waiting for link */
6616 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
6617 		next_event_offset = HZ / 10;
6618 	else
6619 		next_event_offset = HZ * 2;
6620 
6621 	/* Reset the timer */
6622 	mod_timer(&adapter->service_timer, next_event_offset + jiffies);
6623 
6624 	ixgbe_service_event_schedule(adapter);
6625 }
6626 
6627 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
6628 {
6629 	if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
6630 		return;
6631 
6632 	adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
6633 
6634 	/* If we're already down, removing or resetting, just bail */
6635 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6636 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
6637 	    test_bit(__IXGBE_RESETTING, &adapter->state))
6638 		return;
6639 
6640 	ixgbe_dump(adapter);
6641 	netdev_err(adapter->netdev, "Reset adapter\n");
6642 	adapter->tx_timeout_count++;
6643 
6644 	rtnl_lock();
6645 	ixgbe_reinit_locked(adapter);
6646 	rtnl_unlock();
6647 }
6648 
6649 /**
6650  * ixgbe_service_task - manages and runs subtasks
6651  * @work: pointer to work_struct containing our data
6652  **/
6653 static void ixgbe_service_task(struct work_struct *work)
6654 {
6655 	struct ixgbe_adapter *adapter = container_of(work,
6656 						     struct ixgbe_adapter,
6657 						     service_task);
6658 	if (ixgbe_removed(adapter->hw.hw_addr)) {
6659 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
6660 			rtnl_lock();
6661 			ixgbe_down(adapter);
6662 			rtnl_unlock();
6663 		}
6664 		ixgbe_service_event_complete(adapter);
6665 		return;
6666 	}
6667 	ixgbe_reset_subtask(adapter);
6668 	ixgbe_sfp_detection_subtask(adapter);
6669 	ixgbe_sfp_link_config_subtask(adapter);
6670 	ixgbe_check_overtemp_subtask(adapter);
6671 	ixgbe_watchdog_subtask(adapter);
6672 	ixgbe_fdir_reinit_subtask(adapter);
6673 	ixgbe_check_hang_subtask(adapter);
6674 
6675 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
6676 		ixgbe_ptp_overflow_check(adapter);
6677 		ixgbe_ptp_rx_hang(adapter);
6678 	}
6679 
6680 	ixgbe_service_event_complete(adapter);
6681 }
6682 
6683 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
6684 		     struct ixgbe_tx_buffer *first,
6685 		     u8 *hdr_len)
6686 {
6687 	struct sk_buff *skb = first->skb;
6688 	u32 vlan_macip_lens, type_tucmd;
6689 	u32 mss_l4len_idx, l4len;
6690 	int err;
6691 
6692 	if (skb->ip_summed != CHECKSUM_PARTIAL)
6693 		return 0;
6694 
6695 	if (!skb_is_gso(skb))
6696 		return 0;
6697 
6698 	err = skb_cow_head(skb, 0);
6699 	if (err < 0)
6700 		return err;
6701 
6702 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6703 	type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
6704 
6705 	if (first->protocol == htons(ETH_P_IP)) {
6706 		struct iphdr *iph = ip_hdr(skb);
6707 		iph->tot_len = 0;
6708 		iph->check = 0;
6709 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6710 							 iph->daddr, 0,
6711 							 IPPROTO_TCP,
6712 							 0);
6713 		type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6714 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
6715 				   IXGBE_TX_FLAGS_CSUM |
6716 				   IXGBE_TX_FLAGS_IPV4;
6717 	} else if (skb_is_gso_v6(skb)) {
6718 		ipv6_hdr(skb)->payload_len = 0;
6719 		tcp_hdr(skb)->check =
6720 		    ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6721 				     &ipv6_hdr(skb)->daddr,
6722 				     0, IPPROTO_TCP, 0);
6723 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
6724 				   IXGBE_TX_FLAGS_CSUM;
6725 	}
6726 
6727 	/* compute header lengths */
6728 	l4len = tcp_hdrlen(skb);
6729 	*hdr_len = skb_transport_offset(skb) + l4len;
6730 
6731 	/* update gso size and bytecount with header size */
6732 	first->gso_segs = skb_shinfo(skb)->gso_segs;
6733 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
6734 
6735 	/* mss_l4len_id: use 0 as index for TSO */
6736 	mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
6737 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
6738 
6739 	/* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
6740 	vlan_macip_lens = skb_network_header_len(skb);
6741 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6742 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6743 
6744 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
6745 			  mss_l4len_idx);
6746 
6747 	return 1;
6748 }
6749 
6750 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
6751 			  struct ixgbe_tx_buffer *first)
6752 {
6753 	struct sk_buff *skb = first->skb;
6754 	u32 vlan_macip_lens = 0;
6755 	u32 mss_l4len_idx = 0;
6756 	u32 type_tucmd = 0;
6757 
6758 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
6759 		if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
6760 		    !(first->tx_flags & IXGBE_TX_FLAGS_CC))
6761 			return;
6762 	} else {
6763 		u8 l4_hdr = 0;
6764 		switch (first->protocol) {
6765 		case htons(ETH_P_IP):
6766 			vlan_macip_lens |= skb_network_header_len(skb);
6767 			type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6768 			l4_hdr = ip_hdr(skb)->protocol;
6769 			break;
6770 		case htons(ETH_P_IPV6):
6771 			vlan_macip_lens |= skb_network_header_len(skb);
6772 			l4_hdr = ipv6_hdr(skb)->nexthdr;
6773 			break;
6774 		default:
6775 			if (unlikely(net_ratelimit())) {
6776 				dev_warn(tx_ring->dev,
6777 				 "partial checksum but proto=%x!\n",
6778 				 first->protocol);
6779 			}
6780 			break;
6781 		}
6782 
6783 		switch (l4_hdr) {
6784 		case IPPROTO_TCP:
6785 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6786 			mss_l4len_idx = tcp_hdrlen(skb) <<
6787 					IXGBE_ADVTXD_L4LEN_SHIFT;
6788 			break;
6789 		case IPPROTO_SCTP:
6790 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6791 			mss_l4len_idx = sizeof(struct sctphdr) <<
6792 					IXGBE_ADVTXD_L4LEN_SHIFT;
6793 			break;
6794 		case IPPROTO_UDP:
6795 			mss_l4len_idx = sizeof(struct udphdr) <<
6796 					IXGBE_ADVTXD_L4LEN_SHIFT;
6797 			break;
6798 		default:
6799 			if (unlikely(net_ratelimit())) {
6800 				dev_warn(tx_ring->dev,
6801 				 "partial checksum but l4 proto=%x!\n",
6802 				 l4_hdr);
6803 			}
6804 			break;
6805 		}
6806 
6807 		/* update TX checksum flag */
6808 		first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
6809 	}
6810 
6811 	/* vlan_macip_lens: MACLEN, VLAN tag */
6812 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6813 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6814 
6815 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
6816 			  type_tucmd, mss_l4len_idx);
6817 }
6818 
6819 #define IXGBE_SET_FLAG(_input, _flag, _result) \
6820 	((_flag <= _result) ? \
6821 	 ((u32)(_input & _flag) * (_result / _flag)) : \
6822 	 ((u32)(_input & _flag) / (_flag / _result)))
6823 
6824 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
6825 {
6826 	/* set type for advanced descriptor with frame checksum insertion */
6827 	u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
6828 		       IXGBE_ADVTXD_DCMD_DEXT |
6829 		       IXGBE_ADVTXD_DCMD_IFCS;
6830 
6831 	/* set HW vlan bit if vlan is present */
6832 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
6833 				   IXGBE_ADVTXD_DCMD_VLE);
6834 
6835 	/* set segmentation enable bits for TSO/FSO */
6836 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
6837 				   IXGBE_ADVTXD_DCMD_TSE);
6838 
6839 	/* set timestamp bit if present */
6840 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
6841 				   IXGBE_ADVTXD_MAC_TSTAMP);
6842 
6843 	/* insert frame checksum */
6844 	cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
6845 
6846 	return cmd_type;
6847 }
6848 
6849 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
6850 				   u32 tx_flags, unsigned int paylen)
6851 {
6852 	u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
6853 
6854 	/* enable L4 checksum for TSO and TX checksum offload */
6855 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
6856 					IXGBE_TX_FLAGS_CSUM,
6857 					IXGBE_ADVTXD_POPTS_TXSM);
6858 
6859 	/* enble IPv4 checksum for TSO */
6860 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
6861 					IXGBE_TX_FLAGS_IPV4,
6862 					IXGBE_ADVTXD_POPTS_IXSM);
6863 
6864 	/*
6865 	 * Check Context must be set if Tx switch is enabled, which it
6866 	 * always is for case where virtual functions are running
6867 	 */
6868 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
6869 					IXGBE_TX_FLAGS_CC,
6870 					IXGBE_ADVTXD_CC);
6871 
6872 	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6873 }
6874 
6875 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6876 {
6877 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6878 
6879 	/* Herbert's original patch had:
6880 	 *  smp_mb__after_netif_stop_queue();
6881 	 * but since that doesn't exist yet, just open code it.
6882 	 */
6883 	smp_mb();
6884 
6885 	/* We need to check again in a case another CPU has just
6886 	 * made room available.
6887 	 */
6888 	if (likely(ixgbe_desc_unused(tx_ring) < size))
6889 		return -EBUSY;
6890 
6891 	/* A reprieve! - use start_queue because it doesn't call schedule */
6892 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6893 	++tx_ring->tx_stats.restart_queue;
6894 	return 0;
6895 }
6896 
6897 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6898 {
6899 	if (likely(ixgbe_desc_unused(tx_ring) >= size))
6900 		return 0;
6901 
6902 	return __ixgbe_maybe_stop_tx(tx_ring, size);
6903 }
6904 
6905 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
6906 		       IXGBE_TXD_CMD_RS)
6907 
6908 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
6909 			 struct ixgbe_tx_buffer *first,
6910 			 const u8 hdr_len)
6911 {
6912 	struct sk_buff *skb = first->skb;
6913 	struct ixgbe_tx_buffer *tx_buffer;
6914 	union ixgbe_adv_tx_desc *tx_desc;
6915 	struct skb_frag_struct *frag;
6916 	dma_addr_t dma;
6917 	unsigned int data_len, size;
6918 	u32 tx_flags = first->tx_flags;
6919 	u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
6920 	u16 i = tx_ring->next_to_use;
6921 
6922 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
6923 
6924 	ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
6925 
6926 	size = skb_headlen(skb);
6927 	data_len = skb->data_len;
6928 
6929 #ifdef IXGBE_FCOE
6930 	if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6931 		if (data_len < sizeof(struct fcoe_crc_eof)) {
6932 			size -= sizeof(struct fcoe_crc_eof) - data_len;
6933 			data_len = 0;
6934 		} else {
6935 			data_len -= sizeof(struct fcoe_crc_eof);
6936 		}
6937 	}
6938 
6939 #endif
6940 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
6941 
6942 	tx_buffer = first;
6943 
6944 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
6945 		if (dma_mapping_error(tx_ring->dev, dma))
6946 			goto dma_error;
6947 
6948 		/* record length, and DMA address */
6949 		dma_unmap_len_set(tx_buffer, len, size);
6950 		dma_unmap_addr_set(tx_buffer, dma, dma);
6951 
6952 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
6953 
6954 		while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
6955 			tx_desc->read.cmd_type_len =
6956 				cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
6957 
6958 			i++;
6959 			tx_desc++;
6960 			if (i == tx_ring->count) {
6961 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6962 				i = 0;
6963 			}
6964 			tx_desc->read.olinfo_status = 0;
6965 
6966 			dma += IXGBE_MAX_DATA_PER_TXD;
6967 			size -= IXGBE_MAX_DATA_PER_TXD;
6968 
6969 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
6970 		}
6971 
6972 		if (likely(!data_len))
6973 			break;
6974 
6975 		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
6976 
6977 		i++;
6978 		tx_desc++;
6979 		if (i == tx_ring->count) {
6980 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6981 			i = 0;
6982 		}
6983 		tx_desc->read.olinfo_status = 0;
6984 
6985 #ifdef IXGBE_FCOE
6986 		size = min_t(unsigned int, data_len, skb_frag_size(frag));
6987 #else
6988 		size = skb_frag_size(frag);
6989 #endif
6990 		data_len -= size;
6991 
6992 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
6993 				       DMA_TO_DEVICE);
6994 
6995 		tx_buffer = &tx_ring->tx_buffer_info[i];
6996 	}
6997 
6998 	/* write last descriptor with RS and EOP bits */
6999 	cmd_type |= size | IXGBE_TXD_CMD;
7000 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
7001 
7002 	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
7003 
7004 	/* set the timestamp */
7005 	first->time_stamp = jiffies;
7006 
7007 	/*
7008 	 * Force memory writes to complete before letting h/w know there
7009 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
7010 	 * memory model archs, such as IA-64).
7011 	 *
7012 	 * We also need this memory barrier to make certain all of the
7013 	 * status bits have been updated before next_to_watch is written.
7014 	 */
7015 	wmb();
7016 
7017 	/* set next_to_watch value indicating a packet is present */
7018 	first->next_to_watch = tx_desc;
7019 
7020 	i++;
7021 	if (i == tx_ring->count)
7022 		i = 0;
7023 
7024 	tx_ring->next_to_use = i;
7025 
7026 	ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
7027 
7028 	if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) {
7029 		writel(i, tx_ring->tail);
7030 
7031 		/* we need this if more than one processor can write to our tail
7032 		 * at a time, it synchronizes IO on IA64/Altix systems
7033 		 */
7034 		mmiowb();
7035 	}
7036 
7037 	return;
7038 dma_error:
7039 	dev_err(tx_ring->dev, "TX DMA map failed\n");
7040 
7041 	/* clear dma mappings for failed tx_buffer_info map */
7042 	for (;;) {
7043 		tx_buffer = &tx_ring->tx_buffer_info[i];
7044 		ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
7045 		if (tx_buffer == first)
7046 			break;
7047 		if (i == 0)
7048 			i = tx_ring->count;
7049 		i--;
7050 	}
7051 
7052 	tx_ring->next_to_use = i;
7053 }
7054 
7055 static void ixgbe_atr(struct ixgbe_ring *ring,
7056 		      struct ixgbe_tx_buffer *first)
7057 {
7058 	struct ixgbe_q_vector *q_vector = ring->q_vector;
7059 	union ixgbe_atr_hash_dword input = { .dword = 0 };
7060 	union ixgbe_atr_hash_dword common = { .dword = 0 };
7061 	union {
7062 		unsigned char *network;
7063 		struct iphdr *ipv4;
7064 		struct ipv6hdr *ipv6;
7065 	} hdr;
7066 	struct tcphdr *th;
7067 	__be16 vlan_id;
7068 
7069 	/* if ring doesn't have a interrupt vector, cannot perform ATR */
7070 	if (!q_vector)
7071 		return;
7072 
7073 	/* do nothing if sampling is disabled */
7074 	if (!ring->atr_sample_rate)
7075 		return;
7076 
7077 	ring->atr_count++;
7078 
7079 	/* snag network header to get L4 type and address */
7080 	hdr.network = skb_network_header(first->skb);
7081 
7082 	/* Currently only IPv4/IPv6 with TCP is supported */
7083 	if ((first->protocol != htons(ETH_P_IPV6) ||
7084 	     hdr.ipv6->nexthdr != IPPROTO_TCP) &&
7085 	    (first->protocol != htons(ETH_P_IP) ||
7086 	     hdr.ipv4->protocol != IPPROTO_TCP))
7087 		return;
7088 
7089 	th = tcp_hdr(first->skb);
7090 
7091 	/* skip this packet since it is invalid or the socket is closing */
7092 	if (!th || th->fin)
7093 		return;
7094 
7095 	/* sample on all syn packets or once every atr sample count */
7096 	if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
7097 		return;
7098 
7099 	/* reset sample count */
7100 	ring->atr_count = 0;
7101 
7102 	vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
7103 
7104 	/*
7105 	 * src and dst are inverted, think how the receiver sees them
7106 	 *
7107 	 * The input is broken into two sections, a non-compressed section
7108 	 * containing vm_pool, vlan_id, and flow_type.  The rest of the data
7109 	 * is XORed together and stored in the compressed dword.
7110 	 */
7111 	input.formatted.vlan_id = vlan_id;
7112 
7113 	/*
7114 	 * since src port and flex bytes occupy the same word XOR them together
7115 	 * and write the value to source port portion of compressed dword
7116 	 */
7117 	if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
7118 		common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
7119 	else
7120 		common.port.src ^= th->dest ^ first->protocol;
7121 	common.port.dst ^= th->source;
7122 
7123 	if (first->protocol == htons(ETH_P_IP)) {
7124 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
7125 		common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
7126 	} else {
7127 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
7128 		common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
7129 			     hdr.ipv6->saddr.s6_addr32[1] ^
7130 			     hdr.ipv6->saddr.s6_addr32[2] ^
7131 			     hdr.ipv6->saddr.s6_addr32[3] ^
7132 			     hdr.ipv6->daddr.s6_addr32[0] ^
7133 			     hdr.ipv6->daddr.s6_addr32[1] ^
7134 			     hdr.ipv6->daddr.s6_addr32[2] ^
7135 			     hdr.ipv6->daddr.s6_addr32[3];
7136 	}
7137 
7138 	/* This assumes the Rx queue and Tx queue are bound to the same CPU */
7139 	ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
7140 					      input, common, ring->queue_index);
7141 }
7142 
7143 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
7144 			      void *accel_priv, select_queue_fallback_t fallback)
7145 {
7146 	struct ixgbe_fwd_adapter *fwd_adapter = accel_priv;
7147 #ifdef IXGBE_FCOE
7148 	struct ixgbe_adapter *adapter;
7149 	struct ixgbe_ring_feature *f;
7150 	int txq;
7151 #endif
7152 
7153 	if (fwd_adapter)
7154 		return skb->queue_mapping + fwd_adapter->tx_base_queue;
7155 
7156 #ifdef IXGBE_FCOE
7157 
7158 	/*
7159 	 * only execute the code below if protocol is FCoE
7160 	 * or FIP and we have FCoE enabled on the adapter
7161 	 */
7162 	switch (vlan_get_protocol(skb)) {
7163 	case htons(ETH_P_FCOE):
7164 	case htons(ETH_P_FIP):
7165 		adapter = netdev_priv(dev);
7166 
7167 		if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7168 			break;
7169 	default:
7170 		return fallback(dev, skb);
7171 	}
7172 
7173 	f = &adapter->ring_feature[RING_F_FCOE];
7174 
7175 	txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
7176 					   smp_processor_id();
7177 
7178 	while (txq >= f->indices)
7179 		txq -= f->indices;
7180 
7181 	return txq + f->offset;
7182 #else
7183 	return fallback(dev, skb);
7184 #endif
7185 }
7186 
7187 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
7188 			  struct ixgbe_adapter *adapter,
7189 			  struct ixgbe_ring *tx_ring)
7190 {
7191 	struct ixgbe_tx_buffer *first;
7192 	int tso;
7193 	u32 tx_flags = 0;
7194 	unsigned short f;
7195 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
7196 	__be16 protocol = skb->protocol;
7197 	u8 hdr_len = 0;
7198 
7199 	/*
7200 	 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
7201 	 *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
7202 	 *       + 2 desc gap to keep tail from touching head,
7203 	 *       + 1 desc for context descriptor,
7204 	 * otherwise try next time
7205 	 */
7206 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
7207 		count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
7208 
7209 	if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
7210 		tx_ring->tx_stats.tx_busy++;
7211 		return NETDEV_TX_BUSY;
7212 	}
7213 
7214 	/* record the location of the first descriptor for this packet */
7215 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
7216 	first->skb = skb;
7217 	first->bytecount = skb->len;
7218 	first->gso_segs = 1;
7219 
7220 	/* if we have a HW VLAN tag being added default to the HW one */
7221 	if (vlan_tx_tag_present(skb)) {
7222 		tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
7223 		tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
7224 	/* else if it is a SW VLAN check the next protocol and store the tag */
7225 	} else if (protocol == htons(ETH_P_8021Q)) {
7226 		struct vlan_hdr *vhdr, _vhdr;
7227 		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
7228 		if (!vhdr)
7229 			goto out_drop;
7230 
7231 		protocol = vhdr->h_vlan_encapsulated_proto;
7232 		tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
7233 				  IXGBE_TX_FLAGS_VLAN_SHIFT;
7234 		tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
7235 	}
7236 
7237 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
7238 	    adapter->ptp_clock &&
7239 	    !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
7240 				   &adapter->state)) {
7241 		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
7242 		tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
7243 
7244 		/* schedule check for Tx timestamp */
7245 		adapter->ptp_tx_skb = skb_get(skb);
7246 		adapter->ptp_tx_start = jiffies;
7247 		schedule_work(&adapter->ptp_tx_work);
7248 	}
7249 
7250 	skb_tx_timestamp(skb);
7251 
7252 #ifdef CONFIG_PCI_IOV
7253 	/*
7254 	 * Use the l2switch_enable flag - would be false if the DMA
7255 	 * Tx switch had been disabled.
7256 	 */
7257 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7258 		tx_flags |= IXGBE_TX_FLAGS_CC;
7259 
7260 #endif
7261 	/* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
7262 	if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
7263 	    ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
7264 	     (skb->priority != TC_PRIO_CONTROL))) {
7265 		tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
7266 		tx_flags |= (skb->priority & 0x7) <<
7267 					IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
7268 		if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
7269 			struct vlan_ethhdr *vhdr;
7270 
7271 			if (skb_cow_head(skb, 0))
7272 				goto out_drop;
7273 			vhdr = (struct vlan_ethhdr *)skb->data;
7274 			vhdr->h_vlan_TCI = htons(tx_flags >>
7275 						 IXGBE_TX_FLAGS_VLAN_SHIFT);
7276 		} else {
7277 			tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
7278 		}
7279 	}
7280 
7281 	/* record initial flags and protocol */
7282 	first->tx_flags = tx_flags;
7283 	first->protocol = protocol;
7284 
7285 #ifdef IXGBE_FCOE
7286 	/* setup tx offload for FCoE */
7287 	if ((protocol == htons(ETH_P_FCOE)) &&
7288 	    (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
7289 		tso = ixgbe_fso(tx_ring, first, &hdr_len);
7290 		if (tso < 0)
7291 			goto out_drop;
7292 
7293 		goto xmit_fcoe;
7294 	}
7295 
7296 #endif /* IXGBE_FCOE */
7297 	tso = ixgbe_tso(tx_ring, first, &hdr_len);
7298 	if (tso < 0)
7299 		goto out_drop;
7300 	else if (!tso)
7301 		ixgbe_tx_csum(tx_ring, first);
7302 
7303 	/* add the ATR filter if ATR is on */
7304 	if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
7305 		ixgbe_atr(tx_ring, first);
7306 
7307 #ifdef IXGBE_FCOE
7308 xmit_fcoe:
7309 #endif /* IXGBE_FCOE */
7310 	ixgbe_tx_map(tx_ring, first, hdr_len);
7311 
7312 	return NETDEV_TX_OK;
7313 
7314 out_drop:
7315 	dev_kfree_skb_any(first->skb);
7316 	first->skb = NULL;
7317 
7318 	return NETDEV_TX_OK;
7319 }
7320 
7321 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
7322 				      struct net_device *netdev,
7323 				      struct ixgbe_ring *ring)
7324 {
7325 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7326 	struct ixgbe_ring *tx_ring;
7327 
7328 	/*
7329 	 * The minimum packet size for olinfo paylen is 17 so pad the skb
7330 	 * in order to meet this minimum size requirement.
7331 	 */
7332 	if (skb_put_padto(skb, 17))
7333 		return NETDEV_TX_OK;
7334 
7335 	tx_ring = ring ? ring : adapter->tx_ring[skb->queue_mapping];
7336 
7337 	return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
7338 }
7339 
7340 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
7341 				    struct net_device *netdev)
7342 {
7343 	return __ixgbe_xmit_frame(skb, netdev, NULL);
7344 }
7345 
7346 /**
7347  * ixgbe_set_mac - Change the Ethernet Address of the NIC
7348  * @netdev: network interface device structure
7349  * @p: pointer to an address structure
7350  *
7351  * Returns 0 on success, negative on failure
7352  **/
7353 static int ixgbe_set_mac(struct net_device *netdev, void *p)
7354 {
7355 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7356 	struct ixgbe_hw *hw = &adapter->hw;
7357 	struct sockaddr *addr = p;
7358 	int ret;
7359 
7360 	if (!is_valid_ether_addr(addr->sa_data))
7361 		return -EADDRNOTAVAIL;
7362 
7363 	ixgbe_del_mac_filter(adapter, hw->mac.addr, VMDQ_P(0));
7364 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
7365 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
7366 
7367 	ret = ixgbe_add_mac_filter(adapter, hw->mac.addr, VMDQ_P(0));
7368 	return ret > 0 ? 0 : ret;
7369 }
7370 
7371 static int
7372 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
7373 {
7374 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7375 	struct ixgbe_hw *hw = &adapter->hw;
7376 	u16 value;
7377 	int rc;
7378 
7379 	if (prtad != hw->phy.mdio.prtad)
7380 		return -EINVAL;
7381 	rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
7382 	if (!rc)
7383 		rc = value;
7384 	return rc;
7385 }
7386 
7387 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
7388 			    u16 addr, u16 value)
7389 {
7390 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7391 	struct ixgbe_hw *hw = &adapter->hw;
7392 
7393 	if (prtad != hw->phy.mdio.prtad)
7394 		return -EINVAL;
7395 	return hw->phy.ops.write_reg(hw, addr, devad, value);
7396 }
7397 
7398 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
7399 {
7400 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7401 
7402 	switch (cmd) {
7403 	case SIOCSHWTSTAMP:
7404 		return ixgbe_ptp_set_ts_config(adapter, req);
7405 	case SIOCGHWTSTAMP:
7406 		return ixgbe_ptp_get_ts_config(adapter, req);
7407 	default:
7408 		return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
7409 	}
7410 }
7411 
7412 /**
7413  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
7414  * netdev->dev_addrs
7415  * @netdev: network interface device structure
7416  *
7417  * Returns non-zero on failure
7418  **/
7419 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
7420 {
7421 	int err = 0;
7422 	struct ixgbe_adapter *adapter = netdev_priv(dev);
7423 	struct ixgbe_hw *hw = &adapter->hw;
7424 
7425 	if (is_valid_ether_addr(hw->mac.san_addr)) {
7426 		rtnl_lock();
7427 		err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
7428 		rtnl_unlock();
7429 
7430 		/* update SAN MAC vmdq pool selection */
7431 		hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
7432 	}
7433 	return err;
7434 }
7435 
7436 /**
7437  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
7438  * netdev->dev_addrs
7439  * @netdev: network interface device structure
7440  *
7441  * Returns non-zero on failure
7442  **/
7443 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
7444 {
7445 	int err = 0;
7446 	struct ixgbe_adapter *adapter = netdev_priv(dev);
7447 	struct ixgbe_mac_info *mac = &adapter->hw.mac;
7448 
7449 	if (is_valid_ether_addr(mac->san_addr)) {
7450 		rtnl_lock();
7451 		err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
7452 		rtnl_unlock();
7453 	}
7454 	return err;
7455 }
7456 
7457 #ifdef CONFIG_NET_POLL_CONTROLLER
7458 /*
7459  * Polling 'interrupt' - used by things like netconsole to send skbs
7460  * without having to re-enable interrupts. It's not called while
7461  * the interrupt routine is executing.
7462  */
7463 static void ixgbe_netpoll(struct net_device *netdev)
7464 {
7465 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7466 	int i;
7467 
7468 	/* if interface is down do nothing */
7469 	if (test_bit(__IXGBE_DOWN, &adapter->state))
7470 		return;
7471 
7472 	adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
7473 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
7474 		for (i = 0; i < adapter->num_q_vectors; i++)
7475 			ixgbe_msix_clean_rings(0, adapter->q_vector[i]);
7476 	} else {
7477 		ixgbe_intr(adapter->pdev->irq, netdev);
7478 	}
7479 	adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
7480 }
7481 
7482 #endif
7483 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
7484 						   struct rtnl_link_stats64 *stats)
7485 {
7486 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7487 	int i;
7488 
7489 	rcu_read_lock();
7490 	for (i = 0; i < adapter->num_rx_queues; i++) {
7491 		struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
7492 		u64 bytes, packets;
7493 		unsigned int start;
7494 
7495 		if (ring) {
7496 			do {
7497 				start = u64_stats_fetch_begin_irq(&ring->syncp);
7498 				packets = ring->stats.packets;
7499 				bytes   = ring->stats.bytes;
7500 			} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
7501 			stats->rx_packets += packets;
7502 			stats->rx_bytes   += bytes;
7503 		}
7504 	}
7505 
7506 	for (i = 0; i < adapter->num_tx_queues; i++) {
7507 		struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
7508 		u64 bytes, packets;
7509 		unsigned int start;
7510 
7511 		if (ring) {
7512 			do {
7513 				start = u64_stats_fetch_begin_irq(&ring->syncp);
7514 				packets = ring->stats.packets;
7515 				bytes   = ring->stats.bytes;
7516 			} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
7517 			stats->tx_packets += packets;
7518 			stats->tx_bytes   += bytes;
7519 		}
7520 	}
7521 	rcu_read_unlock();
7522 	/* following stats updated by ixgbe_watchdog_task() */
7523 	stats->multicast	= netdev->stats.multicast;
7524 	stats->rx_errors	= netdev->stats.rx_errors;
7525 	stats->rx_length_errors	= netdev->stats.rx_length_errors;
7526 	stats->rx_crc_errors	= netdev->stats.rx_crc_errors;
7527 	stats->rx_missed_errors	= netdev->stats.rx_missed_errors;
7528 	return stats;
7529 }
7530 
7531 #ifdef CONFIG_IXGBE_DCB
7532 /**
7533  * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
7534  * @adapter: pointer to ixgbe_adapter
7535  * @tc: number of traffic classes currently enabled
7536  *
7537  * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
7538  * 802.1Q priority maps to a packet buffer that exists.
7539  */
7540 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
7541 {
7542 	struct ixgbe_hw *hw = &adapter->hw;
7543 	u32 reg, rsave;
7544 	int i;
7545 
7546 	/* 82598 have a static priority to TC mapping that can not
7547 	 * be changed so no validation is needed.
7548 	 */
7549 	if (hw->mac.type == ixgbe_mac_82598EB)
7550 		return;
7551 
7552 	reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
7553 	rsave = reg;
7554 
7555 	for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
7556 		u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
7557 
7558 		/* If up2tc is out of bounds default to zero */
7559 		if (up2tc > tc)
7560 			reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
7561 	}
7562 
7563 	if (reg != rsave)
7564 		IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
7565 
7566 	return;
7567 }
7568 
7569 /**
7570  * ixgbe_set_prio_tc_map - Configure netdev prio tc map
7571  * @adapter: Pointer to adapter struct
7572  *
7573  * Populate the netdev user priority to tc map
7574  */
7575 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
7576 {
7577 	struct net_device *dev = adapter->netdev;
7578 	struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
7579 	struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
7580 	u8 prio;
7581 
7582 	for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
7583 		u8 tc = 0;
7584 
7585 		if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
7586 			tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
7587 		else if (ets)
7588 			tc = ets->prio_tc[prio];
7589 
7590 		netdev_set_prio_tc_map(dev, prio, tc);
7591 	}
7592 }
7593 
7594 #endif /* CONFIG_IXGBE_DCB */
7595 /**
7596  * ixgbe_setup_tc - configure net_device for multiple traffic classes
7597  *
7598  * @netdev: net device to configure
7599  * @tc: number of traffic classes to enable
7600  */
7601 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
7602 {
7603 	struct ixgbe_adapter *adapter = netdev_priv(dev);
7604 	struct ixgbe_hw *hw = &adapter->hw;
7605 	bool pools;
7606 
7607 	/* Hardware supports up to 8 traffic classes */
7608 	if (tc > adapter->dcb_cfg.num_tcs.pg_tcs ||
7609 	    (hw->mac.type == ixgbe_mac_82598EB &&
7610 	     tc < MAX_TRAFFIC_CLASS))
7611 		return -EINVAL;
7612 
7613 	pools = (find_first_zero_bit(&adapter->fwd_bitmask, 32) > 1);
7614 	if (tc && pools && adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS)
7615 		return -EBUSY;
7616 
7617 	/* Hardware has to reinitialize queues and interrupts to
7618 	 * match packet buffer alignment. Unfortunately, the
7619 	 * hardware is not flexible enough to do this dynamically.
7620 	 */
7621 	if (netif_running(dev))
7622 		ixgbe_close(dev);
7623 	ixgbe_clear_interrupt_scheme(adapter);
7624 
7625 #ifdef CONFIG_IXGBE_DCB
7626 	if (tc) {
7627 		netdev_set_num_tc(dev, tc);
7628 		ixgbe_set_prio_tc_map(adapter);
7629 
7630 		adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
7631 
7632 		if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
7633 			adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
7634 			adapter->hw.fc.requested_mode = ixgbe_fc_none;
7635 		}
7636 	} else {
7637 		netdev_reset_tc(dev);
7638 
7639 		if (adapter->hw.mac.type == ixgbe_mac_82598EB)
7640 			adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
7641 
7642 		adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
7643 
7644 		adapter->temp_dcb_cfg.pfc_mode_enable = false;
7645 		adapter->dcb_cfg.pfc_mode_enable = false;
7646 	}
7647 
7648 	ixgbe_validate_rtr(adapter, tc);
7649 
7650 #endif /* CONFIG_IXGBE_DCB */
7651 	ixgbe_init_interrupt_scheme(adapter);
7652 
7653 	if (netif_running(dev))
7654 		return ixgbe_open(dev);
7655 
7656 	return 0;
7657 }
7658 
7659 #ifdef CONFIG_PCI_IOV
7660 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
7661 {
7662 	struct net_device *netdev = adapter->netdev;
7663 
7664 	rtnl_lock();
7665 	ixgbe_setup_tc(netdev, netdev_get_num_tc(netdev));
7666 	rtnl_unlock();
7667 }
7668 
7669 #endif
7670 void ixgbe_do_reset(struct net_device *netdev)
7671 {
7672 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7673 
7674 	if (netif_running(netdev))
7675 		ixgbe_reinit_locked(adapter);
7676 	else
7677 		ixgbe_reset(adapter);
7678 }
7679 
7680 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
7681 					    netdev_features_t features)
7682 {
7683 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7684 
7685 	/* If Rx checksum is disabled, then RSC/LRO should also be disabled */
7686 	if (!(features & NETIF_F_RXCSUM))
7687 		features &= ~NETIF_F_LRO;
7688 
7689 	/* Turn off LRO if not RSC capable */
7690 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
7691 		features &= ~NETIF_F_LRO;
7692 
7693 	return features;
7694 }
7695 
7696 static int ixgbe_set_features(struct net_device *netdev,
7697 			      netdev_features_t features)
7698 {
7699 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
7700 	netdev_features_t changed = netdev->features ^ features;
7701 	bool need_reset = false;
7702 
7703 	/* Make sure RSC matches LRO, reset if change */
7704 	if (!(features & NETIF_F_LRO)) {
7705 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7706 			need_reset = true;
7707 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
7708 	} else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
7709 		   !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
7710 		if (adapter->rx_itr_setting == 1 ||
7711 		    adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
7712 			adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
7713 			need_reset = true;
7714 		} else if ((changed ^ features) & NETIF_F_LRO) {
7715 			e_info(probe, "rx-usecs set too low, "
7716 			       "disabling RSC\n");
7717 		}
7718 	}
7719 
7720 	/*
7721 	 * Check if Flow Director n-tuple support was enabled or disabled.  If
7722 	 * the state changed, we need to reset.
7723 	 */
7724 	switch (features & NETIF_F_NTUPLE) {
7725 	case NETIF_F_NTUPLE:
7726 		/* turn off ATR, enable perfect filters and reset */
7727 		if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
7728 			need_reset = true;
7729 
7730 		adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
7731 		adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7732 		break;
7733 	default:
7734 		/* turn off perfect filters, enable ATR and reset */
7735 		if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7736 			need_reset = true;
7737 
7738 		adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7739 
7740 		/* We cannot enable ATR if SR-IOV is enabled */
7741 		if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7742 			break;
7743 
7744 		/* We cannot enable ATR if we have 2 or more traffic classes */
7745 		if (netdev_get_num_tc(netdev) > 1)
7746 			break;
7747 
7748 		/* We cannot enable ATR if RSS is disabled */
7749 		if (adapter->ring_feature[RING_F_RSS].limit <= 1)
7750 			break;
7751 
7752 		/* A sample rate of 0 indicates ATR disabled */
7753 		if (!adapter->atr_sample_rate)
7754 			break;
7755 
7756 		adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
7757 		break;
7758 	}
7759 
7760 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
7761 		ixgbe_vlan_strip_enable(adapter);
7762 	else
7763 		ixgbe_vlan_strip_disable(adapter);
7764 
7765 	if (changed & NETIF_F_RXALL)
7766 		need_reset = true;
7767 
7768 	netdev->features = features;
7769 	if (need_reset)
7770 		ixgbe_do_reset(netdev);
7771 
7772 	return 0;
7773 }
7774 
7775 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
7776 			     struct net_device *dev,
7777 			     const unsigned char *addr, u16 vid,
7778 			     u16 flags)
7779 {
7780 	/* guarantee we can provide a unique filter for the unicast address */
7781 	if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
7782 		if (IXGBE_MAX_PF_MACVLANS <= netdev_uc_count(dev))
7783 			return -ENOMEM;
7784 	}
7785 
7786 	return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
7787 }
7788 
7789 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
7790 				    struct nlmsghdr *nlh)
7791 {
7792 	struct ixgbe_adapter *adapter = netdev_priv(dev);
7793 	struct nlattr *attr, *br_spec;
7794 	int rem;
7795 
7796 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
7797 		return -EOPNOTSUPP;
7798 
7799 	br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
7800 	if (!br_spec)
7801 		return -EINVAL;
7802 
7803 	nla_for_each_nested(attr, br_spec, rem) {
7804 		__u16 mode;
7805 		u32 reg = 0;
7806 
7807 		if (nla_type(attr) != IFLA_BRIDGE_MODE)
7808 			continue;
7809 
7810 		if (nla_len(attr) < sizeof(mode))
7811 			return -EINVAL;
7812 
7813 		mode = nla_get_u16(attr);
7814 		if (mode == BRIDGE_MODE_VEPA) {
7815 			reg = 0;
7816 			adapter->flags2 &= ~IXGBE_FLAG2_BRIDGE_MODE_VEB;
7817 		} else if (mode == BRIDGE_MODE_VEB) {
7818 			reg = IXGBE_PFDTXGSWC_VT_LBEN;
7819 			adapter->flags2 |= IXGBE_FLAG2_BRIDGE_MODE_VEB;
7820 		} else
7821 			return -EINVAL;
7822 
7823 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, reg);
7824 
7825 		e_info(drv, "enabling bridge mode: %s\n",
7826 			mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
7827 	}
7828 
7829 	return 0;
7830 }
7831 
7832 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
7833 				    struct net_device *dev,
7834 				    u32 filter_mask)
7835 {
7836 	struct ixgbe_adapter *adapter = netdev_priv(dev);
7837 	u16 mode;
7838 
7839 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
7840 		return 0;
7841 
7842 	if (adapter->flags2 & IXGBE_FLAG2_BRIDGE_MODE_VEB)
7843 		mode = BRIDGE_MODE_VEB;
7844 	else
7845 		mode = BRIDGE_MODE_VEPA;
7846 
7847 	return ndo_dflt_bridge_getlink(skb, pid, seq, dev, mode, 0, 0);
7848 }
7849 
7850 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
7851 {
7852 	struct ixgbe_fwd_adapter *fwd_adapter = NULL;
7853 	struct ixgbe_adapter *adapter = netdev_priv(pdev);
7854 	int used_pools = adapter->num_vfs + adapter->num_rx_pools;
7855 	unsigned int limit;
7856 	int pool, err;
7857 
7858 	/* Hardware has a limited number of available pools. Each VF, and the
7859 	 * PF require a pool. Check to ensure we don't attempt to use more
7860 	 * then the available number of pools.
7861 	 */
7862 	if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
7863 		return ERR_PTR(-EINVAL);
7864 
7865 #ifdef CONFIG_RPS
7866 	if (vdev->num_rx_queues != vdev->num_tx_queues) {
7867 		netdev_info(pdev, "%s: Only supports a single queue count for TX and RX\n",
7868 			    vdev->name);
7869 		return ERR_PTR(-EINVAL);
7870 	}
7871 #endif
7872 	/* Check for hardware restriction on number of rx/tx queues */
7873 	if (vdev->num_tx_queues > IXGBE_MAX_L2A_QUEUES ||
7874 	    vdev->num_tx_queues == IXGBE_BAD_L2A_QUEUE) {
7875 		netdev_info(pdev,
7876 			    "%s: Supports RX/TX Queue counts 1,2, and 4\n",
7877 			    pdev->name);
7878 		return ERR_PTR(-EINVAL);
7879 	}
7880 
7881 	if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
7882 	      adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS - 1) ||
7883 	    (adapter->num_rx_pools > IXGBE_MAX_MACVLANS))
7884 		return ERR_PTR(-EBUSY);
7885 
7886 	fwd_adapter = kcalloc(1, sizeof(struct ixgbe_fwd_adapter), GFP_KERNEL);
7887 	if (!fwd_adapter)
7888 		return ERR_PTR(-ENOMEM);
7889 
7890 	pool = find_first_zero_bit(&adapter->fwd_bitmask, 32);
7891 	adapter->num_rx_pools++;
7892 	set_bit(pool, &adapter->fwd_bitmask);
7893 	limit = find_last_bit(&adapter->fwd_bitmask, 32);
7894 
7895 	/* Enable VMDq flag so device will be set in VM mode */
7896 	adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED | IXGBE_FLAG_SRIOV_ENABLED;
7897 	adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
7898 	adapter->ring_feature[RING_F_RSS].limit = vdev->num_tx_queues;
7899 
7900 	/* Force reinit of ring allocation with VMDQ enabled */
7901 	err = ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev));
7902 	if (err)
7903 		goto fwd_add_err;
7904 	fwd_adapter->pool = pool;
7905 	fwd_adapter->real_adapter = adapter;
7906 	err = ixgbe_fwd_ring_up(vdev, fwd_adapter);
7907 	if (err)
7908 		goto fwd_add_err;
7909 	netif_tx_start_all_queues(vdev);
7910 	return fwd_adapter;
7911 fwd_add_err:
7912 	/* unwind counter and free adapter struct */
7913 	netdev_info(pdev,
7914 		    "%s: dfwd hardware acceleration failed\n", vdev->name);
7915 	clear_bit(pool, &adapter->fwd_bitmask);
7916 	adapter->num_rx_pools--;
7917 	kfree(fwd_adapter);
7918 	return ERR_PTR(err);
7919 }
7920 
7921 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
7922 {
7923 	struct ixgbe_fwd_adapter *fwd_adapter = priv;
7924 	struct ixgbe_adapter *adapter = fwd_adapter->real_adapter;
7925 	unsigned int limit;
7926 
7927 	clear_bit(fwd_adapter->pool, &adapter->fwd_bitmask);
7928 	adapter->num_rx_pools--;
7929 
7930 	limit = find_last_bit(&adapter->fwd_bitmask, 32);
7931 	adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
7932 	ixgbe_fwd_ring_down(fwd_adapter->netdev, fwd_adapter);
7933 	ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev));
7934 	netdev_dbg(pdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
7935 		   fwd_adapter->pool, adapter->num_rx_pools,
7936 		   fwd_adapter->rx_base_queue,
7937 		   fwd_adapter->rx_base_queue + adapter->num_rx_queues_per_pool,
7938 		   adapter->fwd_bitmask);
7939 	kfree(fwd_adapter);
7940 }
7941 
7942 static const struct net_device_ops ixgbe_netdev_ops = {
7943 	.ndo_open		= ixgbe_open,
7944 	.ndo_stop		= ixgbe_close,
7945 	.ndo_start_xmit		= ixgbe_xmit_frame,
7946 	.ndo_select_queue	= ixgbe_select_queue,
7947 	.ndo_set_rx_mode	= ixgbe_set_rx_mode,
7948 	.ndo_validate_addr	= eth_validate_addr,
7949 	.ndo_set_mac_address	= ixgbe_set_mac,
7950 	.ndo_change_mtu		= ixgbe_change_mtu,
7951 	.ndo_tx_timeout		= ixgbe_tx_timeout,
7952 	.ndo_vlan_rx_add_vid	= ixgbe_vlan_rx_add_vid,
7953 	.ndo_vlan_rx_kill_vid	= ixgbe_vlan_rx_kill_vid,
7954 	.ndo_do_ioctl		= ixgbe_ioctl,
7955 	.ndo_set_vf_mac		= ixgbe_ndo_set_vf_mac,
7956 	.ndo_set_vf_vlan	= ixgbe_ndo_set_vf_vlan,
7957 	.ndo_set_vf_rate	= ixgbe_ndo_set_vf_bw,
7958 	.ndo_set_vf_spoofchk	= ixgbe_ndo_set_vf_spoofchk,
7959 	.ndo_get_vf_config	= ixgbe_ndo_get_vf_config,
7960 	.ndo_get_stats64	= ixgbe_get_stats64,
7961 #ifdef CONFIG_IXGBE_DCB
7962 	.ndo_setup_tc		= ixgbe_setup_tc,
7963 #endif
7964 #ifdef CONFIG_NET_POLL_CONTROLLER
7965 	.ndo_poll_controller	= ixgbe_netpoll,
7966 #endif
7967 #ifdef CONFIG_NET_RX_BUSY_POLL
7968 	.ndo_busy_poll		= ixgbe_low_latency_recv,
7969 #endif
7970 #ifdef IXGBE_FCOE
7971 	.ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
7972 	.ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
7973 	.ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
7974 	.ndo_fcoe_enable = ixgbe_fcoe_enable,
7975 	.ndo_fcoe_disable = ixgbe_fcoe_disable,
7976 	.ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
7977 	.ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
7978 #endif /* IXGBE_FCOE */
7979 	.ndo_set_features = ixgbe_set_features,
7980 	.ndo_fix_features = ixgbe_fix_features,
7981 	.ndo_fdb_add		= ixgbe_ndo_fdb_add,
7982 	.ndo_bridge_setlink	= ixgbe_ndo_bridge_setlink,
7983 	.ndo_bridge_getlink	= ixgbe_ndo_bridge_getlink,
7984 	.ndo_dfwd_add_station	= ixgbe_fwd_add,
7985 	.ndo_dfwd_del_station	= ixgbe_fwd_del,
7986 };
7987 
7988 /**
7989  * ixgbe_enumerate_functions - Get the number of ports this device has
7990  * @adapter: adapter structure
7991  *
7992  * This function enumerates the phsyical functions co-located on a single slot,
7993  * in order to determine how many ports a device has. This is most useful in
7994  * determining the required GT/s of PCIe bandwidth necessary for optimal
7995  * performance.
7996  **/
7997 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
7998 {
7999 	struct pci_dev *entry, *pdev = adapter->pdev;
8000 	int physfns = 0;
8001 
8002 	/* Some cards can not use the generic count PCIe functions method,
8003 	 * because they are behind a parent switch, so we hardcode these with
8004 	 * the correct number of functions.
8005 	 */
8006 	if (ixgbe_pcie_from_parent(&adapter->hw))
8007 		physfns = 4;
8008 
8009 	list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
8010 		/* don't count virtual functions */
8011 		if (entry->is_virtfn)
8012 			continue;
8013 
8014 		/* When the devices on the bus don't all match our device ID,
8015 		 * we can't reliably determine the correct number of
8016 		 * functions. This can occur if a function has been direct
8017 		 * attached to a virtual machine using VT-d, for example. In
8018 		 * this case, simply return -1 to indicate this.
8019 		 */
8020 		if ((entry->vendor != pdev->vendor) ||
8021 		    (entry->device != pdev->device))
8022 			return -1;
8023 
8024 		physfns++;
8025 	}
8026 
8027 	return physfns;
8028 }
8029 
8030 /**
8031  * ixgbe_wol_supported - Check whether device supports WoL
8032  * @hw: hw specific details
8033  * @device_id: the device ID
8034  * @subdev_id: the subsystem device ID
8035  *
8036  * This function is used by probe and ethtool to determine
8037  * which devices have WoL support
8038  *
8039  **/
8040 int ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
8041 			u16 subdevice_id)
8042 {
8043 	struct ixgbe_hw *hw = &adapter->hw;
8044 	u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
8045 	int is_wol_supported = 0;
8046 
8047 	switch (device_id) {
8048 	case IXGBE_DEV_ID_82599_SFP:
8049 		/* Only these subdevices could supports WOL */
8050 		switch (subdevice_id) {
8051 		case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
8052 		case IXGBE_SUBDEV_ID_82599_560FLR:
8053 			/* only support first port */
8054 			if (hw->bus.func != 0)
8055 				break;
8056 		case IXGBE_SUBDEV_ID_82599_SP_560FLR:
8057 		case IXGBE_SUBDEV_ID_82599_SFP:
8058 		case IXGBE_SUBDEV_ID_82599_RNDC:
8059 		case IXGBE_SUBDEV_ID_82599_ECNA_DP:
8060 		case IXGBE_SUBDEV_ID_82599_LOM_SFP:
8061 			is_wol_supported = 1;
8062 			break;
8063 		}
8064 		break;
8065 	case IXGBE_DEV_ID_82599EN_SFP:
8066 		/* Only this subdevice supports WOL */
8067 		switch (subdevice_id) {
8068 		case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
8069 			is_wol_supported = 1;
8070 			break;
8071 		}
8072 		break;
8073 	case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
8074 		/* All except this subdevice support WOL */
8075 		if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
8076 			is_wol_supported = 1;
8077 		break;
8078 	case IXGBE_DEV_ID_82599_KX4:
8079 		is_wol_supported = 1;
8080 		break;
8081 	case IXGBE_DEV_ID_X540T:
8082 	case IXGBE_DEV_ID_X540T1:
8083 		/* check eeprom to see if enabled wol */
8084 		if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
8085 		    ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
8086 		     (hw->bus.func == 0))) {
8087 			is_wol_supported = 1;
8088 		}
8089 		break;
8090 	}
8091 
8092 	return is_wol_supported;
8093 }
8094 
8095 /**
8096  * ixgbe_get_platform_mac_addr - Look up MAC address in Open Firmware / IDPROM
8097  * @adapter: Pointer to adapter struct
8098  */
8099 static void ixgbe_get_platform_mac_addr(struct ixgbe_adapter *adapter)
8100 {
8101 #ifdef CONFIG_OF
8102 	struct device_node *dp = pci_device_to_OF_node(adapter->pdev);
8103 	struct ixgbe_hw *hw = &adapter->hw;
8104 	const unsigned char *addr;
8105 
8106 	addr = of_get_mac_address(dp);
8107 	if (addr) {
8108 		ether_addr_copy(hw->mac.perm_addr, addr);
8109 		return;
8110 	}
8111 #endif /* CONFIG_OF */
8112 
8113 #ifdef CONFIG_SPARC
8114 	ether_addr_copy(hw->mac.perm_addr, idprom->id_ethaddr);
8115 #endif /* CONFIG_SPARC */
8116 }
8117 
8118 /**
8119  * ixgbe_probe - Device Initialization Routine
8120  * @pdev: PCI device information struct
8121  * @ent: entry in ixgbe_pci_tbl
8122  *
8123  * Returns 0 on success, negative on failure
8124  *
8125  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
8126  * The OS initialization, configuring of the adapter private structure,
8127  * and a hardware reset occur.
8128  **/
8129 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
8130 {
8131 	struct net_device *netdev;
8132 	struct ixgbe_adapter *adapter = NULL;
8133 	struct ixgbe_hw *hw;
8134 	const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
8135 	int i, err, pci_using_dac, expected_gts;
8136 	unsigned int indices = MAX_TX_QUEUES;
8137 	u8 part_str[IXGBE_PBANUM_LENGTH];
8138 	bool disable_dev = false;
8139 #ifdef IXGBE_FCOE
8140 	u16 device_caps;
8141 #endif
8142 	u32 eec;
8143 
8144 	/* Catch broken hardware that put the wrong VF device ID in
8145 	 * the PCIe SR-IOV capability.
8146 	 */
8147 	if (pdev->is_virtfn) {
8148 		WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
8149 		     pci_name(pdev), pdev->vendor, pdev->device);
8150 		return -EINVAL;
8151 	}
8152 
8153 	err = pci_enable_device_mem(pdev);
8154 	if (err)
8155 		return err;
8156 
8157 	if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
8158 		pci_using_dac = 1;
8159 	} else {
8160 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
8161 		if (err) {
8162 			dev_err(&pdev->dev,
8163 				"No usable DMA configuration, aborting\n");
8164 			goto err_dma;
8165 		}
8166 		pci_using_dac = 0;
8167 	}
8168 
8169 	err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
8170 					   IORESOURCE_MEM), ixgbe_driver_name);
8171 	if (err) {
8172 		dev_err(&pdev->dev,
8173 			"pci_request_selected_regions failed 0x%x\n", err);
8174 		goto err_pci_reg;
8175 	}
8176 
8177 	pci_enable_pcie_error_reporting(pdev);
8178 
8179 	pci_set_master(pdev);
8180 	pci_save_state(pdev);
8181 
8182 	if (ii->mac == ixgbe_mac_82598EB) {
8183 #ifdef CONFIG_IXGBE_DCB
8184 		/* 8 TC w/ 4 queues per TC */
8185 		indices = 4 * MAX_TRAFFIC_CLASS;
8186 #else
8187 		indices = IXGBE_MAX_RSS_INDICES;
8188 #endif
8189 	}
8190 
8191 	netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
8192 	if (!netdev) {
8193 		err = -ENOMEM;
8194 		goto err_alloc_etherdev;
8195 	}
8196 
8197 	SET_NETDEV_DEV(netdev, &pdev->dev);
8198 
8199 	adapter = netdev_priv(netdev);
8200 
8201 	adapter->netdev = netdev;
8202 	adapter->pdev = pdev;
8203 	hw = &adapter->hw;
8204 	hw->back = adapter;
8205 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
8206 
8207 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
8208 			      pci_resource_len(pdev, 0));
8209 	adapter->io_addr = hw->hw_addr;
8210 	if (!hw->hw_addr) {
8211 		err = -EIO;
8212 		goto err_ioremap;
8213 	}
8214 
8215 	netdev->netdev_ops = &ixgbe_netdev_ops;
8216 	ixgbe_set_ethtool_ops(netdev);
8217 	netdev->watchdog_timeo = 5 * HZ;
8218 	strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
8219 
8220 	/* Setup hw api */
8221 	memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
8222 	hw->mac.type  = ii->mac;
8223 
8224 	/* EEPROM */
8225 	memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
8226 	eec = IXGBE_READ_REG(hw, IXGBE_EEC);
8227 	if (ixgbe_removed(hw->hw_addr)) {
8228 		err = -EIO;
8229 		goto err_ioremap;
8230 	}
8231 	/* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
8232 	if (!(eec & (1 << 8)))
8233 		hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
8234 
8235 	/* PHY */
8236 	memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
8237 	hw->phy.sfp_type = ixgbe_sfp_type_unknown;
8238 	/* ixgbe_identify_phy_generic will set prtad and mmds properly */
8239 	hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
8240 	hw->phy.mdio.mmds = 0;
8241 	hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
8242 	hw->phy.mdio.dev = netdev;
8243 	hw->phy.mdio.mdio_read = ixgbe_mdio_read;
8244 	hw->phy.mdio.mdio_write = ixgbe_mdio_write;
8245 
8246 	ii->get_invariants(hw);
8247 
8248 	/* setup the private structure */
8249 	err = ixgbe_sw_init(adapter);
8250 	if (err)
8251 		goto err_sw_init;
8252 
8253 	/* Make it possible the adapter to be woken up via WOL */
8254 	switch (adapter->hw.mac.type) {
8255 	case ixgbe_mac_82599EB:
8256 	case ixgbe_mac_X540:
8257 	case ixgbe_mac_X550:
8258 	case ixgbe_mac_X550EM_x:
8259 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
8260 		break;
8261 	default:
8262 		break;
8263 	}
8264 
8265 	/*
8266 	 * If there is a fan on this device and it has failed log the
8267 	 * failure.
8268 	 */
8269 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
8270 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
8271 		if (esdp & IXGBE_ESDP_SDP1)
8272 			e_crit(probe, "Fan has stopped, replace the adapter\n");
8273 	}
8274 
8275 	if (allow_unsupported_sfp)
8276 		hw->allow_unsupported_sfp = allow_unsupported_sfp;
8277 
8278 	/* reset_hw fills in the perm_addr as well */
8279 	hw->phy.reset_if_overtemp = true;
8280 	err = hw->mac.ops.reset_hw(hw);
8281 	hw->phy.reset_if_overtemp = false;
8282 	if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
8283 	    hw->mac.type == ixgbe_mac_82598EB) {
8284 		err = 0;
8285 	} else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
8286 		e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
8287 		e_dev_err("Reload the driver after installing a supported module.\n");
8288 		goto err_sw_init;
8289 	} else if (err) {
8290 		e_dev_err("HW Init failed: %d\n", err);
8291 		goto err_sw_init;
8292 	}
8293 
8294 #ifdef CONFIG_PCI_IOV
8295 	/* SR-IOV not supported on the 82598 */
8296 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
8297 		goto skip_sriov;
8298 	/* Mailbox */
8299 	ixgbe_init_mbx_params_pf(hw);
8300 	memcpy(&hw->mbx.ops, ii->mbx_ops, sizeof(hw->mbx.ops));
8301 	pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
8302 	ixgbe_enable_sriov(adapter);
8303 skip_sriov:
8304 
8305 #endif
8306 	netdev->features = NETIF_F_SG |
8307 			   NETIF_F_IP_CSUM |
8308 			   NETIF_F_IPV6_CSUM |
8309 			   NETIF_F_HW_VLAN_CTAG_TX |
8310 			   NETIF_F_HW_VLAN_CTAG_RX |
8311 			   NETIF_F_HW_VLAN_CTAG_FILTER |
8312 			   NETIF_F_TSO |
8313 			   NETIF_F_TSO6 |
8314 			   NETIF_F_RXHASH |
8315 			   NETIF_F_RXCSUM;
8316 
8317 	netdev->hw_features = netdev->features | NETIF_F_HW_L2FW_DOFFLOAD;
8318 
8319 	switch (adapter->hw.mac.type) {
8320 	case ixgbe_mac_82599EB:
8321 	case ixgbe_mac_X540:
8322 	case ixgbe_mac_X550:
8323 	case ixgbe_mac_X550EM_x:
8324 		netdev->features |= NETIF_F_SCTP_CSUM;
8325 		netdev->hw_features |= NETIF_F_SCTP_CSUM |
8326 				       NETIF_F_NTUPLE;
8327 		break;
8328 	default:
8329 		break;
8330 	}
8331 
8332 	netdev->hw_features |= NETIF_F_RXALL;
8333 
8334 	netdev->vlan_features |= NETIF_F_TSO;
8335 	netdev->vlan_features |= NETIF_F_TSO6;
8336 	netdev->vlan_features |= NETIF_F_IP_CSUM;
8337 	netdev->vlan_features |= NETIF_F_IPV6_CSUM;
8338 	netdev->vlan_features |= NETIF_F_SG;
8339 
8340 	netdev->priv_flags |= IFF_UNICAST_FLT;
8341 	netdev->priv_flags |= IFF_SUPP_NOFCS;
8342 
8343 #ifdef CONFIG_IXGBE_DCB
8344 	netdev->dcbnl_ops = &dcbnl_ops;
8345 #endif
8346 
8347 #ifdef IXGBE_FCOE
8348 	if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
8349 		unsigned int fcoe_l;
8350 
8351 		if (hw->mac.ops.get_device_caps) {
8352 			hw->mac.ops.get_device_caps(hw, &device_caps);
8353 			if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
8354 				adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
8355 		}
8356 
8357 
8358 		fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
8359 		adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
8360 
8361 		netdev->features |= NETIF_F_FSO |
8362 				    NETIF_F_FCOE_CRC;
8363 
8364 		netdev->vlan_features |= NETIF_F_FSO |
8365 					 NETIF_F_FCOE_CRC |
8366 					 NETIF_F_FCOE_MTU;
8367 	}
8368 #endif /* IXGBE_FCOE */
8369 	if (pci_using_dac) {
8370 		netdev->features |= NETIF_F_HIGHDMA;
8371 		netdev->vlan_features |= NETIF_F_HIGHDMA;
8372 	}
8373 
8374 	if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
8375 		netdev->hw_features |= NETIF_F_LRO;
8376 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
8377 		netdev->features |= NETIF_F_LRO;
8378 
8379 	/* make sure the EEPROM is good */
8380 	if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
8381 		e_dev_err("The EEPROM Checksum Is Not Valid\n");
8382 		err = -EIO;
8383 		goto err_sw_init;
8384 	}
8385 
8386 	ixgbe_get_platform_mac_addr(adapter);
8387 
8388 	memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
8389 
8390 	if (!is_valid_ether_addr(netdev->dev_addr)) {
8391 		e_dev_err("invalid MAC address\n");
8392 		err = -EIO;
8393 		goto err_sw_init;
8394 	}
8395 
8396 	ixgbe_mac_set_default_filter(adapter, hw->mac.perm_addr);
8397 
8398 	setup_timer(&adapter->service_timer, &ixgbe_service_timer,
8399 		    (unsigned long) adapter);
8400 
8401 	if (ixgbe_removed(hw->hw_addr)) {
8402 		err = -EIO;
8403 		goto err_sw_init;
8404 	}
8405 	INIT_WORK(&adapter->service_task, ixgbe_service_task);
8406 	set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
8407 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
8408 
8409 	err = ixgbe_init_interrupt_scheme(adapter);
8410 	if (err)
8411 		goto err_sw_init;
8412 
8413 	/* WOL not supported for all devices */
8414 	adapter->wol = 0;
8415 	hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
8416 	hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
8417 						pdev->subsystem_device);
8418 	if (hw->wol_enabled)
8419 		adapter->wol = IXGBE_WUFC_MAG;
8420 
8421 	device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
8422 
8423 	/* save off EEPROM version number */
8424 	hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
8425 	hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
8426 
8427 	/* pick up the PCI bus settings for reporting later */
8428 	hw->mac.ops.get_bus_info(hw);
8429 	if (ixgbe_pcie_from_parent(hw))
8430 		ixgbe_get_parent_bus_info(adapter);
8431 
8432 	/* calculate the expected PCIe bandwidth required for optimal
8433 	 * performance. Note that some older parts will never have enough
8434 	 * bandwidth due to being older generation PCIe parts. We clamp these
8435 	 * parts to ensure no warning is displayed if it can't be fixed.
8436 	 */
8437 	switch (hw->mac.type) {
8438 	case ixgbe_mac_82598EB:
8439 		expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
8440 		break;
8441 	default:
8442 		expected_gts = ixgbe_enumerate_functions(adapter) * 10;
8443 		break;
8444 	}
8445 
8446 	/* don't check link if we failed to enumerate functions */
8447 	if (expected_gts > 0)
8448 		ixgbe_check_minimum_link(adapter, expected_gts);
8449 
8450 	err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
8451 	if (err)
8452 		strlcpy(part_str, "Unknown", sizeof(part_str));
8453 	if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
8454 		e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
8455 			   hw->mac.type, hw->phy.type, hw->phy.sfp_type,
8456 			   part_str);
8457 	else
8458 		e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
8459 			   hw->mac.type, hw->phy.type, part_str);
8460 
8461 	e_dev_info("%pM\n", netdev->dev_addr);
8462 
8463 	/* reset the hardware with the new settings */
8464 	err = hw->mac.ops.start_hw(hw);
8465 	if (err == IXGBE_ERR_EEPROM_VERSION) {
8466 		/* We are running on a pre-production device, log a warning */
8467 		e_dev_warn("This device is a pre-production adapter/LOM. "
8468 			   "Please be aware there may be issues associated "
8469 			   "with your hardware.  If you are experiencing "
8470 			   "problems please contact your Intel or hardware "
8471 			   "representative who provided you with this "
8472 			   "hardware.\n");
8473 	}
8474 	strcpy(netdev->name, "eth%d");
8475 	err = register_netdev(netdev);
8476 	if (err)
8477 		goto err_register;
8478 
8479 	pci_set_drvdata(pdev, adapter);
8480 
8481 	/* power down the optics for 82599 SFP+ fiber */
8482 	if (hw->mac.ops.disable_tx_laser)
8483 		hw->mac.ops.disable_tx_laser(hw);
8484 
8485 	/* carrier off reporting is important to ethtool even BEFORE open */
8486 	netif_carrier_off(netdev);
8487 
8488 #ifdef CONFIG_IXGBE_DCA
8489 	if (dca_add_requester(&pdev->dev) == 0) {
8490 		adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
8491 		ixgbe_setup_dca(adapter);
8492 	}
8493 #endif
8494 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
8495 		e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
8496 		for (i = 0; i < adapter->num_vfs; i++)
8497 			ixgbe_vf_configuration(pdev, (i | 0x10000000));
8498 	}
8499 
8500 	/* firmware requires driver version to be 0xFFFFFFFF
8501 	 * since os does not support feature
8502 	 */
8503 	if (hw->mac.ops.set_fw_drv_ver)
8504 		hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
8505 					   0xFF);
8506 
8507 	/* add san mac addr to netdev */
8508 	ixgbe_add_sanmac_netdev(netdev);
8509 
8510 	e_dev_info("%s\n", ixgbe_default_device_descr);
8511 
8512 #ifdef CONFIG_IXGBE_HWMON
8513 	if (ixgbe_sysfs_init(adapter))
8514 		e_err(probe, "failed to allocate sysfs resources\n");
8515 #endif /* CONFIG_IXGBE_HWMON */
8516 
8517 	ixgbe_dbg_adapter_init(adapter);
8518 
8519 	/* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
8520 	if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
8521 		hw->mac.ops.setup_link(hw,
8522 			IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
8523 			true);
8524 
8525 	return 0;
8526 
8527 err_register:
8528 	ixgbe_release_hw_control(adapter);
8529 	ixgbe_clear_interrupt_scheme(adapter);
8530 err_sw_init:
8531 	ixgbe_disable_sriov(adapter);
8532 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
8533 	iounmap(adapter->io_addr);
8534 	kfree(adapter->mac_table);
8535 err_ioremap:
8536 	disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
8537 	free_netdev(netdev);
8538 err_alloc_etherdev:
8539 	pci_release_selected_regions(pdev,
8540 				     pci_select_bars(pdev, IORESOURCE_MEM));
8541 err_pci_reg:
8542 err_dma:
8543 	if (!adapter || disable_dev)
8544 		pci_disable_device(pdev);
8545 	return err;
8546 }
8547 
8548 /**
8549  * ixgbe_remove - Device Removal Routine
8550  * @pdev: PCI device information struct
8551  *
8552  * ixgbe_remove is called by the PCI subsystem to alert the driver
8553  * that it should release a PCI device.  The could be caused by a
8554  * Hot-Plug event, or because the driver is going to be removed from
8555  * memory.
8556  **/
8557 static void ixgbe_remove(struct pci_dev *pdev)
8558 {
8559 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8560 	struct net_device *netdev;
8561 	bool disable_dev;
8562 
8563 	/* if !adapter then we already cleaned up in probe */
8564 	if (!adapter)
8565 		return;
8566 
8567 	netdev  = adapter->netdev;
8568 	ixgbe_dbg_adapter_exit(adapter);
8569 
8570 	set_bit(__IXGBE_REMOVING, &adapter->state);
8571 	cancel_work_sync(&adapter->service_task);
8572 
8573 
8574 #ifdef CONFIG_IXGBE_DCA
8575 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
8576 		adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
8577 		dca_remove_requester(&pdev->dev);
8578 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
8579 	}
8580 
8581 #endif
8582 #ifdef CONFIG_IXGBE_HWMON
8583 	ixgbe_sysfs_exit(adapter);
8584 #endif /* CONFIG_IXGBE_HWMON */
8585 
8586 	/* remove the added san mac */
8587 	ixgbe_del_sanmac_netdev(netdev);
8588 
8589 	if (netdev->reg_state == NETREG_REGISTERED)
8590 		unregister_netdev(netdev);
8591 
8592 #ifdef CONFIG_PCI_IOV
8593 	/*
8594 	 * Only disable SR-IOV on unload if the user specified the now
8595 	 * deprecated max_vfs module parameter.
8596 	 */
8597 	if (max_vfs)
8598 		ixgbe_disable_sriov(adapter);
8599 #endif
8600 	ixgbe_clear_interrupt_scheme(adapter);
8601 
8602 	ixgbe_release_hw_control(adapter);
8603 
8604 #ifdef CONFIG_DCB
8605 	kfree(adapter->ixgbe_ieee_pfc);
8606 	kfree(adapter->ixgbe_ieee_ets);
8607 
8608 #endif
8609 	iounmap(adapter->io_addr);
8610 	pci_release_selected_regions(pdev, pci_select_bars(pdev,
8611 				     IORESOURCE_MEM));
8612 
8613 	e_dev_info("complete\n");
8614 
8615 	kfree(adapter->mac_table);
8616 	disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
8617 	free_netdev(netdev);
8618 
8619 	pci_disable_pcie_error_reporting(pdev);
8620 
8621 	if (disable_dev)
8622 		pci_disable_device(pdev);
8623 }
8624 
8625 /**
8626  * ixgbe_io_error_detected - called when PCI error is detected
8627  * @pdev: Pointer to PCI device
8628  * @state: The current pci connection state
8629  *
8630  * This function is called after a PCI bus error affecting
8631  * this device has been detected.
8632  */
8633 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
8634 						pci_channel_state_t state)
8635 {
8636 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8637 	struct net_device *netdev = adapter->netdev;
8638 
8639 #ifdef CONFIG_PCI_IOV
8640 	struct ixgbe_hw *hw = &adapter->hw;
8641 	struct pci_dev *bdev, *vfdev;
8642 	u32 dw0, dw1, dw2, dw3;
8643 	int vf, pos;
8644 	u16 req_id, pf_func;
8645 
8646 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
8647 	    adapter->num_vfs == 0)
8648 		goto skip_bad_vf_detection;
8649 
8650 	bdev = pdev->bus->self;
8651 	while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
8652 		bdev = bdev->bus->self;
8653 
8654 	if (!bdev)
8655 		goto skip_bad_vf_detection;
8656 
8657 	pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
8658 	if (!pos)
8659 		goto skip_bad_vf_detection;
8660 
8661 	dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
8662 	dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
8663 	dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
8664 	dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
8665 	if (ixgbe_removed(hw->hw_addr))
8666 		goto skip_bad_vf_detection;
8667 
8668 	req_id = dw1 >> 16;
8669 	/* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
8670 	if (!(req_id & 0x0080))
8671 		goto skip_bad_vf_detection;
8672 
8673 	pf_func = req_id & 0x01;
8674 	if ((pf_func & 1) == (pdev->devfn & 1)) {
8675 		unsigned int device_id;
8676 
8677 		vf = (req_id & 0x7F) >> 1;
8678 		e_dev_err("VF %d has caused a PCIe error\n", vf);
8679 		e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
8680 				"%8.8x\tdw3: %8.8x\n",
8681 		dw0, dw1, dw2, dw3);
8682 		switch (adapter->hw.mac.type) {
8683 		case ixgbe_mac_82599EB:
8684 			device_id = IXGBE_82599_VF_DEVICE_ID;
8685 			break;
8686 		case ixgbe_mac_X540:
8687 			device_id = IXGBE_X540_VF_DEVICE_ID;
8688 			break;
8689 		case ixgbe_mac_X550:
8690 			device_id = IXGBE_DEV_ID_X550_VF;
8691 			break;
8692 		case ixgbe_mac_X550EM_x:
8693 			device_id = IXGBE_DEV_ID_X550EM_X_VF;
8694 			break;
8695 		default:
8696 			device_id = 0;
8697 			break;
8698 		}
8699 
8700 		/* Find the pci device of the offending VF */
8701 		vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
8702 		while (vfdev) {
8703 			if (vfdev->devfn == (req_id & 0xFF))
8704 				break;
8705 			vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
8706 					       device_id, vfdev);
8707 		}
8708 		/*
8709 		 * There's a slim chance the VF could have been hot plugged,
8710 		 * so if it is no longer present we don't need to issue the
8711 		 * VFLR.  Just clean up the AER in that case.
8712 		 */
8713 		if (vfdev) {
8714 			ixgbe_issue_vf_flr(adapter, vfdev);
8715 			/* Free device reference count */
8716 			pci_dev_put(vfdev);
8717 		}
8718 
8719 		pci_cleanup_aer_uncorrect_error_status(pdev);
8720 	}
8721 
8722 	/*
8723 	 * Even though the error may have occurred on the other port
8724 	 * we still need to increment the vf error reference count for
8725 	 * both ports because the I/O resume function will be called
8726 	 * for both of them.
8727 	 */
8728 	adapter->vferr_refcount++;
8729 
8730 	return PCI_ERS_RESULT_RECOVERED;
8731 
8732 skip_bad_vf_detection:
8733 #endif /* CONFIG_PCI_IOV */
8734 	if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
8735 		return PCI_ERS_RESULT_DISCONNECT;
8736 
8737 	rtnl_lock();
8738 	netif_device_detach(netdev);
8739 
8740 	if (state == pci_channel_io_perm_failure) {
8741 		rtnl_unlock();
8742 		return PCI_ERS_RESULT_DISCONNECT;
8743 	}
8744 
8745 	if (netif_running(netdev))
8746 		ixgbe_down(adapter);
8747 
8748 	if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
8749 		pci_disable_device(pdev);
8750 	rtnl_unlock();
8751 
8752 	/* Request a slot reset. */
8753 	return PCI_ERS_RESULT_NEED_RESET;
8754 }
8755 
8756 /**
8757  * ixgbe_io_slot_reset - called after the pci bus has been reset.
8758  * @pdev: Pointer to PCI device
8759  *
8760  * Restart the card from scratch, as if from a cold-boot.
8761  */
8762 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
8763 {
8764 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8765 	pci_ers_result_t result;
8766 	int err;
8767 
8768 	if (pci_enable_device_mem(pdev)) {
8769 		e_err(probe, "Cannot re-enable PCI device after reset.\n");
8770 		result = PCI_ERS_RESULT_DISCONNECT;
8771 	} else {
8772 		smp_mb__before_atomic();
8773 		clear_bit(__IXGBE_DISABLED, &adapter->state);
8774 		adapter->hw.hw_addr = adapter->io_addr;
8775 		pci_set_master(pdev);
8776 		pci_restore_state(pdev);
8777 		pci_save_state(pdev);
8778 
8779 		pci_wake_from_d3(pdev, false);
8780 
8781 		ixgbe_reset(adapter);
8782 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
8783 		result = PCI_ERS_RESULT_RECOVERED;
8784 	}
8785 
8786 	err = pci_cleanup_aer_uncorrect_error_status(pdev);
8787 	if (err) {
8788 		e_dev_err("pci_cleanup_aer_uncorrect_error_status "
8789 			  "failed 0x%0x\n", err);
8790 		/* non-fatal, continue */
8791 	}
8792 
8793 	return result;
8794 }
8795 
8796 /**
8797  * ixgbe_io_resume - called when traffic can start flowing again.
8798  * @pdev: Pointer to PCI device
8799  *
8800  * This callback is called when the error recovery driver tells us that
8801  * its OK to resume normal operation.
8802  */
8803 static void ixgbe_io_resume(struct pci_dev *pdev)
8804 {
8805 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8806 	struct net_device *netdev = adapter->netdev;
8807 
8808 #ifdef CONFIG_PCI_IOV
8809 	if (adapter->vferr_refcount) {
8810 		e_info(drv, "Resuming after VF err\n");
8811 		adapter->vferr_refcount--;
8812 		return;
8813 	}
8814 
8815 #endif
8816 	if (netif_running(netdev))
8817 		ixgbe_up(adapter);
8818 
8819 	netif_device_attach(netdev);
8820 }
8821 
8822 static const struct pci_error_handlers ixgbe_err_handler = {
8823 	.error_detected = ixgbe_io_error_detected,
8824 	.slot_reset = ixgbe_io_slot_reset,
8825 	.resume = ixgbe_io_resume,
8826 };
8827 
8828 static struct pci_driver ixgbe_driver = {
8829 	.name     = ixgbe_driver_name,
8830 	.id_table = ixgbe_pci_tbl,
8831 	.probe    = ixgbe_probe,
8832 	.remove   = ixgbe_remove,
8833 #ifdef CONFIG_PM
8834 	.suspend  = ixgbe_suspend,
8835 	.resume   = ixgbe_resume,
8836 #endif
8837 	.shutdown = ixgbe_shutdown,
8838 	.sriov_configure = ixgbe_pci_sriov_configure,
8839 	.err_handler = &ixgbe_err_handler
8840 };
8841 
8842 /**
8843  * ixgbe_init_module - Driver Registration Routine
8844  *
8845  * ixgbe_init_module is the first routine called when the driver is
8846  * loaded. All it does is register with the PCI subsystem.
8847  **/
8848 static int __init ixgbe_init_module(void)
8849 {
8850 	int ret;
8851 	pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
8852 	pr_info("%s\n", ixgbe_copyright);
8853 
8854 	ixgbe_dbg_init();
8855 
8856 	ret = pci_register_driver(&ixgbe_driver);
8857 	if (ret) {
8858 		ixgbe_dbg_exit();
8859 		return ret;
8860 	}
8861 
8862 #ifdef CONFIG_IXGBE_DCA
8863 	dca_register_notify(&dca_notifier);
8864 #endif
8865 
8866 	return 0;
8867 }
8868 
8869 module_init(ixgbe_init_module);
8870 
8871 /**
8872  * ixgbe_exit_module - Driver Exit Cleanup Routine
8873  *
8874  * ixgbe_exit_module is called just before the driver is removed
8875  * from memory.
8876  **/
8877 static void __exit ixgbe_exit_module(void)
8878 {
8879 #ifdef CONFIG_IXGBE_DCA
8880 	dca_unregister_notify(&dca_notifier);
8881 #endif
8882 	pci_unregister_driver(&ixgbe_driver);
8883 
8884 	ixgbe_dbg_exit();
8885 
8886 	rcu_barrier(); /* Wait for completion of call_rcu()'s */
8887 }
8888 
8889 #ifdef CONFIG_IXGBE_DCA
8890 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
8891 			    void *p)
8892 {
8893 	int ret_val;
8894 
8895 	ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
8896 					 __ixgbe_notify_dca);
8897 
8898 	return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
8899 }
8900 
8901 #endif /* CONFIG_IXGBE_DCA */
8902 
8903 module_exit(ixgbe_exit_module);
8904 
8905 /* ixgbe_main.c */
8906