1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 1999 - 2018 Intel Corporation. */
3 
4 #include <linux/types.h>
5 #include <linux/module.h>
6 #include <linux/pci.h>
7 #include <linux/netdevice.h>
8 #include <linux/vmalloc.h>
9 #include <linux/string.h>
10 #include <linux/in.h>
11 #include <linux/interrupt.h>
12 #include <linux/ip.h>
13 #include <linux/tcp.h>
14 #include <linux/sctp.h>
15 #include <linux/pkt_sched.h>
16 #include <linux/ipv6.h>
17 #include <linux/slab.h>
18 #include <net/checksum.h>
19 #include <net/ip6_checksum.h>
20 #include <linux/etherdevice.h>
21 #include <linux/ethtool.h>
22 #include <linux/if.h>
23 #include <linux/if_vlan.h>
24 #include <linux/if_macvlan.h>
25 #include <linux/if_bridge.h>
26 #include <linux/prefetch.h>
27 #include <linux/bpf.h>
28 #include <linux/bpf_trace.h>
29 #include <linux/atomic.h>
30 #include <linux/numa.h>
31 #include <scsi/fc/fc_fcoe.h>
32 #include <net/udp_tunnel.h>
33 #include <net/pkt_cls.h>
34 #include <net/tc_act/tc_gact.h>
35 #include <net/tc_act/tc_mirred.h>
36 #include <net/vxlan.h>
37 #include <net/mpls.h>
38 #include <net/xdp_sock_drv.h>
39 #include <net/xfrm.h>
40 
41 #include "ixgbe.h"
42 #include "ixgbe_common.h"
43 #include "ixgbe_dcb_82599.h"
44 #include "ixgbe_phy.h"
45 #include "ixgbe_sriov.h"
46 #include "ixgbe_model.h"
47 #include "ixgbe_txrx_common.h"
48 
49 char ixgbe_driver_name[] = "ixgbe";
50 static const char ixgbe_driver_string[] =
51 			      "Intel(R) 10 Gigabit PCI Express Network Driver";
52 #ifdef IXGBE_FCOE
53 char ixgbe_default_device_descr[] =
54 			      "Intel(R) 10 Gigabit Network Connection";
55 #else
56 static char ixgbe_default_device_descr[] =
57 			      "Intel(R) 10 Gigabit Network Connection";
58 #endif
59 #define DRV_VERSION "5.1.0-k"
60 const char ixgbe_driver_version[] = DRV_VERSION;
61 static const char ixgbe_copyright[] =
62 				"Copyright (c) 1999-2016 Intel Corporation.";
63 
64 static const char ixgbe_overheat_msg[] = "Network adapter has been stopped because it has over heated. Restart the computer. If the problem persists, power off the system and replace the adapter";
65 
66 static const struct ixgbe_info *ixgbe_info_tbl[] = {
67 	[board_82598]		= &ixgbe_82598_info,
68 	[board_82599]		= &ixgbe_82599_info,
69 	[board_X540]		= &ixgbe_X540_info,
70 	[board_X550]		= &ixgbe_X550_info,
71 	[board_X550EM_x]	= &ixgbe_X550EM_x_info,
72 	[board_x550em_x_fw]	= &ixgbe_x550em_x_fw_info,
73 	[board_x550em_a]	= &ixgbe_x550em_a_info,
74 	[board_x550em_a_fw]	= &ixgbe_x550em_a_fw_info,
75 };
76 
77 /* ixgbe_pci_tbl - PCI Device ID Table
78  *
79  * Wildcard entries (PCI_ANY_ID) should come last
80  * Last entry must be all 0s
81  *
82  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
83  *   Class, Class Mask, private data (not used) }
84  */
85 static const struct pci_device_id ixgbe_pci_tbl[] = {
86 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
87 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
88 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
89 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
90 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
91 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
92 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
93 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
94 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
95 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
96 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
97 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
98 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
99 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
100 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
101 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
102 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
103 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
104 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
105 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
106 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
107 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
108 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
109 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
110 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
111 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
112 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_QSFP_SF_QP), board_82599 },
113 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
114 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
115 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
116 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T), board_X550},
117 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T1), board_X550},
118 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KX4), board_X550EM_x},
119 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_XFI), board_X550EM_x},
120 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KR), board_X550EM_x},
121 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_10G_T), board_X550EM_x},
122 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_SFP), board_X550EM_x},
123 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_1G_T), board_x550em_x_fw},
124 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR), board_x550em_a },
125 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR_L), board_x550em_a },
126 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP_N), board_x550em_a },
127 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII), board_x550em_a },
128 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII_L), board_x550em_a },
129 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_10G_T), board_x550em_a},
130 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP), board_x550em_a },
131 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T), board_x550em_a_fw },
132 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T_L), board_x550em_a_fw },
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 v2");
168 MODULE_VERSION(DRV_VERSION);
169 
170 static struct workqueue_struct *ixgbe_wq;
171 
172 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev);
173 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *);
174 
175 static const struct net_device_ops ixgbe_netdev_ops;
176 
177 static bool netif_is_ixgbe(struct net_device *dev)
178 {
179 	return dev && (dev->netdev_ops == &ixgbe_netdev_ops);
180 }
181 
182 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
183 					  u32 reg, u16 *value)
184 {
185 	struct pci_dev *parent_dev;
186 	struct pci_bus *parent_bus;
187 
188 	parent_bus = adapter->pdev->bus->parent;
189 	if (!parent_bus)
190 		return -1;
191 
192 	parent_dev = parent_bus->self;
193 	if (!parent_dev)
194 		return -1;
195 
196 	if (!pci_is_pcie(parent_dev))
197 		return -1;
198 
199 	pcie_capability_read_word(parent_dev, reg, value);
200 	if (*value == IXGBE_FAILED_READ_CFG_WORD &&
201 	    ixgbe_check_cfg_remove(&adapter->hw, parent_dev))
202 		return -1;
203 	return 0;
204 }
205 
206 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
207 {
208 	struct ixgbe_hw *hw = &adapter->hw;
209 	u16 link_status = 0;
210 	int err;
211 
212 	hw->bus.type = ixgbe_bus_type_pci_express;
213 
214 	/* Get the negotiated link width and speed from PCI config space of the
215 	 * parent, as this device is behind a switch
216 	 */
217 	err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
218 
219 	/* assume caller will handle error case */
220 	if (err)
221 		return err;
222 
223 	hw->bus.width = ixgbe_convert_bus_width(link_status);
224 	hw->bus.speed = ixgbe_convert_bus_speed(link_status);
225 
226 	return 0;
227 }
228 
229 /**
230  * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
231  * @hw: hw specific details
232  *
233  * This function is used by probe to determine whether a device's PCI-Express
234  * bandwidth details should be gathered from the parent bus instead of from the
235  * device. Used to ensure that various locations all have the correct device ID
236  * checks.
237  */
238 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
239 {
240 	switch (hw->device_id) {
241 	case IXGBE_DEV_ID_82599_SFP_SF_QP:
242 	case IXGBE_DEV_ID_82599_QSFP_SF_QP:
243 		return true;
244 	default:
245 		return false;
246 	}
247 }
248 
249 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
250 				     int expected_gts)
251 {
252 	struct ixgbe_hw *hw = &adapter->hw;
253 	struct pci_dev *pdev;
254 
255 	/* Some devices are not connected over PCIe and thus do not negotiate
256 	 * speed. These devices do not have valid bus info, and thus any report
257 	 * we generate may not be correct.
258 	 */
259 	if (hw->bus.type == ixgbe_bus_type_internal)
260 		return;
261 
262 	/* determine whether to use the parent device */
263 	if (ixgbe_pcie_from_parent(&adapter->hw))
264 		pdev = adapter->pdev->bus->parent->self;
265 	else
266 		pdev = adapter->pdev;
267 
268 	pcie_print_link_status(pdev);
269 }
270 
271 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
272 {
273 	if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
274 	    !test_bit(__IXGBE_REMOVING, &adapter->state) &&
275 	    !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
276 		queue_work(ixgbe_wq, &adapter->service_task);
277 }
278 
279 static void ixgbe_remove_adapter(struct ixgbe_hw *hw)
280 {
281 	struct ixgbe_adapter *adapter = hw->back;
282 
283 	if (!hw->hw_addr)
284 		return;
285 	hw->hw_addr = NULL;
286 	e_dev_err("Adapter removed\n");
287 	if (test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
288 		ixgbe_service_event_schedule(adapter);
289 }
290 
291 static u32 ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg)
292 {
293 	u8 __iomem *reg_addr;
294 	u32 value;
295 	int i;
296 
297 	reg_addr = READ_ONCE(hw->hw_addr);
298 	if (ixgbe_removed(reg_addr))
299 		return IXGBE_FAILED_READ_REG;
300 
301 	/* Register read of 0xFFFFFFF can indicate the adapter has been removed,
302 	 * so perform several status register reads to determine if the adapter
303 	 * has been removed.
304 	 */
305 	for (i = 0; i < IXGBE_FAILED_READ_RETRIES; i++) {
306 		value = readl(reg_addr + IXGBE_STATUS);
307 		if (value != IXGBE_FAILED_READ_REG)
308 			break;
309 		mdelay(3);
310 	}
311 
312 	if (value == IXGBE_FAILED_READ_REG)
313 		ixgbe_remove_adapter(hw);
314 	else
315 		value = readl(reg_addr + reg);
316 	return value;
317 }
318 
319 /**
320  * ixgbe_read_reg - Read from device register
321  * @hw: hw specific details
322  * @reg: offset of register to read
323  *
324  * Returns : value read or IXGBE_FAILED_READ_REG if removed
325  *
326  * This function is used to read device registers. It checks for device
327  * removal by confirming any read that returns all ones by checking the
328  * status register value for all ones. This function avoids reading from
329  * the hardware if a removal was previously detected in which case it
330  * returns IXGBE_FAILED_READ_REG (all ones).
331  */
332 u32 ixgbe_read_reg(struct ixgbe_hw *hw, u32 reg)
333 {
334 	u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
335 	u32 value;
336 
337 	if (ixgbe_removed(reg_addr))
338 		return IXGBE_FAILED_READ_REG;
339 	if (unlikely(hw->phy.nw_mng_if_sel &
340 		     IXGBE_NW_MNG_IF_SEL_SGMII_ENABLE)) {
341 		struct ixgbe_adapter *adapter;
342 		int i;
343 
344 		for (i = 0; i < 200; ++i) {
345 			value = readl(reg_addr + IXGBE_MAC_SGMII_BUSY);
346 			if (likely(!value))
347 				goto writes_completed;
348 			if (value == IXGBE_FAILED_READ_REG) {
349 				ixgbe_remove_adapter(hw);
350 				return IXGBE_FAILED_READ_REG;
351 			}
352 			udelay(5);
353 		}
354 
355 		adapter = hw->back;
356 		e_warn(hw, "register writes incomplete %08x\n", value);
357 	}
358 
359 writes_completed:
360 	value = readl(reg_addr + reg);
361 	if (unlikely(value == IXGBE_FAILED_READ_REG))
362 		value = ixgbe_check_remove(hw, reg);
363 	return value;
364 }
365 
366 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev)
367 {
368 	u16 value;
369 
370 	pci_read_config_word(pdev, PCI_VENDOR_ID, &value);
371 	if (value == IXGBE_FAILED_READ_CFG_WORD) {
372 		ixgbe_remove_adapter(hw);
373 		return true;
374 	}
375 	return false;
376 }
377 
378 u16 ixgbe_read_pci_cfg_word(struct ixgbe_hw *hw, u32 reg)
379 {
380 	struct ixgbe_adapter *adapter = hw->back;
381 	u16 value;
382 
383 	if (ixgbe_removed(hw->hw_addr))
384 		return IXGBE_FAILED_READ_CFG_WORD;
385 	pci_read_config_word(adapter->pdev, reg, &value);
386 	if (value == IXGBE_FAILED_READ_CFG_WORD &&
387 	    ixgbe_check_cfg_remove(hw, adapter->pdev))
388 		return IXGBE_FAILED_READ_CFG_WORD;
389 	return value;
390 }
391 
392 #ifdef CONFIG_PCI_IOV
393 static u32 ixgbe_read_pci_cfg_dword(struct ixgbe_hw *hw, u32 reg)
394 {
395 	struct ixgbe_adapter *adapter = hw->back;
396 	u32 value;
397 
398 	if (ixgbe_removed(hw->hw_addr))
399 		return IXGBE_FAILED_READ_CFG_DWORD;
400 	pci_read_config_dword(adapter->pdev, reg, &value);
401 	if (value == IXGBE_FAILED_READ_CFG_DWORD &&
402 	    ixgbe_check_cfg_remove(hw, adapter->pdev))
403 		return IXGBE_FAILED_READ_CFG_DWORD;
404 	return value;
405 }
406 #endif /* CONFIG_PCI_IOV */
407 
408 void ixgbe_write_pci_cfg_word(struct ixgbe_hw *hw, u32 reg, u16 value)
409 {
410 	struct ixgbe_adapter *adapter = hw->back;
411 
412 	if (ixgbe_removed(hw->hw_addr))
413 		return;
414 	pci_write_config_word(adapter->pdev, reg, value);
415 }
416 
417 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
418 {
419 	BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
420 
421 	/* flush memory to make sure state is correct before next watchdog */
422 	smp_mb__before_atomic();
423 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
424 }
425 
426 struct ixgbe_reg_info {
427 	u32 ofs;
428 	char *name;
429 };
430 
431 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
432 
433 	/* General Registers */
434 	{IXGBE_CTRL, "CTRL"},
435 	{IXGBE_STATUS, "STATUS"},
436 	{IXGBE_CTRL_EXT, "CTRL_EXT"},
437 
438 	/* Interrupt Registers */
439 	{IXGBE_EICR, "EICR"},
440 
441 	/* RX Registers */
442 	{IXGBE_SRRCTL(0), "SRRCTL"},
443 	{IXGBE_DCA_RXCTRL(0), "DRXCTL"},
444 	{IXGBE_RDLEN(0), "RDLEN"},
445 	{IXGBE_RDH(0), "RDH"},
446 	{IXGBE_RDT(0), "RDT"},
447 	{IXGBE_RXDCTL(0), "RXDCTL"},
448 	{IXGBE_RDBAL(0), "RDBAL"},
449 	{IXGBE_RDBAH(0), "RDBAH"},
450 
451 	/* TX Registers */
452 	{IXGBE_TDBAL(0), "TDBAL"},
453 	{IXGBE_TDBAH(0), "TDBAH"},
454 	{IXGBE_TDLEN(0), "TDLEN"},
455 	{IXGBE_TDH(0), "TDH"},
456 	{IXGBE_TDT(0), "TDT"},
457 	{IXGBE_TXDCTL(0), "TXDCTL"},
458 
459 	/* List Terminator */
460 	{ .name = NULL }
461 };
462 
463 
464 /*
465  * ixgbe_regdump - register printout routine
466  */
467 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
468 {
469 	int i;
470 	char rname[16];
471 	u32 regs[64];
472 
473 	switch (reginfo->ofs) {
474 	case IXGBE_SRRCTL(0):
475 		for (i = 0; i < 64; i++)
476 			regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
477 		break;
478 	case IXGBE_DCA_RXCTRL(0):
479 		for (i = 0; i < 64; i++)
480 			regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
481 		break;
482 	case IXGBE_RDLEN(0):
483 		for (i = 0; i < 64; i++)
484 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
485 		break;
486 	case IXGBE_RDH(0):
487 		for (i = 0; i < 64; i++)
488 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
489 		break;
490 	case IXGBE_RDT(0):
491 		for (i = 0; i < 64; i++)
492 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
493 		break;
494 	case IXGBE_RXDCTL(0):
495 		for (i = 0; i < 64; i++)
496 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
497 		break;
498 	case IXGBE_RDBAL(0):
499 		for (i = 0; i < 64; i++)
500 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
501 		break;
502 	case IXGBE_RDBAH(0):
503 		for (i = 0; i < 64; i++)
504 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
505 		break;
506 	case IXGBE_TDBAL(0):
507 		for (i = 0; i < 64; i++)
508 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
509 		break;
510 	case IXGBE_TDBAH(0):
511 		for (i = 0; i < 64; i++)
512 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
513 		break;
514 	case IXGBE_TDLEN(0):
515 		for (i = 0; i < 64; i++)
516 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
517 		break;
518 	case IXGBE_TDH(0):
519 		for (i = 0; i < 64; i++)
520 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
521 		break;
522 	case IXGBE_TDT(0):
523 		for (i = 0; i < 64; i++)
524 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
525 		break;
526 	case IXGBE_TXDCTL(0):
527 		for (i = 0; i < 64; i++)
528 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
529 		break;
530 	default:
531 		pr_info("%-15s %08x\n",
532 			reginfo->name, IXGBE_READ_REG(hw, reginfo->ofs));
533 		return;
534 	}
535 
536 	i = 0;
537 	while (i < 64) {
538 		int j;
539 		char buf[9 * 8 + 1];
540 		char *p = buf;
541 
542 		snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i, i + 7);
543 		for (j = 0; j < 8; j++)
544 			p += sprintf(p, " %08x", regs[i++]);
545 		pr_err("%-15s%s\n", rname, buf);
546 	}
547 
548 }
549 
550 static void ixgbe_print_buffer(struct ixgbe_ring *ring, int n)
551 {
552 	struct ixgbe_tx_buffer *tx_buffer;
553 
554 	tx_buffer = &ring->tx_buffer_info[ring->next_to_clean];
555 	pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
556 		n, ring->next_to_use, ring->next_to_clean,
557 		(u64)dma_unmap_addr(tx_buffer, dma),
558 		dma_unmap_len(tx_buffer, len),
559 		tx_buffer->next_to_watch,
560 		(u64)tx_buffer->time_stamp);
561 }
562 
563 /*
564  * ixgbe_dump - Print registers, tx-rings and rx-rings
565  */
566 static void ixgbe_dump(struct ixgbe_adapter *adapter)
567 {
568 	struct net_device *netdev = adapter->netdev;
569 	struct ixgbe_hw *hw = &adapter->hw;
570 	struct ixgbe_reg_info *reginfo;
571 	int n = 0;
572 	struct ixgbe_ring *ring;
573 	struct ixgbe_tx_buffer *tx_buffer;
574 	union ixgbe_adv_tx_desc *tx_desc;
575 	struct my_u0 { u64 a; u64 b; } *u0;
576 	struct ixgbe_ring *rx_ring;
577 	union ixgbe_adv_rx_desc *rx_desc;
578 	struct ixgbe_rx_buffer *rx_buffer_info;
579 	int i = 0;
580 
581 	if (!netif_msg_hw(adapter))
582 		return;
583 
584 	/* Print netdevice Info */
585 	if (netdev) {
586 		dev_info(&adapter->pdev->dev, "Net device Info\n");
587 		pr_info("Device Name     state            "
588 			"trans_start\n");
589 		pr_info("%-15s %016lX %016lX\n",
590 			netdev->name,
591 			netdev->state,
592 			dev_trans_start(netdev));
593 	}
594 
595 	/* Print Registers */
596 	dev_info(&adapter->pdev->dev, "Register Dump\n");
597 	pr_info(" Register Name   Value\n");
598 	for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
599 	     reginfo->name; reginfo++) {
600 		ixgbe_regdump(hw, reginfo);
601 	}
602 
603 	/* Print TX Ring Summary */
604 	if (!netdev || !netif_running(netdev))
605 		return;
606 
607 	dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
608 	pr_info(" %s     %s              %s        %s\n",
609 		"Queue [NTU] [NTC] [bi(ntc)->dma  ]",
610 		"leng", "ntw", "timestamp");
611 	for (n = 0; n < adapter->num_tx_queues; n++) {
612 		ring = adapter->tx_ring[n];
613 		ixgbe_print_buffer(ring, n);
614 	}
615 
616 	for (n = 0; n < adapter->num_xdp_queues; n++) {
617 		ring = adapter->xdp_ring[n];
618 		ixgbe_print_buffer(ring, n);
619 	}
620 
621 	/* Print TX Rings */
622 	if (!netif_msg_tx_done(adapter))
623 		goto rx_ring_summary;
624 
625 	dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
626 
627 	/* Transmit Descriptor Formats
628 	 *
629 	 * 82598 Advanced Transmit Descriptor
630 	 *   +--------------------------------------------------------------+
631 	 * 0 |         Buffer Address [63:0]                                |
632 	 *   +--------------------------------------------------------------+
633 	 * 8 |  PAYLEN  | POPTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
634 	 *   +--------------------------------------------------------------+
635 	 *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
636 	 *
637 	 * 82598 Advanced Transmit Descriptor (Write-Back Format)
638 	 *   +--------------------------------------------------------------+
639 	 * 0 |                          RSV [63:0]                          |
640 	 *   +--------------------------------------------------------------+
641 	 * 8 |            RSV           |  STA  |          NXTSEQ           |
642 	 *   +--------------------------------------------------------------+
643 	 *   63                       36 35   32 31                         0
644 	 *
645 	 * 82599+ Advanced Transmit Descriptor
646 	 *   +--------------------------------------------------------------+
647 	 * 0 |         Buffer Address [63:0]                                |
648 	 *   +--------------------------------------------------------------+
649 	 * 8 |PAYLEN  |POPTS|CC|IDX  |STA  |DCMD  |DTYP |MAC  |RSV  |DTALEN |
650 	 *   +--------------------------------------------------------------+
651 	 *   63     46 45 40 39 38 36 35 32 31  24 23 20 19 18 17 16 15     0
652 	 *
653 	 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
654 	 *   +--------------------------------------------------------------+
655 	 * 0 |                          RSV [63:0]                          |
656 	 *   +--------------------------------------------------------------+
657 	 * 8 |            RSV           |  STA  |           RSV             |
658 	 *   +--------------------------------------------------------------+
659 	 *   63                       36 35   32 31                         0
660 	 */
661 
662 	for (n = 0; n < adapter->num_tx_queues; n++) {
663 		ring = adapter->tx_ring[n];
664 		pr_info("------------------------------------\n");
665 		pr_info("TX QUEUE INDEX = %d\n", ring->queue_index);
666 		pr_info("------------------------------------\n");
667 		pr_info("%s%s    %s              %s        %s          %s\n",
668 			"T [desc]     [address 63:0  ] ",
669 			"[PlPOIdStDDt Ln] [bi->dma       ] ",
670 			"leng", "ntw", "timestamp", "bi->skb");
671 
672 		for (i = 0; ring->desc && (i < ring->count); i++) {
673 			tx_desc = IXGBE_TX_DESC(ring, i);
674 			tx_buffer = &ring->tx_buffer_info[i];
675 			u0 = (struct my_u0 *)tx_desc;
676 			if (dma_unmap_len(tx_buffer, len) > 0) {
677 				const char *ring_desc;
678 
679 				if (i == ring->next_to_use &&
680 				    i == ring->next_to_clean)
681 					ring_desc = " NTC/U";
682 				else if (i == ring->next_to_use)
683 					ring_desc = " NTU";
684 				else if (i == ring->next_to_clean)
685 					ring_desc = " NTC";
686 				else
687 					ring_desc = "";
688 				pr_info("T [0x%03X]    %016llX %016llX %016llX %08X %p %016llX %p%s",
689 					i,
690 					le64_to_cpu((__force __le64)u0->a),
691 					le64_to_cpu((__force __le64)u0->b),
692 					(u64)dma_unmap_addr(tx_buffer, dma),
693 					dma_unmap_len(tx_buffer, len),
694 					tx_buffer->next_to_watch,
695 					(u64)tx_buffer->time_stamp,
696 					tx_buffer->skb,
697 					ring_desc);
698 
699 				if (netif_msg_pktdata(adapter) &&
700 				    tx_buffer->skb)
701 					print_hex_dump(KERN_INFO, "",
702 						DUMP_PREFIX_ADDRESS, 16, 1,
703 						tx_buffer->skb->data,
704 						dma_unmap_len(tx_buffer, len),
705 						true);
706 			}
707 		}
708 	}
709 
710 	/* Print RX Rings Summary */
711 rx_ring_summary:
712 	dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
713 	pr_info("Queue [NTU] [NTC]\n");
714 	for (n = 0; n < adapter->num_rx_queues; n++) {
715 		rx_ring = adapter->rx_ring[n];
716 		pr_info("%5d %5X %5X\n",
717 			n, rx_ring->next_to_use, rx_ring->next_to_clean);
718 	}
719 
720 	/* Print RX Rings */
721 	if (!netif_msg_rx_status(adapter))
722 		return;
723 
724 	dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
725 
726 	/* Receive Descriptor Formats
727 	 *
728 	 * 82598 Advanced Receive Descriptor (Read) Format
729 	 *    63                                           1        0
730 	 *    +-----------------------------------------------------+
731 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
732 	 *    +----------------------------------------------+------+
733 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
734 	 *    +-----------------------------------------------------+
735 	 *
736 	 *
737 	 * 82598 Advanced Receive Descriptor (Write-Back) Format
738 	 *
739 	 *   63       48 47    32 31  30      21 20 16 15   4 3     0
740 	 *   +------------------------------------------------------+
741 	 * 0 |       RSS Hash /  |SPH| HDR_LEN  | RSV |Packet|  RSS |
742 	 *   | Packet   | IP     |   |          |     | Type | Type |
743 	 *   | Checksum | Ident  |   |          |     |      |      |
744 	 *   +------------------------------------------------------+
745 	 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
746 	 *   +------------------------------------------------------+
747 	 *   63       48 47    32 31            20 19               0
748 	 *
749 	 * 82599+ Advanced Receive Descriptor (Read) Format
750 	 *    63                                           1        0
751 	 *    +-----------------------------------------------------+
752 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
753 	 *    +----------------------------------------------+------+
754 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
755 	 *    +-----------------------------------------------------+
756 	 *
757 	 *
758 	 * 82599+ Advanced Receive Descriptor (Write-Back) Format
759 	 *
760 	 *   63       48 47    32 31  30      21 20 17 16   4 3     0
761 	 *   +------------------------------------------------------+
762 	 * 0 |RSS / Frag Checksum|SPH| HDR_LEN  |RSC- |Packet|  RSS |
763 	 *   |/ RTT / PCoE_PARAM |   |          | CNT | Type | Type |
764 	 *   |/ Flow Dir Flt ID  |   |          |     |      |      |
765 	 *   +------------------------------------------------------+
766 	 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
767 	 *   +------------------------------------------------------+
768 	 *   63       48 47    32 31          20 19                 0
769 	 */
770 
771 	for (n = 0; n < adapter->num_rx_queues; n++) {
772 		rx_ring = adapter->rx_ring[n];
773 		pr_info("------------------------------------\n");
774 		pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
775 		pr_info("------------------------------------\n");
776 		pr_info("%s%s%s\n",
777 			"R  [desc]      [ PktBuf     A0] ",
778 			"[  HeadBuf   DD] [bi->dma       ] [bi->skb       ] ",
779 			"<-- Adv Rx Read format");
780 		pr_info("%s%s%s\n",
781 			"RWB[desc]      [PcsmIpSHl PtRs] ",
782 			"[vl er S cks ln] ---------------- [bi->skb       ] ",
783 			"<-- Adv Rx Write-Back format");
784 
785 		for (i = 0; i < rx_ring->count; i++) {
786 			const char *ring_desc;
787 
788 			if (i == rx_ring->next_to_use)
789 				ring_desc = " NTU";
790 			else if (i == rx_ring->next_to_clean)
791 				ring_desc = " NTC";
792 			else
793 				ring_desc = "";
794 
795 			rx_buffer_info = &rx_ring->rx_buffer_info[i];
796 			rx_desc = IXGBE_RX_DESC(rx_ring, i);
797 			u0 = (struct my_u0 *)rx_desc;
798 			if (rx_desc->wb.upper.length) {
799 				/* Descriptor Done */
800 				pr_info("RWB[0x%03X]     %016llX %016llX ---------------- %p%s\n",
801 					i,
802 					le64_to_cpu((__force __le64)u0->a),
803 					le64_to_cpu((__force __le64)u0->b),
804 					rx_buffer_info->skb,
805 					ring_desc);
806 			} else {
807 				pr_info("R  [0x%03X]     %016llX %016llX %016llX %p%s\n",
808 					i,
809 					le64_to_cpu((__force __le64)u0->a),
810 					le64_to_cpu((__force __le64)u0->b),
811 					(u64)rx_buffer_info->dma,
812 					rx_buffer_info->skb,
813 					ring_desc);
814 
815 				if (netif_msg_pktdata(adapter) &&
816 				    rx_buffer_info->dma) {
817 					print_hex_dump(KERN_INFO, "",
818 					   DUMP_PREFIX_ADDRESS, 16, 1,
819 					   page_address(rx_buffer_info->page) +
820 						    rx_buffer_info->page_offset,
821 					   ixgbe_rx_bufsz(rx_ring), true);
822 				}
823 			}
824 		}
825 	}
826 }
827 
828 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
829 {
830 	u32 ctrl_ext;
831 
832 	/* Let firmware take over control of h/w */
833 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
834 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
835 			ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
836 }
837 
838 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
839 {
840 	u32 ctrl_ext;
841 
842 	/* Let firmware know the driver has taken over */
843 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
844 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
845 			ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
846 }
847 
848 /**
849  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
850  * @adapter: pointer to adapter struct
851  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
852  * @queue: queue to map the corresponding interrupt to
853  * @msix_vector: the vector to map to the corresponding queue
854  *
855  */
856 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
857 			   u8 queue, u8 msix_vector)
858 {
859 	u32 ivar, index;
860 	struct ixgbe_hw *hw = &adapter->hw;
861 	switch (hw->mac.type) {
862 	case ixgbe_mac_82598EB:
863 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
864 		if (direction == -1)
865 			direction = 0;
866 		index = (((direction * 64) + queue) >> 2) & 0x1F;
867 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
868 		ivar &= ~(0xFF << (8 * (queue & 0x3)));
869 		ivar |= (msix_vector << (8 * (queue & 0x3)));
870 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
871 		break;
872 	case ixgbe_mac_82599EB:
873 	case ixgbe_mac_X540:
874 	case ixgbe_mac_X550:
875 	case ixgbe_mac_X550EM_x:
876 	case ixgbe_mac_x550em_a:
877 		if (direction == -1) {
878 			/* other causes */
879 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
880 			index = ((queue & 1) * 8);
881 			ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
882 			ivar &= ~(0xFF << index);
883 			ivar |= (msix_vector << index);
884 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
885 			break;
886 		} else {
887 			/* tx or rx causes */
888 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
889 			index = ((16 * (queue & 1)) + (8 * direction));
890 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
891 			ivar &= ~(0xFF << index);
892 			ivar |= (msix_vector << index);
893 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
894 			break;
895 		}
896 	default:
897 		break;
898 	}
899 }
900 
901 void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
902 			    u64 qmask)
903 {
904 	u32 mask;
905 
906 	switch (adapter->hw.mac.type) {
907 	case ixgbe_mac_82598EB:
908 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
909 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
910 		break;
911 	case ixgbe_mac_82599EB:
912 	case ixgbe_mac_X540:
913 	case ixgbe_mac_X550:
914 	case ixgbe_mac_X550EM_x:
915 	case ixgbe_mac_x550em_a:
916 		mask = (qmask & 0xFFFFFFFF);
917 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
918 		mask = (qmask >> 32);
919 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
920 		break;
921 	default:
922 		break;
923 	}
924 }
925 
926 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
927 {
928 	struct ixgbe_hw *hw = &adapter->hw;
929 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
930 	int i;
931 	u32 data;
932 
933 	if ((hw->fc.current_mode != ixgbe_fc_full) &&
934 	    (hw->fc.current_mode != ixgbe_fc_rx_pause))
935 		return;
936 
937 	switch (hw->mac.type) {
938 	case ixgbe_mac_82598EB:
939 		data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
940 		break;
941 	default:
942 		data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
943 	}
944 	hwstats->lxoffrxc += data;
945 
946 	/* refill credits (no tx hang) if we received xoff */
947 	if (!data)
948 		return;
949 
950 	for (i = 0; i < adapter->num_tx_queues; i++)
951 		clear_bit(__IXGBE_HANG_CHECK_ARMED,
952 			  &adapter->tx_ring[i]->state);
953 
954 	for (i = 0; i < adapter->num_xdp_queues; i++)
955 		clear_bit(__IXGBE_HANG_CHECK_ARMED,
956 			  &adapter->xdp_ring[i]->state);
957 }
958 
959 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
960 {
961 	struct ixgbe_hw *hw = &adapter->hw;
962 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
963 	u32 xoff[8] = {0};
964 	u8 tc;
965 	int i;
966 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
967 
968 	if (adapter->ixgbe_ieee_pfc)
969 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
970 
971 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
972 		ixgbe_update_xoff_rx_lfc(adapter);
973 		return;
974 	}
975 
976 	/* update stats for each tc, only valid with PFC enabled */
977 	for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
978 		u32 pxoffrxc;
979 
980 		switch (hw->mac.type) {
981 		case ixgbe_mac_82598EB:
982 			pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
983 			break;
984 		default:
985 			pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
986 		}
987 		hwstats->pxoffrxc[i] += pxoffrxc;
988 		/* Get the TC for given UP */
989 		tc = netdev_get_prio_tc_map(adapter->netdev, i);
990 		xoff[tc] += pxoffrxc;
991 	}
992 
993 	/* disarm tx queues that have received xoff frames */
994 	for (i = 0; i < adapter->num_tx_queues; i++) {
995 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
996 
997 		tc = tx_ring->dcb_tc;
998 		if (xoff[tc])
999 			clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1000 	}
1001 
1002 	for (i = 0; i < adapter->num_xdp_queues; i++) {
1003 		struct ixgbe_ring *xdp_ring = adapter->xdp_ring[i];
1004 
1005 		tc = xdp_ring->dcb_tc;
1006 		if (xoff[tc])
1007 			clear_bit(__IXGBE_HANG_CHECK_ARMED, &xdp_ring->state);
1008 	}
1009 }
1010 
1011 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
1012 {
1013 	return ring->stats.packets;
1014 }
1015 
1016 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
1017 {
1018 	unsigned int head, tail;
1019 
1020 	head = ring->next_to_clean;
1021 	tail = ring->next_to_use;
1022 
1023 	return ((head <= tail) ? tail : tail + ring->count) - head;
1024 }
1025 
1026 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
1027 {
1028 	u32 tx_done = ixgbe_get_tx_completed(tx_ring);
1029 	u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
1030 	u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
1031 
1032 	clear_check_for_tx_hang(tx_ring);
1033 
1034 	/*
1035 	 * Check for a hung queue, but be thorough. This verifies
1036 	 * that a transmit has been completed since the previous
1037 	 * check AND there is at least one packet pending. The
1038 	 * ARMED bit is set to indicate a potential hang. The
1039 	 * bit is cleared if a pause frame is received to remove
1040 	 * false hang detection due to PFC or 802.3x frames. By
1041 	 * requiring this to fail twice we avoid races with
1042 	 * pfc clearing the ARMED bit and conditions where we
1043 	 * run the check_tx_hang logic with a transmit completion
1044 	 * pending but without time to complete it yet.
1045 	 */
1046 	if (tx_done_old == tx_done && tx_pending)
1047 		/* make sure it is true for two checks in a row */
1048 		return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
1049 					&tx_ring->state);
1050 	/* update completed stats and continue */
1051 	tx_ring->tx_stats.tx_done_old = tx_done;
1052 	/* reset the countdown */
1053 	clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1054 
1055 	return false;
1056 }
1057 
1058 /**
1059  * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1060  * @adapter: driver private struct
1061  **/
1062 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
1063 {
1064 
1065 	/* Do the reset outside of interrupt context */
1066 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1067 		set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
1068 		e_warn(drv, "initiating reset due to tx timeout\n");
1069 		ixgbe_service_event_schedule(adapter);
1070 	}
1071 }
1072 
1073 /**
1074  * ixgbe_tx_maxrate - callback to set the maximum per-queue bitrate
1075  * @netdev: network interface device structure
1076  * @queue_index: Tx queue to set
1077  * @maxrate: desired maximum transmit bitrate
1078  **/
1079 static int ixgbe_tx_maxrate(struct net_device *netdev,
1080 			    int queue_index, u32 maxrate)
1081 {
1082 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
1083 	struct ixgbe_hw *hw = &adapter->hw;
1084 	u32 bcnrc_val = ixgbe_link_mbps(adapter);
1085 
1086 	if (!maxrate)
1087 		return 0;
1088 
1089 	/* Calculate the rate factor values to set */
1090 	bcnrc_val <<= IXGBE_RTTBCNRC_RF_INT_SHIFT;
1091 	bcnrc_val /= maxrate;
1092 
1093 	/* clear everything but the rate factor */
1094 	bcnrc_val &= IXGBE_RTTBCNRC_RF_INT_MASK |
1095 	IXGBE_RTTBCNRC_RF_DEC_MASK;
1096 
1097 	/* enable the rate scheduler */
1098 	bcnrc_val |= IXGBE_RTTBCNRC_RS_ENA;
1099 
1100 	IXGBE_WRITE_REG(hw, IXGBE_RTTDQSEL, queue_index);
1101 	IXGBE_WRITE_REG(hw, IXGBE_RTTBCNRC, bcnrc_val);
1102 
1103 	return 0;
1104 }
1105 
1106 /**
1107  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1108  * @q_vector: structure containing interrupt and ring information
1109  * @tx_ring: tx ring to clean
1110  * @napi_budget: Used to determine if we are in netpoll
1111  **/
1112 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
1113 			       struct ixgbe_ring *tx_ring, int napi_budget)
1114 {
1115 	struct ixgbe_adapter *adapter = q_vector->adapter;
1116 	struct ixgbe_tx_buffer *tx_buffer;
1117 	union ixgbe_adv_tx_desc *tx_desc;
1118 	unsigned int total_bytes = 0, total_packets = 0, total_ipsec = 0;
1119 	unsigned int budget = q_vector->tx.work_limit;
1120 	unsigned int i = tx_ring->next_to_clean;
1121 
1122 	if (test_bit(__IXGBE_DOWN, &adapter->state))
1123 		return true;
1124 
1125 	tx_buffer = &tx_ring->tx_buffer_info[i];
1126 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
1127 	i -= tx_ring->count;
1128 
1129 	do {
1130 		union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1131 
1132 		/* if next_to_watch is not set then there is no work pending */
1133 		if (!eop_desc)
1134 			break;
1135 
1136 		/* prevent any other reads prior to eop_desc */
1137 		smp_rmb();
1138 
1139 		/* if DD is not set pending work has not been completed */
1140 		if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
1141 			break;
1142 
1143 		/* clear next_to_watch to prevent false hangs */
1144 		tx_buffer->next_to_watch = NULL;
1145 
1146 		/* update the statistics for this packet */
1147 		total_bytes += tx_buffer->bytecount;
1148 		total_packets += tx_buffer->gso_segs;
1149 		if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_IPSEC)
1150 			total_ipsec++;
1151 
1152 		/* free the skb */
1153 		if (ring_is_xdp(tx_ring))
1154 			xdp_return_frame(tx_buffer->xdpf);
1155 		else
1156 			napi_consume_skb(tx_buffer->skb, napi_budget);
1157 
1158 		/* unmap skb header data */
1159 		dma_unmap_single(tx_ring->dev,
1160 				 dma_unmap_addr(tx_buffer, dma),
1161 				 dma_unmap_len(tx_buffer, len),
1162 				 DMA_TO_DEVICE);
1163 
1164 		/* clear tx_buffer data */
1165 		dma_unmap_len_set(tx_buffer, len, 0);
1166 
1167 		/* unmap remaining buffers */
1168 		while (tx_desc != eop_desc) {
1169 			tx_buffer++;
1170 			tx_desc++;
1171 			i++;
1172 			if (unlikely(!i)) {
1173 				i -= tx_ring->count;
1174 				tx_buffer = tx_ring->tx_buffer_info;
1175 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1176 			}
1177 
1178 			/* unmap any remaining paged data */
1179 			if (dma_unmap_len(tx_buffer, len)) {
1180 				dma_unmap_page(tx_ring->dev,
1181 					       dma_unmap_addr(tx_buffer, dma),
1182 					       dma_unmap_len(tx_buffer, len),
1183 					       DMA_TO_DEVICE);
1184 				dma_unmap_len_set(tx_buffer, len, 0);
1185 			}
1186 		}
1187 
1188 		/* move us one more past the eop_desc for start of next pkt */
1189 		tx_buffer++;
1190 		tx_desc++;
1191 		i++;
1192 		if (unlikely(!i)) {
1193 			i -= tx_ring->count;
1194 			tx_buffer = tx_ring->tx_buffer_info;
1195 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1196 		}
1197 
1198 		/* issue prefetch for next Tx descriptor */
1199 		prefetch(tx_desc);
1200 
1201 		/* update budget accounting */
1202 		budget--;
1203 	} while (likely(budget));
1204 
1205 	i += tx_ring->count;
1206 	tx_ring->next_to_clean = i;
1207 	u64_stats_update_begin(&tx_ring->syncp);
1208 	tx_ring->stats.bytes += total_bytes;
1209 	tx_ring->stats.packets += total_packets;
1210 	u64_stats_update_end(&tx_ring->syncp);
1211 	q_vector->tx.total_bytes += total_bytes;
1212 	q_vector->tx.total_packets += total_packets;
1213 	adapter->tx_ipsec += total_ipsec;
1214 
1215 	if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1216 		/* schedule immediate reset if we believe we hung */
1217 		struct ixgbe_hw *hw = &adapter->hw;
1218 		e_err(drv, "Detected Tx Unit Hang %s\n"
1219 			"  Tx Queue             <%d>\n"
1220 			"  TDH, TDT             <%x>, <%x>\n"
1221 			"  next_to_use          <%x>\n"
1222 			"  next_to_clean        <%x>\n"
1223 			"tx_buffer_info[next_to_clean]\n"
1224 			"  time_stamp           <%lx>\n"
1225 			"  jiffies              <%lx>\n",
1226 			ring_is_xdp(tx_ring) ? "(XDP)" : "",
1227 			tx_ring->queue_index,
1228 			IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1229 			IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1230 			tx_ring->next_to_use, i,
1231 			tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1232 
1233 		if (!ring_is_xdp(tx_ring))
1234 			netif_stop_subqueue(tx_ring->netdev,
1235 					    tx_ring->queue_index);
1236 
1237 		e_info(probe,
1238 		       "tx hang %d detected on queue %d, resetting adapter\n",
1239 			adapter->tx_timeout_count + 1, tx_ring->queue_index);
1240 
1241 		/* schedule immediate reset if we believe we hung */
1242 		ixgbe_tx_timeout_reset(adapter);
1243 
1244 		/* the adapter is about to reset, no point in enabling stuff */
1245 		return true;
1246 	}
1247 
1248 	if (ring_is_xdp(tx_ring))
1249 		return !!budget;
1250 
1251 	netdev_tx_completed_queue(txring_txq(tx_ring),
1252 				  total_packets, total_bytes);
1253 
1254 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1255 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1256 		     (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1257 		/* Make sure that anybody stopping the queue after this
1258 		 * sees the new next_to_clean.
1259 		 */
1260 		smp_mb();
1261 		if (__netif_subqueue_stopped(tx_ring->netdev,
1262 					     tx_ring->queue_index)
1263 		    && !test_bit(__IXGBE_DOWN, &adapter->state)) {
1264 			netif_wake_subqueue(tx_ring->netdev,
1265 					    tx_ring->queue_index);
1266 			++tx_ring->tx_stats.restart_queue;
1267 		}
1268 	}
1269 
1270 	return !!budget;
1271 }
1272 
1273 #ifdef CONFIG_IXGBE_DCA
1274 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1275 				struct ixgbe_ring *tx_ring,
1276 				int cpu)
1277 {
1278 	struct ixgbe_hw *hw = &adapter->hw;
1279 	u32 txctrl = 0;
1280 	u16 reg_offset;
1281 
1282 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1283 		txctrl = dca3_get_tag(tx_ring->dev, cpu);
1284 
1285 	switch (hw->mac.type) {
1286 	case ixgbe_mac_82598EB:
1287 		reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1288 		break;
1289 	case ixgbe_mac_82599EB:
1290 	case ixgbe_mac_X540:
1291 		reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1292 		txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1293 		break;
1294 	default:
1295 		/* for unknown hardware do not write register */
1296 		return;
1297 	}
1298 
1299 	/*
1300 	 * We can enable relaxed ordering for reads, but not writes when
1301 	 * DCA is enabled.  This is due to a known issue in some chipsets
1302 	 * which will cause the DCA tag to be cleared.
1303 	 */
1304 	txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1305 		  IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1306 		  IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1307 
1308 	IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1309 }
1310 
1311 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1312 				struct ixgbe_ring *rx_ring,
1313 				int cpu)
1314 {
1315 	struct ixgbe_hw *hw = &adapter->hw;
1316 	u32 rxctrl = 0;
1317 	u8 reg_idx = rx_ring->reg_idx;
1318 
1319 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1320 		rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1321 
1322 	switch (hw->mac.type) {
1323 	case ixgbe_mac_82599EB:
1324 	case ixgbe_mac_X540:
1325 		rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1326 		break;
1327 	default:
1328 		break;
1329 	}
1330 
1331 	/*
1332 	 * We can enable relaxed ordering for reads, but not writes when
1333 	 * DCA is enabled.  This is due to a known issue in some chipsets
1334 	 * which will cause the DCA tag to be cleared.
1335 	 */
1336 	rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1337 		  IXGBE_DCA_RXCTRL_DATA_DCA_EN |
1338 		  IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1339 
1340 	IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1341 }
1342 
1343 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1344 {
1345 	struct ixgbe_adapter *adapter = q_vector->adapter;
1346 	struct ixgbe_ring *ring;
1347 	int cpu = get_cpu();
1348 
1349 	if (q_vector->cpu == cpu)
1350 		goto out_no_update;
1351 
1352 	ixgbe_for_each_ring(ring, q_vector->tx)
1353 		ixgbe_update_tx_dca(adapter, ring, cpu);
1354 
1355 	ixgbe_for_each_ring(ring, q_vector->rx)
1356 		ixgbe_update_rx_dca(adapter, ring, cpu);
1357 
1358 	q_vector->cpu = cpu;
1359 out_no_update:
1360 	put_cpu();
1361 }
1362 
1363 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1364 {
1365 	int i;
1366 
1367 	/* always use CB2 mode, difference is masked in the CB driver */
1368 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1369 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1370 				IXGBE_DCA_CTRL_DCA_MODE_CB2);
1371 	else
1372 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1373 				IXGBE_DCA_CTRL_DCA_DISABLE);
1374 
1375 	for (i = 0; i < adapter->num_q_vectors; i++) {
1376 		adapter->q_vector[i]->cpu = -1;
1377 		ixgbe_update_dca(adapter->q_vector[i]);
1378 	}
1379 }
1380 
1381 static int __ixgbe_notify_dca(struct device *dev, void *data)
1382 {
1383 	struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1384 	unsigned long event = *(unsigned long *)data;
1385 
1386 	if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1387 		return 0;
1388 
1389 	switch (event) {
1390 	case DCA_PROVIDER_ADD:
1391 		/* if we're already enabled, don't do it again */
1392 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1393 			break;
1394 		if (dca_add_requester(dev) == 0) {
1395 			adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1396 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1397 					IXGBE_DCA_CTRL_DCA_MODE_CB2);
1398 			break;
1399 		}
1400 		/* fall through - DCA is disabled. */
1401 	case DCA_PROVIDER_REMOVE:
1402 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1403 			dca_remove_requester(dev);
1404 			adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1405 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1406 					IXGBE_DCA_CTRL_DCA_DISABLE);
1407 		}
1408 		break;
1409 	}
1410 
1411 	return 0;
1412 }
1413 
1414 #endif /* CONFIG_IXGBE_DCA */
1415 
1416 #define IXGBE_RSS_L4_TYPES_MASK \
1417 	((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
1418 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
1419 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
1420 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
1421 
1422 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1423 				 union ixgbe_adv_rx_desc *rx_desc,
1424 				 struct sk_buff *skb)
1425 {
1426 	u16 rss_type;
1427 
1428 	if (!(ring->netdev->features & NETIF_F_RXHASH))
1429 		return;
1430 
1431 	rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
1432 		   IXGBE_RXDADV_RSSTYPE_MASK;
1433 
1434 	if (!rss_type)
1435 		return;
1436 
1437 	skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1438 		     (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
1439 		     PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1440 }
1441 
1442 #ifdef IXGBE_FCOE
1443 /**
1444  * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1445  * @ring: structure containing ring specific data
1446  * @rx_desc: advanced rx descriptor
1447  *
1448  * Returns : true if it is FCoE pkt
1449  */
1450 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1451 				    union ixgbe_adv_rx_desc *rx_desc)
1452 {
1453 	__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1454 
1455 	return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1456 	       ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1457 		(cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1458 			     IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1459 }
1460 
1461 #endif /* IXGBE_FCOE */
1462 /**
1463  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1464  * @ring: structure containing ring specific data
1465  * @rx_desc: current Rx descriptor being processed
1466  * @skb: skb currently being received and modified
1467  **/
1468 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1469 				     union ixgbe_adv_rx_desc *rx_desc,
1470 				     struct sk_buff *skb)
1471 {
1472 	__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1473 	bool encap_pkt = false;
1474 
1475 	skb_checksum_none_assert(skb);
1476 
1477 	/* Rx csum disabled */
1478 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
1479 		return;
1480 
1481 	/* check for VXLAN and Geneve packets */
1482 	if (pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_VXLAN)) {
1483 		encap_pkt = true;
1484 		skb->encapsulation = 1;
1485 	}
1486 
1487 	/* if IP and error */
1488 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1489 	    ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1490 		ring->rx_stats.csum_err++;
1491 		return;
1492 	}
1493 
1494 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1495 		return;
1496 
1497 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1498 		/*
1499 		 * 82599 errata, UDP frames with a 0 checksum can be marked as
1500 		 * checksum errors.
1501 		 */
1502 		if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1503 		    test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1504 			return;
1505 
1506 		ring->rx_stats.csum_err++;
1507 		return;
1508 	}
1509 
1510 	/* It must be a TCP or UDP packet with a valid checksum */
1511 	skb->ip_summed = CHECKSUM_UNNECESSARY;
1512 	if (encap_pkt) {
1513 		if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_OUTERIPCS))
1514 			return;
1515 
1516 		if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_OUTERIPER)) {
1517 			skb->ip_summed = CHECKSUM_NONE;
1518 			return;
1519 		}
1520 		/* If we checked the outer header let the stack know */
1521 		skb->csum_level = 1;
1522 	}
1523 }
1524 
1525 static inline unsigned int ixgbe_rx_offset(struct ixgbe_ring *rx_ring)
1526 {
1527 	return ring_uses_build_skb(rx_ring) ? IXGBE_SKB_PAD : 0;
1528 }
1529 
1530 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1531 				    struct ixgbe_rx_buffer *bi)
1532 {
1533 	struct page *page = bi->page;
1534 	dma_addr_t dma;
1535 
1536 	/* since we are recycling buffers we should seldom need to alloc */
1537 	if (likely(page))
1538 		return true;
1539 
1540 	/* alloc new page for storage */
1541 	page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
1542 	if (unlikely(!page)) {
1543 		rx_ring->rx_stats.alloc_rx_page_failed++;
1544 		return false;
1545 	}
1546 
1547 	/* map page for use */
1548 	dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1549 				 ixgbe_rx_pg_size(rx_ring),
1550 				 DMA_FROM_DEVICE,
1551 				 IXGBE_RX_DMA_ATTR);
1552 
1553 	/*
1554 	 * if mapping failed free memory back to system since
1555 	 * there isn't much point in holding memory we can't use
1556 	 */
1557 	if (dma_mapping_error(rx_ring->dev, dma)) {
1558 		__free_pages(page, ixgbe_rx_pg_order(rx_ring));
1559 
1560 		rx_ring->rx_stats.alloc_rx_page_failed++;
1561 		return false;
1562 	}
1563 
1564 	bi->dma = dma;
1565 	bi->page = page;
1566 	bi->page_offset = ixgbe_rx_offset(rx_ring);
1567 	page_ref_add(page, USHRT_MAX - 1);
1568 	bi->pagecnt_bias = USHRT_MAX;
1569 	rx_ring->rx_stats.alloc_rx_page++;
1570 
1571 	return true;
1572 }
1573 
1574 /**
1575  * ixgbe_alloc_rx_buffers - Replace used receive buffers
1576  * @rx_ring: ring to place buffers on
1577  * @cleaned_count: number of buffers to replace
1578  **/
1579 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1580 {
1581 	union ixgbe_adv_rx_desc *rx_desc;
1582 	struct ixgbe_rx_buffer *bi;
1583 	u16 i = rx_ring->next_to_use;
1584 	u16 bufsz;
1585 
1586 	/* nothing to do */
1587 	if (!cleaned_count)
1588 		return;
1589 
1590 	rx_desc = IXGBE_RX_DESC(rx_ring, i);
1591 	bi = &rx_ring->rx_buffer_info[i];
1592 	i -= rx_ring->count;
1593 
1594 	bufsz = ixgbe_rx_bufsz(rx_ring);
1595 
1596 	do {
1597 		if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1598 			break;
1599 
1600 		/* sync the buffer for use by the device */
1601 		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1602 						 bi->page_offset, bufsz,
1603 						 DMA_FROM_DEVICE);
1604 
1605 		/*
1606 		 * Refresh the desc even if buffer_addrs didn't change
1607 		 * because each write-back erases this info.
1608 		 */
1609 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1610 
1611 		rx_desc++;
1612 		bi++;
1613 		i++;
1614 		if (unlikely(!i)) {
1615 			rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1616 			bi = rx_ring->rx_buffer_info;
1617 			i -= rx_ring->count;
1618 		}
1619 
1620 		/* clear the length for the next_to_use descriptor */
1621 		rx_desc->wb.upper.length = 0;
1622 
1623 		cleaned_count--;
1624 	} while (cleaned_count);
1625 
1626 	i += rx_ring->count;
1627 
1628 	if (rx_ring->next_to_use != i) {
1629 		rx_ring->next_to_use = i;
1630 
1631 		/* update next to alloc since we have filled the ring */
1632 		rx_ring->next_to_alloc = i;
1633 
1634 		/* Force memory writes to complete before letting h/w
1635 		 * know there are new descriptors to fetch.  (Only
1636 		 * applicable for weak-ordered memory model archs,
1637 		 * such as IA-64).
1638 		 */
1639 		wmb();
1640 		writel(i, rx_ring->tail);
1641 	}
1642 }
1643 
1644 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1645 				   struct sk_buff *skb)
1646 {
1647 	u16 hdr_len = skb_headlen(skb);
1648 
1649 	/* set gso_size to avoid messing up TCP MSS */
1650 	skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1651 						 IXGBE_CB(skb)->append_cnt);
1652 	skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1653 }
1654 
1655 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1656 				   struct sk_buff *skb)
1657 {
1658 	/* if append_cnt is 0 then frame is not RSC */
1659 	if (!IXGBE_CB(skb)->append_cnt)
1660 		return;
1661 
1662 	rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1663 	rx_ring->rx_stats.rsc_flush++;
1664 
1665 	ixgbe_set_rsc_gso_size(rx_ring, skb);
1666 
1667 	/* gso_size is computed using append_cnt so always clear it last */
1668 	IXGBE_CB(skb)->append_cnt = 0;
1669 }
1670 
1671 /**
1672  * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1673  * @rx_ring: rx descriptor ring packet is being transacted on
1674  * @rx_desc: pointer to the EOP Rx descriptor
1675  * @skb: pointer to current skb being populated
1676  *
1677  * This function checks the ring, descriptor, and packet information in
1678  * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1679  * other fields within the skb.
1680  **/
1681 void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1682 			      union ixgbe_adv_rx_desc *rx_desc,
1683 			      struct sk_buff *skb)
1684 {
1685 	struct net_device *dev = rx_ring->netdev;
1686 	u32 flags = rx_ring->q_vector->adapter->flags;
1687 
1688 	ixgbe_update_rsc_stats(rx_ring, skb);
1689 
1690 	ixgbe_rx_hash(rx_ring, rx_desc, skb);
1691 
1692 	ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1693 
1694 	if (unlikely(flags & IXGBE_FLAG_RX_HWTSTAMP_ENABLED))
1695 		ixgbe_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
1696 
1697 	if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1698 	    ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1699 		u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1700 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1701 	}
1702 
1703 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_STAT_SECP))
1704 		ixgbe_ipsec_rx(rx_ring, rx_desc, skb);
1705 
1706 	/* record Rx queue, or update MACVLAN statistics */
1707 	if (netif_is_ixgbe(dev))
1708 		skb_record_rx_queue(skb, rx_ring->queue_index);
1709 	else
1710 		macvlan_count_rx(netdev_priv(dev), skb->len + ETH_HLEN, true,
1711 				 false);
1712 
1713 	skb->protocol = eth_type_trans(skb, dev);
1714 }
1715 
1716 void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1717 		  struct sk_buff *skb)
1718 {
1719 	napi_gro_receive(&q_vector->napi, skb);
1720 }
1721 
1722 /**
1723  * ixgbe_is_non_eop - process handling of non-EOP buffers
1724  * @rx_ring: Rx ring being processed
1725  * @rx_desc: Rx descriptor for current buffer
1726  * @skb: Current socket buffer containing buffer in progress
1727  *
1728  * This function updates next to clean.  If the buffer is an EOP buffer
1729  * this function exits returning false, otherwise it will place the
1730  * sk_buff in the next buffer to be chained and return true indicating
1731  * that this is in fact a non-EOP buffer.
1732  **/
1733 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1734 			     union ixgbe_adv_rx_desc *rx_desc,
1735 			     struct sk_buff *skb)
1736 {
1737 	u32 ntc = rx_ring->next_to_clean + 1;
1738 
1739 	/* fetch, update, and store next to clean */
1740 	ntc = (ntc < rx_ring->count) ? ntc : 0;
1741 	rx_ring->next_to_clean = ntc;
1742 
1743 	prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1744 
1745 	/* update RSC append count if present */
1746 	if (ring_is_rsc_enabled(rx_ring)) {
1747 		__le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1748 				     cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1749 
1750 		if (unlikely(rsc_enabled)) {
1751 			u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1752 
1753 			rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1754 			IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1755 
1756 			/* update ntc based on RSC value */
1757 			ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1758 			ntc &= IXGBE_RXDADV_NEXTP_MASK;
1759 			ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1760 		}
1761 	}
1762 
1763 	/* if we are the last buffer then there is nothing else to do */
1764 	if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1765 		return false;
1766 
1767 	/* place skb in next buffer to be received */
1768 	rx_ring->rx_buffer_info[ntc].skb = skb;
1769 	rx_ring->rx_stats.non_eop_descs++;
1770 
1771 	return true;
1772 }
1773 
1774 /**
1775  * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1776  * @rx_ring: rx descriptor ring packet is being transacted on
1777  * @skb: pointer to current skb being adjusted
1778  *
1779  * This function is an ixgbe specific version of __pskb_pull_tail.  The
1780  * main difference between this version and the original function is that
1781  * this function can make several assumptions about the state of things
1782  * that allow for significant optimizations versus the standard function.
1783  * As a result we can do things like drop a frag and maintain an accurate
1784  * truesize for the skb.
1785  */
1786 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1787 			    struct sk_buff *skb)
1788 {
1789 	skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
1790 	unsigned char *va;
1791 	unsigned int pull_len;
1792 
1793 	/*
1794 	 * it is valid to use page_address instead of kmap since we are
1795 	 * working with pages allocated out of the lomem pool per
1796 	 * alloc_page(GFP_ATOMIC)
1797 	 */
1798 	va = skb_frag_address(frag);
1799 
1800 	/*
1801 	 * we need the header to contain the greater of either ETH_HLEN or
1802 	 * 60 bytes if the skb->len is less than 60 for skb_pad.
1803 	 */
1804 	pull_len = eth_get_headlen(skb->dev, va, IXGBE_RX_HDR_SIZE);
1805 
1806 	/* align pull length to size of long to optimize memcpy performance */
1807 	skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1808 
1809 	/* update all of the pointers */
1810 	skb_frag_size_sub(frag, pull_len);
1811 	skb_frag_off_add(frag, pull_len);
1812 	skb->data_len -= pull_len;
1813 	skb->tail += pull_len;
1814 }
1815 
1816 /**
1817  * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1818  * @rx_ring: rx descriptor ring packet is being transacted on
1819  * @skb: pointer to current skb being updated
1820  *
1821  * This function provides a basic DMA sync up for the first fragment of an
1822  * skb.  The reason for doing this is that the first fragment cannot be
1823  * unmapped until we have reached the end of packet descriptor for a buffer
1824  * chain.
1825  */
1826 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1827 				struct sk_buff *skb)
1828 {
1829 	if (ring_uses_build_skb(rx_ring)) {
1830 		unsigned long offset = (unsigned long)(skb->data) & ~PAGE_MASK;
1831 
1832 		dma_sync_single_range_for_cpu(rx_ring->dev,
1833 					      IXGBE_CB(skb)->dma,
1834 					      offset,
1835 					      skb_headlen(skb),
1836 					      DMA_FROM_DEVICE);
1837 	} else {
1838 		skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
1839 
1840 		dma_sync_single_range_for_cpu(rx_ring->dev,
1841 					      IXGBE_CB(skb)->dma,
1842 					      skb_frag_off(frag),
1843 					      skb_frag_size(frag),
1844 					      DMA_FROM_DEVICE);
1845 	}
1846 
1847 	/* If the page was released, just unmap it. */
1848 	if (unlikely(IXGBE_CB(skb)->page_released)) {
1849 		dma_unmap_page_attrs(rx_ring->dev, IXGBE_CB(skb)->dma,
1850 				     ixgbe_rx_pg_size(rx_ring),
1851 				     DMA_FROM_DEVICE,
1852 				     IXGBE_RX_DMA_ATTR);
1853 	}
1854 }
1855 
1856 /**
1857  * ixgbe_cleanup_headers - Correct corrupted or empty headers
1858  * @rx_ring: rx descriptor ring packet is being transacted on
1859  * @rx_desc: pointer to the EOP Rx descriptor
1860  * @skb: pointer to current skb being fixed
1861  *
1862  * Check if the skb is valid in the XDP case it will be an error pointer.
1863  * Return true in this case to abort processing and advance to next
1864  * descriptor.
1865  *
1866  * Check for corrupted packet headers caused by senders on the local L2
1867  * embedded NIC switch not setting up their Tx Descriptors right.  These
1868  * should be very rare.
1869  *
1870  * Also address the case where we are pulling data in on pages only
1871  * and as such no data is present in the skb header.
1872  *
1873  * In addition if skb is not at least 60 bytes we need to pad it so that
1874  * it is large enough to qualify as a valid Ethernet frame.
1875  *
1876  * Returns true if an error was encountered and skb was freed.
1877  **/
1878 bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1879 			   union ixgbe_adv_rx_desc *rx_desc,
1880 			   struct sk_buff *skb)
1881 {
1882 	struct net_device *netdev = rx_ring->netdev;
1883 
1884 	/* XDP packets use error pointer so abort at this point */
1885 	if (IS_ERR(skb))
1886 		return true;
1887 
1888 	/* Verify netdev is present, and that packet does not have any
1889 	 * errors that would be unacceptable to the netdev.
1890 	 */
1891 	if (!netdev ||
1892 	    (unlikely(ixgbe_test_staterr(rx_desc,
1893 					 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1894 	     !(netdev->features & NETIF_F_RXALL)))) {
1895 		dev_kfree_skb_any(skb);
1896 		return true;
1897 	}
1898 
1899 	/* place header in linear portion of buffer */
1900 	if (!skb_headlen(skb))
1901 		ixgbe_pull_tail(rx_ring, skb);
1902 
1903 #ifdef IXGBE_FCOE
1904 	/* do not attempt to pad FCoE Frames as this will disrupt DDP */
1905 	if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1906 		return false;
1907 
1908 #endif
1909 	/* if eth_skb_pad returns an error the skb was freed */
1910 	if (eth_skb_pad(skb))
1911 		return true;
1912 
1913 	return false;
1914 }
1915 
1916 /**
1917  * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1918  * @rx_ring: rx descriptor ring to store buffers on
1919  * @old_buff: donor buffer to have page reused
1920  *
1921  * Synchronizes page for reuse by the adapter
1922  **/
1923 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1924 				struct ixgbe_rx_buffer *old_buff)
1925 {
1926 	struct ixgbe_rx_buffer *new_buff;
1927 	u16 nta = rx_ring->next_to_alloc;
1928 
1929 	new_buff = &rx_ring->rx_buffer_info[nta];
1930 
1931 	/* update, and store next to alloc */
1932 	nta++;
1933 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1934 
1935 	/* Transfer page from old buffer to new buffer.
1936 	 * Move each member individually to avoid possible store
1937 	 * forwarding stalls and unnecessary copy of skb.
1938 	 */
1939 	new_buff->dma		= old_buff->dma;
1940 	new_buff->page		= old_buff->page;
1941 	new_buff->page_offset	= old_buff->page_offset;
1942 	new_buff->pagecnt_bias	= old_buff->pagecnt_bias;
1943 }
1944 
1945 static inline bool ixgbe_page_is_reserved(struct page *page)
1946 {
1947 	return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1948 }
1949 
1950 static bool ixgbe_can_reuse_rx_page(struct ixgbe_rx_buffer *rx_buffer)
1951 {
1952 	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1953 	struct page *page = rx_buffer->page;
1954 
1955 	/* avoid re-using remote pages */
1956 	if (unlikely(ixgbe_page_is_reserved(page)))
1957 		return false;
1958 
1959 #if (PAGE_SIZE < 8192)
1960 	/* if we are only owner of page we can reuse it */
1961 	if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
1962 		return false;
1963 #else
1964 	/* The last offset is a bit aggressive in that we assume the
1965 	 * worst case of FCoE being enabled and using a 3K buffer.
1966 	 * However this should have minimal impact as the 1K extra is
1967 	 * still less than one buffer in size.
1968 	 */
1969 #define IXGBE_LAST_OFFSET \
1970 	(SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBE_RXBUFFER_3K)
1971 	if (rx_buffer->page_offset > IXGBE_LAST_OFFSET)
1972 		return false;
1973 #endif
1974 
1975 	/* If we have drained the page fragment pool we need to update
1976 	 * the pagecnt_bias and page count so that we fully restock the
1977 	 * number of references the driver holds.
1978 	 */
1979 	if (unlikely(pagecnt_bias == 1)) {
1980 		page_ref_add(page, USHRT_MAX - 1);
1981 		rx_buffer->pagecnt_bias = USHRT_MAX;
1982 	}
1983 
1984 	return true;
1985 }
1986 
1987 /**
1988  * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1989  * @rx_ring: rx descriptor ring to transact packets on
1990  * @rx_buffer: buffer containing page to add
1991  * @skb: sk_buff to place the data into
1992  * @size: size of data in rx_buffer
1993  *
1994  * This function will add the data contained in rx_buffer->page to the skb.
1995  * This is done either through a direct copy if the data in the buffer is
1996  * less than the skb header size, otherwise it will just attach the page as
1997  * a frag to the skb.
1998  *
1999  * The function will then update the page offset if necessary and return
2000  * true if the buffer can be reused by the adapter.
2001  **/
2002 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
2003 			      struct ixgbe_rx_buffer *rx_buffer,
2004 			      struct sk_buff *skb,
2005 			      unsigned int size)
2006 {
2007 #if (PAGE_SIZE < 8192)
2008 	unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2009 #else
2010 	unsigned int truesize = ring_uses_build_skb(rx_ring) ?
2011 				SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) :
2012 				SKB_DATA_ALIGN(size);
2013 #endif
2014 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
2015 			rx_buffer->page_offset, size, truesize);
2016 #if (PAGE_SIZE < 8192)
2017 	rx_buffer->page_offset ^= truesize;
2018 #else
2019 	rx_buffer->page_offset += truesize;
2020 #endif
2021 }
2022 
2023 static struct ixgbe_rx_buffer *ixgbe_get_rx_buffer(struct ixgbe_ring *rx_ring,
2024 						   union ixgbe_adv_rx_desc *rx_desc,
2025 						   struct sk_buff **skb,
2026 						   const unsigned int size)
2027 {
2028 	struct ixgbe_rx_buffer *rx_buffer;
2029 
2030 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
2031 	prefetchw(rx_buffer->page);
2032 	*skb = rx_buffer->skb;
2033 
2034 	/* Delay unmapping of the first packet. It carries the header
2035 	 * information, HW may still access the header after the writeback.
2036 	 * Only unmap it when EOP is reached
2037 	 */
2038 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)) {
2039 		if (!*skb)
2040 			goto skip_sync;
2041 	} else {
2042 		if (*skb)
2043 			ixgbe_dma_sync_frag(rx_ring, *skb);
2044 	}
2045 
2046 	/* we are reusing so sync this buffer for CPU use */
2047 	dma_sync_single_range_for_cpu(rx_ring->dev,
2048 				      rx_buffer->dma,
2049 				      rx_buffer->page_offset,
2050 				      size,
2051 				      DMA_FROM_DEVICE);
2052 skip_sync:
2053 	rx_buffer->pagecnt_bias--;
2054 
2055 	return rx_buffer;
2056 }
2057 
2058 static void ixgbe_put_rx_buffer(struct ixgbe_ring *rx_ring,
2059 				struct ixgbe_rx_buffer *rx_buffer,
2060 				struct sk_buff *skb)
2061 {
2062 	if (ixgbe_can_reuse_rx_page(rx_buffer)) {
2063 		/* hand second half of page back to the ring */
2064 		ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2065 	} else {
2066 		if (!IS_ERR(skb) && IXGBE_CB(skb)->dma == rx_buffer->dma) {
2067 			/* the page has been released from the ring */
2068 			IXGBE_CB(skb)->page_released = true;
2069 		} else {
2070 			/* we are not reusing the buffer so unmap it */
2071 			dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2072 					     ixgbe_rx_pg_size(rx_ring),
2073 					     DMA_FROM_DEVICE,
2074 					     IXGBE_RX_DMA_ATTR);
2075 		}
2076 		__page_frag_cache_drain(rx_buffer->page,
2077 					rx_buffer->pagecnt_bias);
2078 	}
2079 
2080 	/* clear contents of rx_buffer */
2081 	rx_buffer->page = NULL;
2082 	rx_buffer->skb = NULL;
2083 }
2084 
2085 static struct sk_buff *ixgbe_construct_skb(struct ixgbe_ring *rx_ring,
2086 					   struct ixgbe_rx_buffer *rx_buffer,
2087 					   struct xdp_buff *xdp,
2088 					   union ixgbe_adv_rx_desc *rx_desc)
2089 {
2090 	unsigned int size = xdp->data_end - xdp->data;
2091 #if (PAGE_SIZE < 8192)
2092 	unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2093 #else
2094 	unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
2095 					       xdp->data_hard_start);
2096 #endif
2097 	struct sk_buff *skb;
2098 
2099 	/* prefetch first cache line of first page */
2100 	prefetch(xdp->data);
2101 #if L1_CACHE_BYTES < 128
2102 	prefetch(xdp->data + L1_CACHE_BYTES);
2103 #endif
2104 	/* Note, we get here by enabling legacy-rx via:
2105 	 *
2106 	 *    ethtool --set-priv-flags <dev> legacy-rx on
2107 	 *
2108 	 * In this mode, we currently get 0 extra XDP headroom as
2109 	 * opposed to having legacy-rx off, where we process XDP
2110 	 * packets going to stack via ixgbe_build_skb(). The latter
2111 	 * provides us currently with 192 bytes of headroom.
2112 	 *
2113 	 * For ixgbe_construct_skb() mode it means that the
2114 	 * xdp->data_meta will always point to xdp->data, since
2115 	 * the helper cannot expand the head. Should this ever
2116 	 * change in future for legacy-rx mode on, then lets also
2117 	 * add xdp->data_meta handling here.
2118 	 */
2119 
2120 	/* allocate a skb to store the frags */
2121 	skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBE_RX_HDR_SIZE);
2122 	if (unlikely(!skb))
2123 		return NULL;
2124 
2125 	if (size > IXGBE_RX_HDR_SIZE) {
2126 		if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2127 			IXGBE_CB(skb)->dma = rx_buffer->dma;
2128 
2129 		skb_add_rx_frag(skb, 0, rx_buffer->page,
2130 				xdp->data - page_address(rx_buffer->page),
2131 				size, truesize);
2132 #if (PAGE_SIZE < 8192)
2133 		rx_buffer->page_offset ^= truesize;
2134 #else
2135 		rx_buffer->page_offset += truesize;
2136 #endif
2137 	} else {
2138 		memcpy(__skb_put(skb, size),
2139 		       xdp->data, ALIGN(size, sizeof(long)));
2140 		rx_buffer->pagecnt_bias++;
2141 	}
2142 
2143 	return skb;
2144 }
2145 
2146 static struct sk_buff *ixgbe_build_skb(struct ixgbe_ring *rx_ring,
2147 				       struct ixgbe_rx_buffer *rx_buffer,
2148 				       struct xdp_buff *xdp,
2149 				       union ixgbe_adv_rx_desc *rx_desc)
2150 {
2151 	unsigned int metasize = xdp->data - xdp->data_meta;
2152 #if (PAGE_SIZE < 8192)
2153 	unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2154 #else
2155 	unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
2156 				SKB_DATA_ALIGN(xdp->data_end -
2157 					       xdp->data_hard_start);
2158 #endif
2159 	struct sk_buff *skb;
2160 
2161 	/* Prefetch first cache line of first page. If xdp->data_meta
2162 	 * is unused, this points extactly as xdp->data, otherwise we
2163 	 * likely have a consumer accessing first few bytes of meta
2164 	 * data, and then actual data.
2165 	 */
2166 	prefetch(xdp->data_meta);
2167 #if L1_CACHE_BYTES < 128
2168 	prefetch(xdp->data_meta + L1_CACHE_BYTES);
2169 #endif
2170 
2171 	/* build an skb to around the page buffer */
2172 	skb = build_skb(xdp->data_hard_start, truesize);
2173 	if (unlikely(!skb))
2174 		return NULL;
2175 
2176 	/* update pointers within the skb to store the data */
2177 	skb_reserve(skb, xdp->data - xdp->data_hard_start);
2178 	__skb_put(skb, xdp->data_end - xdp->data);
2179 	if (metasize)
2180 		skb_metadata_set(skb, metasize);
2181 
2182 	/* record DMA address if this is the start of a chain of buffers */
2183 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2184 		IXGBE_CB(skb)->dma = rx_buffer->dma;
2185 
2186 	/* update buffer offset */
2187 #if (PAGE_SIZE < 8192)
2188 	rx_buffer->page_offset ^= truesize;
2189 #else
2190 	rx_buffer->page_offset += truesize;
2191 #endif
2192 
2193 	return skb;
2194 }
2195 
2196 static struct sk_buff *ixgbe_run_xdp(struct ixgbe_adapter *adapter,
2197 				     struct ixgbe_ring *rx_ring,
2198 				     struct xdp_buff *xdp)
2199 {
2200 	int err, result = IXGBE_XDP_PASS;
2201 	struct bpf_prog *xdp_prog;
2202 	struct xdp_frame *xdpf;
2203 	u32 act;
2204 
2205 	rcu_read_lock();
2206 	xdp_prog = READ_ONCE(rx_ring->xdp_prog);
2207 
2208 	if (!xdp_prog)
2209 		goto xdp_out;
2210 
2211 	prefetchw(xdp->data_hard_start); /* xdp_frame write */
2212 
2213 	act = bpf_prog_run_xdp(xdp_prog, xdp);
2214 	switch (act) {
2215 	case XDP_PASS:
2216 		break;
2217 	case XDP_TX:
2218 		xdpf = xdp_convert_buff_to_frame(xdp);
2219 		if (unlikely(!xdpf)) {
2220 			result = IXGBE_XDP_CONSUMED;
2221 			break;
2222 		}
2223 		result = ixgbe_xmit_xdp_ring(adapter, xdpf);
2224 		break;
2225 	case XDP_REDIRECT:
2226 		err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
2227 		if (!err)
2228 			result = IXGBE_XDP_REDIR;
2229 		else
2230 			result = IXGBE_XDP_CONSUMED;
2231 		break;
2232 	default:
2233 		bpf_warn_invalid_xdp_action(act);
2234 		/* fallthrough */
2235 	case XDP_ABORTED:
2236 		trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
2237 		/* fallthrough -- handle aborts by dropping packet */
2238 	case XDP_DROP:
2239 		result = IXGBE_XDP_CONSUMED;
2240 		break;
2241 	}
2242 xdp_out:
2243 	rcu_read_unlock();
2244 	return ERR_PTR(-result);
2245 }
2246 
2247 static unsigned int ixgbe_rx_frame_truesize(struct ixgbe_ring *rx_ring,
2248 					    unsigned int size)
2249 {
2250 	unsigned int truesize;
2251 
2252 #if (PAGE_SIZE < 8192)
2253 	truesize = ixgbe_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
2254 #else
2255 	truesize = ring_uses_build_skb(rx_ring) ?
2256 		SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) +
2257 		SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
2258 		SKB_DATA_ALIGN(size);
2259 #endif
2260 	return truesize;
2261 }
2262 
2263 static void ixgbe_rx_buffer_flip(struct ixgbe_ring *rx_ring,
2264 				 struct ixgbe_rx_buffer *rx_buffer,
2265 				 unsigned int size)
2266 {
2267 	unsigned int truesize = ixgbe_rx_frame_truesize(rx_ring, size);
2268 #if (PAGE_SIZE < 8192)
2269 	rx_buffer->page_offset ^= truesize;
2270 #else
2271 	rx_buffer->page_offset += truesize;
2272 #endif
2273 }
2274 
2275 /**
2276  * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2277  * @q_vector: structure containing interrupt and ring information
2278  * @rx_ring: rx descriptor ring to transact packets on
2279  * @budget: Total limit on number of packets to process
2280  *
2281  * This function provides a "bounce buffer" approach to Rx interrupt
2282  * processing.  The advantage to this is that on systems that have
2283  * expensive overhead for IOMMU access this provides a means of avoiding
2284  * it by maintaining the mapping of the page to the syste.
2285  *
2286  * Returns amount of work completed
2287  **/
2288 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2289 			       struct ixgbe_ring *rx_ring,
2290 			       const int budget)
2291 {
2292 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2293 	struct ixgbe_adapter *adapter = q_vector->adapter;
2294 #ifdef IXGBE_FCOE
2295 	int ddp_bytes;
2296 	unsigned int mss = 0;
2297 #endif /* IXGBE_FCOE */
2298 	u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2299 	unsigned int xdp_xmit = 0;
2300 	struct xdp_buff xdp;
2301 
2302 	xdp.rxq = &rx_ring->xdp_rxq;
2303 
2304 	/* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
2305 #if (PAGE_SIZE < 8192)
2306 	xdp.frame_sz = ixgbe_rx_frame_truesize(rx_ring, 0);
2307 #endif
2308 
2309 	while (likely(total_rx_packets < budget)) {
2310 		union ixgbe_adv_rx_desc *rx_desc;
2311 		struct ixgbe_rx_buffer *rx_buffer;
2312 		struct sk_buff *skb;
2313 		unsigned int size;
2314 
2315 		/* return some buffers to hardware, one at a time is too slow */
2316 		if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2317 			ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2318 			cleaned_count = 0;
2319 		}
2320 
2321 		rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2322 		size = le16_to_cpu(rx_desc->wb.upper.length);
2323 		if (!size)
2324 			break;
2325 
2326 		/* This memory barrier is needed to keep us from reading
2327 		 * any other fields out of the rx_desc until we know the
2328 		 * descriptor has been written back
2329 		 */
2330 		dma_rmb();
2331 
2332 		rx_buffer = ixgbe_get_rx_buffer(rx_ring, rx_desc, &skb, size);
2333 
2334 		/* retrieve a buffer from the ring */
2335 		if (!skb) {
2336 			xdp.data = page_address(rx_buffer->page) +
2337 				   rx_buffer->page_offset;
2338 			xdp.data_meta = xdp.data;
2339 			xdp.data_hard_start = xdp.data -
2340 					      ixgbe_rx_offset(rx_ring);
2341 			xdp.data_end = xdp.data + size;
2342 #if (PAGE_SIZE > 4096)
2343 			/* At larger PAGE_SIZE, frame_sz depend on len size */
2344 			xdp.frame_sz = ixgbe_rx_frame_truesize(rx_ring, size);
2345 #endif
2346 			skb = ixgbe_run_xdp(adapter, rx_ring, &xdp);
2347 		}
2348 
2349 		if (IS_ERR(skb)) {
2350 			unsigned int xdp_res = -PTR_ERR(skb);
2351 
2352 			if (xdp_res & (IXGBE_XDP_TX | IXGBE_XDP_REDIR)) {
2353 				xdp_xmit |= xdp_res;
2354 				ixgbe_rx_buffer_flip(rx_ring, rx_buffer, size);
2355 			} else {
2356 				rx_buffer->pagecnt_bias++;
2357 			}
2358 			total_rx_packets++;
2359 			total_rx_bytes += size;
2360 		} else if (skb) {
2361 			ixgbe_add_rx_frag(rx_ring, rx_buffer, skb, size);
2362 		} else if (ring_uses_build_skb(rx_ring)) {
2363 			skb = ixgbe_build_skb(rx_ring, rx_buffer,
2364 					      &xdp, rx_desc);
2365 		} else {
2366 			skb = ixgbe_construct_skb(rx_ring, rx_buffer,
2367 						  &xdp, rx_desc);
2368 		}
2369 
2370 		/* exit if we failed to retrieve a buffer */
2371 		if (!skb) {
2372 			rx_ring->rx_stats.alloc_rx_buff_failed++;
2373 			rx_buffer->pagecnt_bias++;
2374 			break;
2375 		}
2376 
2377 		ixgbe_put_rx_buffer(rx_ring, rx_buffer, skb);
2378 		cleaned_count++;
2379 
2380 		/* place incomplete frames back on ring for completion */
2381 		if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2382 			continue;
2383 
2384 		/* verify the packet layout is correct */
2385 		if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2386 			continue;
2387 
2388 		/* probably a little skewed due to removing CRC */
2389 		total_rx_bytes += skb->len;
2390 
2391 		/* populate checksum, timestamp, VLAN, and protocol */
2392 		ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2393 
2394 #ifdef IXGBE_FCOE
2395 		/* if ddp, not passing to ULD unless for FCP_RSP or error */
2396 		if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2397 			ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2398 			/* include DDPed FCoE data */
2399 			if (ddp_bytes > 0) {
2400 				if (!mss) {
2401 					mss = rx_ring->netdev->mtu -
2402 						sizeof(struct fcoe_hdr) -
2403 						sizeof(struct fc_frame_header) -
2404 						sizeof(struct fcoe_crc_eof);
2405 					if (mss > 512)
2406 						mss &= ~511;
2407 				}
2408 				total_rx_bytes += ddp_bytes;
2409 				total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2410 								 mss);
2411 			}
2412 			if (!ddp_bytes) {
2413 				dev_kfree_skb_any(skb);
2414 				continue;
2415 			}
2416 		}
2417 
2418 #endif /* IXGBE_FCOE */
2419 		ixgbe_rx_skb(q_vector, skb);
2420 
2421 		/* update budget accounting */
2422 		total_rx_packets++;
2423 	}
2424 
2425 	if (xdp_xmit & IXGBE_XDP_REDIR)
2426 		xdp_do_flush_map();
2427 
2428 	if (xdp_xmit & IXGBE_XDP_TX) {
2429 		struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
2430 
2431 		/* Force memory writes to complete before letting h/w
2432 		 * know there are new descriptors to fetch.
2433 		 */
2434 		wmb();
2435 		writel(ring->next_to_use, ring->tail);
2436 	}
2437 
2438 	u64_stats_update_begin(&rx_ring->syncp);
2439 	rx_ring->stats.packets += total_rx_packets;
2440 	rx_ring->stats.bytes += total_rx_bytes;
2441 	u64_stats_update_end(&rx_ring->syncp);
2442 	q_vector->rx.total_packets += total_rx_packets;
2443 	q_vector->rx.total_bytes += total_rx_bytes;
2444 
2445 	return total_rx_packets;
2446 }
2447 
2448 /**
2449  * ixgbe_configure_msix - Configure MSI-X hardware
2450  * @adapter: board private structure
2451  *
2452  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2453  * interrupts.
2454  **/
2455 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2456 {
2457 	struct ixgbe_q_vector *q_vector;
2458 	int v_idx;
2459 	u32 mask;
2460 
2461 	/* Populate MSIX to EITR Select */
2462 	if (adapter->num_vfs > 32) {
2463 		u32 eitrsel = BIT(adapter->num_vfs - 32) - 1;
2464 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2465 	}
2466 
2467 	/*
2468 	 * Populate the IVAR table and set the ITR values to the
2469 	 * corresponding register.
2470 	 */
2471 	for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2472 		struct ixgbe_ring *ring;
2473 		q_vector = adapter->q_vector[v_idx];
2474 
2475 		ixgbe_for_each_ring(ring, q_vector->rx)
2476 			ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2477 
2478 		ixgbe_for_each_ring(ring, q_vector->tx)
2479 			ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2480 
2481 		ixgbe_write_eitr(q_vector);
2482 	}
2483 
2484 	switch (adapter->hw.mac.type) {
2485 	case ixgbe_mac_82598EB:
2486 		ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2487 			       v_idx);
2488 		break;
2489 	case ixgbe_mac_82599EB:
2490 	case ixgbe_mac_X540:
2491 	case ixgbe_mac_X550:
2492 	case ixgbe_mac_X550EM_x:
2493 	case ixgbe_mac_x550em_a:
2494 		ixgbe_set_ivar(adapter, -1, 1, v_idx);
2495 		break;
2496 	default:
2497 		break;
2498 	}
2499 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2500 
2501 	/* set up to autoclear timer, and the vectors */
2502 	mask = IXGBE_EIMS_ENABLE_MASK;
2503 	mask &= ~(IXGBE_EIMS_OTHER |
2504 		  IXGBE_EIMS_MAILBOX |
2505 		  IXGBE_EIMS_LSC);
2506 
2507 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2508 }
2509 
2510 /**
2511  * ixgbe_update_itr - update the dynamic ITR value based on statistics
2512  * @q_vector: structure containing interrupt and ring information
2513  * @ring_container: structure containing ring performance data
2514  *
2515  *      Stores a new ITR value based on packets and byte
2516  *      counts during the last interrupt.  The advantage of per interrupt
2517  *      computation is faster updates and more accurate ITR for the current
2518  *      traffic pattern.  Constants in this function were computed
2519  *      based on theoretical maximum wire speed and thresholds were set based
2520  *      on testing data as well as attempting to minimize response time
2521  *      while increasing bulk throughput.
2522  **/
2523 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2524 			     struct ixgbe_ring_container *ring_container)
2525 {
2526 	unsigned int itr = IXGBE_ITR_ADAPTIVE_MIN_USECS |
2527 			   IXGBE_ITR_ADAPTIVE_LATENCY;
2528 	unsigned int avg_wire_size, packets, bytes;
2529 	unsigned long next_update = jiffies;
2530 
2531 	/* If we don't have any rings just leave ourselves set for maximum
2532 	 * possible latency so we take ourselves out of the equation.
2533 	 */
2534 	if (!ring_container->ring)
2535 		return;
2536 
2537 	/* If we didn't update within up to 1 - 2 jiffies we can assume
2538 	 * that either packets are coming in so slow there hasn't been
2539 	 * any work, or that there is so much work that NAPI is dealing
2540 	 * with interrupt moderation and we don't need to do anything.
2541 	 */
2542 	if (time_after(next_update, ring_container->next_update))
2543 		goto clear_counts;
2544 
2545 	packets = ring_container->total_packets;
2546 
2547 	/* We have no packets to actually measure against. This means
2548 	 * either one of the other queues on this vector is active or
2549 	 * we are a Tx queue doing TSO with too high of an interrupt rate.
2550 	 *
2551 	 * When this occurs just tick up our delay by the minimum value
2552 	 * and hope that this extra delay will prevent us from being called
2553 	 * without any work on our queue.
2554 	 */
2555 	if (!packets) {
2556 		itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2557 		if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2558 			itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2559 		itr += ring_container->itr & IXGBE_ITR_ADAPTIVE_LATENCY;
2560 		goto clear_counts;
2561 	}
2562 
2563 	bytes = ring_container->total_bytes;
2564 
2565 	/* If packets are less than 4 or bytes are less than 9000 assume
2566 	 * insufficient data to use bulk rate limiting approach. We are
2567 	 * likely latency driven.
2568 	 */
2569 	if (packets < 4 && bytes < 9000) {
2570 		itr = IXGBE_ITR_ADAPTIVE_LATENCY;
2571 		goto adjust_by_size;
2572 	}
2573 
2574 	/* Between 4 and 48 we can assume that our current interrupt delay
2575 	 * is only slightly too low. As such we should increase it by a small
2576 	 * fixed amount.
2577 	 */
2578 	if (packets < 48) {
2579 		itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2580 		if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2581 			itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2582 		goto clear_counts;
2583 	}
2584 
2585 	/* Between 48 and 96 is our "goldilocks" zone where we are working
2586 	 * out "just right". Just report that our current ITR is good for us.
2587 	 */
2588 	if (packets < 96) {
2589 		itr = q_vector->itr >> 2;
2590 		goto clear_counts;
2591 	}
2592 
2593 	/* If packet count is 96 or greater we are likely looking at a slight
2594 	 * overrun of the delay we want. Try halving our delay to see if that
2595 	 * will cut the number of packets in half per interrupt.
2596 	 */
2597 	if (packets < 256) {
2598 		itr = q_vector->itr >> 3;
2599 		if (itr < IXGBE_ITR_ADAPTIVE_MIN_USECS)
2600 			itr = IXGBE_ITR_ADAPTIVE_MIN_USECS;
2601 		goto clear_counts;
2602 	}
2603 
2604 	/* The paths below assume we are dealing with a bulk ITR since number
2605 	 * of packets is 256 or greater. We are just going to have to compute
2606 	 * a value and try to bring the count under control, though for smaller
2607 	 * packet sizes there isn't much we can do as NAPI polling will likely
2608 	 * be kicking in sooner rather than later.
2609 	 */
2610 	itr = IXGBE_ITR_ADAPTIVE_BULK;
2611 
2612 adjust_by_size:
2613 	/* If packet counts are 256 or greater we can assume we have a gross
2614 	 * overestimation of what the rate should be. Instead of trying to fine
2615 	 * tune it just use the formula below to try and dial in an exact value
2616 	 * give the current packet size of the frame.
2617 	 */
2618 	avg_wire_size = bytes / packets;
2619 
2620 	/* The following is a crude approximation of:
2621 	 *  wmem_default / (size + overhead) = desired_pkts_per_int
2622 	 *  rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
2623 	 *  (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
2624 	 *
2625 	 * Assuming wmem_default is 212992 and overhead is 640 bytes per
2626 	 * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
2627 	 * formula down to
2628 	 *
2629 	 *  (170 * (size + 24)) / (size + 640) = ITR
2630 	 *
2631 	 * We first do some math on the packet size and then finally bitshift
2632 	 * by 8 after rounding up. We also have to account for PCIe link speed
2633 	 * difference as ITR scales based on this.
2634 	 */
2635 	if (avg_wire_size <= 60) {
2636 		/* Start at 50k ints/sec */
2637 		avg_wire_size = 5120;
2638 	} else if (avg_wire_size <= 316) {
2639 		/* 50K ints/sec to 16K ints/sec */
2640 		avg_wire_size *= 40;
2641 		avg_wire_size += 2720;
2642 	} else if (avg_wire_size <= 1084) {
2643 		/* 16K ints/sec to 9.2K ints/sec */
2644 		avg_wire_size *= 15;
2645 		avg_wire_size += 11452;
2646 	} else if (avg_wire_size < 1968) {
2647 		/* 9.2K ints/sec to 8K ints/sec */
2648 		avg_wire_size *= 5;
2649 		avg_wire_size += 22420;
2650 	} else {
2651 		/* plateau at a limit of 8K ints/sec */
2652 		avg_wire_size = 32256;
2653 	}
2654 
2655 	/* If we are in low latency mode half our delay which doubles the rate
2656 	 * to somewhere between 100K to 16K ints/sec
2657 	 */
2658 	if (itr & IXGBE_ITR_ADAPTIVE_LATENCY)
2659 		avg_wire_size >>= 1;
2660 
2661 	/* Resultant value is 256 times larger than it needs to be. This
2662 	 * gives us room to adjust the value as needed to either increase
2663 	 * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
2664 	 *
2665 	 * Use addition as we have already recorded the new latency flag
2666 	 * for the ITR value.
2667 	 */
2668 	switch (q_vector->adapter->link_speed) {
2669 	case IXGBE_LINK_SPEED_10GB_FULL:
2670 	case IXGBE_LINK_SPEED_100_FULL:
2671 	default:
2672 		itr += DIV_ROUND_UP(avg_wire_size,
2673 				    IXGBE_ITR_ADAPTIVE_MIN_INC * 256) *
2674 		       IXGBE_ITR_ADAPTIVE_MIN_INC;
2675 		break;
2676 	case IXGBE_LINK_SPEED_2_5GB_FULL:
2677 	case IXGBE_LINK_SPEED_1GB_FULL:
2678 	case IXGBE_LINK_SPEED_10_FULL:
2679 		if (avg_wire_size > 8064)
2680 			avg_wire_size = 8064;
2681 		itr += DIV_ROUND_UP(avg_wire_size,
2682 				    IXGBE_ITR_ADAPTIVE_MIN_INC * 64) *
2683 		       IXGBE_ITR_ADAPTIVE_MIN_INC;
2684 		break;
2685 	}
2686 
2687 clear_counts:
2688 	/* write back value */
2689 	ring_container->itr = itr;
2690 
2691 	/* next update should occur within next jiffy */
2692 	ring_container->next_update = next_update + 1;
2693 
2694 	ring_container->total_bytes = 0;
2695 	ring_container->total_packets = 0;
2696 }
2697 
2698 /**
2699  * ixgbe_write_eitr - write EITR register in hardware specific way
2700  * @q_vector: structure containing interrupt and ring information
2701  *
2702  * This function is made to be called by ethtool and by the driver
2703  * when it needs to update EITR registers at runtime.  Hardware
2704  * specific quirks/differences are taken care of here.
2705  */
2706 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2707 {
2708 	struct ixgbe_adapter *adapter = q_vector->adapter;
2709 	struct ixgbe_hw *hw = &adapter->hw;
2710 	int v_idx = q_vector->v_idx;
2711 	u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2712 
2713 	switch (adapter->hw.mac.type) {
2714 	case ixgbe_mac_82598EB:
2715 		/* must write high and low 16 bits to reset counter */
2716 		itr_reg |= (itr_reg << 16);
2717 		break;
2718 	case ixgbe_mac_82599EB:
2719 	case ixgbe_mac_X540:
2720 	case ixgbe_mac_X550:
2721 	case ixgbe_mac_X550EM_x:
2722 	case ixgbe_mac_x550em_a:
2723 		/*
2724 		 * set the WDIS bit to not clear the timer bits and cause an
2725 		 * immediate assertion of the interrupt
2726 		 */
2727 		itr_reg |= IXGBE_EITR_CNT_WDIS;
2728 		break;
2729 	default:
2730 		break;
2731 	}
2732 	IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2733 }
2734 
2735 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2736 {
2737 	u32 new_itr;
2738 
2739 	ixgbe_update_itr(q_vector, &q_vector->tx);
2740 	ixgbe_update_itr(q_vector, &q_vector->rx);
2741 
2742 	/* use the smallest value of new ITR delay calculations */
2743 	new_itr = min(q_vector->rx.itr, q_vector->tx.itr);
2744 
2745 	/* Clear latency flag if set, shift into correct position */
2746 	new_itr &= ~IXGBE_ITR_ADAPTIVE_LATENCY;
2747 	new_itr <<= 2;
2748 
2749 	if (new_itr != q_vector->itr) {
2750 		/* save the algorithm value here */
2751 		q_vector->itr = new_itr;
2752 
2753 		ixgbe_write_eitr(q_vector);
2754 	}
2755 }
2756 
2757 /**
2758  * ixgbe_check_overtemp_subtask - check for over temperature
2759  * @adapter: pointer to adapter
2760  **/
2761 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2762 {
2763 	struct ixgbe_hw *hw = &adapter->hw;
2764 	u32 eicr = adapter->interrupt_event;
2765 	s32 rc;
2766 
2767 	if (test_bit(__IXGBE_DOWN, &adapter->state))
2768 		return;
2769 
2770 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2771 		return;
2772 
2773 	adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2774 
2775 	switch (hw->device_id) {
2776 	case IXGBE_DEV_ID_82599_T3_LOM:
2777 		/*
2778 		 * Since the warning interrupt is for both ports
2779 		 * we don't have to check if:
2780 		 *  - This interrupt wasn't for our port.
2781 		 *  - We may have missed the interrupt so always have to
2782 		 *    check if we  got a LSC
2783 		 */
2784 		if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2785 		    !(eicr & IXGBE_EICR_LSC))
2786 			return;
2787 
2788 		if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2789 			u32 speed;
2790 			bool link_up = false;
2791 
2792 			hw->mac.ops.check_link(hw, &speed, &link_up, false);
2793 
2794 			if (link_up)
2795 				return;
2796 		}
2797 
2798 		/* Check if this is not due to overtemp */
2799 		if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2800 			return;
2801 
2802 		break;
2803 	case IXGBE_DEV_ID_X550EM_A_1G_T:
2804 	case IXGBE_DEV_ID_X550EM_A_1G_T_L:
2805 		rc = hw->phy.ops.check_overtemp(hw);
2806 		if (rc != IXGBE_ERR_OVERTEMP)
2807 			return;
2808 		break;
2809 	default:
2810 		if (adapter->hw.mac.type >= ixgbe_mac_X540)
2811 			return;
2812 		if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2813 			return;
2814 		break;
2815 	}
2816 	e_crit(drv, "%s\n", ixgbe_overheat_msg);
2817 
2818 	adapter->interrupt_event = 0;
2819 }
2820 
2821 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2822 {
2823 	struct ixgbe_hw *hw = &adapter->hw;
2824 
2825 	if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2826 	    (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2827 		e_crit(probe, "Fan has stopped, replace the adapter\n");
2828 		/* write to clear the interrupt */
2829 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2830 	}
2831 }
2832 
2833 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2834 {
2835 	struct ixgbe_hw *hw = &adapter->hw;
2836 
2837 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2838 		return;
2839 
2840 	switch (adapter->hw.mac.type) {
2841 	case ixgbe_mac_82599EB:
2842 		/*
2843 		 * Need to check link state so complete overtemp check
2844 		 * on service task
2845 		 */
2846 		if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2847 		     (eicr & IXGBE_EICR_LSC)) &&
2848 		    (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2849 			adapter->interrupt_event = eicr;
2850 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2851 			ixgbe_service_event_schedule(adapter);
2852 			return;
2853 		}
2854 		return;
2855 	case ixgbe_mac_x550em_a:
2856 		if (eicr & IXGBE_EICR_GPI_SDP0_X550EM_a) {
2857 			adapter->interrupt_event = eicr;
2858 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2859 			ixgbe_service_event_schedule(adapter);
2860 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC,
2861 					IXGBE_EICR_GPI_SDP0_X550EM_a);
2862 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICR,
2863 					IXGBE_EICR_GPI_SDP0_X550EM_a);
2864 		}
2865 		return;
2866 	case ixgbe_mac_X550:
2867 	case ixgbe_mac_X540:
2868 		if (!(eicr & IXGBE_EICR_TS))
2869 			return;
2870 		break;
2871 	default:
2872 		return;
2873 	}
2874 
2875 	e_crit(drv, "%s\n", ixgbe_overheat_msg);
2876 }
2877 
2878 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2879 {
2880 	switch (hw->mac.type) {
2881 	case ixgbe_mac_82598EB:
2882 		if (hw->phy.type == ixgbe_phy_nl)
2883 			return true;
2884 		return false;
2885 	case ixgbe_mac_82599EB:
2886 	case ixgbe_mac_X550EM_x:
2887 	case ixgbe_mac_x550em_a:
2888 		switch (hw->mac.ops.get_media_type(hw)) {
2889 		case ixgbe_media_type_fiber:
2890 		case ixgbe_media_type_fiber_qsfp:
2891 			return true;
2892 		default:
2893 			return false;
2894 		}
2895 	default:
2896 		return false;
2897 	}
2898 }
2899 
2900 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2901 {
2902 	struct ixgbe_hw *hw = &adapter->hw;
2903 	u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw);
2904 
2905 	if (!ixgbe_is_sfp(hw))
2906 		return;
2907 
2908 	/* Later MAC's use different SDP */
2909 	if (hw->mac.type >= ixgbe_mac_X540)
2910 		eicr_mask = IXGBE_EICR_GPI_SDP0_X540;
2911 
2912 	if (eicr & eicr_mask) {
2913 		/* Clear the interrupt */
2914 		IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask);
2915 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2916 			adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2917 			adapter->sfp_poll_time = 0;
2918 			ixgbe_service_event_schedule(adapter);
2919 		}
2920 	}
2921 
2922 	if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
2923 	    (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2924 		/* Clear the interrupt */
2925 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2926 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2927 			adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2928 			ixgbe_service_event_schedule(adapter);
2929 		}
2930 	}
2931 }
2932 
2933 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2934 {
2935 	struct ixgbe_hw *hw = &adapter->hw;
2936 
2937 	adapter->lsc_int++;
2938 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2939 	adapter->link_check_timeout = jiffies;
2940 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2941 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2942 		IXGBE_WRITE_FLUSH(hw);
2943 		ixgbe_service_event_schedule(adapter);
2944 	}
2945 }
2946 
2947 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2948 					   u64 qmask)
2949 {
2950 	u32 mask;
2951 	struct ixgbe_hw *hw = &adapter->hw;
2952 
2953 	switch (hw->mac.type) {
2954 	case ixgbe_mac_82598EB:
2955 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2956 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2957 		break;
2958 	case ixgbe_mac_82599EB:
2959 	case ixgbe_mac_X540:
2960 	case ixgbe_mac_X550:
2961 	case ixgbe_mac_X550EM_x:
2962 	case ixgbe_mac_x550em_a:
2963 		mask = (qmask & 0xFFFFFFFF);
2964 		if (mask)
2965 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2966 		mask = (qmask >> 32);
2967 		if (mask)
2968 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2969 		break;
2970 	default:
2971 		break;
2972 	}
2973 	/* skip the flush */
2974 }
2975 
2976 /**
2977  * ixgbe_irq_enable - Enable default interrupt generation settings
2978  * @adapter: board private structure
2979  * @queues: enable irqs for queues
2980  * @flush: flush register write
2981  **/
2982 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2983 				    bool flush)
2984 {
2985 	struct ixgbe_hw *hw = &adapter->hw;
2986 	u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2987 
2988 	/* don't reenable LSC while waiting for link */
2989 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2990 		mask &= ~IXGBE_EIMS_LSC;
2991 
2992 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2993 		switch (adapter->hw.mac.type) {
2994 		case ixgbe_mac_82599EB:
2995 			mask |= IXGBE_EIMS_GPI_SDP0(hw);
2996 			break;
2997 		case ixgbe_mac_X540:
2998 		case ixgbe_mac_X550:
2999 		case ixgbe_mac_X550EM_x:
3000 		case ixgbe_mac_x550em_a:
3001 			mask |= IXGBE_EIMS_TS;
3002 			break;
3003 		default:
3004 			break;
3005 		}
3006 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3007 		mask |= IXGBE_EIMS_GPI_SDP1(hw);
3008 	switch (adapter->hw.mac.type) {
3009 	case ixgbe_mac_82599EB:
3010 		mask |= IXGBE_EIMS_GPI_SDP1(hw);
3011 		mask |= IXGBE_EIMS_GPI_SDP2(hw);
3012 		/* fall through */
3013 	case ixgbe_mac_X540:
3014 	case ixgbe_mac_X550:
3015 	case ixgbe_mac_X550EM_x:
3016 	case ixgbe_mac_x550em_a:
3017 		if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
3018 		    adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
3019 		    adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP_N)
3020 			mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw);
3021 		if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
3022 			mask |= IXGBE_EICR_GPI_SDP0_X540;
3023 		mask |= IXGBE_EIMS_ECC;
3024 		mask |= IXGBE_EIMS_MAILBOX;
3025 		break;
3026 	default:
3027 		break;
3028 	}
3029 
3030 	if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
3031 	    !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
3032 		mask |= IXGBE_EIMS_FLOW_DIR;
3033 
3034 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
3035 	if (queues)
3036 		ixgbe_irq_enable_queues(adapter, ~0);
3037 	if (flush)
3038 		IXGBE_WRITE_FLUSH(&adapter->hw);
3039 }
3040 
3041 static irqreturn_t ixgbe_msix_other(int irq, void *data)
3042 {
3043 	struct ixgbe_adapter *adapter = data;
3044 	struct ixgbe_hw *hw = &adapter->hw;
3045 	u32 eicr;
3046 
3047 	/*
3048 	 * Workaround for Silicon errata.  Use clear-by-write instead
3049 	 * of clear-by-read.  Reading with EICS will return the
3050 	 * interrupt causes without clearing, which later be done
3051 	 * with the write to EICR.
3052 	 */
3053 	eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
3054 
3055 	/* The lower 16bits of the EICR register are for the queue interrupts
3056 	 * which should be masked here in order to not accidentally clear them if
3057 	 * the bits are high when ixgbe_msix_other is called. There is a race
3058 	 * condition otherwise which results in possible performance loss
3059 	 * especially if the ixgbe_msix_other interrupt is triggering
3060 	 * consistently (as it would when PPS is turned on for the X540 device)
3061 	 */
3062 	eicr &= 0xFFFF0000;
3063 
3064 	IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
3065 
3066 	if (eicr & IXGBE_EICR_LSC)
3067 		ixgbe_check_lsc(adapter);
3068 
3069 	if (eicr & IXGBE_EICR_MAILBOX)
3070 		ixgbe_msg_task(adapter);
3071 
3072 	switch (hw->mac.type) {
3073 	case ixgbe_mac_82599EB:
3074 	case ixgbe_mac_X540:
3075 	case ixgbe_mac_X550:
3076 	case ixgbe_mac_X550EM_x:
3077 	case ixgbe_mac_x550em_a:
3078 		if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
3079 		    (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
3080 			adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
3081 			ixgbe_service_event_schedule(adapter);
3082 			IXGBE_WRITE_REG(hw, IXGBE_EICR,
3083 					IXGBE_EICR_GPI_SDP0_X540);
3084 		}
3085 		if (eicr & IXGBE_EICR_ECC) {
3086 			e_info(link, "Received ECC Err, initiating reset\n");
3087 			set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3088 			ixgbe_service_event_schedule(adapter);
3089 			IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3090 		}
3091 		/* Handle Flow Director Full threshold interrupt */
3092 		if (eicr & IXGBE_EICR_FLOW_DIR) {
3093 			int reinit_count = 0;
3094 			int i;
3095 			for (i = 0; i < adapter->num_tx_queues; i++) {
3096 				struct ixgbe_ring *ring = adapter->tx_ring[i];
3097 				if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
3098 						       &ring->state))
3099 					reinit_count++;
3100 			}
3101 			if (reinit_count) {
3102 				/* no more flow director interrupts until after init */
3103 				IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
3104 				adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
3105 				ixgbe_service_event_schedule(adapter);
3106 			}
3107 		}
3108 		ixgbe_check_sfp_event(adapter, eicr);
3109 		ixgbe_check_overtemp_event(adapter, eicr);
3110 		break;
3111 	default:
3112 		break;
3113 	}
3114 
3115 	ixgbe_check_fan_failure(adapter, eicr);
3116 
3117 	if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3118 		ixgbe_ptp_check_pps_event(adapter);
3119 
3120 	/* re-enable the original interrupt state, no lsc, no queues */
3121 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
3122 		ixgbe_irq_enable(adapter, false, false);
3123 
3124 	return IRQ_HANDLED;
3125 }
3126 
3127 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
3128 {
3129 	struct ixgbe_q_vector *q_vector = data;
3130 
3131 	/* EIAM disabled interrupts (on this vector) for us */
3132 
3133 	if (q_vector->rx.ring || q_vector->tx.ring)
3134 		napi_schedule_irqoff(&q_vector->napi);
3135 
3136 	return IRQ_HANDLED;
3137 }
3138 
3139 /**
3140  * ixgbe_poll - NAPI Rx polling callback
3141  * @napi: structure for representing this polling device
3142  * @budget: how many packets driver is allowed to clean
3143  *
3144  * This function is used for legacy and MSI, NAPI mode
3145  **/
3146 int ixgbe_poll(struct napi_struct *napi, int budget)
3147 {
3148 	struct ixgbe_q_vector *q_vector =
3149 				container_of(napi, struct ixgbe_q_vector, napi);
3150 	struct ixgbe_adapter *adapter = q_vector->adapter;
3151 	struct ixgbe_ring *ring;
3152 	int per_ring_budget, work_done = 0;
3153 	bool clean_complete = true;
3154 
3155 #ifdef CONFIG_IXGBE_DCA
3156 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3157 		ixgbe_update_dca(q_vector);
3158 #endif
3159 
3160 	ixgbe_for_each_ring(ring, q_vector->tx) {
3161 		bool wd = ring->xsk_umem ?
3162 			  ixgbe_clean_xdp_tx_irq(q_vector, ring, budget) :
3163 			  ixgbe_clean_tx_irq(q_vector, ring, budget);
3164 
3165 		if (!wd)
3166 			clean_complete = false;
3167 	}
3168 
3169 	/* Exit if we are called by netpoll */
3170 	if (budget <= 0)
3171 		return budget;
3172 
3173 	/* attempt to distribute budget to each queue fairly, but don't allow
3174 	 * the budget to go below 1 because we'll exit polling */
3175 	if (q_vector->rx.count > 1)
3176 		per_ring_budget = max(budget/q_vector->rx.count, 1);
3177 	else
3178 		per_ring_budget = budget;
3179 
3180 	ixgbe_for_each_ring(ring, q_vector->rx) {
3181 		int cleaned = ring->xsk_umem ?
3182 			      ixgbe_clean_rx_irq_zc(q_vector, ring,
3183 						    per_ring_budget) :
3184 			      ixgbe_clean_rx_irq(q_vector, ring,
3185 						 per_ring_budget);
3186 
3187 		work_done += cleaned;
3188 		if (cleaned >= per_ring_budget)
3189 			clean_complete = false;
3190 	}
3191 
3192 	/* If all work not completed, return budget and keep polling */
3193 	if (!clean_complete)
3194 		return budget;
3195 
3196 	/* all work done, exit the polling mode */
3197 	if (likely(napi_complete_done(napi, work_done))) {
3198 		if (adapter->rx_itr_setting & 1)
3199 			ixgbe_set_itr(q_vector);
3200 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
3201 			ixgbe_irq_enable_queues(adapter,
3202 						BIT_ULL(q_vector->v_idx));
3203 	}
3204 
3205 	return min(work_done, budget - 1);
3206 }
3207 
3208 /**
3209  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
3210  * @adapter: board private structure
3211  *
3212  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
3213  * interrupts from the kernel.
3214  **/
3215 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
3216 {
3217 	struct net_device *netdev = adapter->netdev;
3218 	unsigned int ri = 0, ti = 0;
3219 	int vector, err;
3220 
3221 	for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3222 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3223 		struct msix_entry *entry = &adapter->msix_entries[vector];
3224 
3225 		if (q_vector->tx.ring && q_vector->rx.ring) {
3226 			snprintf(q_vector->name, sizeof(q_vector->name),
3227 				 "%s-TxRx-%u", netdev->name, ri++);
3228 			ti++;
3229 		} else if (q_vector->rx.ring) {
3230 			snprintf(q_vector->name, sizeof(q_vector->name),
3231 				 "%s-rx-%u", netdev->name, ri++);
3232 		} else if (q_vector->tx.ring) {
3233 			snprintf(q_vector->name, sizeof(q_vector->name),
3234 				 "%s-tx-%u", netdev->name, ti++);
3235 		} else {
3236 			/* skip this unused q_vector */
3237 			continue;
3238 		}
3239 		err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
3240 				  q_vector->name, q_vector);
3241 		if (err) {
3242 			e_err(probe, "request_irq failed for MSIX interrupt "
3243 			      "Error: %d\n", err);
3244 			goto free_queue_irqs;
3245 		}
3246 		/* If Flow Director is enabled, set interrupt affinity */
3247 		if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3248 			/* assign the mask for this irq */
3249 			irq_set_affinity_hint(entry->vector,
3250 					      &q_vector->affinity_mask);
3251 		}
3252 	}
3253 
3254 	err = request_irq(adapter->msix_entries[vector].vector,
3255 			  ixgbe_msix_other, 0, netdev->name, adapter);
3256 	if (err) {
3257 		e_err(probe, "request_irq for msix_other failed: %d\n", err);
3258 		goto free_queue_irqs;
3259 	}
3260 
3261 	return 0;
3262 
3263 free_queue_irqs:
3264 	while (vector) {
3265 		vector--;
3266 		irq_set_affinity_hint(adapter->msix_entries[vector].vector,
3267 				      NULL);
3268 		free_irq(adapter->msix_entries[vector].vector,
3269 			 adapter->q_vector[vector]);
3270 	}
3271 	adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3272 	pci_disable_msix(adapter->pdev);
3273 	kfree(adapter->msix_entries);
3274 	adapter->msix_entries = NULL;
3275 	return err;
3276 }
3277 
3278 /**
3279  * ixgbe_intr - legacy mode Interrupt Handler
3280  * @irq: interrupt number
3281  * @data: pointer to a network interface device structure
3282  **/
3283 static irqreturn_t ixgbe_intr(int irq, void *data)
3284 {
3285 	struct ixgbe_adapter *adapter = data;
3286 	struct ixgbe_hw *hw = &adapter->hw;
3287 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3288 	u32 eicr;
3289 
3290 	/*
3291 	 * Workaround for silicon errata #26 on 82598.  Mask the interrupt
3292 	 * before the read of EICR.
3293 	 */
3294 	IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
3295 
3296 	/* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
3297 	 * therefore no explicit interrupt disable is necessary */
3298 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
3299 	if (!eicr) {
3300 		/*
3301 		 * shared interrupt alert!
3302 		 * make sure interrupts are enabled because the read will
3303 		 * have disabled interrupts due to EIAM
3304 		 * finish the workaround of silicon errata on 82598.  Unmask
3305 		 * the interrupt that we masked before the EICR read.
3306 		 */
3307 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
3308 			ixgbe_irq_enable(adapter, true, true);
3309 		return IRQ_NONE;	/* Not our interrupt */
3310 	}
3311 
3312 	if (eicr & IXGBE_EICR_LSC)
3313 		ixgbe_check_lsc(adapter);
3314 
3315 	switch (hw->mac.type) {
3316 	case ixgbe_mac_82599EB:
3317 		ixgbe_check_sfp_event(adapter, eicr);
3318 		/* Fall through */
3319 	case ixgbe_mac_X540:
3320 	case ixgbe_mac_X550:
3321 	case ixgbe_mac_X550EM_x:
3322 	case ixgbe_mac_x550em_a:
3323 		if (eicr & IXGBE_EICR_ECC) {
3324 			e_info(link, "Received ECC Err, initiating reset\n");
3325 			set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3326 			ixgbe_service_event_schedule(adapter);
3327 			IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3328 		}
3329 		ixgbe_check_overtemp_event(adapter, eicr);
3330 		break;
3331 	default:
3332 		break;
3333 	}
3334 
3335 	ixgbe_check_fan_failure(adapter, eicr);
3336 	if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3337 		ixgbe_ptp_check_pps_event(adapter);
3338 
3339 	/* would disable interrupts here but EIAM disabled it */
3340 	napi_schedule_irqoff(&q_vector->napi);
3341 
3342 	/*
3343 	 * re-enable link(maybe) and non-queue interrupts, no flush.
3344 	 * ixgbe_poll will re-enable the queue interrupts
3345 	 */
3346 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
3347 		ixgbe_irq_enable(adapter, false, false);
3348 
3349 	return IRQ_HANDLED;
3350 }
3351 
3352 /**
3353  * ixgbe_request_irq - initialize interrupts
3354  * @adapter: board private structure
3355  *
3356  * Attempts to configure interrupts using the best available
3357  * capabilities of the hardware and kernel.
3358  **/
3359 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
3360 {
3361 	struct net_device *netdev = adapter->netdev;
3362 	int err;
3363 
3364 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3365 		err = ixgbe_request_msix_irqs(adapter);
3366 	else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
3367 		err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
3368 				  netdev->name, adapter);
3369 	else
3370 		err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
3371 				  netdev->name, adapter);
3372 
3373 	if (err)
3374 		e_err(probe, "request_irq failed, Error %d\n", err);
3375 
3376 	return err;
3377 }
3378 
3379 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
3380 {
3381 	int vector;
3382 
3383 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
3384 		free_irq(adapter->pdev->irq, adapter);
3385 		return;
3386 	}
3387 
3388 	if (!adapter->msix_entries)
3389 		return;
3390 
3391 	for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3392 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3393 		struct msix_entry *entry = &adapter->msix_entries[vector];
3394 
3395 		/* free only the irqs that were actually requested */
3396 		if (!q_vector->rx.ring && !q_vector->tx.ring)
3397 			continue;
3398 
3399 		/* clear the affinity_mask in the IRQ descriptor */
3400 		irq_set_affinity_hint(entry->vector, NULL);
3401 
3402 		free_irq(entry->vector, q_vector);
3403 	}
3404 
3405 	free_irq(adapter->msix_entries[vector].vector, adapter);
3406 }
3407 
3408 /**
3409  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3410  * @adapter: board private structure
3411  **/
3412 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3413 {
3414 	switch (adapter->hw.mac.type) {
3415 	case ixgbe_mac_82598EB:
3416 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3417 		break;
3418 	case ixgbe_mac_82599EB:
3419 	case ixgbe_mac_X540:
3420 	case ixgbe_mac_X550:
3421 	case ixgbe_mac_X550EM_x:
3422 	case ixgbe_mac_x550em_a:
3423 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3424 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3425 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3426 		break;
3427 	default:
3428 		break;
3429 	}
3430 	IXGBE_WRITE_FLUSH(&adapter->hw);
3431 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3432 		int vector;
3433 
3434 		for (vector = 0; vector < adapter->num_q_vectors; vector++)
3435 			synchronize_irq(adapter->msix_entries[vector].vector);
3436 
3437 		synchronize_irq(adapter->msix_entries[vector++].vector);
3438 	} else {
3439 		synchronize_irq(adapter->pdev->irq);
3440 	}
3441 }
3442 
3443 /**
3444  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3445  * @adapter: board private structure
3446  *
3447  **/
3448 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3449 {
3450 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3451 
3452 	ixgbe_write_eitr(q_vector);
3453 
3454 	ixgbe_set_ivar(adapter, 0, 0, 0);
3455 	ixgbe_set_ivar(adapter, 1, 0, 0);
3456 
3457 	e_info(hw, "Legacy interrupt IVAR setup done\n");
3458 }
3459 
3460 /**
3461  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3462  * @adapter: board private structure
3463  * @ring: structure containing ring specific data
3464  *
3465  * Configure the Tx descriptor ring after a reset.
3466  **/
3467 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3468 			     struct ixgbe_ring *ring)
3469 {
3470 	struct ixgbe_hw *hw = &adapter->hw;
3471 	u64 tdba = ring->dma;
3472 	int wait_loop = 10;
3473 	u32 txdctl = IXGBE_TXDCTL_ENABLE;
3474 	u8 reg_idx = ring->reg_idx;
3475 
3476 	ring->xsk_umem = NULL;
3477 	if (ring_is_xdp(ring))
3478 		ring->xsk_umem = ixgbe_xsk_umem(adapter, ring);
3479 
3480 	/* disable queue to avoid issues while updating state */
3481 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3482 	IXGBE_WRITE_FLUSH(hw);
3483 
3484 	IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3485 			(tdba & DMA_BIT_MASK(32)));
3486 	IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3487 	IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3488 			ring->count * sizeof(union ixgbe_adv_tx_desc));
3489 	IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3490 	IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3491 	ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3492 
3493 	/*
3494 	 * set WTHRESH to encourage burst writeback, it should not be set
3495 	 * higher than 1 when:
3496 	 * - ITR is 0 as it could cause false TX hangs
3497 	 * - ITR is set to > 100k int/sec and BQL is enabled
3498 	 *
3499 	 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3500 	 * to or less than the number of on chip descriptors, which is
3501 	 * currently 40.
3502 	 */
3503 	if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3504 		txdctl |= 1u << 16;	/* WTHRESH = 1 */
3505 	else
3506 		txdctl |= 8u << 16;	/* WTHRESH = 8 */
3507 
3508 	/*
3509 	 * Setting PTHRESH to 32 both improves performance
3510 	 * and avoids a TX hang with DFP enabled
3511 	 */
3512 	txdctl |= (1u << 8) |	/* HTHRESH = 1 */
3513 		   32;		/* PTHRESH = 32 */
3514 
3515 	/* reinitialize flowdirector state */
3516 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3517 		ring->atr_sample_rate = adapter->atr_sample_rate;
3518 		ring->atr_count = 0;
3519 		set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3520 	} else {
3521 		ring->atr_sample_rate = 0;
3522 	}
3523 
3524 	/* initialize XPS */
3525 	if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3526 		struct ixgbe_q_vector *q_vector = ring->q_vector;
3527 
3528 		if (q_vector)
3529 			netif_set_xps_queue(ring->netdev,
3530 					    &q_vector->affinity_mask,
3531 					    ring->queue_index);
3532 	}
3533 
3534 	clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3535 
3536 	/* reinitialize tx_buffer_info */
3537 	memset(ring->tx_buffer_info, 0,
3538 	       sizeof(struct ixgbe_tx_buffer) * ring->count);
3539 
3540 	/* enable queue */
3541 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3542 
3543 	/* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3544 	if (hw->mac.type == ixgbe_mac_82598EB &&
3545 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3546 		return;
3547 
3548 	/* poll to verify queue is enabled */
3549 	do {
3550 		usleep_range(1000, 2000);
3551 		txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3552 	} while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3553 	if (!wait_loop)
3554 		hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
3555 }
3556 
3557 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3558 {
3559 	struct ixgbe_hw *hw = &adapter->hw;
3560 	u32 rttdcs, mtqc;
3561 	u8 tcs = adapter->hw_tcs;
3562 
3563 	if (hw->mac.type == ixgbe_mac_82598EB)
3564 		return;
3565 
3566 	/* disable the arbiter while setting MTQC */
3567 	rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3568 	rttdcs |= IXGBE_RTTDCS_ARBDIS;
3569 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3570 
3571 	/* set transmit pool layout */
3572 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3573 		mtqc = IXGBE_MTQC_VT_ENA;
3574 		if (tcs > 4)
3575 			mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3576 		else if (tcs > 1)
3577 			mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3578 		else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3579 			 IXGBE_82599_VMDQ_4Q_MASK)
3580 			mtqc |= IXGBE_MTQC_32VF;
3581 		else
3582 			mtqc |= IXGBE_MTQC_64VF;
3583 	} else {
3584 		if (tcs > 4) {
3585 			mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3586 		} else if (tcs > 1) {
3587 			mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3588 		} else {
3589 			u8 max_txq = adapter->num_tx_queues +
3590 				adapter->num_xdp_queues;
3591 			if (max_txq > 63)
3592 				mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3593 			else
3594 				mtqc = IXGBE_MTQC_64Q_1PB;
3595 		}
3596 	}
3597 
3598 	IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3599 
3600 	/* Enable Security TX Buffer IFG for multiple pb */
3601 	if (tcs) {
3602 		u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3603 		sectx |= IXGBE_SECTX_DCB;
3604 		IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3605 	}
3606 
3607 	/* re-enable the arbiter */
3608 	rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3609 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3610 }
3611 
3612 /**
3613  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3614  * @adapter: board private structure
3615  *
3616  * Configure the Tx unit of the MAC after a reset.
3617  **/
3618 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3619 {
3620 	struct ixgbe_hw *hw = &adapter->hw;
3621 	u32 dmatxctl;
3622 	u32 i;
3623 
3624 	ixgbe_setup_mtqc(adapter);
3625 
3626 	if (hw->mac.type != ixgbe_mac_82598EB) {
3627 		/* DMATXCTL.EN must be before Tx queues are enabled */
3628 		dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3629 		dmatxctl |= IXGBE_DMATXCTL_TE;
3630 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3631 	}
3632 
3633 	/* Setup the HW Tx Head and Tail descriptor pointers */
3634 	for (i = 0; i < adapter->num_tx_queues; i++)
3635 		ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3636 	for (i = 0; i < adapter->num_xdp_queues; i++)
3637 		ixgbe_configure_tx_ring(adapter, adapter->xdp_ring[i]);
3638 }
3639 
3640 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3641 				 struct ixgbe_ring *ring)
3642 {
3643 	struct ixgbe_hw *hw = &adapter->hw;
3644 	u8 reg_idx = ring->reg_idx;
3645 	u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3646 
3647 	srrctl |= IXGBE_SRRCTL_DROP_EN;
3648 
3649 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3650 }
3651 
3652 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3653 				  struct ixgbe_ring *ring)
3654 {
3655 	struct ixgbe_hw *hw = &adapter->hw;
3656 	u8 reg_idx = ring->reg_idx;
3657 	u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3658 
3659 	srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3660 
3661 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3662 }
3663 
3664 #ifdef CONFIG_IXGBE_DCB
3665 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3666 #else
3667 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3668 #endif
3669 {
3670 	int i;
3671 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3672 
3673 	if (adapter->ixgbe_ieee_pfc)
3674 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3675 
3676 	/*
3677 	 * We should set the drop enable bit if:
3678 	 *  SR-IOV is enabled
3679 	 *   or
3680 	 *  Number of Rx queues > 1 and flow control is disabled
3681 	 *
3682 	 *  This allows us to avoid head of line blocking for security
3683 	 *  and performance reasons.
3684 	 */
3685 	if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3686 	    !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3687 		for (i = 0; i < adapter->num_rx_queues; i++)
3688 			ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3689 	} else {
3690 		for (i = 0; i < adapter->num_rx_queues; i++)
3691 			ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3692 	}
3693 }
3694 
3695 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3696 
3697 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3698 				   struct ixgbe_ring *rx_ring)
3699 {
3700 	struct ixgbe_hw *hw = &adapter->hw;
3701 	u32 srrctl;
3702 	u8 reg_idx = rx_ring->reg_idx;
3703 
3704 	if (hw->mac.type == ixgbe_mac_82598EB) {
3705 		u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3706 
3707 		/*
3708 		 * if VMDq is not active we must program one srrctl register
3709 		 * per RSS queue since we have enabled RDRXCTL.MVMEN
3710 		 */
3711 		reg_idx &= mask;
3712 	}
3713 
3714 	/* configure header buffer length, needed for RSC */
3715 	srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3716 
3717 	/* configure the packet buffer length */
3718 	if (rx_ring->xsk_umem) {
3719 		u32 xsk_buf_len = xsk_umem_get_rx_frame_size(rx_ring->xsk_umem);
3720 
3721 		/* If the MAC support setting RXDCTL.RLPML, the
3722 		 * SRRCTL[n].BSIZEPKT is set to PAGE_SIZE and
3723 		 * RXDCTL.RLPML is set to the actual UMEM buffer
3724 		 * size. If not, then we are stuck with a 1k buffer
3725 		 * size resolution. In this case frames larger than
3726 		 * the UMEM buffer size viewed in a 1k resolution will
3727 		 * be dropped.
3728 		 */
3729 		if (hw->mac.type != ixgbe_mac_82599EB)
3730 			srrctl |= PAGE_SIZE >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3731 		else
3732 			srrctl |= xsk_buf_len >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3733 	} else if (test_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state)) {
3734 		srrctl |= IXGBE_RXBUFFER_3K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3735 	} else {
3736 		srrctl |= IXGBE_RXBUFFER_2K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3737 	}
3738 
3739 	/* configure descriptor type */
3740 	srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3741 
3742 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3743 }
3744 
3745 /**
3746  * ixgbe_rss_indir_tbl_entries - Return RSS indirection table entries
3747  * @adapter: device handle
3748  *
3749  *  - 82598/82599/X540:     128
3750  *  - X550(non-SRIOV mode): 512
3751  *  - X550(SRIOV mode):     64
3752  */
3753 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3754 {
3755 	if (adapter->hw.mac.type < ixgbe_mac_X550)
3756 		return 128;
3757 	else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3758 		return 64;
3759 	else
3760 		return 512;
3761 }
3762 
3763 /**
3764  * ixgbe_store_key - Write the RSS key to HW
3765  * @adapter: device handle
3766  *
3767  * Write the RSS key stored in adapter.rss_key to HW.
3768  */
3769 void ixgbe_store_key(struct ixgbe_adapter *adapter)
3770 {
3771 	struct ixgbe_hw *hw = &adapter->hw;
3772 	int i;
3773 
3774 	for (i = 0; i < 10; i++)
3775 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3776 }
3777 
3778 /**
3779  * ixgbe_init_rss_key - Initialize adapter RSS key
3780  * @adapter: device handle
3781  *
3782  * Allocates and initializes the RSS key if it is not allocated.
3783  **/
3784 static inline int ixgbe_init_rss_key(struct ixgbe_adapter *adapter)
3785 {
3786 	u32 *rss_key;
3787 
3788 	if (!adapter->rss_key) {
3789 		rss_key = kzalloc(IXGBE_RSS_KEY_SIZE, GFP_KERNEL);
3790 		if (unlikely(!rss_key))
3791 			return -ENOMEM;
3792 
3793 		netdev_rss_key_fill(rss_key, IXGBE_RSS_KEY_SIZE);
3794 		adapter->rss_key = rss_key;
3795 	}
3796 
3797 	return 0;
3798 }
3799 
3800 /**
3801  * ixgbe_store_reta - Write the RETA table to HW
3802  * @adapter: device handle
3803  *
3804  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3805  */
3806 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3807 {
3808 	u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3809 	struct ixgbe_hw *hw = &adapter->hw;
3810 	u32 reta = 0;
3811 	u32 indices_multi;
3812 	u8 *indir_tbl = adapter->rss_indir_tbl;
3813 
3814 	/* Fill out the redirection table as follows:
3815 	 *  - 82598:      8 bit wide entries containing pair of 4 bit RSS
3816 	 *    indices.
3817 	 *  - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3818 	 *  - X550:       8 bit wide entries containing 6 bit RSS index
3819 	 */
3820 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3821 		indices_multi = 0x11;
3822 	else
3823 		indices_multi = 0x1;
3824 
3825 	/* Write redirection table to HW */
3826 	for (i = 0; i < reta_entries; i++) {
3827 		reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3828 		if ((i & 3) == 3) {
3829 			if (i < 128)
3830 				IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3831 			else
3832 				IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3833 						reta);
3834 			reta = 0;
3835 		}
3836 	}
3837 }
3838 
3839 /**
3840  * ixgbe_store_vfreta - Write the RETA table to HW (x550 devices in SRIOV mode)
3841  * @adapter: device handle
3842  *
3843  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3844  */
3845 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3846 {
3847 	u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3848 	struct ixgbe_hw *hw = &adapter->hw;
3849 	u32 vfreta = 0;
3850 
3851 	/* Write redirection table to HW */
3852 	for (i = 0; i < reta_entries; i++) {
3853 		u16 pool = adapter->num_rx_pools;
3854 
3855 		vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3856 		if ((i & 3) != 3)
3857 			continue;
3858 
3859 		while (pool--)
3860 			IXGBE_WRITE_REG(hw,
3861 					IXGBE_PFVFRETA(i >> 2, VMDQ_P(pool)),
3862 					vfreta);
3863 		vfreta = 0;
3864 	}
3865 }
3866 
3867 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3868 {
3869 	u32 i, j;
3870 	u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3871 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3872 
3873 	/* Program table for at least 4 queues w/ SR-IOV so that VFs can
3874 	 * make full use of any rings they may have.  We will use the
3875 	 * PSRTYPE register to control how many rings we use within the PF.
3876 	 */
3877 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 4))
3878 		rss_i = 4;
3879 
3880 	/* Fill out hash function seeds */
3881 	ixgbe_store_key(adapter);
3882 
3883 	/* Fill out redirection table */
3884 	memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3885 
3886 	for (i = 0, j = 0; i < reta_entries; i++, j++) {
3887 		if (j == rss_i)
3888 			j = 0;
3889 
3890 		adapter->rss_indir_tbl[i] = j;
3891 	}
3892 
3893 	ixgbe_store_reta(adapter);
3894 }
3895 
3896 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3897 {
3898 	struct ixgbe_hw *hw = &adapter->hw;
3899 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3900 	int i, j;
3901 
3902 	/* Fill out hash function seeds */
3903 	for (i = 0; i < 10; i++) {
3904 		u16 pool = adapter->num_rx_pools;
3905 
3906 		while (pool--)
3907 			IXGBE_WRITE_REG(hw,
3908 					IXGBE_PFVFRSSRK(i, VMDQ_P(pool)),
3909 					*(adapter->rss_key + i));
3910 	}
3911 
3912 	/* Fill out the redirection table */
3913 	for (i = 0, j = 0; i < 64; i++, j++) {
3914 		if (j == rss_i)
3915 			j = 0;
3916 
3917 		adapter->rss_indir_tbl[i] = j;
3918 	}
3919 
3920 	ixgbe_store_vfreta(adapter);
3921 }
3922 
3923 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3924 {
3925 	struct ixgbe_hw *hw = &adapter->hw;
3926 	u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3927 	u32 rxcsum;
3928 
3929 	/* Disable indicating checksum in descriptor, enables RSS hash */
3930 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3931 	rxcsum |= IXGBE_RXCSUM_PCSD;
3932 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3933 
3934 	if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3935 		if (adapter->ring_feature[RING_F_RSS].mask)
3936 			mrqc = IXGBE_MRQC_RSSEN;
3937 	} else {
3938 		u8 tcs = adapter->hw_tcs;
3939 
3940 		if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3941 			if (tcs > 4)
3942 				mrqc = IXGBE_MRQC_VMDQRT8TCEN;	/* 8 TCs */
3943 			else if (tcs > 1)
3944 				mrqc = IXGBE_MRQC_VMDQRT4TCEN;	/* 4 TCs */
3945 			else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3946 				 IXGBE_82599_VMDQ_4Q_MASK)
3947 				mrqc = IXGBE_MRQC_VMDQRSS32EN;
3948 			else
3949 				mrqc = IXGBE_MRQC_VMDQRSS64EN;
3950 
3951 			/* Enable L3/L4 for Tx Switched packets only for X550,
3952 			 * older devices do not support this feature
3953 			 */
3954 			if (hw->mac.type >= ixgbe_mac_X550)
3955 				mrqc |= IXGBE_MRQC_L3L4TXSWEN;
3956 		} else {
3957 			if (tcs > 4)
3958 				mrqc = IXGBE_MRQC_RTRSS8TCEN;
3959 			else if (tcs > 1)
3960 				mrqc = IXGBE_MRQC_RTRSS4TCEN;
3961 			else
3962 				mrqc = IXGBE_MRQC_RSSEN;
3963 		}
3964 	}
3965 
3966 	/* Perform hash on these packet types */
3967 	rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3968 		     IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3969 		     IXGBE_MRQC_RSS_FIELD_IPV6 |
3970 		     IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3971 
3972 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3973 		rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3974 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3975 		rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3976 
3977 	if ((hw->mac.type >= ixgbe_mac_X550) &&
3978 	    (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
3979 		u16 pool = adapter->num_rx_pools;
3980 
3981 		/* Enable VF RSS mode */
3982 		mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
3983 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3984 
3985 		/* Setup RSS through the VF registers */
3986 		ixgbe_setup_vfreta(adapter);
3987 		vfmrqc = IXGBE_MRQC_RSSEN;
3988 		vfmrqc |= rss_field;
3989 
3990 		while (pool--)
3991 			IXGBE_WRITE_REG(hw,
3992 					IXGBE_PFVFMRQC(VMDQ_P(pool)),
3993 					vfmrqc);
3994 	} else {
3995 		ixgbe_setup_reta(adapter);
3996 		mrqc |= rss_field;
3997 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3998 	}
3999 }
4000 
4001 /**
4002  * ixgbe_configure_rscctl - enable RSC for the indicated ring
4003  * @adapter: address of board private structure
4004  * @ring: structure containing ring specific data
4005  **/
4006 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
4007 				   struct ixgbe_ring *ring)
4008 {
4009 	struct ixgbe_hw *hw = &adapter->hw;
4010 	u32 rscctrl;
4011 	u8 reg_idx = ring->reg_idx;
4012 
4013 	if (!ring_is_rsc_enabled(ring))
4014 		return;
4015 
4016 	rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
4017 	rscctrl |= IXGBE_RSCCTL_RSCEN;
4018 	/*
4019 	 * we must limit the number of descriptors so that the
4020 	 * total size of max desc * buf_len is not greater
4021 	 * than 65536
4022 	 */
4023 	rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
4024 	IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
4025 }
4026 
4027 #define IXGBE_MAX_RX_DESC_POLL 10
4028 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
4029 				       struct ixgbe_ring *ring)
4030 {
4031 	struct ixgbe_hw *hw = &adapter->hw;
4032 	int wait_loop = IXGBE_MAX_RX_DESC_POLL;
4033 	u32 rxdctl;
4034 	u8 reg_idx = ring->reg_idx;
4035 
4036 	if (ixgbe_removed(hw->hw_addr))
4037 		return;
4038 	/* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
4039 	if (hw->mac.type == ixgbe_mac_82598EB &&
4040 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
4041 		return;
4042 
4043 	do {
4044 		usleep_range(1000, 2000);
4045 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4046 	} while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4047 
4048 	if (!wait_loop) {
4049 		e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
4050 		      "the polling period\n", reg_idx);
4051 	}
4052 }
4053 
4054 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
4055 			     struct ixgbe_ring *ring)
4056 {
4057 	struct ixgbe_hw *hw = &adapter->hw;
4058 	union ixgbe_adv_rx_desc *rx_desc;
4059 	u64 rdba = ring->dma;
4060 	u32 rxdctl;
4061 	u8 reg_idx = ring->reg_idx;
4062 
4063 	xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
4064 	ring->xsk_umem = ixgbe_xsk_umem(adapter, ring);
4065 	if (ring->xsk_umem) {
4066 		WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4067 						   MEM_TYPE_XSK_BUFF_POOL,
4068 						   NULL));
4069 		xsk_buff_set_rxq_info(ring->xsk_umem, &ring->xdp_rxq);
4070 	} else {
4071 		WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4072 						   MEM_TYPE_PAGE_SHARED, NULL));
4073 	}
4074 
4075 	/* disable queue to avoid use of these values while updating state */
4076 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4077 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4078 
4079 	/* write value back with RXDCTL.ENABLE bit cleared */
4080 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4081 	IXGBE_WRITE_FLUSH(hw);
4082 
4083 	IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
4084 	IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
4085 	IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
4086 			ring->count * sizeof(union ixgbe_adv_rx_desc));
4087 	/* Force flushing of IXGBE_RDLEN to prevent MDD */
4088 	IXGBE_WRITE_FLUSH(hw);
4089 
4090 	IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
4091 	IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
4092 	ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
4093 
4094 	ixgbe_configure_srrctl(adapter, ring);
4095 	ixgbe_configure_rscctl(adapter, ring);
4096 
4097 	if (hw->mac.type == ixgbe_mac_82598EB) {
4098 		/*
4099 		 * enable cache line friendly hardware writes:
4100 		 * PTHRESH=32 descriptors (half the internal cache),
4101 		 * this also removes ugly rx_no_buffer_count increment
4102 		 * HTHRESH=4 descriptors (to minimize latency on fetch)
4103 		 * WTHRESH=8 burst writeback up to two cache lines
4104 		 */
4105 		rxdctl &= ~0x3FFFFF;
4106 		rxdctl |=  0x080420;
4107 #if (PAGE_SIZE < 8192)
4108 	/* RXDCTL.RLPML does not work on 82599 */
4109 	} else if (hw->mac.type != ixgbe_mac_82599EB) {
4110 		rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4111 			    IXGBE_RXDCTL_RLPML_EN);
4112 
4113 		/* Limit the maximum frame size so we don't overrun the skb.
4114 		 * This can happen in SRIOV mode when the MTU of the VF is
4115 		 * higher than the MTU of the PF.
4116 		 */
4117 		if (ring_uses_build_skb(ring) &&
4118 		    !test_bit(__IXGBE_RX_3K_BUFFER, &ring->state))
4119 			rxdctl |= IXGBE_MAX_2K_FRAME_BUILD_SKB |
4120 				  IXGBE_RXDCTL_RLPML_EN;
4121 #endif
4122 	}
4123 
4124 	if (ring->xsk_umem && hw->mac.type != ixgbe_mac_82599EB) {
4125 		u32 xsk_buf_len = xsk_umem_get_rx_frame_size(ring->xsk_umem);
4126 
4127 		rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4128 			    IXGBE_RXDCTL_RLPML_EN);
4129 		rxdctl |= xsk_buf_len | IXGBE_RXDCTL_RLPML_EN;
4130 
4131 		ring->rx_buf_len = xsk_buf_len;
4132 	}
4133 
4134 	/* initialize rx_buffer_info */
4135 	memset(ring->rx_buffer_info, 0,
4136 	       sizeof(struct ixgbe_rx_buffer) * ring->count);
4137 
4138 	/* initialize Rx descriptor 0 */
4139 	rx_desc = IXGBE_RX_DESC(ring, 0);
4140 	rx_desc->wb.upper.length = 0;
4141 
4142 	/* enable receive descriptor ring */
4143 	rxdctl |= IXGBE_RXDCTL_ENABLE;
4144 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4145 
4146 	ixgbe_rx_desc_queue_enable(adapter, ring);
4147 	if (ring->xsk_umem)
4148 		ixgbe_alloc_rx_buffers_zc(ring, ixgbe_desc_unused(ring));
4149 	else
4150 		ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
4151 }
4152 
4153 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
4154 {
4155 	struct ixgbe_hw *hw = &adapter->hw;
4156 	int rss_i = adapter->ring_feature[RING_F_RSS].indices;
4157 	u16 pool = adapter->num_rx_pools;
4158 
4159 	/* PSRTYPE must be initialized in non 82598 adapters */
4160 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4161 		      IXGBE_PSRTYPE_UDPHDR |
4162 		      IXGBE_PSRTYPE_IPV4HDR |
4163 		      IXGBE_PSRTYPE_L2HDR |
4164 		      IXGBE_PSRTYPE_IPV6HDR;
4165 
4166 	if (hw->mac.type == ixgbe_mac_82598EB)
4167 		return;
4168 
4169 	if (rss_i > 3)
4170 		psrtype |= 2u << 29;
4171 	else if (rss_i > 1)
4172 		psrtype |= 1u << 29;
4173 
4174 	while (pool--)
4175 		IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4176 }
4177 
4178 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
4179 {
4180 	struct ixgbe_hw *hw = &adapter->hw;
4181 	u16 pool = adapter->num_rx_pools;
4182 	u32 reg_offset, vf_shift, vmolr;
4183 	u32 gcr_ext, vmdctl;
4184 	int i;
4185 
4186 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
4187 		return;
4188 
4189 	vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
4190 	vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
4191 	vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
4192 	vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
4193 	vmdctl |= IXGBE_VT_CTL_REPLEN;
4194 	IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
4195 
4196 	/* accept untagged packets until a vlan tag is
4197 	 * specifically set for the VMDQ queue/pool
4198 	 */
4199 	vmolr = IXGBE_VMOLR_AUPE;
4200 	while (pool--)
4201 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(pool)), vmolr);
4202 
4203 	vf_shift = VMDQ_P(0) % 32;
4204 	reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
4205 
4206 	/* Enable only the PF's pool for Tx/Rx */
4207 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), GENMASK(31, vf_shift));
4208 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
4209 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), GENMASK(31, vf_shift));
4210 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
4211 	if (adapter->bridge_mode == BRIDGE_MODE_VEB)
4212 		IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
4213 
4214 	/* Map PF MAC address in RAR Entry 0 to first pool following VFs */
4215 	hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
4216 
4217 	/* clear VLAN promisc flag so VFTA will be updated if necessary */
4218 	adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4219 
4220 	/*
4221 	 * Set up VF register offsets for selected VT Mode,
4222 	 * i.e. 32 or 64 VFs for SR-IOV
4223 	 */
4224 	switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4225 	case IXGBE_82599_VMDQ_8Q_MASK:
4226 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
4227 		break;
4228 	case IXGBE_82599_VMDQ_4Q_MASK:
4229 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
4230 		break;
4231 	default:
4232 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
4233 		break;
4234 	}
4235 
4236 	IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
4237 
4238 	for (i = 0; i < adapter->num_vfs; i++) {
4239 		/* configure spoof checking */
4240 		ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i,
4241 					  adapter->vfinfo[i].spoofchk_enabled);
4242 
4243 		/* Enable/Disable RSS query feature  */
4244 		ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
4245 					  adapter->vfinfo[i].rss_query_enabled);
4246 	}
4247 }
4248 
4249 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
4250 {
4251 	struct ixgbe_hw *hw = &adapter->hw;
4252 	struct net_device *netdev = adapter->netdev;
4253 	int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4254 	struct ixgbe_ring *rx_ring;
4255 	int i;
4256 	u32 mhadd, hlreg0;
4257 
4258 #ifdef IXGBE_FCOE
4259 	/* adjust max frame to be able to do baby jumbo for FCoE */
4260 	if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
4261 	    (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
4262 		max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4263 
4264 #endif /* IXGBE_FCOE */
4265 
4266 	/* adjust max frame to be at least the size of a standard frame */
4267 	if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
4268 		max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
4269 
4270 	mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
4271 	if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
4272 		mhadd &= ~IXGBE_MHADD_MFS_MASK;
4273 		mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
4274 
4275 		IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
4276 	}
4277 
4278 	hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4279 	/* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
4280 	hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4281 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4282 
4283 	/*
4284 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
4285 	 * the Base and Length of the Rx Descriptor Ring
4286 	 */
4287 	for (i = 0; i < adapter->num_rx_queues; i++) {
4288 		rx_ring = adapter->rx_ring[i];
4289 
4290 		clear_ring_rsc_enabled(rx_ring);
4291 		clear_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4292 		clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4293 
4294 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4295 			set_ring_rsc_enabled(rx_ring);
4296 
4297 		if (test_bit(__IXGBE_RX_FCOE, &rx_ring->state))
4298 			set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4299 
4300 		if (adapter->flags2 & IXGBE_FLAG2_RX_LEGACY)
4301 			continue;
4302 
4303 		set_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4304 
4305 #if (PAGE_SIZE < 8192)
4306 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4307 			set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4308 
4309 		if (IXGBE_2K_TOO_SMALL_WITH_PADDING ||
4310 		    (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
4311 			set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4312 #endif
4313 	}
4314 }
4315 
4316 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
4317 {
4318 	struct ixgbe_hw *hw = &adapter->hw;
4319 	u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4320 
4321 	switch (hw->mac.type) {
4322 	case ixgbe_mac_82598EB:
4323 		/*
4324 		 * For VMDq support of different descriptor types or
4325 		 * buffer sizes through the use of multiple SRRCTL
4326 		 * registers, RDRXCTL.MVMEN must be set to 1
4327 		 *
4328 		 * also, the manual doesn't mention it clearly but DCA hints
4329 		 * will only use queue 0's tags unless this bit is set.  Side
4330 		 * effects of setting this bit are only that SRRCTL must be
4331 		 * fully programmed [0..15]
4332 		 */
4333 		rdrxctl |= IXGBE_RDRXCTL_MVMEN;
4334 		break;
4335 	case ixgbe_mac_X550:
4336 	case ixgbe_mac_X550EM_x:
4337 	case ixgbe_mac_x550em_a:
4338 		if (adapter->num_vfs)
4339 			rdrxctl |= IXGBE_RDRXCTL_PSP;
4340 		/* fall through */
4341 	case ixgbe_mac_82599EB:
4342 	case ixgbe_mac_X540:
4343 		/* Disable RSC for ACK packets */
4344 		IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
4345 		   (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
4346 		rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4347 		/* hardware requires some bits to be set by default */
4348 		rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
4349 		rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4350 		break;
4351 	default:
4352 		/* We should do nothing since we don't know this hardware */
4353 		return;
4354 	}
4355 
4356 	IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4357 }
4358 
4359 /**
4360  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
4361  * @adapter: board private structure
4362  *
4363  * Configure the Rx unit of the MAC after a reset.
4364  **/
4365 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
4366 {
4367 	struct ixgbe_hw *hw = &adapter->hw;
4368 	int i;
4369 	u32 rxctrl, rfctl;
4370 
4371 	/* disable receives while setting up the descriptors */
4372 	hw->mac.ops.disable_rx(hw);
4373 
4374 	ixgbe_setup_psrtype(adapter);
4375 	ixgbe_setup_rdrxctl(adapter);
4376 
4377 	/* RSC Setup */
4378 	rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4379 	rfctl &= ~IXGBE_RFCTL_RSC_DIS;
4380 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
4381 		rfctl |= IXGBE_RFCTL_RSC_DIS;
4382 
4383 	/* disable NFS filtering */
4384 	rfctl |= (IXGBE_RFCTL_NFSW_DIS | IXGBE_RFCTL_NFSR_DIS);
4385 	IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4386 
4387 	/* Program registers for the distribution of queues */
4388 	ixgbe_setup_mrqc(adapter);
4389 
4390 	/* set_rx_buffer_len must be called before ring initialization */
4391 	ixgbe_set_rx_buffer_len(adapter);
4392 
4393 	/*
4394 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
4395 	 * the Base and Length of the Rx Descriptor Ring
4396 	 */
4397 	for (i = 0; i < adapter->num_rx_queues; i++)
4398 		ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
4399 
4400 	rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4401 	/* disable drop enable for 82598 parts */
4402 	if (hw->mac.type == ixgbe_mac_82598EB)
4403 		rxctrl |= IXGBE_RXCTRL_DMBYPS;
4404 
4405 	/* enable all receives */
4406 	rxctrl |= IXGBE_RXCTRL_RXEN;
4407 	hw->mac.ops.enable_rx_dma(hw, rxctrl);
4408 }
4409 
4410 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
4411 				 __be16 proto, u16 vid)
4412 {
4413 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4414 	struct ixgbe_hw *hw = &adapter->hw;
4415 
4416 	/* add VID to filter table */
4417 	if (!vid || !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4418 		hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true, !!vid);
4419 
4420 	set_bit(vid, adapter->active_vlans);
4421 
4422 	return 0;
4423 }
4424 
4425 static int ixgbe_find_vlvf_entry(struct ixgbe_hw *hw, u32 vlan)
4426 {
4427 	u32 vlvf;
4428 	int idx;
4429 
4430 	/* short cut the special case */
4431 	if (vlan == 0)
4432 		return 0;
4433 
4434 	/* Search for the vlan id in the VLVF entries */
4435 	for (idx = IXGBE_VLVF_ENTRIES; --idx;) {
4436 		vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(idx));
4437 		if ((vlvf & VLAN_VID_MASK) == vlan)
4438 			break;
4439 	}
4440 
4441 	return idx;
4442 }
4443 
4444 void ixgbe_update_pf_promisc_vlvf(struct ixgbe_adapter *adapter, u32 vid)
4445 {
4446 	struct ixgbe_hw *hw = &adapter->hw;
4447 	u32 bits, word;
4448 	int idx;
4449 
4450 	idx = ixgbe_find_vlvf_entry(hw, vid);
4451 	if (!idx)
4452 		return;
4453 
4454 	/* See if any other pools are set for this VLAN filter
4455 	 * entry other than the PF.
4456 	 */
4457 	word = idx * 2 + (VMDQ_P(0) / 32);
4458 	bits = ~BIT(VMDQ_P(0) % 32);
4459 	bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4460 
4461 	/* Disable the filter so this falls into the default pool. */
4462 	if (!bits && !IXGBE_READ_REG(hw, IXGBE_VLVFB(word ^ 1))) {
4463 		if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4464 			IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), 0);
4465 		IXGBE_WRITE_REG(hw, IXGBE_VLVF(idx), 0);
4466 	}
4467 }
4468 
4469 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
4470 				  __be16 proto, u16 vid)
4471 {
4472 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4473 	struct ixgbe_hw *hw = &adapter->hw;
4474 
4475 	/* remove VID from filter table */
4476 	if (vid && !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4477 		hw->mac.ops.set_vfta(hw, vid, VMDQ_P(0), false, true);
4478 
4479 	clear_bit(vid, adapter->active_vlans);
4480 
4481 	return 0;
4482 }
4483 
4484 /**
4485  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
4486  * @adapter: driver data
4487  */
4488 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
4489 {
4490 	struct ixgbe_hw *hw = &adapter->hw;
4491 	u32 vlnctrl;
4492 	int i, j;
4493 
4494 	switch (hw->mac.type) {
4495 	case ixgbe_mac_82598EB:
4496 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4497 		vlnctrl &= ~IXGBE_VLNCTRL_VME;
4498 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4499 		break;
4500 	case ixgbe_mac_82599EB:
4501 	case ixgbe_mac_X540:
4502 	case ixgbe_mac_X550:
4503 	case ixgbe_mac_X550EM_x:
4504 	case ixgbe_mac_x550em_a:
4505 		for (i = 0; i < adapter->num_rx_queues; i++) {
4506 			struct ixgbe_ring *ring = adapter->rx_ring[i];
4507 
4508 			if (!netif_is_ixgbe(ring->netdev))
4509 				continue;
4510 
4511 			j = ring->reg_idx;
4512 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4513 			vlnctrl &= ~IXGBE_RXDCTL_VME;
4514 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4515 		}
4516 		break;
4517 	default:
4518 		break;
4519 	}
4520 }
4521 
4522 /**
4523  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
4524  * @adapter: driver data
4525  */
4526 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
4527 {
4528 	struct ixgbe_hw *hw = &adapter->hw;
4529 	u32 vlnctrl;
4530 	int i, j;
4531 
4532 	switch (hw->mac.type) {
4533 	case ixgbe_mac_82598EB:
4534 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4535 		vlnctrl |= IXGBE_VLNCTRL_VME;
4536 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4537 		break;
4538 	case ixgbe_mac_82599EB:
4539 	case ixgbe_mac_X540:
4540 	case ixgbe_mac_X550:
4541 	case ixgbe_mac_X550EM_x:
4542 	case ixgbe_mac_x550em_a:
4543 		for (i = 0; i < adapter->num_rx_queues; i++) {
4544 			struct ixgbe_ring *ring = adapter->rx_ring[i];
4545 
4546 			if (!netif_is_ixgbe(ring->netdev))
4547 				continue;
4548 
4549 			j = ring->reg_idx;
4550 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4551 			vlnctrl |= IXGBE_RXDCTL_VME;
4552 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4553 		}
4554 		break;
4555 	default:
4556 		break;
4557 	}
4558 }
4559 
4560 static void ixgbe_vlan_promisc_enable(struct ixgbe_adapter *adapter)
4561 {
4562 	struct ixgbe_hw *hw = &adapter->hw;
4563 	u32 vlnctrl, i;
4564 
4565 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4566 
4567 	if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
4568 	/* For VMDq and SR-IOV we must leave VLAN filtering enabled */
4569 		vlnctrl |= IXGBE_VLNCTRL_VFE;
4570 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4571 	} else {
4572 		vlnctrl &= ~IXGBE_VLNCTRL_VFE;
4573 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4574 		return;
4575 	}
4576 
4577 	/* Nothing to do for 82598 */
4578 	if (hw->mac.type == ixgbe_mac_82598EB)
4579 		return;
4580 
4581 	/* We are already in VLAN promisc, nothing to do */
4582 	if (adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)
4583 		return;
4584 
4585 	/* Set flag so we don't redo unnecessary work */
4586 	adapter->flags2 |= IXGBE_FLAG2_VLAN_PROMISC;
4587 
4588 	/* Add PF to all active pools */
4589 	for (i = IXGBE_VLVF_ENTRIES; --i;) {
4590 		u32 reg_offset = IXGBE_VLVFB(i * 2 + VMDQ_P(0) / 32);
4591 		u32 vlvfb = IXGBE_READ_REG(hw, reg_offset);
4592 
4593 		vlvfb |= BIT(VMDQ_P(0) % 32);
4594 		IXGBE_WRITE_REG(hw, reg_offset, vlvfb);
4595 	}
4596 
4597 	/* Set all bits in the VLAN filter table array */
4598 	for (i = hw->mac.vft_size; i--;)
4599 		IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), ~0U);
4600 }
4601 
4602 #define VFTA_BLOCK_SIZE 8
4603 static void ixgbe_scrub_vfta(struct ixgbe_adapter *adapter, u32 vfta_offset)
4604 {
4605 	struct ixgbe_hw *hw = &adapter->hw;
4606 	u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4607 	u32 vid_start = vfta_offset * 32;
4608 	u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4609 	u32 i, vid, word, bits;
4610 
4611 	for (i = IXGBE_VLVF_ENTRIES; --i;) {
4612 		u32 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(i));
4613 
4614 		/* pull VLAN ID from VLVF */
4615 		vid = vlvf & VLAN_VID_MASK;
4616 
4617 		/* only concern outselves with a certain range */
4618 		if (vid < vid_start || vid >= vid_end)
4619 			continue;
4620 
4621 		if (vlvf) {
4622 			/* record VLAN ID in VFTA */
4623 			vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4624 
4625 			/* if PF is part of this then continue */
4626 			if (test_bit(vid, adapter->active_vlans))
4627 				continue;
4628 		}
4629 
4630 		/* remove PF from the pool */
4631 		word = i * 2 + VMDQ_P(0) / 32;
4632 		bits = ~BIT(VMDQ_P(0) % 32);
4633 		bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4634 		IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), bits);
4635 	}
4636 
4637 	/* extract values from active_vlans and write back to VFTA */
4638 	for (i = VFTA_BLOCK_SIZE; i--;) {
4639 		vid = (vfta_offset + i) * 32;
4640 		word = vid / BITS_PER_LONG;
4641 		bits = vid % BITS_PER_LONG;
4642 
4643 		vfta[i] |= adapter->active_vlans[word] >> bits;
4644 
4645 		IXGBE_WRITE_REG(hw, IXGBE_VFTA(vfta_offset + i), vfta[i]);
4646 	}
4647 }
4648 
4649 static void ixgbe_vlan_promisc_disable(struct ixgbe_adapter *adapter)
4650 {
4651 	struct ixgbe_hw *hw = &adapter->hw;
4652 	u32 vlnctrl, i;
4653 
4654 	/* Set VLAN filtering to enabled */
4655 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4656 	vlnctrl |= IXGBE_VLNCTRL_VFE;
4657 	IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4658 
4659 	if (!(adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) ||
4660 	    hw->mac.type == ixgbe_mac_82598EB)
4661 		return;
4662 
4663 	/* We are not in VLAN promisc, nothing to do */
4664 	if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4665 		return;
4666 
4667 	/* Set flag so we don't redo unnecessary work */
4668 	adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4669 
4670 	for (i = 0; i < hw->mac.vft_size; i += VFTA_BLOCK_SIZE)
4671 		ixgbe_scrub_vfta(adapter, i);
4672 }
4673 
4674 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
4675 {
4676 	u16 vid = 1;
4677 
4678 	ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
4679 
4680 	for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
4681 		ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
4682 }
4683 
4684 /**
4685  * ixgbe_write_mc_addr_list - write multicast addresses to MTA
4686  * @netdev: network interface device structure
4687  *
4688  * Writes multicast address list to the MTA hash table.
4689  * Returns: -ENOMEM on failure
4690  *                0 on no addresses written
4691  *                X on writing X addresses to MTA
4692  **/
4693 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
4694 {
4695 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4696 	struct ixgbe_hw *hw = &adapter->hw;
4697 
4698 	if (!netif_running(netdev))
4699 		return 0;
4700 
4701 	if (hw->mac.ops.update_mc_addr_list)
4702 		hw->mac.ops.update_mc_addr_list(hw, netdev);
4703 	else
4704 		return -ENOMEM;
4705 
4706 #ifdef CONFIG_PCI_IOV
4707 	ixgbe_restore_vf_multicasts(adapter);
4708 #endif
4709 
4710 	return netdev_mc_count(netdev);
4711 }
4712 
4713 #ifdef CONFIG_PCI_IOV
4714 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4715 {
4716 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4717 	struct ixgbe_hw *hw = &adapter->hw;
4718 	int i;
4719 
4720 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4721 		mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4722 
4723 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4724 			hw->mac.ops.set_rar(hw, i,
4725 					    mac_table->addr,
4726 					    mac_table->pool,
4727 					    IXGBE_RAH_AV);
4728 		else
4729 			hw->mac.ops.clear_rar(hw, i);
4730 	}
4731 }
4732 
4733 #endif
4734 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4735 {
4736 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4737 	struct ixgbe_hw *hw = &adapter->hw;
4738 	int i;
4739 
4740 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4741 		if (!(mac_table->state & IXGBE_MAC_STATE_MODIFIED))
4742 			continue;
4743 
4744 		mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4745 
4746 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4747 			hw->mac.ops.set_rar(hw, i,
4748 					    mac_table->addr,
4749 					    mac_table->pool,
4750 					    IXGBE_RAH_AV);
4751 		else
4752 			hw->mac.ops.clear_rar(hw, i);
4753 	}
4754 }
4755 
4756 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4757 {
4758 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4759 	struct ixgbe_hw *hw = &adapter->hw;
4760 	int i;
4761 
4762 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4763 		mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4764 		mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4765 	}
4766 
4767 	ixgbe_sync_mac_table(adapter);
4768 }
4769 
4770 static int ixgbe_available_rars(struct ixgbe_adapter *adapter, u16 pool)
4771 {
4772 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4773 	struct ixgbe_hw *hw = &adapter->hw;
4774 	int i, count = 0;
4775 
4776 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4777 		/* do not count default RAR as available */
4778 		if (mac_table->state & IXGBE_MAC_STATE_DEFAULT)
4779 			continue;
4780 
4781 		/* only count unused and addresses that belong to us */
4782 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE) {
4783 			if (mac_table->pool != pool)
4784 				continue;
4785 		}
4786 
4787 		count++;
4788 	}
4789 
4790 	return count;
4791 }
4792 
4793 /* this function destroys the first RAR entry */
4794 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter)
4795 {
4796 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4797 	struct ixgbe_hw *hw = &adapter->hw;
4798 
4799 	memcpy(&mac_table->addr, hw->mac.addr, ETH_ALEN);
4800 	mac_table->pool = VMDQ_P(0);
4801 
4802 	mac_table->state = IXGBE_MAC_STATE_DEFAULT | IXGBE_MAC_STATE_IN_USE;
4803 
4804 	hw->mac.ops.set_rar(hw, 0, mac_table->addr, mac_table->pool,
4805 			    IXGBE_RAH_AV);
4806 }
4807 
4808 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter,
4809 			 const u8 *addr, u16 pool)
4810 {
4811 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4812 	struct ixgbe_hw *hw = &adapter->hw;
4813 	int i;
4814 
4815 	if (is_zero_ether_addr(addr))
4816 		return -EINVAL;
4817 
4818 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4819 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4820 			continue;
4821 
4822 		ether_addr_copy(mac_table->addr, addr);
4823 		mac_table->pool = pool;
4824 
4825 		mac_table->state |= IXGBE_MAC_STATE_MODIFIED |
4826 				    IXGBE_MAC_STATE_IN_USE;
4827 
4828 		ixgbe_sync_mac_table(adapter);
4829 
4830 		return i;
4831 	}
4832 
4833 	return -ENOMEM;
4834 }
4835 
4836 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter,
4837 			 const u8 *addr, u16 pool)
4838 {
4839 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4840 	struct ixgbe_hw *hw = &adapter->hw;
4841 	int i;
4842 
4843 	if (is_zero_ether_addr(addr))
4844 		return -EINVAL;
4845 
4846 	/* search table for addr, if found clear IN_USE flag and sync */
4847 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4848 		/* we can only delete an entry if it is in use */
4849 		if (!(mac_table->state & IXGBE_MAC_STATE_IN_USE))
4850 			continue;
4851 		/* we only care about entries that belong to the given pool */
4852 		if (mac_table->pool != pool)
4853 			continue;
4854 		/* we only care about a specific MAC address */
4855 		if (!ether_addr_equal(addr, mac_table->addr))
4856 			continue;
4857 
4858 		mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4859 		mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4860 
4861 		ixgbe_sync_mac_table(adapter);
4862 
4863 		return 0;
4864 	}
4865 
4866 	return -ENOMEM;
4867 }
4868 
4869 static int ixgbe_uc_sync(struct net_device *netdev, const unsigned char *addr)
4870 {
4871 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4872 	int ret;
4873 
4874 	ret = ixgbe_add_mac_filter(adapter, addr, VMDQ_P(0));
4875 
4876 	return min_t(int, ret, 0);
4877 }
4878 
4879 static int ixgbe_uc_unsync(struct net_device *netdev, const unsigned char *addr)
4880 {
4881 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4882 
4883 	ixgbe_del_mac_filter(adapter, addr, VMDQ_P(0));
4884 
4885 	return 0;
4886 }
4887 
4888 /**
4889  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4890  * @netdev: network interface device structure
4891  *
4892  * The set_rx_method entry point is called whenever the unicast/multicast
4893  * address list or the network interface flags are updated.  This routine is
4894  * responsible for configuring the hardware for proper unicast, multicast and
4895  * promiscuous mode.
4896  **/
4897 void ixgbe_set_rx_mode(struct net_device *netdev)
4898 {
4899 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4900 	struct ixgbe_hw *hw = &adapter->hw;
4901 	u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4902 	netdev_features_t features = netdev->features;
4903 	int count;
4904 
4905 	/* Check for Promiscuous and All Multicast modes */
4906 	fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4907 
4908 	/* set all bits that we expect to always be set */
4909 	fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4910 	fctrl |= IXGBE_FCTRL_BAM;
4911 	fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4912 	fctrl |= IXGBE_FCTRL_PMCF;
4913 
4914 	/* clear the bits we are changing the status of */
4915 	fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4916 	if (netdev->flags & IFF_PROMISC) {
4917 		hw->addr_ctrl.user_set_promisc = true;
4918 		fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4919 		vmolr |= IXGBE_VMOLR_MPE;
4920 		features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4921 	} else {
4922 		if (netdev->flags & IFF_ALLMULTI) {
4923 			fctrl |= IXGBE_FCTRL_MPE;
4924 			vmolr |= IXGBE_VMOLR_MPE;
4925 		}
4926 		hw->addr_ctrl.user_set_promisc = false;
4927 	}
4928 
4929 	/*
4930 	 * Write addresses to available RAR registers, if there is not
4931 	 * sufficient space to store all the addresses then enable
4932 	 * unicast promiscuous mode
4933 	 */
4934 	if (__dev_uc_sync(netdev, ixgbe_uc_sync, ixgbe_uc_unsync)) {
4935 		fctrl |= IXGBE_FCTRL_UPE;
4936 		vmolr |= IXGBE_VMOLR_ROPE;
4937 	}
4938 
4939 	/* Write addresses to the MTA, if the attempt fails
4940 	 * then we should just turn on promiscuous mode so
4941 	 * that we can at least receive multicast traffic
4942 	 */
4943 	count = ixgbe_write_mc_addr_list(netdev);
4944 	if (count < 0) {
4945 		fctrl |= IXGBE_FCTRL_MPE;
4946 		vmolr |= IXGBE_VMOLR_MPE;
4947 	} else if (count) {
4948 		vmolr |= IXGBE_VMOLR_ROMPE;
4949 	}
4950 
4951 	if (hw->mac.type != ixgbe_mac_82598EB) {
4952 		vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4953 			 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4954 			   IXGBE_VMOLR_ROPE);
4955 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4956 	}
4957 
4958 	/* This is useful for sniffing bad packets. */
4959 	if (features & NETIF_F_RXALL) {
4960 		/* UPE and MPE will be handled by normal PROMISC logic
4961 		 * in e1000e_set_rx_mode */
4962 		fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4963 			  IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4964 			  IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4965 
4966 		fctrl &= ~(IXGBE_FCTRL_DPF);
4967 		/* NOTE:  VLAN filtering is disabled by setting PROMISC */
4968 	}
4969 
4970 	IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4971 
4972 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
4973 		ixgbe_vlan_strip_enable(adapter);
4974 	else
4975 		ixgbe_vlan_strip_disable(adapter);
4976 
4977 	if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
4978 		ixgbe_vlan_promisc_disable(adapter);
4979 	else
4980 		ixgbe_vlan_promisc_enable(adapter);
4981 }
4982 
4983 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
4984 {
4985 	int q_idx;
4986 
4987 	for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4988 		napi_enable(&adapter->q_vector[q_idx]->napi);
4989 }
4990 
4991 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
4992 {
4993 	int q_idx;
4994 
4995 	for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4996 		napi_disable(&adapter->q_vector[q_idx]->napi);
4997 }
4998 
4999 static void ixgbe_clear_udp_tunnel_port(struct ixgbe_adapter *adapter, u32 mask)
5000 {
5001 	struct ixgbe_hw *hw = &adapter->hw;
5002 	u32 vxlanctrl;
5003 
5004 	if (!(adapter->flags & (IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE |
5005 				IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE)))
5006 		return;
5007 
5008 	vxlanctrl = IXGBE_READ_REG(hw, IXGBE_VXLANCTRL) & ~mask;
5009 	IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, vxlanctrl);
5010 
5011 	if (mask & IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK)
5012 		adapter->vxlan_port = 0;
5013 
5014 	if (mask & IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK)
5015 		adapter->geneve_port = 0;
5016 }
5017 
5018 #ifdef CONFIG_IXGBE_DCB
5019 /**
5020  * ixgbe_configure_dcb - Configure DCB hardware
5021  * @adapter: ixgbe adapter struct
5022  *
5023  * This is called by the driver on open to configure the DCB hardware.
5024  * This is also called by the gennetlink interface when reconfiguring
5025  * the DCB state.
5026  */
5027 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
5028 {
5029 	struct ixgbe_hw *hw = &adapter->hw;
5030 	int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
5031 
5032 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
5033 		if (hw->mac.type == ixgbe_mac_82598EB)
5034 			netif_set_gso_max_size(adapter->netdev, 65536);
5035 		return;
5036 	}
5037 
5038 	if (hw->mac.type == ixgbe_mac_82598EB)
5039 		netif_set_gso_max_size(adapter->netdev, 32768);
5040 
5041 #ifdef IXGBE_FCOE
5042 	if (adapter->netdev->features & NETIF_F_FCOE_MTU)
5043 		max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
5044 #endif
5045 
5046 	/* reconfigure the hardware */
5047 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
5048 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5049 						DCB_TX_CONFIG);
5050 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5051 						DCB_RX_CONFIG);
5052 		ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
5053 	} else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
5054 		ixgbe_dcb_hw_ets(&adapter->hw,
5055 				 adapter->ixgbe_ieee_ets,
5056 				 max_frame);
5057 		ixgbe_dcb_hw_pfc_config(&adapter->hw,
5058 					adapter->ixgbe_ieee_pfc->pfc_en,
5059 					adapter->ixgbe_ieee_ets->prio_tc);
5060 	}
5061 
5062 	/* Enable RSS Hash per TC */
5063 	if (hw->mac.type != ixgbe_mac_82598EB) {
5064 		u32 msb = 0;
5065 		u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
5066 
5067 		while (rss_i) {
5068 			msb++;
5069 			rss_i >>= 1;
5070 		}
5071 
5072 		/* write msb to all 8 TCs in one write */
5073 		IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
5074 	}
5075 }
5076 #endif
5077 
5078 /* Additional bittime to account for IXGBE framing */
5079 #define IXGBE_ETH_FRAMING 20
5080 
5081 /**
5082  * ixgbe_hpbthresh - calculate high water mark for flow control
5083  *
5084  * @adapter: board private structure to calculate for
5085  * @pb: packet buffer to calculate
5086  */
5087 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
5088 {
5089 	struct ixgbe_hw *hw = &adapter->hw;
5090 	struct net_device *dev = adapter->netdev;
5091 	int link, tc, kb, marker;
5092 	u32 dv_id, rx_pba;
5093 
5094 	/* Calculate max LAN frame size */
5095 	tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
5096 
5097 #ifdef IXGBE_FCOE
5098 	/* FCoE traffic class uses FCOE jumbo frames */
5099 	if ((dev->features & NETIF_F_FCOE_MTU) &&
5100 	    (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5101 	    (pb == ixgbe_fcoe_get_tc(adapter)))
5102 		tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5103 #endif
5104 
5105 	/* Calculate delay value for device */
5106 	switch (hw->mac.type) {
5107 	case ixgbe_mac_X540:
5108 	case ixgbe_mac_X550:
5109 	case ixgbe_mac_X550EM_x:
5110 	case ixgbe_mac_x550em_a:
5111 		dv_id = IXGBE_DV_X540(link, tc);
5112 		break;
5113 	default:
5114 		dv_id = IXGBE_DV(link, tc);
5115 		break;
5116 	}
5117 
5118 	/* Loopback switch introduces additional latency */
5119 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5120 		dv_id += IXGBE_B2BT(tc);
5121 
5122 	/* Delay value is calculated in bit times convert to KB */
5123 	kb = IXGBE_BT2KB(dv_id);
5124 	rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
5125 
5126 	marker = rx_pba - kb;
5127 
5128 	/* It is possible that the packet buffer is not large enough
5129 	 * to provide required headroom. In this case throw an error
5130 	 * to user and a do the best we can.
5131 	 */
5132 	if (marker < 0) {
5133 		e_warn(drv, "Packet Buffer(%i) can not provide enough"
5134 			    "headroom to support flow control."
5135 			    "Decrease MTU or number of traffic classes\n", pb);
5136 		marker = tc + 1;
5137 	}
5138 
5139 	return marker;
5140 }
5141 
5142 /**
5143  * ixgbe_lpbthresh - calculate low water mark for for flow control
5144  *
5145  * @adapter: board private structure to calculate for
5146  * @pb: packet buffer to calculate
5147  */
5148 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
5149 {
5150 	struct ixgbe_hw *hw = &adapter->hw;
5151 	struct net_device *dev = adapter->netdev;
5152 	int tc;
5153 	u32 dv_id;
5154 
5155 	/* Calculate max LAN frame size */
5156 	tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5157 
5158 #ifdef IXGBE_FCOE
5159 	/* FCoE traffic class uses FCOE jumbo frames */
5160 	if ((dev->features & NETIF_F_FCOE_MTU) &&
5161 	    (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5162 	    (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
5163 		tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5164 #endif
5165 
5166 	/* Calculate delay value for device */
5167 	switch (hw->mac.type) {
5168 	case ixgbe_mac_X540:
5169 	case ixgbe_mac_X550:
5170 	case ixgbe_mac_X550EM_x:
5171 	case ixgbe_mac_x550em_a:
5172 		dv_id = IXGBE_LOW_DV_X540(tc);
5173 		break;
5174 	default:
5175 		dv_id = IXGBE_LOW_DV(tc);
5176 		break;
5177 	}
5178 
5179 	/* Delay value is calculated in bit times convert to KB */
5180 	return IXGBE_BT2KB(dv_id);
5181 }
5182 
5183 /*
5184  * ixgbe_pbthresh_setup - calculate and setup high low water marks
5185  */
5186 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
5187 {
5188 	struct ixgbe_hw *hw = &adapter->hw;
5189 	int num_tc = adapter->hw_tcs;
5190 	int i;
5191 
5192 	if (!num_tc)
5193 		num_tc = 1;
5194 
5195 	for (i = 0; i < num_tc; i++) {
5196 		hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
5197 		hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
5198 
5199 		/* Low water marks must not be larger than high water marks */
5200 		if (hw->fc.low_water[i] > hw->fc.high_water[i])
5201 			hw->fc.low_water[i] = 0;
5202 	}
5203 
5204 	for (; i < MAX_TRAFFIC_CLASS; i++)
5205 		hw->fc.high_water[i] = 0;
5206 }
5207 
5208 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
5209 {
5210 	struct ixgbe_hw *hw = &adapter->hw;
5211 	int hdrm;
5212 	u8 tc = adapter->hw_tcs;
5213 
5214 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5215 	    adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5216 		hdrm = 32 << adapter->fdir_pballoc;
5217 	else
5218 		hdrm = 0;
5219 
5220 	hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
5221 	ixgbe_pbthresh_setup(adapter);
5222 }
5223 
5224 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
5225 {
5226 	struct ixgbe_hw *hw = &adapter->hw;
5227 	struct hlist_node *node2;
5228 	struct ixgbe_fdir_filter *filter;
5229 	u8 queue;
5230 
5231 	spin_lock(&adapter->fdir_perfect_lock);
5232 
5233 	if (!hlist_empty(&adapter->fdir_filter_list))
5234 		ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
5235 
5236 	hlist_for_each_entry_safe(filter, node2,
5237 				  &adapter->fdir_filter_list, fdir_node) {
5238 		if (filter->action == IXGBE_FDIR_DROP_QUEUE) {
5239 			queue = IXGBE_FDIR_DROP_QUEUE;
5240 		} else {
5241 			u32 ring = ethtool_get_flow_spec_ring(filter->action);
5242 			u8 vf = ethtool_get_flow_spec_ring_vf(filter->action);
5243 
5244 			if (!vf && (ring >= adapter->num_rx_queues)) {
5245 				e_err(drv, "FDIR restore failed without VF, ring: %u\n",
5246 				      ring);
5247 				continue;
5248 			} else if (vf &&
5249 				   ((vf > adapter->num_vfs) ||
5250 				     ring >= adapter->num_rx_queues_per_pool)) {
5251 				e_err(drv, "FDIR restore failed with VF, vf: %hhu, ring: %u\n",
5252 				      vf, ring);
5253 				continue;
5254 			}
5255 
5256 			/* Map the ring onto the absolute queue index */
5257 			if (!vf)
5258 				queue = adapter->rx_ring[ring]->reg_idx;
5259 			else
5260 				queue = ((vf - 1) *
5261 					adapter->num_rx_queues_per_pool) + ring;
5262 		}
5263 
5264 		ixgbe_fdir_write_perfect_filter_82599(hw,
5265 				&filter->filter, filter->sw_idx, queue);
5266 	}
5267 
5268 	spin_unlock(&adapter->fdir_perfect_lock);
5269 }
5270 
5271 /**
5272  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
5273  * @rx_ring: ring to free buffers from
5274  **/
5275 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
5276 {
5277 	u16 i = rx_ring->next_to_clean;
5278 	struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
5279 
5280 	if (rx_ring->xsk_umem) {
5281 		ixgbe_xsk_clean_rx_ring(rx_ring);
5282 		goto skip_free;
5283 	}
5284 
5285 	/* Free all the Rx ring sk_buffs */
5286 	while (i != rx_ring->next_to_alloc) {
5287 		if (rx_buffer->skb) {
5288 			struct sk_buff *skb = rx_buffer->skb;
5289 			if (IXGBE_CB(skb)->page_released)
5290 				dma_unmap_page_attrs(rx_ring->dev,
5291 						     IXGBE_CB(skb)->dma,
5292 						     ixgbe_rx_pg_size(rx_ring),
5293 						     DMA_FROM_DEVICE,
5294 						     IXGBE_RX_DMA_ATTR);
5295 			dev_kfree_skb(skb);
5296 		}
5297 
5298 		/* Invalidate cache lines that may have been written to by
5299 		 * device so that we avoid corrupting memory.
5300 		 */
5301 		dma_sync_single_range_for_cpu(rx_ring->dev,
5302 					      rx_buffer->dma,
5303 					      rx_buffer->page_offset,
5304 					      ixgbe_rx_bufsz(rx_ring),
5305 					      DMA_FROM_DEVICE);
5306 
5307 		/* free resources associated with mapping */
5308 		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
5309 				     ixgbe_rx_pg_size(rx_ring),
5310 				     DMA_FROM_DEVICE,
5311 				     IXGBE_RX_DMA_ATTR);
5312 		__page_frag_cache_drain(rx_buffer->page,
5313 					rx_buffer->pagecnt_bias);
5314 
5315 		i++;
5316 		rx_buffer++;
5317 		if (i == rx_ring->count) {
5318 			i = 0;
5319 			rx_buffer = rx_ring->rx_buffer_info;
5320 		}
5321 	}
5322 
5323 skip_free:
5324 	rx_ring->next_to_alloc = 0;
5325 	rx_ring->next_to_clean = 0;
5326 	rx_ring->next_to_use = 0;
5327 }
5328 
5329 static int ixgbe_fwd_ring_up(struct ixgbe_adapter *adapter,
5330 			     struct ixgbe_fwd_adapter *accel)
5331 {
5332 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
5333 	int num_tc = netdev_get_num_tc(adapter->netdev);
5334 	struct net_device *vdev = accel->netdev;
5335 	int i, baseq, err;
5336 
5337 	baseq = accel->pool * adapter->num_rx_queues_per_pool;
5338 	netdev_dbg(vdev, "pool %i:%i queues %i:%i\n",
5339 		   accel->pool, adapter->num_rx_pools,
5340 		   baseq, baseq + adapter->num_rx_queues_per_pool);
5341 
5342 	accel->rx_base_queue = baseq;
5343 	accel->tx_base_queue = baseq;
5344 
5345 	/* record configuration for macvlan interface in vdev */
5346 	for (i = 0; i < num_tc; i++)
5347 		netdev_bind_sb_channel_queue(adapter->netdev, vdev,
5348 					     i, rss_i, baseq + (rss_i * i));
5349 
5350 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5351 		adapter->rx_ring[baseq + i]->netdev = vdev;
5352 
5353 	/* Guarantee all rings are updated before we update the
5354 	 * MAC address filter.
5355 	 */
5356 	wmb();
5357 
5358 	/* ixgbe_add_mac_filter will return an index if it succeeds, so we
5359 	 * need to only treat it as an error value if it is negative.
5360 	 */
5361 	err = ixgbe_add_mac_filter(adapter, vdev->dev_addr,
5362 				   VMDQ_P(accel->pool));
5363 	if (err >= 0)
5364 		return 0;
5365 
5366 	/* if we cannot add the MAC rule then disable the offload */
5367 	macvlan_release_l2fw_offload(vdev);
5368 
5369 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5370 		adapter->rx_ring[baseq + i]->netdev = NULL;
5371 
5372 	netdev_err(vdev, "L2FW offload disabled due to L2 filter error\n");
5373 
5374 	/* unbind the queues and drop the subordinate channel config */
5375 	netdev_unbind_sb_channel(adapter->netdev, vdev);
5376 	netdev_set_sb_channel(vdev, 0);
5377 
5378 	clear_bit(accel->pool, adapter->fwd_bitmask);
5379 	kfree(accel);
5380 
5381 	return err;
5382 }
5383 
5384 static int ixgbe_macvlan_up(struct net_device *vdev, void *data)
5385 {
5386 	struct ixgbe_adapter *adapter = data;
5387 	struct ixgbe_fwd_adapter *accel;
5388 
5389 	if (!netif_is_macvlan(vdev))
5390 		return 0;
5391 
5392 	accel = macvlan_accel_priv(vdev);
5393 	if (!accel)
5394 		return 0;
5395 
5396 	ixgbe_fwd_ring_up(adapter, accel);
5397 
5398 	return 0;
5399 }
5400 
5401 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
5402 {
5403 	netdev_walk_all_upper_dev_rcu(adapter->netdev,
5404 				      ixgbe_macvlan_up, adapter);
5405 }
5406 
5407 static void ixgbe_configure(struct ixgbe_adapter *adapter)
5408 {
5409 	struct ixgbe_hw *hw = &adapter->hw;
5410 
5411 	ixgbe_configure_pb(adapter);
5412 #ifdef CONFIG_IXGBE_DCB
5413 	ixgbe_configure_dcb(adapter);
5414 #endif
5415 	/*
5416 	 * We must restore virtualization before VLANs or else
5417 	 * the VLVF registers will not be populated
5418 	 */
5419 	ixgbe_configure_virtualization(adapter);
5420 
5421 	ixgbe_set_rx_mode(adapter->netdev);
5422 	ixgbe_restore_vlan(adapter);
5423 	ixgbe_ipsec_restore(adapter);
5424 
5425 	switch (hw->mac.type) {
5426 	case ixgbe_mac_82599EB:
5427 	case ixgbe_mac_X540:
5428 		hw->mac.ops.disable_rx_buff(hw);
5429 		break;
5430 	default:
5431 		break;
5432 	}
5433 
5434 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5435 		ixgbe_init_fdir_signature_82599(&adapter->hw,
5436 						adapter->fdir_pballoc);
5437 	} else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
5438 		ixgbe_init_fdir_perfect_82599(&adapter->hw,
5439 					      adapter->fdir_pballoc);
5440 		ixgbe_fdir_filter_restore(adapter);
5441 	}
5442 
5443 	switch (hw->mac.type) {
5444 	case ixgbe_mac_82599EB:
5445 	case ixgbe_mac_X540:
5446 		hw->mac.ops.enable_rx_buff(hw);
5447 		break;
5448 	default:
5449 		break;
5450 	}
5451 
5452 #ifdef CONFIG_IXGBE_DCA
5453 	/* configure DCA */
5454 	if (adapter->flags & IXGBE_FLAG_DCA_CAPABLE)
5455 		ixgbe_setup_dca(adapter);
5456 #endif /* CONFIG_IXGBE_DCA */
5457 
5458 #ifdef IXGBE_FCOE
5459 	/* configure FCoE L2 filters, redirection table, and Rx control */
5460 	ixgbe_configure_fcoe(adapter);
5461 
5462 #endif /* IXGBE_FCOE */
5463 	ixgbe_configure_tx(adapter);
5464 	ixgbe_configure_rx(adapter);
5465 	ixgbe_configure_dfwd(adapter);
5466 }
5467 
5468 /**
5469  * ixgbe_sfp_link_config - set up SFP+ link
5470  * @adapter: pointer to private adapter struct
5471  **/
5472 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
5473 {
5474 	/*
5475 	 * We are assuming the worst case scenario here, and that
5476 	 * is that an SFP was inserted/removed after the reset
5477 	 * but before SFP detection was enabled.  As such the best
5478 	 * solution is to just start searching as soon as we start
5479 	 */
5480 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5481 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5482 
5483 	adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5484 	adapter->sfp_poll_time = 0;
5485 }
5486 
5487 /**
5488  * ixgbe_non_sfp_link_config - set up non-SFP+ link
5489  * @hw: pointer to private hardware struct
5490  *
5491  * Returns 0 on success, negative on failure
5492  **/
5493 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
5494 {
5495 	u32 speed;
5496 	bool autoneg, link_up = false;
5497 	int ret = IXGBE_ERR_LINK_SETUP;
5498 
5499 	if (hw->mac.ops.check_link)
5500 		ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
5501 
5502 	if (ret)
5503 		return ret;
5504 
5505 	speed = hw->phy.autoneg_advertised;
5506 	if ((!speed) && (hw->mac.ops.get_link_capabilities))
5507 		ret = hw->mac.ops.get_link_capabilities(hw, &speed,
5508 							&autoneg);
5509 	if (ret)
5510 		return ret;
5511 
5512 	if (hw->mac.ops.setup_link)
5513 		ret = hw->mac.ops.setup_link(hw, speed, link_up);
5514 
5515 	return ret;
5516 }
5517 
5518 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
5519 {
5520 	struct ixgbe_hw *hw = &adapter->hw;
5521 	u32 gpie = 0;
5522 
5523 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5524 		gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
5525 		       IXGBE_GPIE_OCD;
5526 		gpie |= IXGBE_GPIE_EIAME;
5527 		/*
5528 		 * use EIAM to auto-mask when MSI-X interrupt is asserted
5529 		 * this saves a register write for every interrupt
5530 		 */
5531 		switch (hw->mac.type) {
5532 		case ixgbe_mac_82598EB:
5533 			IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5534 			break;
5535 		case ixgbe_mac_82599EB:
5536 		case ixgbe_mac_X540:
5537 		case ixgbe_mac_X550:
5538 		case ixgbe_mac_X550EM_x:
5539 		case ixgbe_mac_x550em_a:
5540 		default:
5541 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
5542 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
5543 			break;
5544 		}
5545 	} else {
5546 		/* legacy interrupts, use EIAM to auto-mask when reading EICR,
5547 		 * specifically only auto mask tx and rx interrupts */
5548 		IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5549 	}
5550 
5551 	/* XXX: to interrupt immediately for EICS writes, enable this */
5552 	/* gpie |= IXGBE_GPIE_EIMEN; */
5553 
5554 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
5555 		gpie &= ~IXGBE_GPIE_VTMODE_MASK;
5556 
5557 		switch (adapter->ring_feature[RING_F_VMDQ].mask) {
5558 		case IXGBE_82599_VMDQ_8Q_MASK:
5559 			gpie |= IXGBE_GPIE_VTMODE_16;
5560 			break;
5561 		case IXGBE_82599_VMDQ_4Q_MASK:
5562 			gpie |= IXGBE_GPIE_VTMODE_32;
5563 			break;
5564 		default:
5565 			gpie |= IXGBE_GPIE_VTMODE_64;
5566 			break;
5567 		}
5568 	}
5569 
5570 	/* Enable Thermal over heat sensor interrupt */
5571 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
5572 		switch (adapter->hw.mac.type) {
5573 		case ixgbe_mac_82599EB:
5574 			gpie |= IXGBE_SDP0_GPIEN_8259X;
5575 			break;
5576 		default:
5577 			break;
5578 		}
5579 	}
5580 
5581 	/* Enable fan failure interrupt */
5582 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
5583 		gpie |= IXGBE_SDP1_GPIEN(hw);
5584 
5585 	switch (hw->mac.type) {
5586 	case ixgbe_mac_82599EB:
5587 		gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X;
5588 		break;
5589 	case ixgbe_mac_X550EM_x:
5590 	case ixgbe_mac_x550em_a:
5591 		gpie |= IXGBE_SDP0_GPIEN_X540;
5592 		break;
5593 	default:
5594 		break;
5595 	}
5596 
5597 	IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
5598 }
5599 
5600 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
5601 {
5602 	struct ixgbe_hw *hw = &adapter->hw;
5603 	int err;
5604 	u32 ctrl_ext;
5605 
5606 	ixgbe_get_hw_control(adapter);
5607 	ixgbe_setup_gpie(adapter);
5608 
5609 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5610 		ixgbe_configure_msix(adapter);
5611 	else
5612 		ixgbe_configure_msi_and_legacy(adapter);
5613 
5614 	/* enable the optics for 82599 SFP+ fiber */
5615 	if (hw->mac.ops.enable_tx_laser)
5616 		hw->mac.ops.enable_tx_laser(hw);
5617 
5618 	if (hw->phy.ops.set_phy_power)
5619 		hw->phy.ops.set_phy_power(hw, true);
5620 
5621 	smp_mb__before_atomic();
5622 	clear_bit(__IXGBE_DOWN, &adapter->state);
5623 	ixgbe_napi_enable_all(adapter);
5624 
5625 	if (ixgbe_is_sfp(hw)) {
5626 		ixgbe_sfp_link_config(adapter);
5627 	} else {
5628 		err = ixgbe_non_sfp_link_config(hw);
5629 		if (err)
5630 			e_err(probe, "link_config FAILED %d\n", err);
5631 	}
5632 
5633 	/* clear any pending interrupts, may auto mask */
5634 	IXGBE_READ_REG(hw, IXGBE_EICR);
5635 	ixgbe_irq_enable(adapter, true, true);
5636 
5637 	/*
5638 	 * If this adapter has a fan, check to see if we had a failure
5639 	 * before we enabled the interrupt.
5640 	 */
5641 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5642 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5643 		if (esdp & IXGBE_ESDP_SDP1)
5644 			e_crit(drv, "Fan has stopped, replace the adapter\n");
5645 	}
5646 
5647 	/* bring the link up in the watchdog, this could race with our first
5648 	 * link up interrupt but shouldn't be a problem */
5649 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5650 	adapter->link_check_timeout = jiffies;
5651 	mod_timer(&adapter->service_timer, jiffies);
5652 
5653 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
5654 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5655 	ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
5656 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5657 }
5658 
5659 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
5660 {
5661 	WARN_ON(in_interrupt());
5662 	/* put off any impending NetWatchDogTimeout */
5663 	netif_trans_update(adapter->netdev);
5664 
5665 	while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
5666 		usleep_range(1000, 2000);
5667 	if (adapter->hw.phy.type == ixgbe_phy_fw)
5668 		ixgbe_watchdog_link_is_down(adapter);
5669 	ixgbe_down(adapter);
5670 	/*
5671 	 * If SR-IOV enabled then wait a bit before bringing the adapter
5672 	 * back up to give the VFs time to respond to the reset.  The
5673 	 * two second wait is based upon the watchdog timer cycle in
5674 	 * the VF driver.
5675 	 */
5676 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5677 		msleep(2000);
5678 	ixgbe_up(adapter);
5679 	clear_bit(__IXGBE_RESETTING, &adapter->state);
5680 }
5681 
5682 void ixgbe_up(struct ixgbe_adapter *adapter)
5683 {
5684 	/* hardware has been reset, we need to reload some things */
5685 	ixgbe_configure(adapter);
5686 
5687 	ixgbe_up_complete(adapter);
5688 }
5689 
5690 static unsigned long ixgbe_get_completion_timeout(struct ixgbe_adapter *adapter)
5691 {
5692 	u16 devctl2;
5693 
5694 	pcie_capability_read_word(adapter->pdev, PCI_EXP_DEVCTL2, &devctl2);
5695 
5696 	switch (devctl2 & IXGBE_PCIDEVCTRL2_TIMEO_MASK) {
5697 	case IXGBE_PCIDEVCTRL2_17_34s:
5698 	case IXGBE_PCIDEVCTRL2_4_8s:
5699 		/* For now we cap the upper limit on delay to 2 seconds
5700 		 * as we end up going up to 34 seconds of delay in worst
5701 		 * case timeout value.
5702 		 */
5703 	case IXGBE_PCIDEVCTRL2_1_2s:
5704 		return 2000000ul;	/* 2.0 s */
5705 	case IXGBE_PCIDEVCTRL2_260_520ms:
5706 		return 520000ul;	/* 520 ms */
5707 	case IXGBE_PCIDEVCTRL2_65_130ms:
5708 		return 130000ul;	/* 130 ms */
5709 	case IXGBE_PCIDEVCTRL2_16_32ms:
5710 		return 32000ul;		/* 32 ms */
5711 	case IXGBE_PCIDEVCTRL2_1_2ms:
5712 		return 2000ul;		/* 2 ms */
5713 	case IXGBE_PCIDEVCTRL2_50_100us:
5714 		return 100ul;		/* 100 us */
5715 	case IXGBE_PCIDEVCTRL2_16_32ms_def:
5716 		return 32000ul;		/* 32 ms */
5717 	default:
5718 		break;
5719 	}
5720 
5721 	/* We shouldn't need to hit this path, but just in case default as
5722 	 * though completion timeout is not supported and support 32ms.
5723 	 */
5724 	return 32000ul;
5725 }
5726 
5727 void ixgbe_disable_rx(struct ixgbe_adapter *adapter)
5728 {
5729 	unsigned long wait_delay, delay_interval;
5730 	struct ixgbe_hw *hw = &adapter->hw;
5731 	int i, wait_loop;
5732 	u32 rxdctl;
5733 
5734 	/* disable receives */
5735 	hw->mac.ops.disable_rx(hw);
5736 
5737 	if (ixgbe_removed(hw->hw_addr))
5738 		return;
5739 
5740 	/* disable all enabled Rx queues */
5741 	for (i = 0; i < adapter->num_rx_queues; i++) {
5742 		struct ixgbe_ring *ring = adapter->rx_ring[i];
5743 		u8 reg_idx = ring->reg_idx;
5744 
5745 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5746 		rxdctl &= ~IXGBE_RXDCTL_ENABLE;
5747 		rxdctl |= IXGBE_RXDCTL_SWFLSH;
5748 
5749 		/* write value back with RXDCTL.ENABLE bit cleared */
5750 		IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
5751 	}
5752 
5753 	/* RXDCTL.EN may not change on 82598 if link is down, so skip it */
5754 	if (hw->mac.type == ixgbe_mac_82598EB &&
5755 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5756 		return;
5757 
5758 	/* Determine our minimum delay interval. We will increase this value
5759 	 * with each subsequent test. This way if the device returns quickly
5760 	 * we should spend as little time as possible waiting, however as
5761 	 * the time increases we will wait for larger periods of time.
5762 	 *
5763 	 * The trick here is that we increase the interval using the
5764 	 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5765 	 * of that wait is that it totals up to 100x whatever interval we
5766 	 * choose. Since our minimum wait is 100us we can just divide the
5767 	 * total timeout by 100 to get our minimum delay interval.
5768 	 */
5769 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5770 
5771 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
5772 	wait_delay = delay_interval;
5773 
5774 	while (wait_loop--) {
5775 		usleep_range(wait_delay, wait_delay + 10);
5776 		wait_delay += delay_interval * 2;
5777 		rxdctl = 0;
5778 
5779 		/* OR together the reading of all the active RXDCTL registers,
5780 		 * and then test the result. We need the disable to complete
5781 		 * before we start freeing the memory and invalidating the
5782 		 * DMA mappings.
5783 		 */
5784 		for (i = 0; i < adapter->num_rx_queues; i++) {
5785 			struct ixgbe_ring *ring = adapter->rx_ring[i];
5786 			u8 reg_idx = ring->reg_idx;
5787 
5788 			rxdctl |= IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5789 		}
5790 
5791 		if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
5792 			return;
5793 	}
5794 
5795 	e_err(drv,
5796 	      "RXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5797 }
5798 
5799 void ixgbe_disable_tx(struct ixgbe_adapter *adapter)
5800 {
5801 	unsigned long wait_delay, delay_interval;
5802 	struct ixgbe_hw *hw = &adapter->hw;
5803 	int i, wait_loop;
5804 	u32 txdctl;
5805 
5806 	if (ixgbe_removed(hw->hw_addr))
5807 		return;
5808 
5809 	/* disable all enabled Tx queues */
5810 	for (i = 0; i < adapter->num_tx_queues; i++) {
5811 		struct ixgbe_ring *ring = adapter->tx_ring[i];
5812 		u8 reg_idx = ring->reg_idx;
5813 
5814 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5815 	}
5816 
5817 	/* disable all enabled XDP Tx queues */
5818 	for (i = 0; i < adapter->num_xdp_queues; i++) {
5819 		struct ixgbe_ring *ring = adapter->xdp_ring[i];
5820 		u8 reg_idx = ring->reg_idx;
5821 
5822 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5823 	}
5824 
5825 	/* If the link is not up there shouldn't be much in the way of
5826 	 * pending transactions. Those that are left will be flushed out
5827 	 * when the reset logic goes through the flush sequence to clean out
5828 	 * the pending Tx transactions.
5829 	 */
5830 	if (!(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5831 		goto dma_engine_disable;
5832 
5833 	/* Determine our minimum delay interval. We will increase this value
5834 	 * with each subsequent test. This way if the device returns quickly
5835 	 * we should spend as little time as possible waiting, however as
5836 	 * the time increases we will wait for larger periods of time.
5837 	 *
5838 	 * The trick here is that we increase the interval using the
5839 	 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5840 	 * of that wait is that it totals up to 100x whatever interval we
5841 	 * choose. Since our minimum wait is 100us we can just divide the
5842 	 * total timeout by 100 to get our minimum delay interval.
5843 	 */
5844 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5845 
5846 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
5847 	wait_delay = delay_interval;
5848 
5849 	while (wait_loop--) {
5850 		usleep_range(wait_delay, wait_delay + 10);
5851 		wait_delay += delay_interval * 2;
5852 		txdctl = 0;
5853 
5854 		/* OR together the reading of all the active TXDCTL registers,
5855 		 * and then test the result. We need the disable to complete
5856 		 * before we start freeing the memory and invalidating the
5857 		 * DMA mappings.
5858 		 */
5859 		for (i = 0; i < adapter->num_tx_queues; i++) {
5860 			struct ixgbe_ring *ring = adapter->tx_ring[i];
5861 			u8 reg_idx = ring->reg_idx;
5862 
5863 			txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5864 		}
5865 		for (i = 0; i < adapter->num_xdp_queues; i++) {
5866 			struct ixgbe_ring *ring = adapter->xdp_ring[i];
5867 			u8 reg_idx = ring->reg_idx;
5868 
5869 			txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5870 		}
5871 
5872 		if (!(txdctl & IXGBE_TXDCTL_ENABLE))
5873 			goto dma_engine_disable;
5874 	}
5875 
5876 	e_err(drv,
5877 	      "TXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5878 
5879 dma_engine_disable:
5880 	/* Disable the Tx DMA engine on 82599 and later MAC */
5881 	switch (hw->mac.type) {
5882 	case ixgbe_mac_82599EB:
5883 	case ixgbe_mac_X540:
5884 	case ixgbe_mac_X550:
5885 	case ixgbe_mac_X550EM_x:
5886 	case ixgbe_mac_x550em_a:
5887 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5888 				(IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5889 				 ~IXGBE_DMATXCTL_TE));
5890 		/* fall through */
5891 	default:
5892 		break;
5893 	}
5894 }
5895 
5896 void ixgbe_reset(struct ixgbe_adapter *adapter)
5897 {
5898 	struct ixgbe_hw *hw = &adapter->hw;
5899 	struct net_device *netdev = adapter->netdev;
5900 	int err;
5901 
5902 	if (ixgbe_removed(hw->hw_addr))
5903 		return;
5904 	/* lock SFP init bit to prevent race conditions with the watchdog */
5905 	while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5906 		usleep_range(1000, 2000);
5907 
5908 	/* clear all SFP and link config related flags while holding SFP_INIT */
5909 	adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
5910 			     IXGBE_FLAG2_SFP_NEEDS_RESET);
5911 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5912 
5913 	err = hw->mac.ops.init_hw(hw);
5914 	switch (err) {
5915 	case 0:
5916 	case IXGBE_ERR_SFP_NOT_PRESENT:
5917 	case IXGBE_ERR_SFP_NOT_SUPPORTED:
5918 		break;
5919 	case IXGBE_ERR_MASTER_REQUESTS_PENDING:
5920 		e_dev_err("master disable timed out\n");
5921 		break;
5922 	case IXGBE_ERR_EEPROM_VERSION:
5923 		/* We are running on a pre-production device, log a warning */
5924 		e_dev_warn("This device is a pre-production adapter/LOM. "
5925 			   "Please be aware there may be issues associated with "
5926 			   "your hardware.  If you are experiencing problems "
5927 			   "please contact your Intel or hardware "
5928 			   "representative who provided you with this "
5929 			   "hardware.\n");
5930 		break;
5931 	default:
5932 		e_dev_err("Hardware Error: %d\n", err);
5933 	}
5934 
5935 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5936 
5937 	/* flush entries out of MAC table */
5938 	ixgbe_flush_sw_mac_table(adapter);
5939 	__dev_uc_unsync(netdev, NULL);
5940 
5941 	/* do not flush user set addresses */
5942 	ixgbe_mac_set_default_filter(adapter);
5943 
5944 	/* update SAN MAC vmdq pool selection */
5945 	if (hw->mac.san_mac_rar_index)
5946 		hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
5947 
5948 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5949 		ixgbe_ptp_reset(adapter);
5950 
5951 	if (hw->phy.ops.set_phy_power) {
5952 		if (!netif_running(adapter->netdev) && !adapter->wol)
5953 			hw->phy.ops.set_phy_power(hw, false);
5954 		else
5955 			hw->phy.ops.set_phy_power(hw, true);
5956 	}
5957 }
5958 
5959 /**
5960  * ixgbe_clean_tx_ring - Free Tx Buffers
5961  * @tx_ring: ring to be cleaned
5962  **/
5963 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
5964 {
5965 	u16 i = tx_ring->next_to_clean;
5966 	struct ixgbe_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
5967 
5968 	if (tx_ring->xsk_umem) {
5969 		ixgbe_xsk_clean_tx_ring(tx_ring);
5970 		goto out;
5971 	}
5972 
5973 	while (i != tx_ring->next_to_use) {
5974 		union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
5975 
5976 		/* Free all the Tx ring sk_buffs */
5977 		if (ring_is_xdp(tx_ring))
5978 			xdp_return_frame(tx_buffer->xdpf);
5979 		else
5980 			dev_kfree_skb_any(tx_buffer->skb);
5981 
5982 		/* unmap skb header data */
5983 		dma_unmap_single(tx_ring->dev,
5984 				 dma_unmap_addr(tx_buffer, dma),
5985 				 dma_unmap_len(tx_buffer, len),
5986 				 DMA_TO_DEVICE);
5987 
5988 		/* check for eop_desc to determine the end of the packet */
5989 		eop_desc = tx_buffer->next_to_watch;
5990 		tx_desc = IXGBE_TX_DESC(tx_ring, i);
5991 
5992 		/* unmap remaining buffers */
5993 		while (tx_desc != eop_desc) {
5994 			tx_buffer++;
5995 			tx_desc++;
5996 			i++;
5997 			if (unlikely(i == tx_ring->count)) {
5998 				i = 0;
5999 				tx_buffer = tx_ring->tx_buffer_info;
6000 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6001 			}
6002 
6003 			/* unmap any remaining paged data */
6004 			if (dma_unmap_len(tx_buffer, len))
6005 				dma_unmap_page(tx_ring->dev,
6006 					       dma_unmap_addr(tx_buffer, dma),
6007 					       dma_unmap_len(tx_buffer, len),
6008 					       DMA_TO_DEVICE);
6009 		}
6010 
6011 		/* move us one more past the eop_desc for start of next pkt */
6012 		tx_buffer++;
6013 		i++;
6014 		if (unlikely(i == tx_ring->count)) {
6015 			i = 0;
6016 			tx_buffer = tx_ring->tx_buffer_info;
6017 		}
6018 	}
6019 
6020 	/* reset BQL for queue */
6021 	if (!ring_is_xdp(tx_ring))
6022 		netdev_tx_reset_queue(txring_txq(tx_ring));
6023 
6024 out:
6025 	/* reset next_to_use and next_to_clean */
6026 	tx_ring->next_to_use = 0;
6027 	tx_ring->next_to_clean = 0;
6028 }
6029 
6030 /**
6031  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
6032  * @adapter: board private structure
6033  **/
6034 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
6035 {
6036 	int i;
6037 
6038 	for (i = 0; i < adapter->num_rx_queues; i++)
6039 		ixgbe_clean_rx_ring(adapter->rx_ring[i]);
6040 }
6041 
6042 /**
6043  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
6044  * @adapter: board private structure
6045  **/
6046 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
6047 {
6048 	int i;
6049 
6050 	for (i = 0; i < adapter->num_tx_queues; i++)
6051 		ixgbe_clean_tx_ring(adapter->tx_ring[i]);
6052 	for (i = 0; i < adapter->num_xdp_queues; i++)
6053 		ixgbe_clean_tx_ring(adapter->xdp_ring[i]);
6054 }
6055 
6056 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
6057 {
6058 	struct hlist_node *node2;
6059 	struct ixgbe_fdir_filter *filter;
6060 
6061 	spin_lock(&adapter->fdir_perfect_lock);
6062 
6063 	hlist_for_each_entry_safe(filter, node2,
6064 				  &adapter->fdir_filter_list, fdir_node) {
6065 		hlist_del(&filter->fdir_node);
6066 		kfree(filter);
6067 	}
6068 	adapter->fdir_filter_count = 0;
6069 
6070 	spin_unlock(&adapter->fdir_perfect_lock);
6071 }
6072 
6073 void ixgbe_down(struct ixgbe_adapter *adapter)
6074 {
6075 	struct net_device *netdev = adapter->netdev;
6076 	struct ixgbe_hw *hw = &adapter->hw;
6077 	int i;
6078 
6079 	/* signal that we are down to the interrupt handler */
6080 	if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
6081 		return; /* do nothing if already down */
6082 
6083 	/* Shut off incoming Tx traffic */
6084 	netif_tx_stop_all_queues(netdev);
6085 
6086 	/* call carrier off first to avoid false dev_watchdog timeouts */
6087 	netif_carrier_off(netdev);
6088 	netif_tx_disable(netdev);
6089 
6090 	/* Disable Rx */
6091 	ixgbe_disable_rx(adapter);
6092 
6093 	/* synchronize_rcu() needed for pending XDP buffers to drain */
6094 	if (adapter->xdp_ring[0])
6095 		synchronize_rcu();
6096 
6097 	ixgbe_irq_disable(adapter);
6098 
6099 	ixgbe_napi_disable_all(adapter);
6100 
6101 	clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
6102 	adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6103 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6104 
6105 	del_timer_sync(&adapter->service_timer);
6106 
6107 	if (adapter->num_vfs) {
6108 		/* Clear EITR Select mapping */
6109 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
6110 
6111 		/* Mark all the VFs as inactive */
6112 		for (i = 0 ; i < adapter->num_vfs; i++)
6113 			adapter->vfinfo[i].clear_to_send = false;
6114 
6115 		/* ping all the active vfs to let them know we are going down */
6116 		ixgbe_ping_all_vfs(adapter);
6117 
6118 		/* Disable all VFTE/VFRE TX/RX */
6119 		ixgbe_disable_tx_rx(adapter);
6120 	}
6121 
6122 	/* disable transmits in the hardware now that interrupts are off */
6123 	ixgbe_disable_tx(adapter);
6124 
6125 	if (!pci_channel_offline(adapter->pdev))
6126 		ixgbe_reset(adapter);
6127 
6128 	/* power down the optics for 82599 SFP+ fiber */
6129 	if (hw->mac.ops.disable_tx_laser)
6130 		hw->mac.ops.disable_tx_laser(hw);
6131 
6132 	ixgbe_clean_all_tx_rings(adapter);
6133 	ixgbe_clean_all_rx_rings(adapter);
6134 }
6135 
6136 /**
6137  * ixgbe_eee_capable - helper function to determine EEE support on X550
6138  * @adapter: board private structure
6139  */
6140 static void ixgbe_set_eee_capable(struct ixgbe_adapter *adapter)
6141 {
6142 	struct ixgbe_hw *hw = &adapter->hw;
6143 
6144 	switch (hw->device_id) {
6145 	case IXGBE_DEV_ID_X550EM_A_1G_T:
6146 	case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6147 		if (!hw->phy.eee_speeds_supported)
6148 			break;
6149 		adapter->flags2 |= IXGBE_FLAG2_EEE_CAPABLE;
6150 		if (!hw->phy.eee_speeds_advertised)
6151 			break;
6152 		adapter->flags2 |= IXGBE_FLAG2_EEE_ENABLED;
6153 		break;
6154 	default:
6155 		adapter->flags2 &= ~IXGBE_FLAG2_EEE_CAPABLE;
6156 		adapter->flags2 &= ~IXGBE_FLAG2_EEE_ENABLED;
6157 		break;
6158 	}
6159 }
6160 
6161 /**
6162  * ixgbe_tx_timeout - Respond to a Tx Hang
6163  * @netdev: network interface device structure
6164  **/
6165 static void ixgbe_tx_timeout(struct net_device *netdev, unsigned int txqueue)
6166 {
6167 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6168 
6169 	/* Do the reset outside of interrupt context */
6170 	ixgbe_tx_timeout_reset(adapter);
6171 }
6172 
6173 #ifdef CONFIG_IXGBE_DCB
6174 static void ixgbe_init_dcb(struct ixgbe_adapter *adapter)
6175 {
6176 	struct ixgbe_hw *hw = &adapter->hw;
6177 	struct tc_configuration *tc;
6178 	int j;
6179 
6180 	switch (hw->mac.type) {
6181 	case ixgbe_mac_82598EB:
6182 	case ixgbe_mac_82599EB:
6183 		adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
6184 		adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
6185 		break;
6186 	case ixgbe_mac_X540:
6187 	case ixgbe_mac_X550:
6188 		adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
6189 		adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
6190 		break;
6191 	case ixgbe_mac_X550EM_x:
6192 	case ixgbe_mac_x550em_a:
6193 	default:
6194 		adapter->dcb_cfg.num_tcs.pg_tcs = DEF_TRAFFIC_CLASS;
6195 		adapter->dcb_cfg.num_tcs.pfc_tcs = DEF_TRAFFIC_CLASS;
6196 		break;
6197 	}
6198 
6199 	/* Configure DCB traffic classes */
6200 	for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
6201 		tc = &adapter->dcb_cfg.tc_config[j];
6202 		tc->path[DCB_TX_CONFIG].bwg_id = 0;
6203 		tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
6204 		tc->path[DCB_RX_CONFIG].bwg_id = 0;
6205 		tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
6206 		tc->dcb_pfc = pfc_disabled;
6207 	}
6208 
6209 	/* Initialize default user to priority mapping, UPx->TC0 */
6210 	tc = &adapter->dcb_cfg.tc_config[0];
6211 	tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
6212 	tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
6213 
6214 	adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
6215 	adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
6216 	adapter->dcb_cfg.pfc_mode_enable = false;
6217 	adapter->dcb_set_bitmap = 0x00;
6218 	if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
6219 		adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
6220 	memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
6221 	       sizeof(adapter->temp_dcb_cfg));
6222 }
6223 #endif
6224 
6225 /**
6226  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
6227  * @adapter: board private structure to initialize
6228  * @ii: pointer to ixgbe_info for device
6229  *
6230  * ixgbe_sw_init initializes the Adapter private data structure.
6231  * Fields are initialized based on PCI device information and
6232  * OS network device settings (MTU size).
6233  **/
6234 static int ixgbe_sw_init(struct ixgbe_adapter *adapter,
6235 			 const struct ixgbe_info *ii)
6236 {
6237 	struct ixgbe_hw *hw = &adapter->hw;
6238 	struct pci_dev *pdev = adapter->pdev;
6239 	unsigned int rss, fdir;
6240 	u32 fwsm;
6241 	int i;
6242 
6243 	/* PCI config space info */
6244 
6245 	hw->vendor_id = pdev->vendor;
6246 	hw->device_id = pdev->device;
6247 	hw->revision_id = pdev->revision;
6248 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
6249 	hw->subsystem_device_id = pdev->subsystem_device;
6250 
6251 	/* get_invariants needs the device IDs */
6252 	ii->get_invariants(hw);
6253 
6254 	/* Set common capability flags and settings */
6255 	rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
6256 	adapter->ring_feature[RING_F_RSS].limit = rss;
6257 	adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
6258 	adapter->max_q_vectors = MAX_Q_VECTORS_82599;
6259 	adapter->atr_sample_rate = 20;
6260 	fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
6261 	adapter->ring_feature[RING_F_FDIR].limit = fdir;
6262 	adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
6263 	adapter->ring_feature[RING_F_VMDQ].limit = 1;
6264 #ifdef CONFIG_IXGBE_DCA
6265 	adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
6266 #endif
6267 #ifdef CONFIG_IXGBE_DCB
6268 	adapter->flags |= IXGBE_FLAG_DCB_CAPABLE;
6269 	adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6270 #endif
6271 #ifdef IXGBE_FCOE
6272 	adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
6273 	adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6274 #ifdef CONFIG_IXGBE_DCB
6275 	/* Default traffic class to use for FCoE */
6276 	adapter->fcoe.up = IXGBE_FCOE_DEFTC;
6277 #endif /* CONFIG_IXGBE_DCB */
6278 #endif /* IXGBE_FCOE */
6279 
6280 	/* initialize static ixgbe jump table entries */
6281 	adapter->jump_tables[0] = kzalloc(sizeof(*adapter->jump_tables[0]),
6282 					  GFP_KERNEL);
6283 	if (!adapter->jump_tables[0])
6284 		return -ENOMEM;
6285 	adapter->jump_tables[0]->mat = ixgbe_ipv4_fields;
6286 
6287 	for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++)
6288 		adapter->jump_tables[i] = NULL;
6289 
6290 	adapter->mac_table = kcalloc(hw->mac.num_rar_entries,
6291 				     sizeof(struct ixgbe_mac_addr),
6292 				     GFP_KERNEL);
6293 	if (!adapter->mac_table)
6294 		return -ENOMEM;
6295 
6296 	if (ixgbe_init_rss_key(adapter))
6297 		return -ENOMEM;
6298 
6299 	adapter->af_xdp_zc_qps = bitmap_zalloc(MAX_XDP_QUEUES, GFP_KERNEL);
6300 	if (!adapter->af_xdp_zc_qps)
6301 		return -ENOMEM;
6302 
6303 	/* Set MAC specific capability flags and exceptions */
6304 	switch (hw->mac.type) {
6305 	case ixgbe_mac_82598EB:
6306 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
6307 
6308 		if (hw->device_id == IXGBE_DEV_ID_82598AT)
6309 			adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
6310 
6311 		adapter->max_q_vectors = MAX_Q_VECTORS_82598;
6312 		adapter->ring_feature[RING_F_FDIR].limit = 0;
6313 		adapter->atr_sample_rate = 0;
6314 		adapter->fdir_pballoc = 0;
6315 #ifdef IXGBE_FCOE
6316 		adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6317 		adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6318 #ifdef CONFIG_IXGBE_DCB
6319 		adapter->fcoe.up = 0;
6320 #endif /* IXGBE_DCB */
6321 #endif /* IXGBE_FCOE */
6322 		break;
6323 	case ixgbe_mac_82599EB:
6324 		if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
6325 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6326 		break;
6327 	case ixgbe_mac_X540:
6328 		fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
6329 		if (fwsm & IXGBE_FWSM_TS_ENABLED)
6330 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6331 		break;
6332 	case ixgbe_mac_x550em_a:
6333 		adapter->flags |= IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE;
6334 		switch (hw->device_id) {
6335 		case IXGBE_DEV_ID_X550EM_A_1G_T:
6336 		case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6337 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6338 			break;
6339 		default:
6340 			break;
6341 		}
6342 	/* fall through */
6343 	case ixgbe_mac_X550EM_x:
6344 #ifdef CONFIG_IXGBE_DCB
6345 		adapter->flags &= ~IXGBE_FLAG_DCB_CAPABLE;
6346 #endif
6347 #ifdef IXGBE_FCOE
6348 		adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6349 #ifdef CONFIG_IXGBE_DCB
6350 		adapter->fcoe.up = 0;
6351 #endif /* IXGBE_DCB */
6352 #endif /* IXGBE_FCOE */
6353 	/* Fall Through */
6354 	case ixgbe_mac_X550:
6355 		if (hw->mac.type == ixgbe_mac_X550)
6356 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6357 #ifdef CONFIG_IXGBE_DCA
6358 		adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
6359 #endif
6360 		adapter->flags |= IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE;
6361 		break;
6362 	default:
6363 		break;
6364 	}
6365 
6366 #ifdef IXGBE_FCOE
6367 	/* FCoE support exists, always init the FCoE lock */
6368 	spin_lock_init(&adapter->fcoe.lock);
6369 
6370 #endif
6371 	/* n-tuple support exists, always init our spinlock */
6372 	spin_lock_init(&adapter->fdir_perfect_lock);
6373 
6374 #ifdef CONFIG_IXGBE_DCB
6375 	ixgbe_init_dcb(adapter);
6376 #endif
6377 	ixgbe_init_ipsec_offload(adapter);
6378 
6379 	/* default flow control settings */
6380 	hw->fc.requested_mode = ixgbe_fc_full;
6381 	hw->fc.current_mode = ixgbe_fc_full;	/* init for ethtool output */
6382 	ixgbe_pbthresh_setup(adapter);
6383 	hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
6384 	hw->fc.send_xon = true;
6385 	hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
6386 
6387 #ifdef CONFIG_PCI_IOV
6388 	if (max_vfs > 0)
6389 		e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
6390 
6391 	/* assign number of SR-IOV VFs */
6392 	if (hw->mac.type != ixgbe_mac_82598EB) {
6393 		if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
6394 			max_vfs = 0;
6395 			e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
6396 		}
6397 	}
6398 #endif /* CONFIG_PCI_IOV */
6399 
6400 	/* enable itr by default in dynamic mode */
6401 	adapter->rx_itr_setting = 1;
6402 	adapter->tx_itr_setting = 1;
6403 
6404 	/* set default ring sizes */
6405 	adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
6406 	adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
6407 
6408 	/* set default work limits */
6409 	adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
6410 
6411 	/* initialize eeprom parameters */
6412 	if (ixgbe_init_eeprom_params_generic(hw)) {
6413 		e_dev_err("EEPROM initialization failed\n");
6414 		return -EIO;
6415 	}
6416 
6417 	/* PF holds first pool slot */
6418 	set_bit(0, adapter->fwd_bitmask);
6419 	set_bit(__IXGBE_DOWN, &adapter->state);
6420 
6421 	return 0;
6422 }
6423 
6424 /**
6425  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
6426  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
6427  *
6428  * Return 0 on success, negative on failure
6429  **/
6430 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
6431 {
6432 	struct device *dev = tx_ring->dev;
6433 	int orig_node = dev_to_node(dev);
6434 	int ring_node = NUMA_NO_NODE;
6435 	int size;
6436 
6437 	size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
6438 
6439 	if (tx_ring->q_vector)
6440 		ring_node = tx_ring->q_vector->numa_node;
6441 
6442 	tx_ring->tx_buffer_info = vmalloc_node(size, ring_node);
6443 	if (!tx_ring->tx_buffer_info)
6444 		tx_ring->tx_buffer_info = vmalloc(size);
6445 	if (!tx_ring->tx_buffer_info)
6446 		goto err;
6447 
6448 	/* round up to nearest 4K */
6449 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
6450 	tx_ring->size = ALIGN(tx_ring->size, 4096);
6451 
6452 	set_dev_node(dev, ring_node);
6453 	tx_ring->desc = dma_alloc_coherent(dev,
6454 					   tx_ring->size,
6455 					   &tx_ring->dma,
6456 					   GFP_KERNEL);
6457 	set_dev_node(dev, orig_node);
6458 	if (!tx_ring->desc)
6459 		tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
6460 						   &tx_ring->dma, GFP_KERNEL);
6461 	if (!tx_ring->desc)
6462 		goto err;
6463 
6464 	tx_ring->next_to_use = 0;
6465 	tx_ring->next_to_clean = 0;
6466 	return 0;
6467 
6468 err:
6469 	vfree(tx_ring->tx_buffer_info);
6470 	tx_ring->tx_buffer_info = NULL;
6471 	dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
6472 	return -ENOMEM;
6473 }
6474 
6475 /**
6476  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
6477  * @adapter: board private structure
6478  *
6479  * If this function returns with an error, then it's possible one or
6480  * more of the rings is populated (while the rest are not).  It is the
6481  * callers duty to clean those orphaned rings.
6482  *
6483  * Return 0 on success, negative on failure
6484  **/
6485 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
6486 {
6487 	int i, j = 0, err = 0;
6488 
6489 	for (i = 0; i < adapter->num_tx_queues; i++) {
6490 		err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
6491 		if (!err)
6492 			continue;
6493 
6494 		e_err(probe, "Allocation for Tx Queue %u failed\n", i);
6495 		goto err_setup_tx;
6496 	}
6497 	for (j = 0; j < adapter->num_xdp_queues; j++) {
6498 		err = ixgbe_setup_tx_resources(adapter->xdp_ring[j]);
6499 		if (!err)
6500 			continue;
6501 
6502 		e_err(probe, "Allocation for Tx Queue %u failed\n", j);
6503 		goto err_setup_tx;
6504 	}
6505 
6506 	return 0;
6507 err_setup_tx:
6508 	/* rewind the index freeing the rings as we go */
6509 	while (j--)
6510 		ixgbe_free_tx_resources(adapter->xdp_ring[j]);
6511 	while (i--)
6512 		ixgbe_free_tx_resources(adapter->tx_ring[i]);
6513 	return err;
6514 }
6515 
6516 /**
6517  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
6518  * @adapter: pointer to ixgbe_adapter
6519  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
6520  *
6521  * Returns 0 on success, negative on failure
6522  **/
6523 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
6524 			     struct ixgbe_ring *rx_ring)
6525 {
6526 	struct device *dev = rx_ring->dev;
6527 	int orig_node = dev_to_node(dev);
6528 	int ring_node = NUMA_NO_NODE;
6529 	int size;
6530 
6531 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
6532 
6533 	if (rx_ring->q_vector)
6534 		ring_node = rx_ring->q_vector->numa_node;
6535 
6536 	rx_ring->rx_buffer_info = vmalloc_node(size, ring_node);
6537 	if (!rx_ring->rx_buffer_info)
6538 		rx_ring->rx_buffer_info = vmalloc(size);
6539 	if (!rx_ring->rx_buffer_info)
6540 		goto err;
6541 
6542 	/* Round up to nearest 4K */
6543 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
6544 	rx_ring->size = ALIGN(rx_ring->size, 4096);
6545 
6546 	set_dev_node(dev, ring_node);
6547 	rx_ring->desc = dma_alloc_coherent(dev,
6548 					   rx_ring->size,
6549 					   &rx_ring->dma,
6550 					   GFP_KERNEL);
6551 	set_dev_node(dev, orig_node);
6552 	if (!rx_ring->desc)
6553 		rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
6554 						   &rx_ring->dma, GFP_KERNEL);
6555 	if (!rx_ring->desc)
6556 		goto err;
6557 
6558 	rx_ring->next_to_clean = 0;
6559 	rx_ring->next_to_use = 0;
6560 
6561 	/* XDP RX-queue info */
6562 	if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
6563 			     rx_ring->queue_index) < 0)
6564 		goto err;
6565 
6566 	rx_ring->xdp_prog = adapter->xdp_prog;
6567 
6568 	return 0;
6569 err:
6570 	vfree(rx_ring->rx_buffer_info);
6571 	rx_ring->rx_buffer_info = NULL;
6572 	dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
6573 	return -ENOMEM;
6574 }
6575 
6576 /**
6577  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
6578  * @adapter: board private structure
6579  *
6580  * If this function returns with an error, then it's possible one or
6581  * more of the rings is populated (while the rest are not).  It is the
6582  * callers duty to clean those orphaned rings.
6583  *
6584  * Return 0 on success, negative on failure
6585  **/
6586 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
6587 {
6588 	int i, err = 0;
6589 
6590 	for (i = 0; i < adapter->num_rx_queues; i++) {
6591 		err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
6592 		if (!err)
6593 			continue;
6594 
6595 		e_err(probe, "Allocation for Rx Queue %u failed\n", i);
6596 		goto err_setup_rx;
6597 	}
6598 
6599 #ifdef IXGBE_FCOE
6600 	err = ixgbe_setup_fcoe_ddp_resources(adapter);
6601 	if (!err)
6602 #endif
6603 		return 0;
6604 err_setup_rx:
6605 	/* rewind the index freeing the rings as we go */
6606 	while (i--)
6607 		ixgbe_free_rx_resources(adapter->rx_ring[i]);
6608 	return err;
6609 }
6610 
6611 /**
6612  * ixgbe_free_tx_resources - Free Tx Resources per Queue
6613  * @tx_ring: Tx descriptor ring for a specific queue
6614  *
6615  * Free all transmit software resources
6616  **/
6617 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
6618 {
6619 	ixgbe_clean_tx_ring(tx_ring);
6620 
6621 	vfree(tx_ring->tx_buffer_info);
6622 	tx_ring->tx_buffer_info = NULL;
6623 
6624 	/* if not set, then don't free */
6625 	if (!tx_ring->desc)
6626 		return;
6627 
6628 	dma_free_coherent(tx_ring->dev, tx_ring->size,
6629 			  tx_ring->desc, tx_ring->dma);
6630 
6631 	tx_ring->desc = NULL;
6632 }
6633 
6634 /**
6635  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
6636  * @adapter: board private structure
6637  *
6638  * Free all transmit software resources
6639  **/
6640 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
6641 {
6642 	int i;
6643 
6644 	for (i = 0; i < adapter->num_tx_queues; i++)
6645 		if (adapter->tx_ring[i]->desc)
6646 			ixgbe_free_tx_resources(adapter->tx_ring[i]);
6647 	for (i = 0; i < adapter->num_xdp_queues; i++)
6648 		if (adapter->xdp_ring[i]->desc)
6649 			ixgbe_free_tx_resources(adapter->xdp_ring[i]);
6650 }
6651 
6652 /**
6653  * ixgbe_free_rx_resources - Free Rx Resources
6654  * @rx_ring: ring to clean the resources from
6655  *
6656  * Free all receive software resources
6657  **/
6658 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
6659 {
6660 	ixgbe_clean_rx_ring(rx_ring);
6661 
6662 	rx_ring->xdp_prog = NULL;
6663 	xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
6664 	vfree(rx_ring->rx_buffer_info);
6665 	rx_ring->rx_buffer_info = NULL;
6666 
6667 	/* if not set, then don't free */
6668 	if (!rx_ring->desc)
6669 		return;
6670 
6671 	dma_free_coherent(rx_ring->dev, rx_ring->size,
6672 			  rx_ring->desc, rx_ring->dma);
6673 
6674 	rx_ring->desc = NULL;
6675 }
6676 
6677 /**
6678  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
6679  * @adapter: board private structure
6680  *
6681  * Free all receive software resources
6682  **/
6683 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
6684 {
6685 	int i;
6686 
6687 #ifdef IXGBE_FCOE
6688 	ixgbe_free_fcoe_ddp_resources(adapter);
6689 
6690 #endif
6691 	for (i = 0; i < adapter->num_rx_queues; i++)
6692 		if (adapter->rx_ring[i]->desc)
6693 			ixgbe_free_rx_resources(adapter->rx_ring[i]);
6694 }
6695 
6696 /**
6697  * ixgbe_change_mtu - Change the Maximum Transfer Unit
6698  * @netdev: network interface device structure
6699  * @new_mtu: new value for maximum frame size
6700  *
6701  * Returns 0 on success, negative on failure
6702  **/
6703 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
6704 {
6705 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6706 
6707 	if (adapter->xdp_prog) {
6708 		int new_frame_size = new_mtu + ETH_HLEN + ETH_FCS_LEN +
6709 				     VLAN_HLEN;
6710 		int i;
6711 
6712 		for (i = 0; i < adapter->num_rx_queues; i++) {
6713 			struct ixgbe_ring *ring = adapter->rx_ring[i];
6714 
6715 			if (new_frame_size > ixgbe_rx_bufsz(ring)) {
6716 				e_warn(probe, "Requested MTU size is not supported with XDP\n");
6717 				return -EINVAL;
6718 			}
6719 		}
6720 	}
6721 
6722 	/*
6723 	 * For 82599EB we cannot allow legacy VFs to enable their receive
6724 	 * paths when MTU greater than 1500 is configured.  So display a
6725 	 * warning that legacy VFs will be disabled.
6726 	 */
6727 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
6728 	    (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
6729 	    (new_mtu > ETH_DATA_LEN))
6730 		e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
6731 
6732 	netdev_dbg(netdev, "changing MTU from %d to %d\n",
6733 		   netdev->mtu, new_mtu);
6734 
6735 	/* must set new MTU before calling down or up */
6736 	netdev->mtu = new_mtu;
6737 
6738 	if (netif_running(netdev))
6739 		ixgbe_reinit_locked(adapter);
6740 
6741 	return 0;
6742 }
6743 
6744 /**
6745  * ixgbe_open - Called when a network interface is made active
6746  * @netdev: network interface device structure
6747  *
6748  * Returns 0 on success, negative value on failure
6749  *
6750  * The open entry point is called when a network interface is made
6751  * active by the system (IFF_UP).  At this point all resources needed
6752  * for transmit and receive operations are allocated, the interrupt
6753  * handler is registered with the OS, the watchdog timer is started,
6754  * and the stack is notified that the interface is ready.
6755  **/
6756 int ixgbe_open(struct net_device *netdev)
6757 {
6758 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6759 	struct ixgbe_hw *hw = &adapter->hw;
6760 	int err, queues;
6761 
6762 	/* disallow open during test */
6763 	if (test_bit(__IXGBE_TESTING, &adapter->state))
6764 		return -EBUSY;
6765 
6766 	netif_carrier_off(netdev);
6767 
6768 	/* allocate transmit descriptors */
6769 	err = ixgbe_setup_all_tx_resources(adapter);
6770 	if (err)
6771 		goto err_setup_tx;
6772 
6773 	/* allocate receive descriptors */
6774 	err = ixgbe_setup_all_rx_resources(adapter);
6775 	if (err)
6776 		goto err_setup_rx;
6777 
6778 	ixgbe_configure(adapter);
6779 
6780 	err = ixgbe_request_irq(adapter);
6781 	if (err)
6782 		goto err_req_irq;
6783 
6784 	/* Notify the stack of the actual queue counts. */
6785 	queues = adapter->num_tx_queues;
6786 	err = netif_set_real_num_tx_queues(netdev, queues);
6787 	if (err)
6788 		goto err_set_queues;
6789 
6790 	queues = adapter->num_rx_queues;
6791 	err = netif_set_real_num_rx_queues(netdev, queues);
6792 	if (err)
6793 		goto err_set_queues;
6794 
6795 	ixgbe_ptp_init(adapter);
6796 
6797 	ixgbe_up_complete(adapter);
6798 
6799 	ixgbe_clear_udp_tunnel_port(adapter, IXGBE_VXLANCTRL_ALL_UDPPORT_MASK);
6800 	udp_tunnel_get_rx_info(netdev);
6801 
6802 	return 0;
6803 
6804 err_set_queues:
6805 	ixgbe_free_irq(adapter);
6806 err_req_irq:
6807 	ixgbe_free_all_rx_resources(adapter);
6808 	if (hw->phy.ops.set_phy_power && !adapter->wol)
6809 		hw->phy.ops.set_phy_power(&adapter->hw, false);
6810 err_setup_rx:
6811 	ixgbe_free_all_tx_resources(adapter);
6812 err_setup_tx:
6813 	ixgbe_reset(adapter);
6814 
6815 	return err;
6816 }
6817 
6818 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
6819 {
6820 	ixgbe_ptp_suspend(adapter);
6821 
6822 	if (adapter->hw.phy.ops.enter_lplu) {
6823 		adapter->hw.phy.reset_disable = true;
6824 		ixgbe_down(adapter);
6825 		adapter->hw.phy.ops.enter_lplu(&adapter->hw);
6826 		adapter->hw.phy.reset_disable = false;
6827 	} else {
6828 		ixgbe_down(adapter);
6829 	}
6830 
6831 	ixgbe_free_irq(adapter);
6832 
6833 	ixgbe_free_all_tx_resources(adapter);
6834 	ixgbe_free_all_rx_resources(adapter);
6835 }
6836 
6837 /**
6838  * ixgbe_close - Disables a network interface
6839  * @netdev: network interface device structure
6840  *
6841  * Returns 0, this is not allowed to fail
6842  *
6843  * The close entry point is called when an interface is de-activated
6844  * by the OS.  The hardware is still under the drivers control, but
6845  * needs to be disabled.  A global MAC reset is issued to stop the
6846  * hardware, and all transmit and receive resources are freed.
6847  **/
6848 int ixgbe_close(struct net_device *netdev)
6849 {
6850 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6851 
6852 	ixgbe_ptp_stop(adapter);
6853 
6854 	if (netif_device_present(netdev))
6855 		ixgbe_close_suspend(adapter);
6856 
6857 	ixgbe_fdir_filter_exit(adapter);
6858 
6859 	ixgbe_release_hw_control(adapter);
6860 
6861 	return 0;
6862 }
6863 
6864 #ifdef CONFIG_PM
6865 static int ixgbe_resume(struct pci_dev *pdev)
6866 {
6867 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6868 	struct net_device *netdev = adapter->netdev;
6869 	u32 err;
6870 
6871 	adapter->hw.hw_addr = adapter->io_addr;
6872 	pci_set_power_state(pdev, PCI_D0);
6873 	pci_restore_state(pdev);
6874 	/*
6875 	 * pci_restore_state clears dev->state_saved so call
6876 	 * pci_save_state to restore it.
6877 	 */
6878 	pci_save_state(pdev);
6879 
6880 	err = pci_enable_device_mem(pdev);
6881 	if (err) {
6882 		e_dev_err("Cannot enable PCI device from suspend\n");
6883 		return err;
6884 	}
6885 	smp_mb__before_atomic();
6886 	clear_bit(__IXGBE_DISABLED, &adapter->state);
6887 	pci_set_master(pdev);
6888 
6889 	pci_wake_from_d3(pdev, false);
6890 
6891 	ixgbe_reset(adapter);
6892 
6893 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6894 
6895 	rtnl_lock();
6896 	err = ixgbe_init_interrupt_scheme(adapter);
6897 	if (!err && netif_running(netdev))
6898 		err = ixgbe_open(netdev);
6899 
6900 
6901 	if (!err)
6902 		netif_device_attach(netdev);
6903 	rtnl_unlock();
6904 
6905 	return err;
6906 }
6907 #endif /* CONFIG_PM */
6908 
6909 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
6910 {
6911 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6912 	struct net_device *netdev = adapter->netdev;
6913 	struct ixgbe_hw *hw = &adapter->hw;
6914 	u32 ctrl;
6915 	u32 wufc = adapter->wol;
6916 #ifdef CONFIG_PM
6917 	int retval = 0;
6918 #endif
6919 
6920 	rtnl_lock();
6921 	netif_device_detach(netdev);
6922 
6923 	if (netif_running(netdev))
6924 		ixgbe_close_suspend(adapter);
6925 
6926 	ixgbe_clear_interrupt_scheme(adapter);
6927 	rtnl_unlock();
6928 
6929 #ifdef CONFIG_PM
6930 	retval = pci_save_state(pdev);
6931 	if (retval)
6932 		return retval;
6933 
6934 #endif
6935 	if (hw->mac.ops.stop_link_on_d3)
6936 		hw->mac.ops.stop_link_on_d3(hw);
6937 
6938 	if (wufc) {
6939 		u32 fctrl;
6940 
6941 		ixgbe_set_rx_mode(netdev);
6942 
6943 		/* enable the optics for 82599 SFP+ fiber as we can WoL */
6944 		if (hw->mac.ops.enable_tx_laser)
6945 			hw->mac.ops.enable_tx_laser(hw);
6946 
6947 		/* enable the reception of multicast packets */
6948 		fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6949 		fctrl |= IXGBE_FCTRL_MPE;
6950 		IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
6951 
6952 		ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
6953 		ctrl |= IXGBE_CTRL_GIO_DIS;
6954 		IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
6955 
6956 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
6957 	} else {
6958 		IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
6959 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
6960 	}
6961 
6962 	switch (hw->mac.type) {
6963 	case ixgbe_mac_82598EB:
6964 		pci_wake_from_d3(pdev, false);
6965 		break;
6966 	case ixgbe_mac_82599EB:
6967 	case ixgbe_mac_X540:
6968 	case ixgbe_mac_X550:
6969 	case ixgbe_mac_X550EM_x:
6970 	case ixgbe_mac_x550em_a:
6971 		pci_wake_from_d3(pdev, !!wufc);
6972 		break;
6973 	default:
6974 		break;
6975 	}
6976 
6977 	*enable_wake = !!wufc;
6978 	if (hw->phy.ops.set_phy_power && !*enable_wake)
6979 		hw->phy.ops.set_phy_power(hw, false);
6980 
6981 	ixgbe_release_hw_control(adapter);
6982 
6983 	if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
6984 		pci_disable_device(pdev);
6985 
6986 	return 0;
6987 }
6988 
6989 #ifdef CONFIG_PM
6990 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
6991 {
6992 	int retval;
6993 	bool wake;
6994 
6995 	retval = __ixgbe_shutdown(pdev, &wake);
6996 	if (retval)
6997 		return retval;
6998 
6999 	if (wake) {
7000 		pci_prepare_to_sleep(pdev);
7001 	} else {
7002 		pci_wake_from_d3(pdev, false);
7003 		pci_set_power_state(pdev, PCI_D3hot);
7004 	}
7005 
7006 	return 0;
7007 }
7008 #endif /* CONFIG_PM */
7009 
7010 static void ixgbe_shutdown(struct pci_dev *pdev)
7011 {
7012 	bool wake;
7013 
7014 	__ixgbe_shutdown(pdev, &wake);
7015 
7016 	if (system_state == SYSTEM_POWER_OFF) {
7017 		pci_wake_from_d3(pdev, wake);
7018 		pci_set_power_state(pdev, PCI_D3hot);
7019 	}
7020 }
7021 
7022 /**
7023  * ixgbe_update_stats - Update the board statistics counters.
7024  * @adapter: board private structure
7025  **/
7026 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
7027 {
7028 	struct net_device *netdev = adapter->netdev;
7029 	struct ixgbe_hw *hw = &adapter->hw;
7030 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
7031 	u64 total_mpc = 0;
7032 	u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
7033 	u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
7034 	u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
7035 	u64 alloc_rx_page = 0;
7036 	u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
7037 
7038 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7039 	    test_bit(__IXGBE_RESETTING, &adapter->state))
7040 		return;
7041 
7042 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
7043 		u64 rsc_count = 0;
7044 		u64 rsc_flush = 0;
7045 		for (i = 0; i < adapter->num_rx_queues; i++) {
7046 			rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
7047 			rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
7048 		}
7049 		adapter->rsc_total_count = rsc_count;
7050 		adapter->rsc_total_flush = rsc_flush;
7051 	}
7052 
7053 	for (i = 0; i < adapter->num_rx_queues; i++) {
7054 		struct ixgbe_ring *rx_ring = READ_ONCE(adapter->rx_ring[i]);
7055 
7056 		if (!rx_ring)
7057 			continue;
7058 		non_eop_descs += rx_ring->rx_stats.non_eop_descs;
7059 		alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
7060 		alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
7061 		alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
7062 		hw_csum_rx_error += rx_ring->rx_stats.csum_err;
7063 		bytes += rx_ring->stats.bytes;
7064 		packets += rx_ring->stats.packets;
7065 	}
7066 	adapter->non_eop_descs = non_eop_descs;
7067 	adapter->alloc_rx_page = alloc_rx_page;
7068 	adapter->alloc_rx_page_failed = alloc_rx_page_failed;
7069 	adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
7070 	adapter->hw_csum_rx_error = hw_csum_rx_error;
7071 	netdev->stats.rx_bytes = bytes;
7072 	netdev->stats.rx_packets = packets;
7073 
7074 	bytes = 0;
7075 	packets = 0;
7076 	/* gather some stats to the adapter struct that are per queue */
7077 	for (i = 0; i < adapter->num_tx_queues; i++) {
7078 		struct ixgbe_ring *tx_ring = READ_ONCE(adapter->tx_ring[i]);
7079 
7080 		if (!tx_ring)
7081 			continue;
7082 		restart_queue += tx_ring->tx_stats.restart_queue;
7083 		tx_busy += tx_ring->tx_stats.tx_busy;
7084 		bytes += tx_ring->stats.bytes;
7085 		packets += tx_ring->stats.packets;
7086 	}
7087 	for (i = 0; i < adapter->num_xdp_queues; i++) {
7088 		struct ixgbe_ring *xdp_ring = READ_ONCE(adapter->xdp_ring[i]);
7089 
7090 		if (!xdp_ring)
7091 			continue;
7092 		restart_queue += xdp_ring->tx_stats.restart_queue;
7093 		tx_busy += xdp_ring->tx_stats.tx_busy;
7094 		bytes += xdp_ring->stats.bytes;
7095 		packets += xdp_ring->stats.packets;
7096 	}
7097 	adapter->restart_queue = restart_queue;
7098 	adapter->tx_busy = tx_busy;
7099 	netdev->stats.tx_bytes = bytes;
7100 	netdev->stats.tx_packets = packets;
7101 
7102 	hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
7103 
7104 	/* 8 register reads */
7105 	for (i = 0; i < 8; i++) {
7106 		/* for packet buffers not used, the register should read 0 */
7107 		mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
7108 		missed_rx += mpc;
7109 		hwstats->mpc[i] += mpc;
7110 		total_mpc += hwstats->mpc[i];
7111 		hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
7112 		hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
7113 		switch (hw->mac.type) {
7114 		case ixgbe_mac_82598EB:
7115 			hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
7116 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
7117 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
7118 			hwstats->pxonrxc[i] +=
7119 				IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
7120 			break;
7121 		case ixgbe_mac_82599EB:
7122 		case ixgbe_mac_X540:
7123 		case ixgbe_mac_X550:
7124 		case ixgbe_mac_X550EM_x:
7125 		case ixgbe_mac_x550em_a:
7126 			hwstats->pxonrxc[i] +=
7127 				IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
7128 			break;
7129 		default:
7130 			break;
7131 		}
7132 	}
7133 
7134 	/*16 register reads */
7135 	for (i = 0; i < 16; i++) {
7136 		hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
7137 		hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
7138 		if ((hw->mac.type == ixgbe_mac_82599EB) ||
7139 		    (hw->mac.type == ixgbe_mac_X540) ||
7140 		    (hw->mac.type == ixgbe_mac_X550) ||
7141 		    (hw->mac.type == ixgbe_mac_X550EM_x) ||
7142 		    (hw->mac.type == ixgbe_mac_x550em_a)) {
7143 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
7144 			IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
7145 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
7146 			IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
7147 		}
7148 	}
7149 
7150 	hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
7151 	/* work around hardware counting issue */
7152 	hwstats->gprc -= missed_rx;
7153 
7154 	ixgbe_update_xoff_received(adapter);
7155 
7156 	/* 82598 hardware only has a 32 bit counter in the high register */
7157 	switch (hw->mac.type) {
7158 	case ixgbe_mac_82598EB:
7159 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
7160 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
7161 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
7162 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
7163 		break;
7164 	case ixgbe_mac_X540:
7165 	case ixgbe_mac_X550:
7166 	case ixgbe_mac_X550EM_x:
7167 	case ixgbe_mac_x550em_a:
7168 		/* OS2BMC stats are X540 and later */
7169 		hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
7170 		hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
7171 		hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
7172 		hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
7173 		/* fall through */
7174 	case ixgbe_mac_82599EB:
7175 		for (i = 0; i < 16; i++)
7176 			adapter->hw_rx_no_dma_resources +=
7177 					     IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
7178 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
7179 		IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
7180 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
7181 		IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
7182 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
7183 		IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
7184 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
7185 		hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
7186 		hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
7187 #ifdef IXGBE_FCOE
7188 		hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
7189 		hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
7190 		hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
7191 		hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
7192 		hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
7193 		hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
7194 		/* Add up per cpu counters for total ddp aloc fail */
7195 		if (adapter->fcoe.ddp_pool) {
7196 			struct ixgbe_fcoe *fcoe = &adapter->fcoe;
7197 			struct ixgbe_fcoe_ddp_pool *ddp_pool;
7198 			unsigned int cpu;
7199 			u64 noddp = 0, noddp_ext_buff = 0;
7200 			for_each_possible_cpu(cpu) {
7201 				ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
7202 				noddp += ddp_pool->noddp;
7203 				noddp_ext_buff += ddp_pool->noddp_ext_buff;
7204 			}
7205 			hwstats->fcoe_noddp = noddp;
7206 			hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
7207 		}
7208 #endif /* IXGBE_FCOE */
7209 		break;
7210 	default:
7211 		break;
7212 	}
7213 	bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
7214 	hwstats->bprc += bprc;
7215 	hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
7216 	if (hw->mac.type == ixgbe_mac_82598EB)
7217 		hwstats->mprc -= bprc;
7218 	hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
7219 	hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
7220 	hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
7221 	hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
7222 	hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
7223 	hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
7224 	hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
7225 	hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
7226 	lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
7227 	hwstats->lxontxc += lxon;
7228 	lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
7229 	hwstats->lxofftxc += lxoff;
7230 	hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
7231 	hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
7232 	/*
7233 	 * 82598 errata - tx of flow control packets is included in tx counters
7234 	 */
7235 	xon_off_tot = lxon + lxoff;
7236 	hwstats->gptc -= xon_off_tot;
7237 	hwstats->mptc -= xon_off_tot;
7238 	hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
7239 	hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
7240 	hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
7241 	hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
7242 	hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
7243 	hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
7244 	hwstats->ptc64 -= xon_off_tot;
7245 	hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
7246 	hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
7247 	hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
7248 	hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
7249 	hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
7250 	hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
7251 
7252 	/* Fill out the OS statistics structure */
7253 	netdev->stats.multicast = hwstats->mprc;
7254 
7255 	/* Rx Errors */
7256 	netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
7257 	netdev->stats.rx_dropped = 0;
7258 	netdev->stats.rx_length_errors = hwstats->rlec;
7259 	netdev->stats.rx_crc_errors = hwstats->crcerrs;
7260 	netdev->stats.rx_missed_errors = total_mpc;
7261 }
7262 
7263 /**
7264  * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
7265  * @adapter: pointer to the device adapter structure
7266  **/
7267 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
7268 {
7269 	struct ixgbe_hw *hw = &adapter->hw;
7270 	int i;
7271 
7272 	if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
7273 		return;
7274 
7275 	adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
7276 
7277 	/* if interface is down do nothing */
7278 	if (test_bit(__IXGBE_DOWN, &adapter->state))
7279 		return;
7280 
7281 	/* do nothing if we are not using signature filters */
7282 	if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
7283 		return;
7284 
7285 	adapter->fdir_overflow++;
7286 
7287 	if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
7288 		for (i = 0; i < adapter->num_tx_queues; i++)
7289 			set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7290 				&(adapter->tx_ring[i]->state));
7291 		for (i = 0; i < adapter->num_xdp_queues; i++)
7292 			set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7293 				&adapter->xdp_ring[i]->state);
7294 		/* re-enable flow director interrupts */
7295 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
7296 	} else {
7297 		e_err(probe, "failed to finish FDIR re-initialization, "
7298 		      "ignored adding FDIR ATR filters\n");
7299 	}
7300 }
7301 
7302 /**
7303  * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
7304  * @adapter: pointer to the device adapter structure
7305  *
7306  * This function serves two purposes.  First it strobes the interrupt lines
7307  * in order to make certain interrupts are occurring.  Secondly it sets the
7308  * bits needed to check for TX hangs.  As a result we should immediately
7309  * determine if a hang has occurred.
7310  */
7311 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
7312 {
7313 	struct ixgbe_hw *hw = &adapter->hw;
7314 	u64 eics = 0;
7315 	int i;
7316 
7317 	/* If we're down, removing or resetting, just bail */
7318 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7319 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
7320 	    test_bit(__IXGBE_RESETTING, &adapter->state))
7321 		return;
7322 
7323 	/* Force detection of hung controller */
7324 	if (netif_carrier_ok(adapter->netdev)) {
7325 		for (i = 0; i < adapter->num_tx_queues; i++)
7326 			set_check_for_tx_hang(adapter->tx_ring[i]);
7327 		for (i = 0; i < adapter->num_xdp_queues; i++)
7328 			set_check_for_tx_hang(adapter->xdp_ring[i]);
7329 	}
7330 
7331 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
7332 		/*
7333 		 * for legacy and MSI interrupts don't set any bits
7334 		 * that are enabled for EIAM, because this operation
7335 		 * would set *both* EIMS and EICS for any bit in EIAM
7336 		 */
7337 		IXGBE_WRITE_REG(hw, IXGBE_EICS,
7338 			(IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
7339 	} else {
7340 		/* get one bit for every active tx/rx interrupt vector */
7341 		for (i = 0; i < adapter->num_q_vectors; i++) {
7342 			struct ixgbe_q_vector *qv = adapter->q_vector[i];
7343 			if (qv->rx.ring || qv->tx.ring)
7344 				eics |= BIT_ULL(i);
7345 		}
7346 	}
7347 
7348 	/* Cause software interrupt to ensure rings are cleaned */
7349 	ixgbe_irq_rearm_queues(adapter, eics);
7350 }
7351 
7352 /**
7353  * ixgbe_watchdog_update_link - update the link status
7354  * @adapter: pointer to the device adapter structure
7355  **/
7356 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
7357 {
7358 	struct ixgbe_hw *hw = &adapter->hw;
7359 	u32 link_speed = adapter->link_speed;
7360 	bool link_up = adapter->link_up;
7361 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
7362 
7363 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
7364 		return;
7365 
7366 	if (hw->mac.ops.check_link) {
7367 		hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
7368 	} else {
7369 		/* always assume link is up, if no check link function */
7370 		link_speed = IXGBE_LINK_SPEED_10GB_FULL;
7371 		link_up = true;
7372 	}
7373 
7374 	if (adapter->ixgbe_ieee_pfc)
7375 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
7376 
7377 	if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
7378 		hw->mac.ops.fc_enable(hw);
7379 		ixgbe_set_rx_drop_en(adapter);
7380 	}
7381 
7382 	if (link_up ||
7383 	    time_after(jiffies, (adapter->link_check_timeout +
7384 				 IXGBE_TRY_LINK_TIMEOUT))) {
7385 		adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
7386 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
7387 		IXGBE_WRITE_FLUSH(hw);
7388 	}
7389 
7390 	adapter->link_up = link_up;
7391 	adapter->link_speed = link_speed;
7392 }
7393 
7394 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
7395 {
7396 #ifdef CONFIG_IXGBE_DCB
7397 	struct net_device *netdev = adapter->netdev;
7398 	struct dcb_app app = {
7399 			      .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
7400 			      .protocol = 0,
7401 			     };
7402 	u8 up = 0;
7403 
7404 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
7405 		up = dcb_ieee_getapp_mask(netdev, &app);
7406 
7407 	adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
7408 #endif
7409 }
7410 
7411 /**
7412  * ixgbe_watchdog_link_is_up - update netif_carrier status and
7413  *                             print link up message
7414  * @adapter: pointer to the device adapter structure
7415  **/
7416 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
7417 {
7418 	struct net_device *netdev = adapter->netdev;
7419 	struct ixgbe_hw *hw = &adapter->hw;
7420 	u32 link_speed = adapter->link_speed;
7421 	const char *speed_str;
7422 	bool flow_rx, flow_tx;
7423 
7424 	/* only continue if link was previously down */
7425 	if (netif_carrier_ok(netdev))
7426 		return;
7427 
7428 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7429 
7430 	switch (hw->mac.type) {
7431 	case ixgbe_mac_82598EB: {
7432 		u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
7433 		u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
7434 		flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
7435 		flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
7436 	}
7437 		break;
7438 	case ixgbe_mac_X540:
7439 	case ixgbe_mac_X550:
7440 	case ixgbe_mac_X550EM_x:
7441 	case ixgbe_mac_x550em_a:
7442 	case ixgbe_mac_82599EB: {
7443 		u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
7444 		u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
7445 		flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
7446 		flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
7447 	}
7448 		break;
7449 	default:
7450 		flow_tx = false;
7451 		flow_rx = false;
7452 		break;
7453 	}
7454 
7455 	adapter->last_rx_ptp_check = jiffies;
7456 
7457 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7458 		ixgbe_ptp_start_cyclecounter(adapter);
7459 
7460 	switch (link_speed) {
7461 	case IXGBE_LINK_SPEED_10GB_FULL:
7462 		speed_str = "10 Gbps";
7463 		break;
7464 	case IXGBE_LINK_SPEED_5GB_FULL:
7465 		speed_str = "5 Gbps";
7466 		break;
7467 	case IXGBE_LINK_SPEED_2_5GB_FULL:
7468 		speed_str = "2.5 Gbps";
7469 		break;
7470 	case IXGBE_LINK_SPEED_1GB_FULL:
7471 		speed_str = "1 Gbps";
7472 		break;
7473 	case IXGBE_LINK_SPEED_100_FULL:
7474 		speed_str = "100 Mbps";
7475 		break;
7476 	case IXGBE_LINK_SPEED_10_FULL:
7477 		speed_str = "10 Mbps";
7478 		break;
7479 	default:
7480 		speed_str = "unknown speed";
7481 		break;
7482 	}
7483 	e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str,
7484 	       ((flow_rx && flow_tx) ? "RX/TX" :
7485 	       (flow_rx ? "RX" :
7486 	       (flow_tx ? "TX" : "None"))));
7487 
7488 	netif_carrier_on(netdev);
7489 	ixgbe_check_vf_rate_limit(adapter);
7490 
7491 	/* enable transmits */
7492 	netif_tx_wake_all_queues(adapter->netdev);
7493 
7494 	/* update the default user priority for VFs */
7495 	ixgbe_update_default_up(adapter);
7496 
7497 	/* ping all the active vfs to let them know link has changed */
7498 	ixgbe_ping_all_vfs(adapter);
7499 }
7500 
7501 /**
7502  * ixgbe_watchdog_link_is_down - update netif_carrier status and
7503  *                               print link down message
7504  * @adapter: pointer to the adapter structure
7505  **/
7506 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
7507 {
7508 	struct net_device *netdev = adapter->netdev;
7509 	struct ixgbe_hw *hw = &adapter->hw;
7510 
7511 	adapter->link_up = false;
7512 	adapter->link_speed = 0;
7513 
7514 	/* only continue if link was up previously */
7515 	if (!netif_carrier_ok(netdev))
7516 		return;
7517 
7518 	/* poll for SFP+ cable when link is down */
7519 	if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
7520 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
7521 
7522 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7523 		ixgbe_ptp_start_cyclecounter(adapter);
7524 
7525 	e_info(drv, "NIC Link is Down\n");
7526 	netif_carrier_off(netdev);
7527 
7528 	/* ping all the active vfs to let them know link has changed */
7529 	ixgbe_ping_all_vfs(adapter);
7530 }
7531 
7532 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
7533 {
7534 	int i;
7535 
7536 	for (i = 0; i < adapter->num_tx_queues; i++) {
7537 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7538 
7539 		if (tx_ring->next_to_use != tx_ring->next_to_clean)
7540 			return true;
7541 	}
7542 
7543 	for (i = 0; i < adapter->num_xdp_queues; i++) {
7544 		struct ixgbe_ring *ring = adapter->xdp_ring[i];
7545 
7546 		if (ring->next_to_use != ring->next_to_clean)
7547 			return true;
7548 	}
7549 
7550 	return false;
7551 }
7552 
7553 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
7554 {
7555 	struct ixgbe_hw *hw = &adapter->hw;
7556 	struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
7557 	u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
7558 
7559 	int i, j;
7560 
7561 	if (!adapter->num_vfs)
7562 		return false;
7563 
7564 	/* resetting the PF is only needed for MAC before X550 */
7565 	if (hw->mac.type >= ixgbe_mac_X550)
7566 		return false;
7567 
7568 	for (i = 0; i < adapter->num_vfs; i++) {
7569 		for (j = 0; j < q_per_pool; j++) {
7570 			u32 h, t;
7571 
7572 			h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
7573 			t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
7574 
7575 			if (h != t)
7576 				return true;
7577 		}
7578 	}
7579 
7580 	return false;
7581 }
7582 
7583 /**
7584  * ixgbe_watchdog_flush_tx - flush queues on link down
7585  * @adapter: pointer to the device adapter structure
7586  **/
7587 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
7588 {
7589 	if (!netif_carrier_ok(adapter->netdev)) {
7590 		if (ixgbe_ring_tx_pending(adapter) ||
7591 		    ixgbe_vf_tx_pending(adapter)) {
7592 			/* We've lost link, so the controller stops DMA,
7593 			 * but we've got queued Tx work that's never going
7594 			 * to get done, so reset controller to flush Tx.
7595 			 * (Do the reset outside of interrupt context).
7596 			 */
7597 			e_warn(drv, "initiating reset to clear Tx work after link loss\n");
7598 			set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
7599 		}
7600 	}
7601 }
7602 
7603 #ifdef CONFIG_PCI_IOV
7604 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
7605 {
7606 	struct ixgbe_hw *hw = &adapter->hw;
7607 	struct pci_dev *pdev = adapter->pdev;
7608 	unsigned int vf;
7609 	u32 gpc;
7610 
7611 	if (!(netif_carrier_ok(adapter->netdev)))
7612 		return;
7613 
7614 	gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
7615 	if (gpc) /* If incrementing then no need for the check below */
7616 		return;
7617 	/* Check to see if a bad DMA write target from an errant or
7618 	 * malicious VF has caused a PCIe error.  If so then we can
7619 	 * issue a VFLR to the offending VF(s) and then resume without
7620 	 * requesting a full slot reset.
7621 	 */
7622 
7623 	if (!pdev)
7624 		return;
7625 
7626 	/* check status reg for all VFs owned by this PF */
7627 	for (vf = 0; vf < adapter->num_vfs; ++vf) {
7628 		struct pci_dev *vfdev = adapter->vfinfo[vf].vfdev;
7629 		u16 status_reg;
7630 
7631 		if (!vfdev)
7632 			continue;
7633 		pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
7634 		if (status_reg != IXGBE_FAILED_READ_CFG_WORD &&
7635 		    status_reg & PCI_STATUS_REC_MASTER_ABORT)
7636 			pcie_flr(vfdev);
7637 	}
7638 }
7639 
7640 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
7641 {
7642 	u32 ssvpc;
7643 
7644 	/* Do not perform spoof check for 82598 or if not in IOV mode */
7645 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7646 	    adapter->num_vfs == 0)
7647 		return;
7648 
7649 	ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
7650 
7651 	/*
7652 	 * ssvpc register is cleared on read, if zero then no
7653 	 * spoofed packets in the last interval.
7654 	 */
7655 	if (!ssvpc)
7656 		return;
7657 
7658 	e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
7659 }
7660 #else
7661 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
7662 {
7663 }
7664 
7665 static void
7666 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
7667 {
7668 }
7669 #endif /* CONFIG_PCI_IOV */
7670 
7671 
7672 /**
7673  * ixgbe_watchdog_subtask - check and bring link up
7674  * @adapter: pointer to the device adapter structure
7675  **/
7676 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
7677 {
7678 	/* if interface is down, removing or resetting, do nothing */
7679 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7680 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
7681 	    test_bit(__IXGBE_RESETTING, &adapter->state))
7682 		return;
7683 
7684 	ixgbe_watchdog_update_link(adapter);
7685 
7686 	if (adapter->link_up)
7687 		ixgbe_watchdog_link_is_up(adapter);
7688 	else
7689 		ixgbe_watchdog_link_is_down(adapter);
7690 
7691 	ixgbe_check_for_bad_vf(adapter);
7692 	ixgbe_spoof_check(adapter);
7693 	ixgbe_update_stats(adapter);
7694 
7695 	ixgbe_watchdog_flush_tx(adapter);
7696 }
7697 
7698 /**
7699  * ixgbe_sfp_detection_subtask - poll for SFP+ cable
7700  * @adapter: the ixgbe adapter structure
7701  **/
7702 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
7703 {
7704 	struct ixgbe_hw *hw = &adapter->hw;
7705 	s32 err;
7706 
7707 	/* not searching for SFP so there is nothing to do here */
7708 	if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
7709 	    !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7710 		return;
7711 
7712 	if (adapter->sfp_poll_time &&
7713 	    time_after(adapter->sfp_poll_time, jiffies))
7714 		return; /* If not yet time to poll for SFP */
7715 
7716 	/* someone else is in init, wait until next service event */
7717 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7718 		return;
7719 
7720 	adapter->sfp_poll_time = jiffies + IXGBE_SFP_POLL_JIFFIES - 1;
7721 
7722 	err = hw->phy.ops.identify_sfp(hw);
7723 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7724 		goto sfp_out;
7725 
7726 	if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
7727 		/* If no cable is present, then we need to reset
7728 		 * the next time we find a good cable. */
7729 		adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
7730 	}
7731 
7732 	/* exit on error */
7733 	if (err)
7734 		goto sfp_out;
7735 
7736 	/* exit if reset not needed */
7737 	if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7738 		goto sfp_out;
7739 
7740 	adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
7741 
7742 	/*
7743 	 * A module may be identified correctly, but the EEPROM may not have
7744 	 * support for that module.  setup_sfp() will fail in that case, so
7745 	 * we should not allow that module to load.
7746 	 */
7747 	if (hw->mac.type == ixgbe_mac_82598EB)
7748 		err = hw->phy.ops.reset(hw);
7749 	else
7750 		err = hw->mac.ops.setup_sfp(hw);
7751 
7752 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7753 		goto sfp_out;
7754 
7755 	adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
7756 	e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
7757 
7758 sfp_out:
7759 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7760 
7761 	if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
7762 	    (adapter->netdev->reg_state == NETREG_REGISTERED)) {
7763 		e_dev_err("failed to initialize because an unsupported "
7764 			  "SFP+ module type was detected.\n");
7765 		e_dev_err("Reload the driver after installing a "
7766 			  "supported module.\n");
7767 		unregister_netdev(adapter->netdev);
7768 	}
7769 }
7770 
7771 /**
7772  * ixgbe_sfp_link_config_subtask - set up link SFP after module install
7773  * @adapter: the ixgbe adapter structure
7774  **/
7775 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
7776 {
7777 	struct ixgbe_hw *hw = &adapter->hw;
7778 	u32 cap_speed;
7779 	u32 speed;
7780 	bool autoneg = false;
7781 
7782 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
7783 		return;
7784 
7785 	/* someone else is in init, wait until next service event */
7786 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7787 		return;
7788 
7789 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
7790 
7791 	hw->mac.ops.get_link_capabilities(hw, &cap_speed, &autoneg);
7792 
7793 	/* advertise highest capable link speed */
7794 	if (!autoneg && (cap_speed & IXGBE_LINK_SPEED_10GB_FULL))
7795 		speed = IXGBE_LINK_SPEED_10GB_FULL;
7796 	else
7797 		speed = cap_speed & (IXGBE_LINK_SPEED_10GB_FULL |
7798 				     IXGBE_LINK_SPEED_1GB_FULL);
7799 
7800 	if (hw->mac.ops.setup_link)
7801 		hw->mac.ops.setup_link(hw, speed, true);
7802 
7803 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
7804 	adapter->link_check_timeout = jiffies;
7805 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7806 }
7807 
7808 /**
7809  * ixgbe_service_timer - Timer Call-back
7810  * @t: pointer to timer_list structure
7811  **/
7812 static void ixgbe_service_timer(struct timer_list *t)
7813 {
7814 	struct ixgbe_adapter *adapter = from_timer(adapter, t, service_timer);
7815 	unsigned long next_event_offset;
7816 
7817 	/* poll faster when waiting for link */
7818 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
7819 		next_event_offset = HZ / 10;
7820 	else
7821 		next_event_offset = HZ * 2;
7822 
7823 	/* Reset the timer */
7824 	mod_timer(&adapter->service_timer, next_event_offset + jiffies);
7825 
7826 	ixgbe_service_event_schedule(adapter);
7827 }
7828 
7829 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter)
7830 {
7831 	struct ixgbe_hw *hw = &adapter->hw;
7832 	u32 status;
7833 
7834 	if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT))
7835 		return;
7836 
7837 	adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT;
7838 
7839 	if (!hw->phy.ops.handle_lasi)
7840 		return;
7841 
7842 	status = hw->phy.ops.handle_lasi(&adapter->hw);
7843 	if (status != IXGBE_ERR_OVERTEMP)
7844 		return;
7845 
7846 	e_crit(drv, "%s\n", ixgbe_overheat_msg);
7847 }
7848 
7849 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
7850 {
7851 	if (!test_and_clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state))
7852 		return;
7853 
7854 	rtnl_lock();
7855 	/* If we're already down, removing or resetting, just bail */
7856 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7857 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
7858 	    test_bit(__IXGBE_RESETTING, &adapter->state)) {
7859 		rtnl_unlock();
7860 		return;
7861 	}
7862 
7863 	ixgbe_dump(adapter);
7864 	netdev_err(adapter->netdev, "Reset adapter\n");
7865 	adapter->tx_timeout_count++;
7866 
7867 	ixgbe_reinit_locked(adapter);
7868 	rtnl_unlock();
7869 }
7870 
7871 /**
7872  * ixgbe_check_fw_error - Check firmware for errors
7873  * @adapter: the adapter private structure
7874  *
7875  * Check firmware errors in register FWSM
7876  */
7877 static bool ixgbe_check_fw_error(struct ixgbe_adapter *adapter)
7878 {
7879 	struct ixgbe_hw *hw = &adapter->hw;
7880 	u32 fwsm;
7881 
7882 	/* read fwsm.ext_err_ind register and log errors */
7883 	fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
7884 
7885 	if (fwsm & IXGBE_FWSM_EXT_ERR_IND_MASK ||
7886 	    !(fwsm & IXGBE_FWSM_FW_VAL_BIT))
7887 		e_dev_warn("Warning firmware error detected FWSM: 0x%08X\n",
7888 			   fwsm);
7889 
7890 	if (hw->mac.ops.fw_recovery_mode && hw->mac.ops.fw_recovery_mode(hw)) {
7891 		e_dev_err("Firmware recovery mode detected. Limiting functionality. Refer to the Intel(R) Ethernet Adapters and Devices User Guide for details on firmware recovery mode.\n");
7892 		return true;
7893 	}
7894 
7895 	return false;
7896 }
7897 
7898 /**
7899  * ixgbe_service_task - manages and runs subtasks
7900  * @work: pointer to work_struct containing our data
7901  **/
7902 static void ixgbe_service_task(struct work_struct *work)
7903 {
7904 	struct ixgbe_adapter *adapter = container_of(work,
7905 						     struct ixgbe_adapter,
7906 						     service_task);
7907 	if (ixgbe_removed(adapter->hw.hw_addr)) {
7908 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
7909 			rtnl_lock();
7910 			ixgbe_down(adapter);
7911 			rtnl_unlock();
7912 		}
7913 		ixgbe_service_event_complete(adapter);
7914 		return;
7915 	}
7916 	if (ixgbe_check_fw_error(adapter)) {
7917 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
7918 			unregister_netdev(adapter->netdev);
7919 		ixgbe_service_event_complete(adapter);
7920 		return;
7921 	}
7922 	if (adapter->flags2 & IXGBE_FLAG2_UDP_TUN_REREG_NEEDED) {
7923 		rtnl_lock();
7924 		adapter->flags2 &= ~IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
7925 		udp_tunnel_get_rx_info(adapter->netdev);
7926 		rtnl_unlock();
7927 	}
7928 	ixgbe_reset_subtask(adapter);
7929 	ixgbe_phy_interrupt_subtask(adapter);
7930 	ixgbe_sfp_detection_subtask(adapter);
7931 	ixgbe_sfp_link_config_subtask(adapter);
7932 	ixgbe_check_overtemp_subtask(adapter);
7933 	ixgbe_watchdog_subtask(adapter);
7934 	ixgbe_fdir_reinit_subtask(adapter);
7935 	ixgbe_check_hang_subtask(adapter);
7936 
7937 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
7938 		ixgbe_ptp_overflow_check(adapter);
7939 		if (adapter->flags & IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER)
7940 			ixgbe_ptp_rx_hang(adapter);
7941 		ixgbe_ptp_tx_hang(adapter);
7942 	}
7943 
7944 	ixgbe_service_event_complete(adapter);
7945 }
7946 
7947 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
7948 		     struct ixgbe_tx_buffer *first,
7949 		     u8 *hdr_len,
7950 		     struct ixgbe_ipsec_tx_data *itd)
7951 {
7952 	u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
7953 	struct sk_buff *skb = first->skb;
7954 	union {
7955 		struct iphdr *v4;
7956 		struct ipv6hdr *v6;
7957 		unsigned char *hdr;
7958 	} ip;
7959 	union {
7960 		struct tcphdr *tcp;
7961 		struct udphdr *udp;
7962 		unsigned char *hdr;
7963 	} l4;
7964 	u32 paylen, l4_offset;
7965 	u32 fceof_saidx = 0;
7966 	int err;
7967 
7968 	if (skb->ip_summed != CHECKSUM_PARTIAL)
7969 		return 0;
7970 
7971 	if (!skb_is_gso(skb))
7972 		return 0;
7973 
7974 	err = skb_cow_head(skb, 0);
7975 	if (err < 0)
7976 		return err;
7977 
7978 	if (eth_p_mpls(first->protocol))
7979 		ip.hdr = skb_inner_network_header(skb);
7980 	else
7981 		ip.hdr = skb_network_header(skb);
7982 	l4.hdr = skb_checksum_start(skb);
7983 
7984 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
7985 	type_tucmd = (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ?
7986 		      IXGBE_ADVTXD_TUCMD_L4T_UDP : IXGBE_ADVTXD_TUCMD_L4T_TCP;
7987 
7988 	/* initialize outer IP header fields */
7989 	if (ip.v4->version == 4) {
7990 		unsigned char *csum_start = skb_checksum_start(skb);
7991 		unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
7992 		int len = csum_start - trans_start;
7993 
7994 		/* IP header will have to cancel out any data that
7995 		 * is not a part of the outer IP header, so set to
7996 		 * a reverse csum if needed, else init check to 0.
7997 		 */
7998 		ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
7999 					   csum_fold(csum_partial(trans_start,
8000 								  len, 0)) : 0;
8001 		type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
8002 
8003 		ip.v4->tot_len = 0;
8004 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
8005 				   IXGBE_TX_FLAGS_CSUM |
8006 				   IXGBE_TX_FLAGS_IPV4;
8007 	} else {
8008 		ip.v6->payload_len = 0;
8009 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
8010 				   IXGBE_TX_FLAGS_CSUM;
8011 	}
8012 
8013 	/* determine offset of inner transport header */
8014 	l4_offset = l4.hdr - skb->data;
8015 
8016 	/* remove payload length from inner checksum */
8017 	paylen = skb->len - l4_offset;
8018 
8019 	if (type_tucmd & IXGBE_ADVTXD_TUCMD_L4T_TCP) {
8020 		/* compute length of segmentation header */
8021 		*hdr_len = (l4.tcp->doff * 4) + l4_offset;
8022 		csum_replace_by_diff(&l4.tcp->check,
8023 				     (__force __wsum)htonl(paylen));
8024 	} else {
8025 		/* compute length of segmentation header */
8026 		*hdr_len = sizeof(*l4.udp) + l4_offset;
8027 		csum_replace_by_diff(&l4.udp->check,
8028 				     (__force __wsum)htonl(paylen));
8029 	}
8030 
8031 	/* update gso size and bytecount with header size */
8032 	first->gso_segs = skb_shinfo(skb)->gso_segs;
8033 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
8034 
8035 	/* mss_l4len_id: use 0 as index for TSO */
8036 	mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
8037 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
8038 
8039 	fceof_saidx |= itd->sa_idx;
8040 	type_tucmd |= itd->flags | itd->trailer_len;
8041 
8042 	/* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
8043 	vlan_macip_lens = l4.hdr - ip.hdr;
8044 	vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
8045 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8046 
8047 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
8048 			  mss_l4len_idx);
8049 
8050 	return 1;
8051 }
8052 
8053 static inline bool ixgbe_ipv6_csum_is_sctp(struct sk_buff *skb)
8054 {
8055 	unsigned int offset = 0;
8056 
8057 	ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
8058 
8059 	return offset == skb_checksum_start_offset(skb);
8060 }
8061 
8062 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
8063 			  struct ixgbe_tx_buffer *first,
8064 			  struct ixgbe_ipsec_tx_data *itd)
8065 {
8066 	struct sk_buff *skb = first->skb;
8067 	u32 vlan_macip_lens = 0;
8068 	u32 fceof_saidx = 0;
8069 	u32 type_tucmd = 0;
8070 
8071 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
8072 csum_failed:
8073 		if (!(first->tx_flags & (IXGBE_TX_FLAGS_HW_VLAN |
8074 					 IXGBE_TX_FLAGS_CC)))
8075 			return;
8076 		goto no_csum;
8077 	}
8078 
8079 	switch (skb->csum_offset) {
8080 	case offsetof(struct tcphdr, check):
8081 		type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
8082 		/* fall through */
8083 	case offsetof(struct udphdr, check):
8084 		break;
8085 	case offsetof(struct sctphdr, checksum):
8086 		/* validate that this is actually an SCTP request */
8087 		if (((first->protocol == htons(ETH_P_IP)) &&
8088 		     (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
8089 		    ((first->protocol == htons(ETH_P_IPV6)) &&
8090 		     ixgbe_ipv6_csum_is_sctp(skb))) {
8091 			type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
8092 			break;
8093 		}
8094 		/* fall through */
8095 	default:
8096 		skb_checksum_help(skb);
8097 		goto csum_failed;
8098 	}
8099 
8100 	/* update TX checksum flag */
8101 	first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
8102 	vlan_macip_lens = skb_checksum_start_offset(skb) -
8103 			  skb_network_offset(skb);
8104 no_csum:
8105 	/* vlan_macip_lens: MACLEN, VLAN tag */
8106 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
8107 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8108 
8109 	fceof_saidx |= itd->sa_idx;
8110 	type_tucmd |= itd->flags | itd->trailer_len;
8111 
8112 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd, 0);
8113 }
8114 
8115 #define IXGBE_SET_FLAG(_input, _flag, _result) \
8116 	((_flag <= _result) ? \
8117 	 ((u32)(_input & _flag) * (_result / _flag)) : \
8118 	 ((u32)(_input & _flag) / (_flag / _result)))
8119 
8120 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
8121 {
8122 	/* set type for advanced descriptor with frame checksum insertion */
8123 	u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8124 		       IXGBE_ADVTXD_DCMD_DEXT |
8125 		       IXGBE_ADVTXD_DCMD_IFCS;
8126 
8127 	/* set HW vlan bit if vlan is present */
8128 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
8129 				   IXGBE_ADVTXD_DCMD_VLE);
8130 
8131 	/* set segmentation enable bits for TSO/FSO */
8132 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
8133 				   IXGBE_ADVTXD_DCMD_TSE);
8134 
8135 	/* set timestamp bit if present */
8136 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
8137 				   IXGBE_ADVTXD_MAC_TSTAMP);
8138 
8139 	/* insert frame checksum */
8140 	cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
8141 
8142 	return cmd_type;
8143 }
8144 
8145 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
8146 				   u32 tx_flags, unsigned int paylen)
8147 {
8148 	u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
8149 
8150 	/* enable L4 checksum for TSO and TX checksum offload */
8151 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8152 					IXGBE_TX_FLAGS_CSUM,
8153 					IXGBE_ADVTXD_POPTS_TXSM);
8154 
8155 	/* enable IPv4 checksum for TSO */
8156 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8157 					IXGBE_TX_FLAGS_IPV4,
8158 					IXGBE_ADVTXD_POPTS_IXSM);
8159 
8160 	/* enable IPsec */
8161 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8162 					IXGBE_TX_FLAGS_IPSEC,
8163 					IXGBE_ADVTXD_POPTS_IPSEC);
8164 
8165 	/*
8166 	 * Check Context must be set if Tx switch is enabled, which it
8167 	 * always is for case where virtual functions are running
8168 	 */
8169 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8170 					IXGBE_TX_FLAGS_CC,
8171 					IXGBE_ADVTXD_CC);
8172 
8173 	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
8174 }
8175 
8176 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8177 {
8178 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
8179 
8180 	/* Herbert's original patch had:
8181 	 *  smp_mb__after_netif_stop_queue();
8182 	 * but since that doesn't exist yet, just open code it.
8183 	 */
8184 	smp_mb();
8185 
8186 	/* We need to check again in a case another CPU has just
8187 	 * made room available.
8188 	 */
8189 	if (likely(ixgbe_desc_unused(tx_ring) < size))
8190 		return -EBUSY;
8191 
8192 	/* A reprieve! - use start_queue because it doesn't call schedule */
8193 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
8194 	++tx_ring->tx_stats.restart_queue;
8195 	return 0;
8196 }
8197 
8198 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8199 {
8200 	if (likely(ixgbe_desc_unused(tx_ring) >= size))
8201 		return 0;
8202 
8203 	return __ixgbe_maybe_stop_tx(tx_ring, size);
8204 }
8205 
8206 static int ixgbe_tx_map(struct ixgbe_ring *tx_ring,
8207 			struct ixgbe_tx_buffer *first,
8208 			const u8 hdr_len)
8209 {
8210 	struct sk_buff *skb = first->skb;
8211 	struct ixgbe_tx_buffer *tx_buffer;
8212 	union ixgbe_adv_tx_desc *tx_desc;
8213 	skb_frag_t *frag;
8214 	dma_addr_t dma;
8215 	unsigned int data_len, size;
8216 	u32 tx_flags = first->tx_flags;
8217 	u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
8218 	u16 i = tx_ring->next_to_use;
8219 
8220 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
8221 
8222 	ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
8223 
8224 	size = skb_headlen(skb);
8225 	data_len = skb->data_len;
8226 
8227 #ifdef IXGBE_FCOE
8228 	if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
8229 		if (data_len < sizeof(struct fcoe_crc_eof)) {
8230 			size -= sizeof(struct fcoe_crc_eof) - data_len;
8231 			data_len = 0;
8232 		} else {
8233 			data_len -= sizeof(struct fcoe_crc_eof);
8234 		}
8235 	}
8236 
8237 #endif
8238 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
8239 
8240 	tx_buffer = first;
8241 
8242 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
8243 		if (dma_mapping_error(tx_ring->dev, dma))
8244 			goto dma_error;
8245 
8246 		/* record length, and DMA address */
8247 		dma_unmap_len_set(tx_buffer, len, size);
8248 		dma_unmap_addr_set(tx_buffer, dma, dma);
8249 
8250 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
8251 
8252 		while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
8253 			tx_desc->read.cmd_type_len =
8254 				cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
8255 
8256 			i++;
8257 			tx_desc++;
8258 			if (i == tx_ring->count) {
8259 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8260 				i = 0;
8261 			}
8262 			tx_desc->read.olinfo_status = 0;
8263 
8264 			dma += IXGBE_MAX_DATA_PER_TXD;
8265 			size -= IXGBE_MAX_DATA_PER_TXD;
8266 
8267 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
8268 		}
8269 
8270 		if (likely(!data_len))
8271 			break;
8272 
8273 		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
8274 
8275 		i++;
8276 		tx_desc++;
8277 		if (i == tx_ring->count) {
8278 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8279 			i = 0;
8280 		}
8281 		tx_desc->read.olinfo_status = 0;
8282 
8283 #ifdef IXGBE_FCOE
8284 		size = min_t(unsigned int, data_len, skb_frag_size(frag));
8285 #else
8286 		size = skb_frag_size(frag);
8287 #endif
8288 		data_len -= size;
8289 
8290 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
8291 				       DMA_TO_DEVICE);
8292 
8293 		tx_buffer = &tx_ring->tx_buffer_info[i];
8294 	}
8295 
8296 	/* write last descriptor with RS and EOP bits */
8297 	cmd_type |= size | IXGBE_TXD_CMD;
8298 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8299 
8300 	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
8301 
8302 	/* set the timestamp */
8303 	first->time_stamp = jiffies;
8304 
8305 	skb_tx_timestamp(skb);
8306 
8307 	/*
8308 	 * Force memory writes to complete before letting h/w know there
8309 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
8310 	 * memory model archs, such as IA-64).
8311 	 *
8312 	 * We also need this memory barrier to make certain all of the
8313 	 * status bits have been updated before next_to_watch is written.
8314 	 */
8315 	wmb();
8316 
8317 	/* set next_to_watch value indicating a packet is present */
8318 	first->next_to_watch = tx_desc;
8319 
8320 	i++;
8321 	if (i == tx_ring->count)
8322 		i = 0;
8323 
8324 	tx_ring->next_to_use = i;
8325 
8326 	ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
8327 
8328 	if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
8329 		writel(i, tx_ring->tail);
8330 	}
8331 
8332 	return 0;
8333 dma_error:
8334 	dev_err(tx_ring->dev, "TX DMA map failed\n");
8335 
8336 	/* clear dma mappings for failed tx_buffer_info map */
8337 	for (;;) {
8338 		tx_buffer = &tx_ring->tx_buffer_info[i];
8339 		if (dma_unmap_len(tx_buffer, len))
8340 			dma_unmap_page(tx_ring->dev,
8341 				       dma_unmap_addr(tx_buffer, dma),
8342 				       dma_unmap_len(tx_buffer, len),
8343 				       DMA_TO_DEVICE);
8344 		dma_unmap_len_set(tx_buffer, len, 0);
8345 		if (tx_buffer == first)
8346 			break;
8347 		if (i == 0)
8348 			i += tx_ring->count;
8349 		i--;
8350 	}
8351 
8352 	dev_kfree_skb_any(first->skb);
8353 	first->skb = NULL;
8354 
8355 	tx_ring->next_to_use = i;
8356 
8357 	return -1;
8358 }
8359 
8360 static void ixgbe_atr(struct ixgbe_ring *ring,
8361 		      struct ixgbe_tx_buffer *first)
8362 {
8363 	struct ixgbe_q_vector *q_vector = ring->q_vector;
8364 	union ixgbe_atr_hash_dword input = { .dword = 0 };
8365 	union ixgbe_atr_hash_dword common = { .dword = 0 };
8366 	union {
8367 		unsigned char *network;
8368 		struct iphdr *ipv4;
8369 		struct ipv6hdr *ipv6;
8370 	} hdr;
8371 	struct tcphdr *th;
8372 	unsigned int hlen;
8373 	struct sk_buff *skb;
8374 	__be16 vlan_id;
8375 	int l4_proto;
8376 
8377 	/* if ring doesn't have a interrupt vector, cannot perform ATR */
8378 	if (!q_vector)
8379 		return;
8380 
8381 	/* do nothing if sampling is disabled */
8382 	if (!ring->atr_sample_rate)
8383 		return;
8384 
8385 	ring->atr_count++;
8386 
8387 	/* currently only IPv4/IPv6 with TCP is supported */
8388 	if ((first->protocol != htons(ETH_P_IP)) &&
8389 	    (first->protocol != htons(ETH_P_IPV6)))
8390 		return;
8391 
8392 	/* snag network header to get L4 type and address */
8393 	skb = first->skb;
8394 	hdr.network = skb_network_header(skb);
8395 	if (unlikely(hdr.network <= skb->data))
8396 		return;
8397 	if (skb->encapsulation &&
8398 	    first->protocol == htons(ETH_P_IP) &&
8399 	    hdr.ipv4->protocol == IPPROTO_UDP) {
8400 		struct ixgbe_adapter *adapter = q_vector->adapter;
8401 
8402 		if (unlikely(skb_tail_pointer(skb) < hdr.network +
8403 			     VXLAN_HEADROOM))
8404 			return;
8405 
8406 		/* verify the port is recognized as VXLAN */
8407 		if (adapter->vxlan_port &&
8408 		    udp_hdr(skb)->dest == adapter->vxlan_port)
8409 			hdr.network = skb_inner_network_header(skb);
8410 
8411 		if (adapter->geneve_port &&
8412 		    udp_hdr(skb)->dest == adapter->geneve_port)
8413 			hdr.network = skb_inner_network_header(skb);
8414 	}
8415 
8416 	/* Make sure we have at least [minimum IPv4 header + TCP]
8417 	 * or [IPv6 header] bytes
8418 	 */
8419 	if (unlikely(skb_tail_pointer(skb) < hdr.network + 40))
8420 		return;
8421 
8422 	/* Currently only IPv4/IPv6 with TCP is supported */
8423 	switch (hdr.ipv4->version) {
8424 	case IPVERSION:
8425 		/* access ihl as u8 to avoid unaligned access on ia64 */
8426 		hlen = (hdr.network[0] & 0x0F) << 2;
8427 		l4_proto = hdr.ipv4->protocol;
8428 		break;
8429 	case 6:
8430 		hlen = hdr.network - skb->data;
8431 		l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
8432 		hlen -= hdr.network - skb->data;
8433 		break;
8434 	default:
8435 		return;
8436 	}
8437 
8438 	if (l4_proto != IPPROTO_TCP)
8439 		return;
8440 
8441 	if (unlikely(skb_tail_pointer(skb) < hdr.network +
8442 		     hlen + sizeof(struct tcphdr)))
8443 		return;
8444 
8445 	th = (struct tcphdr *)(hdr.network + hlen);
8446 
8447 	/* skip this packet since the socket is closing */
8448 	if (th->fin)
8449 		return;
8450 
8451 	/* sample on all syn packets or once every atr sample count */
8452 	if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
8453 		return;
8454 
8455 	/* reset sample count */
8456 	ring->atr_count = 0;
8457 
8458 	vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
8459 
8460 	/*
8461 	 * src and dst are inverted, think how the receiver sees them
8462 	 *
8463 	 * The input is broken into two sections, a non-compressed section
8464 	 * containing vm_pool, vlan_id, and flow_type.  The rest of the data
8465 	 * is XORed together and stored in the compressed dword.
8466 	 */
8467 	input.formatted.vlan_id = vlan_id;
8468 
8469 	/*
8470 	 * since src port and flex bytes occupy the same word XOR them together
8471 	 * and write the value to source port portion of compressed dword
8472 	 */
8473 	if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
8474 		common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
8475 	else
8476 		common.port.src ^= th->dest ^ first->protocol;
8477 	common.port.dst ^= th->source;
8478 
8479 	switch (hdr.ipv4->version) {
8480 	case IPVERSION:
8481 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
8482 		common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
8483 		break;
8484 	case 6:
8485 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
8486 		common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
8487 			     hdr.ipv6->saddr.s6_addr32[1] ^
8488 			     hdr.ipv6->saddr.s6_addr32[2] ^
8489 			     hdr.ipv6->saddr.s6_addr32[3] ^
8490 			     hdr.ipv6->daddr.s6_addr32[0] ^
8491 			     hdr.ipv6->daddr.s6_addr32[1] ^
8492 			     hdr.ipv6->daddr.s6_addr32[2] ^
8493 			     hdr.ipv6->daddr.s6_addr32[3];
8494 		break;
8495 	default:
8496 		break;
8497 	}
8498 
8499 	if (hdr.network != skb_network_header(skb))
8500 		input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
8501 
8502 	/* This assumes the Rx queue and Tx queue are bound to the same CPU */
8503 	ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
8504 					      input, common, ring->queue_index);
8505 }
8506 
8507 #ifdef IXGBE_FCOE
8508 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
8509 			      struct net_device *sb_dev)
8510 {
8511 	struct ixgbe_adapter *adapter;
8512 	struct ixgbe_ring_feature *f;
8513 	int txq;
8514 
8515 	if (sb_dev) {
8516 		u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
8517 		struct net_device *vdev = sb_dev;
8518 
8519 		txq = vdev->tc_to_txq[tc].offset;
8520 		txq += reciprocal_scale(skb_get_hash(skb),
8521 					vdev->tc_to_txq[tc].count);
8522 
8523 		return txq;
8524 	}
8525 
8526 	/*
8527 	 * only execute the code below if protocol is FCoE
8528 	 * or FIP and we have FCoE enabled on the adapter
8529 	 */
8530 	switch (vlan_get_protocol(skb)) {
8531 	case htons(ETH_P_FCOE):
8532 	case htons(ETH_P_FIP):
8533 		adapter = netdev_priv(dev);
8534 
8535 		if (!sb_dev && (adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
8536 			break;
8537 		/* fall through */
8538 	default:
8539 		return netdev_pick_tx(dev, skb, sb_dev);
8540 	}
8541 
8542 	f = &adapter->ring_feature[RING_F_FCOE];
8543 
8544 	txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
8545 					   smp_processor_id();
8546 
8547 	while (txq >= f->indices)
8548 		txq -= f->indices;
8549 
8550 	return txq + f->offset;
8551 }
8552 
8553 #endif
8554 int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
8555 			struct xdp_frame *xdpf)
8556 {
8557 	struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
8558 	struct ixgbe_tx_buffer *tx_buffer;
8559 	union ixgbe_adv_tx_desc *tx_desc;
8560 	u32 len, cmd_type;
8561 	dma_addr_t dma;
8562 	u16 i;
8563 
8564 	len = xdpf->len;
8565 
8566 	if (unlikely(!ixgbe_desc_unused(ring)))
8567 		return IXGBE_XDP_CONSUMED;
8568 
8569 	dma = dma_map_single(ring->dev, xdpf->data, len, DMA_TO_DEVICE);
8570 	if (dma_mapping_error(ring->dev, dma))
8571 		return IXGBE_XDP_CONSUMED;
8572 
8573 	/* record the location of the first descriptor for this packet */
8574 	tx_buffer = &ring->tx_buffer_info[ring->next_to_use];
8575 	tx_buffer->bytecount = len;
8576 	tx_buffer->gso_segs = 1;
8577 	tx_buffer->protocol = 0;
8578 
8579 	i = ring->next_to_use;
8580 	tx_desc = IXGBE_TX_DESC(ring, i);
8581 
8582 	dma_unmap_len_set(tx_buffer, len, len);
8583 	dma_unmap_addr_set(tx_buffer, dma, dma);
8584 	tx_buffer->xdpf = xdpf;
8585 
8586 	tx_desc->read.buffer_addr = cpu_to_le64(dma);
8587 
8588 	/* put descriptor type bits */
8589 	cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8590 		   IXGBE_ADVTXD_DCMD_DEXT |
8591 		   IXGBE_ADVTXD_DCMD_IFCS;
8592 	cmd_type |= len | IXGBE_TXD_CMD;
8593 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8594 	tx_desc->read.olinfo_status =
8595 		cpu_to_le32(len << IXGBE_ADVTXD_PAYLEN_SHIFT);
8596 
8597 	/* Avoid any potential race with xdp_xmit and cleanup */
8598 	smp_wmb();
8599 
8600 	/* set next_to_watch value indicating a packet is present */
8601 	i++;
8602 	if (i == ring->count)
8603 		i = 0;
8604 
8605 	tx_buffer->next_to_watch = tx_desc;
8606 	ring->next_to_use = i;
8607 
8608 	return IXGBE_XDP_TX;
8609 }
8610 
8611 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
8612 			  struct ixgbe_adapter *adapter,
8613 			  struct ixgbe_ring *tx_ring)
8614 {
8615 	struct ixgbe_tx_buffer *first;
8616 	int tso;
8617 	u32 tx_flags = 0;
8618 	unsigned short f;
8619 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
8620 	struct ixgbe_ipsec_tx_data ipsec_tx = { 0 };
8621 	__be16 protocol = skb->protocol;
8622 	u8 hdr_len = 0;
8623 
8624 	/*
8625 	 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
8626 	 *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
8627 	 *       + 2 desc gap to keep tail from touching head,
8628 	 *       + 1 desc for context descriptor,
8629 	 * otherwise try next time
8630 	 */
8631 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
8632 		count += TXD_USE_COUNT(skb_frag_size(
8633 						&skb_shinfo(skb)->frags[f]));
8634 
8635 	if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
8636 		tx_ring->tx_stats.tx_busy++;
8637 		return NETDEV_TX_BUSY;
8638 	}
8639 
8640 	/* record the location of the first descriptor for this packet */
8641 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
8642 	first->skb = skb;
8643 	first->bytecount = skb->len;
8644 	first->gso_segs = 1;
8645 
8646 	/* if we have a HW VLAN tag being added default to the HW one */
8647 	if (skb_vlan_tag_present(skb)) {
8648 		tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
8649 		tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8650 	/* else if it is a SW VLAN check the next protocol and store the tag */
8651 	} else if (protocol == htons(ETH_P_8021Q)) {
8652 		struct vlan_hdr *vhdr, _vhdr;
8653 		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
8654 		if (!vhdr)
8655 			goto out_drop;
8656 
8657 		tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
8658 				  IXGBE_TX_FLAGS_VLAN_SHIFT;
8659 		tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
8660 	}
8661 	protocol = vlan_get_protocol(skb);
8662 
8663 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
8664 	    adapter->ptp_clock) {
8665 		if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
8666 		    !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
8667 					   &adapter->state)) {
8668 			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
8669 			tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
8670 
8671 			/* schedule check for Tx timestamp */
8672 			adapter->ptp_tx_skb = skb_get(skb);
8673 			adapter->ptp_tx_start = jiffies;
8674 			schedule_work(&adapter->ptp_tx_work);
8675 		} else {
8676 			adapter->tx_hwtstamp_skipped++;
8677 		}
8678 	}
8679 
8680 #ifdef CONFIG_PCI_IOV
8681 	/*
8682 	 * Use the l2switch_enable flag - would be false if the DMA
8683 	 * Tx switch had been disabled.
8684 	 */
8685 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
8686 		tx_flags |= IXGBE_TX_FLAGS_CC;
8687 
8688 #endif
8689 	/* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
8690 	if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8691 	    ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
8692 	     (skb->priority != TC_PRIO_CONTROL))) {
8693 		tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
8694 		tx_flags |= (skb->priority & 0x7) <<
8695 					IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
8696 		if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
8697 			struct vlan_ethhdr *vhdr;
8698 
8699 			if (skb_cow_head(skb, 0))
8700 				goto out_drop;
8701 			vhdr = (struct vlan_ethhdr *)skb->data;
8702 			vhdr->h_vlan_TCI = htons(tx_flags >>
8703 						 IXGBE_TX_FLAGS_VLAN_SHIFT);
8704 		} else {
8705 			tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8706 		}
8707 	}
8708 
8709 	/* record initial flags and protocol */
8710 	first->tx_flags = tx_flags;
8711 	first->protocol = protocol;
8712 
8713 #ifdef IXGBE_FCOE
8714 	/* setup tx offload for FCoE */
8715 	if ((protocol == htons(ETH_P_FCOE)) &&
8716 	    (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
8717 		tso = ixgbe_fso(tx_ring, first, &hdr_len);
8718 		if (tso < 0)
8719 			goto out_drop;
8720 
8721 		goto xmit_fcoe;
8722 	}
8723 
8724 #endif /* IXGBE_FCOE */
8725 
8726 #ifdef CONFIG_IXGBE_IPSEC
8727 	if (xfrm_offload(skb) &&
8728 	    !ixgbe_ipsec_tx(tx_ring, first, &ipsec_tx))
8729 		goto out_drop;
8730 #endif
8731 	tso = ixgbe_tso(tx_ring, first, &hdr_len, &ipsec_tx);
8732 	if (tso < 0)
8733 		goto out_drop;
8734 	else if (!tso)
8735 		ixgbe_tx_csum(tx_ring, first, &ipsec_tx);
8736 
8737 	/* add the ATR filter if ATR is on */
8738 	if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
8739 		ixgbe_atr(tx_ring, first);
8740 
8741 #ifdef IXGBE_FCOE
8742 xmit_fcoe:
8743 #endif /* IXGBE_FCOE */
8744 	if (ixgbe_tx_map(tx_ring, first, hdr_len))
8745 		goto cleanup_tx_timestamp;
8746 
8747 	return NETDEV_TX_OK;
8748 
8749 out_drop:
8750 	dev_kfree_skb_any(first->skb);
8751 	first->skb = NULL;
8752 cleanup_tx_timestamp:
8753 	if (unlikely(tx_flags & IXGBE_TX_FLAGS_TSTAMP)) {
8754 		dev_kfree_skb_any(adapter->ptp_tx_skb);
8755 		adapter->ptp_tx_skb = NULL;
8756 		cancel_work_sync(&adapter->ptp_tx_work);
8757 		clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS, &adapter->state);
8758 	}
8759 
8760 	return NETDEV_TX_OK;
8761 }
8762 
8763 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
8764 				      struct net_device *netdev,
8765 				      struct ixgbe_ring *ring)
8766 {
8767 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8768 	struct ixgbe_ring *tx_ring;
8769 
8770 	/*
8771 	 * The minimum packet size for olinfo paylen is 17 so pad the skb
8772 	 * in order to meet this minimum size requirement.
8773 	 */
8774 	if (skb_put_padto(skb, 17))
8775 		return NETDEV_TX_OK;
8776 
8777 	tx_ring = ring ? ring : adapter->tx_ring[skb_get_queue_mapping(skb)];
8778 	if (unlikely(test_bit(__IXGBE_TX_DISABLED, &tx_ring->state)))
8779 		return NETDEV_TX_BUSY;
8780 
8781 	return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
8782 }
8783 
8784 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
8785 				    struct net_device *netdev)
8786 {
8787 	return __ixgbe_xmit_frame(skb, netdev, NULL);
8788 }
8789 
8790 /**
8791  * ixgbe_set_mac - Change the Ethernet Address of the NIC
8792  * @netdev: network interface device structure
8793  * @p: pointer to an address structure
8794  *
8795  * Returns 0 on success, negative on failure
8796  **/
8797 static int ixgbe_set_mac(struct net_device *netdev, void *p)
8798 {
8799 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8800 	struct ixgbe_hw *hw = &adapter->hw;
8801 	struct sockaddr *addr = p;
8802 
8803 	if (!is_valid_ether_addr(addr->sa_data))
8804 		return -EADDRNOTAVAIL;
8805 
8806 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
8807 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
8808 
8809 	ixgbe_mac_set_default_filter(adapter);
8810 
8811 	return 0;
8812 }
8813 
8814 static int
8815 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
8816 {
8817 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8818 	struct ixgbe_hw *hw = &adapter->hw;
8819 	u16 value;
8820 	int rc;
8821 
8822 	if (adapter->mii_bus) {
8823 		int regnum = addr;
8824 
8825 		if (devad != MDIO_DEVAD_NONE)
8826 			regnum |= (devad << 16) | MII_ADDR_C45;
8827 
8828 		return mdiobus_read(adapter->mii_bus, prtad, regnum);
8829 	}
8830 
8831 	if (prtad != hw->phy.mdio.prtad)
8832 		return -EINVAL;
8833 	rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
8834 	if (!rc)
8835 		rc = value;
8836 	return rc;
8837 }
8838 
8839 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
8840 			    u16 addr, u16 value)
8841 {
8842 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8843 	struct ixgbe_hw *hw = &adapter->hw;
8844 
8845 	if (adapter->mii_bus) {
8846 		int regnum = addr;
8847 
8848 		if (devad != MDIO_DEVAD_NONE)
8849 			regnum |= (devad << 16) | MII_ADDR_C45;
8850 
8851 		return mdiobus_write(adapter->mii_bus, prtad, regnum, value);
8852 	}
8853 
8854 	if (prtad != hw->phy.mdio.prtad)
8855 		return -EINVAL;
8856 	return hw->phy.ops.write_reg(hw, addr, devad, value);
8857 }
8858 
8859 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
8860 {
8861 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8862 
8863 	switch (cmd) {
8864 	case SIOCSHWTSTAMP:
8865 		return ixgbe_ptp_set_ts_config(adapter, req);
8866 	case SIOCGHWTSTAMP:
8867 		return ixgbe_ptp_get_ts_config(adapter, req);
8868 	case SIOCGMIIPHY:
8869 		if (!adapter->hw.phy.ops.read_reg)
8870 			return -EOPNOTSUPP;
8871 		/* fall through */
8872 	default:
8873 		return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
8874 	}
8875 }
8876 
8877 /**
8878  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
8879  * netdev->dev_addrs
8880  * @dev: network interface device structure
8881  *
8882  * Returns non-zero on failure
8883  **/
8884 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
8885 {
8886 	int err = 0;
8887 	struct ixgbe_adapter *adapter = netdev_priv(dev);
8888 	struct ixgbe_hw *hw = &adapter->hw;
8889 
8890 	if (is_valid_ether_addr(hw->mac.san_addr)) {
8891 		rtnl_lock();
8892 		err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
8893 		rtnl_unlock();
8894 
8895 		/* update SAN MAC vmdq pool selection */
8896 		hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
8897 	}
8898 	return err;
8899 }
8900 
8901 /**
8902  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
8903  * netdev->dev_addrs
8904  * @dev: network interface device structure
8905  *
8906  * Returns non-zero on failure
8907  **/
8908 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
8909 {
8910 	int err = 0;
8911 	struct ixgbe_adapter *adapter = netdev_priv(dev);
8912 	struct ixgbe_mac_info *mac = &adapter->hw.mac;
8913 
8914 	if (is_valid_ether_addr(mac->san_addr)) {
8915 		rtnl_lock();
8916 		err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
8917 		rtnl_unlock();
8918 	}
8919 	return err;
8920 }
8921 
8922 static void ixgbe_get_ring_stats64(struct rtnl_link_stats64 *stats,
8923 				   struct ixgbe_ring *ring)
8924 {
8925 	u64 bytes, packets;
8926 	unsigned int start;
8927 
8928 	if (ring) {
8929 		do {
8930 			start = u64_stats_fetch_begin_irq(&ring->syncp);
8931 			packets = ring->stats.packets;
8932 			bytes   = ring->stats.bytes;
8933 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8934 		stats->tx_packets += packets;
8935 		stats->tx_bytes   += bytes;
8936 	}
8937 }
8938 
8939 static void ixgbe_get_stats64(struct net_device *netdev,
8940 			      struct rtnl_link_stats64 *stats)
8941 {
8942 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8943 	int i;
8944 
8945 	rcu_read_lock();
8946 	for (i = 0; i < adapter->num_rx_queues; i++) {
8947 		struct ixgbe_ring *ring = READ_ONCE(adapter->rx_ring[i]);
8948 		u64 bytes, packets;
8949 		unsigned int start;
8950 
8951 		if (ring) {
8952 			do {
8953 				start = u64_stats_fetch_begin_irq(&ring->syncp);
8954 				packets = ring->stats.packets;
8955 				bytes   = ring->stats.bytes;
8956 			} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8957 			stats->rx_packets += packets;
8958 			stats->rx_bytes   += bytes;
8959 		}
8960 	}
8961 
8962 	for (i = 0; i < adapter->num_tx_queues; i++) {
8963 		struct ixgbe_ring *ring = READ_ONCE(adapter->tx_ring[i]);
8964 
8965 		ixgbe_get_ring_stats64(stats, ring);
8966 	}
8967 	for (i = 0; i < adapter->num_xdp_queues; i++) {
8968 		struct ixgbe_ring *ring = READ_ONCE(adapter->xdp_ring[i]);
8969 
8970 		ixgbe_get_ring_stats64(stats, ring);
8971 	}
8972 	rcu_read_unlock();
8973 
8974 	/* following stats updated by ixgbe_watchdog_task() */
8975 	stats->multicast	= netdev->stats.multicast;
8976 	stats->rx_errors	= netdev->stats.rx_errors;
8977 	stats->rx_length_errors	= netdev->stats.rx_length_errors;
8978 	stats->rx_crc_errors	= netdev->stats.rx_crc_errors;
8979 	stats->rx_missed_errors	= netdev->stats.rx_missed_errors;
8980 }
8981 
8982 #ifdef CONFIG_IXGBE_DCB
8983 /**
8984  * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
8985  * @adapter: pointer to ixgbe_adapter
8986  * @tc: number of traffic classes currently enabled
8987  *
8988  * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
8989  * 802.1Q priority maps to a packet buffer that exists.
8990  */
8991 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
8992 {
8993 	struct ixgbe_hw *hw = &adapter->hw;
8994 	u32 reg, rsave;
8995 	int i;
8996 
8997 	/* 82598 have a static priority to TC mapping that can not
8998 	 * be changed so no validation is needed.
8999 	 */
9000 	if (hw->mac.type == ixgbe_mac_82598EB)
9001 		return;
9002 
9003 	reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
9004 	rsave = reg;
9005 
9006 	for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
9007 		u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
9008 
9009 		/* If up2tc is out of bounds default to zero */
9010 		if (up2tc > tc)
9011 			reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
9012 	}
9013 
9014 	if (reg != rsave)
9015 		IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
9016 
9017 	return;
9018 }
9019 
9020 /**
9021  * ixgbe_set_prio_tc_map - Configure netdev prio tc map
9022  * @adapter: Pointer to adapter struct
9023  *
9024  * Populate the netdev user priority to tc map
9025  */
9026 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
9027 {
9028 	struct net_device *dev = adapter->netdev;
9029 	struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
9030 	struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
9031 	u8 prio;
9032 
9033 	for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
9034 		u8 tc = 0;
9035 
9036 		if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
9037 			tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
9038 		else if (ets)
9039 			tc = ets->prio_tc[prio];
9040 
9041 		netdev_set_prio_tc_map(dev, prio, tc);
9042 	}
9043 }
9044 
9045 #endif /* CONFIG_IXGBE_DCB */
9046 static int ixgbe_reassign_macvlan_pool(struct net_device *vdev, void *data)
9047 {
9048 	struct ixgbe_adapter *adapter = data;
9049 	struct ixgbe_fwd_adapter *accel;
9050 	int pool;
9051 
9052 	/* we only care about macvlans... */
9053 	if (!netif_is_macvlan(vdev))
9054 		return 0;
9055 
9056 	/* that have hardware offload enabled... */
9057 	accel = macvlan_accel_priv(vdev);
9058 	if (!accel)
9059 		return 0;
9060 
9061 	/* If we can relocate to a different bit do so */
9062 	pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
9063 	if (pool < adapter->num_rx_pools) {
9064 		set_bit(pool, adapter->fwd_bitmask);
9065 		accel->pool = pool;
9066 		return 0;
9067 	}
9068 
9069 	/* if we cannot find a free pool then disable the offload */
9070 	netdev_err(vdev, "L2FW offload disabled due to lack of queue resources\n");
9071 	macvlan_release_l2fw_offload(vdev);
9072 
9073 	/* unbind the queues and drop the subordinate channel config */
9074 	netdev_unbind_sb_channel(adapter->netdev, vdev);
9075 	netdev_set_sb_channel(vdev, 0);
9076 
9077 	kfree(accel);
9078 
9079 	return 0;
9080 }
9081 
9082 static void ixgbe_defrag_macvlan_pools(struct net_device *dev)
9083 {
9084 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9085 
9086 	/* flush any stale bits out of the fwd bitmask */
9087 	bitmap_clear(adapter->fwd_bitmask, 1, 63);
9088 
9089 	/* walk through upper devices reassigning pools */
9090 	netdev_walk_all_upper_dev_rcu(dev, ixgbe_reassign_macvlan_pool,
9091 				      adapter);
9092 }
9093 
9094 /**
9095  * ixgbe_setup_tc - configure net_device for multiple traffic classes
9096  *
9097  * @dev: net device to configure
9098  * @tc: number of traffic classes to enable
9099  */
9100 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
9101 {
9102 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9103 	struct ixgbe_hw *hw = &adapter->hw;
9104 
9105 	/* Hardware supports up to 8 traffic classes */
9106 	if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
9107 		return -EINVAL;
9108 
9109 	if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
9110 		return -EINVAL;
9111 
9112 	/* Hardware has to reinitialize queues and interrupts to
9113 	 * match packet buffer alignment. Unfortunately, the
9114 	 * hardware is not flexible enough to do this dynamically.
9115 	 */
9116 	if (netif_running(dev))
9117 		ixgbe_close(dev);
9118 	else
9119 		ixgbe_reset(adapter);
9120 
9121 	ixgbe_clear_interrupt_scheme(adapter);
9122 
9123 #ifdef CONFIG_IXGBE_DCB
9124 	if (tc) {
9125 		if (adapter->xdp_prog) {
9126 			e_warn(probe, "DCB is not supported with XDP\n");
9127 
9128 			ixgbe_init_interrupt_scheme(adapter);
9129 			if (netif_running(dev))
9130 				ixgbe_open(dev);
9131 			return -EINVAL;
9132 		}
9133 
9134 		netdev_set_num_tc(dev, tc);
9135 		ixgbe_set_prio_tc_map(adapter);
9136 
9137 		adapter->hw_tcs = tc;
9138 		adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
9139 
9140 		if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
9141 			adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
9142 			adapter->hw.fc.requested_mode = ixgbe_fc_none;
9143 		}
9144 	} else {
9145 		netdev_reset_tc(dev);
9146 
9147 		if (adapter->hw.mac.type == ixgbe_mac_82598EB)
9148 			adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
9149 
9150 		adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
9151 		adapter->hw_tcs = tc;
9152 
9153 		adapter->temp_dcb_cfg.pfc_mode_enable = false;
9154 		adapter->dcb_cfg.pfc_mode_enable = false;
9155 	}
9156 
9157 	ixgbe_validate_rtr(adapter, tc);
9158 
9159 #endif /* CONFIG_IXGBE_DCB */
9160 	ixgbe_init_interrupt_scheme(adapter);
9161 
9162 	ixgbe_defrag_macvlan_pools(dev);
9163 
9164 	if (netif_running(dev))
9165 		return ixgbe_open(dev);
9166 
9167 	return 0;
9168 }
9169 
9170 static int ixgbe_delete_clsu32(struct ixgbe_adapter *adapter,
9171 			       struct tc_cls_u32_offload *cls)
9172 {
9173 	u32 hdl = cls->knode.handle;
9174 	u32 uhtid = TC_U32_USERHTID(cls->knode.handle);
9175 	u32 loc = cls->knode.handle & 0xfffff;
9176 	int err = 0, i, j;
9177 	struct ixgbe_jump_table *jump = NULL;
9178 
9179 	if (loc > IXGBE_MAX_HW_ENTRIES)
9180 		return -EINVAL;
9181 
9182 	if ((uhtid != 0x800) && (uhtid >= IXGBE_MAX_LINK_HANDLE))
9183 		return -EINVAL;
9184 
9185 	/* Clear this filter in the link data it is associated with */
9186 	if (uhtid != 0x800) {
9187 		jump = adapter->jump_tables[uhtid];
9188 		if (!jump)
9189 			return -EINVAL;
9190 		if (!test_bit(loc - 1, jump->child_loc_map))
9191 			return -EINVAL;
9192 		clear_bit(loc - 1, jump->child_loc_map);
9193 	}
9194 
9195 	/* Check if the filter being deleted is a link */
9196 	for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9197 		jump = adapter->jump_tables[i];
9198 		if (jump && jump->link_hdl == hdl) {
9199 			/* Delete filters in the hardware in the child hash
9200 			 * table associated with this link
9201 			 */
9202 			for (j = 0; j < IXGBE_MAX_HW_ENTRIES; j++) {
9203 				if (!test_bit(j, jump->child_loc_map))
9204 					continue;
9205 				spin_lock(&adapter->fdir_perfect_lock);
9206 				err = ixgbe_update_ethtool_fdir_entry(adapter,
9207 								      NULL,
9208 								      j + 1);
9209 				spin_unlock(&adapter->fdir_perfect_lock);
9210 				clear_bit(j, jump->child_loc_map);
9211 			}
9212 			/* Remove resources for this link */
9213 			kfree(jump->input);
9214 			kfree(jump->mask);
9215 			kfree(jump);
9216 			adapter->jump_tables[i] = NULL;
9217 			return err;
9218 		}
9219 	}
9220 
9221 	spin_lock(&adapter->fdir_perfect_lock);
9222 	err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, loc);
9223 	spin_unlock(&adapter->fdir_perfect_lock);
9224 	return err;
9225 }
9226 
9227 static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter *adapter,
9228 					    struct tc_cls_u32_offload *cls)
9229 {
9230 	u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9231 
9232 	if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9233 		return -EINVAL;
9234 
9235 	/* This ixgbe devices do not support hash tables at the moment
9236 	 * so abort when given hash tables.
9237 	 */
9238 	if (cls->hnode.divisor > 0)
9239 		return -EINVAL;
9240 
9241 	set_bit(uhtid - 1, &adapter->tables);
9242 	return 0;
9243 }
9244 
9245 static int ixgbe_configure_clsu32_del_hnode(struct ixgbe_adapter *adapter,
9246 					    struct tc_cls_u32_offload *cls)
9247 {
9248 	u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9249 
9250 	if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9251 		return -EINVAL;
9252 
9253 	clear_bit(uhtid - 1, &adapter->tables);
9254 	return 0;
9255 }
9256 
9257 #ifdef CONFIG_NET_CLS_ACT
9258 struct upper_walk_data {
9259 	struct ixgbe_adapter *adapter;
9260 	u64 action;
9261 	int ifindex;
9262 	u8 queue;
9263 };
9264 
9265 static int get_macvlan_queue(struct net_device *upper, void *_data)
9266 {
9267 	if (netif_is_macvlan(upper)) {
9268 		struct ixgbe_fwd_adapter *vadapter = macvlan_accel_priv(upper);
9269 		struct upper_walk_data *data = _data;
9270 		struct ixgbe_adapter *adapter = data->adapter;
9271 		int ifindex = data->ifindex;
9272 
9273 		if (vadapter && upper->ifindex == ifindex) {
9274 			data->queue = adapter->rx_ring[vadapter->rx_base_queue]->reg_idx;
9275 			data->action = data->queue;
9276 			return 1;
9277 		}
9278 	}
9279 
9280 	return 0;
9281 }
9282 
9283 static int handle_redirect_action(struct ixgbe_adapter *adapter, int ifindex,
9284 				  u8 *queue, u64 *action)
9285 {
9286 	struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
9287 	unsigned int num_vfs = adapter->num_vfs, vf;
9288 	struct upper_walk_data data;
9289 	struct net_device *upper;
9290 
9291 	/* redirect to a SRIOV VF */
9292 	for (vf = 0; vf < num_vfs; ++vf) {
9293 		upper = pci_get_drvdata(adapter->vfinfo[vf].vfdev);
9294 		if (upper->ifindex == ifindex) {
9295 			*queue = vf * __ALIGN_MASK(1, ~vmdq->mask);
9296 			*action = vf + 1;
9297 			*action <<= ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF;
9298 			return 0;
9299 		}
9300 	}
9301 
9302 	/* redirect to a offloaded macvlan netdev */
9303 	data.adapter = adapter;
9304 	data.ifindex = ifindex;
9305 	data.action = 0;
9306 	data.queue = 0;
9307 	if (netdev_walk_all_upper_dev_rcu(adapter->netdev,
9308 					  get_macvlan_queue, &data)) {
9309 		*action = data.action;
9310 		*queue = data.queue;
9311 
9312 		return 0;
9313 	}
9314 
9315 	return -EINVAL;
9316 }
9317 
9318 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9319 			    struct tcf_exts *exts, u64 *action, u8 *queue)
9320 {
9321 	const struct tc_action *a;
9322 	int i;
9323 
9324 	if (!tcf_exts_has_actions(exts))
9325 		return -EINVAL;
9326 
9327 	tcf_exts_for_each_action(i, a, exts) {
9328 		/* Drop action */
9329 		if (is_tcf_gact_shot(a)) {
9330 			*action = IXGBE_FDIR_DROP_QUEUE;
9331 			*queue = IXGBE_FDIR_DROP_QUEUE;
9332 			return 0;
9333 		}
9334 
9335 		/* Redirect to a VF or a offloaded macvlan */
9336 		if (is_tcf_mirred_egress_redirect(a)) {
9337 			struct net_device *dev = tcf_mirred_dev(a);
9338 
9339 			if (!dev)
9340 				return -EINVAL;
9341 			return handle_redirect_action(adapter, dev->ifindex,
9342 						      queue, action);
9343 		}
9344 
9345 		return -EINVAL;
9346 	}
9347 
9348 	return -EINVAL;
9349 }
9350 #else
9351 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9352 			    struct tcf_exts *exts, u64 *action, u8 *queue)
9353 {
9354 	return -EINVAL;
9355 }
9356 #endif /* CONFIG_NET_CLS_ACT */
9357 
9358 static int ixgbe_clsu32_build_input(struct ixgbe_fdir_filter *input,
9359 				    union ixgbe_atr_input *mask,
9360 				    struct tc_cls_u32_offload *cls,
9361 				    struct ixgbe_mat_field *field_ptr,
9362 				    struct ixgbe_nexthdr *nexthdr)
9363 {
9364 	int i, j, off;
9365 	__be32 val, m;
9366 	bool found_entry = false, found_jump_field = false;
9367 
9368 	for (i = 0; i < cls->knode.sel->nkeys; i++) {
9369 		off = cls->knode.sel->keys[i].off;
9370 		val = cls->knode.sel->keys[i].val;
9371 		m = cls->knode.sel->keys[i].mask;
9372 
9373 		for (j = 0; field_ptr[j].val; j++) {
9374 			if (field_ptr[j].off == off) {
9375 				field_ptr[j].val(input, mask, (__force u32)val,
9376 						 (__force u32)m);
9377 				input->filter.formatted.flow_type |=
9378 					field_ptr[j].type;
9379 				found_entry = true;
9380 				break;
9381 			}
9382 		}
9383 		if (nexthdr) {
9384 			if (nexthdr->off == cls->knode.sel->keys[i].off &&
9385 			    nexthdr->val ==
9386 			    (__force u32)cls->knode.sel->keys[i].val &&
9387 			    nexthdr->mask ==
9388 			    (__force u32)cls->knode.sel->keys[i].mask)
9389 				found_jump_field = true;
9390 			else
9391 				continue;
9392 		}
9393 	}
9394 
9395 	if (nexthdr && !found_jump_field)
9396 		return -EINVAL;
9397 
9398 	if (!found_entry)
9399 		return 0;
9400 
9401 	mask->formatted.flow_type = IXGBE_ATR_L4TYPE_IPV6_MASK |
9402 				    IXGBE_ATR_L4TYPE_MASK;
9403 
9404 	if (input->filter.formatted.flow_type == IXGBE_ATR_FLOW_TYPE_IPV4)
9405 		mask->formatted.flow_type &= IXGBE_ATR_L4TYPE_IPV6_MASK;
9406 
9407 	return 0;
9408 }
9409 
9410 static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
9411 				  struct tc_cls_u32_offload *cls)
9412 {
9413 	__be16 protocol = cls->common.protocol;
9414 	u32 loc = cls->knode.handle & 0xfffff;
9415 	struct ixgbe_hw *hw = &adapter->hw;
9416 	struct ixgbe_mat_field *field_ptr;
9417 	struct ixgbe_fdir_filter *input = NULL;
9418 	union ixgbe_atr_input *mask = NULL;
9419 	struct ixgbe_jump_table *jump = NULL;
9420 	int i, err = -EINVAL;
9421 	u8 queue;
9422 	u32 uhtid, link_uhtid;
9423 
9424 	uhtid = TC_U32_USERHTID(cls->knode.handle);
9425 	link_uhtid = TC_U32_USERHTID(cls->knode.link_handle);
9426 
9427 	/* At the moment cls_u32 jumps to network layer and skips past
9428 	 * L2 headers. The canonical method to match L2 frames is to use
9429 	 * negative values. However this is error prone at best but really
9430 	 * just broken because there is no way to "know" what sort of hdr
9431 	 * is in front of the network layer. Fix cls_u32 to support L2
9432 	 * headers when needed.
9433 	 */
9434 	if (protocol != htons(ETH_P_IP))
9435 		return err;
9436 
9437 	if (loc >= ((1024 << adapter->fdir_pballoc) - 2)) {
9438 		e_err(drv, "Location out of range\n");
9439 		return err;
9440 	}
9441 
9442 	/* cls u32 is a graph starting at root node 0x800. The driver tracks
9443 	 * links and also the fields used to advance the parser across each
9444 	 * link (e.g. nexthdr/eat parameters from 'tc'). This way we can map
9445 	 * the u32 graph onto the hardware parse graph denoted in ixgbe_model.h
9446 	 * To add support for new nodes update ixgbe_model.h parse structures
9447 	 * this function _should_ be generic try not to hardcode values here.
9448 	 */
9449 	if (uhtid == 0x800) {
9450 		field_ptr = (adapter->jump_tables[0])->mat;
9451 	} else {
9452 		if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9453 			return err;
9454 		if (!adapter->jump_tables[uhtid])
9455 			return err;
9456 		field_ptr = (adapter->jump_tables[uhtid])->mat;
9457 	}
9458 
9459 	if (!field_ptr)
9460 		return err;
9461 
9462 	/* At this point we know the field_ptr is valid and need to either
9463 	 * build cls_u32 link or attach filter. Because adding a link to
9464 	 * a handle that does not exist is invalid and the same for adding
9465 	 * rules to handles that don't exist.
9466 	 */
9467 
9468 	if (link_uhtid) {
9469 		struct ixgbe_nexthdr *nexthdr = ixgbe_ipv4_jumps;
9470 
9471 		if (link_uhtid >= IXGBE_MAX_LINK_HANDLE)
9472 			return err;
9473 
9474 		if (!test_bit(link_uhtid - 1, &adapter->tables))
9475 			return err;
9476 
9477 		/* Multiple filters as links to the same hash table are not
9478 		 * supported. To add a new filter with the same next header
9479 		 * but different match/jump conditions, create a new hash table
9480 		 * and link to it.
9481 		 */
9482 		if (adapter->jump_tables[link_uhtid] &&
9483 		    (adapter->jump_tables[link_uhtid])->link_hdl) {
9484 			e_err(drv, "Link filter exists for link: %x\n",
9485 			      link_uhtid);
9486 			return err;
9487 		}
9488 
9489 		for (i = 0; nexthdr[i].jump; i++) {
9490 			if (nexthdr[i].o != cls->knode.sel->offoff ||
9491 			    nexthdr[i].s != cls->knode.sel->offshift ||
9492 			    nexthdr[i].m !=
9493 			    (__force u32)cls->knode.sel->offmask)
9494 				return err;
9495 
9496 			jump = kzalloc(sizeof(*jump), GFP_KERNEL);
9497 			if (!jump)
9498 				return -ENOMEM;
9499 			input = kzalloc(sizeof(*input), GFP_KERNEL);
9500 			if (!input) {
9501 				err = -ENOMEM;
9502 				goto free_jump;
9503 			}
9504 			mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9505 			if (!mask) {
9506 				err = -ENOMEM;
9507 				goto free_input;
9508 			}
9509 			jump->input = input;
9510 			jump->mask = mask;
9511 			jump->link_hdl = cls->knode.handle;
9512 
9513 			err = ixgbe_clsu32_build_input(input, mask, cls,
9514 						       field_ptr, &nexthdr[i]);
9515 			if (!err) {
9516 				jump->mat = nexthdr[i].jump;
9517 				adapter->jump_tables[link_uhtid] = jump;
9518 				break;
9519 			} else {
9520 				kfree(mask);
9521 				kfree(input);
9522 				kfree(jump);
9523 			}
9524 		}
9525 		return 0;
9526 	}
9527 
9528 	input = kzalloc(sizeof(*input), GFP_KERNEL);
9529 	if (!input)
9530 		return -ENOMEM;
9531 	mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9532 	if (!mask) {
9533 		err = -ENOMEM;
9534 		goto free_input;
9535 	}
9536 
9537 	if ((uhtid != 0x800) && (adapter->jump_tables[uhtid])) {
9538 		if ((adapter->jump_tables[uhtid])->input)
9539 			memcpy(input, (adapter->jump_tables[uhtid])->input,
9540 			       sizeof(*input));
9541 		if ((adapter->jump_tables[uhtid])->mask)
9542 			memcpy(mask, (adapter->jump_tables[uhtid])->mask,
9543 			       sizeof(*mask));
9544 
9545 		/* Lookup in all child hash tables if this location is already
9546 		 * filled with a filter
9547 		 */
9548 		for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9549 			struct ixgbe_jump_table *link = adapter->jump_tables[i];
9550 
9551 			if (link && (test_bit(loc - 1, link->child_loc_map))) {
9552 				e_err(drv, "Filter exists in location: %x\n",
9553 				      loc);
9554 				err = -EINVAL;
9555 				goto err_out;
9556 			}
9557 		}
9558 	}
9559 	err = ixgbe_clsu32_build_input(input, mask, cls, field_ptr, NULL);
9560 	if (err)
9561 		goto err_out;
9562 
9563 	err = parse_tc_actions(adapter, cls->knode.exts, &input->action,
9564 			       &queue);
9565 	if (err < 0)
9566 		goto err_out;
9567 
9568 	input->sw_idx = loc;
9569 
9570 	spin_lock(&adapter->fdir_perfect_lock);
9571 
9572 	if (hlist_empty(&adapter->fdir_filter_list)) {
9573 		memcpy(&adapter->fdir_mask, mask, sizeof(*mask));
9574 		err = ixgbe_fdir_set_input_mask_82599(hw, mask);
9575 		if (err)
9576 			goto err_out_w_lock;
9577 	} else if (memcmp(&adapter->fdir_mask, mask, sizeof(*mask))) {
9578 		err = -EINVAL;
9579 		goto err_out_w_lock;
9580 	}
9581 
9582 	ixgbe_atr_compute_perfect_hash_82599(&input->filter, mask);
9583 	err = ixgbe_fdir_write_perfect_filter_82599(hw, &input->filter,
9584 						    input->sw_idx, queue);
9585 	if (!err)
9586 		ixgbe_update_ethtool_fdir_entry(adapter, input, input->sw_idx);
9587 	spin_unlock(&adapter->fdir_perfect_lock);
9588 
9589 	if ((uhtid != 0x800) && (adapter->jump_tables[uhtid]))
9590 		set_bit(loc - 1, (adapter->jump_tables[uhtid])->child_loc_map);
9591 
9592 	kfree(mask);
9593 	return err;
9594 err_out_w_lock:
9595 	spin_unlock(&adapter->fdir_perfect_lock);
9596 err_out:
9597 	kfree(mask);
9598 free_input:
9599 	kfree(input);
9600 free_jump:
9601 	kfree(jump);
9602 	return err;
9603 }
9604 
9605 static int ixgbe_setup_tc_cls_u32(struct ixgbe_adapter *adapter,
9606 				  struct tc_cls_u32_offload *cls_u32)
9607 {
9608 	switch (cls_u32->command) {
9609 	case TC_CLSU32_NEW_KNODE:
9610 	case TC_CLSU32_REPLACE_KNODE:
9611 		return ixgbe_configure_clsu32(adapter, cls_u32);
9612 	case TC_CLSU32_DELETE_KNODE:
9613 		return ixgbe_delete_clsu32(adapter, cls_u32);
9614 	case TC_CLSU32_NEW_HNODE:
9615 	case TC_CLSU32_REPLACE_HNODE:
9616 		return ixgbe_configure_clsu32_add_hnode(adapter, cls_u32);
9617 	case TC_CLSU32_DELETE_HNODE:
9618 		return ixgbe_configure_clsu32_del_hnode(adapter, cls_u32);
9619 	default:
9620 		return -EOPNOTSUPP;
9621 	}
9622 }
9623 
9624 static int ixgbe_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
9625 				   void *cb_priv)
9626 {
9627 	struct ixgbe_adapter *adapter = cb_priv;
9628 
9629 	if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
9630 		return -EOPNOTSUPP;
9631 
9632 	switch (type) {
9633 	case TC_SETUP_CLSU32:
9634 		return ixgbe_setup_tc_cls_u32(adapter, type_data);
9635 	default:
9636 		return -EOPNOTSUPP;
9637 	}
9638 }
9639 
9640 static int ixgbe_setup_tc_mqprio(struct net_device *dev,
9641 				 struct tc_mqprio_qopt *mqprio)
9642 {
9643 	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
9644 	return ixgbe_setup_tc(dev, mqprio->num_tc);
9645 }
9646 
9647 static LIST_HEAD(ixgbe_block_cb_list);
9648 
9649 static int __ixgbe_setup_tc(struct net_device *dev, enum tc_setup_type type,
9650 			    void *type_data)
9651 {
9652 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9653 
9654 	switch (type) {
9655 	case TC_SETUP_BLOCK:
9656 		return flow_block_cb_setup_simple(type_data,
9657 						  &ixgbe_block_cb_list,
9658 						  ixgbe_setup_tc_block_cb,
9659 						  adapter, adapter, true);
9660 	case TC_SETUP_QDISC_MQPRIO:
9661 		return ixgbe_setup_tc_mqprio(dev, type_data);
9662 	default:
9663 		return -EOPNOTSUPP;
9664 	}
9665 }
9666 
9667 #ifdef CONFIG_PCI_IOV
9668 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
9669 {
9670 	struct net_device *netdev = adapter->netdev;
9671 
9672 	rtnl_lock();
9673 	ixgbe_setup_tc(netdev, adapter->hw_tcs);
9674 	rtnl_unlock();
9675 }
9676 
9677 #endif
9678 void ixgbe_do_reset(struct net_device *netdev)
9679 {
9680 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
9681 
9682 	if (netif_running(netdev))
9683 		ixgbe_reinit_locked(adapter);
9684 	else
9685 		ixgbe_reset(adapter);
9686 }
9687 
9688 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
9689 					    netdev_features_t features)
9690 {
9691 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
9692 
9693 	/* If Rx checksum is disabled, then RSC/LRO should also be disabled */
9694 	if (!(features & NETIF_F_RXCSUM))
9695 		features &= ~NETIF_F_LRO;
9696 
9697 	/* Turn off LRO if not RSC capable */
9698 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
9699 		features &= ~NETIF_F_LRO;
9700 
9701 	if (adapter->xdp_prog && (features & NETIF_F_LRO)) {
9702 		e_dev_err("LRO is not supported with XDP\n");
9703 		features &= ~NETIF_F_LRO;
9704 	}
9705 
9706 	return features;
9707 }
9708 
9709 static void ixgbe_reset_l2fw_offload(struct ixgbe_adapter *adapter)
9710 {
9711 	int rss = min_t(int, ixgbe_max_rss_indices(adapter),
9712 			num_online_cpus());
9713 
9714 	/* go back to full RSS if we're not running SR-IOV */
9715 	if (!adapter->ring_feature[RING_F_VMDQ].offset)
9716 		adapter->flags &= ~(IXGBE_FLAG_VMDQ_ENABLED |
9717 				    IXGBE_FLAG_SRIOV_ENABLED);
9718 
9719 	adapter->ring_feature[RING_F_RSS].limit = rss;
9720 	adapter->ring_feature[RING_F_VMDQ].limit = 1;
9721 
9722 	ixgbe_setup_tc(adapter->netdev, adapter->hw_tcs);
9723 }
9724 
9725 static int ixgbe_set_features(struct net_device *netdev,
9726 			      netdev_features_t features)
9727 {
9728 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
9729 	netdev_features_t changed = netdev->features ^ features;
9730 	bool need_reset = false;
9731 
9732 	/* Make sure RSC matches LRO, reset if change */
9733 	if (!(features & NETIF_F_LRO)) {
9734 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
9735 			need_reset = true;
9736 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
9737 	} else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
9738 		   !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
9739 		if (adapter->rx_itr_setting == 1 ||
9740 		    adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
9741 			adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
9742 			need_reset = true;
9743 		} else if ((changed ^ features) & NETIF_F_LRO) {
9744 			e_info(probe, "rx-usecs set too low, "
9745 			       "disabling RSC\n");
9746 		}
9747 	}
9748 
9749 	/*
9750 	 * Check if Flow Director n-tuple support or hw_tc support was
9751 	 * enabled or disabled.  If the state changed, we need to reset.
9752 	 */
9753 	if ((features & NETIF_F_NTUPLE) || (features & NETIF_F_HW_TC)) {
9754 		/* turn off ATR, enable perfect filters and reset */
9755 		if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
9756 			need_reset = true;
9757 
9758 		adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
9759 		adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9760 	} else {
9761 		/* turn off perfect filters, enable ATR and reset */
9762 		if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
9763 			need_reset = true;
9764 
9765 		adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9766 
9767 		/* We cannot enable ATR if SR-IOV is enabled */
9768 		if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED ||
9769 		    /* We cannot enable ATR if we have 2 or more tcs */
9770 		    (adapter->hw_tcs > 1) ||
9771 		    /* We cannot enable ATR if RSS is disabled */
9772 		    (adapter->ring_feature[RING_F_RSS].limit <= 1) ||
9773 		    /* A sample rate of 0 indicates ATR disabled */
9774 		    (!adapter->atr_sample_rate))
9775 			; /* do nothing not supported */
9776 		else /* otherwise supported and set the flag */
9777 			adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
9778 	}
9779 
9780 	if (changed & NETIF_F_RXALL)
9781 		need_reset = true;
9782 
9783 	netdev->features = features;
9784 
9785 	if ((adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE)) {
9786 		if (features & NETIF_F_RXCSUM) {
9787 			adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9788 		} else {
9789 			u32 port_mask = IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK;
9790 
9791 			ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9792 		}
9793 	}
9794 
9795 	if ((adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE)) {
9796 		if (features & NETIF_F_RXCSUM) {
9797 			adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9798 		} else {
9799 			u32 port_mask = IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK;
9800 
9801 			ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9802 		}
9803 	}
9804 
9805 	if ((changed & NETIF_F_HW_L2FW_DOFFLOAD) && adapter->num_rx_pools > 1)
9806 		ixgbe_reset_l2fw_offload(adapter);
9807 	else if (need_reset)
9808 		ixgbe_do_reset(netdev);
9809 	else if (changed & (NETIF_F_HW_VLAN_CTAG_RX |
9810 			    NETIF_F_HW_VLAN_CTAG_FILTER))
9811 		ixgbe_set_rx_mode(netdev);
9812 
9813 	return 1;
9814 }
9815 
9816 /**
9817  * ixgbe_add_udp_tunnel_port - Get notifications about adding UDP tunnel ports
9818  * @dev: The port's netdev
9819  * @ti: Tunnel endpoint information
9820  **/
9821 static void ixgbe_add_udp_tunnel_port(struct net_device *dev,
9822 				      struct udp_tunnel_info *ti)
9823 {
9824 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9825 	struct ixgbe_hw *hw = &adapter->hw;
9826 	__be16 port = ti->port;
9827 	u32 port_shift = 0;
9828 	u32 reg;
9829 
9830 	if (ti->sa_family != AF_INET)
9831 		return;
9832 
9833 	switch (ti->type) {
9834 	case UDP_TUNNEL_TYPE_VXLAN:
9835 		if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
9836 			return;
9837 
9838 		if (adapter->vxlan_port == port)
9839 			return;
9840 
9841 		if (adapter->vxlan_port) {
9842 			netdev_info(dev,
9843 				    "VXLAN port %d set, not adding port %d\n",
9844 				    ntohs(adapter->vxlan_port),
9845 				    ntohs(port));
9846 			return;
9847 		}
9848 
9849 		adapter->vxlan_port = port;
9850 		break;
9851 	case UDP_TUNNEL_TYPE_GENEVE:
9852 		if (!(adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE))
9853 			return;
9854 
9855 		if (adapter->geneve_port == port)
9856 			return;
9857 
9858 		if (adapter->geneve_port) {
9859 			netdev_info(dev,
9860 				    "GENEVE port %d set, not adding port %d\n",
9861 				    ntohs(adapter->geneve_port),
9862 				    ntohs(port));
9863 			return;
9864 		}
9865 
9866 		port_shift = IXGBE_VXLANCTRL_GENEVE_UDPPORT_SHIFT;
9867 		adapter->geneve_port = port;
9868 		break;
9869 	default:
9870 		return;
9871 	}
9872 
9873 	reg = IXGBE_READ_REG(hw, IXGBE_VXLANCTRL) | ntohs(port) << port_shift;
9874 	IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, reg);
9875 }
9876 
9877 /**
9878  * ixgbe_del_udp_tunnel_port - Get notifications about removing UDP tunnel ports
9879  * @dev: The port's netdev
9880  * @ti: Tunnel endpoint information
9881  **/
9882 static void ixgbe_del_udp_tunnel_port(struct net_device *dev,
9883 				      struct udp_tunnel_info *ti)
9884 {
9885 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9886 	u32 port_mask;
9887 
9888 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN &&
9889 	    ti->type != UDP_TUNNEL_TYPE_GENEVE)
9890 		return;
9891 
9892 	if (ti->sa_family != AF_INET)
9893 		return;
9894 
9895 	switch (ti->type) {
9896 	case UDP_TUNNEL_TYPE_VXLAN:
9897 		if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
9898 			return;
9899 
9900 		if (adapter->vxlan_port != ti->port) {
9901 			netdev_info(dev, "VXLAN port %d not found\n",
9902 				    ntohs(ti->port));
9903 			return;
9904 		}
9905 
9906 		port_mask = IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK;
9907 		break;
9908 	case UDP_TUNNEL_TYPE_GENEVE:
9909 		if (!(adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE))
9910 			return;
9911 
9912 		if (adapter->geneve_port != ti->port) {
9913 			netdev_info(dev, "GENEVE port %d not found\n",
9914 				    ntohs(ti->port));
9915 			return;
9916 		}
9917 
9918 		port_mask = IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK;
9919 		break;
9920 	default:
9921 		return;
9922 	}
9923 
9924 	ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9925 	adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9926 }
9927 
9928 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
9929 			     struct net_device *dev,
9930 			     const unsigned char *addr, u16 vid,
9931 			     u16 flags,
9932 			     struct netlink_ext_ack *extack)
9933 {
9934 	/* guarantee we can provide a unique filter for the unicast address */
9935 	if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
9936 		struct ixgbe_adapter *adapter = netdev_priv(dev);
9937 		u16 pool = VMDQ_P(0);
9938 
9939 		if (netdev_uc_count(dev) >= ixgbe_available_rars(adapter, pool))
9940 			return -ENOMEM;
9941 	}
9942 
9943 	return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
9944 }
9945 
9946 /**
9947  * ixgbe_configure_bridge_mode - set various bridge modes
9948  * @adapter: the private structure
9949  * @mode: requested bridge mode
9950  *
9951  * Configure some settings require for various bridge modes.
9952  **/
9953 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
9954 				       __u16 mode)
9955 {
9956 	struct ixgbe_hw *hw = &adapter->hw;
9957 	unsigned int p, num_pools;
9958 	u32 vmdctl;
9959 
9960 	switch (mode) {
9961 	case BRIDGE_MODE_VEPA:
9962 		/* disable Tx loopback, rely on switch hairpin mode */
9963 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
9964 
9965 		/* must enable Rx switching replication to allow multicast
9966 		 * packet reception on all VFs, and to enable source address
9967 		 * pruning.
9968 		 */
9969 		vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9970 		vmdctl |= IXGBE_VT_CTL_REPLEN;
9971 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9972 
9973 		/* enable Rx source address pruning. Note, this requires
9974 		 * replication to be enabled or else it does nothing.
9975 		 */
9976 		num_pools = adapter->num_vfs + adapter->num_rx_pools;
9977 		for (p = 0; p < num_pools; p++) {
9978 			if (hw->mac.ops.set_source_address_pruning)
9979 				hw->mac.ops.set_source_address_pruning(hw,
9980 								       true,
9981 								       p);
9982 		}
9983 		break;
9984 	case BRIDGE_MODE_VEB:
9985 		/* enable Tx loopback for internal VF/PF communication */
9986 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
9987 				IXGBE_PFDTXGSWC_VT_LBEN);
9988 
9989 		/* disable Rx switching replication unless we have SR-IOV
9990 		 * virtual functions
9991 		 */
9992 		vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9993 		if (!adapter->num_vfs)
9994 			vmdctl &= ~IXGBE_VT_CTL_REPLEN;
9995 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9996 
9997 		/* disable Rx source address pruning, since we don't expect to
9998 		 * be receiving external loopback of our transmitted frames.
9999 		 */
10000 		num_pools = adapter->num_vfs + adapter->num_rx_pools;
10001 		for (p = 0; p < num_pools; p++) {
10002 			if (hw->mac.ops.set_source_address_pruning)
10003 				hw->mac.ops.set_source_address_pruning(hw,
10004 								       false,
10005 								       p);
10006 		}
10007 		break;
10008 	default:
10009 		return -EINVAL;
10010 	}
10011 
10012 	adapter->bridge_mode = mode;
10013 
10014 	e_info(drv, "enabling bridge mode: %s\n",
10015 	       mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
10016 
10017 	return 0;
10018 }
10019 
10020 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
10021 				    struct nlmsghdr *nlh, u16 flags,
10022 				    struct netlink_ext_ack *extack)
10023 {
10024 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10025 	struct nlattr *attr, *br_spec;
10026 	int rem;
10027 
10028 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
10029 		return -EOPNOTSUPP;
10030 
10031 	br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
10032 	if (!br_spec)
10033 		return -EINVAL;
10034 
10035 	nla_for_each_nested(attr, br_spec, rem) {
10036 		int status;
10037 		__u16 mode;
10038 
10039 		if (nla_type(attr) != IFLA_BRIDGE_MODE)
10040 			continue;
10041 
10042 		if (nla_len(attr) < sizeof(mode))
10043 			return -EINVAL;
10044 
10045 		mode = nla_get_u16(attr);
10046 		status = ixgbe_configure_bridge_mode(adapter, mode);
10047 		if (status)
10048 			return status;
10049 
10050 		break;
10051 	}
10052 
10053 	return 0;
10054 }
10055 
10056 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
10057 				    struct net_device *dev,
10058 				    u32 filter_mask, int nlflags)
10059 {
10060 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10061 
10062 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
10063 		return 0;
10064 
10065 	return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
10066 				       adapter->bridge_mode, 0, 0, nlflags,
10067 				       filter_mask, NULL);
10068 }
10069 
10070 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
10071 {
10072 	struct ixgbe_adapter *adapter = netdev_priv(pdev);
10073 	struct ixgbe_fwd_adapter *accel;
10074 	int tcs = adapter->hw_tcs ? : 1;
10075 	int pool, err;
10076 
10077 	if (adapter->xdp_prog) {
10078 		e_warn(probe, "L2FW offload is not supported with XDP\n");
10079 		return ERR_PTR(-EINVAL);
10080 	}
10081 
10082 	/* The hardware supported by ixgbe only filters on the destination MAC
10083 	 * address. In order to avoid issues we only support offloading modes
10084 	 * where the hardware can actually provide the functionality.
10085 	 */
10086 	if (!macvlan_supports_dest_filter(vdev))
10087 		return ERR_PTR(-EMEDIUMTYPE);
10088 
10089 	/* We need to lock down the macvlan to be a single queue device so that
10090 	 * we can reuse the tc_to_txq field in the macvlan netdev to represent
10091 	 * the queue mapping to our netdev.
10092 	 */
10093 	if (netif_is_multiqueue(vdev))
10094 		return ERR_PTR(-ERANGE);
10095 
10096 	pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
10097 	if (pool == adapter->num_rx_pools) {
10098 		u16 used_pools = adapter->num_vfs + adapter->num_rx_pools;
10099 		u16 reserved_pools;
10100 
10101 		if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
10102 		     adapter->num_rx_pools >= (MAX_TX_QUEUES / tcs)) ||
10103 		    adapter->num_rx_pools > IXGBE_MAX_MACVLANS)
10104 			return ERR_PTR(-EBUSY);
10105 
10106 		/* Hardware has a limited number of available pools. Each VF,
10107 		 * and the PF require a pool. Check to ensure we don't
10108 		 * attempt to use more then the available number of pools.
10109 		 */
10110 		if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
10111 			return ERR_PTR(-EBUSY);
10112 
10113 		/* Enable VMDq flag so device will be set in VM mode */
10114 		adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED |
10115 				  IXGBE_FLAG_SRIOV_ENABLED;
10116 
10117 		/* Try to reserve as many queues per pool as possible,
10118 		 * we start with the configurations that support 4 queues
10119 		 * per pools, followed by 2, and then by just 1 per pool.
10120 		 */
10121 		if (used_pools < 32 && adapter->num_rx_pools < 16)
10122 			reserved_pools = min_t(u16,
10123 					       32 - used_pools,
10124 					       16 - adapter->num_rx_pools);
10125 		else if (adapter->num_rx_pools < 32)
10126 			reserved_pools = min_t(u16,
10127 					       64 - used_pools,
10128 					       32 - adapter->num_rx_pools);
10129 		else
10130 			reserved_pools = 64 - used_pools;
10131 
10132 
10133 		if (!reserved_pools)
10134 			return ERR_PTR(-EBUSY);
10135 
10136 		adapter->ring_feature[RING_F_VMDQ].limit += reserved_pools;
10137 
10138 		/* Force reinit of ring allocation with VMDQ enabled */
10139 		err = ixgbe_setup_tc(pdev, adapter->hw_tcs);
10140 		if (err)
10141 			return ERR_PTR(err);
10142 
10143 		if (pool >= adapter->num_rx_pools)
10144 			return ERR_PTR(-ENOMEM);
10145 	}
10146 
10147 	accel = kzalloc(sizeof(*accel), GFP_KERNEL);
10148 	if (!accel)
10149 		return ERR_PTR(-ENOMEM);
10150 
10151 	set_bit(pool, adapter->fwd_bitmask);
10152 	netdev_set_sb_channel(vdev, pool);
10153 	accel->pool = pool;
10154 	accel->netdev = vdev;
10155 
10156 	if (!netif_running(pdev))
10157 		return accel;
10158 
10159 	err = ixgbe_fwd_ring_up(adapter, accel);
10160 	if (err)
10161 		return ERR_PTR(err);
10162 
10163 	return accel;
10164 }
10165 
10166 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
10167 {
10168 	struct ixgbe_fwd_adapter *accel = priv;
10169 	struct ixgbe_adapter *adapter = netdev_priv(pdev);
10170 	unsigned int rxbase = accel->rx_base_queue;
10171 	unsigned int i;
10172 
10173 	/* delete unicast filter associated with offloaded interface */
10174 	ixgbe_del_mac_filter(adapter, accel->netdev->dev_addr,
10175 			     VMDQ_P(accel->pool));
10176 
10177 	/* Allow remaining Rx packets to get flushed out of the
10178 	 * Rx FIFO before we drop the netdev for the ring.
10179 	 */
10180 	usleep_range(10000, 20000);
10181 
10182 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
10183 		struct ixgbe_ring *ring = adapter->rx_ring[rxbase + i];
10184 		struct ixgbe_q_vector *qv = ring->q_vector;
10185 
10186 		/* Make sure we aren't processing any packets and clear
10187 		 * netdev to shut down the ring.
10188 		 */
10189 		if (netif_running(adapter->netdev))
10190 			napi_synchronize(&qv->napi);
10191 		ring->netdev = NULL;
10192 	}
10193 
10194 	/* unbind the queues and drop the subordinate channel config */
10195 	netdev_unbind_sb_channel(pdev, accel->netdev);
10196 	netdev_set_sb_channel(accel->netdev, 0);
10197 
10198 	clear_bit(accel->pool, adapter->fwd_bitmask);
10199 	kfree(accel);
10200 }
10201 
10202 #define IXGBE_MAX_MAC_HDR_LEN		127
10203 #define IXGBE_MAX_NETWORK_HDR_LEN	511
10204 
10205 static netdev_features_t
10206 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
10207 		     netdev_features_t features)
10208 {
10209 	unsigned int network_hdr_len, mac_hdr_len;
10210 
10211 	/* Make certain the headers can be described by a context descriptor */
10212 	mac_hdr_len = skb_network_header(skb) - skb->data;
10213 	if (unlikely(mac_hdr_len > IXGBE_MAX_MAC_HDR_LEN))
10214 		return features & ~(NETIF_F_HW_CSUM |
10215 				    NETIF_F_SCTP_CRC |
10216 				    NETIF_F_GSO_UDP_L4 |
10217 				    NETIF_F_HW_VLAN_CTAG_TX |
10218 				    NETIF_F_TSO |
10219 				    NETIF_F_TSO6);
10220 
10221 	network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
10222 	if (unlikely(network_hdr_len >  IXGBE_MAX_NETWORK_HDR_LEN))
10223 		return features & ~(NETIF_F_HW_CSUM |
10224 				    NETIF_F_SCTP_CRC |
10225 				    NETIF_F_GSO_UDP_L4 |
10226 				    NETIF_F_TSO |
10227 				    NETIF_F_TSO6);
10228 
10229 	/* We can only support IPV4 TSO in tunnels if we can mangle the
10230 	 * inner IP ID field, so strip TSO if MANGLEID is not supported.
10231 	 * IPsec offoad sets skb->encapsulation but still can handle
10232 	 * the TSO, so it's the exception.
10233 	 */
10234 	if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) {
10235 #ifdef CONFIG_IXGBE_IPSEC
10236 		if (!secpath_exists(skb))
10237 #endif
10238 			features &= ~NETIF_F_TSO;
10239 	}
10240 
10241 	return features;
10242 }
10243 
10244 static int ixgbe_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
10245 {
10246 	int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
10247 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10248 	struct bpf_prog *old_prog;
10249 	bool need_reset;
10250 
10251 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
10252 		return -EINVAL;
10253 
10254 	if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
10255 		return -EINVAL;
10256 
10257 	/* verify ixgbe ring attributes are sufficient for XDP */
10258 	for (i = 0; i < adapter->num_rx_queues; i++) {
10259 		struct ixgbe_ring *ring = adapter->rx_ring[i];
10260 
10261 		if (ring_is_rsc_enabled(ring))
10262 			return -EINVAL;
10263 
10264 		if (frame_size > ixgbe_rx_bufsz(ring))
10265 			return -EINVAL;
10266 	}
10267 
10268 	if (nr_cpu_ids > MAX_XDP_QUEUES)
10269 		return -ENOMEM;
10270 
10271 	old_prog = xchg(&adapter->xdp_prog, prog);
10272 	need_reset = (!!prog != !!old_prog);
10273 
10274 	/* If transitioning XDP modes reconfigure rings */
10275 	if (need_reset) {
10276 		int err;
10277 
10278 		if (!prog)
10279 			/* Wait until ndo_xsk_wakeup completes. */
10280 			synchronize_rcu();
10281 		err = ixgbe_setup_tc(dev, adapter->hw_tcs);
10282 
10283 		if (err) {
10284 			rcu_assign_pointer(adapter->xdp_prog, old_prog);
10285 			return -EINVAL;
10286 		}
10287 	} else {
10288 		for (i = 0; i < adapter->num_rx_queues; i++)
10289 			(void)xchg(&adapter->rx_ring[i]->xdp_prog,
10290 			    adapter->xdp_prog);
10291 	}
10292 
10293 	if (old_prog)
10294 		bpf_prog_put(old_prog);
10295 
10296 	/* Kick start the NAPI context if there is an AF_XDP socket open
10297 	 * on that queue id. This so that receiving will start.
10298 	 */
10299 	if (need_reset && prog)
10300 		for (i = 0; i < adapter->num_rx_queues; i++)
10301 			if (adapter->xdp_ring[i]->xsk_umem)
10302 				(void)ixgbe_xsk_wakeup(adapter->netdev, i,
10303 						       XDP_WAKEUP_RX);
10304 
10305 	return 0;
10306 }
10307 
10308 static int ixgbe_xdp(struct net_device *dev, struct netdev_bpf *xdp)
10309 {
10310 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10311 
10312 	switch (xdp->command) {
10313 	case XDP_SETUP_PROG:
10314 		return ixgbe_xdp_setup(dev, xdp->prog);
10315 	case XDP_QUERY_PROG:
10316 		xdp->prog_id = adapter->xdp_prog ?
10317 			adapter->xdp_prog->aux->id : 0;
10318 		return 0;
10319 	case XDP_SETUP_XSK_UMEM:
10320 		return ixgbe_xsk_umem_setup(adapter, xdp->xsk.umem,
10321 					    xdp->xsk.queue_id);
10322 
10323 	default:
10324 		return -EINVAL;
10325 	}
10326 }
10327 
10328 void ixgbe_xdp_ring_update_tail(struct ixgbe_ring *ring)
10329 {
10330 	/* Force memory writes to complete before letting h/w know there
10331 	 * are new descriptors to fetch.
10332 	 */
10333 	wmb();
10334 	writel(ring->next_to_use, ring->tail);
10335 }
10336 
10337 static int ixgbe_xdp_xmit(struct net_device *dev, int n,
10338 			  struct xdp_frame **frames, u32 flags)
10339 {
10340 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10341 	struct ixgbe_ring *ring;
10342 	int drops = 0;
10343 	int i;
10344 
10345 	if (unlikely(test_bit(__IXGBE_DOWN, &adapter->state)))
10346 		return -ENETDOWN;
10347 
10348 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
10349 		return -EINVAL;
10350 
10351 	/* During program transitions its possible adapter->xdp_prog is assigned
10352 	 * but ring has not been configured yet. In this case simply abort xmit.
10353 	 */
10354 	ring = adapter->xdp_prog ? adapter->xdp_ring[smp_processor_id()] : NULL;
10355 	if (unlikely(!ring))
10356 		return -ENXIO;
10357 
10358 	if (unlikely(test_bit(__IXGBE_TX_DISABLED, &ring->state)))
10359 		return -ENXIO;
10360 
10361 	for (i = 0; i < n; i++) {
10362 		struct xdp_frame *xdpf = frames[i];
10363 		int err;
10364 
10365 		err = ixgbe_xmit_xdp_ring(adapter, xdpf);
10366 		if (err != IXGBE_XDP_TX) {
10367 			xdp_return_frame_rx_napi(xdpf);
10368 			drops++;
10369 		}
10370 	}
10371 
10372 	if (unlikely(flags & XDP_XMIT_FLUSH))
10373 		ixgbe_xdp_ring_update_tail(ring);
10374 
10375 	return n - drops;
10376 }
10377 
10378 static const struct net_device_ops ixgbe_netdev_ops = {
10379 	.ndo_open		= ixgbe_open,
10380 	.ndo_stop		= ixgbe_close,
10381 	.ndo_start_xmit		= ixgbe_xmit_frame,
10382 	.ndo_set_rx_mode	= ixgbe_set_rx_mode,
10383 	.ndo_validate_addr	= eth_validate_addr,
10384 	.ndo_set_mac_address	= ixgbe_set_mac,
10385 	.ndo_change_mtu		= ixgbe_change_mtu,
10386 	.ndo_tx_timeout		= ixgbe_tx_timeout,
10387 	.ndo_set_tx_maxrate	= ixgbe_tx_maxrate,
10388 	.ndo_vlan_rx_add_vid	= ixgbe_vlan_rx_add_vid,
10389 	.ndo_vlan_rx_kill_vid	= ixgbe_vlan_rx_kill_vid,
10390 	.ndo_do_ioctl		= ixgbe_ioctl,
10391 	.ndo_set_vf_mac		= ixgbe_ndo_set_vf_mac,
10392 	.ndo_set_vf_vlan	= ixgbe_ndo_set_vf_vlan,
10393 	.ndo_set_vf_rate	= ixgbe_ndo_set_vf_bw,
10394 	.ndo_set_vf_spoofchk	= ixgbe_ndo_set_vf_spoofchk,
10395 	.ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
10396 	.ndo_set_vf_trust	= ixgbe_ndo_set_vf_trust,
10397 	.ndo_get_vf_config	= ixgbe_ndo_get_vf_config,
10398 	.ndo_get_stats64	= ixgbe_get_stats64,
10399 	.ndo_setup_tc		= __ixgbe_setup_tc,
10400 #ifdef IXGBE_FCOE
10401 	.ndo_select_queue	= ixgbe_select_queue,
10402 	.ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
10403 	.ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
10404 	.ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
10405 	.ndo_fcoe_enable = ixgbe_fcoe_enable,
10406 	.ndo_fcoe_disable = ixgbe_fcoe_disable,
10407 	.ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
10408 	.ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
10409 #endif /* IXGBE_FCOE */
10410 	.ndo_set_features = ixgbe_set_features,
10411 	.ndo_fix_features = ixgbe_fix_features,
10412 	.ndo_fdb_add		= ixgbe_ndo_fdb_add,
10413 	.ndo_bridge_setlink	= ixgbe_ndo_bridge_setlink,
10414 	.ndo_bridge_getlink	= ixgbe_ndo_bridge_getlink,
10415 	.ndo_dfwd_add_station	= ixgbe_fwd_add,
10416 	.ndo_dfwd_del_station	= ixgbe_fwd_del,
10417 	.ndo_udp_tunnel_add	= ixgbe_add_udp_tunnel_port,
10418 	.ndo_udp_tunnel_del	= ixgbe_del_udp_tunnel_port,
10419 	.ndo_features_check	= ixgbe_features_check,
10420 	.ndo_bpf		= ixgbe_xdp,
10421 	.ndo_xdp_xmit		= ixgbe_xdp_xmit,
10422 	.ndo_xsk_wakeup         = ixgbe_xsk_wakeup,
10423 };
10424 
10425 static void ixgbe_disable_txr_hw(struct ixgbe_adapter *adapter,
10426 				 struct ixgbe_ring *tx_ring)
10427 {
10428 	unsigned long wait_delay, delay_interval;
10429 	struct ixgbe_hw *hw = &adapter->hw;
10430 	u8 reg_idx = tx_ring->reg_idx;
10431 	int wait_loop;
10432 	u32 txdctl;
10433 
10434 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
10435 
10436 	/* delay mechanism from ixgbe_disable_tx */
10437 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10438 
10439 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
10440 	wait_delay = delay_interval;
10441 
10442 	while (wait_loop--) {
10443 		usleep_range(wait_delay, wait_delay + 10);
10444 		wait_delay += delay_interval * 2;
10445 		txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
10446 
10447 		if (!(txdctl & IXGBE_TXDCTL_ENABLE))
10448 			return;
10449 	}
10450 
10451 	e_err(drv, "TXDCTL.ENABLE not cleared within the polling period\n");
10452 }
10453 
10454 static void ixgbe_disable_txr(struct ixgbe_adapter *adapter,
10455 			      struct ixgbe_ring *tx_ring)
10456 {
10457 	set_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10458 	ixgbe_disable_txr_hw(adapter, tx_ring);
10459 }
10460 
10461 static void ixgbe_disable_rxr_hw(struct ixgbe_adapter *adapter,
10462 				 struct ixgbe_ring *rx_ring)
10463 {
10464 	unsigned long wait_delay, delay_interval;
10465 	struct ixgbe_hw *hw = &adapter->hw;
10466 	u8 reg_idx = rx_ring->reg_idx;
10467 	int wait_loop;
10468 	u32 rxdctl;
10469 
10470 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10471 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
10472 	rxdctl |= IXGBE_RXDCTL_SWFLSH;
10473 
10474 	/* write value back with RXDCTL.ENABLE bit cleared */
10475 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
10476 
10477 	/* RXDCTL.EN may not change on 82598 if link is down, so skip it */
10478 	if (hw->mac.type == ixgbe_mac_82598EB &&
10479 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
10480 		return;
10481 
10482 	/* delay mechanism from ixgbe_disable_rx */
10483 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10484 
10485 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
10486 	wait_delay = delay_interval;
10487 
10488 	while (wait_loop--) {
10489 		usleep_range(wait_delay, wait_delay + 10);
10490 		wait_delay += delay_interval * 2;
10491 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10492 
10493 		if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
10494 			return;
10495 	}
10496 
10497 	e_err(drv, "RXDCTL.ENABLE not cleared within the polling period\n");
10498 }
10499 
10500 static void ixgbe_reset_txr_stats(struct ixgbe_ring *tx_ring)
10501 {
10502 	memset(&tx_ring->stats, 0, sizeof(tx_ring->stats));
10503 	memset(&tx_ring->tx_stats, 0, sizeof(tx_ring->tx_stats));
10504 }
10505 
10506 static void ixgbe_reset_rxr_stats(struct ixgbe_ring *rx_ring)
10507 {
10508 	memset(&rx_ring->stats, 0, sizeof(rx_ring->stats));
10509 	memset(&rx_ring->rx_stats, 0, sizeof(rx_ring->rx_stats));
10510 }
10511 
10512 /**
10513  * ixgbe_txrx_ring_disable - Disable Rx/Tx/XDP Tx rings
10514  * @adapter: adapter structure
10515  * @ring: ring index
10516  *
10517  * This function disables a certain Rx/Tx/XDP Tx ring. The function
10518  * assumes that the netdev is running.
10519  **/
10520 void ixgbe_txrx_ring_disable(struct ixgbe_adapter *adapter, int ring)
10521 {
10522 	struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10523 
10524 	rx_ring = adapter->rx_ring[ring];
10525 	tx_ring = adapter->tx_ring[ring];
10526 	xdp_ring = adapter->xdp_ring[ring];
10527 
10528 	ixgbe_disable_txr(adapter, tx_ring);
10529 	if (xdp_ring)
10530 		ixgbe_disable_txr(adapter, xdp_ring);
10531 	ixgbe_disable_rxr_hw(adapter, rx_ring);
10532 
10533 	if (xdp_ring)
10534 		synchronize_rcu();
10535 
10536 	/* Rx/Tx/XDP Tx share the same napi context. */
10537 	napi_disable(&rx_ring->q_vector->napi);
10538 
10539 	ixgbe_clean_tx_ring(tx_ring);
10540 	if (xdp_ring)
10541 		ixgbe_clean_tx_ring(xdp_ring);
10542 	ixgbe_clean_rx_ring(rx_ring);
10543 
10544 	ixgbe_reset_txr_stats(tx_ring);
10545 	if (xdp_ring)
10546 		ixgbe_reset_txr_stats(xdp_ring);
10547 	ixgbe_reset_rxr_stats(rx_ring);
10548 }
10549 
10550 /**
10551  * ixgbe_txrx_ring_enable - Enable Rx/Tx/XDP Tx rings
10552  * @adapter: adapter structure
10553  * @ring: ring index
10554  *
10555  * This function enables a certain Rx/Tx/XDP Tx ring. The function
10556  * assumes that the netdev is running.
10557  **/
10558 void ixgbe_txrx_ring_enable(struct ixgbe_adapter *adapter, int ring)
10559 {
10560 	struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10561 
10562 	rx_ring = adapter->rx_ring[ring];
10563 	tx_ring = adapter->tx_ring[ring];
10564 	xdp_ring = adapter->xdp_ring[ring];
10565 
10566 	/* Rx/Tx/XDP Tx share the same napi context. */
10567 	napi_enable(&rx_ring->q_vector->napi);
10568 
10569 	ixgbe_configure_tx_ring(adapter, tx_ring);
10570 	if (xdp_ring)
10571 		ixgbe_configure_tx_ring(adapter, xdp_ring);
10572 	ixgbe_configure_rx_ring(adapter, rx_ring);
10573 
10574 	clear_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10575 	if (xdp_ring)
10576 		clear_bit(__IXGBE_TX_DISABLED, &xdp_ring->state);
10577 }
10578 
10579 /**
10580  * ixgbe_enumerate_functions - Get the number of ports this device has
10581  * @adapter: adapter structure
10582  *
10583  * This function enumerates the phsyical functions co-located on a single slot,
10584  * in order to determine how many ports a device has. This is most useful in
10585  * determining the required GT/s of PCIe bandwidth necessary for optimal
10586  * performance.
10587  **/
10588 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
10589 {
10590 	struct pci_dev *entry, *pdev = adapter->pdev;
10591 	int physfns = 0;
10592 
10593 	/* Some cards can not use the generic count PCIe functions method,
10594 	 * because they are behind a parent switch, so we hardcode these with
10595 	 * the correct number of functions.
10596 	 */
10597 	if (ixgbe_pcie_from_parent(&adapter->hw))
10598 		physfns = 4;
10599 
10600 	list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
10601 		/* don't count virtual functions */
10602 		if (entry->is_virtfn)
10603 			continue;
10604 
10605 		/* When the devices on the bus don't all match our device ID,
10606 		 * we can't reliably determine the correct number of
10607 		 * functions. This can occur if a function has been direct
10608 		 * attached to a virtual machine using VT-d, for example. In
10609 		 * this case, simply return -1 to indicate this.
10610 		 */
10611 		if ((entry->vendor != pdev->vendor) ||
10612 		    (entry->device != pdev->device))
10613 			return -1;
10614 
10615 		physfns++;
10616 	}
10617 
10618 	return physfns;
10619 }
10620 
10621 /**
10622  * ixgbe_wol_supported - Check whether device supports WoL
10623  * @adapter: the adapter private structure
10624  * @device_id: the device ID
10625  * @subdevice_id: the subsystem device ID
10626  *
10627  * This function is used by probe and ethtool to determine
10628  * which devices have WoL support
10629  *
10630  **/
10631 bool ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
10632 			 u16 subdevice_id)
10633 {
10634 	struct ixgbe_hw *hw = &adapter->hw;
10635 	u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
10636 
10637 	/* WOL not supported on 82598 */
10638 	if (hw->mac.type == ixgbe_mac_82598EB)
10639 		return false;
10640 
10641 	/* check eeprom to see if WOL is enabled for X540 and newer */
10642 	if (hw->mac.type >= ixgbe_mac_X540) {
10643 		if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
10644 		    ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
10645 		     (hw->bus.func == 0)))
10646 			return true;
10647 	}
10648 
10649 	/* WOL is determined based on device IDs for 82599 MACs */
10650 	switch (device_id) {
10651 	case IXGBE_DEV_ID_82599_SFP:
10652 		/* Only these subdevices could supports WOL */
10653 		switch (subdevice_id) {
10654 		case IXGBE_SUBDEV_ID_82599_560FLR:
10655 		case IXGBE_SUBDEV_ID_82599_LOM_SNAP6:
10656 		case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
10657 		case IXGBE_SUBDEV_ID_82599_SFP_2OCP:
10658 			/* only support first port */
10659 			if (hw->bus.func != 0)
10660 				break;
10661 			/* fall through */
10662 		case IXGBE_SUBDEV_ID_82599_SP_560FLR:
10663 		case IXGBE_SUBDEV_ID_82599_SFP:
10664 		case IXGBE_SUBDEV_ID_82599_RNDC:
10665 		case IXGBE_SUBDEV_ID_82599_ECNA_DP:
10666 		case IXGBE_SUBDEV_ID_82599_SFP_1OCP:
10667 		case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM1:
10668 		case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM2:
10669 			return true;
10670 		}
10671 		break;
10672 	case IXGBE_DEV_ID_82599EN_SFP:
10673 		/* Only these subdevices support WOL */
10674 		switch (subdevice_id) {
10675 		case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
10676 			return true;
10677 		}
10678 		break;
10679 	case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
10680 		/* All except this subdevice support WOL */
10681 		if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
10682 			return true;
10683 		break;
10684 	case IXGBE_DEV_ID_82599_KX4:
10685 		return  true;
10686 	default:
10687 		break;
10688 	}
10689 
10690 	return false;
10691 }
10692 
10693 /**
10694  * ixgbe_set_fw_version - Set FW version
10695  * @adapter: the adapter private structure
10696  *
10697  * This function is used by probe and ethtool to determine the FW version to
10698  * format to display. The FW version is taken from the EEPROM/NVM.
10699  */
10700 static void ixgbe_set_fw_version(struct ixgbe_adapter *adapter)
10701 {
10702 	struct ixgbe_hw *hw = &adapter->hw;
10703 	struct ixgbe_nvm_version nvm_ver;
10704 
10705 	ixgbe_get_oem_prod_version(hw, &nvm_ver);
10706 	if (nvm_ver.oem_valid) {
10707 		snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10708 			 "%x.%x.%x", nvm_ver.oem_major, nvm_ver.oem_minor,
10709 			 nvm_ver.oem_release);
10710 		return;
10711 	}
10712 
10713 	ixgbe_get_etk_id(hw, &nvm_ver);
10714 	ixgbe_get_orom_version(hw, &nvm_ver);
10715 
10716 	if (nvm_ver.or_valid) {
10717 		snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10718 			 "0x%08x, %d.%d.%d", nvm_ver.etk_id, nvm_ver.or_major,
10719 			 nvm_ver.or_build, nvm_ver.or_patch);
10720 		return;
10721 	}
10722 
10723 	/* Set ETrack ID format */
10724 	snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10725 		 "0x%08x", nvm_ver.etk_id);
10726 }
10727 
10728 /**
10729  * ixgbe_probe - Device Initialization Routine
10730  * @pdev: PCI device information struct
10731  * @ent: entry in ixgbe_pci_tbl
10732  *
10733  * Returns 0 on success, negative on failure
10734  *
10735  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
10736  * The OS initialization, configuring of the adapter private structure,
10737  * and a hardware reset occur.
10738  **/
10739 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
10740 {
10741 	struct net_device *netdev;
10742 	struct ixgbe_adapter *adapter = NULL;
10743 	struct ixgbe_hw *hw;
10744 	const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
10745 	int i, err, pci_using_dac, expected_gts;
10746 	unsigned int indices = MAX_TX_QUEUES;
10747 	u8 part_str[IXGBE_PBANUM_LENGTH];
10748 	bool disable_dev = false;
10749 #ifdef IXGBE_FCOE
10750 	u16 device_caps;
10751 #endif
10752 	u32 eec;
10753 
10754 	/* Catch broken hardware that put the wrong VF device ID in
10755 	 * the PCIe SR-IOV capability.
10756 	 */
10757 	if (pdev->is_virtfn) {
10758 		WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
10759 		     pci_name(pdev), pdev->vendor, pdev->device);
10760 		return -EINVAL;
10761 	}
10762 
10763 	err = pci_enable_device_mem(pdev);
10764 	if (err)
10765 		return err;
10766 
10767 	if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
10768 		pci_using_dac = 1;
10769 	} else {
10770 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
10771 		if (err) {
10772 			dev_err(&pdev->dev,
10773 				"No usable DMA configuration, aborting\n");
10774 			goto err_dma;
10775 		}
10776 		pci_using_dac = 0;
10777 	}
10778 
10779 	err = pci_request_mem_regions(pdev, ixgbe_driver_name);
10780 	if (err) {
10781 		dev_err(&pdev->dev,
10782 			"pci_request_selected_regions failed 0x%x\n", err);
10783 		goto err_pci_reg;
10784 	}
10785 
10786 	pci_enable_pcie_error_reporting(pdev);
10787 
10788 	pci_set_master(pdev);
10789 	pci_save_state(pdev);
10790 
10791 	if (ii->mac == ixgbe_mac_82598EB) {
10792 #ifdef CONFIG_IXGBE_DCB
10793 		/* 8 TC w/ 4 queues per TC */
10794 		indices = 4 * MAX_TRAFFIC_CLASS;
10795 #else
10796 		indices = IXGBE_MAX_RSS_INDICES;
10797 #endif
10798 	}
10799 
10800 	netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
10801 	if (!netdev) {
10802 		err = -ENOMEM;
10803 		goto err_alloc_etherdev;
10804 	}
10805 
10806 	SET_NETDEV_DEV(netdev, &pdev->dev);
10807 
10808 	adapter = netdev_priv(netdev);
10809 
10810 	adapter->netdev = netdev;
10811 	adapter->pdev = pdev;
10812 	hw = &adapter->hw;
10813 	hw->back = adapter;
10814 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
10815 
10816 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
10817 			      pci_resource_len(pdev, 0));
10818 	adapter->io_addr = hw->hw_addr;
10819 	if (!hw->hw_addr) {
10820 		err = -EIO;
10821 		goto err_ioremap;
10822 	}
10823 
10824 	netdev->netdev_ops = &ixgbe_netdev_ops;
10825 	ixgbe_set_ethtool_ops(netdev);
10826 	netdev->watchdog_timeo = 5 * HZ;
10827 	strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
10828 
10829 	/* Setup hw api */
10830 	hw->mac.ops   = *ii->mac_ops;
10831 	hw->mac.type  = ii->mac;
10832 	hw->mvals     = ii->mvals;
10833 	if (ii->link_ops)
10834 		hw->link.ops  = *ii->link_ops;
10835 
10836 	/* EEPROM */
10837 	hw->eeprom.ops = *ii->eeprom_ops;
10838 	eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
10839 	if (ixgbe_removed(hw->hw_addr)) {
10840 		err = -EIO;
10841 		goto err_ioremap;
10842 	}
10843 	/* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
10844 	if (!(eec & BIT(8)))
10845 		hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
10846 
10847 	/* PHY */
10848 	hw->phy.ops = *ii->phy_ops;
10849 	hw->phy.sfp_type = ixgbe_sfp_type_unknown;
10850 	/* ixgbe_identify_phy_generic will set prtad and mmds properly */
10851 	hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
10852 	hw->phy.mdio.mmds = 0;
10853 	hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
10854 	hw->phy.mdio.dev = netdev;
10855 	hw->phy.mdio.mdio_read = ixgbe_mdio_read;
10856 	hw->phy.mdio.mdio_write = ixgbe_mdio_write;
10857 
10858 	/* setup the private structure */
10859 	err = ixgbe_sw_init(adapter, ii);
10860 	if (err)
10861 		goto err_sw_init;
10862 
10863 	/* Make sure the SWFW semaphore is in a valid state */
10864 	if (hw->mac.ops.init_swfw_sync)
10865 		hw->mac.ops.init_swfw_sync(hw);
10866 
10867 	/* Make it possible the adapter to be woken up via WOL */
10868 	switch (adapter->hw.mac.type) {
10869 	case ixgbe_mac_82599EB:
10870 	case ixgbe_mac_X540:
10871 	case ixgbe_mac_X550:
10872 	case ixgbe_mac_X550EM_x:
10873 	case ixgbe_mac_x550em_a:
10874 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
10875 		break;
10876 	default:
10877 		break;
10878 	}
10879 
10880 	/*
10881 	 * If there is a fan on this device and it has failed log the
10882 	 * failure.
10883 	 */
10884 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
10885 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
10886 		if (esdp & IXGBE_ESDP_SDP1)
10887 			e_crit(probe, "Fan has stopped, replace the adapter\n");
10888 	}
10889 
10890 	if (allow_unsupported_sfp)
10891 		hw->allow_unsupported_sfp = allow_unsupported_sfp;
10892 
10893 	/* reset_hw fills in the perm_addr as well */
10894 	hw->phy.reset_if_overtemp = true;
10895 	err = hw->mac.ops.reset_hw(hw);
10896 	hw->phy.reset_if_overtemp = false;
10897 	ixgbe_set_eee_capable(adapter);
10898 	if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
10899 		err = 0;
10900 	} else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
10901 		e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
10902 		e_dev_err("Reload the driver after installing a supported module.\n");
10903 		goto err_sw_init;
10904 	} else if (err) {
10905 		e_dev_err("HW Init failed: %d\n", err);
10906 		goto err_sw_init;
10907 	}
10908 
10909 #ifdef CONFIG_PCI_IOV
10910 	/* SR-IOV not supported on the 82598 */
10911 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
10912 		goto skip_sriov;
10913 	/* Mailbox */
10914 	ixgbe_init_mbx_params_pf(hw);
10915 	hw->mbx.ops = ii->mbx_ops;
10916 	pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
10917 	ixgbe_enable_sriov(adapter, max_vfs);
10918 skip_sriov:
10919 
10920 #endif
10921 	netdev->features = NETIF_F_SG |
10922 			   NETIF_F_TSO |
10923 			   NETIF_F_TSO6 |
10924 			   NETIF_F_RXHASH |
10925 			   NETIF_F_RXCSUM |
10926 			   NETIF_F_HW_CSUM;
10927 
10928 #define IXGBE_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
10929 				    NETIF_F_GSO_GRE_CSUM | \
10930 				    NETIF_F_GSO_IPXIP4 | \
10931 				    NETIF_F_GSO_IPXIP6 | \
10932 				    NETIF_F_GSO_UDP_TUNNEL | \
10933 				    NETIF_F_GSO_UDP_TUNNEL_CSUM)
10934 
10935 	netdev->gso_partial_features = IXGBE_GSO_PARTIAL_FEATURES;
10936 	netdev->features |= NETIF_F_GSO_PARTIAL |
10937 			    IXGBE_GSO_PARTIAL_FEATURES;
10938 
10939 	if (hw->mac.type >= ixgbe_mac_82599EB)
10940 		netdev->features |= NETIF_F_SCTP_CRC | NETIF_F_GSO_UDP_L4;
10941 
10942 #ifdef CONFIG_IXGBE_IPSEC
10943 #define IXGBE_ESP_FEATURES	(NETIF_F_HW_ESP | \
10944 				 NETIF_F_HW_ESP_TX_CSUM | \
10945 				 NETIF_F_GSO_ESP)
10946 
10947 	if (adapter->ipsec)
10948 		netdev->features |= IXGBE_ESP_FEATURES;
10949 #endif
10950 	/* copy netdev features into list of user selectable features */
10951 	netdev->hw_features |= netdev->features |
10952 			       NETIF_F_HW_VLAN_CTAG_FILTER |
10953 			       NETIF_F_HW_VLAN_CTAG_RX |
10954 			       NETIF_F_HW_VLAN_CTAG_TX |
10955 			       NETIF_F_RXALL |
10956 			       NETIF_F_HW_L2FW_DOFFLOAD;
10957 
10958 	if (hw->mac.type >= ixgbe_mac_82599EB)
10959 		netdev->hw_features |= NETIF_F_NTUPLE |
10960 				       NETIF_F_HW_TC;
10961 
10962 	if (pci_using_dac)
10963 		netdev->features |= NETIF_F_HIGHDMA;
10964 
10965 	netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
10966 	netdev->hw_enc_features |= netdev->vlan_features;
10967 	netdev->mpls_features |= NETIF_F_SG |
10968 				 NETIF_F_TSO |
10969 				 NETIF_F_TSO6 |
10970 				 NETIF_F_HW_CSUM;
10971 	netdev->mpls_features |= IXGBE_GSO_PARTIAL_FEATURES;
10972 
10973 	/* set this bit last since it cannot be part of vlan_features */
10974 	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
10975 			    NETIF_F_HW_VLAN_CTAG_RX |
10976 			    NETIF_F_HW_VLAN_CTAG_TX;
10977 
10978 	netdev->priv_flags |= IFF_UNICAST_FLT;
10979 	netdev->priv_flags |= IFF_SUPP_NOFCS;
10980 
10981 	/* MTU range: 68 - 9710 */
10982 	netdev->min_mtu = ETH_MIN_MTU;
10983 	netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE - (ETH_HLEN + ETH_FCS_LEN);
10984 
10985 #ifdef CONFIG_IXGBE_DCB
10986 	if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
10987 		netdev->dcbnl_ops = &ixgbe_dcbnl_ops;
10988 #endif
10989 
10990 #ifdef IXGBE_FCOE
10991 	if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
10992 		unsigned int fcoe_l;
10993 
10994 		if (hw->mac.ops.get_device_caps) {
10995 			hw->mac.ops.get_device_caps(hw, &device_caps);
10996 			if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
10997 				adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
10998 		}
10999 
11000 
11001 		fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
11002 		adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
11003 
11004 		netdev->features |= NETIF_F_FSO |
11005 				    NETIF_F_FCOE_CRC;
11006 
11007 		netdev->vlan_features |= NETIF_F_FSO |
11008 					 NETIF_F_FCOE_CRC |
11009 					 NETIF_F_FCOE_MTU;
11010 	}
11011 #endif /* IXGBE_FCOE */
11012 	if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
11013 		netdev->hw_features |= NETIF_F_LRO;
11014 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
11015 		netdev->features |= NETIF_F_LRO;
11016 
11017 	if (ixgbe_check_fw_error(adapter)) {
11018 		err = -EIO;
11019 		goto err_sw_init;
11020 	}
11021 
11022 	/* make sure the EEPROM is good */
11023 	if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
11024 		e_dev_err("The EEPROM Checksum Is Not Valid\n");
11025 		err = -EIO;
11026 		goto err_sw_init;
11027 	}
11028 
11029 	eth_platform_get_mac_address(&adapter->pdev->dev,
11030 				     adapter->hw.mac.perm_addr);
11031 
11032 	memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
11033 
11034 	if (!is_valid_ether_addr(netdev->dev_addr)) {
11035 		e_dev_err("invalid MAC address\n");
11036 		err = -EIO;
11037 		goto err_sw_init;
11038 	}
11039 
11040 	/* Set hw->mac.addr to permanent MAC address */
11041 	ether_addr_copy(hw->mac.addr, hw->mac.perm_addr);
11042 	ixgbe_mac_set_default_filter(adapter);
11043 
11044 	timer_setup(&adapter->service_timer, ixgbe_service_timer, 0);
11045 
11046 	if (ixgbe_removed(hw->hw_addr)) {
11047 		err = -EIO;
11048 		goto err_sw_init;
11049 	}
11050 	INIT_WORK(&adapter->service_task, ixgbe_service_task);
11051 	set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
11052 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
11053 
11054 	err = ixgbe_init_interrupt_scheme(adapter);
11055 	if (err)
11056 		goto err_sw_init;
11057 
11058 	for (i = 0; i < adapter->num_rx_queues; i++)
11059 		u64_stats_init(&adapter->rx_ring[i]->syncp);
11060 	for (i = 0; i < adapter->num_tx_queues; i++)
11061 		u64_stats_init(&adapter->tx_ring[i]->syncp);
11062 	for (i = 0; i < adapter->num_xdp_queues; i++)
11063 		u64_stats_init(&adapter->xdp_ring[i]->syncp);
11064 
11065 	/* WOL not supported for all devices */
11066 	adapter->wol = 0;
11067 	hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
11068 	hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
11069 						pdev->subsystem_device);
11070 	if (hw->wol_enabled)
11071 		adapter->wol = IXGBE_WUFC_MAG;
11072 
11073 	device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
11074 
11075 	/* save off EEPROM version number */
11076 	ixgbe_set_fw_version(adapter);
11077 
11078 	/* pick up the PCI bus settings for reporting later */
11079 	if (ixgbe_pcie_from_parent(hw))
11080 		ixgbe_get_parent_bus_info(adapter);
11081 	else
11082 		 hw->mac.ops.get_bus_info(hw);
11083 
11084 	/* calculate the expected PCIe bandwidth required for optimal
11085 	 * performance. Note that some older parts will never have enough
11086 	 * bandwidth due to being older generation PCIe parts. We clamp these
11087 	 * parts to ensure no warning is displayed if it can't be fixed.
11088 	 */
11089 	switch (hw->mac.type) {
11090 	case ixgbe_mac_82598EB:
11091 		expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
11092 		break;
11093 	default:
11094 		expected_gts = ixgbe_enumerate_functions(adapter) * 10;
11095 		break;
11096 	}
11097 
11098 	/* don't check link if we failed to enumerate functions */
11099 	if (expected_gts > 0)
11100 		ixgbe_check_minimum_link(adapter, expected_gts);
11101 
11102 	err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
11103 	if (err)
11104 		strlcpy(part_str, "Unknown", sizeof(part_str));
11105 	if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
11106 		e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
11107 			   hw->mac.type, hw->phy.type, hw->phy.sfp_type,
11108 			   part_str);
11109 	else
11110 		e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
11111 			   hw->mac.type, hw->phy.type, part_str);
11112 
11113 	e_dev_info("%pM\n", netdev->dev_addr);
11114 
11115 	/* reset the hardware with the new settings */
11116 	err = hw->mac.ops.start_hw(hw);
11117 	if (err == IXGBE_ERR_EEPROM_VERSION) {
11118 		/* We are running on a pre-production device, log a warning */
11119 		e_dev_warn("This device is a pre-production adapter/LOM. "
11120 			   "Please be aware there may be issues associated "
11121 			   "with your hardware.  If you are experiencing "
11122 			   "problems please contact your Intel or hardware "
11123 			   "representative who provided you with this "
11124 			   "hardware.\n");
11125 	}
11126 	strcpy(netdev->name, "eth%d");
11127 	pci_set_drvdata(pdev, adapter);
11128 	err = register_netdev(netdev);
11129 	if (err)
11130 		goto err_register;
11131 
11132 
11133 	/* power down the optics for 82599 SFP+ fiber */
11134 	if (hw->mac.ops.disable_tx_laser)
11135 		hw->mac.ops.disable_tx_laser(hw);
11136 
11137 	/* carrier off reporting is important to ethtool even BEFORE open */
11138 	netif_carrier_off(netdev);
11139 
11140 #ifdef CONFIG_IXGBE_DCA
11141 	if (dca_add_requester(&pdev->dev) == 0) {
11142 		adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
11143 		ixgbe_setup_dca(adapter);
11144 	}
11145 #endif
11146 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
11147 		e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
11148 		for (i = 0; i < adapter->num_vfs; i++)
11149 			ixgbe_vf_configuration(pdev, (i | 0x10000000));
11150 	}
11151 
11152 	/* firmware requires driver version to be 0xFFFFFFFF
11153 	 * since os does not support feature
11154 	 */
11155 	if (hw->mac.ops.set_fw_drv_ver)
11156 		hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF, 0xFF,
11157 					   sizeof(ixgbe_driver_version) - 1,
11158 					   ixgbe_driver_version);
11159 
11160 	/* add san mac addr to netdev */
11161 	ixgbe_add_sanmac_netdev(netdev);
11162 
11163 	e_dev_info("%s\n", ixgbe_default_device_descr);
11164 
11165 #ifdef CONFIG_IXGBE_HWMON
11166 	if (ixgbe_sysfs_init(adapter))
11167 		e_err(probe, "failed to allocate sysfs resources\n");
11168 #endif /* CONFIG_IXGBE_HWMON */
11169 
11170 	ixgbe_dbg_adapter_init(adapter);
11171 
11172 	/* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
11173 	if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
11174 		hw->mac.ops.setup_link(hw,
11175 			IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
11176 			true);
11177 
11178 	ixgbe_mii_bus_init(hw);
11179 
11180 	return 0;
11181 
11182 err_register:
11183 	ixgbe_release_hw_control(adapter);
11184 	ixgbe_clear_interrupt_scheme(adapter);
11185 err_sw_init:
11186 	ixgbe_disable_sriov(adapter);
11187 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
11188 	iounmap(adapter->io_addr);
11189 	kfree(adapter->jump_tables[0]);
11190 	kfree(adapter->mac_table);
11191 	kfree(adapter->rss_key);
11192 	bitmap_free(adapter->af_xdp_zc_qps);
11193 err_ioremap:
11194 	disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11195 	free_netdev(netdev);
11196 err_alloc_etherdev:
11197 	pci_release_mem_regions(pdev);
11198 err_pci_reg:
11199 err_dma:
11200 	if (!adapter || disable_dev)
11201 		pci_disable_device(pdev);
11202 	return err;
11203 }
11204 
11205 /**
11206  * ixgbe_remove - Device Removal Routine
11207  * @pdev: PCI device information struct
11208  *
11209  * ixgbe_remove is called by the PCI subsystem to alert the driver
11210  * that it should release a PCI device.  The could be caused by a
11211  * Hot-Plug event, or because the driver is going to be removed from
11212  * memory.
11213  **/
11214 static void ixgbe_remove(struct pci_dev *pdev)
11215 {
11216 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11217 	struct net_device *netdev;
11218 	bool disable_dev;
11219 	int i;
11220 
11221 	/* if !adapter then we already cleaned up in probe */
11222 	if (!adapter)
11223 		return;
11224 
11225 	netdev  = adapter->netdev;
11226 	ixgbe_dbg_adapter_exit(adapter);
11227 
11228 	set_bit(__IXGBE_REMOVING, &adapter->state);
11229 	cancel_work_sync(&adapter->service_task);
11230 
11231 	if (adapter->mii_bus)
11232 		mdiobus_unregister(adapter->mii_bus);
11233 
11234 #ifdef CONFIG_IXGBE_DCA
11235 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
11236 		adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
11237 		dca_remove_requester(&pdev->dev);
11238 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
11239 				IXGBE_DCA_CTRL_DCA_DISABLE);
11240 	}
11241 
11242 #endif
11243 #ifdef CONFIG_IXGBE_HWMON
11244 	ixgbe_sysfs_exit(adapter);
11245 #endif /* CONFIG_IXGBE_HWMON */
11246 
11247 	/* remove the added san mac */
11248 	ixgbe_del_sanmac_netdev(netdev);
11249 
11250 #ifdef CONFIG_PCI_IOV
11251 	ixgbe_disable_sriov(adapter);
11252 #endif
11253 	if (netdev->reg_state == NETREG_REGISTERED)
11254 		unregister_netdev(netdev);
11255 
11256 	ixgbe_stop_ipsec_offload(adapter);
11257 	ixgbe_clear_interrupt_scheme(adapter);
11258 
11259 	ixgbe_release_hw_control(adapter);
11260 
11261 #ifdef CONFIG_DCB
11262 	kfree(adapter->ixgbe_ieee_pfc);
11263 	kfree(adapter->ixgbe_ieee_ets);
11264 
11265 #endif
11266 	iounmap(adapter->io_addr);
11267 	pci_release_mem_regions(pdev);
11268 
11269 	e_dev_info("complete\n");
11270 
11271 	for (i = 0; i < IXGBE_MAX_LINK_HANDLE; i++) {
11272 		if (adapter->jump_tables[i]) {
11273 			kfree(adapter->jump_tables[i]->input);
11274 			kfree(adapter->jump_tables[i]->mask);
11275 		}
11276 		kfree(adapter->jump_tables[i]);
11277 	}
11278 
11279 	kfree(adapter->mac_table);
11280 	kfree(adapter->rss_key);
11281 	bitmap_free(adapter->af_xdp_zc_qps);
11282 	disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11283 	free_netdev(netdev);
11284 
11285 	pci_disable_pcie_error_reporting(pdev);
11286 
11287 	if (disable_dev)
11288 		pci_disable_device(pdev);
11289 }
11290 
11291 /**
11292  * ixgbe_io_error_detected - called when PCI error is detected
11293  * @pdev: Pointer to PCI device
11294  * @state: The current pci connection state
11295  *
11296  * This function is called after a PCI bus error affecting
11297  * this device has been detected.
11298  */
11299 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
11300 						pci_channel_state_t state)
11301 {
11302 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11303 	struct net_device *netdev = adapter->netdev;
11304 
11305 #ifdef CONFIG_PCI_IOV
11306 	struct ixgbe_hw *hw = &adapter->hw;
11307 	struct pci_dev *bdev, *vfdev;
11308 	u32 dw0, dw1, dw2, dw3;
11309 	int vf, pos;
11310 	u16 req_id, pf_func;
11311 
11312 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
11313 	    adapter->num_vfs == 0)
11314 		goto skip_bad_vf_detection;
11315 
11316 	bdev = pdev->bus->self;
11317 	while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
11318 		bdev = bdev->bus->self;
11319 
11320 	if (!bdev)
11321 		goto skip_bad_vf_detection;
11322 
11323 	pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
11324 	if (!pos)
11325 		goto skip_bad_vf_detection;
11326 
11327 	dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
11328 	dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
11329 	dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
11330 	dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
11331 	if (ixgbe_removed(hw->hw_addr))
11332 		goto skip_bad_vf_detection;
11333 
11334 	req_id = dw1 >> 16;
11335 	/* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
11336 	if (!(req_id & 0x0080))
11337 		goto skip_bad_vf_detection;
11338 
11339 	pf_func = req_id & 0x01;
11340 	if ((pf_func & 1) == (pdev->devfn & 1)) {
11341 		unsigned int device_id;
11342 
11343 		vf = (req_id & 0x7F) >> 1;
11344 		e_dev_err("VF %d has caused a PCIe error\n", vf);
11345 		e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
11346 				"%8.8x\tdw3: %8.8x\n",
11347 		dw0, dw1, dw2, dw3);
11348 		switch (adapter->hw.mac.type) {
11349 		case ixgbe_mac_82599EB:
11350 			device_id = IXGBE_82599_VF_DEVICE_ID;
11351 			break;
11352 		case ixgbe_mac_X540:
11353 			device_id = IXGBE_X540_VF_DEVICE_ID;
11354 			break;
11355 		case ixgbe_mac_X550:
11356 			device_id = IXGBE_DEV_ID_X550_VF;
11357 			break;
11358 		case ixgbe_mac_X550EM_x:
11359 			device_id = IXGBE_DEV_ID_X550EM_X_VF;
11360 			break;
11361 		case ixgbe_mac_x550em_a:
11362 			device_id = IXGBE_DEV_ID_X550EM_A_VF;
11363 			break;
11364 		default:
11365 			device_id = 0;
11366 			break;
11367 		}
11368 
11369 		/* Find the pci device of the offending VF */
11370 		vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
11371 		while (vfdev) {
11372 			if (vfdev->devfn == (req_id & 0xFF))
11373 				break;
11374 			vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
11375 					       device_id, vfdev);
11376 		}
11377 		/*
11378 		 * There's a slim chance the VF could have been hot plugged,
11379 		 * so if it is no longer present we don't need to issue the
11380 		 * VFLR.  Just clean up the AER in that case.
11381 		 */
11382 		if (vfdev) {
11383 			pcie_flr(vfdev);
11384 			/* Free device reference count */
11385 			pci_dev_put(vfdev);
11386 		}
11387 	}
11388 
11389 	/*
11390 	 * Even though the error may have occurred on the other port
11391 	 * we still need to increment the vf error reference count for
11392 	 * both ports because the I/O resume function will be called
11393 	 * for both of them.
11394 	 */
11395 	adapter->vferr_refcount++;
11396 
11397 	return PCI_ERS_RESULT_RECOVERED;
11398 
11399 skip_bad_vf_detection:
11400 #endif /* CONFIG_PCI_IOV */
11401 	if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
11402 		return PCI_ERS_RESULT_DISCONNECT;
11403 
11404 	if (!netif_device_present(netdev))
11405 		return PCI_ERS_RESULT_DISCONNECT;
11406 
11407 	rtnl_lock();
11408 	netif_device_detach(netdev);
11409 
11410 	if (netif_running(netdev))
11411 		ixgbe_close_suspend(adapter);
11412 
11413 	if (state == pci_channel_io_perm_failure) {
11414 		rtnl_unlock();
11415 		return PCI_ERS_RESULT_DISCONNECT;
11416 	}
11417 
11418 	if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
11419 		pci_disable_device(pdev);
11420 	rtnl_unlock();
11421 
11422 	/* Request a slot reset. */
11423 	return PCI_ERS_RESULT_NEED_RESET;
11424 }
11425 
11426 /**
11427  * ixgbe_io_slot_reset - called after the pci bus has been reset.
11428  * @pdev: Pointer to PCI device
11429  *
11430  * Restart the card from scratch, as if from a cold-boot.
11431  */
11432 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
11433 {
11434 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11435 	pci_ers_result_t result;
11436 
11437 	if (pci_enable_device_mem(pdev)) {
11438 		e_err(probe, "Cannot re-enable PCI device after reset.\n");
11439 		result = PCI_ERS_RESULT_DISCONNECT;
11440 	} else {
11441 		smp_mb__before_atomic();
11442 		clear_bit(__IXGBE_DISABLED, &adapter->state);
11443 		adapter->hw.hw_addr = adapter->io_addr;
11444 		pci_set_master(pdev);
11445 		pci_restore_state(pdev);
11446 		pci_save_state(pdev);
11447 
11448 		pci_wake_from_d3(pdev, false);
11449 
11450 		ixgbe_reset(adapter);
11451 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
11452 		result = PCI_ERS_RESULT_RECOVERED;
11453 	}
11454 
11455 	return result;
11456 }
11457 
11458 /**
11459  * ixgbe_io_resume - called when traffic can start flowing again.
11460  * @pdev: Pointer to PCI device
11461  *
11462  * This callback is called when the error recovery driver tells us that
11463  * its OK to resume normal operation.
11464  */
11465 static void ixgbe_io_resume(struct pci_dev *pdev)
11466 {
11467 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11468 	struct net_device *netdev = adapter->netdev;
11469 
11470 #ifdef CONFIG_PCI_IOV
11471 	if (adapter->vferr_refcount) {
11472 		e_info(drv, "Resuming after VF err\n");
11473 		adapter->vferr_refcount--;
11474 		return;
11475 	}
11476 
11477 #endif
11478 	rtnl_lock();
11479 	if (netif_running(netdev))
11480 		ixgbe_open(netdev);
11481 
11482 	netif_device_attach(netdev);
11483 	rtnl_unlock();
11484 }
11485 
11486 static const struct pci_error_handlers ixgbe_err_handler = {
11487 	.error_detected = ixgbe_io_error_detected,
11488 	.slot_reset = ixgbe_io_slot_reset,
11489 	.resume = ixgbe_io_resume,
11490 };
11491 
11492 static struct pci_driver ixgbe_driver = {
11493 	.name     = ixgbe_driver_name,
11494 	.id_table = ixgbe_pci_tbl,
11495 	.probe    = ixgbe_probe,
11496 	.remove   = ixgbe_remove,
11497 #ifdef CONFIG_PM
11498 	.suspend  = ixgbe_suspend,
11499 	.resume   = ixgbe_resume,
11500 #endif
11501 	.shutdown = ixgbe_shutdown,
11502 	.sriov_configure = ixgbe_pci_sriov_configure,
11503 	.err_handler = &ixgbe_err_handler
11504 };
11505 
11506 /**
11507  * ixgbe_init_module - Driver Registration Routine
11508  *
11509  * ixgbe_init_module is the first routine called when the driver is
11510  * loaded. All it does is register with the PCI subsystem.
11511  **/
11512 static int __init ixgbe_init_module(void)
11513 {
11514 	int ret;
11515 	pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
11516 	pr_info("%s\n", ixgbe_copyright);
11517 
11518 	ixgbe_wq = create_singlethread_workqueue(ixgbe_driver_name);
11519 	if (!ixgbe_wq) {
11520 		pr_err("%s: Failed to create workqueue\n", ixgbe_driver_name);
11521 		return -ENOMEM;
11522 	}
11523 
11524 	ixgbe_dbg_init();
11525 
11526 	ret = pci_register_driver(&ixgbe_driver);
11527 	if (ret) {
11528 		destroy_workqueue(ixgbe_wq);
11529 		ixgbe_dbg_exit();
11530 		return ret;
11531 	}
11532 
11533 #ifdef CONFIG_IXGBE_DCA
11534 	dca_register_notify(&dca_notifier);
11535 #endif
11536 
11537 	return 0;
11538 }
11539 
11540 module_init(ixgbe_init_module);
11541 
11542 /**
11543  * ixgbe_exit_module - Driver Exit Cleanup Routine
11544  *
11545  * ixgbe_exit_module is called just before the driver is removed
11546  * from memory.
11547  **/
11548 static void __exit ixgbe_exit_module(void)
11549 {
11550 #ifdef CONFIG_IXGBE_DCA
11551 	dca_unregister_notify(&dca_notifier);
11552 #endif
11553 	pci_unregister_driver(&ixgbe_driver);
11554 
11555 	ixgbe_dbg_exit();
11556 	if (ixgbe_wq) {
11557 		destroy_workqueue(ixgbe_wq);
11558 		ixgbe_wq = NULL;
11559 	}
11560 }
11561 
11562 #ifdef CONFIG_IXGBE_DCA
11563 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
11564 			    void *p)
11565 {
11566 	int ret_val;
11567 
11568 	ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
11569 					 __ixgbe_notify_dca);
11570 
11571 	return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
11572 }
11573 
11574 #endif /* CONFIG_IXGBE_DCA */
11575 
11576 module_exit(ixgbe_exit_module);
11577 
11578 /* ixgbe_main.c */
11579