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 <generated/utsrelease.h>
32 #include <scsi/fc/fc_fcoe.h>
33 #include <net/udp_tunnel.h>
34 #include <net/pkt_cls.h>
35 #include <net/tc_act/tc_gact.h>
36 #include <net/tc_act/tc_mirred.h>
37 #include <net/vxlan.h>
38 #include <net/mpls.h>
39 #include <net/xdp_sock_drv.h>
40 #include <net/xfrm.h>
41 
42 #include "ixgbe.h"
43 #include "ixgbe_common.h"
44 #include "ixgbe_dcb_82599.h"
45 #include "ixgbe_phy.h"
46 #include "ixgbe_sriov.h"
47 #include "ixgbe_model.h"
48 #include "ixgbe_txrx_common.h"
49 
50 char ixgbe_driver_name[] = "ixgbe";
51 static const char ixgbe_driver_string[] =
52 			      "Intel(R) 10 Gigabit PCI Express Network Driver";
53 #ifdef IXGBE_FCOE
54 char ixgbe_default_device_descr[] =
55 			      "Intel(R) 10 Gigabit Network Connection";
56 #else
57 static char ixgbe_default_device_descr[] =
58 			      "Intel(R) 10 Gigabit Network Connection";
59 #endif
60 static const char ixgbe_copyright[] =
61 				"Copyright (c) 1999-2016 Intel Corporation.";
62 
63 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";
64 
65 static const struct ixgbe_info *ixgbe_info_tbl[] = {
66 	[board_82598]		= &ixgbe_82598_info,
67 	[board_82599]		= &ixgbe_82599_info,
68 	[board_X540]		= &ixgbe_X540_info,
69 	[board_X550]		= &ixgbe_X550_info,
70 	[board_X550EM_x]	= &ixgbe_X550EM_x_info,
71 	[board_x550em_x_fw]	= &ixgbe_x550em_x_fw_info,
72 	[board_x550em_a]	= &ixgbe_x550em_a_info,
73 	[board_x550em_a_fw]	= &ixgbe_x550em_a_fw_info,
74 };
75 
76 /* ixgbe_pci_tbl - PCI Device ID Table
77  *
78  * Wildcard entries (PCI_ANY_ID) should come last
79  * Last entry must be all 0s
80  *
81  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
82  *   Class, Class Mask, private data (not used) }
83  */
84 static const struct pci_device_id ixgbe_pci_tbl[] = {
85 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
86 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
87 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
88 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
89 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
90 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
91 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
92 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
93 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
94 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
95 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
96 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
97 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
98 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
99 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
100 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
101 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
102 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
103 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
104 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
105 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
106 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
107 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
108 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
109 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
110 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
111 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_QSFP_SF_QP), board_82599 },
112 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
113 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
114 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
115 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T), board_X550},
116 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T1), board_X550},
117 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KX4), board_X550EM_x},
118 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_XFI), board_X550EM_x},
119 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KR), board_X550EM_x},
120 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_10G_T), board_X550EM_x},
121 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_SFP), board_X550EM_x},
122 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_1G_T), board_x550em_x_fw},
123 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR), board_x550em_a },
124 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR_L), board_x550em_a },
125 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP_N), board_x550em_a },
126 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII), board_x550em_a },
127 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII_L), board_x550em_a },
128 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_10G_T), board_x550em_a},
129 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP), board_x550em_a },
130 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T), board_x550em_a_fw },
131 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T_L), board_x550em_a_fw },
132 	/* required last entry */
133 	{0, }
134 };
135 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
136 
137 #ifdef CONFIG_IXGBE_DCA
138 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
139 			    void *p);
140 static struct notifier_block dca_notifier = {
141 	.notifier_call = ixgbe_notify_dca,
142 	.next          = NULL,
143 	.priority      = 0
144 };
145 #endif
146 
147 #ifdef CONFIG_PCI_IOV
148 static unsigned int max_vfs;
149 module_param(max_vfs, uint, 0);
150 MODULE_PARM_DESC(max_vfs,
151 		 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63. (Deprecated)");
152 #endif /* CONFIG_PCI_IOV */
153 
154 static unsigned int allow_unsupported_sfp;
155 module_param(allow_unsupported_sfp, uint, 0);
156 MODULE_PARM_DESC(allow_unsupported_sfp,
157 		 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
158 
159 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
160 static int debug = -1;
161 module_param(debug, int, 0);
162 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
163 
164 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
165 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
166 MODULE_LICENSE("GPL v2");
167 
168 static struct workqueue_struct *ixgbe_wq;
169 
170 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev);
171 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *);
172 
173 static const struct net_device_ops ixgbe_netdev_ops;
174 
175 static bool netif_is_ixgbe(struct net_device *dev)
176 {
177 	return dev && (dev->netdev_ops == &ixgbe_netdev_ops);
178 }
179 
180 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
181 					  u32 reg, u16 *value)
182 {
183 	struct pci_dev *parent_dev;
184 	struct pci_bus *parent_bus;
185 
186 	parent_bus = adapter->pdev->bus->parent;
187 	if (!parent_bus)
188 		return -1;
189 
190 	parent_dev = parent_bus->self;
191 	if (!parent_dev)
192 		return -1;
193 
194 	if (!pci_is_pcie(parent_dev))
195 		return -1;
196 
197 	pcie_capability_read_word(parent_dev, reg, value);
198 	if (*value == IXGBE_FAILED_READ_CFG_WORD &&
199 	    ixgbe_check_cfg_remove(&adapter->hw, parent_dev))
200 		return -1;
201 	return 0;
202 }
203 
204 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
205 {
206 	struct ixgbe_hw *hw = &adapter->hw;
207 	u16 link_status = 0;
208 	int err;
209 
210 	hw->bus.type = ixgbe_bus_type_pci_express;
211 
212 	/* Get the negotiated link width and speed from PCI config space of the
213 	 * parent, as this device is behind a switch
214 	 */
215 	err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
216 
217 	/* assume caller will handle error case */
218 	if (err)
219 		return err;
220 
221 	hw->bus.width = ixgbe_convert_bus_width(link_status);
222 	hw->bus.speed = ixgbe_convert_bus_speed(link_status);
223 
224 	return 0;
225 }
226 
227 /**
228  * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
229  * @hw: hw specific details
230  *
231  * This function is used by probe to determine whether a device's PCI-Express
232  * bandwidth details should be gathered from the parent bus instead of from the
233  * device. Used to ensure that various locations all have the correct device ID
234  * checks.
235  */
236 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
237 {
238 	switch (hw->device_id) {
239 	case IXGBE_DEV_ID_82599_SFP_SF_QP:
240 	case IXGBE_DEV_ID_82599_QSFP_SF_QP:
241 		return true;
242 	default:
243 		return false;
244 	}
245 }
246 
247 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
248 				     int expected_gts)
249 {
250 	struct ixgbe_hw *hw = &adapter->hw;
251 	struct pci_dev *pdev;
252 
253 	/* Some devices are not connected over PCIe and thus do not negotiate
254 	 * speed. These devices do not have valid bus info, and thus any report
255 	 * we generate may not be correct.
256 	 */
257 	if (hw->bus.type == ixgbe_bus_type_internal)
258 		return;
259 
260 	/* determine whether to use the parent device */
261 	if (ixgbe_pcie_from_parent(&adapter->hw))
262 		pdev = adapter->pdev->bus->parent->self;
263 	else
264 		pdev = adapter->pdev;
265 
266 	pcie_print_link_status(pdev);
267 }
268 
269 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
270 {
271 	if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
272 	    !test_bit(__IXGBE_REMOVING, &adapter->state) &&
273 	    !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
274 		queue_work(ixgbe_wq, &adapter->service_task);
275 }
276 
277 static void ixgbe_remove_adapter(struct ixgbe_hw *hw)
278 {
279 	struct ixgbe_adapter *adapter = hw->back;
280 
281 	if (!hw->hw_addr)
282 		return;
283 	hw->hw_addr = NULL;
284 	e_dev_err("Adapter removed\n");
285 	if (test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
286 		ixgbe_service_event_schedule(adapter);
287 }
288 
289 static u32 ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg)
290 {
291 	u8 __iomem *reg_addr;
292 	u32 value;
293 	int i;
294 
295 	reg_addr = READ_ONCE(hw->hw_addr);
296 	if (ixgbe_removed(reg_addr))
297 		return IXGBE_FAILED_READ_REG;
298 
299 	/* Register read of 0xFFFFFFF can indicate the adapter has been removed,
300 	 * so perform several status register reads to determine if the adapter
301 	 * has been removed.
302 	 */
303 	for (i = 0; i < IXGBE_FAILED_READ_RETRIES; i++) {
304 		value = readl(reg_addr + IXGBE_STATUS);
305 		if (value != IXGBE_FAILED_READ_REG)
306 			break;
307 		mdelay(3);
308 	}
309 
310 	if (value == IXGBE_FAILED_READ_REG)
311 		ixgbe_remove_adapter(hw);
312 	else
313 		value = readl(reg_addr + reg);
314 	return value;
315 }
316 
317 /**
318  * ixgbe_read_reg - Read from device register
319  * @hw: hw specific details
320  * @reg: offset of register to read
321  *
322  * Returns : value read or IXGBE_FAILED_READ_REG if removed
323  *
324  * This function is used to read device registers. It checks for device
325  * removal by confirming any read that returns all ones by checking the
326  * status register value for all ones. This function avoids reading from
327  * the hardware if a removal was previously detected in which case it
328  * returns IXGBE_FAILED_READ_REG (all ones).
329  */
330 u32 ixgbe_read_reg(struct ixgbe_hw *hw, u32 reg)
331 {
332 	u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
333 	u32 value;
334 
335 	if (ixgbe_removed(reg_addr))
336 		return IXGBE_FAILED_READ_REG;
337 	if (unlikely(hw->phy.nw_mng_if_sel &
338 		     IXGBE_NW_MNG_IF_SEL_SGMII_ENABLE)) {
339 		struct ixgbe_adapter *adapter;
340 		int i;
341 
342 		for (i = 0; i < 200; ++i) {
343 			value = readl(reg_addr + IXGBE_MAC_SGMII_BUSY);
344 			if (likely(!value))
345 				goto writes_completed;
346 			if (value == IXGBE_FAILED_READ_REG) {
347 				ixgbe_remove_adapter(hw);
348 				return IXGBE_FAILED_READ_REG;
349 			}
350 			udelay(5);
351 		}
352 
353 		adapter = hw->back;
354 		e_warn(hw, "register writes incomplete %08x\n", value);
355 	}
356 
357 writes_completed:
358 	value = readl(reg_addr + reg);
359 	if (unlikely(value == IXGBE_FAILED_READ_REG))
360 		value = ixgbe_check_remove(hw, reg);
361 	return value;
362 }
363 
364 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev)
365 {
366 	u16 value;
367 
368 	pci_read_config_word(pdev, PCI_VENDOR_ID, &value);
369 	if (value == IXGBE_FAILED_READ_CFG_WORD) {
370 		ixgbe_remove_adapter(hw);
371 		return true;
372 	}
373 	return false;
374 }
375 
376 u16 ixgbe_read_pci_cfg_word(struct ixgbe_hw *hw, u32 reg)
377 {
378 	struct ixgbe_adapter *adapter = hw->back;
379 	u16 value;
380 
381 	if (ixgbe_removed(hw->hw_addr))
382 		return IXGBE_FAILED_READ_CFG_WORD;
383 	pci_read_config_word(adapter->pdev, reg, &value);
384 	if (value == IXGBE_FAILED_READ_CFG_WORD &&
385 	    ixgbe_check_cfg_remove(hw, adapter->pdev))
386 		return IXGBE_FAILED_READ_CFG_WORD;
387 	return value;
388 }
389 
390 #ifdef CONFIG_PCI_IOV
391 static u32 ixgbe_read_pci_cfg_dword(struct ixgbe_hw *hw, u32 reg)
392 {
393 	struct ixgbe_adapter *adapter = hw->back;
394 	u32 value;
395 
396 	if (ixgbe_removed(hw->hw_addr))
397 		return IXGBE_FAILED_READ_CFG_DWORD;
398 	pci_read_config_dword(adapter->pdev, reg, &value);
399 	if (value == IXGBE_FAILED_READ_CFG_DWORD &&
400 	    ixgbe_check_cfg_remove(hw, adapter->pdev))
401 		return IXGBE_FAILED_READ_CFG_DWORD;
402 	return value;
403 }
404 #endif /* CONFIG_PCI_IOV */
405 
406 void ixgbe_write_pci_cfg_word(struct ixgbe_hw *hw, u32 reg, u16 value)
407 {
408 	struct ixgbe_adapter *adapter = hw->back;
409 
410 	if (ixgbe_removed(hw->hw_addr))
411 		return;
412 	pci_write_config_word(adapter->pdev, reg, value);
413 }
414 
415 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
416 {
417 	BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
418 
419 	/* flush memory to make sure state is correct before next watchdog */
420 	smp_mb__before_atomic();
421 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
422 }
423 
424 struct ixgbe_reg_info {
425 	u32 ofs;
426 	char *name;
427 };
428 
429 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
430 
431 	/* General Registers */
432 	{IXGBE_CTRL, "CTRL"},
433 	{IXGBE_STATUS, "STATUS"},
434 	{IXGBE_CTRL_EXT, "CTRL_EXT"},
435 
436 	/* Interrupt Registers */
437 	{IXGBE_EICR, "EICR"},
438 
439 	/* RX Registers */
440 	{IXGBE_SRRCTL(0), "SRRCTL"},
441 	{IXGBE_DCA_RXCTRL(0), "DRXCTL"},
442 	{IXGBE_RDLEN(0), "RDLEN"},
443 	{IXGBE_RDH(0), "RDH"},
444 	{IXGBE_RDT(0), "RDT"},
445 	{IXGBE_RXDCTL(0), "RXDCTL"},
446 	{IXGBE_RDBAL(0), "RDBAL"},
447 	{IXGBE_RDBAH(0), "RDBAH"},
448 
449 	/* TX Registers */
450 	{IXGBE_TDBAL(0), "TDBAL"},
451 	{IXGBE_TDBAH(0), "TDBAH"},
452 	{IXGBE_TDLEN(0), "TDLEN"},
453 	{IXGBE_TDH(0), "TDH"},
454 	{IXGBE_TDT(0), "TDT"},
455 	{IXGBE_TXDCTL(0), "TXDCTL"},
456 
457 	/* List Terminator */
458 	{ .name = NULL }
459 };
460 
461 
462 /*
463  * ixgbe_regdump - register printout routine
464  */
465 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
466 {
467 	int i;
468 	char rname[16];
469 	u32 regs[64];
470 
471 	switch (reginfo->ofs) {
472 	case IXGBE_SRRCTL(0):
473 		for (i = 0; i < 64; i++)
474 			regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
475 		break;
476 	case IXGBE_DCA_RXCTRL(0):
477 		for (i = 0; i < 64; i++)
478 			regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
479 		break;
480 	case IXGBE_RDLEN(0):
481 		for (i = 0; i < 64; i++)
482 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
483 		break;
484 	case IXGBE_RDH(0):
485 		for (i = 0; i < 64; i++)
486 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
487 		break;
488 	case IXGBE_RDT(0):
489 		for (i = 0; i < 64; i++)
490 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
491 		break;
492 	case IXGBE_RXDCTL(0):
493 		for (i = 0; i < 64; i++)
494 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
495 		break;
496 	case IXGBE_RDBAL(0):
497 		for (i = 0; i < 64; i++)
498 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
499 		break;
500 	case IXGBE_RDBAH(0):
501 		for (i = 0; i < 64; i++)
502 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
503 		break;
504 	case IXGBE_TDBAL(0):
505 		for (i = 0; i < 64; i++)
506 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
507 		break;
508 	case IXGBE_TDBAH(0):
509 		for (i = 0; i < 64; i++)
510 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
511 		break;
512 	case IXGBE_TDLEN(0):
513 		for (i = 0; i < 64; i++)
514 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
515 		break;
516 	case IXGBE_TDH(0):
517 		for (i = 0; i < 64; i++)
518 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
519 		break;
520 	case IXGBE_TDT(0):
521 		for (i = 0; i < 64; i++)
522 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
523 		break;
524 	case IXGBE_TXDCTL(0):
525 		for (i = 0; i < 64; i++)
526 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
527 		break;
528 	default:
529 		pr_info("%-15s %08x\n",
530 			reginfo->name, IXGBE_READ_REG(hw, reginfo->ofs));
531 		return;
532 	}
533 
534 	i = 0;
535 	while (i < 64) {
536 		int j;
537 		char buf[9 * 8 + 1];
538 		char *p = buf;
539 
540 		snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i, i + 7);
541 		for (j = 0; j < 8; j++)
542 			p += sprintf(p, " %08x", regs[i++]);
543 		pr_err("%-15s%s\n", rname, buf);
544 	}
545 
546 }
547 
548 static void ixgbe_print_buffer(struct ixgbe_ring *ring, int n)
549 {
550 	struct ixgbe_tx_buffer *tx_buffer;
551 
552 	tx_buffer = &ring->tx_buffer_info[ring->next_to_clean];
553 	pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
554 		n, ring->next_to_use, ring->next_to_clean,
555 		(u64)dma_unmap_addr(tx_buffer, dma),
556 		dma_unmap_len(tx_buffer, len),
557 		tx_buffer->next_to_watch,
558 		(u64)tx_buffer->time_stamp);
559 }
560 
561 /*
562  * ixgbe_dump - Print registers, tx-rings and rx-rings
563  */
564 static void ixgbe_dump(struct ixgbe_adapter *adapter)
565 {
566 	struct net_device *netdev = adapter->netdev;
567 	struct ixgbe_hw *hw = &adapter->hw;
568 	struct ixgbe_reg_info *reginfo;
569 	int n = 0;
570 	struct ixgbe_ring *ring;
571 	struct ixgbe_tx_buffer *tx_buffer;
572 	union ixgbe_adv_tx_desc *tx_desc;
573 	struct my_u0 { u64 a; u64 b; } *u0;
574 	struct ixgbe_ring *rx_ring;
575 	union ixgbe_adv_rx_desc *rx_desc;
576 	struct ixgbe_rx_buffer *rx_buffer_info;
577 	int i = 0;
578 
579 	if (!netif_msg_hw(adapter))
580 		return;
581 
582 	/* Print netdevice Info */
583 	if (netdev) {
584 		dev_info(&adapter->pdev->dev, "Net device Info\n");
585 		pr_info("Device Name     state            "
586 			"trans_start\n");
587 		pr_info("%-15s %016lX %016lX\n",
588 			netdev->name,
589 			netdev->state,
590 			dev_trans_start(netdev));
591 	}
592 
593 	/* Print Registers */
594 	dev_info(&adapter->pdev->dev, "Register Dump\n");
595 	pr_info(" Register Name   Value\n");
596 	for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
597 	     reginfo->name; reginfo++) {
598 		ixgbe_regdump(hw, reginfo);
599 	}
600 
601 	/* Print TX Ring Summary */
602 	if (!netdev || !netif_running(netdev))
603 		return;
604 
605 	dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
606 	pr_info(" %s     %s              %s        %s\n",
607 		"Queue [NTU] [NTC] [bi(ntc)->dma  ]",
608 		"leng", "ntw", "timestamp");
609 	for (n = 0; n < adapter->num_tx_queues; n++) {
610 		ring = adapter->tx_ring[n];
611 		ixgbe_print_buffer(ring, n);
612 	}
613 
614 	for (n = 0; n < adapter->num_xdp_queues; n++) {
615 		ring = adapter->xdp_ring[n];
616 		ixgbe_print_buffer(ring, n);
617 	}
618 
619 	/* Print TX Rings */
620 	if (!netif_msg_tx_done(adapter))
621 		goto rx_ring_summary;
622 
623 	dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
624 
625 	/* Transmit Descriptor Formats
626 	 *
627 	 * 82598 Advanced Transmit Descriptor
628 	 *   +--------------------------------------------------------------+
629 	 * 0 |         Buffer Address [63:0]                                |
630 	 *   +--------------------------------------------------------------+
631 	 * 8 |  PAYLEN  | POPTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
632 	 *   +--------------------------------------------------------------+
633 	 *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
634 	 *
635 	 * 82598 Advanced Transmit Descriptor (Write-Back Format)
636 	 *   +--------------------------------------------------------------+
637 	 * 0 |                          RSV [63:0]                          |
638 	 *   +--------------------------------------------------------------+
639 	 * 8 |            RSV           |  STA  |          NXTSEQ           |
640 	 *   +--------------------------------------------------------------+
641 	 *   63                       36 35   32 31                         0
642 	 *
643 	 * 82599+ Advanced Transmit Descriptor
644 	 *   +--------------------------------------------------------------+
645 	 * 0 |         Buffer Address [63:0]                                |
646 	 *   +--------------------------------------------------------------+
647 	 * 8 |PAYLEN  |POPTS|CC|IDX  |STA  |DCMD  |DTYP |MAC  |RSV  |DTALEN |
648 	 *   +--------------------------------------------------------------+
649 	 *   63     46 45 40 39 38 36 35 32 31  24 23 20 19 18 17 16 15     0
650 	 *
651 	 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
652 	 *   +--------------------------------------------------------------+
653 	 * 0 |                          RSV [63:0]                          |
654 	 *   +--------------------------------------------------------------+
655 	 * 8 |            RSV           |  STA  |           RSV             |
656 	 *   +--------------------------------------------------------------+
657 	 *   63                       36 35   32 31                         0
658 	 */
659 
660 	for (n = 0; n < adapter->num_tx_queues; n++) {
661 		ring = adapter->tx_ring[n];
662 		pr_info("------------------------------------\n");
663 		pr_info("TX QUEUE INDEX = %d\n", ring->queue_index);
664 		pr_info("------------------------------------\n");
665 		pr_info("%s%s    %s              %s        %s          %s\n",
666 			"T [desc]     [address 63:0  ] ",
667 			"[PlPOIdStDDt Ln] [bi->dma       ] ",
668 			"leng", "ntw", "timestamp", "bi->skb");
669 
670 		for (i = 0; ring->desc && (i < ring->count); i++) {
671 			tx_desc = IXGBE_TX_DESC(ring, i);
672 			tx_buffer = &ring->tx_buffer_info[i];
673 			u0 = (struct my_u0 *)tx_desc;
674 			if (dma_unmap_len(tx_buffer, len) > 0) {
675 				const char *ring_desc;
676 
677 				if (i == ring->next_to_use &&
678 				    i == ring->next_to_clean)
679 					ring_desc = " NTC/U";
680 				else if (i == ring->next_to_use)
681 					ring_desc = " NTU";
682 				else if (i == ring->next_to_clean)
683 					ring_desc = " NTC";
684 				else
685 					ring_desc = "";
686 				pr_info("T [0x%03X]    %016llX %016llX %016llX %08X %p %016llX %p%s",
687 					i,
688 					le64_to_cpu((__force __le64)u0->a),
689 					le64_to_cpu((__force __le64)u0->b),
690 					(u64)dma_unmap_addr(tx_buffer, dma),
691 					dma_unmap_len(tx_buffer, len),
692 					tx_buffer->next_to_watch,
693 					(u64)tx_buffer->time_stamp,
694 					tx_buffer->skb,
695 					ring_desc);
696 
697 				if (netif_msg_pktdata(adapter) &&
698 				    tx_buffer->skb)
699 					print_hex_dump(KERN_INFO, "",
700 						DUMP_PREFIX_ADDRESS, 16, 1,
701 						tx_buffer->skb->data,
702 						dma_unmap_len(tx_buffer, len),
703 						true);
704 			}
705 		}
706 	}
707 
708 	/* Print RX Rings Summary */
709 rx_ring_summary:
710 	dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
711 	pr_info("Queue [NTU] [NTC]\n");
712 	for (n = 0; n < adapter->num_rx_queues; n++) {
713 		rx_ring = adapter->rx_ring[n];
714 		pr_info("%5d %5X %5X\n",
715 			n, rx_ring->next_to_use, rx_ring->next_to_clean);
716 	}
717 
718 	/* Print RX Rings */
719 	if (!netif_msg_rx_status(adapter))
720 		return;
721 
722 	dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
723 
724 	/* Receive Descriptor Formats
725 	 *
726 	 * 82598 Advanced Receive Descriptor (Read) Format
727 	 *    63                                           1        0
728 	 *    +-----------------------------------------------------+
729 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
730 	 *    +----------------------------------------------+------+
731 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
732 	 *    +-----------------------------------------------------+
733 	 *
734 	 *
735 	 * 82598 Advanced Receive Descriptor (Write-Back) Format
736 	 *
737 	 *   63       48 47    32 31  30      21 20 16 15   4 3     0
738 	 *   +------------------------------------------------------+
739 	 * 0 |       RSS Hash /  |SPH| HDR_LEN  | RSV |Packet|  RSS |
740 	 *   | Packet   | IP     |   |          |     | Type | Type |
741 	 *   | Checksum | Ident  |   |          |     |      |      |
742 	 *   +------------------------------------------------------+
743 	 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
744 	 *   +------------------------------------------------------+
745 	 *   63       48 47    32 31            20 19               0
746 	 *
747 	 * 82599+ Advanced Receive Descriptor (Read) Format
748 	 *    63                                           1        0
749 	 *    +-----------------------------------------------------+
750 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
751 	 *    +----------------------------------------------+------+
752 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
753 	 *    +-----------------------------------------------------+
754 	 *
755 	 *
756 	 * 82599+ Advanced Receive Descriptor (Write-Back) Format
757 	 *
758 	 *   63       48 47    32 31  30      21 20 17 16   4 3     0
759 	 *   +------------------------------------------------------+
760 	 * 0 |RSS / Frag Checksum|SPH| HDR_LEN  |RSC- |Packet|  RSS |
761 	 *   |/ RTT / PCoE_PARAM |   |          | CNT | Type | Type |
762 	 *   |/ Flow Dir Flt ID  |   |          |     |      |      |
763 	 *   +------------------------------------------------------+
764 	 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
765 	 *   +------------------------------------------------------+
766 	 *   63       48 47    32 31          20 19                 0
767 	 */
768 
769 	for (n = 0; n < adapter->num_rx_queues; n++) {
770 		rx_ring = adapter->rx_ring[n];
771 		pr_info("------------------------------------\n");
772 		pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
773 		pr_info("------------------------------------\n");
774 		pr_info("%s%s%s\n",
775 			"R  [desc]      [ PktBuf     A0] ",
776 			"[  HeadBuf   DD] [bi->dma       ] [bi->skb       ] ",
777 			"<-- Adv Rx Read format");
778 		pr_info("%s%s%s\n",
779 			"RWB[desc]      [PcsmIpSHl PtRs] ",
780 			"[vl er S cks ln] ---------------- [bi->skb       ] ",
781 			"<-- Adv Rx Write-Back format");
782 
783 		for (i = 0; i < rx_ring->count; i++) {
784 			const char *ring_desc;
785 
786 			if (i == rx_ring->next_to_use)
787 				ring_desc = " NTU";
788 			else if (i == rx_ring->next_to_clean)
789 				ring_desc = " NTC";
790 			else
791 				ring_desc = "";
792 
793 			rx_buffer_info = &rx_ring->rx_buffer_info[i];
794 			rx_desc = IXGBE_RX_DESC(rx_ring, i);
795 			u0 = (struct my_u0 *)rx_desc;
796 			if (rx_desc->wb.upper.length) {
797 				/* Descriptor Done */
798 				pr_info("RWB[0x%03X]     %016llX %016llX ---------------- %p%s\n",
799 					i,
800 					le64_to_cpu((__force __le64)u0->a),
801 					le64_to_cpu((__force __le64)u0->b),
802 					rx_buffer_info->skb,
803 					ring_desc);
804 			} else {
805 				pr_info("R  [0x%03X]     %016llX %016llX %016llX %p%s\n",
806 					i,
807 					le64_to_cpu((__force __le64)u0->a),
808 					le64_to_cpu((__force __le64)u0->b),
809 					(u64)rx_buffer_info->dma,
810 					rx_buffer_info->skb,
811 					ring_desc);
812 
813 				if (netif_msg_pktdata(adapter) &&
814 				    rx_buffer_info->dma) {
815 					print_hex_dump(KERN_INFO, "",
816 					   DUMP_PREFIX_ADDRESS, 16, 1,
817 					   page_address(rx_buffer_info->page) +
818 						    rx_buffer_info->page_offset,
819 					   ixgbe_rx_bufsz(rx_ring), true);
820 				}
821 			}
822 		}
823 	}
824 }
825 
826 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
827 {
828 	u32 ctrl_ext;
829 
830 	/* Let firmware take over control of h/w */
831 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
832 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
833 			ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
834 }
835 
836 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
837 {
838 	u32 ctrl_ext;
839 
840 	/* Let firmware know the driver has taken over */
841 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
842 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
843 			ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
844 }
845 
846 /**
847  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
848  * @adapter: pointer to adapter struct
849  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
850  * @queue: queue to map the corresponding interrupt to
851  * @msix_vector: the vector to map to the corresponding queue
852  *
853  */
854 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
855 			   u8 queue, u8 msix_vector)
856 {
857 	u32 ivar, index;
858 	struct ixgbe_hw *hw = &adapter->hw;
859 	switch (hw->mac.type) {
860 	case ixgbe_mac_82598EB:
861 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
862 		if (direction == -1)
863 			direction = 0;
864 		index = (((direction * 64) + queue) >> 2) & 0x1F;
865 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
866 		ivar &= ~(0xFF << (8 * (queue & 0x3)));
867 		ivar |= (msix_vector << (8 * (queue & 0x3)));
868 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
869 		break;
870 	case ixgbe_mac_82599EB:
871 	case ixgbe_mac_X540:
872 	case ixgbe_mac_X550:
873 	case ixgbe_mac_X550EM_x:
874 	case ixgbe_mac_x550em_a:
875 		if (direction == -1) {
876 			/* other causes */
877 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
878 			index = ((queue & 1) * 8);
879 			ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
880 			ivar &= ~(0xFF << index);
881 			ivar |= (msix_vector << index);
882 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
883 			break;
884 		} else {
885 			/* tx or rx causes */
886 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
887 			index = ((16 * (queue & 1)) + (8 * direction));
888 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
889 			ivar &= ~(0xFF << index);
890 			ivar |= (msix_vector << index);
891 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
892 			break;
893 		}
894 	default:
895 		break;
896 	}
897 }
898 
899 void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
900 			    u64 qmask)
901 {
902 	u32 mask;
903 
904 	switch (adapter->hw.mac.type) {
905 	case ixgbe_mac_82598EB:
906 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
907 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
908 		break;
909 	case ixgbe_mac_82599EB:
910 	case ixgbe_mac_X540:
911 	case ixgbe_mac_X550:
912 	case ixgbe_mac_X550EM_x:
913 	case ixgbe_mac_x550em_a:
914 		mask = (qmask & 0xFFFFFFFF);
915 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
916 		mask = (qmask >> 32);
917 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
918 		break;
919 	default:
920 		break;
921 	}
922 }
923 
924 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
925 {
926 	struct ixgbe_hw *hw = &adapter->hw;
927 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
928 	int i;
929 	u32 data;
930 
931 	if ((hw->fc.current_mode != ixgbe_fc_full) &&
932 	    (hw->fc.current_mode != ixgbe_fc_rx_pause))
933 		return;
934 
935 	switch (hw->mac.type) {
936 	case ixgbe_mac_82598EB:
937 		data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
938 		break;
939 	default:
940 		data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
941 	}
942 	hwstats->lxoffrxc += data;
943 
944 	/* refill credits (no tx hang) if we received xoff */
945 	if (!data)
946 		return;
947 
948 	for (i = 0; i < adapter->num_tx_queues; i++)
949 		clear_bit(__IXGBE_HANG_CHECK_ARMED,
950 			  &adapter->tx_ring[i]->state);
951 
952 	for (i = 0; i < adapter->num_xdp_queues; i++)
953 		clear_bit(__IXGBE_HANG_CHECK_ARMED,
954 			  &adapter->xdp_ring[i]->state);
955 }
956 
957 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
958 {
959 	struct ixgbe_hw *hw = &adapter->hw;
960 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
961 	u32 xoff[8] = {0};
962 	u8 tc;
963 	int i;
964 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
965 
966 	if (adapter->ixgbe_ieee_pfc)
967 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
968 
969 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
970 		ixgbe_update_xoff_rx_lfc(adapter);
971 		return;
972 	}
973 
974 	/* update stats for each tc, only valid with PFC enabled */
975 	for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
976 		u32 pxoffrxc;
977 
978 		switch (hw->mac.type) {
979 		case ixgbe_mac_82598EB:
980 			pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
981 			break;
982 		default:
983 			pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
984 		}
985 		hwstats->pxoffrxc[i] += pxoffrxc;
986 		/* Get the TC for given UP */
987 		tc = netdev_get_prio_tc_map(adapter->netdev, i);
988 		xoff[tc] += pxoffrxc;
989 	}
990 
991 	/* disarm tx queues that have received xoff frames */
992 	for (i = 0; i < adapter->num_tx_queues; i++) {
993 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
994 
995 		tc = tx_ring->dcb_tc;
996 		if (xoff[tc])
997 			clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
998 	}
999 
1000 	for (i = 0; i < adapter->num_xdp_queues; i++) {
1001 		struct ixgbe_ring *xdp_ring = adapter->xdp_ring[i];
1002 
1003 		tc = xdp_ring->dcb_tc;
1004 		if (xoff[tc])
1005 			clear_bit(__IXGBE_HANG_CHECK_ARMED, &xdp_ring->state);
1006 	}
1007 }
1008 
1009 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
1010 {
1011 	return ring->stats.packets;
1012 }
1013 
1014 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
1015 {
1016 	unsigned int head, tail;
1017 
1018 	head = ring->next_to_clean;
1019 	tail = ring->next_to_use;
1020 
1021 	return ((head <= tail) ? tail : tail + ring->count) - head;
1022 }
1023 
1024 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
1025 {
1026 	u32 tx_done = ixgbe_get_tx_completed(tx_ring);
1027 	u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
1028 	u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
1029 
1030 	clear_check_for_tx_hang(tx_ring);
1031 
1032 	/*
1033 	 * Check for a hung queue, but be thorough. This verifies
1034 	 * that a transmit has been completed since the previous
1035 	 * check AND there is at least one packet pending. The
1036 	 * ARMED bit is set to indicate a potential hang. The
1037 	 * bit is cleared if a pause frame is received to remove
1038 	 * false hang detection due to PFC or 802.3x frames. By
1039 	 * requiring this to fail twice we avoid races with
1040 	 * pfc clearing the ARMED bit and conditions where we
1041 	 * run the check_tx_hang logic with a transmit completion
1042 	 * pending but without time to complete it yet.
1043 	 */
1044 	if (tx_done_old == tx_done && tx_pending)
1045 		/* make sure it is true for two checks in a row */
1046 		return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
1047 					&tx_ring->state);
1048 	/* update completed stats and continue */
1049 	tx_ring->tx_stats.tx_done_old = tx_done;
1050 	/* reset the countdown */
1051 	clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1052 
1053 	return false;
1054 }
1055 
1056 /**
1057  * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1058  * @adapter: driver private struct
1059  **/
1060 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
1061 {
1062 
1063 	/* Do the reset outside of interrupt context */
1064 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1065 		set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
1066 		e_warn(drv, "initiating reset due to tx timeout\n");
1067 		ixgbe_service_event_schedule(adapter);
1068 	}
1069 }
1070 
1071 /**
1072  * ixgbe_tx_maxrate - callback to set the maximum per-queue bitrate
1073  * @netdev: network interface device structure
1074  * @queue_index: Tx queue to set
1075  * @maxrate: desired maximum transmit bitrate
1076  **/
1077 static int ixgbe_tx_maxrate(struct net_device *netdev,
1078 			    int queue_index, u32 maxrate)
1079 {
1080 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
1081 	struct ixgbe_hw *hw = &adapter->hw;
1082 	u32 bcnrc_val = ixgbe_link_mbps(adapter);
1083 
1084 	if (!maxrate)
1085 		return 0;
1086 
1087 	/* Calculate the rate factor values to set */
1088 	bcnrc_val <<= IXGBE_RTTBCNRC_RF_INT_SHIFT;
1089 	bcnrc_val /= maxrate;
1090 
1091 	/* clear everything but the rate factor */
1092 	bcnrc_val &= IXGBE_RTTBCNRC_RF_INT_MASK |
1093 	IXGBE_RTTBCNRC_RF_DEC_MASK;
1094 
1095 	/* enable the rate scheduler */
1096 	bcnrc_val |= IXGBE_RTTBCNRC_RS_ENA;
1097 
1098 	IXGBE_WRITE_REG(hw, IXGBE_RTTDQSEL, queue_index);
1099 	IXGBE_WRITE_REG(hw, IXGBE_RTTBCNRC, bcnrc_val);
1100 
1101 	return 0;
1102 }
1103 
1104 /**
1105  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1106  * @q_vector: structure containing interrupt and ring information
1107  * @tx_ring: tx ring to clean
1108  * @napi_budget: Used to determine if we are in netpoll
1109  **/
1110 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
1111 			       struct ixgbe_ring *tx_ring, int napi_budget)
1112 {
1113 	struct ixgbe_adapter *adapter = q_vector->adapter;
1114 	struct ixgbe_tx_buffer *tx_buffer;
1115 	union ixgbe_adv_tx_desc *tx_desc;
1116 	unsigned int total_bytes = 0, total_packets = 0, total_ipsec = 0;
1117 	unsigned int budget = q_vector->tx.work_limit;
1118 	unsigned int i = tx_ring->next_to_clean;
1119 
1120 	if (test_bit(__IXGBE_DOWN, &adapter->state))
1121 		return true;
1122 
1123 	tx_buffer = &tx_ring->tx_buffer_info[i];
1124 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
1125 	i -= tx_ring->count;
1126 
1127 	do {
1128 		union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1129 
1130 		/* if next_to_watch is not set then there is no work pending */
1131 		if (!eop_desc)
1132 			break;
1133 
1134 		/* prevent any other reads prior to eop_desc */
1135 		smp_rmb();
1136 
1137 		/* if DD is not set pending work has not been completed */
1138 		if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
1139 			break;
1140 
1141 		/* clear next_to_watch to prevent false hangs */
1142 		tx_buffer->next_to_watch = NULL;
1143 
1144 		/* update the statistics for this packet */
1145 		total_bytes += tx_buffer->bytecount;
1146 		total_packets += tx_buffer->gso_segs;
1147 		if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_IPSEC)
1148 			total_ipsec++;
1149 
1150 		/* free the skb */
1151 		if (ring_is_xdp(tx_ring))
1152 			xdp_return_frame(tx_buffer->xdpf);
1153 		else
1154 			napi_consume_skb(tx_buffer->skb, napi_budget);
1155 
1156 		/* unmap skb header data */
1157 		dma_unmap_single(tx_ring->dev,
1158 				 dma_unmap_addr(tx_buffer, dma),
1159 				 dma_unmap_len(tx_buffer, len),
1160 				 DMA_TO_DEVICE);
1161 
1162 		/* clear tx_buffer data */
1163 		dma_unmap_len_set(tx_buffer, len, 0);
1164 
1165 		/* unmap remaining buffers */
1166 		while (tx_desc != eop_desc) {
1167 			tx_buffer++;
1168 			tx_desc++;
1169 			i++;
1170 			if (unlikely(!i)) {
1171 				i -= tx_ring->count;
1172 				tx_buffer = tx_ring->tx_buffer_info;
1173 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1174 			}
1175 
1176 			/* unmap any remaining paged data */
1177 			if (dma_unmap_len(tx_buffer, len)) {
1178 				dma_unmap_page(tx_ring->dev,
1179 					       dma_unmap_addr(tx_buffer, dma),
1180 					       dma_unmap_len(tx_buffer, len),
1181 					       DMA_TO_DEVICE);
1182 				dma_unmap_len_set(tx_buffer, len, 0);
1183 			}
1184 		}
1185 
1186 		/* move us one more past the eop_desc for start of next pkt */
1187 		tx_buffer++;
1188 		tx_desc++;
1189 		i++;
1190 		if (unlikely(!i)) {
1191 			i -= tx_ring->count;
1192 			tx_buffer = tx_ring->tx_buffer_info;
1193 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1194 		}
1195 
1196 		/* issue prefetch for next Tx descriptor */
1197 		prefetch(tx_desc);
1198 
1199 		/* update budget accounting */
1200 		budget--;
1201 	} while (likely(budget));
1202 
1203 	i += tx_ring->count;
1204 	tx_ring->next_to_clean = i;
1205 	u64_stats_update_begin(&tx_ring->syncp);
1206 	tx_ring->stats.bytes += total_bytes;
1207 	tx_ring->stats.packets += total_packets;
1208 	u64_stats_update_end(&tx_ring->syncp);
1209 	q_vector->tx.total_bytes += total_bytes;
1210 	q_vector->tx.total_packets += total_packets;
1211 	adapter->tx_ipsec += total_ipsec;
1212 
1213 	if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1214 		/* schedule immediate reset if we believe we hung */
1215 		struct ixgbe_hw *hw = &adapter->hw;
1216 		e_err(drv, "Detected Tx Unit Hang %s\n"
1217 			"  Tx Queue             <%d>\n"
1218 			"  TDH, TDT             <%x>, <%x>\n"
1219 			"  next_to_use          <%x>\n"
1220 			"  next_to_clean        <%x>\n"
1221 			"tx_buffer_info[next_to_clean]\n"
1222 			"  time_stamp           <%lx>\n"
1223 			"  jiffies              <%lx>\n",
1224 			ring_is_xdp(tx_ring) ? "(XDP)" : "",
1225 			tx_ring->queue_index,
1226 			IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1227 			IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1228 			tx_ring->next_to_use, i,
1229 			tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1230 
1231 		if (!ring_is_xdp(tx_ring))
1232 			netif_stop_subqueue(tx_ring->netdev,
1233 					    tx_ring->queue_index);
1234 
1235 		e_info(probe,
1236 		       "tx hang %d detected on queue %d, resetting adapter\n",
1237 			adapter->tx_timeout_count + 1, tx_ring->queue_index);
1238 
1239 		/* schedule immediate reset if we believe we hung */
1240 		ixgbe_tx_timeout_reset(adapter);
1241 
1242 		/* the adapter is about to reset, no point in enabling stuff */
1243 		return true;
1244 	}
1245 
1246 	if (ring_is_xdp(tx_ring))
1247 		return !!budget;
1248 
1249 	netdev_tx_completed_queue(txring_txq(tx_ring),
1250 				  total_packets, total_bytes);
1251 
1252 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1253 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1254 		     (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1255 		/* Make sure that anybody stopping the queue after this
1256 		 * sees the new next_to_clean.
1257 		 */
1258 		smp_mb();
1259 		if (__netif_subqueue_stopped(tx_ring->netdev,
1260 					     tx_ring->queue_index)
1261 		    && !test_bit(__IXGBE_DOWN, &adapter->state)) {
1262 			netif_wake_subqueue(tx_ring->netdev,
1263 					    tx_ring->queue_index);
1264 			++tx_ring->tx_stats.restart_queue;
1265 		}
1266 	}
1267 
1268 	return !!budget;
1269 }
1270 
1271 #ifdef CONFIG_IXGBE_DCA
1272 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1273 				struct ixgbe_ring *tx_ring,
1274 				int cpu)
1275 {
1276 	struct ixgbe_hw *hw = &adapter->hw;
1277 	u32 txctrl = 0;
1278 	u16 reg_offset;
1279 
1280 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1281 		txctrl = dca3_get_tag(tx_ring->dev, cpu);
1282 
1283 	switch (hw->mac.type) {
1284 	case ixgbe_mac_82598EB:
1285 		reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1286 		break;
1287 	case ixgbe_mac_82599EB:
1288 	case ixgbe_mac_X540:
1289 		reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1290 		txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1291 		break;
1292 	default:
1293 		/* for unknown hardware do not write register */
1294 		return;
1295 	}
1296 
1297 	/*
1298 	 * We can enable relaxed ordering for reads, but not writes when
1299 	 * DCA is enabled.  This is due to a known issue in some chipsets
1300 	 * which will cause the DCA tag to be cleared.
1301 	 */
1302 	txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1303 		  IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1304 		  IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1305 
1306 	IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1307 }
1308 
1309 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1310 				struct ixgbe_ring *rx_ring,
1311 				int cpu)
1312 {
1313 	struct ixgbe_hw *hw = &adapter->hw;
1314 	u32 rxctrl = 0;
1315 	u8 reg_idx = rx_ring->reg_idx;
1316 
1317 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1318 		rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1319 
1320 	switch (hw->mac.type) {
1321 	case ixgbe_mac_82599EB:
1322 	case ixgbe_mac_X540:
1323 		rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1324 		break;
1325 	default:
1326 		break;
1327 	}
1328 
1329 	/*
1330 	 * We can enable relaxed ordering for reads, but not writes when
1331 	 * DCA is enabled.  This is due to a known issue in some chipsets
1332 	 * which will cause the DCA tag to be cleared.
1333 	 */
1334 	rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1335 		  IXGBE_DCA_RXCTRL_DATA_DCA_EN |
1336 		  IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1337 
1338 	IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1339 }
1340 
1341 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1342 {
1343 	struct ixgbe_adapter *adapter = q_vector->adapter;
1344 	struct ixgbe_ring *ring;
1345 	int cpu = get_cpu();
1346 
1347 	if (q_vector->cpu == cpu)
1348 		goto out_no_update;
1349 
1350 	ixgbe_for_each_ring(ring, q_vector->tx)
1351 		ixgbe_update_tx_dca(adapter, ring, cpu);
1352 
1353 	ixgbe_for_each_ring(ring, q_vector->rx)
1354 		ixgbe_update_rx_dca(adapter, ring, cpu);
1355 
1356 	q_vector->cpu = cpu;
1357 out_no_update:
1358 	put_cpu();
1359 }
1360 
1361 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1362 {
1363 	int i;
1364 
1365 	/* always use CB2 mode, difference is masked in the CB driver */
1366 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1367 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1368 				IXGBE_DCA_CTRL_DCA_MODE_CB2);
1369 	else
1370 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1371 				IXGBE_DCA_CTRL_DCA_DISABLE);
1372 
1373 	for (i = 0; i < adapter->num_q_vectors; i++) {
1374 		adapter->q_vector[i]->cpu = -1;
1375 		ixgbe_update_dca(adapter->q_vector[i]);
1376 	}
1377 }
1378 
1379 static int __ixgbe_notify_dca(struct device *dev, void *data)
1380 {
1381 	struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1382 	unsigned long event = *(unsigned long *)data;
1383 
1384 	if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1385 		return 0;
1386 
1387 	switch (event) {
1388 	case DCA_PROVIDER_ADD:
1389 		/* if we're already enabled, don't do it again */
1390 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1391 			break;
1392 		if (dca_add_requester(dev) == 0) {
1393 			adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1394 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1395 					IXGBE_DCA_CTRL_DCA_MODE_CB2);
1396 			break;
1397 		}
1398 		fallthrough; /* DCA is disabled. */
1399 	case DCA_PROVIDER_REMOVE:
1400 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1401 			dca_remove_requester(dev);
1402 			adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1403 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1404 					IXGBE_DCA_CTRL_DCA_DISABLE);
1405 		}
1406 		break;
1407 	}
1408 
1409 	return 0;
1410 }
1411 
1412 #endif /* CONFIG_IXGBE_DCA */
1413 
1414 #define IXGBE_RSS_L4_TYPES_MASK \
1415 	((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
1416 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
1417 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
1418 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
1419 
1420 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1421 				 union ixgbe_adv_rx_desc *rx_desc,
1422 				 struct sk_buff *skb)
1423 {
1424 	u16 rss_type;
1425 
1426 	if (!(ring->netdev->features & NETIF_F_RXHASH))
1427 		return;
1428 
1429 	rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
1430 		   IXGBE_RXDADV_RSSTYPE_MASK;
1431 
1432 	if (!rss_type)
1433 		return;
1434 
1435 	skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1436 		     (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
1437 		     PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1438 }
1439 
1440 #ifdef IXGBE_FCOE
1441 /**
1442  * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1443  * @ring: structure containing ring specific data
1444  * @rx_desc: advanced rx descriptor
1445  *
1446  * Returns : true if it is FCoE pkt
1447  */
1448 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1449 				    union ixgbe_adv_rx_desc *rx_desc)
1450 {
1451 	__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1452 
1453 	return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1454 	       ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1455 		(cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1456 			     IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1457 }
1458 
1459 #endif /* IXGBE_FCOE */
1460 /**
1461  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1462  * @ring: structure containing ring specific data
1463  * @rx_desc: current Rx descriptor being processed
1464  * @skb: skb currently being received and modified
1465  **/
1466 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1467 				     union ixgbe_adv_rx_desc *rx_desc,
1468 				     struct sk_buff *skb)
1469 {
1470 	__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1471 	bool encap_pkt = false;
1472 
1473 	skb_checksum_none_assert(skb);
1474 
1475 	/* Rx csum disabled */
1476 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
1477 		return;
1478 
1479 	/* check for VXLAN and Geneve packets */
1480 	if (pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_VXLAN)) {
1481 		encap_pkt = true;
1482 		skb->encapsulation = 1;
1483 	}
1484 
1485 	/* if IP and error */
1486 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1487 	    ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1488 		ring->rx_stats.csum_err++;
1489 		return;
1490 	}
1491 
1492 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1493 		return;
1494 
1495 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1496 		/*
1497 		 * 82599 errata, UDP frames with a 0 checksum can be marked as
1498 		 * checksum errors.
1499 		 */
1500 		if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1501 		    test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1502 			return;
1503 
1504 		ring->rx_stats.csum_err++;
1505 		return;
1506 	}
1507 
1508 	/* It must be a TCP or UDP packet with a valid checksum */
1509 	skb->ip_summed = CHECKSUM_UNNECESSARY;
1510 	if (encap_pkt) {
1511 		if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_OUTERIPCS))
1512 			return;
1513 
1514 		if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_OUTERIPER)) {
1515 			skb->ip_summed = CHECKSUM_NONE;
1516 			return;
1517 		}
1518 		/* If we checked the outer header let the stack know */
1519 		skb->csum_level = 1;
1520 	}
1521 }
1522 
1523 static unsigned int ixgbe_rx_offset(struct ixgbe_ring *rx_ring)
1524 {
1525 	return ring_uses_build_skb(rx_ring) ? IXGBE_SKB_PAD : 0;
1526 }
1527 
1528 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1529 				    struct ixgbe_rx_buffer *bi)
1530 {
1531 	struct page *page = bi->page;
1532 	dma_addr_t dma;
1533 
1534 	/* since we are recycling buffers we should seldom need to alloc */
1535 	if (likely(page))
1536 		return true;
1537 
1538 	/* alloc new page for storage */
1539 	page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
1540 	if (unlikely(!page)) {
1541 		rx_ring->rx_stats.alloc_rx_page_failed++;
1542 		return false;
1543 	}
1544 
1545 	/* map page for use */
1546 	dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1547 				 ixgbe_rx_pg_size(rx_ring),
1548 				 DMA_FROM_DEVICE,
1549 				 IXGBE_RX_DMA_ATTR);
1550 
1551 	/*
1552 	 * if mapping failed free memory back to system since
1553 	 * there isn't much point in holding memory we can't use
1554 	 */
1555 	if (dma_mapping_error(rx_ring->dev, dma)) {
1556 		__free_pages(page, ixgbe_rx_pg_order(rx_ring));
1557 
1558 		rx_ring->rx_stats.alloc_rx_page_failed++;
1559 		return false;
1560 	}
1561 
1562 	bi->dma = dma;
1563 	bi->page = page;
1564 	bi->page_offset = rx_ring->rx_offset;
1565 	page_ref_add(page, USHRT_MAX - 1);
1566 	bi->pagecnt_bias = USHRT_MAX;
1567 	rx_ring->rx_stats.alloc_rx_page++;
1568 
1569 	return true;
1570 }
1571 
1572 /**
1573  * ixgbe_alloc_rx_buffers - Replace used receive buffers
1574  * @rx_ring: ring to place buffers on
1575  * @cleaned_count: number of buffers to replace
1576  **/
1577 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1578 {
1579 	union ixgbe_adv_rx_desc *rx_desc;
1580 	struct ixgbe_rx_buffer *bi;
1581 	u16 i = rx_ring->next_to_use;
1582 	u16 bufsz;
1583 
1584 	/* nothing to do */
1585 	if (!cleaned_count)
1586 		return;
1587 
1588 	rx_desc = IXGBE_RX_DESC(rx_ring, i);
1589 	bi = &rx_ring->rx_buffer_info[i];
1590 	i -= rx_ring->count;
1591 
1592 	bufsz = ixgbe_rx_bufsz(rx_ring);
1593 
1594 	do {
1595 		if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1596 			break;
1597 
1598 		/* sync the buffer for use by the device */
1599 		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1600 						 bi->page_offset, bufsz,
1601 						 DMA_FROM_DEVICE);
1602 
1603 		/*
1604 		 * Refresh the desc even if buffer_addrs didn't change
1605 		 * because each write-back erases this info.
1606 		 */
1607 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1608 
1609 		rx_desc++;
1610 		bi++;
1611 		i++;
1612 		if (unlikely(!i)) {
1613 			rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1614 			bi = rx_ring->rx_buffer_info;
1615 			i -= rx_ring->count;
1616 		}
1617 
1618 		/* clear the length for the next_to_use descriptor */
1619 		rx_desc->wb.upper.length = 0;
1620 
1621 		cleaned_count--;
1622 	} while (cleaned_count);
1623 
1624 	i += rx_ring->count;
1625 
1626 	if (rx_ring->next_to_use != i) {
1627 		rx_ring->next_to_use = i;
1628 
1629 		/* update next to alloc since we have filled the ring */
1630 		rx_ring->next_to_alloc = i;
1631 
1632 		/* Force memory writes to complete before letting h/w
1633 		 * know there are new descriptors to fetch.  (Only
1634 		 * applicable for weak-ordered memory model archs,
1635 		 * such as IA-64).
1636 		 */
1637 		wmb();
1638 		writel(i, rx_ring->tail);
1639 	}
1640 }
1641 
1642 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1643 				   struct sk_buff *skb)
1644 {
1645 	u16 hdr_len = skb_headlen(skb);
1646 
1647 	/* set gso_size to avoid messing up TCP MSS */
1648 	skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1649 						 IXGBE_CB(skb)->append_cnt);
1650 	skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1651 }
1652 
1653 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1654 				   struct sk_buff *skb)
1655 {
1656 	/* if append_cnt is 0 then frame is not RSC */
1657 	if (!IXGBE_CB(skb)->append_cnt)
1658 		return;
1659 
1660 	rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1661 	rx_ring->rx_stats.rsc_flush++;
1662 
1663 	ixgbe_set_rsc_gso_size(rx_ring, skb);
1664 
1665 	/* gso_size is computed using append_cnt so always clear it last */
1666 	IXGBE_CB(skb)->append_cnt = 0;
1667 }
1668 
1669 /**
1670  * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1671  * @rx_ring: rx descriptor ring packet is being transacted on
1672  * @rx_desc: pointer to the EOP Rx descriptor
1673  * @skb: pointer to current skb being populated
1674  *
1675  * This function checks the ring, descriptor, and packet information in
1676  * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1677  * other fields within the skb.
1678  **/
1679 void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1680 			      union ixgbe_adv_rx_desc *rx_desc,
1681 			      struct sk_buff *skb)
1682 {
1683 	struct net_device *dev = rx_ring->netdev;
1684 	u32 flags = rx_ring->q_vector->adapter->flags;
1685 
1686 	ixgbe_update_rsc_stats(rx_ring, skb);
1687 
1688 	ixgbe_rx_hash(rx_ring, rx_desc, skb);
1689 
1690 	ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1691 
1692 	if (unlikely(flags & IXGBE_FLAG_RX_HWTSTAMP_ENABLED))
1693 		ixgbe_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
1694 
1695 	if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1696 	    ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1697 		u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1698 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1699 	}
1700 
1701 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_STAT_SECP))
1702 		ixgbe_ipsec_rx(rx_ring, rx_desc, skb);
1703 
1704 	/* record Rx queue, or update MACVLAN statistics */
1705 	if (netif_is_ixgbe(dev))
1706 		skb_record_rx_queue(skb, rx_ring->queue_index);
1707 	else
1708 		macvlan_count_rx(netdev_priv(dev), skb->len + ETH_HLEN, true,
1709 				 false);
1710 
1711 	skb->protocol = eth_type_trans(skb, dev);
1712 }
1713 
1714 void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1715 		  struct sk_buff *skb)
1716 {
1717 	napi_gro_receive(&q_vector->napi, skb);
1718 }
1719 
1720 /**
1721  * ixgbe_is_non_eop - process handling of non-EOP buffers
1722  * @rx_ring: Rx ring being processed
1723  * @rx_desc: Rx descriptor for current buffer
1724  * @skb: Current socket buffer containing buffer in progress
1725  *
1726  * This function updates next to clean.  If the buffer is an EOP buffer
1727  * this function exits returning false, otherwise it will place the
1728  * sk_buff in the next buffer to be chained and return true indicating
1729  * that this is in fact a non-EOP buffer.
1730  **/
1731 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1732 			     union ixgbe_adv_rx_desc *rx_desc,
1733 			     struct sk_buff *skb)
1734 {
1735 	u32 ntc = rx_ring->next_to_clean + 1;
1736 
1737 	/* fetch, update, and store next to clean */
1738 	ntc = (ntc < rx_ring->count) ? ntc : 0;
1739 	rx_ring->next_to_clean = ntc;
1740 
1741 	prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1742 
1743 	/* update RSC append count if present */
1744 	if (ring_is_rsc_enabled(rx_ring)) {
1745 		__le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1746 				     cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1747 
1748 		if (unlikely(rsc_enabled)) {
1749 			u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1750 
1751 			rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1752 			IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1753 
1754 			/* update ntc based on RSC value */
1755 			ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1756 			ntc &= IXGBE_RXDADV_NEXTP_MASK;
1757 			ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1758 		}
1759 	}
1760 
1761 	/* if we are the last buffer then there is nothing else to do */
1762 	if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1763 		return false;
1764 
1765 	/* place skb in next buffer to be received */
1766 	rx_ring->rx_buffer_info[ntc].skb = skb;
1767 	rx_ring->rx_stats.non_eop_descs++;
1768 
1769 	return true;
1770 }
1771 
1772 /**
1773  * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1774  * @rx_ring: rx descriptor ring packet is being transacted on
1775  * @skb: pointer to current skb being adjusted
1776  *
1777  * This function is an ixgbe specific version of __pskb_pull_tail.  The
1778  * main difference between this version and the original function is that
1779  * this function can make several assumptions about the state of things
1780  * that allow for significant optimizations versus the standard function.
1781  * As a result we can do things like drop a frag and maintain an accurate
1782  * truesize for the skb.
1783  */
1784 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1785 			    struct sk_buff *skb)
1786 {
1787 	skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
1788 	unsigned char *va;
1789 	unsigned int pull_len;
1790 
1791 	/*
1792 	 * it is valid to use page_address instead of kmap since we are
1793 	 * working with pages allocated out of the lomem pool per
1794 	 * alloc_page(GFP_ATOMIC)
1795 	 */
1796 	va = skb_frag_address(frag);
1797 
1798 	/*
1799 	 * we need the header to contain the greater of either ETH_HLEN or
1800 	 * 60 bytes if the skb->len is less than 60 for skb_pad.
1801 	 */
1802 	pull_len = eth_get_headlen(skb->dev, va, IXGBE_RX_HDR_SIZE);
1803 
1804 	/* align pull length to size of long to optimize memcpy performance */
1805 	skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1806 
1807 	/* update all of the pointers */
1808 	skb_frag_size_sub(frag, pull_len);
1809 	skb_frag_off_add(frag, pull_len);
1810 	skb->data_len -= pull_len;
1811 	skb->tail += pull_len;
1812 }
1813 
1814 /**
1815  * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1816  * @rx_ring: rx descriptor ring packet is being transacted on
1817  * @skb: pointer to current skb being updated
1818  *
1819  * This function provides a basic DMA sync up for the first fragment of an
1820  * skb.  The reason for doing this is that the first fragment cannot be
1821  * unmapped until we have reached the end of packet descriptor for a buffer
1822  * chain.
1823  */
1824 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1825 				struct sk_buff *skb)
1826 {
1827 	if (ring_uses_build_skb(rx_ring)) {
1828 		unsigned long offset = (unsigned long)(skb->data) & ~PAGE_MASK;
1829 
1830 		dma_sync_single_range_for_cpu(rx_ring->dev,
1831 					      IXGBE_CB(skb)->dma,
1832 					      offset,
1833 					      skb_headlen(skb),
1834 					      DMA_FROM_DEVICE);
1835 	} else {
1836 		skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
1837 
1838 		dma_sync_single_range_for_cpu(rx_ring->dev,
1839 					      IXGBE_CB(skb)->dma,
1840 					      skb_frag_off(frag),
1841 					      skb_frag_size(frag),
1842 					      DMA_FROM_DEVICE);
1843 	}
1844 
1845 	/* If the page was released, just unmap it. */
1846 	if (unlikely(IXGBE_CB(skb)->page_released)) {
1847 		dma_unmap_page_attrs(rx_ring->dev, IXGBE_CB(skb)->dma,
1848 				     ixgbe_rx_pg_size(rx_ring),
1849 				     DMA_FROM_DEVICE,
1850 				     IXGBE_RX_DMA_ATTR);
1851 	}
1852 }
1853 
1854 /**
1855  * ixgbe_cleanup_headers - Correct corrupted or empty headers
1856  * @rx_ring: rx descriptor ring packet is being transacted on
1857  * @rx_desc: pointer to the EOP Rx descriptor
1858  * @skb: pointer to current skb being fixed
1859  *
1860  * Check if the skb is valid in the XDP case it will be an error pointer.
1861  * Return true in this case to abort processing and advance to next
1862  * descriptor.
1863  *
1864  * Check for corrupted packet headers caused by senders on the local L2
1865  * embedded NIC switch not setting up their Tx Descriptors right.  These
1866  * should be very rare.
1867  *
1868  * Also address the case where we are pulling data in on pages only
1869  * and as such no data is present in the skb header.
1870  *
1871  * In addition if skb is not at least 60 bytes we need to pad it so that
1872  * it is large enough to qualify as a valid Ethernet frame.
1873  *
1874  * Returns true if an error was encountered and skb was freed.
1875  **/
1876 bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1877 			   union ixgbe_adv_rx_desc *rx_desc,
1878 			   struct sk_buff *skb)
1879 {
1880 	struct net_device *netdev = rx_ring->netdev;
1881 
1882 	/* XDP packets use error pointer so abort at this point */
1883 	if (IS_ERR(skb))
1884 		return true;
1885 
1886 	/* Verify netdev is present, and that packet does not have any
1887 	 * errors that would be unacceptable to the netdev.
1888 	 */
1889 	if (!netdev ||
1890 	    (unlikely(ixgbe_test_staterr(rx_desc,
1891 					 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1892 	     !(netdev->features & NETIF_F_RXALL)))) {
1893 		dev_kfree_skb_any(skb);
1894 		return true;
1895 	}
1896 
1897 	/* place header in linear portion of buffer */
1898 	if (!skb_headlen(skb))
1899 		ixgbe_pull_tail(rx_ring, skb);
1900 
1901 #ifdef IXGBE_FCOE
1902 	/* do not attempt to pad FCoE Frames as this will disrupt DDP */
1903 	if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1904 		return false;
1905 
1906 #endif
1907 	/* if eth_skb_pad returns an error the skb was freed */
1908 	if (eth_skb_pad(skb))
1909 		return true;
1910 
1911 	return false;
1912 }
1913 
1914 /**
1915  * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1916  * @rx_ring: rx descriptor ring to store buffers on
1917  * @old_buff: donor buffer to have page reused
1918  *
1919  * Synchronizes page for reuse by the adapter
1920  **/
1921 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1922 				struct ixgbe_rx_buffer *old_buff)
1923 {
1924 	struct ixgbe_rx_buffer *new_buff;
1925 	u16 nta = rx_ring->next_to_alloc;
1926 
1927 	new_buff = &rx_ring->rx_buffer_info[nta];
1928 
1929 	/* update, and store next to alloc */
1930 	nta++;
1931 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1932 
1933 	/* Transfer page from old buffer to new buffer.
1934 	 * Move each member individually to avoid possible store
1935 	 * forwarding stalls and unnecessary copy of skb.
1936 	 */
1937 	new_buff->dma		= old_buff->dma;
1938 	new_buff->page		= old_buff->page;
1939 	new_buff->page_offset	= old_buff->page_offset;
1940 	new_buff->pagecnt_bias	= old_buff->pagecnt_bias;
1941 }
1942 
1943 static bool ixgbe_can_reuse_rx_page(struct ixgbe_rx_buffer *rx_buffer,
1944 				    int rx_buffer_pgcnt)
1945 {
1946 	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1947 	struct page *page = rx_buffer->page;
1948 
1949 	/* avoid re-using remote and pfmemalloc pages */
1950 	if (!dev_page_is_reusable(page))
1951 		return false;
1952 
1953 #if (PAGE_SIZE < 8192)
1954 	/* if we are only owner of page we can reuse it */
1955 	if (unlikely((rx_buffer_pgcnt - pagecnt_bias) > 1))
1956 		return false;
1957 #else
1958 	/* The last offset is a bit aggressive in that we assume the
1959 	 * worst case of FCoE being enabled and using a 3K buffer.
1960 	 * However this should have minimal impact as the 1K extra is
1961 	 * still less than one buffer in size.
1962 	 */
1963 #define IXGBE_LAST_OFFSET \
1964 	(SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBE_RXBUFFER_3K)
1965 	if (rx_buffer->page_offset > IXGBE_LAST_OFFSET)
1966 		return false;
1967 #endif
1968 
1969 	/* If we have drained the page fragment pool we need to update
1970 	 * the pagecnt_bias and page count so that we fully restock the
1971 	 * number of references the driver holds.
1972 	 */
1973 	if (unlikely(pagecnt_bias == 1)) {
1974 		page_ref_add(page, USHRT_MAX - 1);
1975 		rx_buffer->pagecnt_bias = USHRT_MAX;
1976 	}
1977 
1978 	return true;
1979 }
1980 
1981 /**
1982  * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1983  * @rx_ring: rx descriptor ring to transact packets on
1984  * @rx_buffer: buffer containing page to add
1985  * @skb: sk_buff to place the data into
1986  * @size: size of data in rx_buffer
1987  *
1988  * This function will add the data contained in rx_buffer->page to the skb.
1989  * This is done either through a direct copy if the data in the buffer is
1990  * less than the skb header size, otherwise it will just attach the page as
1991  * a frag to the skb.
1992  *
1993  * The function will then update the page offset if necessary and return
1994  * true if the buffer can be reused by the adapter.
1995  **/
1996 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1997 			      struct ixgbe_rx_buffer *rx_buffer,
1998 			      struct sk_buff *skb,
1999 			      unsigned int size)
2000 {
2001 #if (PAGE_SIZE < 8192)
2002 	unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2003 #else
2004 	unsigned int truesize = rx_ring->rx_offset ?
2005 				SKB_DATA_ALIGN(rx_ring->rx_offset + size) :
2006 				SKB_DATA_ALIGN(size);
2007 #endif
2008 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
2009 			rx_buffer->page_offset, size, truesize);
2010 #if (PAGE_SIZE < 8192)
2011 	rx_buffer->page_offset ^= truesize;
2012 #else
2013 	rx_buffer->page_offset += truesize;
2014 #endif
2015 }
2016 
2017 static struct ixgbe_rx_buffer *ixgbe_get_rx_buffer(struct ixgbe_ring *rx_ring,
2018 						   union ixgbe_adv_rx_desc *rx_desc,
2019 						   struct sk_buff **skb,
2020 						   const unsigned int size,
2021 						   int *rx_buffer_pgcnt)
2022 {
2023 	struct ixgbe_rx_buffer *rx_buffer;
2024 
2025 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
2026 	*rx_buffer_pgcnt =
2027 #if (PAGE_SIZE < 8192)
2028 		page_count(rx_buffer->page);
2029 #else
2030 		0;
2031 #endif
2032 	prefetchw(rx_buffer->page);
2033 	*skb = rx_buffer->skb;
2034 
2035 	/* Delay unmapping of the first packet. It carries the header
2036 	 * information, HW may still access the header after the writeback.
2037 	 * Only unmap it when EOP is reached
2038 	 */
2039 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)) {
2040 		if (!*skb)
2041 			goto skip_sync;
2042 	} else {
2043 		if (*skb)
2044 			ixgbe_dma_sync_frag(rx_ring, *skb);
2045 	}
2046 
2047 	/* we are reusing so sync this buffer for CPU use */
2048 	dma_sync_single_range_for_cpu(rx_ring->dev,
2049 				      rx_buffer->dma,
2050 				      rx_buffer->page_offset,
2051 				      size,
2052 				      DMA_FROM_DEVICE);
2053 skip_sync:
2054 	rx_buffer->pagecnt_bias--;
2055 
2056 	return rx_buffer;
2057 }
2058 
2059 static void ixgbe_put_rx_buffer(struct ixgbe_ring *rx_ring,
2060 				struct ixgbe_rx_buffer *rx_buffer,
2061 				struct sk_buff *skb,
2062 				int rx_buffer_pgcnt)
2063 {
2064 	if (ixgbe_can_reuse_rx_page(rx_buffer, rx_buffer_pgcnt)) {
2065 		/* hand second half of page back to the ring */
2066 		ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2067 	} else {
2068 		if (!IS_ERR(skb) && IXGBE_CB(skb)->dma == rx_buffer->dma) {
2069 			/* the page has been released from the ring */
2070 			IXGBE_CB(skb)->page_released = true;
2071 		} else {
2072 			/* we are not reusing the buffer so unmap it */
2073 			dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2074 					     ixgbe_rx_pg_size(rx_ring),
2075 					     DMA_FROM_DEVICE,
2076 					     IXGBE_RX_DMA_ATTR);
2077 		}
2078 		__page_frag_cache_drain(rx_buffer->page,
2079 					rx_buffer->pagecnt_bias);
2080 	}
2081 
2082 	/* clear contents of rx_buffer */
2083 	rx_buffer->page = NULL;
2084 	rx_buffer->skb = NULL;
2085 }
2086 
2087 static struct sk_buff *ixgbe_construct_skb(struct ixgbe_ring *rx_ring,
2088 					   struct ixgbe_rx_buffer *rx_buffer,
2089 					   struct xdp_buff *xdp,
2090 					   union ixgbe_adv_rx_desc *rx_desc)
2091 {
2092 	unsigned int size = xdp->data_end - xdp->data;
2093 #if (PAGE_SIZE < 8192)
2094 	unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2095 #else
2096 	unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
2097 					       xdp->data_hard_start);
2098 #endif
2099 	struct sk_buff *skb;
2100 
2101 	/* prefetch first cache line of first page */
2102 	net_prefetch(xdp->data);
2103 
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 	net_prefetch(xdp->data_meta);
2167 
2168 	/* build an skb to around the page buffer */
2169 	skb = build_skb(xdp->data_hard_start, truesize);
2170 	if (unlikely(!skb))
2171 		return NULL;
2172 
2173 	/* update pointers within the skb to store the data */
2174 	skb_reserve(skb, xdp->data - xdp->data_hard_start);
2175 	__skb_put(skb, xdp->data_end - xdp->data);
2176 	if (metasize)
2177 		skb_metadata_set(skb, metasize);
2178 
2179 	/* record DMA address if this is the start of a chain of buffers */
2180 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2181 		IXGBE_CB(skb)->dma = rx_buffer->dma;
2182 
2183 	/* update buffer offset */
2184 #if (PAGE_SIZE < 8192)
2185 	rx_buffer->page_offset ^= truesize;
2186 #else
2187 	rx_buffer->page_offset += truesize;
2188 #endif
2189 
2190 	return skb;
2191 }
2192 
2193 static struct sk_buff *ixgbe_run_xdp(struct ixgbe_adapter *adapter,
2194 				     struct ixgbe_ring *rx_ring,
2195 				     struct xdp_buff *xdp)
2196 {
2197 	int err, result = IXGBE_XDP_PASS;
2198 	struct bpf_prog *xdp_prog;
2199 	struct xdp_frame *xdpf;
2200 	u32 act;
2201 
2202 	rcu_read_lock();
2203 	xdp_prog = READ_ONCE(rx_ring->xdp_prog);
2204 
2205 	if (!xdp_prog)
2206 		goto xdp_out;
2207 
2208 	prefetchw(xdp->data_hard_start); /* xdp_frame write */
2209 
2210 	act = bpf_prog_run_xdp(xdp_prog, xdp);
2211 	switch (act) {
2212 	case XDP_PASS:
2213 		break;
2214 	case XDP_TX:
2215 		xdpf = xdp_convert_buff_to_frame(xdp);
2216 		if (unlikely(!xdpf)) {
2217 			result = IXGBE_XDP_CONSUMED;
2218 			break;
2219 		}
2220 		result = ixgbe_xmit_xdp_ring(adapter, xdpf);
2221 		break;
2222 	case XDP_REDIRECT:
2223 		err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
2224 		if (!err)
2225 			result = IXGBE_XDP_REDIR;
2226 		else
2227 			result = IXGBE_XDP_CONSUMED;
2228 		break;
2229 	default:
2230 		bpf_warn_invalid_xdp_action(act);
2231 		fallthrough;
2232 	case XDP_ABORTED:
2233 		trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
2234 		fallthrough; /* handle aborts by dropping packet */
2235 	case XDP_DROP:
2236 		result = IXGBE_XDP_CONSUMED;
2237 		break;
2238 	}
2239 xdp_out:
2240 	rcu_read_unlock();
2241 	return ERR_PTR(-result);
2242 }
2243 
2244 static unsigned int ixgbe_rx_frame_truesize(struct ixgbe_ring *rx_ring,
2245 					    unsigned int size)
2246 {
2247 	unsigned int truesize;
2248 
2249 #if (PAGE_SIZE < 8192)
2250 	truesize = ixgbe_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
2251 #else
2252 	truesize = rx_ring->rx_offset ?
2253 		SKB_DATA_ALIGN(rx_ring->rx_offset + size) +
2254 		SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
2255 		SKB_DATA_ALIGN(size);
2256 #endif
2257 	return truesize;
2258 }
2259 
2260 static void ixgbe_rx_buffer_flip(struct ixgbe_ring *rx_ring,
2261 				 struct ixgbe_rx_buffer *rx_buffer,
2262 				 unsigned int size)
2263 {
2264 	unsigned int truesize = ixgbe_rx_frame_truesize(rx_ring, size);
2265 #if (PAGE_SIZE < 8192)
2266 	rx_buffer->page_offset ^= truesize;
2267 #else
2268 	rx_buffer->page_offset += truesize;
2269 #endif
2270 }
2271 
2272 /**
2273  * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2274  * @q_vector: structure containing interrupt and ring information
2275  * @rx_ring: rx descriptor ring to transact packets on
2276  * @budget: Total limit on number of packets to process
2277  *
2278  * This function provides a "bounce buffer" approach to Rx interrupt
2279  * processing.  The advantage to this is that on systems that have
2280  * expensive overhead for IOMMU access this provides a means of avoiding
2281  * it by maintaining the mapping of the page to the syste.
2282  *
2283  * Returns amount of work completed
2284  **/
2285 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2286 			       struct ixgbe_ring *rx_ring,
2287 			       const int budget)
2288 {
2289 	unsigned int total_rx_bytes = 0, total_rx_packets = 0, frame_sz = 0;
2290 	struct ixgbe_adapter *adapter = q_vector->adapter;
2291 #ifdef IXGBE_FCOE
2292 	int ddp_bytes;
2293 	unsigned int mss = 0;
2294 #endif /* IXGBE_FCOE */
2295 	u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2296 	unsigned int offset = rx_ring->rx_offset;
2297 	unsigned int xdp_xmit = 0;
2298 	struct xdp_buff xdp;
2299 
2300 	/* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
2301 #if (PAGE_SIZE < 8192)
2302 	frame_sz = ixgbe_rx_frame_truesize(rx_ring, 0);
2303 #endif
2304 	xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
2305 
2306 	while (likely(total_rx_packets < budget)) {
2307 		union ixgbe_adv_rx_desc *rx_desc;
2308 		struct ixgbe_rx_buffer *rx_buffer;
2309 		struct sk_buff *skb;
2310 		int rx_buffer_pgcnt;
2311 		unsigned int size;
2312 
2313 		/* return some buffers to hardware, one at a time is too slow */
2314 		if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2315 			ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2316 			cleaned_count = 0;
2317 		}
2318 
2319 		rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2320 		size = le16_to_cpu(rx_desc->wb.upper.length);
2321 		if (!size)
2322 			break;
2323 
2324 		/* This memory barrier is needed to keep us from reading
2325 		 * any other fields out of the rx_desc until we know the
2326 		 * descriptor has been written back
2327 		 */
2328 		dma_rmb();
2329 
2330 		rx_buffer = ixgbe_get_rx_buffer(rx_ring, rx_desc, &skb, size, &rx_buffer_pgcnt);
2331 
2332 		/* retrieve a buffer from the ring */
2333 		if (!skb) {
2334 			unsigned char *hard_start;
2335 
2336 			hard_start = page_address(rx_buffer->page) +
2337 				     rx_buffer->page_offset - offset;
2338 			xdp_prepare_buff(&xdp, hard_start, offset, size, true);
2339 #if (PAGE_SIZE > 4096)
2340 			/* At larger PAGE_SIZE, frame_sz depend on len size */
2341 			xdp.frame_sz = ixgbe_rx_frame_truesize(rx_ring, size);
2342 #endif
2343 			skb = ixgbe_run_xdp(adapter, rx_ring, &xdp);
2344 		}
2345 
2346 		if (IS_ERR(skb)) {
2347 			unsigned int xdp_res = -PTR_ERR(skb);
2348 
2349 			if (xdp_res & (IXGBE_XDP_TX | IXGBE_XDP_REDIR)) {
2350 				xdp_xmit |= xdp_res;
2351 				ixgbe_rx_buffer_flip(rx_ring, rx_buffer, size);
2352 			} else {
2353 				rx_buffer->pagecnt_bias++;
2354 			}
2355 			total_rx_packets++;
2356 			total_rx_bytes += size;
2357 		} else if (skb) {
2358 			ixgbe_add_rx_frag(rx_ring, rx_buffer, skb, size);
2359 		} else if (ring_uses_build_skb(rx_ring)) {
2360 			skb = ixgbe_build_skb(rx_ring, rx_buffer,
2361 					      &xdp, rx_desc);
2362 		} else {
2363 			skb = ixgbe_construct_skb(rx_ring, rx_buffer,
2364 						  &xdp, rx_desc);
2365 		}
2366 
2367 		/* exit if we failed to retrieve a buffer */
2368 		if (!skb) {
2369 			rx_ring->rx_stats.alloc_rx_buff_failed++;
2370 			rx_buffer->pagecnt_bias++;
2371 			break;
2372 		}
2373 
2374 		ixgbe_put_rx_buffer(rx_ring, rx_buffer, skb, rx_buffer_pgcnt);
2375 		cleaned_count++;
2376 
2377 		/* place incomplete frames back on ring for completion */
2378 		if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2379 			continue;
2380 
2381 		/* verify the packet layout is correct */
2382 		if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2383 			continue;
2384 
2385 		/* probably a little skewed due to removing CRC */
2386 		total_rx_bytes += skb->len;
2387 
2388 		/* populate checksum, timestamp, VLAN, and protocol */
2389 		ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2390 
2391 #ifdef IXGBE_FCOE
2392 		/* if ddp, not passing to ULD unless for FCP_RSP or error */
2393 		if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2394 			ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2395 			/* include DDPed FCoE data */
2396 			if (ddp_bytes > 0) {
2397 				if (!mss) {
2398 					mss = rx_ring->netdev->mtu -
2399 						sizeof(struct fcoe_hdr) -
2400 						sizeof(struct fc_frame_header) -
2401 						sizeof(struct fcoe_crc_eof);
2402 					if (mss > 512)
2403 						mss &= ~511;
2404 				}
2405 				total_rx_bytes += ddp_bytes;
2406 				total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2407 								 mss);
2408 			}
2409 			if (!ddp_bytes) {
2410 				dev_kfree_skb_any(skb);
2411 				continue;
2412 			}
2413 		}
2414 
2415 #endif /* IXGBE_FCOE */
2416 		ixgbe_rx_skb(q_vector, skb);
2417 
2418 		/* update budget accounting */
2419 		total_rx_packets++;
2420 	}
2421 
2422 	if (xdp_xmit & IXGBE_XDP_REDIR)
2423 		xdp_do_flush_map();
2424 
2425 	if (xdp_xmit & IXGBE_XDP_TX) {
2426 		struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
2427 
2428 		/* Force memory writes to complete before letting h/w
2429 		 * know there are new descriptors to fetch.
2430 		 */
2431 		wmb();
2432 		writel(ring->next_to_use, ring->tail);
2433 	}
2434 
2435 	u64_stats_update_begin(&rx_ring->syncp);
2436 	rx_ring->stats.packets += total_rx_packets;
2437 	rx_ring->stats.bytes += total_rx_bytes;
2438 	u64_stats_update_end(&rx_ring->syncp);
2439 	q_vector->rx.total_packets += total_rx_packets;
2440 	q_vector->rx.total_bytes += total_rx_bytes;
2441 
2442 	return total_rx_packets;
2443 }
2444 
2445 /**
2446  * ixgbe_configure_msix - Configure MSI-X hardware
2447  * @adapter: board private structure
2448  *
2449  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2450  * interrupts.
2451  **/
2452 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2453 {
2454 	struct ixgbe_q_vector *q_vector;
2455 	int v_idx;
2456 	u32 mask;
2457 
2458 	/* Populate MSIX to EITR Select */
2459 	if (adapter->num_vfs > 32) {
2460 		u32 eitrsel = BIT(adapter->num_vfs - 32) - 1;
2461 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2462 	}
2463 
2464 	/*
2465 	 * Populate the IVAR table and set the ITR values to the
2466 	 * corresponding register.
2467 	 */
2468 	for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2469 		struct ixgbe_ring *ring;
2470 		q_vector = adapter->q_vector[v_idx];
2471 
2472 		ixgbe_for_each_ring(ring, q_vector->rx)
2473 			ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2474 
2475 		ixgbe_for_each_ring(ring, q_vector->tx)
2476 			ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2477 
2478 		ixgbe_write_eitr(q_vector);
2479 	}
2480 
2481 	switch (adapter->hw.mac.type) {
2482 	case ixgbe_mac_82598EB:
2483 		ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2484 			       v_idx);
2485 		break;
2486 	case ixgbe_mac_82599EB:
2487 	case ixgbe_mac_X540:
2488 	case ixgbe_mac_X550:
2489 	case ixgbe_mac_X550EM_x:
2490 	case ixgbe_mac_x550em_a:
2491 		ixgbe_set_ivar(adapter, -1, 1, v_idx);
2492 		break;
2493 	default:
2494 		break;
2495 	}
2496 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2497 
2498 	/* set up to autoclear timer, and the vectors */
2499 	mask = IXGBE_EIMS_ENABLE_MASK;
2500 	mask &= ~(IXGBE_EIMS_OTHER |
2501 		  IXGBE_EIMS_MAILBOX |
2502 		  IXGBE_EIMS_LSC);
2503 
2504 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2505 }
2506 
2507 /**
2508  * ixgbe_update_itr - update the dynamic ITR value based on statistics
2509  * @q_vector: structure containing interrupt and ring information
2510  * @ring_container: structure containing ring performance data
2511  *
2512  *      Stores a new ITR value based on packets and byte
2513  *      counts during the last interrupt.  The advantage of per interrupt
2514  *      computation is faster updates and more accurate ITR for the current
2515  *      traffic pattern.  Constants in this function were computed
2516  *      based on theoretical maximum wire speed and thresholds were set based
2517  *      on testing data as well as attempting to minimize response time
2518  *      while increasing bulk throughput.
2519  **/
2520 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2521 			     struct ixgbe_ring_container *ring_container)
2522 {
2523 	unsigned int itr = IXGBE_ITR_ADAPTIVE_MIN_USECS |
2524 			   IXGBE_ITR_ADAPTIVE_LATENCY;
2525 	unsigned int avg_wire_size, packets, bytes;
2526 	unsigned long next_update = jiffies;
2527 
2528 	/* If we don't have any rings just leave ourselves set for maximum
2529 	 * possible latency so we take ourselves out of the equation.
2530 	 */
2531 	if (!ring_container->ring)
2532 		return;
2533 
2534 	/* If we didn't update within up to 1 - 2 jiffies we can assume
2535 	 * that either packets are coming in so slow there hasn't been
2536 	 * any work, or that there is so much work that NAPI is dealing
2537 	 * with interrupt moderation and we don't need to do anything.
2538 	 */
2539 	if (time_after(next_update, ring_container->next_update))
2540 		goto clear_counts;
2541 
2542 	packets = ring_container->total_packets;
2543 
2544 	/* We have no packets to actually measure against. This means
2545 	 * either one of the other queues on this vector is active or
2546 	 * we are a Tx queue doing TSO with too high of an interrupt rate.
2547 	 *
2548 	 * When this occurs just tick up our delay by the minimum value
2549 	 * and hope that this extra delay will prevent us from being called
2550 	 * without any work on our queue.
2551 	 */
2552 	if (!packets) {
2553 		itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2554 		if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2555 			itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2556 		itr += ring_container->itr & IXGBE_ITR_ADAPTIVE_LATENCY;
2557 		goto clear_counts;
2558 	}
2559 
2560 	bytes = ring_container->total_bytes;
2561 
2562 	/* If packets are less than 4 or bytes are less than 9000 assume
2563 	 * insufficient data to use bulk rate limiting approach. We are
2564 	 * likely latency driven.
2565 	 */
2566 	if (packets < 4 && bytes < 9000) {
2567 		itr = IXGBE_ITR_ADAPTIVE_LATENCY;
2568 		goto adjust_by_size;
2569 	}
2570 
2571 	/* Between 4 and 48 we can assume that our current interrupt delay
2572 	 * is only slightly too low. As such we should increase it by a small
2573 	 * fixed amount.
2574 	 */
2575 	if (packets < 48) {
2576 		itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2577 		if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2578 			itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2579 		goto clear_counts;
2580 	}
2581 
2582 	/* Between 48 and 96 is our "goldilocks" zone where we are working
2583 	 * out "just right". Just report that our current ITR is good for us.
2584 	 */
2585 	if (packets < 96) {
2586 		itr = q_vector->itr >> 2;
2587 		goto clear_counts;
2588 	}
2589 
2590 	/* If packet count is 96 or greater we are likely looking at a slight
2591 	 * overrun of the delay we want. Try halving our delay to see if that
2592 	 * will cut the number of packets in half per interrupt.
2593 	 */
2594 	if (packets < 256) {
2595 		itr = q_vector->itr >> 3;
2596 		if (itr < IXGBE_ITR_ADAPTIVE_MIN_USECS)
2597 			itr = IXGBE_ITR_ADAPTIVE_MIN_USECS;
2598 		goto clear_counts;
2599 	}
2600 
2601 	/* The paths below assume we are dealing with a bulk ITR since number
2602 	 * of packets is 256 or greater. We are just going to have to compute
2603 	 * a value and try to bring the count under control, though for smaller
2604 	 * packet sizes there isn't much we can do as NAPI polling will likely
2605 	 * be kicking in sooner rather than later.
2606 	 */
2607 	itr = IXGBE_ITR_ADAPTIVE_BULK;
2608 
2609 adjust_by_size:
2610 	/* If packet counts are 256 or greater we can assume we have a gross
2611 	 * overestimation of what the rate should be. Instead of trying to fine
2612 	 * tune it just use the formula below to try and dial in an exact value
2613 	 * give the current packet size of the frame.
2614 	 */
2615 	avg_wire_size = bytes / packets;
2616 
2617 	/* The following is a crude approximation of:
2618 	 *  wmem_default / (size + overhead) = desired_pkts_per_int
2619 	 *  rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
2620 	 *  (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
2621 	 *
2622 	 * Assuming wmem_default is 212992 and overhead is 640 bytes per
2623 	 * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
2624 	 * formula down to
2625 	 *
2626 	 *  (170 * (size + 24)) / (size + 640) = ITR
2627 	 *
2628 	 * We first do some math on the packet size and then finally bitshift
2629 	 * by 8 after rounding up. We also have to account for PCIe link speed
2630 	 * difference as ITR scales based on this.
2631 	 */
2632 	if (avg_wire_size <= 60) {
2633 		/* Start at 50k ints/sec */
2634 		avg_wire_size = 5120;
2635 	} else if (avg_wire_size <= 316) {
2636 		/* 50K ints/sec to 16K ints/sec */
2637 		avg_wire_size *= 40;
2638 		avg_wire_size += 2720;
2639 	} else if (avg_wire_size <= 1084) {
2640 		/* 16K ints/sec to 9.2K ints/sec */
2641 		avg_wire_size *= 15;
2642 		avg_wire_size += 11452;
2643 	} else if (avg_wire_size < 1968) {
2644 		/* 9.2K ints/sec to 8K ints/sec */
2645 		avg_wire_size *= 5;
2646 		avg_wire_size += 22420;
2647 	} else {
2648 		/* plateau at a limit of 8K ints/sec */
2649 		avg_wire_size = 32256;
2650 	}
2651 
2652 	/* If we are in low latency mode half our delay which doubles the rate
2653 	 * to somewhere between 100K to 16K ints/sec
2654 	 */
2655 	if (itr & IXGBE_ITR_ADAPTIVE_LATENCY)
2656 		avg_wire_size >>= 1;
2657 
2658 	/* Resultant value is 256 times larger than it needs to be. This
2659 	 * gives us room to adjust the value as needed to either increase
2660 	 * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
2661 	 *
2662 	 * Use addition as we have already recorded the new latency flag
2663 	 * for the ITR value.
2664 	 */
2665 	switch (q_vector->adapter->link_speed) {
2666 	case IXGBE_LINK_SPEED_10GB_FULL:
2667 	case IXGBE_LINK_SPEED_100_FULL:
2668 	default:
2669 		itr += DIV_ROUND_UP(avg_wire_size,
2670 				    IXGBE_ITR_ADAPTIVE_MIN_INC * 256) *
2671 		       IXGBE_ITR_ADAPTIVE_MIN_INC;
2672 		break;
2673 	case IXGBE_LINK_SPEED_2_5GB_FULL:
2674 	case IXGBE_LINK_SPEED_1GB_FULL:
2675 	case IXGBE_LINK_SPEED_10_FULL:
2676 		if (avg_wire_size > 8064)
2677 			avg_wire_size = 8064;
2678 		itr += DIV_ROUND_UP(avg_wire_size,
2679 				    IXGBE_ITR_ADAPTIVE_MIN_INC * 64) *
2680 		       IXGBE_ITR_ADAPTIVE_MIN_INC;
2681 		break;
2682 	}
2683 
2684 clear_counts:
2685 	/* write back value */
2686 	ring_container->itr = itr;
2687 
2688 	/* next update should occur within next jiffy */
2689 	ring_container->next_update = next_update + 1;
2690 
2691 	ring_container->total_bytes = 0;
2692 	ring_container->total_packets = 0;
2693 }
2694 
2695 /**
2696  * ixgbe_write_eitr - write EITR register in hardware specific way
2697  * @q_vector: structure containing interrupt and ring information
2698  *
2699  * This function is made to be called by ethtool and by the driver
2700  * when it needs to update EITR registers at runtime.  Hardware
2701  * specific quirks/differences are taken care of here.
2702  */
2703 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2704 {
2705 	struct ixgbe_adapter *adapter = q_vector->adapter;
2706 	struct ixgbe_hw *hw = &adapter->hw;
2707 	int v_idx = q_vector->v_idx;
2708 	u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2709 
2710 	switch (adapter->hw.mac.type) {
2711 	case ixgbe_mac_82598EB:
2712 		/* must write high and low 16 bits to reset counter */
2713 		itr_reg |= (itr_reg << 16);
2714 		break;
2715 	case ixgbe_mac_82599EB:
2716 	case ixgbe_mac_X540:
2717 	case ixgbe_mac_X550:
2718 	case ixgbe_mac_X550EM_x:
2719 	case ixgbe_mac_x550em_a:
2720 		/*
2721 		 * set the WDIS bit to not clear the timer bits and cause an
2722 		 * immediate assertion of the interrupt
2723 		 */
2724 		itr_reg |= IXGBE_EITR_CNT_WDIS;
2725 		break;
2726 	default:
2727 		break;
2728 	}
2729 	IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2730 }
2731 
2732 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2733 {
2734 	u32 new_itr;
2735 
2736 	ixgbe_update_itr(q_vector, &q_vector->tx);
2737 	ixgbe_update_itr(q_vector, &q_vector->rx);
2738 
2739 	/* use the smallest value of new ITR delay calculations */
2740 	new_itr = min(q_vector->rx.itr, q_vector->tx.itr);
2741 
2742 	/* Clear latency flag if set, shift into correct position */
2743 	new_itr &= ~IXGBE_ITR_ADAPTIVE_LATENCY;
2744 	new_itr <<= 2;
2745 
2746 	if (new_itr != q_vector->itr) {
2747 		/* save the algorithm value here */
2748 		q_vector->itr = new_itr;
2749 
2750 		ixgbe_write_eitr(q_vector);
2751 	}
2752 }
2753 
2754 /**
2755  * ixgbe_check_overtemp_subtask - check for over temperature
2756  * @adapter: pointer to adapter
2757  **/
2758 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2759 {
2760 	struct ixgbe_hw *hw = &adapter->hw;
2761 	u32 eicr = adapter->interrupt_event;
2762 	s32 rc;
2763 
2764 	if (test_bit(__IXGBE_DOWN, &adapter->state))
2765 		return;
2766 
2767 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2768 		return;
2769 
2770 	adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2771 
2772 	switch (hw->device_id) {
2773 	case IXGBE_DEV_ID_82599_T3_LOM:
2774 		/*
2775 		 * Since the warning interrupt is for both ports
2776 		 * we don't have to check if:
2777 		 *  - This interrupt wasn't for our port.
2778 		 *  - We may have missed the interrupt so always have to
2779 		 *    check if we  got a LSC
2780 		 */
2781 		if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2782 		    !(eicr & IXGBE_EICR_LSC))
2783 			return;
2784 
2785 		if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2786 			u32 speed;
2787 			bool link_up = false;
2788 
2789 			hw->mac.ops.check_link(hw, &speed, &link_up, false);
2790 
2791 			if (link_up)
2792 				return;
2793 		}
2794 
2795 		/* Check if this is not due to overtemp */
2796 		if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2797 			return;
2798 
2799 		break;
2800 	case IXGBE_DEV_ID_X550EM_A_1G_T:
2801 	case IXGBE_DEV_ID_X550EM_A_1G_T_L:
2802 		rc = hw->phy.ops.check_overtemp(hw);
2803 		if (rc != IXGBE_ERR_OVERTEMP)
2804 			return;
2805 		break;
2806 	default:
2807 		if (adapter->hw.mac.type >= ixgbe_mac_X540)
2808 			return;
2809 		if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2810 			return;
2811 		break;
2812 	}
2813 	e_crit(drv, "%s\n", ixgbe_overheat_msg);
2814 
2815 	adapter->interrupt_event = 0;
2816 }
2817 
2818 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2819 {
2820 	struct ixgbe_hw *hw = &adapter->hw;
2821 
2822 	if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2823 	    (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2824 		e_crit(probe, "Fan has stopped, replace the adapter\n");
2825 		/* write to clear the interrupt */
2826 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2827 	}
2828 }
2829 
2830 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2831 {
2832 	struct ixgbe_hw *hw = &adapter->hw;
2833 
2834 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2835 		return;
2836 
2837 	switch (adapter->hw.mac.type) {
2838 	case ixgbe_mac_82599EB:
2839 		/*
2840 		 * Need to check link state so complete overtemp check
2841 		 * on service task
2842 		 */
2843 		if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2844 		     (eicr & IXGBE_EICR_LSC)) &&
2845 		    (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2846 			adapter->interrupt_event = eicr;
2847 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2848 			ixgbe_service_event_schedule(adapter);
2849 			return;
2850 		}
2851 		return;
2852 	case ixgbe_mac_x550em_a:
2853 		if (eicr & IXGBE_EICR_GPI_SDP0_X550EM_a) {
2854 			adapter->interrupt_event = eicr;
2855 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2856 			ixgbe_service_event_schedule(adapter);
2857 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC,
2858 					IXGBE_EICR_GPI_SDP0_X550EM_a);
2859 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICR,
2860 					IXGBE_EICR_GPI_SDP0_X550EM_a);
2861 		}
2862 		return;
2863 	case ixgbe_mac_X550:
2864 	case ixgbe_mac_X540:
2865 		if (!(eicr & IXGBE_EICR_TS))
2866 			return;
2867 		break;
2868 	default:
2869 		return;
2870 	}
2871 
2872 	e_crit(drv, "%s\n", ixgbe_overheat_msg);
2873 }
2874 
2875 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2876 {
2877 	switch (hw->mac.type) {
2878 	case ixgbe_mac_82598EB:
2879 		if (hw->phy.type == ixgbe_phy_nl)
2880 			return true;
2881 		return false;
2882 	case ixgbe_mac_82599EB:
2883 	case ixgbe_mac_X550EM_x:
2884 	case ixgbe_mac_x550em_a:
2885 		switch (hw->mac.ops.get_media_type(hw)) {
2886 		case ixgbe_media_type_fiber:
2887 		case ixgbe_media_type_fiber_qsfp:
2888 			return true;
2889 		default:
2890 			return false;
2891 		}
2892 	default:
2893 		return false;
2894 	}
2895 }
2896 
2897 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2898 {
2899 	struct ixgbe_hw *hw = &adapter->hw;
2900 	u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw);
2901 
2902 	if (!ixgbe_is_sfp(hw))
2903 		return;
2904 
2905 	/* Later MAC's use different SDP */
2906 	if (hw->mac.type >= ixgbe_mac_X540)
2907 		eicr_mask = IXGBE_EICR_GPI_SDP0_X540;
2908 
2909 	if (eicr & eicr_mask) {
2910 		/* Clear the interrupt */
2911 		IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask);
2912 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2913 			adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2914 			adapter->sfp_poll_time = 0;
2915 			ixgbe_service_event_schedule(adapter);
2916 		}
2917 	}
2918 
2919 	if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
2920 	    (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2921 		/* Clear the interrupt */
2922 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2923 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2924 			adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2925 			ixgbe_service_event_schedule(adapter);
2926 		}
2927 	}
2928 }
2929 
2930 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2931 {
2932 	struct ixgbe_hw *hw = &adapter->hw;
2933 
2934 	adapter->lsc_int++;
2935 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2936 	adapter->link_check_timeout = jiffies;
2937 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2938 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2939 		IXGBE_WRITE_FLUSH(hw);
2940 		ixgbe_service_event_schedule(adapter);
2941 	}
2942 }
2943 
2944 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2945 					   u64 qmask)
2946 {
2947 	u32 mask;
2948 	struct ixgbe_hw *hw = &adapter->hw;
2949 
2950 	switch (hw->mac.type) {
2951 	case ixgbe_mac_82598EB:
2952 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2953 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2954 		break;
2955 	case ixgbe_mac_82599EB:
2956 	case ixgbe_mac_X540:
2957 	case ixgbe_mac_X550:
2958 	case ixgbe_mac_X550EM_x:
2959 	case ixgbe_mac_x550em_a:
2960 		mask = (qmask & 0xFFFFFFFF);
2961 		if (mask)
2962 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2963 		mask = (qmask >> 32);
2964 		if (mask)
2965 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2966 		break;
2967 	default:
2968 		break;
2969 	}
2970 	/* skip the flush */
2971 }
2972 
2973 /**
2974  * ixgbe_irq_enable - Enable default interrupt generation settings
2975  * @adapter: board private structure
2976  * @queues: enable irqs for queues
2977  * @flush: flush register write
2978  **/
2979 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2980 				    bool flush)
2981 {
2982 	struct ixgbe_hw *hw = &adapter->hw;
2983 	u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2984 
2985 	/* don't reenable LSC while waiting for link */
2986 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2987 		mask &= ~IXGBE_EIMS_LSC;
2988 
2989 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2990 		switch (adapter->hw.mac.type) {
2991 		case ixgbe_mac_82599EB:
2992 			mask |= IXGBE_EIMS_GPI_SDP0(hw);
2993 			break;
2994 		case ixgbe_mac_X540:
2995 		case ixgbe_mac_X550:
2996 		case ixgbe_mac_X550EM_x:
2997 		case ixgbe_mac_x550em_a:
2998 			mask |= IXGBE_EIMS_TS;
2999 			break;
3000 		default:
3001 			break;
3002 		}
3003 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3004 		mask |= IXGBE_EIMS_GPI_SDP1(hw);
3005 	switch (adapter->hw.mac.type) {
3006 	case ixgbe_mac_82599EB:
3007 		mask |= IXGBE_EIMS_GPI_SDP1(hw);
3008 		mask |= IXGBE_EIMS_GPI_SDP2(hw);
3009 		fallthrough;
3010 	case ixgbe_mac_X540:
3011 	case ixgbe_mac_X550:
3012 	case ixgbe_mac_X550EM_x:
3013 	case ixgbe_mac_x550em_a:
3014 		if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
3015 		    adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
3016 		    adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP_N)
3017 			mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw);
3018 		if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
3019 			mask |= IXGBE_EICR_GPI_SDP0_X540;
3020 		mask |= IXGBE_EIMS_ECC;
3021 		mask |= IXGBE_EIMS_MAILBOX;
3022 		break;
3023 	default:
3024 		break;
3025 	}
3026 
3027 	if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
3028 	    !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
3029 		mask |= IXGBE_EIMS_FLOW_DIR;
3030 
3031 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
3032 	if (queues)
3033 		ixgbe_irq_enable_queues(adapter, ~0);
3034 	if (flush)
3035 		IXGBE_WRITE_FLUSH(&adapter->hw);
3036 }
3037 
3038 static irqreturn_t ixgbe_msix_other(int irq, void *data)
3039 {
3040 	struct ixgbe_adapter *adapter = data;
3041 	struct ixgbe_hw *hw = &adapter->hw;
3042 	u32 eicr;
3043 
3044 	/*
3045 	 * Workaround for Silicon errata.  Use clear-by-write instead
3046 	 * of clear-by-read.  Reading with EICS will return the
3047 	 * interrupt causes without clearing, which later be done
3048 	 * with the write to EICR.
3049 	 */
3050 	eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
3051 
3052 	/* The lower 16bits of the EICR register are for the queue interrupts
3053 	 * which should be masked here in order to not accidentally clear them if
3054 	 * the bits are high when ixgbe_msix_other is called. There is a race
3055 	 * condition otherwise which results in possible performance loss
3056 	 * especially if the ixgbe_msix_other interrupt is triggering
3057 	 * consistently (as it would when PPS is turned on for the X540 device)
3058 	 */
3059 	eicr &= 0xFFFF0000;
3060 
3061 	IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
3062 
3063 	if (eicr & IXGBE_EICR_LSC)
3064 		ixgbe_check_lsc(adapter);
3065 
3066 	if (eicr & IXGBE_EICR_MAILBOX)
3067 		ixgbe_msg_task(adapter);
3068 
3069 	switch (hw->mac.type) {
3070 	case ixgbe_mac_82599EB:
3071 	case ixgbe_mac_X540:
3072 	case ixgbe_mac_X550:
3073 	case ixgbe_mac_X550EM_x:
3074 	case ixgbe_mac_x550em_a:
3075 		if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
3076 		    (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
3077 			adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
3078 			ixgbe_service_event_schedule(adapter);
3079 			IXGBE_WRITE_REG(hw, IXGBE_EICR,
3080 					IXGBE_EICR_GPI_SDP0_X540);
3081 		}
3082 		if (eicr & IXGBE_EICR_ECC) {
3083 			e_info(link, "Received ECC Err, initiating reset\n");
3084 			set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3085 			ixgbe_service_event_schedule(adapter);
3086 			IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3087 		}
3088 		/* Handle Flow Director Full threshold interrupt */
3089 		if (eicr & IXGBE_EICR_FLOW_DIR) {
3090 			int reinit_count = 0;
3091 			int i;
3092 			for (i = 0; i < adapter->num_tx_queues; i++) {
3093 				struct ixgbe_ring *ring = adapter->tx_ring[i];
3094 				if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
3095 						       &ring->state))
3096 					reinit_count++;
3097 			}
3098 			if (reinit_count) {
3099 				/* no more flow director interrupts until after init */
3100 				IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
3101 				adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
3102 				ixgbe_service_event_schedule(adapter);
3103 			}
3104 		}
3105 		ixgbe_check_sfp_event(adapter, eicr);
3106 		ixgbe_check_overtemp_event(adapter, eicr);
3107 		break;
3108 	default:
3109 		break;
3110 	}
3111 
3112 	ixgbe_check_fan_failure(adapter, eicr);
3113 
3114 	if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3115 		ixgbe_ptp_check_pps_event(adapter);
3116 
3117 	/* re-enable the original interrupt state, no lsc, no queues */
3118 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
3119 		ixgbe_irq_enable(adapter, false, false);
3120 
3121 	return IRQ_HANDLED;
3122 }
3123 
3124 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
3125 {
3126 	struct ixgbe_q_vector *q_vector = data;
3127 
3128 	/* EIAM disabled interrupts (on this vector) for us */
3129 
3130 	if (q_vector->rx.ring || q_vector->tx.ring)
3131 		napi_schedule_irqoff(&q_vector->napi);
3132 
3133 	return IRQ_HANDLED;
3134 }
3135 
3136 /**
3137  * ixgbe_poll - NAPI Rx polling callback
3138  * @napi: structure for representing this polling device
3139  * @budget: how many packets driver is allowed to clean
3140  *
3141  * This function is used for legacy and MSI, NAPI mode
3142  **/
3143 int ixgbe_poll(struct napi_struct *napi, int budget)
3144 {
3145 	struct ixgbe_q_vector *q_vector =
3146 				container_of(napi, struct ixgbe_q_vector, napi);
3147 	struct ixgbe_adapter *adapter = q_vector->adapter;
3148 	struct ixgbe_ring *ring;
3149 	int per_ring_budget, work_done = 0;
3150 	bool clean_complete = true;
3151 
3152 #ifdef CONFIG_IXGBE_DCA
3153 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3154 		ixgbe_update_dca(q_vector);
3155 #endif
3156 
3157 	ixgbe_for_each_ring(ring, q_vector->tx) {
3158 		bool wd = ring->xsk_pool ?
3159 			  ixgbe_clean_xdp_tx_irq(q_vector, ring, budget) :
3160 			  ixgbe_clean_tx_irq(q_vector, ring, budget);
3161 
3162 		if (!wd)
3163 			clean_complete = false;
3164 	}
3165 
3166 	/* Exit if we are called by netpoll */
3167 	if (budget <= 0)
3168 		return budget;
3169 
3170 	/* attempt to distribute budget to each queue fairly, but don't allow
3171 	 * the budget to go below 1 because we'll exit polling */
3172 	if (q_vector->rx.count > 1)
3173 		per_ring_budget = max(budget/q_vector->rx.count, 1);
3174 	else
3175 		per_ring_budget = budget;
3176 
3177 	ixgbe_for_each_ring(ring, q_vector->rx) {
3178 		int cleaned = ring->xsk_pool ?
3179 			      ixgbe_clean_rx_irq_zc(q_vector, ring,
3180 						    per_ring_budget) :
3181 			      ixgbe_clean_rx_irq(q_vector, ring,
3182 						 per_ring_budget);
3183 
3184 		work_done += cleaned;
3185 		if (cleaned >= per_ring_budget)
3186 			clean_complete = false;
3187 	}
3188 
3189 	/* If all work not completed, return budget and keep polling */
3190 	if (!clean_complete)
3191 		return budget;
3192 
3193 	/* all work done, exit the polling mode */
3194 	if (likely(napi_complete_done(napi, work_done))) {
3195 		if (adapter->rx_itr_setting & 1)
3196 			ixgbe_set_itr(q_vector);
3197 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
3198 			ixgbe_irq_enable_queues(adapter,
3199 						BIT_ULL(q_vector->v_idx));
3200 	}
3201 
3202 	return min(work_done, budget - 1);
3203 }
3204 
3205 /**
3206  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
3207  * @adapter: board private structure
3208  *
3209  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
3210  * interrupts from the kernel.
3211  **/
3212 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
3213 {
3214 	struct net_device *netdev = adapter->netdev;
3215 	unsigned int ri = 0, ti = 0;
3216 	int vector, err;
3217 
3218 	for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3219 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3220 		struct msix_entry *entry = &adapter->msix_entries[vector];
3221 
3222 		if (q_vector->tx.ring && q_vector->rx.ring) {
3223 			snprintf(q_vector->name, sizeof(q_vector->name),
3224 				 "%s-TxRx-%u", netdev->name, ri++);
3225 			ti++;
3226 		} else if (q_vector->rx.ring) {
3227 			snprintf(q_vector->name, sizeof(q_vector->name),
3228 				 "%s-rx-%u", netdev->name, ri++);
3229 		} else if (q_vector->tx.ring) {
3230 			snprintf(q_vector->name, sizeof(q_vector->name),
3231 				 "%s-tx-%u", netdev->name, ti++);
3232 		} else {
3233 			/* skip this unused q_vector */
3234 			continue;
3235 		}
3236 		err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
3237 				  q_vector->name, q_vector);
3238 		if (err) {
3239 			e_err(probe, "request_irq failed for MSIX interrupt "
3240 			      "Error: %d\n", err);
3241 			goto free_queue_irqs;
3242 		}
3243 		/* If Flow Director is enabled, set interrupt affinity */
3244 		if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3245 			/* assign the mask for this irq */
3246 			irq_set_affinity_hint(entry->vector,
3247 					      &q_vector->affinity_mask);
3248 		}
3249 	}
3250 
3251 	err = request_irq(adapter->msix_entries[vector].vector,
3252 			  ixgbe_msix_other, 0, netdev->name, adapter);
3253 	if (err) {
3254 		e_err(probe, "request_irq for msix_other failed: %d\n", err);
3255 		goto free_queue_irqs;
3256 	}
3257 
3258 	return 0;
3259 
3260 free_queue_irqs:
3261 	while (vector) {
3262 		vector--;
3263 		irq_set_affinity_hint(adapter->msix_entries[vector].vector,
3264 				      NULL);
3265 		free_irq(adapter->msix_entries[vector].vector,
3266 			 adapter->q_vector[vector]);
3267 	}
3268 	adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3269 	pci_disable_msix(adapter->pdev);
3270 	kfree(adapter->msix_entries);
3271 	adapter->msix_entries = NULL;
3272 	return err;
3273 }
3274 
3275 /**
3276  * ixgbe_intr - legacy mode Interrupt Handler
3277  * @irq: interrupt number
3278  * @data: pointer to a network interface device structure
3279  **/
3280 static irqreturn_t ixgbe_intr(int irq, void *data)
3281 {
3282 	struct ixgbe_adapter *adapter = data;
3283 	struct ixgbe_hw *hw = &adapter->hw;
3284 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3285 	u32 eicr;
3286 
3287 	/*
3288 	 * Workaround for silicon errata #26 on 82598.  Mask the interrupt
3289 	 * before the read of EICR.
3290 	 */
3291 	IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
3292 
3293 	/* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
3294 	 * therefore no explicit interrupt disable is necessary */
3295 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
3296 	if (!eicr) {
3297 		/*
3298 		 * shared interrupt alert!
3299 		 * make sure interrupts are enabled because the read will
3300 		 * have disabled interrupts due to EIAM
3301 		 * finish the workaround of silicon errata on 82598.  Unmask
3302 		 * the interrupt that we masked before the EICR read.
3303 		 */
3304 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
3305 			ixgbe_irq_enable(adapter, true, true);
3306 		return IRQ_NONE;	/* Not our interrupt */
3307 	}
3308 
3309 	if (eicr & IXGBE_EICR_LSC)
3310 		ixgbe_check_lsc(adapter);
3311 
3312 	switch (hw->mac.type) {
3313 	case ixgbe_mac_82599EB:
3314 		ixgbe_check_sfp_event(adapter, eicr);
3315 		fallthrough;
3316 	case ixgbe_mac_X540:
3317 	case ixgbe_mac_X550:
3318 	case ixgbe_mac_X550EM_x:
3319 	case ixgbe_mac_x550em_a:
3320 		if (eicr & IXGBE_EICR_ECC) {
3321 			e_info(link, "Received ECC Err, initiating reset\n");
3322 			set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3323 			ixgbe_service_event_schedule(adapter);
3324 			IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3325 		}
3326 		ixgbe_check_overtemp_event(adapter, eicr);
3327 		break;
3328 	default:
3329 		break;
3330 	}
3331 
3332 	ixgbe_check_fan_failure(adapter, eicr);
3333 	if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3334 		ixgbe_ptp_check_pps_event(adapter);
3335 
3336 	/* would disable interrupts here but EIAM disabled it */
3337 	napi_schedule_irqoff(&q_vector->napi);
3338 
3339 	/*
3340 	 * re-enable link(maybe) and non-queue interrupts, no flush.
3341 	 * ixgbe_poll will re-enable the queue interrupts
3342 	 */
3343 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
3344 		ixgbe_irq_enable(adapter, false, false);
3345 
3346 	return IRQ_HANDLED;
3347 }
3348 
3349 /**
3350  * ixgbe_request_irq - initialize interrupts
3351  * @adapter: board private structure
3352  *
3353  * Attempts to configure interrupts using the best available
3354  * capabilities of the hardware and kernel.
3355  **/
3356 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
3357 {
3358 	struct net_device *netdev = adapter->netdev;
3359 	int err;
3360 
3361 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3362 		err = ixgbe_request_msix_irqs(adapter);
3363 	else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
3364 		err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
3365 				  netdev->name, adapter);
3366 	else
3367 		err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
3368 				  netdev->name, adapter);
3369 
3370 	if (err)
3371 		e_err(probe, "request_irq failed, Error %d\n", err);
3372 
3373 	return err;
3374 }
3375 
3376 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
3377 {
3378 	int vector;
3379 
3380 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
3381 		free_irq(adapter->pdev->irq, adapter);
3382 		return;
3383 	}
3384 
3385 	if (!adapter->msix_entries)
3386 		return;
3387 
3388 	for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3389 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3390 		struct msix_entry *entry = &adapter->msix_entries[vector];
3391 
3392 		/* free only the irqs that were actually requested */
3393 		if (!q_vector->rx.ring && !q_vector->tx.ring)
3394 			continue;
3395 
3396 		/* clear the affinity_mask in the IRQ descriptor */
3397 		irq_set_affinity_hint(entry->vector, NULL);
3398 
3399 		free_irq(entry->vector, q_vector);
3400 	}
3401 
3402 	free_irq(adapter->msix_entries[vector].vector, adapter);
3403 }
3404 
3405 /**
3406  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3407  * @adapter: board private structure
3408  **/
3409 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3410 {
3411 	switch (adapter->hw.mac.type) {
3412 	case ixgbe_mac_82598EB:
3413 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3414 		break;
3415 	case ixgbe_mac_82599EB:
3416 	case ixgbe_mac_X540:
3417 	case ixgbe_mac_X550:
3418 	case ixgbe_mac_X550EM_x:
3419 	case ixgbe_mac_x550em_a:
3420 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3421 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3422 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3423 		break;
3424 	default:
3425 		break;
3426 	}
3427 	IXGBE_WRITE_FLUSH(&adapter->hw);
3428 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3429 		int vector;
3430 
3431 		for (vector = 0; vector < adapter->num_q_vectors; vector++)
3432 			synchronize_irq(adapter->msix_entries[vector].vector);
3433 
3434 		synchronize_irq(adapter->msix_entries[vector++].vector);
3435 	} else {
3436 		synchronize_irq(adapter->pdev->irq);
3437 	}
3438 }
3439 
3440 /**
3441  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3442  * @adapter: board private structure
3443  *
3444  **/
3445 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3446 {
3447 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3448 
3449 	ixgbe_write_eitr(q_vector);
3450 
3451 	ixgbe_set_ivar(adapter, 0, 0, 0);
3452 	ixgbe_set_ivar(adapter, 1, 0, 0);
3453 
3454 	e_info(hw, "Legacy interrupt IVAR setup done\n");
3455 }
3456 
3457 /**
3458  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3459  * @adapter: board private structure
3460  * @ring: structure containing ring specific data
3461  *
3462  * Configure the Tx descriptor ring after a reset.
3463  **/
3464 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3465 			     struct ixgbe_ring *ring)
3466 {
3467 	struct ixgbe_hw *hw = &adapter->hw;
3468 	u64 tdba = ring->dma;
3469 	int wait_loop = 10;
3470 	u32 txdctl = IXGBE_TXDCTL_ENABLE;
3471 	u8 reg_idx = ring->reg_idx;
3472 
3473 	ring->xsk_pool = NULL;
3474 	if (ring_is_xdp(ring))
3475 		ring->xsk_pool = ixgbe_xsk_pool(adapter, ring);
3476 
3477 	/* disable queue to avoid issues while updating state */
3478 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3479 	IXGBE_WRITE_FLUSH(hw);
3480 
3481 	IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3482 			(tdba & DMA_BIT_MASK(32)));
3483 	IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3484 	IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3485 			ring->count * sizeof(union ixgbe_adv_tx_desc));
3486 	IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3487 	IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3488 	ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3489 
3490 	/*
3491 	 * set WTHRESH to encourage burst writeback, it should not be set
3492 	 * higher than 1 when:
3493 	 * - ITR is 0 as it could cause false TX hangs
3494 	 * - ITR is set to > 100k int/sec and BQL is enabled
3495 	 *
3496 	 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3497 	 * to or less than the number of on chip descriptors, which is
3498 	 * currently 40.
3499 	 */
3500 	if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3501 		txdctl |= 1u << 16;	/* WTHRESH = 1 */
3502 	else
3503 		txdctl |= 8u << 16;	/* WTHRESH = 8 */
3504 
3505 	/*
3506 	 * Setting PTHRESH to 32 both improves performance
3507 	 * and avoids a TX hang with DFP enabled
3508 	 */
3509 	txdctl |= (1u << 8) |	/* HTHRESH = 1 */
3510 		   32;		/* PTHRESH = 32 */
3511 
3512 	/* reinitialize flowdirector state */
3513 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3514 		ring->atr_sample_rate = adapter->atr_sample_rate;
3515 		ring->atr_count = 0;
3516 		set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3517 	} else {
3518 		ring->atr_sample_rate = 0;
3519 	}
3520 
3521 	/* initialize XPS */
3522 	if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3523 		struct ixgbe_q_vector *q_vector = ring->q_vector;
3524 
3525 		if (q_vector)
3526 			netif_set_xps_queue(ring->netdev,
3527 					    &q_vector->affinity_mask,
3528 					    ring->queue_index);
3529 	}
3530 
3531 	clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3532 
3533 	/* reinitialize tx_buffer_info */
3534 	memset(ring->tx_buffer_info, 0,
3535 	       sizeof(struct ixgbe_tx_buffer) * ring->count);
3536 
3537 	/* enable queue */
3538 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3539 
3540 	/* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3541 	if (hw->mac.type == ixgbe_mac_82598EB &&
3542 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3543 		return;
3544 
3545 	/* poll to verify queue is enabled */
3546 	do {
3547 		usleep_range(1000, 2000);
3548 		txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3549 	} while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3550 	if (!wait_loop)
3551 		hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
3552 }
3553 
3554 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3555 {
3556 	struct ixgbe_hw *hw = &adapter->hw;
3557 	u32 rttdcs, mtqc;
3558 	u8 tcs = adapter->hw_tcs;
3559 
3560 	if (hw->mac.type == ixgbe_mac_82598EB)
3561 		return;
3562 
3563 	/* disable the arbiter while setting MTQC */
3564 	rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3565 	rttdcs |= IXGBE_RTTDCS_ARBDIS;
3566 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3567 
3568 	/* set transmit pool layout */
3569 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3570 		mtqc = IXGBE_MTQC_VT_ENA;
3571 		if (tcs > 4)
3572 			mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3573 		else if (tcs > 1)
3574 			mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3575 		else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3576 			 IXGBE_82599_VMDQ_4Q_MASK)
3577 			mtqc |= IXGBE_MTQC_32VF;
3578 		else
3579 			mtqc |= IXGBE_MTQC_64VF;
3580 	} else {
3581 		if (tcs > 4) {
3582 			mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3583 		} else if (tcs > 1) {
3584 			mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3585 		} else {
3586 			u8 max_txq = adapter->num_tx_queues +
3587 				adapter->num_xdp_queues;
3588 			if (max_txq > 63)
3589 				mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3590 			else
3591 				mtqc = IXGBE_MTQC_64Q_1PB;
3592 		}
3593 	}
3594 
3595 	IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3596 
3597 	/* Enable Security TX Buffer IFG for multiple pb */
3598 	if (tcs) {
3599 		u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3600 		sectx |= IXGBE_SECTX_DCB;
3601 		IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3602 	}
3603 
3604 	/* re-enable the arbiter */
3605 	rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3606 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3607 }
3608 
3609 /**
3610  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3611  * @adapter: board private structure
3612  *
3613  * Configure the Tx unit of the MAC after a reset.
3614  **/
3615 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3616 {
3617 	struct ixgbe_hw *hw = &adapter->hw;
3618 	u32 dmatxctl;
3619 	u32 i;
3620 
3621 	ixgbe_setup_mtqc(adapter);
3622 
3623 	if (hw->mac.type != ixgbe_mac_82598EB) {
3624 		/* DMATXCTL.EN must be before Tx queues are enabled */
3625 		dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3626 		dmatxctl |= IXGBE_DMATXCTL_TE;
3627 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3628 	}
3629 
3630 	/* Setup the HW Tx Head and Tail descriptor pointers */
3631 	for (i = 0; i < adapter->num_tx_queues; i++)
3632 		ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3633 	for (i = 0; i < adapter->num_xdp_queues; i++)
3634 		ixgbe_configure_tx_ring(adapter, adapter->xdp_ring[i]);
3635 }
3636 
3637 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3638 				 struct ixgbe_ring *ring)
3639 {
3640 	struct ixgbe_hw *hw = &adapter->hw;
3641 	u8 reg_idx = ring->reg_idx;
3642 	u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3643 
3644 	srrctl |= IXGBE_SRRCTL_DROP_EN;
3645 
3646 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3647 }
3648 
3649 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3650 				  struct ixgbe_ring *ring)
3651 {
3652 	struct ixgbe_hw *hw = &adapter->hw;
3653 	u8 reg_idx = ring->reg_idx;
3654 	u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3655 
3656 	srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3657 
3658 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3659 }
3660 
3661 #ifdef CONFIG_IXGBE_DCB
3662 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3663 #else
3664 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3665 #endif
3666 {
3667 	int i;
3668 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3669 
3670 	if (adapter->ixgbe_ieee_pfc)
3671 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3672 
3673 	/*
3674 	 * We should set the drop enable bit if:
3675 	 *  SR-IOV is enabled
3676 	 *   or
3677 	 *  Number of Rx queues > 1 and flow control is disabled
3678 	 *
3679 	 *  This allows us to avoid head of line blocking for security
3680 	 *  and performance reasons.
3681 	 */
3682 	if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3683 	    !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3684 		for (i = 0; i < adapter->num_rx_queues; i++)
3685 			ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3686 	} else {
3687 		for (i = 0; i < adapter->num_rx_queues; i++)
3688 			ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3689 	}
3690 }
3691 
3692 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3693 
3694 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3695 				   struct ixgbe_ring *rx_ring)
3696 {
3697 	struct ixgbe_hw *hw = &adapter->hw;
3698 	u32 srrctl;
3699 	u8 reg_idx = rx_ring->reg_idx;
3700 
3701 	if (hw->mac.type == ixgbe_mac_82598EB) {
3702 		u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3703 
3704 		/*
3705 		 * if VMDq is not active we must program one srrctl register
3706 		 * per RSS queue since we have enabled RDRXCTL.MVMEN
3707 		 */
3708 		reg_idx &= mask;
3709 	}
3710 
3711 	/* configure header buffer length, needed for RSC */
3712 	srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3713 
3714 	/* configure the packet buffer length */
3715 	if (rx_ring->xsk_pool) {
3716 		u32 xsk_buf_len = xsk_pool_get_rx_frame_size(rx_ring->xsk_pool);
3717 
3718 		/* If the MAC support setting RXDCTL.RLPML, the
3719 		 * SRRCTL[n].BSIZEPKT is set to PAGE_SIZE and
3720 		 * RXDCTL.RLPML is set to the actual UMEM buffer
3721 		 * size. If not, then we are stuck with a 1k buffer
3722 		 * size resolution. In this case frames larger than
3723 		 * the UMEM buffer size viewed in a 1k resolution will
3724 		 * be dropped.
3725 		 */
3726 		if (hw->mac.type != ixgbe_mac_82599EB)
3727 			srrctl |= PAGE_SIZE >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3728 		else
3729 			srrctl |= xsk_buf_len >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3730 	} else if (test_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state)) {
3731 		srrctl |= IXGBE_RXBUFFER_3K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3732 	} else {
3733 		srrctl |= IXGBE_RXBUFFER_2K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3734 	}
3735 
3736 	/* configure descriptor type */
3737 	srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3738 
3739 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3740 }
3741 
3742 /**
3743  * ixgbe_rss_indir_tbl_entries - Return RSS indirection table entries
3744  * @adapter: device handle
3745  *
3746  *  - 82598/82599/X540:     128
3747  *  - X550(non-SRIOV mode): 512
3748  *  - X550(SRIOV mode):     64
3749  */
3750 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3751 {
3752 	if (adapter->hw.mac.type < ixgbe_mac_X550)
3753 		return 128;
3754 	else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3755 		return 64;
3756 	else
3757 		return 512;
3758 }
3759 
3760 /**
3761  * ixgbe_store_key - Write the RSS key to HW
3762  * @adapter: device handle
3763  *
3764  * Write the RSS key stored in adapter.rss_key to HW.
3765  */
3766 void ixgbe_store_key(struct ixgbe_adapter *adapter)
3767 {
3768 	struct ixgbe_hw *hw = &adapter->hw;
3769 	int i;
3770 
3771 	for (i = 0; i < 10; i++)
3772 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3773 }
3774 
3775 /**
3776  * ixgbe_init_rss_key - Initialize adapter RSS key
3777  * @adapter: device handle
3778  *
3779  * Allocates and initializes the RSS key if it is not allocated.
3780  **/
3781 static inline int ixgbe_init_rss_key(struct ixgbe_adapter *adapter)
3782 {
3783 	u32 *rss_key;
3784 
3785 	if (!adapter->rss_key) {
3786 		rss_key = kzalloc(IXGBE_RSS_KEY_SIZE, GFP_KERNEL);
3787 		if (unlikely(!rss_key))
3788 			return -ENOMEM;
3789 
3790 		netdev_rss_key_fill(rss_key, IXGBE_RSS_KEY_SIZE);
3791 		adapter->rss_key = rss_key;
3792 	}
3793 
3794 	return 0;
3795 }
3796 
3797 /**
3798  * ixgbe_store_reta - Write the RETA table to HW
3799  * @adapter: device handle
3800  *
3801  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3802  */
3803 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3804 {
3805 	u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3806 	struct ixgbe_hw *hw = &adapter->hw;
3807 	u32 reta = 0;
3808 	u32 indices_multi;
3809 	u8 *indir_tbl = adapter->rss_indir_tbl;
3810 
3811 	/* Fill out the redirection table as follows:
3812 	 *  - 82598:      8 bit wide entries containing pair of 4 bit RSS
3813 	 *    indices.
3814 	 *  - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3815 	 *  - X550:       8 bit wide entries containing 6 bit RSS index
3816 	 */
3817 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3818 		indices_multi = 0x11;
3819 	else
3820 		indices_multi = 0x1;
3821 
3822 	/* Write redirection table to HW */
3823 	for (i = 0; i < reta_entries; i++) {
3824 		reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3825 		if ((i & 3) == 3) {
3826 			if (i < 128)
3827 				IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3828 			else
3829 				IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3830 						reta);
3831 			reta = 0;
3832 		}
3833 	}
3834 }
3835 
3836 /**
3837  * ixgbe_store_vfreta - Write the RETA table to HW (x550 devices in SRIOV mode)
3838  * @adapter: device handle
3839  *
3840  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3841  */
3842 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3843 {
3844 	u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3845 	struct ixgbe_hw *hw = &adapter->hw;
3846 	u32 vfreta = 0;
3847 
3848 	/* Write redirection table to HW */
3849 	for (i = 0; i < reta_entries; i++) {
3850 		u16 pool = adapter->num_rx_pools;
3851 
3852 		vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3853 		if ((i & 3) != 3)
3854 			continue;
3855 
3856 		while (pool--)
3857 			IXGBE_WRITE_REG(hw,
3858 					IXGBE_PFVFRETA(i >> 2, VMDQ_P(pool)),
3859 					vfreta);
3860 		vfreta = 0;
3861 	}
3862 }
3863 
3864 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3865 {
3866 	u32 i, j;
3867 	u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3868 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3869 
3870 	/* Program table for at least 4 queues w/ SR-IOV so that VFs can
3871 	 * make full use of any rings they may have.  We will use the
3872 	 * PSRTYPE register to control how many rings we use within the PF.
3873 	 */
3874 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 4))
3875 		rss_i = 4;
3876 
3877 	/* Fill out hash function seeds */
3878 	ixgbe_store_key(adapter);
3879 
3880 	/* Fill out redirection table */
3881 	memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3882 
3883 	for (i = 0, j = 0; i < reta_entries; i++, j++) {
3884 		if (j == rss_i)
3885 			j = 0;
3886 
3887 		adapter->rss_indir_tbl[i] = j;
3888 	}
3889 
3890 	ixgbe_store_reta(adapter);
3891 }
3892 
3893 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3894 {
3895 	struct ixgbe_hw *hw = &adapter->hw;
3896 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3897 	int i, j;
3898 
3899 	/* Fill out hash function seeds */
3900 	for (i = 0; i < 10; i++) {
3901 		u16 pool = adapter->num_rx_pools;
3902 
3903 		while (pool--)
3904 			IXGBE_WRITE_REG(hw,
3905 					IXGBE_PFVFRSSRK(i, VMDQ_P(pool)),
3906 					*(adapter->rss_key + i));
3907 	}
3908 
3909 	/* Fill out the redirection table */
3910 	for (i = 0, j = 0; i < 64; i++, j++) {
3911 		if (j == rss_i)
3912 			j = 0;
3913 
3914 		adapter->rss_indir_tbl[i] = j;
3915 	}
3916 
3917 	ixgbe_store_vfreta(adapter);
3918 }
3919 
3920 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3921 {
3922 	struct ixgbe_hw *hw = &adapter->hw;
3923 	u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3924 	u32 rxcsum;
3925 
3926 	/* Disable indicating checksum in descriptor, enables RSS hash */
3927 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3928 	rxcsum |= IXGBE_RXCSUM_PCSD;
3929 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3930 
3931 	if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3932 		if (adapter->ring_feature[RING_F_RSS].mask)
3933 			mrqc = IXGBE_MRQC_RSSEN;
3934 	} else {
3935 		u8 tcs = adapter->hw_tcs;
3936 
3937 		if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3938 			if (tcs > 4)
3939 				mrqc = IXGBE_MRQC_VMDQRT8TCEN;	/* 8 TCs */
3940 			else if (tcs > 1)
3941 				mrqc = IXGBE_MRQC_VMDQRT4TCEN;	/* 4 TCs */
3942 			else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3943 				 IXGBE_82599_VMDQ_4Q_MASK)
3944 				mrqc = IXGBE_MRQC_VMDQRSS32EN;
3945 			else
3946 				mrqc = IXGBE_MRQC_VMDQRSS64EN;
3947 
3948 			/* Enable L3/L4 for Tx Switched packets only for X550,
3949 			 * older devices do not support this feature
3950 			 */
3951 			if (hw->mac.type >= ixgbe_mac_X550)
3952 				mrqc |= IXGBE_MRQC_L3L4TXSWEN;
3953 		} else {
3954 			if (tcs > 4)
3955 				mrqc = IXGBE_MRQC_RTRSS8TCEN;
3956 			else if (tcs > 1)
3957 				mrqc = IXGBE_MRQC_RTRSS4TCEN;
3958 			else
3959 				mrqc = IXGBE_MRQC_RSSEN;
3960 		}
3961 	}
3962 
3963 	/* Perform hash on these packet types */
3964 	rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3965 		     IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3966 		     IXGBE_MRQC_RSS_FIELD_IPV6 |
3967 		     IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3968 
3969 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3970 		rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3971 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3972 		rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3973 
3974 	if ((hw->mac.type >= ixgbe_mac_X550) &&
3975 	    (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
3976 		u16 pool = adapter->num_rx_pools;
3977 
3978 		/* Enable VF RSS mode */
3979 		mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
3980 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3981 
3982 		/* Setup RSS through the VF registers */
3983 		ixgbe_setup_vfreta(adapter);
3984 		vfmrqc = IXGBE_MRQC_RSSEN;
3985 		vfmrqc |= rss_field;
3986 
3987 		while (pool--)
3988 			IXGBE_WRITE_REG(hw,
3989 					IXGBE_PFVFMRQC(VMDQ_P(pool)),
3990 					vfmrqc);
3991 	} else {
3992 		ixgbe_setup_reta(adapter);
3993 		mrqc |= rss_field;
3994 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3995 	}
3996 }
3997 
3998 /**
3999  * ixgbe_configure_rscctl - enable RSC for the indicated ring
4000  * @adapter: address of board private structure
4001  * @ring: structure containing ring specific data
4002  **/
4003 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
4004 				   struct ixgbe_ring *ring)
4005 {
4006 	struct ixgbe_hw *hw = &adapter->hw;
4007 	u32 rscctrl;
4008 	u8 reg_idx = ring->reg_idx;
4009 
4010 	if (!ring_is_rsc_enabled(ring))
4011 		return;
4012 
4013 	rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
4014 	rscctrl |= IXGBE_RSCCTL_RSCEN;
4015 	/*
4016 	 * we must limit the number of descriptors so that the
4017 	 * total size of max desc * buf_len is not greater
4018 	 * than 65536
4019 	 */
4020 	rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
4021 	IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
4022 }
4023 
4024 #define IXGBE_MAX_RX_DESC_POLL 10
4025 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
4026 				       struct ixgbe_ring *ring)
4027 {
4028 	struct ixgbe_hw *hw = &adapter->hw;
4029 	int wait_loop = IXGBE_MAX_RX_DESC_POLL;
4030 	u32 rxdctl;
4031 	u8 reg_idx = ring->reg_idx;
4032 
4033 	if (ixgbe_removed(hw->hw_addr))
4034 		return;
4035 	/* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
4036 	if (hw->mac.type == ixgbe_mac_82598EB &&
4037 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
4038 		return;
4039 
4040 	do {
4041 		usleep_range(1000, 2000);
4042 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4043 	} while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4044 
4045 	if (!wait_loop) {
4046 		e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
4047 		      "the polling period\n", reg_idx);
4048 	}
4049 }
4050 
4051 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
4052 			     struct ixgbe_ring *ring)
4053 {
4054 	struct ixgbe_hw *hw = &adapter->hw;
4055 	union ixgbe_adv_rx_desc *rx_desc;
4056 	u64 rdba = ring->dma;
4057 	u32 rxdctl;
4058 	u8 reg_idx = ring->reg_idx;
4059 
4060 	xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
4061 	ring->xsk_pool = ixgbe_xsk_pool(adapter, ring);
4062 	if (ring->xsk_pool) {
4063 		WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4064 						   MEM_TYPE_XSK_BUFF_POOL,
4065 						   NULL));
4066 		xsk_pool_set_rxq_info(ring->xsk_pool, &ring->xdp_rxq);
4067 	} else {
4068 		WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4069 						   MEM_TYPE_PAGE_SHARED, NULL));
4070 	}
4071 
4072 	/* disable queue to avoid use of these values while updating state */
4073 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4074 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4075 
4076 	/* write value back with RXDCTL.ENABLE bit cleared */
4077 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4078 	IXGBE_WRITE_FLUSH(hw);
4079 
4080 	IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
4081 	IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
4082 	IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
4083 			ring->count * sizeof(union ixgbe_adv_rx_desc));
4084 	/* Force flushing of IXGBE_RDLEN to prevent MDD */
4085 	IXGBE_WRITE_FLUSH(hw);
4086 
4087 	IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
4088 	IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
4089 	ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
4090 
4091 	ixgbe_configure_srrctl(adapter, ring);
4092 	ixgbe_configure_rscctl(adapter, ring);
4093 
4094 	if (hw->mac.type == ixgbe_mac_82598EB) {
4095 		/*
4096 		 * enable cache line friendly hardware writes:
4097 		 * PTHRESH=32 descriptors (half the internal cache),
4098 		 * this also removes ugly rx_no_buffer_count increment
4099 		 * HTHRESH=4 descriptors (to minimize latency on fetch)
4100 		 * WTHRESH=8 burst writeback up to two cache lines
4101 		 */
4102 		rxdctl &= ~0x3FFFFF;
4103 		rxdctl |=  0x080420;
4104 #if (PAGE_SIZE < 8192)
4105 	/* RXDCTL.RLPML does not work on 82599 */
4106 	} else if (hw->mac.type != ixgbe_mac_82599EB) {
4107 		rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4108 			    IXGBE_RXDCTL_RLPML_EN);
4109 
4110 		/* Limit the maximum frame size so we don't overrun the skb.
4111 		 * This can happen in SRIOV mode when the MTU of the VF is
4112 		 * higher than the MTU of the PF.
4113 		 */
4114 		if (ring_uses_build_skb(ring) &&
4115 		    !test_bit(__IXGBE_RX_3K_BUFFER, &ring->state))
4116 			rxdctl |= IXGBE_MAX_2K_FRAME_BUILD_SKB |
4117 				  IXGBE_RXDCTL_RLPML_EN;
4118 #endif
4119 	}
4120 
4121 	if (ring->xsk_pool && hw->mac.type != ixgbe_mac_82599EB) {
4122 		u32 xsk_buf_len = xsk_pool_get_rx_frame_size(ring->xsk_pool);
4123 
4124 		rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4125 			    IXGBE_RXDCTL_RLPML_EN);
4126 		rxdctl |= xsk_buf_len | IXGBE_RXDCTL_RLPML_EN;
4127 
4128 		ring->rx_buf_len = xsk_buf_len;
4129 	}
4130 
4131 	/* initialize rx_buffer_info */
4132 	memset(ring->rx_buffer_info, 0,
4133 	       sizeof(struct ixgbe_rx_buffer) * ring->count);
4134 
4135 	/* initialize Rx descriptor 0 */
4136 	rx_desc = IXGBE_RX_DESC(ring, 0);
4137 	rx_desc->wb.upper.length = 0;
4138 
4139 	/* enable receive descriptor ring */
4140 	rxdctl |= IXGBE_RXDCTL_ENABLE;
4141 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4142 
4143 	ixgbe_rx_desc_queue_enable(adapter, ring);
4144 	if (ring->xsk_pool)
4145 		ixgbe_alloc_rx_buffers_zc(ring, ixgbe_desc_unused(ring));
4146 	else
4147 		ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
4148 }
4149 
4150 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
4151 {
4152 	struct ixgbe_hw *hw = &adapter->hw;
4153 	int rss_i = adapter->ring_feature[RING_F_RSS].indices;
4154 	u16 pool = adapter->num_rx_pools;
4155 
4156 	/* PSRTYPE must be initialized in non 82598 adapters */
4157 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4158 		      IXGBE_PSRTYPE_UDPHDR |
4159 		      IXGBE_PSRTYPE_IPV4HDR |
4160 		      IXGBE_PSRTYPE_L2HDR |
4161 		      IXGBE_PSRTYPE_IPV6HDR;
4162 
4163 	if (hw->mac.type == ixgbe_mac_82598EB)
4164 		return;
4165 
4166 	if (rss_i > 3)
4167 		psrtype |= 2u << 29;
4168 	else if (rss_i > 1)
4169 		psrtype |= 1u << 29;
4170 
4171 	while (pool--)
4172 		IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4173 }
4174 
4175 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
4176 {
4177 	struct ixgbe_hw *hw = &adapter->hw;
4178 	u16 pool = adapter->num_rx_pools;
4179 	u32 reg_offset, vf_shift, vmolr;
4180 	u32 gcr_ext, vmdctl;
4181 	int i;
4182 
4183 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
4184 		return;
4185 
4186 	vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
4187 	vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
4188 	vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
4189 	vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
4190 	vmdctl |= IXGBE_VT_CTL_REPLEN;
4191 	IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
4192 
4193 	/* accept untagged packets until a vlan tag is
4194 	 * specifically set for the VMDQ queue/pool
4195 	 */
4196 	vmolr = IXGBE_VMOLR_AUPE;
4197 	while (pool--)
4198 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(pool)), vmolr);
4199 
4200 	vf_shift = VMDQ_P(0) % 32;
4201 	reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
4202 
4203 	/* Enable only the PF's pool for Tx/Rx */
4204 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), GENMASK(31, vf_shift));
4205 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
4206 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), GENMASK(31, vf_shift));
4207 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
4208 	if (adapter->bridge_mode == BRIDGE_MODE_VEB)
4209 		IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
4210 
4211 	/* Map PF MAC address in RAR Entry 0 to first pool following VFs */
4212 	hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
4213 
4214 	/* clear VLAN promisc flag so VFTA will be updated if necessary */
4215 	adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4216 
4217 	/*
4218 	 * Set up VF register offsets for selected VT Mode,
4219 	 * i.e. 32 or 64 VFs for SR-IOV
4220 	 */
4221 	switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4222 	case IXGBE_82599_VMDQ_8Q_MASK:
4223 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
4224 		break;
4225 	case IXGBE_82599_VMDQ_4Q_MASK:
4226 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
4227 		break;
4228 	default:
4229 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
4230 		break;
4231 	}
4232 
4233 	IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
4234 
4235 	for (i = 0; i < adapter->num_vfs; i++) {
4236 		/* configure spoof checking */
4237 		ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i,
4238 					  adapter->vfinfo[i].spoofchk_enabled);
4239 
4240 		/* Enable/Disable RSS query feature  */
4241 		ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
4242 					  adapter->vfinfo[i].rss_query_enabled);
4243 	}
4244 }
4245 
4246 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
4247 {
4248 	struct ixgbe_hw *hw = &adapter->hw;
4249 	struct net_device *netdev = adapter->netdev;
4250 	int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4251 	struct ixgbe_ring *rx_ring;
4252 	int i;
4253 	u32 mhadd, hlreg0;
4254 
4255 #ifdef IXGBE_FCOE
4256 	/* adjust max frame to be able to do baby jumbo for FCoE */
4257 	if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
4258 	    (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
4259 		max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4260 
4261 #endif /* IXGBE_FCOE */
4262 
4263 	/* adjust max frame to be at least the size of a standard frame */
4264 	if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
4265 		max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
4266 
4267 	mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
4268 	if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
4269 		mhadd &= ~IXGBE_MHADD_MFS_MASK;
4270 		mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
4271 
4272 		IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
4273 	}
4274 
4275 	hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4276 	/* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
4277 	hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4278 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4279 
4280 	/*
4281 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
4282 	 * the Base and Length of the Rx Descriptor Ring
4283 	 */
4284 	for (i = 0; i < adapter->num_rx_queues; i++) {
4285 		rx_ring = adapter->rx_ring[i];
4286 
4287 		clear_ring_rsc_enabled(rx_ring);
4288 		clear_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4289 		clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4290 
4291 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4292 			set_ring_rsc_enabled(rx_ring);
4293 
4294 		if (test_bit(__IXGBE_RX_FCOE, &rx_ring->state))
4295 			set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4296 
4297 		if (adapter->flags2 & IXGBE_FLAG2_RX_LEGACY)
4298 			continue;
4299 
4300 		set_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4301 
4302 #if (PAGE_SIZE < 8192)
4303 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4304 			set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4305 
4306 		if (IXGBE_2K_TOO_SMALL_WITH_PADDING ||
4307 		    (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
4308 			set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4309 #endif
4310 	}
4311 }
4312 
4313 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
4314 {
4315 	struct ixgbe_hw *hw = &adapter->hw;
4316 	u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4317 
4318 	switch (hw->mac.type) {
4319 	case ixgbe_mac_82598EB:
4320 		/*
4321 		 * For VMDq support of different descriptor types or
4322 		 * buffer sizes through the use of multiple SRRCTL
4323 		 * registers, RDRXCTL.MVMEN must be set to 1
4324 		 *
4325 		 * also, the manual doesn't mention it clearly but DCA hints
4326 		 * will only use queue 0's tags unless this bit is set.  Side
4327 		 * effects of setting this bit are only that SRRCTL must be
4328 		 * fully programmed [0..15]
4329 		 */
4330 		rdrxctl |= IXGBE_RDRXCTL_MVMEN;
4331 		break;
4332 	case ixgbe_mac_X550:
4333 	case ixgbe_mac_X550EM_x:
4334 	case ixgbe_mac_x550em_a:
4335 		if (adapter->num_vfs)
4336 			rdrxctl |= IXGBE_RDRXCTL_PSP;
4337 		fallthrough;
4338 	case ixgbe_mac_82599EB:
4339 	case ixgbe_mac_X540:
4340 		/* Disable RSC for ACK packets */
4341 		IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
4342 		   (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
4343 		rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4344 		/* hardware requires some bits to be set by default */
4345 		rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
4346 		rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4347 		break;
4348 	default:
4349 		/* We should do nothing since we don't know this hardware */
4350 		return;
4351 	}
4352 
4353 	IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4354 }
4355 
4356 /**
4357  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
4358  * @adapter: board private structure
4359  *
4360  * Configure the Rx unit of the MAC after a reset.
4361  **/
4362 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
4363 {
4364 	struct ixgbe_hw *hw = &adapter->hw;
4365 	int i;
4366 	u32 rxctrl, rfctl;
4367 
4368 	/* disable receives while setting up the descriptors */
4369 	hw->mac.ops.disable_rx(hw);
4370 
4371 	ixgbe_setup_psrtype(adapter);
4372 	ixgbe_setup_rdrxctl(adapter);
4373 
4374 	/* RSC Setup */
4375 	rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4376 	rfctl &= ~IXGBE_RFCTL_RSC_DIS;
4377 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
4378 		rfctl |= IXGBE_RFCTL_RSC_DIS;
4379 
4380 	/* disable NFS filtering */
4381 	rfctl |= (IXGBE_RFCTL_NFSW_DIS | IXGBE_RFCTL_NFSR_DIS);
4382 	IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4383 
4384 	/* Program registers for the distribution of queues */
4385 	ixgbe_setup_mrqc(adapter);
4386 
4387 	/* set_rx_buffer_len must be called before ring initialization */
4388 	ixgbe_set_rx_buffer_len(adapter);
4389 
4390 	/*
4391 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
4392 	 * the Base and Length of the Rx Descriptor Ring
4393 	 */
4394 	for (i = 0; i < adapter->num_rx_queues; i++)
4395 		ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
4396 
4397 	rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4398 	/* disable drop enable for 82598 parts */
4399 	if (hw->mac.type == ixgbe_mac_82598EB)
4400 		rxctrl |= IXGBE_RXCTRL_DMBYPS;
4401 
4402 	/* enable all receives */
4403 	rxctrl |= IXGBE_RXCTRL_RXEN;
4404 	hw->mac.ops.enable_rx_dma(hw, rxctrl);
4405 }
4406 
4407 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
4408 				 __be16 proto, u16 vid)
4409 {
4410 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4411 	struct ixgbe_hw *hw = &adapter->hw;
4412 
4413 	/* add VID to filter table */
4414 	if (!vid || !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4415 		hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true, !!vid);
4416 
4417 	set_bit(vid, adapter->active_vlans);
4418 
4419 	return 0;
4420 }
4421 
4422 static int ixgbe_find_vlvf_entry(struct ixgbe_hw *hw, u32 vlan)
4423 {
4424 	u32 vlvf;
4425 	int idx;
4426 
4427 	/* short cut the special case */
4428 	if (vlan == 0)
4429 		return 0;
4430 
4431 	/* Search for the vlan id in the VLVF entries */
4432 	for (idx = IXGBE_VLVF_ENTRIES; --idx;) {
4433 		vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(idx));
4434 		if ((vlvf & VLAN_VID_MASK) == vlan)
4435 			break;
4436 	}
4437 
4438 	return idx;
4439 }
4440 
4441 void ixgbe_update_pf_promisc_vlvf(struct ixgbe_adapter *adapter, u32 vid)
4442 {
4443 	struct ixgbe_hw *hw = &adapter->hw;
4444 	u32 bits, word;
4445 	int idx;
4446 
4447 	idx = ixgbe_find_vlvf_entry(hw, vid);
4448 	if (!idx)
4449 		return;
4450 
4451 	/* See if any other pools are set for this VLAN filter
4452 	 * entry other than the PF.
4453 	 */
4454 	word = idx * 2 + (VMDQ_P(0) / 32);
4455 	bits = ~BIT(VMDQ_P(0) % 32);
4456 	bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4457 
4458 	/* Disable the filter so this falls into the default pool. */
4459 	if (!bits && !IXGBE_READ_REG(hw, IXGBE_VLVFB(word ^ 1))) {
4460 		if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4461 			IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), 0);
4462 		IXGBE_WRITE_REG(hw, IXGBE_VLVF(idx), 0);
4463 	}
4464 }
4465 
4466 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
4467 				  __be16 proto, u16 vid)
4468 {
4469 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4470 	struct ixgbe_hw *hw = &adapter->hw;
4471 
4472 	/* remove VID from filter table */
4473 	if (vid && !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4474 		hw->mac.ops.set_vfta(hw, vid, VMDQ_P(0), false, true);
4475 
4476 	clear_bit(vid, adapter->active_vlans);
4477 
4478 	return 0;
4479 }
4480 
4481 /**
4482  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
4483  * @adapter: driver data
4484  */
4485 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
4486 {
4487 	struct ixgbe_hw *hw = &adapter->hw;
4488 	u32 vlnctrl;
4489 	int i, j;
4490 
4491 	switch (hw->mac.type) {
4492 	case ixgbe_mac_82598EB:
4493 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4494 		vlnctrl &= ~IXGBE_VLNCTRL_VME;
4495 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4496 		break;
4497 	case ixgbe_mac_82599EB:
4498 	case ixgbe_mac_X540:
4499 	case ixgbe_mac_X550:
4500 	case ixgbe_mac_X550EM_x:
4501 	case ixgbe_mac_x550em_a:
4502 		for (i = 0; i < adapter->num_rx_queues; i++) {
4503 			struct ixgbe_ring *ring = adapter->rx_ring[i];
4504 
4505 			if (!netif_is_ixgbe(ring->netdev))
4506 				continue;
4507 
4508 			j = ring->reg_idx;
4509 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4510 			vlnctrl &= ~IXGBE_RXDCTL_VME;
4511 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4512 		}
4513 		break;
4514 	default:
4515 		break;
4516 	}
4517 }
4518 
4519 /**
4520  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
4521  * @adapter: driver data
4522  */
4523 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
4524 {
4525 	struct ixgbe_hw *hw = &adapter->hw;
4526 	u32 vlnctrl;
4527 	int i, j;
4528 
4529 	switch (hw->mac.type) {
4530 	case ixgbe_mac_82598EB:
4531 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4532 		vlnctrl |= IXGBE_VLNCTRL_VME;
4533 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4534 		break;
4535 	case ixgbe_mac_82599EB:
4536 	case ixgbe_mac_X540:
4537 	case ixgbe_mac_X550:
4538 	case ixgbe_mac_X550EM_x:
4539 	case ixgbe_mac_x550em_a:
4540 		for (i = 0; i < adapter->num_rx_queues; i++) {
4541 			struct ixgbe_ring *ring = adapter->rx_ring[i];
4542 
4543 			if (!netif_is_ixgbe(ring->netdev))
4544 				continue;
4545 
4546 			j = ring->reg_idx;
4547 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4548 			vlnctrl |= IXGBE_RXDCTL_VME;
4549 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4550 		}
4551 		break;
4552 	default:
4553 		break;
4554 	}
4555 }
4556 
4557 static void ixgbe_vlan_promisc_enable(struct ixgbe_adapter *adapter)
4558 {
4559 	struct ixgbe_hw *hw = &adapter->hw;
4560 	u32 vlnctrl, i;
4561 
4562 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4563 
4564 	if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
4565 	/* For VMDq and SR-IOV we must leave VLAN filtering enabled */
4566 		vlnctrl |= IXGBE_VLNCTRL_VFE;
4567 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4568 	} else {
4569 		vlnctrl &= ~IXGBE_VLNCTRL_VFE;
4570 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4571 		return;
4572 	}
4573 
4574 	/* Nothing to do for 82598 */
4575 	if (hw->mac.type == ixgbe_mac_82598EB)
4576 		return;
4577 
4578 	/* We are already in VLAN promisc, nothing to do */
4579 	if (adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)
4580 		return;
4581 
4582 	/* Set flag so we don't redo unnecessary work */
4583 	adapter->flags2 |= IXGBE_FLAG2_VLAN_PROMISC;
4584 
4585 	/* Add PF to all active pools */
4586 	for (i = IXGBE_VLVF_ENTRIES; --i;) {
4587 		u32 reg_offset = IXGBE_VLVFB(i * 2 + VMDQ_P(0) / 32);
4588 		u32 vlvfb = IXGBE_READ_REG(hw, reg_offset);
4589 
4590 		vlvfb |= BIT(VMDQ_P(0) % 32);
4591 		IXGBE_WRITE_REG(hw, reg_offset, vlvfb);
4592 	}
4593 
4594 	/* Set all bits in the VLAN filter table array */
4595 	for (i = hw->mac.vft_size; i--;)
4596 		IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), ~0U);
4597 }
4598 
4599 #define VFTA_BLOCK_SIZE 8
4600 static void ixgbe_scrub_vfta(struct ixgbe_adapter *adapter, u32 vfta_offset)
4601 {
4602 	struct ixgbe_hw *hw = &adapter->hw;
4603 	u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4604 	u32 vid_start = vfta_offset * 32;
4605 	u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4606 	u32 i, vid, word, bits;
4607 
4608 	for (i = IXGBE_VLVF_ENTRIES; --i;) {
4609 		u32 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(i));
4610 
4611 		/* pull VLAN ID from VLVF */
4612 		vid = vlvf & VLAN_VID_MASK;
4613 
4614 		/* only concern outselves with a certain range */
4615 		if (vid < vid_start || vid >= vid_end)
4616 			continue;
4617 
4618 		if (vlvf) {
4619 			/* record VLAN ID in VFTA */
4620 			vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4621 
4622 			/* if PF is part of this then continue */
4623 			if (test_bit(vid, adapter->active_vlans))
4624 				continue;
4625 		}
4626 
4627 		/* remove PF from the pool */
4628 		word = i * 2 + VMDQ_P(0) / 32;
4629 		bits = ~BIT(VMDQ_P(0) % 32);
4630 		bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4631 		IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), bits);
4632 	}
4633 
4634 	/* extract values from active_vlans and write back to VFTA */
4635 	for (i = VFTA_BLOCK_SIZE; i--;) {
4636 		vid = (vfta_offset + i) * 32;
4637 		word = vid / BITS_PER_LONG;
4638 		bits = vid % BITS_PER_LONG;
4639 
4640 		vfta[i] |= adapter->active_vlans[word] >> bits;
4641 
4642 		IXGBE_WRITE_REG(hw, IXGBE_VFTA(vfta_offset + i), vfta[i]);
4643 	}
4644 }
4645 
4646 static void ixgbe_vlan_promisc_disable(struct ixgbe_adapter *adapter)
4647 {
4648 	struct ixgbe_hw *hw = &adapter->hw;
4649 	u32 vlnctrl, i;
4650 
4651 	/* Set VLAN filtering to enabled */
4652 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4653 	vlnctrl |= IXGBE_VLNCTRL_VFE;
4654 	IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4655 
4656 	if (!(adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) ||
4657 	    hw->mac.type == ixgbe_mac_82598EB)
4658 		return;
4659 
4660 	/* We are not in VLAN promisc, nothing to do */
4661 	if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4662 		return;
4663 
4664 	/* Set flag so we don't redo unnecessary work */
4665 	adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4666 
4667 	for (i = 0; i < hw->mac.vft_size; i += VFTA_BLOCK_SIZE)
4668 		ixgbe_scrub_vfta(adapter, i);
4669 }
4670 
4671 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
4672 {
4673 	u16 vid = 1;
4674 
4675 	ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
4676 
4677 	for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
4678 		ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
4679 }
4680 
4681 /**
4682  * ixgbe_write_mc_addr_list - write multicast addresses to MTA
4683  * @netdev: network interface device structure
4684  *
4685  * Writes multicast address list to the MTA hash table.
4686  * Returns: -ENOMEM on failure
4687  *                0 on no addresses written
4688  *                X on writing X addresses to MTA
4689  **/
4690 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
4691 {
4692 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4693 	struct ixgbe_hw *hw = &adapter->hw;
4694 
4695 	if (!netif_running(netdev))
4696 		return 0;
4697 
4698 	if (hw->mac.ops.update_mc_addr_list)
4699 		hw->mac.ops.update_mc_addr_list(hw, netdev);
4700 	else
4701 		return -ENOMEM;
4702 
4703 #ifdef CONFIG_PCI_IOV
4704 	ixgbe_restore_vf_multicasts(adapter);
4705 #endif
4706 
4707 	return netdev_mc_count(netdev);
4708 }
4709 
4710 #ifdef CONFIG_PCI_IOV
4711 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4712 {
4713 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4714 	struct ixgbe_hw *hw = &adapter->hw;
4715 	int i;
4716 
4717 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4718 		mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4719 
4720 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4721 			hw->mac.ops.set_rar(hw, i,
4722 					    mac_table->addr,
4723 					    mac_table->pool,
4724 					    IXGBE_RAH_AV);
4725 		else
4726 			hw->mac.ops.clear_rar(hw, i);
4727 	}
4728 }
4729 
4730 #endif
4731 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4732 {
4733 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4734 	struct ixgbe_hw *hw = &adapter->hw;
4735 	int i;
4736 
4737 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4738 		if (!(mac_table->state & IXGBE_MAC_STATE_MODIFIED))
4739 			continue;
4740 
4741 		mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4742 
4743 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4744 			hw->mac.ops.set_rar(hw, i,
4745 					    mac_table->addr,
4746 					    mac_table->pool,
4747 					    IXGBE_RAH_AV);
4748 		else
4749 			hw->mac.ops.clear_rar(hw, i);
4750 	}
4751 }
4752 
4753 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4754 {
4755 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4756 	struct ixgbe_hw *hw = &adapter->hw;
4757 	int i;
4758 
4759 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4760 		mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4761 		mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4762 	}
4763 
4764 	ixgbe_sync_mac_table(adapter);
4765 }
4766 
4767 static int ixgbe_available_rars(struct ixgbe_adapter *adapter, u16 pool)
4768 {
4769 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4770 	struct ixgbe_hw *hw = &adapter->hw;
4771 	int i, count = 0;
4772 
4773 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4774 		/* do not count default RAR as available */
4775 		if (mac_table->state & IXGBE_MAC_STATE_DEFAULT)
4776 			continue;
4777 
4778 		/* only count unused and addresses that belong to us */
4779 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE) {
4780 			if (mac_table->pool != pool)
4781 				continue;
4782 		}
4783 
4784 		count++;
4785 	}
4786 
4787 	return count;
4788 }
4789 
4790 /* this function destroys the first RAR entry */
4791 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter)
4792 {
4793 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4794 	struct ixgbe_hw *hw = &adapter->hw;
4795 
4796 	memcpy(&mac_table->addr, hw->mac.addr, ETH_ALEN);
4797 	mac_table->pool = VMDQ_P(0);
4798 
4799 	mac_table->state = IXGBE_MAC_STATE_DEFAULT | IXGBE_MAC_STATE_IN_USE;
4800 
4801 	hw->mac.ops.set_rar(hw, 0, mac_table->addr, mac_table->pool,
4802 			    IXGBE_RAH_AV);
4803 }
4804 
4805 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter,
4806 			 const u8 *addr, u16 pool)
4807 {
4808 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4809 	struct ixgbe_hw *hw = &adapter->hw;
4810 	int i;
4811 
4812 	if (is_zero_ether_addr(addr))
4813 		return -EINVAL;
4814 
4815 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4816 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4817 			continue;
4818 
4819 		ether_addr_copy(mac_table->addr, addr);
4820 		mac_table->pool = pool;
4821 
4822 		mac_table->state |= IXGBE_MAC_STATE_MODIFIED |
4823 				    IXGBE_MAC_STATE_IN_USE;
4824 
4825 		ixgbe_sync_mac_table(adapter);
4826 
4827 		return i;
4828 	}
4829 
4830 	return -ENOMEM;
4831 }
4832 
4833 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter,
4834 			 const u8 *addr, u16 pool)
4835 {
4836 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4837 	struct ixgbe_hw *hw = &adapter->hw;
4838 	int i;
4839 
4840 	if (is_zero_ether_addr(addr))
4841 		return -EINVAL;
4842 
4843 	/* search table for addr, if found clear IN_USE flag and sync */
4844 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4845 		/* we can only delete an entry if it is in use */
4846 		if (!(mac_table->state & IXGBE_MAC_STATE_IN_USE))
4847 			continue;
4848 		/* we only care about entries that belong to the given pool */
4849 		if (mac_table->pool != pool)
4850 			continue;
4851 		/* we only care about a specific MAC address */
4852 		if (!ether_addr_equal(addr, mac_table->addr))
4853 			continue;
4854 
4855 		mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4856 		mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4857 
4858 		ixgbe_sync_mac_table(adapter);
4859 
4860 		return 0;
4861 	}
4862 
4863 	return -ENOMEM;
4864 }
4865 
4866 static int ixgbe_uc_sync(struct net_device *netdev, const unsigned char *addr)
4867 {
4868 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4869 	int ret;
4870 
4871 	ret = ixgbe_add_mac_filter(adapter, addr, VMDQ_P(0));
4872 
4873 	return min_t(int, ret, 0);
4874 }
4875 
4876 static int ixgbe_uc_unsync(struct net_device *netdev, const unsigned char *addr)
4877 {
4878 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4879 
4880 	ixgbe_del_mac_filter(adapter, addr, VMDQ_P(0));
4881 
4882 	return 0;
4883 }
4884 
4885 /**
4886  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4887  * @netdev: network interface device structure
4888  *
4889  * The set_rx_method entry point is called whenever the unicast/multicast
4890  * address list or the network interface flags are updated.  This routine is
4891  * responsible for configuring the hardware for proper unicast, multicast and
4892  * promiscuous mode.
4893  **/
4894 void ixgbe_set_rx_mode(struct net_device *netdev)
4895 {
4896 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4897 	struct ixgbe_hw *hw = &adapter->hw;
4898 	u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4899 	netdev_features_t features = netdev->features;
4900 	int count;
4901 
4902 	/* Check for Promiscuous and All Multicast modes */
4903 	fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4904 
4905 	/* set all bits that we expect to always be set */
4906 	fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4907 	fctrl |= IXGBE_FCTRL_BAM;
4908 	fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4909 	fctrl |= IXGBE_FCTRL_PMCF;
4910 
4911 	/* clear the bits we are changing the status of */
4912 	fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4913 	if (netdev->flags & IFF_PROMISC) {
4914 		hw->addr_ctrl.user_set_promisc = true;
4915 		fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4916 		vmolr |= IXGBE_VMOLR_MPE;
4917 		features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4918 	} else {
4919 		if (netdev->flags & IFF_ALLMULTI) {
4920 			fctrl |= IXGBE_FCTRL_MPE;
4921 			vmolr |= IXGBE_VMOLR_MPE;
4922 		}
4923 		hw->addr_ctrl.user_set_promisc = false;
4924 	}
4925 
4926 	/*
4927 	 * Write addresses to available RAR registers, if there is not
4928 	 * sufficient space to store all the addresses then enable
4929 	 * unicast promiscuous mode
4930 	 */
4931 	if (__dev_uc_sync(netdev, ixgbe_uc_sync, ixgbe_uc_unsync)) {
4932 		fctrl |= IXGBE_FCTRL_UPE;
4933 		vmolr |= IXGBE_VMOLR_ROPE;
4934 	}
4935 
4936 	/* Write addresses to the MTA, if the attempt fails
4937 	 * then we should just turn on promiscuous mode so
4938 	 * that we can at least receive multicast traffic
4939 	 */
4940 	count = ixgbe_write_mc_addr_list(netdev);
4941 	if (count < 0) {
4942 		fctrl |= IXGBE_FCTRL_MPE;
4943 		vmolr |= IXGBE_VMOLR_MPE;
4944 	} else if (count) {
4945 		vmolr |= IXGBE_VMOLR_ROMPE;
4946 	}
4947 
4948 	if (hw->mac.type != ixgbe_mac_82598EB) {
4949 		vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4950 			 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4951 			   IXGBE_VMOLR_ROPE);
4952 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4953 	}
4954 
4955 	/* This is useful for sniffing bad packets. */
4956 	if (features & NETIF_F_RXALL) {
4957 		/* UPE and MPE will be handled by normal PROMISC logic
4958 		 * in e1000e_set_rx_mode */
4959 		fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4960 			  IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4961 			  IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4962 
4963 		fctrl &= ~(IXGBE_FCTRL_DPF);
4964 		/* NOTE:  VLAN filtering is disabled by setting PROMISC */
4965 	}
4966 
4967 	IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4968 
4969 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
4970 		ixgbe_vlan_strip_enable(adapter);
4971 	else
4972 		ixgbe_vlan_strip_disable(adapter);
4973 
4974 	if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
4975 		ixgbe_vlan_promisc_disable(adapter);
4976 	else
4977 		ixgbe_vlan_promisc_enable(adapter);
4978 }
4979 
4980 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
4981 {
4982 	int q_idx;
4983 
4984 	for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4985 		napi_enable(&adapter->q_vector[q_idx]->napi);
4986 }
4987 
4988 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
4989 {
4990 	int q_idx;
4991 
4992 	for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4993 		napi_disable(&adapter->q_vector[q_idx]->napi);
4994 }
4995 
4996 static int ixgbe_udp_tunnel_sync(struct net_device *dev, unsigned int table)
4997 {
4998 	struct ixgbe_adapter *adapter = netdev_priv(dev);
4999 	struct ixgbe_hw *hw = &adapter->hw;
5000 	struct udp_tunnel_info ti;
5001 
5002 	udp_tunnel_nic_get_port(dev, table, 0, &ti);
5003 	if (ti.type == UDP_TUNNEL_TYPE_VXLAN)
5004 		adapter->vxlan_port = ti.port;
5005 	else
5006 		adapter->geneve_port = ti.port;
5007 
5008 	IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL,
5009 			ntohs(adapter->vxlan_port) |
5010 			ntohs(adapter->geneve_port) <<
5011 				IXGBE_VXLANCTRL_GENEVE_UDPPORT_SHIFT);
5012 	return 0;
5013 }
5014 
5015 static const struct udp_tunnel_nic_info ixgbe_udp_tunnels_x550 = {
5016 	.sync_table	= ixgbe_udp_tunnel_sync,
5017 	.flags		= UDP_TUNNEL_NIC_INFO_IPV4_ONLY,
5018 	.tables		= {
5019 		{ .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN,  },
5020 	},
5021 };
5022 
5023 static const struct udp_tunnel_nic_info ixgbe_udp_tunnels_x550em_a = {
5024 	.sync_table	= ixgbe_udp_tunnel_sync,
5025 	.flags		= UDP_TUNNEL_NIC_INFO_IPV4_ONLY,
5026 	.tables		= {
5027 		{ .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN,  },
5028 		{ .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_GENEVE, },
5029 	},
5030 };
5031 
5032 #ifdef CONFIG_IXGBE_DCB
5033 /**
5034  * ixgbe_configure_dcb - Configure DCB hardware
5035  * @adapter: ixgbe adapter struct
5036  *
5037  * This is called by the driver on open to configure the DCB hardware.
5038  * This is also called by the gennetlink interface when reconfiguring
5039  * the DCB state.
5040  */
5041 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
5042 {
5043 	struct ixgbe_hw *hw = &adapter->hw;
5044 	int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
5045 
5046 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
5047 		if (hw->mac.type == ixgbe_mac_82598EB)
5048 			netif_set_gso_max_size(adapter->netdev, 65536);
5049 		return;
5050 	}
5051 
5052 	if (hw->mac.type == ixgbe_mac_82598EB)
5053 		netif_set_gso_max_size(adapter->netdev, 32768);
5054 
5055 #ifdef IXGBE_FCOE
5056 	if (adapter->netdev->features & NETIF_F_FCOE_MTU)
5057 		max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
5058 #endif
5059 
5060 	/* reconfigure the hardware */
5061 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
5062 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5063 						DCB_TX_CONFIG);
5064 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5065 						DCB_RX_CONFIG);
5066 		ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
5067 	} else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
5068 		ixgbe_dcb_hw_ets(&adapter->hw,
5069 				 adapter->ixgbe_ieee_ets,
5070 				 max_frame);
5071 		ixgbe_dcb_hw_pfc_config(&adapter->hw,
5072 					adapter->ixgbe_ieee_pfc->pfc_en,
5073 					adapter->ixgbe_ieee_ets->prio_tc);
5074 	}
5075 
5076 	/* Enable RSS Hash per TC */
5077 	if (hw->mac.type != ixgbe_mac_82598EB) {
5078 		u32 msb = 0;
5079 		u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
5080 
5081 		while (rss_i) {
5082 			msb++;
5083 			rss_i >>= 1;
5084 		}
5085 
5086 		/* write msb to all 8 TCs in one write */
5087 		IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
5088 	}
5089 }
5090 #endif
5091 
5092 /* Additional bittime to account for IXGBE framing */
5093 #define IXGBE_ETH_FRAMING 20
5094 
5095 /**
5096  * ixgbe_hpbthresh - calculate high water mark for flow control
5097  *
5098  * @adapter: board private structure to calculate for
5099  * @pb: packet buffer to calculate
5100  */
5101 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
5102 {
5103 	struct ixgbe_hw *hw = &adapter->hw;
5104 	struct net_device *dev = adapter->netdev;
5105 	int link, tc, kb, marker;
5106 	u32 dv_id, rx_pba;
5107 
5108 	/* Calculate max LAN frame size */
5109 	tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
5110 
5111 #ifdef IXGBE_FCOE
5112 	/* FCoE traffic class uses FCOE jumbo frames */
5113 	if ((dev->features & NETIF_F_FCOE_MTU) &&
5114 	    (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5115 	    (pb == ixgbe_fcoe_get_tc(adapter)))
5116 		tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5117 #endif
5118 
5119 	/* Calculate delay value for device */
5120 	switch (hw->mac.type) {
5121 	case ixgbe_mac_X540:
5122 	case ixgbe_mac_X550:
5123 	case ixgbe_mac_X550EM_x:
5124 	case ixgbe_mac_x550em_a:
5125 		dv_id = IXGBE_DV_X540(link, tc);
5126 		break;
5127 	default:
5128 		dv_id = IXGBE_DV(link, tc);
5129 		break;
5130 	}
5131 
5132 	/* Loopback switch introduces additional latency */
5133 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5134 		dv_id += IXGBE_B2BT(tc);
5135 
5136 	/* Delay value is calculated in bit times convert to KB */
5137 	kb = IXGBE_BT2KB(dv_id);
5138 	rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
5139 
5140 	marker = rx_pba - kb;
5141 
5142 	/* It is possible that the packet buffer is not large enough
5143 	 * to provide required headroom. In this case throw an error
5144 	 * to user and a do the best we can.
5145 	 */
5146 	if (marker < 0) {
5147 		e_warn(drv, "Packet Buffer(%i) can not provide enough"
5148 			    "headroom to support flow control."
5149 			    "Decrease MTU or number of traffic classes\n", pb);
5150 		marker = tc + 1;
5151 	}
5152 
5153 	return marker;
5154 }
5155 
5156 /**
5157  * ixgbe_lpbthresh - calculate low water mark for for flow control
5158  *
5159  * @adapter: board private structure to calculate for
5160  * @pb: packet buffer to calculate
5161  */
5162 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
5163 {
5164 	struct ixgbe_hw *hw = &adapter->hw;
5165 	struct net_device *dev = adapter->netdev;
5166 	int tc;
5167 	u32 dv_id;
5168 
5169 	/* Calculate max LAN frame size */
5170 	tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5171 
5172 #ifdef IXGBE_FCOE
5173 	/* FCoE traffic class uses FCOE jumbo frames */
5174 	if ((dev->features & NETIF_F_FCOE_MTU) &&
5175 	    (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5176 	    (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
5177 		tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5178 #endif
5179 
5180 	/* Calculate delay value for device */
5181 	switch (hw->mac.type) {
5182 	case ixgbe_mac_X540:
5183 	case ixgbe_mac_X550:
5184 	case ixgbe_mac_X550EM_x:
5185 	case ixgbe_mac_x550em_a:
5186 		dv_id = IXGBE_LOW_DV_X540(tc);
5187 		break;
5188 	default:
5189 		dv_id = IXGBE_LOW_DV(tc);
5190 		break;
5191 	}
5192 
5193 	/* Delay value is calculated in bit times convert to KB */
5194 	return IXGBE_BT2KB(dv_id);
5195 }
5196 
5197 /*
5198  * ixgbe_pbthresh_setup - calculate and setup high low water marks
5199  */
5200 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
5201 {
5202 	struct ixgbe_hw *hw = &adapter->hw;
5203 	int num_tc = adapter->hw_tcs;
5204 	int i;
5205 
5206 	if (!num_tc)
5207 		num_tc = 1;
5208 
5209 	for (i = 0; i < num_tc; i++) {
5210 		hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
5211 		hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
5212 
5213 		/* Low water marks must not be larger than high water marks */
5214 		if (hw->fc.low_water[i] > hw->fc.high_water[i])
5215 			hw->fc.low_water[i] = 0;
5216 	}
5217 
5218 	for (; i < MAX_TRAFFIC_CLASS; i++)
5219 		hw->fc.high_water[i] = 0;
5220 }
5221 
5222 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
5223 {
5224 	struct ixgbe_hw *hw = &adapter->hw;
5225 	int hdrm;
5226 	u8 tc = adapter->hw_tcs;
5227 
5228 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5229 	    adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5230 		hdrm = 32 << adapter->fdir_pballoc;
5231 	else
5232 		hdrm = 0;
5233 
5234 	hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
5235 	ixgbe_pbthresh_setup(adapter);
5236 }
5237 
5238 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
5239 {
5240 	struct ixgbe_hw *hw = &adapter->hw;
5241 	struct hlist_node *node2;
5242 	struct ixgbe_fdir_filter *filter;
5243 	u8 queue;
5244 
5245 	spin_lock(&adapter->fdir_perfect_lock);
5246 
5247 	if (!hlist_empty(&adapter->fdir_filter_list))
5248 		ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
5249 
5250 	hlist_for_each_entry_safe(filter, node2,
5251 				  &adapter->fdir_filter_list, fdir_node) {
5252 		if (filter->action == IXGBE_FDIR_DROP_QUEUE) {
5253 			queue = IXGBE_FDIR_DROP_QUEUE;
5254 		} else {
5255 			u32 ring = ethtool_get_flow_spec_ring(filter->action);
5256 			u8 vf = ethtool_get_flow_spec_ring_vf(filter->action);
5257 
5258 			if (!vf && (ring >= adapter->num_rx_queues)) {
5259 				e_err(drv, "FDIR restore failed without VF, ring: %u\n",
5260 				      ring);
5261 				continue;
5262 			} else if (vf &&
5263 				   ((vf > adapter->num_vfs) ||
5264 				     ring >= adapter->num_rx_queues_per_pool)) {
5265 				e_err(drv, "FDIR restore failed with VF, vf: %hhu, ring: %u\n",
5266 				      vf, ring);
5267 				continue;
5268 			}
5269 
5270 			/* Map the ring onto the absolute queue index */
5271 			if (!vf)
5272 				queue = adapter->rx_ring[ring]->reg_idx;
5273 			else
5274 				queue = ((vf - 1) *
5275 					adapter->num_rx_queues_per_pool) + ring;
5276 		}
5277 
5278 		ixgbe_fdir_write_perfect_filter_82599(hw,
5279 				&filter->filter, filter->sw_idx, queue);
5280 	}
5281 
5282 	spin_unlock(&adapter->fdir_perfect_lock);
5283 }
5284 
5285 /**
5286  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
5287  * @rx_ring: ring to free buffers from
5288  **/
5289 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
5290 {
5291 	u16 i = rx_ring->next_to_clean;
5292 	struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
5293 
5294 	if (rx_ring->xsk_pool) {
5295 		ixgbe_xsk_clean_rx_ring(rx_ring);
5296 		goto skip_free;
5297 	}
5298 
5299 	/* Free all the Rx ring sk_buffs */
5300 	while (i != rx_ring->next_to_alloc) {
5301 		if (rx_buffer->skb) {
5302 			struct sk_buff *skb = rx_buffer->skb;
5303 			if (IXGBE_CB(skb)->page_released)
5304 				dma_unmap_page_attrs(rx_ring->dev,
5305 						     IXGBE_CB(skb)->dma,
5306 						     ixgbe_rx_pg_size(rx_ring),
5307 						     DMA_FROM_DEVICE,
5308 						     IXGBE_RX_DMA_ATTR);
5309 			dev_kfree_skb(skb);
5310 		}
5311 
5312 		/* Invalidate cache lines that may have been written to by
5313 		 * device so that we avoid corrupting memory.
5314 		 */
5315 		dma_sync_single_range_for_cpu(rx_ring->dev,
5316 					      rx_buffer->dma,
5317 					      rx_buffer->page_offset,
5318 					      ixgbe_rx_bufsz(rx_ring),
5319 					      DMA_FROM_DEVICE);
5320 
5321 		/* free resources associated with mapping */
5322 		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
5323 				     ixgbe_rx_pg_size(rx_ring),
5324 				     DMA_FROM_DEVICE,
5325 				     IXGBE_RX_DMA_ATTR);
5326 		__page_frag_cache_drain(rx_buffer->page,
5327 					rx_buffer->pagecnt_bias);
5328 
5329 		i++;
5330 		rx_buffer++;
5331 		if (i == rx_ring->count) {
5332 			i = 0;
5333 			rx_buffer = rx_ring->rx_buffer_info;
5334 		}
5335 	}
5336 
5337 skip_free:
5338 	rx_ring->next_to_alloc = 0;
5339 	rx_ring->next_to_clean = 0;
5340 	rx_ring->next_to_use = 0;
5341 }
5342 
5343 static int ixgbe_fwd_ring_up(struct ixgbe_adapter *adapter,
5344 			     struct ixgbe_fwd_adapter *accel)
5345 {
5346 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
5347 	int num_tc = netdev_get_num_tc(adapter->netdev);
5348 	struct net_device *vdev = accel->netdev;
5349 	int i, baseq, err;
5350 
5351 	baseq = accel->pool * adapter->num_rx_queues_per_pool;
5352 	netdev_dbg(vdev, "pool %i:%i queues %i:%i\n",
5353 		   accel->pool, adapter->num_rx_pools,
5354 		   baseq, baseq + adapter->num_rx_queues_per_pool);
5355 
5356 	accel->rx_base_queue = baseq;
5357 	accel->tx_base_queue = baseq;
5358 
5359 	/* record configuration for macvlan interface in vdev */
5360 	for (i = 0; i < num_tc; i++)
5361 		netdev_bind_sb_channel_queue(adapter->netdev, vdev,
5362 					     i, rss_i, baseq + (rss_i * i));
5363 
5364 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5365 		adapter->rx_ring[baseq + i]->netdev = vdev;
5366 
5367 	/* Guarantee all rings are updated before we update the
5368 	 * MAC address filter.
5369 	 */
5370 	wmb();
5371 
5372 	/* ixgbe_add_mac_filter will return an index if it succeeds, so we
5373 	 * need to only treat it as an error value if it is negative.
5374 	 */
5375 	err = ixgbe_add_mac_filter(adapter, vdev->dev_addr,
5376 				   VMDQ_P(accel->pool));
5377 	if (err >= 0)
5378 		return 0;
5379 
5380 	/* if we cannot add the MAC rule then disable the offload */
5381 	macvlan_release_l2fw_offload(vdev);
5382 
5383 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5384 		adapter->rx_ring[baseq + i]->netdev = NULL;
5385 
5386 	netdev_err(vdev, "L2FW offload disabled due to L2 filter error\n");
5387 
5388 	/* unbind the queues and drop the subordinate channel config */
5389 	netdev_unbind_sb_channel(adapter->netdev, vdev);
5390 	netdev_set_sb_channel(vdev, 0);
5391 
5392 	clear_bit(accel->pool, adapter->fwd_bitmask);
5393 	kfree(accel);
5394 
5395 	return err;
5396 }
5397 
5398 static int ixgbe_macvlan_up(struct net_device *vdev,
5399 			    struct netdev_nested_priv *priv)
5400 {
5401 	struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)priv->data;
5402 	struct ixgbe_fwd_adapter *accel;
5403 
5404 	if (!netif_is_macvlan(vdev))
5405 		return 0;
5406 
5407 	accel = macvlan_accel_priv(vdev);
5408 	if (!accel)
5409 		return 0;
5410 
5411 	ixgbe_fwd_ring_up(adapter, accel);
5412 
5413 	return 0;
5414 }
5415 
5416 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
5417 {
5418 	struct netdev_nested_priv priv = {
5419 		.data = (void *)adapter,
5420 	};
5421 
5422 	netdev_walk_all_upper_dev_rcu(adapter->netdev,
5423 				      ixgbe_macvlan_up, &priv);
5424 }
5425 
5426 static void ixgbe_configure(struct ixgbe_adapter *adapter)
5427 {
5428 	struct ixgbe_hw *hw = &adapter->hw;
5429 
5430 	ixgbe_configure_pb(adapter);
5431 #ifdef CONFIG_IXGBE_DCB
5432 	ixgbe_configure_dcb(adapter);
5433 #endif
5434 	/*
5435 	 * We must restore virtualization before VLANs or else
5436 	 * the VLVF registers will not be populated
5437 	 */
5438 	ixgbe_configure_virtualization(adapter);
5439 
5440 	ixgbe_set_rx_mode(adapter->netdev);
5441 	ixgbe_restore_vlan(adapter);
5442 	ixgbe_ipsec_restore(adapter);
5443 
5444 	switch (hw->mac.type) {
5445 	case ixgbe_mac_82599EB:
5446 	case ixgbe_mac_X540:
5447 		hw->mac.ops.disable_rx_buff(hw);
5448 		break;
5449 	default:
5450 		break;
5451 	}
5452 
5453 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5454 		ixgbe_init_fdir_signature_82599(&adapter->hw,
5455 						adapter->fdir_pballoc);
5456 	} else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
5457 		ixgbe_init_fdir_perfect_82599(&adapter->hw,
5458 					      adapter->fdir_pballoc);
5459 		ixgbe_fdir_filter_restore(adapter);
5460 	}
5461 
5462 	switch (hw->mac.type) {
5463 	case ixgbe_mac_82599EB:
5464 	case ixgbe_mac_X540:
5465 		hw->mac.ops.enable_rx_buff(hw);
5466 		break;
5467 	default:
5468 		break;
5469 	}
5470 
5471 #ifdef CONFIG_IXGBE_DCA
5472 	/* configure DCA */
5473 	if (adapter->flags & IXGBE_FLAG_DCA_CAPABLE)
5474 		ixgbe_setup_dca(adapter);
5475 #endif /* CONFIG_IXGBE_DCA */
5476 
5477 #ifdef IXGBE_FCOE
5478 	/* configure FCoE L2 filters, redirection table, and Rx control */
5479 	ixgbe_configure_fcoe(adapter);
5480 
5481 #endif /* IXGBE_FCOE */
5482 	ixgbe_configure_tx(adapter);
5483 	ixgbe_configure_rx(adapter);
5484 	ixgbe_configure_dfwd(adapter);
5485 }
5486 
5487 /**
5488  * ixgbe_sfp_link_config - set up SFP+ link
5489  * @adapter: pointer to private adapter struct
5490  **/
5491 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
5492 {
5493 	/*
5494 	 * We are assuming the worst case scenario here, and that
5495 	 * is that an SFP was inserted/removed after the reset
5496 	 * but before SFP detection was enabled.  As such the best
5497 	 * solution is to just start searching as soon as we start
5498 	 */
5499 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5500 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5501 
5502 	adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5503 	adapter->sfp_poll_time = 0;
5504 }
5505 
5506 /**
5507  * ixgbe_non_sfp_link_config - set up non-SFP+ link
5508  * @hw: pointer to private hardware struct
5509  *
5510  * Returns 0 on success, negative on failure
5511  **/
5512 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
5513 {
5514 	u32 speed;
5515 	bool autoneg, link_up = false;
5516 	int ret = IXGBE_ERR_LINK_SETUP;
5517 
5518 	if (hw->mac.ops.check_link)
5519 		ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
5520 
5521 	if (ret)
5522 		return ret;
5523 
5524 	speed = hw->phy.autoneg_advertised;
5525 	if (!speed && hw->mac.ops.get_link_capabilities) {
5526 		ret = hw->mac.ops.get_link_capabilities(hw, &speed,
5527 							&autoneg);
5528 		speed &= ~(IXGBE_LINK_SPEED_5GB_FULL |
5529 			   IXGBE_LINK_SPEED_2_5GB_FULL);
5530 	}
5531 
5532 	if (ret)
5533 		return ret;
5534 
5535 	if (hw->mac.ops.setup_link)
5536 		ret = hw->mac.ops.setup_link(hw, speed, link_up);
5537 
5538 	return ret;
5539 }
5540 
5541 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
5542 {
5543 	struct ixgbe_hw *hw = &adapter->hw;
5544 	u32 gpie = 0;
5545 
5546 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5547 		gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
5548 		       IXGBE_GPIE_OCD;
5549 		gpie |= IXGBE_GPIE_EIAME;
5550 		/*
5551 		 * use EIAM to auto-mask when MSI-X interrupt is asserted
5552 		 * this saves a register write for every interrupt
5553 		 */
5554 		switch (hw->mac.type) {
5555 		case ixgbe_mac_82598EB:
5556 			IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5557 			break;
5558 		case ixgbe_mac_82599EB:
5559 		case ixgbe_mac_X540:
5560 		case ixgbe_mac_X550:
5561 		case ixgbe_mac_X550EM_x:
5562 		case ixgbe_mac_x550em_a:
5563 		default:
5564 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
5565 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
5566 			break;
5567 		}
5568 	} else {
5569 		/* legacy interrupts, use EIAM to auto-mask when reading EICR,
5570 		 * specifically only auto mask tx and rx interrupts */
5571 		IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5572 	}
5573 
5574 	/* XXX: to interrupt immediately for EICS writes, enable this */
5575 	/* gpie |= IXGBE_GPIE_EIMEN; */
5576 
5577 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
5578 		gpie &= ~IXGBE_GPIE_VTMODE_MASK;
5579 
5580 		switch (adapter->ring_feature[RING_F_VMDQ].mask) {
5581 		case IXGBE_82599_VMDQ_8Q_MASK:
5582 			gpie |= IXGBE_GPIE_VTMODE_16;
5583 			break;
5584 		case IXGBE_82599_VMDQ_4Q_MASK:
5585 			gpie |= IXGBE_GPIE_VTMODE_32;
5586 			break;
5587 		default:
5588 			gpie |= IXGBE_GPIE_VTMODE_64;
5589 			break;
5590 		}
5591 	}
5592 
5593 	/* Enable Thermal over heat sensor interrupt */
5594 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
5595 		switch (adapter->hw.mac.type) {
5596 		case ixgbe_mac_82599EB:
5597 			gpie |= IXGBE_SDP0_GPIEN_8259X;
5598 			break;
5599 		default:
5600 			break;
5601 		}
5602 	}
5603 
5604 	/* Enable fan failure interrupt */
5605 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
5606 		gpie |= IXGBE_SDP1_GPIEN(hw);
5607 
5608 	switch (hw->mac.type) {
5609 	case ixgbe_mac_82599EB:
5610 		gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X;
5611 		break;
5612 	case ixgbe_mac_X550EM_x:
5613 	case ixgbe_mac_x550em_a:
5614 		gpie |= IXGBE_SDP0_GPIEN_X540;
5615 		break;
5616 	default:
5617 		break;
5618 	}
5619 
5620 	IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
5621 }
5622 
5623 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
5624 {
5625 	struct ixgbe_hw *hw = &adapter->hw;
5626 	int err;
5627 	u32 ctrl_ext;
5628 
5629 	ixgbe_get_hw_control(adapter);
5630 	ixgbe_setup_gpie(adapter);
5631 
5632 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5633 		ixgbe_configure_msix(adapter);
5634 	else
5635 		ixgbe_configure_msi_and_legacy(adapter);
5636 
5637 	/* enable the optics for 82599 SFP+ fiber */
5638 	if (hw->mac.ops.enable_tx_laser)
5639 		hw->mac.ops.enable_tx_laser(hw);
5640 
5641 	if (hw->phy.ops.set_phy_power)
5642 		hw->phy.ops.set_phy_power(hw, true);
5643 
5644 	smp_mb__before_atomic();
5645 	clear_bit(__IXGBE_DOWN, &adapter->state);
5646 	ixgbe_napi_enable_all(adapter);
5647 
5648 	if (ixgbe_is_sfp(hw)) {
5649 		ixgbe_sfp_link_config(adapter);
5650 	} else {
5651 		err = ixgbe_non_sfp_link_config(hw);
5652 		if (err)
5653 			e_err(probe, "link_config FAILED %d\n", err);
5654 	}
5655 
5656 	/* clear any pending interrupts, may auto mask */
5657 	IXGBE_READ_REG(hw, IXGBE_EICR);
5658 	ixgbe_irq_enable(adapter, true, true);
5659 
5660 	/*
5661 	 * If this adapter has a fan, check to see if we had a failure
5662 	 * before we enabled the interrupt.
5663 	 */
5664 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5665 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5666 		if (esdp & IXGBE_ESDP_SDP1)
5667 			e_crit(drv, "Fan has stopped, replace the adapter\n");
5668 	}
5669 
5670 	/* bring the link up in the watchdog, this could race with our first
5671 	 * link up interrupt but shouldn't be a problem */
5672 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5673 	adapter->link_check_timeout = jiffies;
5674 	mod_timer(&adapter->service_timer, jiffies);
5675 
5676 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
5677 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5678 	ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
5679 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5680 }
5681 
5682 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
5683 {
5684 	/* put off any impending NetWatchDogTimeout */
5685 	netif_trans_update(adapter->netdev);
5686 
5687 	while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
5688 		usleep_range(1000, 2000);
5689 	if (adapter->hw.phy.type == ixgbe_phy_fw)
5690 		ixgbe_watchdog_link_is_down(adapter);
5691 	ixgbe_down(adapter);
5692 	/*
5693 	 * If SR-IOV enabled then wait a bit before bringing the adapter
5694 	 * back up to give the VFs time to respond to the reset.  The
5695 	 * two second wait is based upon the watchdog timer cycle in
5696 	 * the VF driver.
5697 	 */
5698 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5699 		msleep(2000);
5700 	ixgbe_up(adapter);
5701 	clear_bit(__IXGBE_RESETTING, &adapter->state);
5702 }
5703 
5704 void ixgbe_up(struct ixgbe_adapter *adapter)
5705 {
5706 	/* hardware has been reset, we need to reload some things */
5707 	ixgbe_configure(adapter);
5708 
5709 	ixgbe_up_complete(adapter);
5710 }
5711 
5712 static unsigned long ixgbe_get_completion_timeout(struct ixgbe_adapter *adapter)
5713 {
5714 	u16 devctl2;
5715 
5716 	pcie_capability_read_word(adapter->pdev, PCI_EXP_DEVCTL2, &devctl2);
5717 
5718 	switch (devctl2 & IXGBE_PCIDEVCTRL2_TIMEO_MASK) {
5719 	case IXGBE_PCIDEVCTRL2_17_34s:
5720 	case IXGBE_PCIDEVCTRL2_4_8s:
5721 		/* For now we cap the upper limit on delay to 2 seconds
5722 		 * as we end up going up to 34 seconds of delay in worst
5723 		 * case timeout value.
5724 		 */
5725 	case IXGBE_PCIDEVCTRL2_1_2s:
5726 		return 2000000ul;	/* 2.0 s */
5727 	case IXGBE_PCIDEVCTRL2_260_520ms:
5728 		return 520000ul;	/* 520 ms */
5729 	case IXGBE_PCIDEVCTRL2_65_130ms:
5730 		return 130000ul;	/* 130 ms */
5731 	case IXGBE_PCIDEVCTRL2_16_32ms:
5732 		return 32000ul;		/* 32 ms */
5733 	case IXGBE_PCIDEVCTRL2_1_2ms:
5734 		return 2000ul;		/* 2 ms */
5735 	case IXGBE_PCIDEVCTRL2_50_100us:
5736 		return 100ul;		/* 100 us */
5737 	case IXGBE_PCIDEVCTRL2_16_32ms_def:
5738 		return 32000ul;		/* 32 ms */
5739 	default:
5740 		break;
5741 	}
5742 
5743 	/* We shouldn't need to hit this path, but just in case default as
5744 	 * though completion timeout is not supported and support 32ms.
5745 	 */
5746 	return 32000ul;
5747 }
5748 
5749 void ixgbe_disable_rx(struct ixgbe_adapter *adapter)
5750 {
5751 	unsigned long wait_delay, delay_interval;
5752 	struct ixgbe_hw *hw = &adapter->hw;
5753 	int i, wait_loop;
5754 	u32 rxdctl;
5755 
5756 	/* disable receives */
5757 	hw->mac.ops.disable_rx(hw);
5758 
5759 	if (ixgbe_removed(hw->hw_addr))
5760 		return;
5761 
5762 	/* disable all enabled Rx queues */
5763 	for (i = 0; i < adapter->num_rx_queues; i++) {
5764 		struct ixgbe_ring *ring = adapter->rx_ring[i];
5765 		u8 reg_idx = ring->reg_idx;
5766 
5767 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5768 		rxdctl &= ~IXGBE_RXDCTL_ENABLE;
5769 		rxdctl |= IXGBE_RXDCTL_SWFLSH;
5770 
5771 		/* write value back with RXDCTL.ENABLE bit cleared */
5772 		IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
5773 	}
5774 
5775 	/* RXDCTL.EN may not change on 82598 if link is down, so skip it */
5776 	if (hw->mac.type == ixgbe_mac_82598EB &&
5777 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5778 		return;
5779 
5780 	/* Determine our minimum delay interval. We will increase this value
5781 	 * with each subsequent test. This way if the device returns quickly
5782 	 * we should spend as little time as possible waiting, however as
5783 	 * the time increases we will wait for larger periods of time.
5784 	 *
5785 	 * The trick here is that we increase the interval using the
5786 	 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5787 	 * of that wait is that it totals up to 100x whatever interval we
5788 	 * choose. Since our minimum wait is 100us we can just divide the
5789 	 * total timeout by 100 to get our minimum delay interval.
5790 	 */
5791 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5792 
5793 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
5794 	wait_delay = delay_interval;
5795 
5796 	while (wait_loop--) {
5797 		usleep_range(wait_delay, wait_delay + 10);
5798 		wait_delay += delay_interval * 2;
5799 		rxdctl = 0;
5800 
5801 		/* OR together the reading of all the active RXDCTL registers,
5802 		 * and then test the result. We need the disable to complete
5803 		 * before we start freeing the memory and invalidating the
5804 		 * DMA mappings.
5805 		 */
5806 		for (i = 0; i < adapter->num_rx_queues; i++) {
5807 			struct ixgbe_ring *ring = adapter->rx_ring[i];
5808 			u8 reg_idx = ring->reg_idx;
5809 
5810 			rxdctl |= IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5811 		}
5812 
5813 		if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
5814 			return;
5815 	}
5816 
5817 	e_err(drv,
5818 	      "RXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5819 }
5820 
5821 void ixgbe_disable_tx(struct ixgbe_adapter *adapter)
5822 {
5823 	unsigned long wait_delay, delay_interval;
5824 	struct ixgbe_hw *hw = &adapter->hw;
5825 	int i, wait_loop;
5826 	u32 txdctl;
5827 
5828 	if (ixgbe_removed(hw->hw_addr))
5829 		return;
5830 
5831 	/* disable all enabled Tx queues */
5832 	for (i = 0; i < adapter->num_tx_queues; i++) {
5833 		struct ixgbe_ring *ring = adapter->tx_ring[i];
5834 		u8 reg_idx = ring->reg_idx;
5835 
5836 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5837 	}
5838 
5839 	/* disable all enabled XDP Tx queues */
5840 	for (i = 0; i < adapter->num_xdp_queues; i++) {
5841 		struct ixgbe_ring *ring = adapter->xdp_ring[i];
5842 		u8 reg_idx = ring->reg_idx;
5843 
5844 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5845 	}
5846 
5847 	/* If the link is not up there shouldn't be much in the way of
5848 	 * pending transactions. Those that are left will be flushed out
5849 	 * when the reset logic goes through the flush sequence to clean out
5850 	 * the pending Tx transactions.
5851 	 */
5852 	if (!(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5853 		goto dma_engine_disable;
5854 
5855 	/* Determine our minimum delay interval. We will increase this value
5856 	 * with each subsequent test. This way if the device returns quickly
5857 	 * we should spend as little time as possible waiting, however as
5858 	 * the time increases we will wait for larger periods of time.
5859 	 *
5860 	 * The trick here is that we increase the interval using the
5861 	 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5862 	 * of that wait is that it totals up to 100x whatever interval we
5863 	 * choose. Since our minimum wait is 100us we can just divide the
5864 	 * total timeout by 100 to get our minimum delay interval.
5865 	 */
5866 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5867 
5868 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
5869 	wait_delay = delay_interval;
5870 
5871 	while (wait_loop--) {
5872 		usleep_range(wait_delay, wait_delay + 10);
5873 		wait_delay += delay_interval * 2;
5874 		txdctl = 0;
5875 
5876 		/* OR together the reading of all the active TXDCTL registers,
5877 		 * and then test the result. We need the disable to complete
5878 		 * before we start freeing the memory and invalidating the
5879 		 * DMA mappings.
5880 		 */
5881 		for (i = 0; i < adapter->num_tx_queues; i++) {
5882 			struct ixgbe_ring *ring = adapter->tx_ring[i];
5883 			u8 reg_idx = ring->reg_idx;
5884 
5885 			txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5886 		}
5887 		for (i = 0; i < adapter->num_xdp_queues; i++) {
5888 			struct ixgbe_ring *ring = adapter->xdp_ring[i];
5889 			u8 reg_idx = ring->reg_idx;
5890 
5891 			txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5892 		}
5893 
5894 		if (!(txdctl & IXGBE_TXDCTL_ENABLE))
5895 			goto dma_engine_disable;
5896 	}
5897 
5898 	e_err(drv,
5899 	      "TXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5900 
5901 dma_engine_disable:
5902 	/* Disable the Tx DMA engine on 82599 and later MAC */
5903 	switch (hw->mac.type) {
5904 	case ixgbe_mac_82599EB:
5905 	case ixgbe_mac_X540:
5906 	case ixgbe_mac_X550:
5907 	case ixgbe_mac_X550EM_x:
5908 	case ixgbe_mac_x550em_a:
5909 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5910 				(IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5911 				 ~IXGBE_DMATXCTL_TE));
5912 		fallthrough;
5913 	default:
5914 		break;
5915 	}
5916 }
5917 
5918 void ixgbe_reset(struct ixgbe_adapter *adapter)
5919 {
5920 	struct ixgbe_hw *hw = &adapter->hw;
5921 	struct net_device *netdev = adapter->netdev;
5922 	int err;
5923 
5924 	if (ixgbe_removed(hw->hw_addr))
5925 		return;
5926 	/* lock SFP init bit to prevent race conditions with the watchdog */
5927 	while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5928 		usleep_range(1000, 2000);
5929 
5930 	/* clear all SFP and link config related flags while holding SFP_INIT */
5931 	adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
5932 			     IXGBE_FLAG2_SFP_NEEDS_RESET);
5933 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5934 
5935 	err = hw->mac.ops.init_hw(hw);
5936 	switch (err) {
5937 	case 0:
5938 	case IXGBE_ERR_SFP_NOT_PRESENT:
5939 	case IXGBE_ERR_SFP_NOT_SUPPORTED:
5940 		break;
5941 	case IXGBE_ERR_MASTER_REQUESTS_PENDING:
5942 		e_dev_err("master disable timed out\n");
5943 		break;
5944 	case IXGBE_ERR_EEPROM_VERSION:
5945 		/* We are running on a pre-production device, log a warning */
5946 		e_dev_warn("This device is a pre-production adapter/LOM. "
5947 			   "Please be aware there may be issues associated with "
5948 			   "your hardware.  If you are experiencing problems "
5949 			   "please contact your Intel or hardware "
5950 			   "representative who provided you with this "
5951 			   "hardware.\n");
5952 		break;
5953 	default:
5954 		e_dev_err("Hardware Error: %d\n", err);
5955 	}
5956 
5957 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5958 
5959 	/* flush entries out of MAC table */
5960 	ixgbe_flush_sw_mac_table(adapter);
5961 	__dev_uc_unsync(netdev, NULL);
5962 
5963 	/* do not flush user set addresses */
5964 	ixgbe_mac_set_default_filter(adapter);
5965 
5966 	/* update SAN MAC vmdq pool selection */
5967 	if (hw->mac.san_mac_rar_index)
5968 		hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
5969 
5970 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5971 		ixgbe_ptp_reset(adapter);
5972 
5973 	if (hw->phy.ops.set_phy_power) {
5974 		if (!netif_running(adapter->netdev) && !adapter->wol)
5975 			hw->phy.ops.set_phy_power(hw, false);
5976 		else
5977 			hw->phy.ops.set_phy_power(hw, true);
5978 	}
5979 }
5980 
5981 /**
5982  * ixgbe_clean_tx_ring - Free Tx Buffers
5983  * @tx_ring: ring to be cleaned
5984  **/
5985 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
5986 {
5987 	u16 i = tx_ring->next_to_clean;
5988 	struct ixgbe_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
5989 
5990 	if (tx_ring->xsk_pool) {
5991 		ixgbe_xsk_clean_tx_ring(tx_ring);
5992 		goto out;
5993 	}
5994 
5995 	while (i != tx_ring->next_to_use) {
5996 		union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
5997 
5998 		/* Free all the Tx ring sk_buffs */
5999 		if (ring_is_xdp(tx_ring))
6000 			xdp_return_frame(tx_buffer->xdpf);
6001 		else
6002 			dev_kfree_skb_any(tx_buffer->skb);
6003 
6004 		/* unmap skb header data */
6005 		dma_unmap_single(tx_ring->dev,
6006 				 dma_unmap_addr(tx_buffer, dma),
6007 				 dma_unmap_len(tx_buffer, len),
6008 				 DMA_TO_DEVICE);
6009 
6010 		/* check for eop_desc to determine the end of the packet */
6011 		eop_desc = tx_buffer->next_to_watch;
6012 		tx_desc = IXGBE_TX_DESC(tx_ring, i);
6013 
6014 		/* unmap remaining buffers */
6015 		while (tx_desc != eop_desc) {
6016 			tx_buffer++;
6017 			tx_desc++;
6018 			i++;
6019 			if (unlikely(i == tx_ring->count)) {
6020 				i = 0;
6021 				tx_buffer = tx_ring->tx_buffer_info;
6022 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6023 			}
6024 
6025 			/* unmap any remaining paged data */
6026 			if (dma_unmap_len(tx_buffer, len))
6027 				dma_unmap_page(tx_ring->dev,
6028 					       dma_unmap_addr(tx_buffer, dma),
6029 					       dma_unmap_len(tx_buffer, len),
6030 					       DMA_TO_DEVICE);
6031 		}
6032 
6033 		/* move us one more past the eop_desc for start of next pkt */
6034 		tx_buffer++;
6035 		i++;
6036 		if (unlikely(i == tx_ring->count)) {
6037 			i = 0;
6038 			tx_buffer = tx_ring->tx_buffer_info;
6039 		}
6040 	}
6041 
6042 	/* reset BQL for queue */
6043 	if (!ring_is_xdp(tx_ring))
6044 		netdev_tx_reset_queue(txring_txq(tx_ring));
6045 
6046 out:
6047 	/* reset next_to_use and next_to_clean */
6048 	tx_ring->next_to_use = 0;
6049 	tx_ring->next_to_clean = 0;
6050 }
6051 
6052 /**
6053  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
6054  * @adapter: board private structure
6055  **/
6056 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
6057 {
6058 	int i;
6059 
6060 	for (i = 0; i < adapter->num_rx_queues; i++)
6061 		ixgbe_clean_rx_ring(adapter->rx_ring[i]);
6062 }
6063 
6064 /**
6065  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
6066  * @adapter: board private structure
6067  **/
6068 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
6069 {
6070 	int i;
6071 
6072 	for (i = 0; i < adapter->num_tx_queues; i++)
6073 		ixgbe_clean_tx_ring(adapter->tx_ring[i]);
6074 	for (i = 0; i < adapter->num_xdp_queues; i++)
6075 		ixgbe_clean_tx_ring(adapter->xdp_ring[i]);
6076 }
6077 
6078 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
6079 {
6080 	struct hlist_node *node2;
6081 	struct ixgbe_fdir_filter *filter;
6082 
6083 	spin_lock(&adapter->fdir_perfect_lock);
6084 
6085 	hlist_for_each_entry_safe(filter, node2,
6086 				  &adapter->fdir_filter_list, fdir_node) {
6087 		hlist_del(&filter->fdir_node);
6088 		kfree(filter);
6089 	}
6090 	adapter->fdir_filter_count = 0;
6091 
6092 	spin_unlock(&adapter->fdir_perfect_lock);
6093 }
6094 
6095 void ixgbe_down(struct ixgbe_adapter *adapter)
6096 {
6097 	struct net_device *netdev = adapter->netdev;
6098 	struct ixgbe_hw *hw = &adapter->hw;
6099 	int i;
6100 
6101 	/* signal that we are down to the interrupt handler */
6102 	if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
6103 		return; /* do nothing if already down */
6104 
6105 	/* Shut off incoming Tx traffic */
6106 	netif_tx_stop_all_queues(netdev);
6107 
6108 	/* call carrier off first to avoid false dev_watchdog timeouts */
6109 	netif_carrier_off(netdev);
6110 	netif_tx_disable(netdev);
6111 
6112 	/* Disable Rx */
6113 	ixgbe_disable_rx(adapter);
6114 
6115 	/* synchronize_rcu() needed for pending XDP buffers to drain */
6116 	if (adapter->xdp_ring[0])
6117 		synchronize_rcu();
6118 
6119 	ixgbe_irq_disable(adapter);
6120 
6121 	ixgbe_napi_disable_all(adapter);
6122 
6123 	clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
6124 	adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6125 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6126 
6127 	del_timer_sync(&adapter->service_timer);
6128 
6129 	if (adapter->num_vfs) {
6130 		/* Clear EITR Select mapping */
6131 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
6132 
6133 		/* Mark all the VFs as inactive */
6134 		for (i = 0 ; i < adapter->num_vfs; i++)
6135 			adapter->vfinfo[i].clear_to_send = false;
6136 
6137 		/* ping all the active vfs to let them know we are going down */
6138 		ixgbe_ping_all_vfs(adapter);
6139 
6140 		/* Disable all VFTE/VFRE TX/RX */
6141 		ixgbe_disable_tx_rx(adapter);
6142 	}
6143 
6144 	/* disable transmits in the hardware now that interrupts are off */
6145 	ixgbe_disable_tx(adapter);
6146 
6147 	if (!pci_channel_offline(adapter->pdev))
6148 		ixgbe_reset(adapter);
6149 
6150 	/* power down the optics for 82599 SFP+ fiber */
6151 	if (hw->mac.ops.disable_tx_laser)
6152 		hw->mac.ops.disable_tx_laser(hw);
6153 
6154 	ixgbe_clean_all_tx_rings(adapter);
6155 	ixgbe_clean_all_rx_rings(adapter);
6156 }
6157 
6158 /**
6159  * ixgbe_eee_capable - helper function to determine EEE support on X550
6160  * @adapter: board private structure
6161  */
6162 static void ixgbe_set_eee_capable(struct ixgbe_adapter *adapter)
6163 {
6164 	struct ixgbe_hw *hw = &adapter->hw;
6165 
6166 	switch (hw->device_id) {
6167 	case IXGBE_DEV_ID_X550EM_A_1G_T:
6168 	case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6169 		if (!hw->phy.eee_speeds_supported)
6170 			break;
6171 		adapter->flags2 |= IXGBE_FLAG2_EEE_CAPABLE;
6172 		if (!hw->phy.eee_speeds_advertised)
6173 			break;
6174 		adapter->flags2 |= IXGBE_FLAG2_EEE_ENABLED;
6175 		break;
6176 	default:
6177 		adapter->flags2 &= ~IXGBE_FLAG2_EEE_CAPABLE;
6178 		adapter->flags2 &= ~IXGBE_FLAG2_EEE_ENABLED;
6179 		break;
6180 	}
6181 }
6182 
6183 /**
6184  * ixgbe_tx_timeout - Respond to a Tx Hang
6185  * @netdev: network interface device structure
6186  * @txqueue: queue number that timed out
6187  **/
6188 static void ixgbe_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
6189 {
6190 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6191 
6192 	/* Do the reset outside of interrupt context */
6193 	ixgbe_tx_timeout_reset(adapter);
6194 }
6195 
6196 #ifdef CONFIG_IXGBE_DCB
6197 static void ixgbe_init_dcb(struct ixgbe_adapter *adapter)
6198 {
6199 	struct ixgbe_hw *hw = &adapter->hw;
6200 	struct tc_configuration *tc;
6201 	int j;
6202 
6203 	switch (hw->mac.type) {
6204 	case ixgbe_mac_82598EB:
6205 	case ixgbe_mac_82599EB:
6206 		adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
6207 		adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
6208 		break;
6209 	case ixgbe_mac_X540:
6210 	case ixgbe_mac_X550:
6211 		adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
6212 		adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
6213 		break;
6214 	case ixgbe_mac_X550EM_x:
6215 	case ixgbe_mac_x550em_a:
6216 	default:
6217 		adapter->dcb_cfg.num_tcs.pg_tcs = DEF_TRAFFIC_CLASS;
6218 		adapter->dcb_cfg.num_tcs.pfc_tcs = DEF_TRAFFIC_CLASS;
6219 		break;
6220 	}
6221 
6222 	/* Configure DCB traffic classes */
6223 	for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
6224 		tc = &adapter->dcb_cfg.tc_config[j];
6225 		tc->path[DCB_TX_CONFIG].bwg_id = 0;
6226 		tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
6227 		tc->path[DCB_RX_CONFIG].bwg_id = 0;
6228 		tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
6229 		tc->dcb_pfc = pfc_disabled;
6230 	}
6231 
6232 	/* Initialize default user to priority mapping, UPx->TC0 */
6233 	tc = &adapter->dcb_cfg.tc_config[0];
6234 	tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
6235 	tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
6236 
6237 	adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
6238 	adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
6239 	adapter->dcb_cfg.pfc_mode_enable = false;
6240 	adapter->dcb_set_bitmap = 0x00;
6241 	if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
6242 		adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
6243 	memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
6244 	       sizeof(adapter->temp_dcb_cfg));
6245 }
6246 #endif
6247 
6248 /**
6249  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
6250  * @adapter: board private structure to initialize
6251  * @ii: pointer to ixgbe_info for device
6252  *
6253  * ixgbe_sw_init initializes the Adapter private data structure.
6254  * Fields are initialized based on PCI device information and
6255  * OS network device settings (MTU size).
6256  **/
6257 static int ixgbe_sw_init(struct ixgbe_adapter *adapter,
6258 			 const struct ixgbe_info *ii)
6259 {
6260 	struct ixgbe_hw *hw = &adapter->hw;
6261 	struct pci_dev *pdev = adapter->pdev;
6262 	unsigned int rss, fdir;
6263 	u32 fwsm;
6264 	int i;
6265 
6266 	/* PCI config space info */
6267 
6268 	hw->vendor_id = pdev->vendor;
6269 	hw->device_id = pdev->device;
6270 	hw->revision_id = pdev->revision;
6271 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
6272 	hw->subsystem_device_id = pdev->subsystem_device;
6273 
6274 	/* get_invariants needs the device IDs */
6275 	ii->get_invariants(hw);
6276 
6277 	/* Set common capability flags and settings */
6278 	rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
6279 	adapter->ring_feature[RING_F_RSS].limit = rss;
6280 	adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
6281 	adapter->max_q_vectors = MAX_Q_VECTORS_82599;
6282 	adapter->atr_sample_rate = 20;
6283 	fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
6284 	adapter->ring_feature[RING_F_FDIR].limit = fdir;
6285 	adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
6286 	adapter->ring_feature[RING_F_VMDQ].limit = 1;
6287 #ifdef CONFIG_IXGBE_DCA
6288 	adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
6289 #endif
6290 #ifdef CONFIG_IXGBE_DCB
6291 	adapter->flags |= IXGBE_FLAG_DCB_CAPABLE;
6292 	adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6293 #endif
6294 #ifdef IXGBE_FCOE
6295 	adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
6296 	adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6297 #ifdef CONFIG_IXGBE_DCB
6298 	/* Default traffic class to use for FCoE */
6299 	adapter->fcoe.up = IXGBE_FCOE_DEFTC;
6300 #endif /* CONFIG_IXGBE_DCB */
6301 #endif /* IXGBE_FCOE */
6302 
6303 	/* initialize static ixgbe jump table entries */
6304 	adapter->jump_tables[0] = kzalloc(sizeof(*adapter->jump_tables[0]),
6305 					  GFP_KERNEL);
6306 	if (!adapter->jump_tables[0])
6307 		return -ENOMEM;
6308 	adapter->jump_tables[0]->mat = ixgbe_ipv4_fields;
6309 
6310 	for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++)
6311 		adapter->jump_tables[i] = NULL;
6312 
6313 	adapter->mac_table = kcalloc(hw->mac.num_rar_entries,
6314 				     sizeof(struct ixgbe_mac_addr),
6315 				     GFP_KERNEL);
6316 	if (!adapter->mac_table)
6317 		return -ENOMEM;
6318 
6319 	if (ixgbe_init_rss_key(adapter))
6320 		return -ENOMEM;
6321 
6322 	adapter->af_xdp_zc_qps = bitmap_zalloc(MAX_XDP_QUEUES, GFP_KERNEL);
6323 	if (!adapter->af_xdp_zc_qps)
6324 		return -ENOMEM;
6325 
6326 	/* Set MAC specific capability flags and exceptions */
6327 	switch (hw->mac.type) {
6328 	case ixgbe_mac_82598EB:
6329 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
6330 
6331 		if (hw->device_id == IXGBE_DEV_ID_82598AT)
6332 			adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
6333 
6334 		adapter->max_q_vectors = MAX_Q_VECTORS_82598;
6335 		adapter->ring_feature[RING_F_FDIR].limit = 0;
6336 		adapter->atr_sample_rate = 0;
6337 		adapter->fdir_pballoc = 0;
6338 #ifdef IXGBE_FCOE
6339 		adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6340 		adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6341 #ifdef CONFIG_IXGBE_DCB
6342 		adapter->fcoe.up = 0;
6343 #endif /* IXGBE_DCB */
6344 #endif /* IXGBE_FCOE */
6345 		break;
6346 	case ixgbe_mac_82599EB:
6347 		if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
6348 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6349 		break;
6350 	case ixgbe_mac_X540:
6351 		fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
6352 		if (fwsm & IXGBE_FWSM_TS_ENABLED)
6353 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6354 		break;
6355 	case ixgbe_mac_x550em_a:
6356 		switch (hw->device_id) {
6357 		case IXGBE_DEV_ID_X550EM_A_1G_T:
6358 		case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6359 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6360 			break;
6361 		default:
6362 			break;
6363 		}
6364 		fallthrough;
6365 	case ixgbe_mac_X550EM_x:
6366 #ifdef CONFIG_IXGBE_DCB
6367 		adapter->flags &= ~IXGBE_FLAG_DCB_CAPABLE;
6368 #endif
6369 #ifdef IXGBE_FCOE
6370 		adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6371 #ifdef CONFIG_IXGBE_DCB
6372 		adapter->fcoe.up = 0;
6373 #endif /* IXGBE_DCB */
6374 #endif /* IXGBE_FCOE */
6375 		fallthrough;
6376 	case ixgbe_mac_X550:
6377 		if (hw->mac.type == ixgbe_mac_X550)
6378 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6379 #ifdef CONFIG_IXGBE_DCA
6380 		adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
6381 #endif
6382 		break;
6383 	default:
6384 		break;
6385 	}
6386 
6387 #ifdef IXGBE_FCOE
6388 	/* FCoE support exists, always init the FCoE lock */
6389 	spin_lock_init(&adapter->fcoe.lock);
6390 
6391 #endif
6392 	/* n-tuple support exists, always init our spinlock */
6393 	spin_lock_init(&adapter->fdir_perfect_lock);
6394 
6395 #ifdef CONFIG_IXGBE_DCB
6396 	ixgbe_init_dcb(adapter);
6397 #endif
6398 	ixgbe_init_ipsec_offload(adapter);
6399 
6400 	/* default flow control settings */
6401 	hw->fc.requested_mode = ixgbe_fc_full;
6402 	hw->fc.current_mode = ixgbe_fc_full;	/* init for ethtool output */
6403 	ixgbe_pbthresh_setup(adapter);
6404 	hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
6405 	hw->fc.send_xon = true;
6406 	hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
6407 
6408 #ifdef CONFIG_PCI_IOV
6409 	if (max_vfs > 0)
6410 		e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
6411 
6412 	/* assign number of SR-IOV VFs */
6413 	if (hw->mac.type != ixgbe_mac_82598EB) {
6414 		if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
6415 			max_vfs = 0;
6416 			e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
6417 		}
6418 	}
6419 #endif /* CONFIG_PCI_IOV */
6420 
6421 	/* enable itr by default in dynamic mode */
6422 	adapter->rx_itr_setting = 1;
6423 	adapter->tx_itr_setting = 1;
6424 
6425 	/* set default ring sizes */
6426 	adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
6427 	adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
6428 
6429 	/* set default work limits */
6430 	adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
6431 
6432 	/* initialize eeprom parameters */
6433 	if (ixgbe_init_eeprom_params_generic(hw)) {
6434 		e_dev_err("EEPROM initialization failed\n");
6435 		return -EIO;
6436 	}
6437 
6438 	/* PF holds first pool slot */
6439 	set_bit(0, adapter->fwd_bitmask);
6440 	set_bit(__IXGBE_DOWN, &adapter->state);
6441 
6442 	return 0;
6443 }
6444 
6445 /**
6446  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
6447  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
6448  *
6449  * Return 0 on success, negative on failure
6450  **/
6451 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
6452 {
6453 	struct device *dev = tx_ring->dev;
6454 	int orig_node = dev_to_node(dev);
6455 	int ring_node = NUMA_NO_NODE;
6456 	int size;
6457 
6458 	size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
6459 
6460 	if (tx_ring->q_vector)
6461 		ring_node = tx_ring->q_vector->numa_node;
6462 
6463 	tx_ring->tx_buffer_info = vmalloc_node(size, ring_node);
6464 	if (!tx_ring->tx_buffer_info)
6465 		tx_ring->tx_buffer_info = vmalloc(size);
6466 	if (!tx_ring->tx_buffer_info)
6467 		goto err;
6468 
6469 	/* round up to nearest 4K */
6470 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
6471 	tx_ring->size = ALIGN(tx_ring->size, 4096);
6472 
6473 	set_dev_node(dev, ring_node);
6474 	tx_ring->desc = dma_alloc_coherent(dev,
6475 					   tx_ring->size,
6476 					   &tx_ring->dma,
6477 					   GFP_KERNEL);
6478 	set_dev_node(dev, orig_node);
6479 	if (!tx_ring->desc)
6480 		tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
6481 						   &tx_ring->dma, GFP_KERNEL);
6482 	if (!tx_ring->desc)
6483 		goto err;
6484 
6485 	tx_ring->next_to_use = 0;
6486 	tx_ring->next_to_clean = 0;
6487 	return 0;
6488 
6489 err:
6490 	vfree(tx_ring->tx_buffer_info);
6491 	tx_ring->tx_buffer_info = NULL;
6492 	dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
6493 	return -ENOMEM;
6494 }
6495 
6496 /**
6497  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
6498  * @adapter: board private structure
6499  *
6500  * If this function returns with an error, then it's possible one or
6501  * more of the rings is populated (while the rest are not).  It is the
6502  * callers duty to clean those orphaned rings.
6503  *
6504  * Return 0 on success, negative on failure
6505  **/
6506 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
6507 {
6508 	int i, j = 0, err = 0;
6509 
6510 	for (i = 0; i < adapter->num_tx_queues; i++) {
6511 		err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
6512 		if (!err)
6513 			continue;
6514 
6515 		e_err(probe, "Allocation for Tx Queue %u failed\n", i);
6516 		goto err_setup_tx;
6517 	}
6518 	for (j = 0; j < adapter->num_xdp_queues; j++) {
6519 		err = ixgbe_setup_tx_resources(adapter->xdp_ring[j]);
6520 		if (!err)
6521 			continue;
6522 
6523 		e_err(probe, "Allocation for Tx Queue %u failed\n", j);
6524 		goto err_setup_tx;
6525 	}
6526 
6527 	return 0;
6528 err_setup_tx:
6529 	/* rewind the index freeing the rings as we go */
6530 	while (j--)
6531 		ixgbe_free_tx_resources(adapter->xdp_ring[j]);
6532 	while (i--)
6533 		ixgbe_free_tx_resources(adapter->tx_ring[i]);
6534 	return err;
6535 }
6536 
6537 /**
6538  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
6539  * @adapter: pointer to ixgbe_adapter
6540  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
6541  *
6542  * Returns 0 on success, negative on failure
6543  **/
6544 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
6545 			     struct ixgbe_ring *rx_ring)
6546 {
6547 	struct device *dev = rx_ring->dev;
6548 	int orig_node = dev_to_node(dev);
6549 	int ring_node = NUMA_NO_NODE;
6550 	int size;
6551 
6552 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
6553 
6554 	if (rx_ring->q_vector)
6555 		ring_node = rx_ring->q_vector->numa_node;
6556 
6557 	rx_ring->rx_buffer_info = vmalloc_node(size, ring_node);
6558 	if (!rx_ring->rx_buffer_info)
6559 		rx_ring->rx_buffer_info = vmalloc(size);
6560 	if (!rx_ring->rx_buffer_info)
6561 		goto err;
6562 
6563 	/* Round up to nearest 4K */
6564 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
6565 	rx_ring->size = ALIGN(rx_ring->size, 4096);
6566 
6567 	set_dev_node(dev, ring_node);
6568 	rx_ring->desc = dma_alloc_coherent(dev,
6569 					   rx_ring->size,
6570 					   &rx_ring->dma,
6571 					   GFP_KERNEL);
6572 	set_dev_node(dev, orig_node);
6573 	if (!rx_ring->desc)
6574 		rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
6575 						   &rx_ring->dma, GFP_KERNEL);
6576 	if (!rx_ring->desc)
6577 		goto err;
6578 
6579 	rx_ring->next_to_clean = 0;
6580 	rx_ring->next_to_use = 0;
6581 	rx_ring->rx_offset = ixgbe_rx_offset(rx_ring);
6582 
6583 	/* XDP RX-queue info */
6584 	if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
6585 			     rx_ring->queue_index, rx_ring->q_vector->napi.napi_id) < 0)
6586 		goto err;
6587 
6588 	rx_ring->xdp_prog = adapter->xdp_prog;
6589 
6590 	return 0;
6591 err:
6592 	vfree(rx_ring->rx_buffer_info);
6593 	rx_ring->rx_buffer_info = NULL;
6594 	dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
6595 	return -ENOMEM;
6596 }
6597 
6598 /**
6599  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
6600  * @adapter: board private structure
6601  *
6602  * If this function returns with an error, then it's possible one or
6603  * more of the rings is populated (while the rest are not).  It is the
6604  * callers duty to clean those orphaned rings.
6605  *
6606  * Return 0 on success, negative on failure
6607  **/
6608 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
6609 {
6610 	int i, err = 0;
6611 
6612 	for (i = 0; i < adapter->num_rx_queues; i++) {
6613 		err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
6614 		if (!err)
6615 			continue;
6616 
6617 		e_err(probe, "Allocation for Rx Queue %u failed\n", i);
6618 		goto err_setup_rx;
6619 	}
6620 
6621 #ifdef IXGBE_FCOE
6622 	err = ixgbe_setup_fcoe_ddp_resources(adapter);
6623 	if (!err)
6624 #endif
6625 		return 0;
6626 err_setup_rx:
6627 	/* rewind the index freeing the rings as we go */
6628 	while (i--)
6629 		ixgbe_free_rx_resources(adapter->rx_ring[i]);
6630 	return err;
6631 }
6632 
6633 /**
6634  * ixgbe_free_tx_resources - Free Tx Resources per Queue
6635  * @tx_ring: Tx descriptor ring for a specific queue
6636  *
6637  * Free all transmit software resources
6638  **/
6639 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
6640 {
6641 	ixgbe_clean_tx_ring(tx_ring);
6642 
6643 	vfree(tx_ring->tx_buffer_info);
6644 	tx_ring->tx_buffer_info = NULL;
6645 
6646 	/* if not set, then don't free */
6647 	if (!tx_ring->desc)
6648 		return;
6649 
6650 	dma_free_coherent(tx_ring->dev, tx_ring->size,
6651 			  tx_ring->desc, tx_ring->dma);
6652 
6653 	tx_ring->desc = NULL;
6654 }
6655 
6656 /**
6657  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
6658  * @adapter: board private structure
6659  *
6660  * Free all transmit software resources
6661  **/
6662 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
6663 {
6664 	int i;
6665 
6666 	for (i = 0; i < adapter->num_tx_queues; i++)
6667 		if (adapter->tx_ring[i]->desc)
6668 			ixgbe_free_tx_resources(adapter->tx_ring[i]);
6669 	for (i = 0; i < adapter->num_xdp_queues; i++)
6670 		if (adapter->xdp_ring[i]->desc)
6671 			ixgbe_free_tx_resources(adapter->xdp_ring[i]);
6672 }
6673 
6674 /**
6675  * ixgbe_free_rx_resources - Free Rx Resources
6676  * @rx_ring: ring to clean the resources from
6677  *
6678  * Free all receive software resources
6679  **/
6680 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
6681 {
6682 	ixgbe_clean_rx_ring(rx_ring);
6683 
6684 	rx_ring->xdp_prog = NULL;
6685 	xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
6686 	vfree(rx_ring->rx_buffer_info);
6687 	rx_ring->rx_buffer_info = NULL;
6688 
6689 	/* if not set, then don't free */
6690 	if (!rx_ring->desc)
6691 		return;
6692 
6693 	dma_free_coherent(rx_ring->dev, rx_ring->size,
6694 			  rx_ring->desc, rx_ring->dma);
6695 
6696 	rx_ring->desc = NULL;
6697 }
6698 
6699 /**
6700  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
6701  * @adapter: board private structure
6702  *
6703  * Free all receive software resources
6704  **/
6705 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
6706 {
6707 	int i;
6708 
6709 #ifdef IXGBE_FCOE
6710 	ixgbe_free_fcoe_ddp_resources(adapter);
6711 
6712 #endif
6713 	for (i = 0; i < adapter->num_rx_queues; i++)
6714 		if (adapter->rx_ring[i]->desc)
6715 			ixgbe_free_rx_resources(adapter->rx_ring[i]);
6716 }
6717 
6718 /**
6719  * ixgbe_change_mtu - Change the Maximum Transfer Unit
6720  * @netdev: network interface device structure
6721  * @new_mtu: new value for maximum frame size
6722  *
6723  * Returns 0 on success, negative on failure
6724  **/
6725 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
6726 {
6727 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6728 
6729 	if (adapter->xdp_prog) {
6730 		int new_frame_size = new_mtu + ETH_HLEN + ETH_FCS_LEN +
6731 				     VLAN_HLEN;
6732 		int i;
6733 
6734 		for (i = 0; i < adapter->num_rx_queues; i++) {
6735 			struct ixgbe_ring *ring = adapter->rx_ring[i];
6736 
6737 			if (new_frame_size > ixgbe_rx_bufsz(ring)) {
6738 				e_warn(probe, "Requested MTU size is not supported with XDP\n");
6739 				return -EINVAL;
6740 			}
6741 		}
6742 	}
6743 
6744 	/*
6745 	 * For 82599EB we cannot allow legacy VFs to enable their receive
6746 	 * paths when MTU greater than 1500 is configured.  So display a
6747 	 * warning that legacy VFs will be disabled.
6748 	 */
6749 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
6750 	    (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
6751 	    (new_mtu > ETH_DATA_LEN))
6752 		e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
6753 
6754 	netdev_dbg(netdev, "changing MTU from %d to %d\n",
6755 		   netdev->mtu, new_mtu);
6756 
6757 	/* must set new MTU before calling down or up */
6758 	netdev->mtu = new_mtu;
6759 
6760 	if (netif_running(netdev))
6761 		ixgbe_reinit_locked(adapter);
6762 
6763 	return 0;
6764 }
6765 
6766 /**
6767  * ixgbe_open - Called when a network interface is made active
6768  * @netdev: network interface device structure
6769  *
6770  * Returns 0 on success, negative value on failure
6771  *
6772  * The open entry point is called when a network interface is made
6773  * active by the system (IFF_UP).  At this point all resources needed
6774  * for transmit and receive operations are allocated, the interrupt
6775  * handler is registered with the OS, the watchdog timer is started,
6776  * and the stack is notified that the interface is ready.
6777  **/
6778 int ixgbe_open(struct net_device *netdev)
6779 {
6780 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6781 	struct ixgbe_hw *hw = &adapter->hw;
6782 	int err, queues;
6783 
6784 	/* disallow open during test */
6785 	if (test_bit(__IXGBE_TESTING, &adapter->state))
6786 		return -EBUSY;
6787 
6788 	netif_carrier_off(netdev);
6789 
6790 	/* allocate transmit descriptors */
6791 	err = ixgbe_setup_all_tx_resources(adapter);
6792 	if (err)
6793 		goto err_setup_tx;
6794 
6795 	/* allocate receive descriptors */
6796 	err = ixgbe_setup_all_rx_resources(adapter);
6797 	if (err)
6798 		goto err_setup_rx;
6799 
6800 	ixgbe_configure(adapter);
6801 
6802 	err = ixgbe_request_irq(adapter);
6803 	if (err)
6804 		goto err_req_irq;
6805 
6806 	/* Notify the stack of the actual queue counts. */
6807 	queues = adapter->num_tx_queues;
6808 	err = netif_set_real_num_tx_queues(netdev, queues);
6809 	if (err)
6810 		goto err_set_queues;
6811 
6812 	queues = adapter->num_rx_queues;
6813 	err = netif_set_real_num_rx_queues(netdev, queues);
6814 	if (err)
6815 		goto err_set_queues;
6816 
6817 	ixgbe_ptp_init(adapter);
6818 
6819 	ixgbe_up_complete(adapter);
6820 
6821 	udp_tunnel_nic_reset_ntf(netdev);
6822 
6823 	return 0;
6824 
6825 err_set_queues:
6826 	ixgbe_free_irq(adapter);
6827 err_req_irq:
6828 	ixgbe_free_all_rx_resources(adapter);
6829 	if (hw->phy.ops.set_phy_power && !adapter->wol)
6830 		hw->phy.ops.set_phy_power(&adapter->hw, false);
6831 err_setup_rx:
6832 	ixgbe_free_all_tx_resources(adapter);
6833 err_setup_tx:
6834 	ixgbe_reset(adapter);
6835 
6836 	return err;
6837 }
6838 
6839 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
6840 {
6841 	ixgbe_ptp_suspend(adapter);
6842 
6843 	if (adapter->hw.phy.ops.enter_lplu) {
6844 		adapter->hw.phy.reset_disable = true;
6845 		ixgbe_down(adapter);
6846 		adapter->hw.phy.ops.enter_lplu(&adapter->hw);
6847 		adapter->hw.phy.reset_disable = false;
6848 	} else {
6849 		ixgbe_down(adapter);
6850 	}
6851 
6852 	ixgbe_free_irq(adapter);
6853 
6854 	ixgbe_free_all_tx_resources(adapter);
6855 	ixgbe_free_all_rx_resources(adapter);
6856 }
6857 
6858 /**
6859  * ixgbe_close - Disables a network interface
6860  * @netdev: network interface device structure
6861  *
6862  * Returns 0, this is not allowed to fail
6863  *
6864  * The close entry point is called when an interface is de-activated
6865  * by the OS.  The hardware is still under the drivers control, but
6866  * needs to be disabled.  A global MAC reset is issued to stop the
6867  * hardware, and all transmit and receive resources are freed.
6868  **/
6869 int ixgbe_close(struct net_device *netdev)
6870 {
6871 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6872 
6873 	ixgbe_ptp_stop(adapter);
6874 
6875 	if (netif_device_present(netdev))
6876 		ixgbe_close_suspend(adapter);
6877 
6878 	ixgbe_fdir_filter_exit(adapter);
6879 
6880 	ixgbe_release_hw_control(adapter);
6881 
6882 	return 0;
6883 }
6884 
6885 static int __maybe_unused ixgbe_resume(struct device *dev_d)
6886 {
6887 	struct pci_dev *pdev = to_pci_dev(dev_d);
6888 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6889 	struct net_device *netdev = adapter->netdev;
6890 	u32 err;
6891 
6892 	adapter->hw.hw_addr = adapter->io_addr;
6893 
6894 	smp_mb__before_atomic();
6895 	clear_bit(__IXGBE_DISABLED, &adapter->state);
6896 	pci_set_master(pdev);
6897 
6898 	device_wakeup_disable(dev_d);
6899 
6900 	ixgbe_reset(adapter);
6901 
6902 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6903 
6904 	rtnl_lock();
6905 	err = ixgbe_init_interrupt_scheme(adapter);
6906 	if (!err && netif_running(netdev))
6907 		err = ixgbe_open(netdev);
6908 
6909 
6910 	if (!err)
6911 		netif_device_attach(netdev);
6912 	rtnl_unlock();
6913 
6914 	return err;
6915 }
6916 
6917 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
6918 {
6919 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6920 	struct net_device *netdev = adapter->netdev;
6921 	struct ixgbe_hw *hw = &adapter->hw;
6922 	u32 ctrl;
6923 	u32 wufc = adapter->wol;
6924 
6925 	rtnl_lock();
6926 	netif_device_detach(netdev);
6927 
6928 	if (netif_running(netdev))
6929 		ixgbe_close_suspend(adapter);
6930 
6931 	ixgbe_clear_interrupt_scheme(adapter);
6932 	rtnl_unlock();
6933 
6934 	if (hw->mac.ops.stop_link_on_d3)
6935 		hw->mac.ops.stop_link_on_d3(hw);
6936 
6937 	if (wufc) {
6938 		u32 fctrl;
6939 
6940 		ixgbe_set_rx_mode(netdev);
6941 
6942 		/* enable the optics for 82599 SFP+ fiber as we can WoL */
6943 		if (hw->mac.ops.enable_tx_laser)
6944 			hw->mac.ops.enable_tx_laser(hw);
6945 
6946 		/* enable the reception of multicast packets */
6947 		fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6948 		fctrl |= IXGBE_FCTRL_MPE;
6949 		IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
6950 
6951 		ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
6952 		ctrl |= IXGBE_CTRL_GIO_DIS;
6953 		IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
6954 
6955 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
6956 	} else {
6957 		IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
6958 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
6959 	}
6960 
6961 	switch (hw->mac.type) {
6962 	case ixgbe_mac_82598EB:
6963 		pci_wake_from_d3(pdev, false);
6964 		break;
6965 	case ixgbe_mac_82599EB:
6966 	case ixgbe_mac_X540:
6967 	case ixgbe_mac_X550:
6968 	case ixgbe_mac_X550EM_x:
6969 	case ixgbe_mac_x550em_a:
6970 		pci_wake_from_d3(pdev, !!wufc);
6971 		break;
6972 	default:
6973 		break;
6974 	}
6975 
6976 	*enable_wake = !!wufc;
6977 	if (hw->phy.ops.set_phy_power && !*enable_wake)
6978 		hw->phy.ops.set_phy_power(hw, false);
6979 
6980 	ixgbe_release_hw_control(adapter);
6981 
6982 	if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
6983 		pci_disable_device(pdev);
6984 
6985 	return 0;
6986 }
6987 
6988 static int __maybe_unused ixgbe_suspend(struct device *dev_d)
6989 {
6990 	struct pci_dev *pdev = to_pci_dev(dev_d);
6991 	int retval;
6992 	bool wake;
6993 
6994 	retval = __ixgbe_shutdown(pdev, &wake);
6995 
6996 	device_set_wakeup_enable(dev_d, wake);
6997 
6998 	return retval;
6999 }
7000 
7001 static void ixgbe_shutdown(struct pci_dev *pdev)
7002 {
7003 	bool wake;
7004 
7005 	__ixgbe_shutdown(pdev, &wake);
7006 
7007 	if (system_state == SYSTEM_POWER_OFF) {
7008 		pci_wake_from_d3(pdev, wake);
7009 		pci_set_power_state(pdev, PCI_D3hot);
7010 	}
7011 }
7012 
7013 /**
7014  * ixgbe_update_stats - Update the board statistics counters.
7015  * @adapter: board private structure
7016  **/
7017 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
7018 {
7019 	struct net_device *netdev = adapter->netdev;
7020 	struct ixgbe_hw *hw = &adapter->hw;
7021 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
7022 	u64 total_mpc = 0;
7023 	u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
7024 	u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
7025 	u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
7026 	u64 alloc_rx_page = 0;
7027 	u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
7028 
7029 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7030 	    test_bit(__IXGBE_RESETTING, &adapter->state))
7031 		return;
7032 
7033 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
7034 		u64 rsc_count = 0;
7035 		u64 rsc_flush = 0;
7036 		for (i = 0; i < adapter->num_rx_queues; i++) {
7037 			rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
7038 			rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
7039 		}
7040 		adapter->rsc_total_count = rsc_count;
7041 		adapter->rsc_total_flush = rsc_flush;
7042 	}
7043 
7044 	for (i = 0; i < adapter->num_rx_queues; i++) {
7045 		struct ixgbe_ring *rx_ring = READ_ONCE(adapter->rx_ring[i]);
7046 
7047 		if (!rx_ring)
7048 			continue;
7049 		non_eop_descs += rx_ring->rx_stats.non_eop_descs;
7050 		alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
7051 		alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
7052 		alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
7053 		hw_csum_rx_error += rx_ring->rx_stats.csum_err;
7054 		bytes += rx_ring->stats.bytes;
7055 		packets += rx_ring->stats.packets;
7056 	}
7057 	adapter->non_eop_descs = non_eop_descs;
7058 	adapter->alloc_rx_page = alloc_rx_page;
7059 	adapter->alloc_rx_page_failed = alloc_rx_page_failed;
7060 	adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
7061 	adapter->hw_csum_rx_error = hw_csum_rx_error;
7062 	netdev->stats.rx_bytes = bytes;
7063 	netdev->stats.rx_packets = packets;
7064 
7065 	bytes = 0;
7066 	packets = 0;
7067 	/* gather some stats to the adapter struct that are per queue */
7068 	for (i = 0; i < adapter->num_tx_queues; i++) {
7069 		struct ixgbe_ring *tx_ring = READ_ONCE(adapter->tx_ring[i]);
7070 
7071 		if (!tx_ring)
7072 			continue;
7073 		restart_queue += tx_ring->tx_stats.restart_queue;
7074 		tx_busy += tx_ring->tx_stats.tx_busy;
7075 		bytes += tx_ring->stats.bytes;
7076 		packets += tx_ring->stats.packets;
7077 	}
7078 	for (i = 0; i < adapter->num_xdp_queues; i++) {
7079 		struct ixgbe_ring *xdp_ring = READ_ONCE(adapter->xdp_ring[i]);
7080 
7081 		if (!xdp_ring)
7082 			continue;
7083 		restart_queue += xdp_ring->tx_stats.restart_queue;
7084 		tx_busy += xdp_ring->tx_stats.tx_busy;
7085 		bytes += xdp_ring->stats.bytes;
7086 		packets += xdp_ring->stats.packets;
7087 	}
7088 	adapter->restart_queue = restart_queue;
7089 	adapter->tx_busy = tx_busy;
7090 	netdev->stats.tx_bytes = bytes;
7091 	netdev->stats.tx_packets = packets;
7092 
7093 	hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
7094 
7095 	/* 8 register reads */
7096 	for (i = 0; i < 8; i++) {
7097 		/* for packet buffers not used, the register should read 0 */
7098 		mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
7099 		missed_rx += mpc;
7100 		hwstats->mpc[i] += mpc;
7101 		total_mpc += hwstats->mpc[i];
7102 		hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
7103 		hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
7104 		switch (hw->mac.type) {
7105 		case ixgbe_mac_82598EB:
7106 			hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
7107 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
7108 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
7109 			hwstats->pxonrxc[i] +=
7110 				IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
7111 			break;
7112 		case ixgbe_mac_82599EB:
7113 		case ixgbe_mac_X540:
7114 		case ixgbe_mac_X550:
7115 		case ixgbe_mac_X550EM_x:
7116 		case ixgbe_mac_x550em_a:
7117 			hwstats->pxonrxc[i] +=
7118 				IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
7119 			break;
7120 		default:
7121 			break;
7122 		}
7123 	}
7124 
7125 	/*16 register reads */
7126 	for (i = 0; i < 16; i++) {
7127 		hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
7128 		hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
7129 		if ((hw->mac.type == ixgbe_mac_82599EB) ||
7130 		    (hw->mac.type == ixgbe_mac_X540) ||
7131 		    (hw->mac.type == ixgbe_mac_X550) ||
7132 		    (hw->mac.type == ixgbe_mac_X550EM_x) ||
7133 		    (hw->mac.type == ixgbe_mac_x550em_a)) {
7134 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
7135 			IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
7136 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
7137 			IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
7138 		}
7139 	}
7140 
7141 	hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
7142 	/* work around hardware counting issue */
7143 	hwstats->gprc -= missed_rx;
7144 
7145 	ixgbe_update_xoff_received(adapter);
7146 
7147 	/* 82598 hardware only has a 32 bit counter in the high register */
7148 	switch (hw->mac.type) {
7149 	case ixgbe_mac_82598EB:
7150 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
7151 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
7152 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
7153 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
7154 		break;
7155 	case ixgbe_mac_X540:
7156 	case ixgbe_mac_X550:
7157 	case ixgbe_mac_X550EM_x:
7158 	case ixgbe_mac_x550em_a:
7159 		/* OS2BMC stats are X540 and later */
7160 		hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
7161 		hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
7162 		hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
7163 		hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
7164 		fallthrough;
7165 	case ixgbe_mac_82599EB:
7166 		for (i = 0; i < 16; i++)
7167 			adapter->hw_rx_no_dma_resources +=
7168 					     IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
7169 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
7170 		IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
7171 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
7172 		IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
7173 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
7174 		IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
7175 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
7176 		hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
7177 		hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
7178 #ifdef IXGBE_FCOE
7179 		hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
7180 		hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
7181 		hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
7182 		hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
7183 		hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
7184 		hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
7185 		/* Add up per cpu counters for total ddp aloc fail */
7186 		if (adapter->fcoe.ddp_pool) {
7187 			struct ixgbe_fcoe *fcoe = &adapter->fcoe;
7188 			struct ixgbe_fcoe_ddp_pool *ddp_pool;
7189 			unsigned int cpu;
7190 			u64 noddp = 0, noddp_ext_buff = 0;
7191 			for_each_possible_cpu(cpu) {
7192 				ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
7193 				noddp += ddp_pool->noddp;
7194 				noddp_ext_buff += ddp_pool->noddp_ext_buff;
7195 			}
7196 			hwstats->fcoe_noddp = noddp;
7197 			hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
7198 		}
7199 #endif /* IXGBE_FCOE */
7200 		break;
7201 	default:
7202 		break;
7203 	}
7204 	bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
7205 	hwstats->bprc += bprc;
7206 	hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
7207 	if (hw->mac.type == ixgbe_mac_82598EB)
7208 		hwstats->mprc -= bprc;
7209 	hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
7210 	hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
7211 	hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
7212 	hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
7213 	hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
7214 	hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
7215 	hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
7216 	hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
7217 	lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
7218 	hwstats->lxontxc += lxon;
7219 	lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
7220 	hwstats->lxofftxc += lxoff;
7221 	hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
7222 	hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
7223 	/*
7224 	 * 82598 errata - tx of flow control packets is included in tx counters
7225 	 */
7226 	xon_off_tot = lxon + lxoff;
7227 	hwstats->gptc -= xon_off_tot;
7228 	hwstats->mptc -= xon_off_tot;
7229 	hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
7230 	hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
7231 	hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
7232 	hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
7233 	hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
7234 	hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
7235 	hwstats->ptc64 -= xon_off_tot;
7236 	hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
7237 	hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
7238 	hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
7239 	hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
7240 	hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
7241 	hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
7242 
7243 	/* Fill out the OS statistics structure */
7244 	netdev->stats.multicast = hwstats->mprc;
7245 
7246 	/* Rx Errors */
7247 	netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
7248 	netdev->stats.rx_dropped = 0;
7249 	netdev->stats.rx_length_errors = hwstats->rlec;
7250 	netdev->stats.rx_crc_errors = hwstats->crcerrs;
7251 	netdev->stats.rx_missed_errors = total_mpc;
7252 }
7253 
7254 /**
7255  * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
7256  * @adapter: pointer to the device adapter structure
7257  **/
7258 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
7259 {
7260 	struct ixgbe_hw *hw = &adapter->hw;
7261 	int i;
7262 
7263 	if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
7264 		return;
7265 
7266 	adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
7267 
7268 	/* if interface is down do nothing */
7269 	if (test_bit(__IXGBE_DOWN, &adapter->state))
7270 		return;
7271 
7272 	/* do nothing if we are not using signature filters */
7273 	if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
7274 		return;
7275 
7276 	adapter->fdir_overflow++;
7277 
7278 	if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
7279 		for (i = 0; i < adapter->num_tx_queues; i++)
7280 			set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7281 				&(adapter->tx_ring[i]->state));
7282 		for (i = 0; i < adapter->num_xdp_queues; i++)
7283 			set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7284 				&adapter->xdp_ring[i]->state);
7285 		/* re-enable flow director interrupts */
7286 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
7287 	} else {
7288 		e_err(probe, "failed to finish FDIR re-initialization, "
7289 		      "ignored adding FDIR ATR filters\n");
7290 	}
7291 }
7292 
7293 /**
7294  * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
7295  * @adapter: pointer to the device adapter structure
7296  *
7297  * This function serves two purposes.  First it strobes the interrupt lines
7298  * in order to make certain interrupts are occurring.  Secondly it sets the
7299  * bits needed to check for TX hangs.  As a result we should immediately
7300  * determine if a hang has occurred.
7301  */
7302 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
7303 {
7304 	struct ixgbe_hw *hw = &adapter->hw;
7305 	u64 eics = 0;
7306 	int i;
7307 
7308 	/* If we're down, removing or resetting, just bail */
7309 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7310 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
7311 	    test_bit(__IXGBE_RESETTING, &adapter->state))
7312 		return;
7313 
7314 	/* Force detection of hung controller */
7315 	if (netif_carrier_ok(adapter->netdev)) {
7316 		for (i = 0; i < adapter->num_tx_queues; i++)
7317 			set_check_for_tx_hang(adapter->tx_ring[i]);
7318 		for (i = 0; i < adapter->num_xdp_queues; i++)
7319 			set_check_for_tx_hang(adapter->xdp_ring[i]);
7320 	}
7321 
7322 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
7323 		/*
7324 		 * for legacy and MSI interrupts don't set any bits
7325 		 * that are enabled for EIAM, because this operation
7326 		 * would set *both* EIMS and EICS for any bit in EIAM
7327 		 */
7328 		IXGBE_WRITE_REG(hw, IXGBE_EICS,
7329 			(IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
7330 	} else {
7331 		/* get one bit for every active tx/rx interrupt vector */
7332 		for (i = 0; i < adapter->num_q_vectors; i++) {
7333 			struct ixgbe_q_vector *qv = adapter->q_vector[i];
7334 			if (qv->rx.ring || qv->tx.ring)
7335 				eics |= BIT_ULL(i);
7336 		}
7337 	}
7338 
7339 	/* Cause software interrupt to ensure rings are cleaned */
7340 	ixgbe_irq_rearm_queues(adapter, eics);
7341 }
7342 
7343 /**
7344  * ixgbe_watchdog_update_link - update the link status
7345  * @adapter: pointer to the device adapter structure
7346  **/
7347 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
7348 {
7349 	struct ixgbe_hw *hw = &adapter->hw;
7350 	u32 link_speed = adapter->link_speed;
7351 	bool link_up = adapter->link_up;
7352 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
7353 
7354 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
7355 		return;
7356 
7357 	if (hw->mac.ops.check_link) {
7358 		hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
7359 	} else {
7360 		/* always assume link is up, if no check link function */
7361 		link_speed = IXGBE_LINK_SPEED_10GB_FULL;
7362 		link_up = true;
7363 	}
7364 
7365 	if (adapter->ixgbe_ieee_pfc)
7366 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
7367 
7368 	if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
7369 		hw->mac.ops.fc_enable(hw);
7370 		ixgbe_set_rx_drop_en(adapter);
7371 	}
7372 
7373 	if (link_up ||
7374 	    time_after(jiffies, (adapter->link_check_timeout +
7375 				 IXGBE_TRY_LINK_TIMEOUT))) {
7376 		adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
7377 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
7378 		IXGBE_WRITE_FLUSH(hw);
7379 	}
7380 
7381 	adapter->link_up = link_up;
7382 	adapter->link_speed = link_speed;
7383 }
7384 
7385 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
7386 {
7387 #ifdef CONFIG_IXGBE_DCB
7388 	struct net_device *netdev = adapter->netdev;
7389 	struct dcb_app app = {
7390 			      .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
7391 			      .protocol = 0,
7392 			     };
7393 	u8 up = 0;
7394 
7395 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
7396 		up = dcb_ieee_getapp_mask(netdev, &app);
7397 
7398 	adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
7399 #endif
7400 }
7401 
7402 /**
7403  * ixgbe_watchdog_link_is_up - update netif_carrier status and
7404  *                             print link up message
7405  * @adapter: pointer to the device adapter structure
7406  **/
7407 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
7408 {
7409 	struct net_device *netdev = adapter->netdev;
7410 	struct ixgbe_hw *hw = &adapter->hw;
7411 	u32 link_speed = adapter->link_speed;
7412 	const char *speed_str;
7413 	bool flow_rx, flow_tx;
7414 
7415 	/* only continue if link was previously down */
7416 	if (netif_carrier_ok(netdev))
7417 		return;
7418 
7419 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7420 
7421 	switch (hw->mac.type) {
7422 	case ixgbe_mac_82598EB: {
7423 		u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
7424 		u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
7425 		flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
7426 		flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
7427 	}
7428 		break;
7429 	case ixgbe_mac_X540:
7430 	case ixgbe_mac_X550:
7431 	case ixgbe_mac_X550EM_x:
7432 	case ixgbe_mac_x550em_a:
7433 	case ixgbe_mac_82599EB: {
7434 		u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
7435 		u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
7436 		flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
7437 		flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
7438 	}
7439 		break;
7440 	default:
7441 		flow_tx = false;
7442 		flow_rx = false;
7443 		break;
7444 	}
7445 
7446 	adapter->last_rx_ptp_check = jiffies;
7447 
7448 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7449 		ixgbe_ptp_start_cyclecounter(adapter);
7450 
7451 	switch (link_speed) {
7452 	case IXGBE_LINK_SPEED_10GB_FULL:
7453 		speed_str = "10 Gbps";
7454 		break;
7455 	case IXGBE_LINK_SPEED_5GB_FULL:
7456 		speed_str = "5 Gbps";
7457 		break;
7458 	case IXGBE_LINK_SPEED_2_5GB_FULL:
7459 		speed_str = "2.5 Gbps";
7460 		break;
7461 	case IXGBE_LINK_SPEED_1GB_FULL:
7462 		speed_str = "1 Gbps";
7463 		break;
7464 	case IXGBE_LINK_SPEED_100_FULL:
7465 		speed_str = "100 Mbps";
7466 		break;
7467 	case IXGBE_LINK_SPEED_10_FULL:
7468 		speed_str = "10 Mbps";
7469 		break;
7470 	default:
7471 		speed_str = "unknown speed";
7472 		break;
7473 	}
7474 	e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str,
7475 	       ((flow_rx && flow_tx) ? "RX/TX" :
7476 	       (flow_rx ? "RX" :
7477 	       (flow_tx ? "TX" : "None"))));
7478 
7479 	netif_carrier_on(netdev);
7480 	ixgbe_check_vf_rate_limit(adapter);
7481 
7482 	/* enable transmits */
7483 	netif_tx_wake_all_queues(adapter->netdev);
7484 
7485 	/* update the default user priority for VFs */
7486 	ixgbe_update_default_up(adapter);
7487 
7488 	/* ping all the active vfs to let them know link has changed */
7489 	ixgbe_ping_all_vfs(adapter);
7490 }
7491 
7492 /**
7493  * ixgbe_watchdog_link_is_down - update netif_carrier status and
7494  *                               print link down message
7495  * @adapter: pointer to the adapter structure
7496  **/
7497 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
7498 {
7499 	struct net_device *netdev = adapter->netdev;
7500 	struct ixgbe_hw *hw = &adapter->hw;
7501 
7502 	adapter->link_up = false;
7503 	adapter->link_speed = 0;
7504 
7505 	/* only continue if link was up previously */
7506 	if (!netif_carrier_ok(netdev))
7507 		return;
7508 
7509 	/* poll for SFP+ cable when link is down */
7510 	if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
7511 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
7512 
7513 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7514 		ixgbe_ptp_start_cyclecounter(adapter);
7515 
7516 	e_info(drv, "NIC Link is Down\n");
7517 	netif_carrier_off(netdev);
7518 
7519 	/* ping all the active vfs to let them know link has changed */
7520 	ixgbe_ping_all_vfs(adapter);
7521 }
7522 
7523 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
7524 {
7525 	int i;
7526 
7527 	for (i = 0; i < adapter->num_tx_queues; i++) {
7528 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7529 
7530 		if (tx_ring->next_to_use != tx_ring->next_to_clean)
7531 			return true;
7532 	}
7533 
7534 	for (i = 0; i < adapter->num_xdp_queues; i++) {
7535 		struct ixgbe_ring *ring = adapter->xdp_ring[i];
7536 
7537 		if (ring->next_to_use != ring->next_to_clean)
7538 			return true;
7539 	}
7540 
7541 	return false;
7542 }
7543 
7544 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
7545 {
7546 	struct ixgbe_hw *hw = &adapter->hw;
7547 	struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
7548 	u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
7549 
7550 	int i, j;
7551 
7552 	if (!adapter->num_vfs)
7553 		return false;
7554 
7555 	/* resetting the PF is only needed for MAC before X550 */
7556 	if (hw->mac.type >= ixgbe_mac_X550)
7557 		return false;
7558 
7559 	for (i = 0; i < adapter->num_vfs; i++) {
7560 		for (j = 0; j < q_per_pool; j++) {
7561 			u32 h, t;
7562 
7563 			h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
7564 			t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
7565 
7566 			if (h != t)
7567 				return true;
7568 		}
7569 	}
7570 
7571 	return false;
7572 }
7573 
7574 /**
7575  * ixgbe_watchdog_flush_tx - flush queues on link down
7576  * @adapter: pointer to the device adapter structure
7577  **/
7578 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
7579 {
7580 	if (!netif_carrier_ok(adapter->netdev)) {
7581 		if (ixgbe_ring_tx_pending(adapter) ||
7582 		    ixgbe_vf_tx_pending(adapter)) {
7583 			/* We've lost link, so the controller stops DMA,
7584 			 * but we've got queued Tx work that's never going
7585 			 * to get done, so reset controller to flush Tx.
7586 			 * (Do the reset outside of interrupt context).
7587 			 */
7588 			e_warn(drv, "initiating reset to clear Tx work after link loss\n");
7589 			set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
7590 		}
7591 	}
7592 }
7593 
7594 #ifdef CONFIG_PCI_IOV
7595 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
7596 {
7597 	struct ixgbe_hw *hw = &adapter->hw;
7598 	struct pci_dev *pdev = adapter->pdev;
7599 	unsigned int vf;
7600 	u32 gpc;
7601 
7602 	if (!(netif_carrier_ok(adapter->netdev)))
7603 		return;
7604 
7605 	gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
7606 	if (gpc) /* If incrementing then no need for the check below */
7607 		return;
7608 	/* Check to see if a bad DMA write target from an errant or
7609 	 * malicious VF has caused a PCIe error.  If so then we can
7610 	 * issue a VFLR to the offending VF(s) and then resume without
7611 	 * requesting a full slot reset.
7612 	 */
7613 
7614 	if (!pdev)
7615 		return;
7616 
7617 	/* check status reg for all VFs owned by this PF */
7618 	for (vf = 0; vf < adapter->num_vfs; ++vf) {
7619 		struct pci_dev *vfdev = adapter->vfinfo[vf].vfdev;
7620 		u16 status_reg;
7621 
7622 		if (!vfdev)
7623 			continue;
7624 		pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
7625 		if (status_reg != IXGBE_FAILED_READ_CFG_WORD &&
7626 		    status_reg & PCI_STATUS_REC_MASTER_ABORT)
7627 			pcie_flr(vfdev);
7628 	}
7629 }
7630 
7631 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
7632 {
7633 	u32 ssvpc;
7634 
7635 	/* Do not perform spoof check for 82598 or if not in IOV mode */
7636 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7637 	    adapter->num_vfs == 0)
7638 		return;
7639 
7640 	ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
7641 
7642 	/*
7643 	 * ssvpc register is cleared on read, if zero then no
7644 	 * spoofed packets in the last interval.
7645 	 */
7646 	if (!ssvpc)
7647 		return;
7648 
7649 	e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
7650 }
7651 #else
7652 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
7653 {
7654 }
7655 
7656 static void
7657 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
7658 {
7659 }
7660 #endif /* CONFIG_PCI_IOV */
7661 
7662 
7663 /**
7664  * ixgbe_watchdog_subtask - check and bring link up
7665  * @adapter: pointer to the device adapter structure
7666  **/
7667 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
7668 {
7669 	/* if interface is down, removing or resetting, do nothing */
7670 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7671 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
7672 	    test_bit(__IXGBE_RESETTING, &adapter->state))
7673 		return;
7674 
7675 	ixgbe_watchdog_update_link(adapter);
7676 
7677 	if (adapter->link_up)
7678 		ixgbe_watchdog_link_is_up(adapter);
7679 	else
7680 		ixgbe_watchdog_link_is_down(adapter);
7681 
7682 	ixgbe_check_for_bad_vf(adapter);
7683 	ixgbe_spoof_check(adapter);
7684 	ixgbe_update_stats(adapter);
7685 
7686 	ixgbe_watchdog_flush_tx(adapter);
7687 }
7688 
7689 /**
7690  * ixgbe_sfp_detection_subtask - poll for SFP+ cable
7691  * @adapter: the ixgbe adapter structure
7692  **/
7693 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
7694 {
7695 	struct ixgbe_hw *hw = &adapter->hw;
7696 	s32 err;
7697 
7698 	/* not searching for SFP so there is nothing to do here */
7699 	if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
7700 	    !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7701 		return;
7702 
7703 	if (adapter->sfp_poll_time &&
7704 	    time_after(adapter->sfp_poll_time, jiffies))
7705 		return; /* If not yet time to poll for SFP */
7706 
7707 	/* someone else is in init, wait until next service event */
7708 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7709 		return;
7710 
7711 	adapter->sfp_poll_time = jiffies + IXGBE_SFP_POLL_JIFFIES - 1;
7712 
7713 	err = hw->phy.ops.identify_sfp(hw);
7714 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7715 		goto sfp_out;
7716 
7717 	if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
7718 		/* If no cable is present, then we need to reset
7719 		 * the next time we find a good cable. */
7720 		adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
7721 	}
7722 
7723 	/* exit on error */
7724 	if (err)
7725 		goto sfp_out;
7726 
7727 	/* exit if reset not needed */
7728 	if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7729 		goto sfp_out;
7730 
7731 	adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
7732 
7733 	/*
7734 	 * A module may be identified correctly, but the EEPROM may not have
7735 	 * support for that module.  setup_sfp() will fail in that case, so
7736 	 * we should not allow that module to load.
7737 	 */
7738 	if (hw->mac.type == ixgbe_mac_82598EB)
7739 		err = hw->phy.ops.reset(hw);
7740 	else
7741 		err = hw->mac.ops.setup_sfp(hw);
7742 
7743 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7744 		goto sfp_out;
7745 
7746 	adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
7747 	e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
7748 
7749 sfp_out:
7750 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7751 
7752 	if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
7753 	    (adapter->netdev->reg_state == NETREG_REGISTERED)) {
7754 		e_dev_err("failed to initialize because an unsupported "
7755 			  "SFP+ module type was detected.\n");
7756 		e_dev_err("Reload the driver after installing a "
7757 			  "supported module.\n");
7758 		unregister_netdev(adapter->netdev);
7759 	}
7760 }
7761 
7762 /**
7763  * ixgbe_sfp_link_config_subtask - set up link SFP after module install
7764  * @adapter: the ixgbe adapter structure
7765  **/
7766 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
7767 {
7768 	struct ixgbe_hw *hw = &adapter->hw;
7769 	u32 cap_speed;
7770 	u32 speed;
7771 	bool autoneg = false;
7772 
7773 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
7774 		return;
7775 
7776 	/* someone else is in init, wait until next service event */
7777 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7778 		return;
7779 
7780 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
7781 
7782 	hw->mac.ops.get_link_capabilities(hw, &cap_speed, &autoneg);
7783 
7784 	/* advertise highest capable link speed */
7785 	if (!autoneg && (cap_speed & IXGBE_LINK_SPEED_10GB_FULL))
7786 		speed = IXGBE_LINK_SPEED_10GB_FULL;
7787 	else
7788 		speed = cap_speed & (IXGBE_LINK_SPEED_10GB_FULL |
7789 				     IXGBE_LINK_SPEED_1GB_FULL);
7790 
7791 	if (hw->mac.ops.setup_link)
7792 		hw->mac.ops.setup_link(hw, speed, true);
7793 
7794 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
7795 	adapter->link_check_timeout = jiffies;
7796 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7797 }
7798 
7799 /**
7800  * ixgbe_service_timer - Timer Call-back
7801  * @t: pointer to timer_list structure
7802  **/
7803 static void ixgbe_service_timer(struct timer_list *t)
7804 {
7805 	struct ixgbe_adapter *adapter = from_timer(adapter, t, service_timer);
7806 	unsigned long next_event_offset;
7807 
7808 	/* poll faster when waiting for link */
7809 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
7810 		next_event_offset = HZ / 10;
7811 	else
7812 		next_event_offset = HZ * 2;
7813 
7814 	/* Reset the timer */
7815 	mod_timer(&adapter->service_timer, next_event_offset + jiffies);
7816 
7817 	ixgbe_service_event_schedule(adapter);
7818 }
7819 
7820 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter)
7821 {
7822 	struct ixgbe_hw *hw = &adapter->hw;
7823 	u32 status;
7824 
7825 	if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT))
7826 		return;
7827 
7828 	adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT;
7829 
7830 	if (!hw->phy.ops.handle_lasi)
7831 		return;
7832 
7833 	status = hw->phy.ops.handle_lasi(&adapter->hw);
7834 	if (status != IXGBE_ERR_OVERTEMP)
7835 		return;
7836 
7837 	e_crit(drv, "%s\n", ixgbe_overheat_msg);
7838 }
7839 
7840 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
7841 {
7842 	if (!test_and_clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state))
7843 		return;
7844 
7845 	rtnl_lock();
7846 	/* If we're already down, removing or resetting, just bail */
7847 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7848 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
7849 	    test_bit(__IXGBE_RESETTING, &adapter->state)) {
7850 		rtnl_unlock();
7851 		return;
7852 	}
7853 
7854 	ixgbe_dump(adapter);
7855 	netdev_err(adapter->netdev, "Reset adapter\n");
7856 	adapter->tx_timeout_count++;
7857 
7858 	ixgbe_reinit_locked(adapter);
7859 	rtnl_unlock();
7860 }
7861 
7862 /**
7863  * ixgbe_check_fw_error - Check firmware for errors
7864  * @adapter: the adapter private structure
7865  *
7866  * Check firmware errors in register FWSM
7867  */
7868 static bool ixgbe_check_fw_error(struct ixgbe_adapter *adapter)
7869 {
7870 	struct ixgbe_hw *hw = &adapter->hw;
7871 	u32 fwsm;
7872 
7873 	/* read fwsm.ext_err_ind register and log errors */
7874 	fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
7875 
7876 	if (fwsm & IXGBE_FWSM_EXT_ERR_IND_MASK ||
7877 	    !(fwsm & IXGBE_FWSM_FW_VAL_BIT))
7878 		e_dev_warn("Warning firmware error detected FWSM: 0x%08X\n",
7879 			   fwsm);
7880 
7881 	if (hw->mac.ops.fw_recovery_mode && hw->mac.ops.fw_recovery_mode(hw)) {
7882 		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");
7883 		return true;
7884 	}
7885 
7886 	return false;
7887 }
7888 
7889 /**
7890  * ixgbe_service_task - manages and runs subtasks
7891  * @work: pointer to work_struct containing our data
7892  **/
7893 static void ixgbe_service_task(struct work_struct *work)
7894 {
7895 	struct ixgbe_adapter *adapter = container_of(work,
7896 						     struct ixgbe_adapter,
7897 						     service_task);
7898 	if (ixgbe_removed(adapter->hw.hw_addr)) {
7899 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
7900 			rtnl_lock();
7901 			ixgbe_down(adapter);
7902 			rtnl_unlock();
7903 		}
7904 		ixgbe_service_event_complete(adapter);
7905 		return;
7906 	}
7907 	if (ixgbe_check_fw_error(adapter)) {
7908 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
7909 			unregister_netdev(adapter->netdev);
7910 		ixgbe_service_event_complete(adapter);
7911 		return;
7912 	}
7913 	ixgbe_reset_subtask(adapter);
7914 	ixgbe_phy_interrupt_subtask(adapter);
7915 	ixgbe_sfp_detection_subtask(adapter);
7916 	ixgbe_sfp_link_config_subtask(adapter);
7917 	ixgbe_check_overtemp_subtask(adapter);
7918 	ixgbe_watchdog_subtask(adapter);
7919 	ixgbe_fdir_reinit_subtask(adapter);
7920 	ixgbe_check_hang_subtask(adapter);
7921 
7922 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
7923 		ixgbe_ptp_overflow_check(adapter);
7924 		if (adapter->flags & IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER)
7925 			ixgbe_ptp_rx_hang(adapter);
7926 		ixgbe_ptp_tx_hang(adapter);
7927 	}
7928 
7929 	ixgbe_service_event_complete(adapter);
7930 }
7931 
7932 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
7933 		     struct ixgbe_tx_buffer *first,
7934 		     u8 *hdr_len,
7935 		     struct ixgbe_ipsec_tx_data *itd)
7936 {
7937 	u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
7938 	struct sk_buff *skb = first->skb;
7939 	union {
7940 		struct iphdr *v4;
7941 		struct ipv6hdr *v6;
7942 		unsigned char *hdr;
7943 	} ip;
7944 	union {
7945 		struct tcphdr *tcp;
7946 		struct udphdr *udp;
7947 		unsigned char *hdr;
7948 	} l4;
7949 	u32 paylen, l4_offset;
7950 	u32 fceof_saidx = 0;
7951 	int err;
7952 
7953 	if (skb->ip_summed != CHECKSUM_PARTIAL)
7954 		return 0;
7955 
7956 	if (!skb_is_gso(skb))
7957 		return 0;
7958 
7959 	err = skb_cow_head(skb, 0);
7960 	if (err < 0)
7961 		return err;
7962 
7963 	if (eth_p_mpls(first->protocol))
7964 		ip.hdr = skb_inner_network_header(skb);
7965 	else
7966 		ip.hdr = skb_network_header(skb);
7967 	l4.hdr = skb_checksum_start(skb);
7968 
7969 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
7970 	type_tucmd = (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ?
7971 		      IXGBE_ADVTXD_TUCMD_L4T_UDP : IXGBE_ADVTXD_TUCMD_L4T_TCP;
7972 
7973 	/* initialize outer IP header fields */
7974 	if (ip.v4->version == 4) {
7975 		unsigned char *csum_start = skb_checksum_start(skb);
7976 		unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
7977 		int len = csum_start - trans_start;
7978 
7979 		/* IP header will have to cancel out any data that
7980 		 * is not a part of the outer IP header, so set to
7981 		 * a reverse csum if needed, else init check to 0.
7982 		 */
7983 		ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
7984 					   csum_fold(csum_partial(trans_start,
7985 								  len, 0)) : 0;
7986 		type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
7987 
7988 		ip.v4->tot_len = 0;
7989 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
7990 				   IXGBE_TX_FLAGS_CSUM |
7991 				   IXGBE_TX_FLAGS_IPV4;
7992 	} else {
7993 		ip.v6->payload_len = 0;
7994 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
7995 				   IXGBE_TX_FLAGS_CSUM;
7996 	}
7997 
7998 	/* determine offset of inner transport header */
7999 	l4_offset = l4.hdr - skb->data;
8000 
8001 	/* remove payload length from inner checksum */
8002 	paylen = skb->len - l4_offset;
8003 
8004 	if (type_tucmd & IXGBE_ADVTXD_TUCMD_L4T_TCP) {
8005 		/* compute length of segmentation header */
8006 		*hdr_len = (l4.tcp->doff * 4) + l4_offset;
8007 		csum_replace_by_diff(&l4.tcp->check,
8008 				     (__force __wsum)htonl(paylen));
8009 	} else {
8010 		/* compute length of segmentation header */
8011 		*hdr_len = sizeof(*l4.udp) + l4_offset;
8012 		csum_replace_by_diff(&l4.udp->check,
8013 				     (__force __wsum)htonl(paylen));
8014 	}
8015 
8016 	/* update gso size and bytecount with header size */
8017 	first->gso_segs = skb_shinfo(skb)->gso_segs;
8018 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
8019 
8020 	/* mss_l4len_id: use 0 as index for TSO */
8021 	mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
8022 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
8023 
8024 	fceof_saidx |= itd->sa_idx;
8025 	type_tucmd |= itd->flags | itd->trailer_len;
8026 
8027 	/* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
8028 	vlan_macip_lens = l4.hdr - ip.hdr;
8029 	vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
8030 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8031 
8032 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
8033 			  mss_l4len_idx);
8034 
8035 	return 1;
8036 }
8037 
8038 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
8039 			  struct ixgbe_tx_buffer *first,
8040 			  struct ixgbe_ipsec_tx_data *itd)
8041 {
8042 	struct sk_buff *skb = first->skb;
8043 	u32 vlan_macip_lens = 0;
8044 	u32 fceof_saidx = 0;
8045 	u32 type_tucmd = 0;
8046 
8047 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
8048 csum_failed:
8049 		if (!(first->tx_flags & (IXGBE_TX_FLAGS_HW_VLAN |
8050 					 IXGBE_TX_FLAGS_CC)))
8051 			return;
8052 		goto no_csum;
8053 	}
8054 
8055 	switch (skb->csum_offset) {
8056 	case offsetof(struct tcphdr, check):
8057 		type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
8058 		fallthrough;
8059 	case offsetof(struct udphdr, check):
8060 		break;
8061 	case offsetof(struct sctphdr, checksum):
8062 		/* validate that this is actually an SCTP request */
8063 		if (skb_csum_is_sctp(skb)) {
8064 			type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
8065 			break;
8066 		}
8067 		fallthrough;
8068 	default:
8069 		skb_checksum_help(skb);
8070 		goto csum_failed;
8071 	}
8072 
8073 	/* update TX checksum flag */
8074 	first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
8075 	vlan_macip_lens = skb_checksum_start_offset(skb) -
8076 			  skb_network_offset(skb);
8077 no_csum:
8078 	/* vlan_macip_lens: MACLEN, VLAN tag */
8079 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
8080 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8081 
8082 	fceof_saidx |= itd->sa_idx;
8083 	type_tucmd |= itd->flags | itd->trailer_len;
8084 
8085 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd, 0);
8086 }
8087 
8088 #define IXGBE_SET_FLAG(_input, _flag, _result) \
8089 	((_flag <= _result) ? \
8090 	 ((u32)(_input & _flag) * (_result / _flag)) : \
8091 	 ((u32)(_input & _flag) / (_flag / _result)))
8092 
8093 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
8094 {
8095 	/* set type for advanced descriptor with frame checksum insertion */
8096 	u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8097 		       IXGBE_ADVTXD_DCMD_DEXT |
8098 		       IXGBE_ADVTXD_DCMD_IFCS;
8099 
8100 	/* set HW vlan bit if vlan is present */
8101 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
8102 				   IXGBE_ADVTXD_DCMD_VLE);
8103 
8104 	/* set segmentation enable bits for TSO/FSO */
8105 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
8106 				   IXGBE_ADVTXD_DCMD_TSE);
8107 
8108 	/* set timestamp bit if present */
8109 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
8110 				   IXGBE_ADVTXD_MAC_TSTAMP);
8111 
8112 	/* insert frame checksum */
8113 	cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
8114 
8115 	return cmd_type;
8116 }
8117 
8118 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
8119 				   u32 tx_flags, unsigned int paylen)
8120 {
8121 	u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
8122 
8123 	/* enable L4 checksum for TSO and TX checksum offload */
8124 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8125 					IXGBE_TX_FLAGS_CSUM,
8126 					IXGBE_ADVTXD_POPTS_TXSM);
8127 
8128 	/* enable IPv4 checksum for TSO */
8129 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8130 					IXGBE_TX_FLAGS_IPV4,
8131 					IXGBE_ADVTXD_POPTS_IXSM);
8132 
8133 	/* enable IPsec */
8134 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8135 					IXGBE_TX_FLAGS_IPSEC,
8136 					IXGBE_ADVTXD_POPTS_IPSEC);
8137 
8138 	/*
8139 	 * Check Context must be set if Tx switch is enabled, which it
8140 	 * always is for case where virtual functions are running
8141 	 */
8142 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8143 					IXGBE_TX_FLAGS_CC,
8144 					IXGBE_ADVTXD_CC);
8145 
8146 	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
8147 }
8148 
8149 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8150 {
8151 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
8152 
8153 	/* Herbert's original patch had:
8154 	 *  smp_mb__after_netif_stop_queue();
8155 	 * but since that doesn't exist yet, just open code it.
8156 	 */
8157 	smp_mb();
8158 
8159 	/* We need to check again in a case another CPU has just
8160 	 * made room available.
8161 	 */
8162 	if (likely(ixgbe_desc_unused(tx_ring) < size))
8163 		return -EBUSY;
8164 
8165 	/* A reprieve! - use start_queue because it doesn't call schedule */
8166 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
8167 	++tx_ring->tx_stats.restart_queue;
8168 	return 0;
8169 }
8170 
8171 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8172 {
8173 	if (likely(ixgbe_desc_unused(tx_ring) >= size))
8174 		return 0;
8175 
8176 	return __ixgbe_maybe_stop_tx(tx_ring, size);
8177 }
8178 
8179 static int ixgbe_tx_map(struct ixgbe_ring *tx_ring,
8180 			struct ixgbe_tx_buffer *first,
8181 			const u8 hdr_len)
8182 {
8183 	struct sk_buff *skb = first->skb;
8184 	struct ixgbe_tx_buffer *tx_buffer;
8185 	union ixgbe_adv_tx_desc *tx_desc;
8186 	skb_frag_t *frag;
8187 	dma_addr_t dma;
8188 	unsigned int data_len, size;
8189 	u32 tx_flags = first->tx_flags;
8190 	u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
8191 	u16 i = tx_ring->next_to_use;
8192 
8193 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
8194 
8195 	ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
8196 
8197 	size = skb_headlen(skb);
8198 	data_len = skb->data_len;
8199 
8200 #ifdef IXGBE_FCOE
8201 	if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
8202 		if (data_len < sizeof(struct fcoe_crc_eof)) {
8203 			size -= sizeof(struct fcoe_crc_eof) - data_len;
8204 			data_len = 0;
8205 		} else {
8206 			data_len -= sizeof(struct fcoe_crc_eof);
8207 		}
8208 	}
8209 
8210 #endif
8211 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
8212 
8213 	tx_buffer = first;
8214 
8215 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
8216 		if (dma_mapping_error(tx_ring->dev, dma))
8217 			goto dma_error;
8218 
8219 		/* record length, and DMA address */
8220 		dma_unmap_len_set(tx_buffer, len, size);
8221 		dma_unmap_addr_set(tx_buffer, dma, dma);
8222 
8223 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
8224 
8225 		while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
8226 			tx_desc->read.cmd_type_len =
8227 				cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
8228 
8229 			i++;
8230 			tx_desc++;
8231 			if (i == tx_ring->count) {
8232 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8233 				i = 0;
8234 			}
8235 			tx_desc->read.olinfo_status = 0;
8236 
8237 			dma += IXGBE_MAX_DATA_PER_TXD;
8238 			size -= IXGBE_MAX_DATA_PER_TXD;
8239 
8240 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
8241 		}
8242 
8243 		if (likely(!data_len))
8244 			break;
8245 
8246 		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
8247 
8248 		i++;
8249 		tx_desc++;
8250 		if (i == tx_ring->count) {
8251 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8252 			i = 0;
8253 		}
8254 		tx_desc->read.olinfo_status = 0;
8255 
8256 #ifdef IXGBE_FCOE
8257 		size = min_t(unsigned int, data_len, skb_frag_size(frag));
8258 #else
8259 		size = skb_frag_size(frag);
8260 #endif
8261 		data_len -= size;
8262 
8263 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
8264 				       DMA_TO_DEVICE);
8265 
8266 		tx_buffer = &tx_ring->tx_buffer_info[i];
8267 	}
8268 
8269 	/* write last descriptor with RS and EOP bits */
8270 	cmd_type |= size | IXGBE_TXD_CMD;
8271 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8272 
8273 	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
8274 
8275 	/* set the timestamp */
8276 	first->time_stamp = jiffies;
8277 
8278 	skb_tx_timestamp(skb);
8279 
8280 	/*
8281 	 * Force memory writes to complete before letting h/w know there
8282 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
8283 	 * memory model archs, such as IA-64).
8284 	 *
8285 	 * We also need this memory barrier to make certain all of the
8286 	 * status bits have been updated before next_to_watch is written.
8287 	 */
8288 	wmb();
8289 
8290 	/* set next_to_watch value indicating a packet is present */
8291 	first->next_to_watch = tx_desc;
8292 
8293 	i++;
8294 	if (i == tx_ring->count)
8295 		i = 0;
8296 
8297 	tx_ring->next_to_use = i;
8298 
8299 	ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
8300 
8301 	if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
8302 		writel(i, tx_ring->tail);
8303 	}
8304 
8305 	return 0;
8306 dma_error:
8307 	dev_err(tx_ring->dev, "TX DMA map failed\n");
8308 
8309 	/* clear dma mappings for failed tx_buffer_info map */
8310 	for (;;) {
8311 		tx_buffer = &tx_ring->tx_buffer_info[i];
8312 		if (dma_unmap_len(tx_buffer, len))
8313 			dma_unmap_page(tx_ring->dev,
8314 				       dma_unmap_addr(tx_buffer, dma),
8315 				       dma_unmap_len(tx_buffer, len),
8316 				       DMA_TO_DEVICE);
8317 		dma_unmap_len_set(tx_buffer, len, 0);
8318 		if (tx_buffer == first)
8319 			break;
8320 		if (i == 0)
8321 			i += tx_ring->count;
8322 		i--;
8323 	}
8324 
8325 	dev_kfree_skb_any(first->skb);
8326 	first->skb = NULL;
8327 
8328 	tx_ring->next_to_use = i;
8329 
8330 	return -1;
8331 }
8332 
8333 static void ixgbe_atr(struct ixgbe_ring *ring,
8334 		      struct ixgbe_tx_buffer *first)
8335 {
8336 	struct ixgbe_q_vector *q_vector = ring->q_vector;
8337 	union ixgbe_atr_hash_dword input = { .dword = 0 };
8338 	union ixgbe_atr_hash_dword common = { .dword = 0 };
8339 	union {
8340 		unsigned char *network;
8341 		struct iphdr *ipv4;
8342 		struct ipv6hdr *ipv6;
8343 	} hdr;
8344 	struct tcphdr *th;
8345 	unsigned int hlen;
8346 	struct sk_buff *skb;
8347 	__be16 vlan_id;
8348 	int l4_proto;
8349 
8350 	/* if ring doesn't have a interrupt vector, cannot perform ATR */
8351 	if (!q_vector)
8352 		return;
8353 
8354 	/* do nothing if sampling is disabled */
8355 	if (!ring->atr_sample_rate)
8356 		return;
8357 
8358 	ring->atr_count++;
8359 
8360 	/* currently only IPv4/IPv6 with TCP is supported */
8361 	if ((first->protocol != htons(ETH_P_IP)) &&
8362 	    (first->protocol != htons(ETH_P_IPV6)))
8363 		return;
8364 
8365 	/* snag network header to get L4 type and address */
8366 	skb = first->skb;
8367 	hdr.network = skb_network_header(skb);
8368 	if (unlikely(hdr.network <= skb->data))
8369 		return;
8370 	if (skb->encapsulation &&
8371 	    first->protocol == htons(ETH_P_IP) &&
8372 	    hdr.ipv4->protocol == IPPROTO_UDP) {
8373 		struct ixgbe_adapter *adapter = q_vector->adapter;
8374 
8375 		if (unlikely(skb_tail_pointer(skb) < hdr.network +
8376 			     VXLAN_HEADROOM))
8377 			return;
8378 
8379 		/* verify the port is recognized as VXLAN */
8380 		if (adapter->vxlan_port &&
8381 		    udp_hdr(skb)->dest == adapter->vxlan_port)
8382 			hdr.network = skb_inner_network_header(skb);
8383 
8384 		if (adapter->geneve_port &&
8385 		    udp_hdr(skb)->dest == adapter->geneve_port)
8386 			hdr.network = skb_inner_network_header(skb);
8387 	}
8388 
8389 	/* Make sure we have at least [minimum IPv4 header + TCP]
8390 	 * or [IPv6 header] bytes
8391 	 */
8392 	if (unlikely(skb_tail_pointer(skb) < hdr.network + 40))
8393 		return;
8394 
8395 	/* Currently only IPv4/IPv6 with TCP is supported */
8396 	switch (hdr.ipv4->version) {
8397 	case IPVERSION:
8398 		/* access ihl as u8 to avoid unaligned access on ia64 */
8399 		hlen = (hdr.network[0] & 0x0F) << 2;
8400 		l4_proto = hdr.ipv4->protocol;
8401 		break;
8402 	case 6:
8403 		hlen = hdr.network - skb->data;
8404 		l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
8405 		hlen -= hdr.network - skb->data;
8406 		break;
8407 	default:
8408 		return;
8409 	}
8410 
8411 	if (l4_proto != IPPROTO_TCP)
8412 		return;
8413 
8414 	if (unlikely(skb_tail_pointer(skb) < hdr.network +
8415 		     hlen + sizeof(struct tcphdr)))
8416 		return;
8417 
8418 	th = (struct tcphdr *)(hdr.network + hlen);
8419 
8420 	/* skip this packet since the socket is closing */
8421 	if (th->fin)
8422 		return;
8423 
8424 	/* sample on all syn packets or once every atr sample count */
8425 	if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
8426 		return;
8427 
8428 	/* reset sample count */
8429 	ring->atr_count = 0;
8430 
8431 	vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
8432 
8433 	/*
8434 	 * src and dst are inverted, think how the receiver sees them
8435 	 *
8436 	 * The input is broken into two sections, a non-compressed section
8437 	 * containing vm_pool, vlan_id, and flow_type.  The rest of the data
8438 	 * is XORed together and stored in the compressed dword.
8439 	 */
8440 	input.formatted.vlan_id = vlan_id;
8441 
8442 	/*
8443 	 * since src port and flex bytes occupy the same word XOR them together
8444 	 * and write the value to source port portion of compressed dword
8445 	 */
8446 	if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
8447 		common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
8448 	else
8449 		common.port.src ^= th->dest ^ first->protocol;
8450 	common.port.dst ^= th->source;
8451 
8452 	switch (hdr.ipv4->version) {
8453 	case IPVERSION:
8454 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
8455 		common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
8456 		break;
8457 	case 6:
8458 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
8459 		common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
8460 			     hdr.ipv6->saddr.s6_addr32[1] ^
8461 			     hdr.ipv6->saddr.s6_addr32[2] ^
8462 			     hdr.ipv6->saddr.s6_addr32[3] ^
8463 			     hdr.ipv6->daddr.s6_addr32[0] ^
8464 			     hdr.ipv6->daddr.s6_addr32[1] ^
8465 			     hdr.ipv6->daddr.s6_addr32[2] ^
8466 			     hdr.ipv6->daddr.s6_addr32[3];
8467 		break;
8468 	default:
8469 		break;
8470 	}
8471 
8472 	if (hdr.network != skb_network_header(skb))
8473 		input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
8474 
8475 	/* This assumes the Rx queue and Tx queue are bound to the same CPU */
8476 	ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
8477 					      input, common, ring->queue_index);
8478 }
8479 
8480 #ifdef IXGBE_FCOE
8481 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
8482 			      struct net_device *sb_dev)
8483 {
8484 	struct ixgbe_adapter *adapter;
8485 	struct ixgbe_ring_feature *f;
8486 	int txq;
8487 
8488 	if (sb_dev) {
8489 		u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
8490 		struct net_device *vdev = sb_dev;
8491 
8492 		txq = vdev->tc_to_txq[tc].offset;
8493 		txq += reciprocal_scale(skb_get_hash(skb),
8494 					vdev->tc_to_txq[tc].count);
8495 
8496 		return txq;
8497 	}
8498 
8499 	/*
8500 	 * only execute the code below if protocol is FCoE
8501 	 * or FIP and we have FCoE enabled on the adapter
8502 	 */
8503 	switch (vlan_get_protocol(skb)) {
8504 	case htons(ETH_P_FCOE):
8505 	case htons(ETH_P_FIP):
8506 		adapter = netdev_priv(dev);
8507 
8508 		if (!sb_dev && (adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
8509 			break;
8510 		fallthrough;
8511 	default:
8512 		return netdev_pick_tx(dev, skb, sb_dev);
8513 	}
8514 
8515 	f = &adapter->ring_feature[RING_F_FCOE];
8516 
8517 	txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
8518 					   smp_processor_id();
8519 
8520 	while (txq >= f->indices)
8521 		txq -= f->indices;
8522 
8523 	return txq + f->offset;
8524 }
8525 
8526 #endif
8527 int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
8528 			struct xdp_frame *xdpf)
8529 {
8530 	struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
8531 	struct ixgbe_tx_buffer *tx_buffer;
8532 	union ixgbe_adv_tx_desc *tx_desc;
8533 	u32 len, cmd_type;
8534 	dma_addr_t dma;
8535 	u16 i;
8536 
8537 	len = xdpf->len;
8538 
8539 	if (unlikely(!ixgbe_desc_unused(ring)))
8540 		return IXGBE_XDP_CONSUMED;
8541 
8542 	dma = dma_map_single(ring->dev, xdpf->data, len, DMA_TO_DEVICE);
8543 	if (dma_mapping_error(ring->dev, dma))
8544 		return IXGBE_XDP_CONSUMED;
8545 
8546 	/* record the location of the first descriptor for this packet */
8547 	tx_buffer = &ring->tx_buffer_info[ring->next_to_use];
8548 	tx_buffer->bytecount = len;
8549 	tx_buffer->gso_segs = 1;
8550 	tx_buffer->protocol = 0;
8551 
8552 	i = ring->next_to_use;
8553 	tx_desc = IXGBE_TX_DESC(ring, i);
8554 
8555 	dma_unmap_len_set(tx_buffer, len, len);
8556 	dma_unmap_addr_set(tx_buffer, dma, dma);
8557 	tx_buffer->xdpf = xdpf;
8558 
8559 	tx_desc->read.buffer_addr = cpu_to_le64(dma);
8560 
8561 	/* put descriptor type bits */
8562 	cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8563 		   IXGBE_ADVTXD_DCMD_DEXT |
8564 		   IXGBE_ADVTXD_DCMD_IFCS;
8565 	cmd_type |= len | IXGBE_TXD_CMD;
8566 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8567 	tx_desc->read.olinfo_status =
8568 		cpu_to_le32(len << IXGBE_ADVTXD_PAYLEN_SHIFT);
8569 
8570 	/* Avoid any potential race with xdp_xmit and cleanup */
8571 	smp_wmb();
8572 
8573 	/* set next_to_watch value indicating a packet is present */
8574 	i++;
8575 	if (i == ring->count)
8576 		i = 0;
8577 
8578 	tx_buffer->next_to_watch = tx_desc;
8579 	ring->next_to_use = i;
8580 
8581 	return IXGBE_XDP_TX;
8582 }
8583 
8584 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
8585 			  struct ixgbe_adapter *adapter,
8586 			  struct ixgbe_ring *tx_ring)
8587 {
8588 	struct ixgbe_tx_buffer *first;
8589 	int tso;
8590 	u32 tx_flags = 0;
8591 	unsigned short f;
8592 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
8593 	struct ixgbe_ipsec_tx_data ipsec_tx = { 0 };
8594 	__be16 protocol = skb->protocol;
8595 	u8 hdr_len = 0;
8596 
8597 	/*
8598 	 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
8599 	 *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
8600 	 *       + 2 desc gap to keep tail from touching head,
8601 	 *       + 1 desc for context descriptor,
8602 	 * otherwise try next time
8603 	 */
8604 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
8605 		count += TXD_USE_COUNT(skb_frag_size(
8606 						&skb_shinfo(skb)->frags[f]));
8607 
8608 	if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
8609 		tx_ring->tx_stats.tx_busy++;
8610 		return NETDEV_TX_BUSY;
8611 	}
8612 
8613 	/* record the location of the first descriptor for this packet */
8614 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
8615 	first->skb = skb;
8616 	first->bytecount = skb->len;
8617 	first->gso_segs = 1;
8618 
8619 	/* if we have a HW VLAN tag being added default to the HW one */
8620 	if (skb_vlan_tag_present(skb)) {
8621 		tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
8622 		tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8623 	/* else if it is a SW VLAN check the next protocol and store the tag */
8624 	} else if (protocol == htons(ETH_P_8021Q)) {
8625 		struct vlan_hdr *vhdr, _vhdr;
8626 		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
8627 		if (!vhdr)
8628 			goto out_drop;
8629 
8630 		tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
8631 				  IXGBE_TX_FLAGS_VLAN_SHIFT;
8632 		tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
8633 	}
8634 	protocol = vlan_get_protocol(skb);
8635 
8636 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
8637 	    adapter->ptp_clock) {
8638 		if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
8639 		    !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
8640 					   &adapter->state)) {
8641 			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
8642 			tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
8643 
8644 			/* schedule check for Tx timestamp */
8645 			adapter->ptp_tx_skb = skb_get(skb);
8646 			adapter->ptp_tx_start = jiffies;
8647 			schedule_work(&adapter->ptp_tx_work);
8648 		} else {
8649 			adapter->tx_hwtstamp_skipped++;
8650 		}
8651 	}
8652 
8653 #ifdef CONFIG_PCI_IOV
8654 	/*
8655 	 * Use the l2switch_enable flag - would be false if the DMA
8656 	 * Tx switch had been disabled.
8657 	 */
8658 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
8659 		tx_flags |= IXGBE_TX_FLAGS_CC;
8660 
8661 #endif
8662 	/* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
8663 	if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8664 	    ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
8665 	     (skb->priority != TC_PRIO_CONTROL))) {
8666 		tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
8667 		tx_flags |= (skb->priority & 0x7) <<
8668 					IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
8669 		if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
8670 			struct vlan_ethhdr *vhdr;
8671 
8672 			if (skb_cow_head(skb, 0))
8673 				goto out_drop;
8674 			vhdr = (struct vlan_ethhdr *)skb->data;
8675 			vhdr->h_vlan_TCI = htons(tx_flags >>
8676 						 IXGBE_TX_FLAGS_VLAN_SHIFT);
8677 		} else {
8678 			tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8679 		}
8680 	}
8681 
8682 	/* record initial flags and protocol */
8683 	first->tx_flags = tx_flags;
8684 	first->protocol = protocol;
8685 
8686 #ifdef IXGBE_FCOE
8687 	/* setup tx offload for FCoE */
8688 	if ((protocol == htons(ETH_P_FCOE)) &&
8689 	    (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
8690 		tso = ixgbe_fso(tx_ring, first, &hdr_len);
8691 		if (tso < 0)
8692 			goto out_drop;
8693 
8694 		goto xmit_fcoe;
8695 	}
8696 
8697 #endif /* IXGBE_FCOE */
8698 
8699 #ifdef CONFIG_IXGBE_IPSEC
8700 	if (xfrm_offload(skb) &&
8701 	    !ixgbe_ipsec_tx(tx_ring, first, &ipsec_tx))
8702 		goto out_drop;
8703 #endif
8704 	tso = ixgbe_tso(tx_ring, first, &hdr_len, &ipsec_tx);
8705 	if (tso < 0)
8706 		goto out_drop;
8707 	else if (!tso)
8708 		ixgbe_tx_csum(tx_ring, first, &ipsec_tx);
8709 
8710 	/* add the ATR filter if ATR is on */
8711 	if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
8712 		ixgbe_atr(tx_ring, first);
8713 
8714 #ifdef IXGBE_FCOE
8715 xmit_fcoe:
8716 #endif /* IXGBE_FCOE */
8717 	if (ixgbe_tx_map(tx_ring, first, hdr_len))
8718 		goto cleanup_tx_timestamp;
8719 
8720 	return NETDEV_TX_OK;
8721 
8722 out_drop:
8723 	dev_kfree_skb_any(first->skb);
8724 	first->skb = NULL;
8725 cleanup_tx_timestamp:
8726 	if (unlikely(tx_flags & IXGBE_TX_FLAGS_TSTAMP)) {
8727 		dev_kfree_skb_any(adapter->ptp_tx_skb);
8728 		adapter->ptp_tx_skb = NULL;
8729 		cancel_work_sync(&adapter->ptp_tx_work);
8730 		clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS, &adapter->state);
8731 	}
8732 
8733 	return NETDEV_TX_OK;
8734 }
8735 
8736 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
8737 				      struct net_device *netdev,
8738 				      struct ixgbe_ring *ring)
8739 {
8740 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8741 	struct ixgbe_ring *tx_ring;
8742 
8743 	/*
8744 	 * The minimum packet size for olinfo paylen is 17 so pad the skb
8745 	 * in order to meet this minimum size requirement.
8746 	 */
8747 	if (skb_put_padto(skb, 17))
8748 		return NETDEV_TX_OK;
8749 
8750 	tx_ring = ring ? ring : adapter->tx_ring[skb_get_queue_mapping(skb)];
8751 	if (unlikely(test_bit(__IXGBE_TX_DISABLED, &tx_ring->state)))
8752 		return NETDEV_TX_BUSY;
8753 
8754 	return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
8755 }
8756 
8757 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
8758 				    struct net_device *netdev)
8759 {
8760 	return __ixgbe_xmit_frame(skb, netdev, NULL);
8761 }
8762 
8763 /**
8764  * ixgbe_set_mac - Change the Ethernet Address of the NIC
8765  * @netdev: network interface device structure
8766  * @p: pointer to an address structure
8767  *
8768  * Returns 0 on success, negative on failure
8769  **/
8770 static int ixgbe_set_mac(struct net_device *netdev, void *p)
8771 {
8772 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8773 	struct ixgbe_hw *hw = &adapter->hw;
8774 	struct sockaddr *addr = p;
8775 
8776 	if (!is_valid_ether_addr(addr->sa_data))
8777 		return -EADDRNOTAVAIL;
8778 
8779 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
8780 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
8781 
8782 	ixgbe_mac_set_default_filter(adapter);
8783 
8784 	return 0;
8785 }
8786 
8787 static int
8788 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
8789 {
8790 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8791 	struct ixgbe_hw *hw = &adapter->hw;
8792 	u16 value;
8793 	int rc;
8794 
8795 	if (adapter->mii_bus) {
8796 		int regnum = addr;
8797 
8798 		if (devad != MDIO_DEVAD_NONE)
8799 			regnum |= (devad << 16) | MII_ADDR_C45;
8800 
8801 		return mdiobus_read(adapter->mii_bus, prtad, regnum);
8802 	}
8803 
8804 	if (prtad != hw->phy.mdio.prtad)
8805 		return -EINVAL;
8806 	rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
8807 	if (!rc)
8808 		rc = value;
8809 	return rc;
8810 }
8811 
8812 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
8813 			    u16 addr, u16 value)
8814 {
8815 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8816 	struct ixgbe_hw *hw = &adapter->hw;
8817 
8818 	if (adapter->mii_bus) {
8819 		int regnum = addr;
8820 
8821 		if (devad != MDIO_DEVAD_NONE)
8822 			regnum |= (devad << 16) | MII_ADDR_C45;
8823 
8824 		return mdiobus_write(adapter->mii_bus, prtad, regnum, value);
8825 	}
8826 
8827 	if (prtad != hw->phy.mdio.prtad)
8828 		return -EINVAL;
8829 	return hw->phy.ops.write_reg(hw, addr, devad, value);
8830 }
8831 
8832 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
8833 {
8834 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8835 
8836 	switch (cmd) {
8837 	case SIOCSHWTSTAMP:
8838 		return ixgbe_ptp_set_ts_config(adapter, req);
8839 	case SIOCGHWTSTAMP:
8840 		return ixgbe_ptp_get_ts_config(adapter, req);
8841 	case SIOCGMIIPHY:
8842 		if (!adapter->hw.phy.ops.read_reg)
8843 			return -EOPNOTSUPP;
8844 		fallthrough;
8845 	default:
8846 		return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
8847 	}
8848 }
8849 
8850 /**
8851  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
8852  * netdev->dev_addrs
8853  * @dev: network interface device structure
8854  *
8855  * Returns non-zero on failure
8856  **/
8857 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
8858 {
8859 	int err = 0;
8860 	struct ixgbe_adapter *adapter = netdev_priv(dev);
8861 	struct ixgbe_hw *hw = &adapter->hw;
8862 
8863 	if (is_valid_ether_addr(hw->mac.san_addr)) {
8864 		rtnl_lock();
8865 		err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
8866 		rtnl_unlock();
8867 
8868 		/* update SAN MAC vmdq pool selection */
8869 		hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
8870 	}
8871 	return err;
8872 }
8873 
8874 /**
8875  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
8876  * netdev->dev_addrs
8877  * @dev: network interface device structure
8878  *
8879  * Returns non-zero on failure
8880  **/
8881 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
8882 {
8883 	int err = 0;
8884 	struct ixgbe_adapter *adapter = netdev_priv(dev);
8885 	struct ixgbe_mac_info *mac = &adapter->hw.mac;
8886 
8887 	if (is_valid_ether_addr(mac->san_addr)) {
8888 		rtnl_lock();
8889 		err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
8890 		rtnl_unlock();
8891 	}
8892 	return err;
8893 }
8894 
8895 static void ixgbe_get_ring_stats64(struct rtnl_link_stats64 *stats,
8896 				   struct ixgbe_ring *ring)
8897 {
8898 	u64 bytes, packets;
8899 	unsigned int start;
8900 
8901 	if (ring) {
8902 		do {
8903 			start = u64_stats_fetch_begin_irq(&ring->syncp);
8904 			packets = ring->stats.packets;
8905 			bytes   = ring->stats.bytes;
8906 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8907 		stats->tx_packets += packets;
8908 		stats->tx_bytes   += bytes;
8909 	}
8910 }
8911 
8912 static void ixgbe_get_stats64(struct net_device *netdev,
8913 			      struct rtnl_link_stats64 *stats)
8914 {
8915 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8916 	int i;
8917 
8918 	rcu_read_lock();
8919 	for (i = 0; i < adapter->num_rx_queues; i++) {
8920 		struct ixgbe_ring *ring = READ_ONCE(adapter->rx_ring[i]);
8921 		u64 bytes, packets;
8922 		unsigned int start;
8923 
8924 		if (ring) {
8925 			do {
8926 				start = u64_stats_fetch_begin_irq(&ring->syncp);
8927 				packets = ring->stats.packets;
8928 				bytes   = ring->stats.bytes;
8929 			} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8930 			stats->rx_packets += packets;
8931 			stats->rx_bytes   += bytes;
8932 		}
8933 	}
8934 
8935 	for (i = 0; i < adapter->num_tx_queues; i++) {
8936 		struct ixgbe_ring *ring = READ_ONCE(adapter->tx_ring[i]);
8937 
8938 		ixgbe_get_ring_stats64(stats, ring);
8939 	}
8940 	for (i = 0; i < adapter->num_xdp_queues; i++) {
8941 		struct ixgbe_ring *ring = READ_ONCE(adapter->xdp_ring[i]);
8942 
8943 		ixgbe_get_ring_stats64(stats, ring);
8944 	}
8945 	rcu_read_unlock();
8946 
8947 	/* following stats updated by ixgbe_watchdog_task() */
8948 	stats->multicast	= netdev->stats.multicast;
8949 	stats->rx_errors	= netdev->stats.rx_errors;
8950 	stats->rx_length_errors	= netdev->stats.rx_length_errors;
8951 	stats->rx_crc_errors	= netdev->stats.rx_crc_errors;
8952 	stats->rx_missed_errors	= netdev->stats.rx_missed_errors;
8953 }
8954 
8955 #ifdef CONFIG_IXGBE_DCB
8956 /**
8957  * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
8958  * @adapter: pointer to ixgbe_adapter
8959  * @tc: number of traffic classes currently enabled
8960  *
8961  * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
8962  * 802.1Q priority maps to a packet buffer that exists.
8963  */
8964 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
8965 {
8966 	struct ixgbe_hw *hw = &adapter->hw;
8967 	u32 reg, rsave;
8968 	int i;
8969 
8970 	/* 82598 have a static priority to TC mapping that can not
8971 	 * be changed so no validation is needed.
8972 	 */
8973 	if (hw->mac.type == ixgbe_mac_82598EB)
8974 		return;
8975 
8976 	reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
8977 	rsave = reg;
8978 
8979 	for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
8980 		u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
8981 
8982 		/* If up2tc is out of bounds default to zero */
8983 		if (up2tc > tc)
8984 			reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
8985 	}
8986 
8987 	if (reg != rsave)
8988 		IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
8989 
8990 	return;
8991 }
8992 
8993 /**
8994  * ixgbe_set_prio_tc_map - Configure netdev prio tc map
8995  * @adapter: Pointer to adapter struct
8996  *
8997  * Populate the netdev user priority to tc map
8998  */
8999 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
9000 {
9001 	struct net_device *dev = adapter->netdev;
9002 	struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
9003 	struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
9004 	u8 prio;
9005 
9006 	for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
9007 		u8 tc = 0;
9008 
9009 		if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
9010 			tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
9011 		else if (ets)
9012 			tc = ets->prio_tc[prio];
9013 
9014 		netdev_set_prio_tc_map(dev, prio, tc);
9015 	}
9016 }
9017 
9018 #endif /* CONFIG_IXGBE_DCB */
9019 static int ixgbe_reassign_macvlan_pool(struct net_device *vdev,
9020 				       struct netdev_nested_priv *priv)
9021 {
9022 	struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)priv->data;
9023 	struct ixgbe_fwd_adapter *accel;
9024 	int pool;
9025 
9026 	/* we only care about macvlans... */
9027 	if (!netif_is_macvlan(vdev))
9028 		return 0;
9029 
9030 	/* that have hardware offload enabled... */
9031 	accel = macvlan_accel_priv(vdev);
9032 	if (!accel)
9033 		return 0;
9034 
9035 	/* If we can relocate to a different bit do so */
9036 	pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
9037 	if (pool < adapter->num_rx_pools) {
9038 		set_bit(pool, adapter->fwd_bitmask);
9039 		accel->pool = pool;
9040 		return 0;
9041 	}
9042 
9043 	/* if we cannot find a free pool then disable the offload */
9044 	netdev_err(vdev, "L2FW offload disabled due to lack of queue resources\n");
9045 	macvlan_release_l2fw_offload(vdev);
9046 
9047 	/* unbind the queues and drop the subordinate channel config */
9048 	netdev_unbind_sb_channel(adapter->netdev, vdev);
9049 	netdev_set_sb_channel(vdev, 0);
9050 
9051 	kfree(accel);
9052 
9053 	return 0;
9054 }
9055 
9056 static void ixgbe_defrag_macvlan_pools(struct net_device *dev)
9057 {
9058 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9059 	struct netdev_nested_priv priv = {
9060 		.data = (void *)adapter,
9061 	};
9062 
9063 	/* flush any stale bits out of the fwd bitmask */
9064 	bitmap_clear(adapter->fwd_bitmask, 1, 63);
9065 
9066 	/* walk through upper devices reassigning pools */
9067 	netdev_walk_all_upper_dev_rcu(dev, ixgbe_reassign_macvlan_pool,
9068 				      &priv);
9069 }
9070 
9071 /**
9072  * ixgbe_setup_tc - configure net_device for multiple traffic classes
9073  *
9074  * @dev: net device to configure
9075  * @tc: number of traffic classes to enable
9076  */
9077 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
9078 {
9079 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9080 	struct ixgbe_hw *hw = &adapter->hw;
9081 
9082 	/* Hardware supports up to 8 traffic classes */
9083 	if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
9084 		return -EINVAL;
9085 
9086 	if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
9087 		return -EINVAL;
9088 
9089 	/* Hardware has to reinitialize queues and interrupts to
9090 	 * match packet buffer alignment. Unfortunately, the
9091 	 * hardware is not flexible enough to do this dynamically.
9092 	 */
9093 	if (netif_running(dev))
9094 		ixgbe_close(dev);
9095 	else
9096 		ixgbe_reset(adapter);
9097 
9098 	ixgbe_clear_interrupt_scheme(adapter);
9099 
9100 #ifdef CONFIG_IXGBE_DCB
9101 	if (tc) {
9102 		if (adapter->xdp_prog) {
9103 			e_warn(probe, "DCB is not supported with XDP\n");
9104 
9105 			ixgbe_init_interrupt_scheme(adapter);
9106 			if (netif_running(dev))
9107 				ixgbe_open(dev);
9108 			return -EINVAL;
9109 		}
9110 
9111 		netdev_set_num_tc(dev, tc);
9112 		ixgbe_set_prio_tc_map(adapter);
9113 
9114 		adapter->hw_tcs = tc;
9115 		adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
9116 
9117 		if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
9118 			adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
9119 			adapter->hw.fc.requested_mode = ixgbe_fc_none;
9120 		}
9121 	} else {
9122 		netdev_reset_tc(dev);
9123 
9124 		if (adapter->hw.mac.type == ixgbe_mac_82598EB)
9125 			adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
9126 
9127 		adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
9128 		adapter->hw_tcs = tc;
9129 
9130 		adapter->temp_dcb_cfg.pfc_mode_enable = false;
9131 		adapter->dcb_cfg.pfc_mode_enable = false;
9132 	}
9133 
9134 	ixgbe_validate_rtr(adapter, tc);
9135 
9136 #endif /* CONFIG_IXGBE_DCB */
9137 	ixgbe_init_interrupt_scheme(adapter);
9138 
9139 	ixgbe_defrag_macvlan_pools(dev);
9140 
9141 	if (netif_running(dev))
9142 		return ixgbe_open(dev);
9143 
9144 	return 0;
9145 }
9146 
9147 static int ixgbe_delete_clsu32(struct ixgbe_adapter *adapter,
9148 			       struct tc_cls_u32_offload *cls)
9149 {
9150 	u32 hdl = cls->knode.handle;
9151 	u32 uhtid = TC_U32_USERHTID(cls->knode.handle);
9152 	u32 loc = cls->knode.handle & 0xfffff;
9153 	int err = 0, i, j;
9154 	struct ixgbe_jump_table *jump = NULL;
9155 
9156 	if (loc > IXGBE_MAX_HW_ENTRIES)
9157 		return -EINVAL;
9158 
9159 	if ((uhtid != 0x800) && (uhtid >= IXGBE_MAX_LINK_HANDLE))
9160 		return -EINVAL;
9161 
9162 	/* Clear this filter in the link data it is associated with */
9163 	if (uhtid != 0x800) {
9164 		jump = adapter->jump_tables[uhtid];
9165 		if (!jump)
9166 			return -EINVAL;
9167 		if (!test_bit(loc - 1, jump->child_loc_map))
9168 			return -EINVAL;
9169 		clear_bit(loc - 1, jump->child_loc_map);
9170 	}
9171 
9172 	/* Check if the filter being deleted is a link */
9173 	for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9174 		jump = adapter->jump_tables[i];
9175 		if (jump && jump->link_hdl == hdl) {
9176 			/* Delete filters in the hardware in the child hash
9177 			 * table associated with this link
9178 			 */
9179 			for (j = 0; j < IXGBE_MAX_HW_ENTRIES; j++) {
9180 				if (!test_bit(j, jump->child_loc_map))
9181 					continue;
9182 				spin_lock(&adapter->fdir_perfect_lock);
9183 				err = ixgbe_update_ethtool_fdir_entry(adapter,
9184 								      NULL,
9185 								      j + 1);
9186 				spin_unlock(&adapter->fdir_perfect_lock);
9187 				clear_bit(j, jump->child_loc_map);
9188 			}
9189 			/* Remove resources for this link */
9190 			kfree(jump->input);
9191 			kfree(jump->mask);
9192 			kfree(jump);
9193 			adapter->jump_tables[i] = NULL;
9194 			return err;
9195 		}
9196 	}
9197 
9198 	spin_lock(&adapter->fdir_perfect_lock);
9199 	err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, loc);
9200 	spin_unlock(&adapter->fdir_perfect_lock);
9201 	return err;
9202 }
9203 
9204 static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter *adapter,
9205 					    struct tc_cls_u32_offload *cls)
9206 {
9207 	u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9208 
9209 	if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9210 		return -EINVAL;
9211 
9212 	/* This ixgbe devices do not support hash tables at the moment
9213 	 * so abort when given hash tables.
9214 	 */
9215 	if (cls->hnode.divisor > 0)
9216 		return -EINVAL;
9217 
9218 	set_bit(uhtid - 1, &adapter->tables);
9219 	return 0;
9220 }
9221 
9222 static int ixgbe_configure_clsu32_del_hnode(struct ixgbe_adapter *adapter,
9223 					    struct tc_cls_u32_offload *cls)
9224 {
9225 	u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9226 
9227 	if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9228 		return -EINVAL;
9229 
9230 	clear_bit(uhtid - 1, &adapter->tables);
9231 	return 0;
9232 }
9233 
9234 #ifdef CONFIG_NET_CLS_ACT
9235 struct upper_walk_data {
9236 	struct ixgbe_adapter *adapter;
9237 	u64 action;
9238 	int ifindex;
9239 	u8 queue;
9240 };
9241 
9242 static int get_macvlan_queue(struct net_device *upper,
9243 			     struct netdev_nested_priv *priv)
9244 {
9245 	if (netif_is_macvlan(upper)) {
9246 		struct ixgbe_fwd_adapter *vadapter = macvlan_accel_priv(upper);
9247 		struct ixgbe_adapter *adapter;
9248 		struct upper_walk_data *data;
9249 		int ifindex;
9250 
9251 		data = (struct upper_walk_data *)priv->data;
9252 		ifindex = data->ifindex;
9253 		adapter = data->adapter;
9254 		if (vadapter && upper->ifindex == ifindex) {
9255 			data->queue = adapter->rx_ring[vadapter->rx_base_queue]->reg_idx;
9256 			data->action = data->queue;
9257 			return 1;
9258 		}
9259 	}
9260 
9261 	return 0;
9262 }
9263 
9264 static int handle_redirect_action(struct ixgbe_adapter *adapter, int ifindex,
9265 				  u8 *queue, u64 *action)
9266 {
9267 	struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
9268 	unsigned int num_vfs = adapter->num_vfs, vf;
9269 	struct netdev_nested_priv priv;
9270 	struct upper_walk_data data;
9271 	struct net_device *upper;
9272 
9273 	/* redirect to a SRIOV VF */
9274 	for (vf = 0; vf < num_vfs; ++vf) {
9275 		upper = pci_get_drvdata(adapter->vfinfo[vf].vfdev);
9276 		if (upper->ifindex == ifindex) {
9277 			*queue = vf * __ALIGN_MASK(1, ~vmdq->mask);
9278 			*action = vf + 1;
9279 			*action <<= ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF;
9280 			return 0;
9281 		}
9282 	}
9283 
9284 	/* redirect to a offloaded macvlan netdev */
9285 	data.adapter = adapter;
9286 	data.ifindex = ifindex;
9287 	data.action = 0;
9288 	data.queue = 0;
9289 	priv.data = (void *)&data;
9290 	if (netdev_walk_all_upper_dev_rcu(adapter->netdev,
9291 					  get_macvlan_queue, &priv)) {
9292 		*action = data.action;
9293 		*queue = data.queue;
9294 
9295 		return 0;
9296 	}
9297 
9298 	return -EINVAL;
9299 }
9300 
9301 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9302 			    struct tcf_exts *exts, u64 *action, u8 *queue)
9303 {
9304 	const struct tc_action *a;
9305 	int i;
9306 
9307 	if (!tcf_exts_has_actions(exts))
9308 		return -EINVAL;
9309 
9310 	tcf_exts_for_each_action(i, a, exts) {
9311 		/* Drop action */
9312 		if (is_tcf_gact_shot(a)) {
9313 			*action = IXGBE_FDIR_DROP_QUEUE;
9314 			*queue = IXGBE_FDIR_DROP_QUEUE;
9315 			return 0;
9316 		}
9317 
9318 		/* Redirect to a VF or a offloaded macvlan */
9319 		if (is_tcf_mirred_egress_redirect(a)) {
9320 			struct net_device *dev = tcf_mirred_dev(a);
9321 
9322 			if (!dev)
9323 				return -EINVAL;
9324 			return handle_redirect_action(adapter, dev->ifindex,
9325 						      queue, action);
9326 		}
9327 
9328 		return -EINVAL;
9329 	}
9330 
9331 	return -EINVAL;
9332 }
9333 #else
9334 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9335 			    struct tcf_exts *exts, u64 *action, u8 *queue)
9336 {
9337 	return -EINVAL;
9338 }
9339 #endif /* CONFIG_NET_CLS_ACT */
9340 
9341 static int ixgbe_clsu32_build_input(struct ixgbe_fdir_filter *input,
9342 				    union ixgbe_atr_input *mask,
9343 				    struct tc_cls_u32_offload *cls,
9344 				    struct ixgbe_mat_field *field_ptr,
9345 				    struct ixgbe_nexthdr *nexthdr)
9346 {
9347 	int i, j, off;
9348 	__be32 val, m;
9349 	bool found_entry = false, found_jump_field = false;
9350 
9351 	for (i = 0; i < cls->knode.sel->nkeys; i++) {
9352 		off = cls->knode.sel->keys[i].off;
9353 		val = cls->knode.sel->keys[i].val;
9354 		m = cls->knode.sel->keys[i].mask;
9355 
9356 		for (j = 0; field_ptr[j].val; j++) {
9357 			if (field_ptr[j].off == off) {
9358 				field_ptr[j].val(input, mask, (__force u32)val,
9359 						 (__force u32)m);
9360 				input->filter.formatted.flow_type |=
9361 					field_ptr[j].type;
9362 				found_entry = true;
9363 				break;
9364 			}
9365 		}
9366 		if (nexthdr) {
9367 			if (nexthdr->off == cls->knode.sel->keys[i].off &&
9368 			    nexthdr->val ==
9369 			    (__force u32)cls->knode.sel->keys[i].val &&
9370 			    nexthdr->mask ==
9371 			    (__force u32)cls->knode.sel->keys[i].mask)
9372 				found_jump_field = true;
9373 			else
9374 				continue;
9375 		}
9376 	}
9377 
9378 	if (nexthdr && !found_jump_field)
9379 		return -EINVAL;
9380 
9381 	if (!found_entry)
9382 		return 0;
9383 
9384 	mask->formatted.flow_type = IXGBE_ATR_L4TYPE_IPV6_MASK |
9385 				    IXGBE_ATR_L4TYPE_MASK;
9386 
9387 	if (input->filter.formatted.flow_type == IXGBE_ATR_FLOW_TYPE_IPV4)
9388 		mask->formatted.flow_type &= IXGBE_ATR_L4TYPE_IPV6_MASK;
9389 
9390 	return 0;
9391 }
9392 
9393 static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
9394 				  struct tc_cls_u32_offload *cls)
9395 {
9396 	__be16 protocol = cls->common.protocol;
9397 	u32 loc = cls->knode.handle & 0xfffff;
9398 	struct ixgbe_hw *hw = &adapter->hw;
9399 	struct ixgbe_mat_field *field_ptr;
9400 	struct ixgbe_fdir_filter *input = NULL;
9401 	union ixgbe_atr_input *mask = NULL;
9402 	struct ixgbe_jump_table *jump = NULL;
9403 	int i, err = -EINVAL;
9404 	u8 queue;
9405 	u32 uhtid, link_uhtid;
9406 
9407 	uhtid = TC_U32_USERHTID(cls->knode.handle);
9408 	link_uhtid = TC_U32_USERHTID(cls->knode.link_handle);
9409 
9410 	/* At the moment cls_u32 jumps to network layer and skips past
9411 	 * L2 headers. The canonical method to match L2 frames is to use
9412 	 * negative values. However this is error prone at best but really
9413 	 * just broken because there is no way to "know" what sort of hdr
9414 	 * is in front of the network layer. Fix cls_u32 to support L2
9415 	 * headers when needed.
9416 	 */
9417 	if (protocol != htons(ETH_P_IP))
9418 		return err;
9419 
9420 	if (loc >= ((1024 << adapter->fdir_pballoc) - 2)) {
9421 		e_err(drv, "Location out of range\n");
9422 		return err;
9423 	}
9424 
9425 	/* cls u32 is a graph starting at root node 0x800. The driver tracks
9426 	 * links and also the fields used to advance the parser across each
9427 	 * link (e.g. nexthdr/eat parameters from 'tc'). This way we can map
9428 	 * the u32 graph onto the hardware parse graph denoted in ixgbe_model.h
9429 	 * To add support for new nodes update ixgbe_model.h parse structures
9430 	 * this function _should_ be generic try not to hardcode values here.
9431 	 */
9432 	if (uhtid == 0x800) {
9433 		field_ptr = (adapter->jump_tables[0])->mat;
9434 	} else {
9435 		if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9436 			return err;
9437 		if (!adapter->jump_tables[uhtid])
9438 			return err;
9439 		field_ptr = (adapter->jump_tables[uhtid])->mat;
9440 	}
9441 
9442 	if (!field_ptr)
9443 		return err;
9444 
9445 	/* At this point we know the field_ptr is valid and need to either
9446 	 * build cls_u32 link or attach filter. Because adding a link to
9447 	 * a handle that does not exist is invalid and the same for adding
9448 	 * rules to handles that don't exist.
9449 	 */
9450 
9451 	if (link_uhtid) {
9452 		struct ixgbe_nexthdr *nexthdr = ixgbe_ipv4_jumps;
9453 
9454 		if (link_uhtid >= IXGBE_MAX_LINK_HANDLE)
9455 			return err;
9456 
9457 		if (!test_bit(link_uhtid - 1, &adapter->tables))
9458 			return err;
9459 
9460 		/* Multiple filters as links to the same hash table are not
9461 		 * supported. To add a new filter with the same next header
9462 		 * but different match/jump conditions, create a new hash table
9463 		 * and link to it.
9464 		 */
9465 		if (adapter->jump_tables[link_uhtid] &&
9466 		    (adapter->jump_tables[link_uhtid])->link_hdl) {
9467 			e_err(drv, "Link filter exists for link: %x\n",
9468 			      link_uhtid);
9469 			return err;
9470 		}
9471 
9472 		for (i = 0; nexthdr[i].jump; i++) {
9473 			if (nexthdr[i].o != cls->knode.sel->offoff ||
9474 			    nexthdr[i].s != cls->knode.sel->offshift ||
9475 			    nexthdr[i].m !=
9476 			    (__force u32)cls->knode.sel->offmask)
9477 				return err;
9478 
9479 			jump = kzalloc(sizeof(*jump), GFP_KERNEL);
9480 			if (!jump)
9481 				return -ENOMEM;
9482 			input = kzalloc(sizeof(*input), GFP_KERNEL);
9483 			if (!input) {
9484 				err = -ENOMEM;
9485 				goto free_jump;
9486 			}
9487 			mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9488 			if (!mask) {
9489 				err = -ENOMEM;
9490 				goto free_input;
9491 			}
9492 			jump->input = input;
9493 			jump->mask = mask;
9494 			jump->link_hdl = cls->knode.handle;
9495 
9496 			err = ixgbe_clsu32_build_input(input, mask, cls,
9497 						       field_ptr, &nexthdr[i]);
9498 			if (!err) {
9499 				jump->mat = nexthdr[i].jump;
9500 				adapter->jump_tables[link_uhtid] = jump;
9501 				break;
9502 			} else {
9503 				kfree(mask);
9504 				kfree(input);
9505 				kfree(jump);
9506 			}
9507 		}
9508 		return 0;
9509 	}
9510 
9511 	input = kzalloc(sizeof(*input), GFP_KERNEL);
9512 	if (!input)
9513 		return -ENOMEM;
9514 	mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9515 	if (!mask) {
9516 		err = -ENOMEM;
9517 		goto free_input;
9518 	}
9519 
9520 	if ((uhtid != 0x800) && (adapter->jump_tables[uhtid])) {
9521 		if ((adapter->jump_tables[uhtid])->input)
9522 			memcpy(input, (adapter->jump_tables[uhtid])->input,
9523 			       sizeof(*input));
9524 		if ((adapter->jump_tables[uhtid])->mask)
9525 			memcpy(mask, (adapter->jump_tables[uhtid])->mask,
9526 			       sizeof(*mask));
9527 
9528 		/* Lookup in all child hash tables if this location is already
9529 		 * filled with a filter
9530 		 */
9531 		for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9532 			struct ixgbe_jump_table *link = adapter->jump_tables[i];
9533 
9534 			if (link && (test_bit(loc - 1, link->child_loc_map))) {
9535 				e_err(drv, "Filter exists in location: %x\n",
9536 				      loc);
9537 				err = -EINVAL;
9538 				goto err_out;
9539 			}
9540 		}
9541 	}
9542 	err = ixgbe_clsu32_build_input(input, mask, cls, field_ptr, NULL);
9543 	if (err)
9544 		goto err_out;
9545 
9546 	err = parse_tc_actions(adapter, cls->knode.exts, &input->action,
9547 			       &queue);
9548 	if (err < 0)
9549 		goto err_out;
9550 
9551 	input->sw_idx = loc;
9552 
9553 	spin_lock(&adapter->fdir_perfect_lock);
9554 
9555 	if (hlist_empty(&adapter->fdir_filter_list)) {
9556 		memcpy(&adapter->fdir_mask, mask, sizeof(*mask));
9557 		err = ixgbe_fdir_set_input_mask_82599(hw, mask);
9558 		if (err)
9559 			goto err_out_w_lock;
9560 	} else if (memcmp(&adapter->fdir_mask, mask, sizeof(*mask))) {
9561 		err = -EINVAL;
9562 		goto err_out_w_lock;
9563 	}
9564 
9565 	ixgbe_atr_compute_perfect_hash_82599(&input->filter, mask);
9566 	err = ixgbe_fdir_write_perfect_filter_82599(hw, &input->filter,
9567 						    input->sw_idx, queue);
9568 	if (!err)
9569 		ixgbe_update_ethtool_fdir_entry(adapter, input, input->sw_idx);
9570 	spin_unlock(&adapter->fdir_perfect_lock);
9571 
9572 	if ((uhtid != 0x800) && (adapter->jump_tables[uhtid]))
9573 		set_bit(loc - 1, (adapter->jump_tables[uhtid])->child_loc_map);
9574 
9575 	kfree(mask);
9576 	return err;
9577 err_out_w_lock:
9578 	spin_unlock(&adapter->fdir_perfect_lock);
9579 err_out:
9580 	kfree(mask);
9581 free_input:
9582 	kfree(input);
9583 free_jump:
9584 	kfree(jump);
9585 	return err;
9586 }
9587 
9588 static int ixgbe_setup_tc_cls_u32(struct ixgbe_adapter *adapter,
9589 				  struct tc_cls_u32_offload *cls_u32)
9590 {
9591 	switch (cls_u32->command) {
9592 	case TC_CLSU32_NEW_KNODE:
9593 	case TC_CLSU32_REPLACE_KNODE:
9594 		return ixgbe_configure_clsu32(adapter, cls_u32);
9595 	case TC_CLSU32_DELETE_KNODE:
9596 		return ixgbe_delete_clsu32(adapter, cls_u32);
9597 	case TC_CLSU32_NEW_HNODE:
9598 	case TC_CLSU32_REPLACE_HNODE:
9599 		return ixgbe_configure_clsu32_add_hnode(adapter, cls_u32);
9600 	case TC_CLSU32_DELETE_HNODE:
9601 		return ixgbe_configure_clsu32_del_hnode(adapter, cls_u32);
9602 	default:
9603 		return -EOPNOTSUPP;
9604 	}
9605 }
9606 
9607 static int ixgbe_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
9608 				   void *cb_priv)
9609 {
9610 	struct ixgbe_adapter *adapter = cb_priv;
9611 
9612 	if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
9613 		return -EOPNOTSUPP;
9614 
9615 	switch (type) {
9616 	case TC_SETUP_CLSU32:
9617 		return ixgbe_setup_tc_cls_u32(adapter, type_data);
9618 	default:
9619 		return -EOPNOTSUPP;
9620 	}
9621 }
9622 
9623 static int ixgbe_setup_tc_mqprio(struct net_device *dev,
9624 				 struct tc_mqprio_qopt *mqprio)
9625 {
9626 	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
9627 	return ixgbe_setup_tc(dev, mqprio->num_tc);
9628 }
9629 
9630 static LIST_HEAD(ixgbe_block_cb_list);
9631 
9632 static int __ixgbe_setup_tc(struct net_device *dev, enum tc_setup_type type,
9633 			    void *type_data)
9634 {
9635 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9636 
9637 	switch (type) {
9638 	case TC_SETUP_BLOCK:
9639 		return flow_block_cb_setup_simple(type_data,
9640 						  &ixgbe_block_cb_list,
9641 						  ixgbe_setup_tc_block_cb,
9642 						  adapter, adapter, true);
9643 	case TC_SETUP_QDISC_MQPRIO:
9644 		return ixgbe_setup_tc_mqprio(dev, type_data);
9645 	default:
9646 		return -EOPNOTSUPP;
9647 	}
9648 }
9649 
9650 #ifdef CONFIG_PCI_IOV
9651 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
9652 {
9653 	struct net_device *netdev = adapter->netdev;
9654 
9655 	rtnl_lock();
9656 	ixgbe_setup_tc(netdev, adapter->hw_tcs);
9657 	rtnl_unlock();
9658 }
9659 
9660 #endif
9661 void ixgbe_do_reset(struct net_device *netdev)
9662 {
9663 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
9664 
9665 	if (netif_running(netdev))
9666 		ixgbe_reinit_locked(adapter);
9667 	else
9668 		ixgbe_reset(adapter);
9669 }
9670 
9671 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
9672 					    netdev_features_t features)
9673 {
9674 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
9675 
9676 	/* If Rx checksum is disabled, then RSC/LRO should also be disabled */
9677 	if (!(features & NETIF_F_RXCSUM))
9678 		features &= ~NETIF_F_LRO;
9679 
9680 	/* Turn off LRO if not RSC capable */
9681 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
9682 		features &= ~NETIF_F_LRO;
9683 
9684 	if (adapter->xdp_prog && (features & NETIF_F_LRO)) {
9685 		e_dev_err("LRO is not supported with XDP\n");
9686 		features &= ~NETIF_F_LRO;
9687 	}
9688 
9689 	return features;
9690 }
9691 
9692 static void ixgbe_reset_l2fw_offload(struct ixgbe_adapter *adapter)
9693 {
9694 	int rss = min_t(int, ixgbe_max_rss_indices(adapter),
9695 			num_online_cpus());
9696 
9697 	/* go back to full RSS if we're not running SR-IOV */
9698 	if (!adapter->ring_feature[RING_F_VMDQ].offset)
9699 		adapter->flags &= ~(IXGBE_FLAG_VMDQ_ENABLED |
9700 				    IXGBE_FLAG_SRIOV_ENABLED);
9701 
9702 	adapter->ring_feature[RING_F_RSS].limit = rss;
9703 	adapter->ring_feature[RING_F_VMDQ].limit = 1;
9704 
9705 	ixgbe_setup_tc(adapter->netdev, adapter->hw_tcs);
9706 }
9707 
9708 static int ixgbe_set_features(struct net_device *netdev,
9709 			      netdev_features_t features)
9710 {
9711 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
9712 	netdev_features_t changed = netdev->features ^ features;
9713 	bool need_reset = false;
9714 
9715 	/* Make sure RSC matches LRO, reset if change */
9716 	if (!(features & NETIF_F_LRO)) {
9717 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
9718 			need_reset = true;
9719 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
9720 	} else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
9721 		   !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
9722 		if (adapter->rx_itr_setting == 1 ||
9723 		    adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
9724 			adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
9725 			need_reset = true;
9726 		} else if ((changed ^ features) & NETIF_F_LRO) {
9727 			e_info(probe, "rx-usecs set too low, "
9728 			       "disabling RSC\n");
9729 		}
9730 	}
9731 
9732 	/*
9733 	 * Check if Flow Director n-tuple support or hw_tc support was
9734 	 * enabled or disabled.  If the state changed, we need to reset.
9735 	 */
9736 	if ((features & NETIF_F_NTUPLE) || (features & NETIF_F_HW_TC)) {
9737 		/* turn off ATR, enable perfect filters and reset */
9738 		if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
9739 			need_reset = true;
9740 
9741 		adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
9742 		adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9743 	} else {
9744 		/* turn off perfect filters, enable ATR and reset */
9745 		if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
9746 			need_reset = true;
9747 
9748 		adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9749 
9750 		/* We cannot enable ATR if SR-IOV is enabled */
9751 		if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED ||
9752 		    /* We cannot enable ATR if we have 2 or more tcs */
9753 		    (adapter->hw_tcs > 1) ||
9754 		    /* We cannot enable ATR if RSS is disabled */
9755 		    (adapter->ring_feature[RING_F_RSS].limit <= 1) ||
9756 		    /* A sample rate of 0 indicates ATR disabled */
9757 		    (!adapter->atr_sample_rate))
9758 			; /* do nothing not supported */
9759 		else /* otherwise supported and set the flag */
9760 			adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
9761 	}
9762 
9763 	if (changed & NETIF_F_RXALL)
9764 		need_reset = true;
9765 
9766 	netdev->features = features;
9767 
9768 	if ((changed & NETIF_F_HW_L2FW_DOFFLOAD) && adapter->num_rx_pools > 1)
9769 		ixgbe_reset_l2fw_offload(adapter);
9770 	else if (need_reset)
9771 		ixgbe_do_reset(netdev);
9772 	else if (changed & (NETIF_F_HW_VLAN_CTAG_RX |
9773 			    NETIF_F_HW_VLAN_CTAG_FILTER))
9774 		ixgbe_set_rx_mode(netdev);
9775 
9776 	return 1;
9777 }
9778 
9779 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
9780 			     struct net_device *dev,
9781 			     const unsigned char *addr, u16 vid,
9782 			     u16 flags,
9783 			     struct netlink_ext_ack *extack)
9784 {
9785 	/* guarantee we can provide a unique filter for the unicast address */
9786 	if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
9787 		struct ixgbe_adapter *adapter = netdev_priv(dev);
9788 		u16 pool = VMDQ_P(0);
9789 
9790 		if (netdev_uc_count(dev) >= ixgbe_available_rars(adapter, pool))
9791 			return -ENOMEM;
9792 	}
9793 
9794 	return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
9795 }
9796 
9797 /**
9798  * ixgbe_configure_bridge_mode - set various bridge modes
9799  * @adapter: the private structure
9800  * @mode: requested bridge mode
9801  *
9802  * Configure some settings require for various bridge modes.
9803  **/
9804 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
9805 				       __u16 mode)
9806 {
9807 	struct ixgbe_hw *hw = &adapter->hw;
9808 	unsigned int p, num_pools;
9809 	u32 vmdctl;
9810 
9811 	switch (mode) {
9812 	case BRIDGE_MODE_VEPA:
9813 		/* disable Tx loopback, rely on switch hairpin mode */
9814 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
9815 
9816 		/* must enable Rx switching replication to allow multicast
9817 		 * packet reception on all VFs, and to enable source address
9818 		 * pruning.
9819 		 */
9820 		vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9821 		vmdctl |= IXGBE_VT_CTL_REPLEN;
9822 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9823 
9824 		/* enable Rx source address pruning. Note, this requires
9825 		 * replication to be enabled or else it does nothing.
9826 		 */
9827 		num_pools = adapter->num_vfs + adapter->num_rx_pools;
9828 		for (p = 0; p < num_pools; p++) {
9829 			if (hw->mac.ops.set_source_address_pruning)
9830 				hw->mac.ops.set_source_address_pruning(hw,
9831 								       true,
9832 								       p);
9833 		}
9834 		break;
9835 	case BRIDGE_MODE_VEB:
9836 		/* enable Tx loopback for internal VF/PF communication */
9837 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
9838 				IXGBE_PFDTXGSWC_VT_LBEN);
9839 
9840 		/* disable Rx switching replication unless we have SR-IOV
9841 		 * virtual functions
9842 		 */
9843 		vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9844 		if (!adapter->num_vfs)
9845 			vmdctl &= ~IXGBE_VT_CTL_REPLEN;
9846 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9847 
9848 		/* disable Rx source address pruning, since we don't expect to
9849 		 * be receiving external loopback of our transmitted frames.
9850 		 */
9851 		num_pools = adapter->num_vfs + adapter->num_rx_pools;
9852 		for (p = 0; p < num_pools; p++) {
9853 			if (hw->mac.ops.set_source_address_pruning)
9854 				hw->mac.ops.set_source_address_pruning(hw,
9855 								       false,
9856 								       p);
9857 		}
9858 		break;
9859 	default:
9860 		return -EINVAL;
9861 	}
9862 
9863 	adapter->bridge_mode = mode;
9864 
9865 	e_info(drv, "enabling bridge mode: %s\n",
9866 	       mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
9867 
9868 	return 0;
9869 }
9870 
9871 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
9872 				    struct nlmsghdr *nlh, u16 flags,
9873 				    struct netlink_ext_ack *extack)
9874 {
9875 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9876 	struct nlattr *attr, *br_spec;
9877 	int rem;
9878 
9879 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
9880 		return -EOPNOTSUPP;
9881 
9882 	br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
9883 	if (!br_spec)
9884 		return -EINVAL;
9885 
9886 	nla_for_each_nested(attr, br_spec, rem) {
9887 		int status;
9888 		__u16 mode;
9889 
9890 		if (nla_type(attr) != IFLA_BRIDGE_MODE)
9891 			continue;
9892 
9893 		if (nla_len(attr) < sizeof(mode))
9894 			return -EINVAL;
9895 
9896 		mode = nla_get_u16(attr);
9897 		status = ixgbe_configure_bridge_mode(adapter, mode);
9898 		if (status)
9899 			return status;
9900 
9901 		break;
9902 	}
9903 
9904 	return 0;
9905 }
9906 
9907 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
9908 				    struct net_device *dev,
9909 				    u32 filter_mask, int nlflags)
9910 {
9911 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9912 
9913 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
9914 		return 0;
9915 
9916 	return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
9917 				       adapter->bridge_mode, 0, 0, nlflags,
9918 				       filter_mask, NULL);
9919 }
9920 
9921 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
9922 {
9923 	struct ixgbe_adapter *adapter = netdev_priv(pdev);
9924 	struct ixgbe_fwd_adapter *accel;
9925 	int tcs = adapter->hw_tcs ? : 1;
9926 	int pool, err;
9927 
9928 	if (adapter->xdp_prog) {
9929 		e_warn(probe, "L2FW offload is not supported with XDP\n");
9930 		return ERR_PTR(-EINVAL);
9931 	}
9932 
9933 	/* The hardware supported by ixgbe only filters on the destination MAC
9934 	 * address. In order to avoid issues we only support offloading modes
9935 	 * where the hardware can actually provide the functionality.
9936 	 */
9937 	if (!macvlan_supports_dest_filter(vdev))
9938 		return ERR_PTR(-EMEDIUMTYPE);
9939 
9940 	/* We need to lock down the macvlan to be a single queue device so that
9941 	 * we can reuse the tc_to_txq field in the macvlan netdev to represent
9942 	 * the queue mapping to our netdev.
9943 	 */
9944 	if (netif_is_multiqueue(vdev))
9945 		return ERR_PTR(-ERANGE);
9946 
9947 	pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
9948 	if (pool == adapter->num_rx_pools) {
9949 		u16 used_pools = adapter->num_vfs + adapter->num_rx_pools;
9950 		u16 reserved_pools;
9951 
9952 		if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
9953 		     adapter->num_rx_pools >= (MAX_TX_QUEUES / tcs)) ||
9954 		    adapter->num_rx_pools > IXGBE_MAX_MACVLANS)
9955 			return ERR_PTR(-EBUSY);
9956 
9957 		/* Hardware has a limited number of available pools. Each VF,
9958 		 * and the PF require a pool. Check to ensure we don't
9959 		 * attempt to use more then the available number of pools.
9960 		 */
9961 		if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
9962 			return ERR_PTR(-EBUSY);
9963 
9964 		/* Enable VMDq flag so device will be set in VM mode */
9965 		adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED |
9966 				  IXGBE_FLAG_SRIOV_ENABLED;
9967 
9968 		/* Try to reserve as many queues per pool as possible,
9969 		 * we start with the configurations that support 4 queues
9970 		 * per pools, followed by 2, and then by just 1 per pool.
9971 		 */
9972 		if (used_pools < 32 && adapter->num_rx_pools < 16)
9973 			reserved_pools = min_t(u16,
9974 					       32 - used_pools,
9975 					       16 - adapter->num_rx_pools);
9976 		else if (adapter->num_rx_pools < 32)
9977 			reserved_pools = min_t(u16,
9978 					       64 - used_pools,
9979 					       32 - adapter->num_rx_pools);
9980 		else
9981 			reserved_pools = 64 - used_pools;
9982 
9983 
9984 		if (!reserved_pools)
9985 			return ERR_PTR(-EBUSY);
9986 
9987 		adapter->ring_feature[RING_F_VMDQ].limit += reserved_pools;
9988 
9989 		/* Force reinit of ring allocation with VMDQ enabled */
9990 		err = ixgbe_setup_tc(pdev, adapter->hw_tcs);
9991 		if (err)
9992 			return ERR_PTR(err);
9993 
9994 		if (pool >= adapter->num_rx_pools)
9995 			return ERR_PTR(-ENOMEM);
9996 	}
9997 
9998 	accel = kzalloc(sizeof(*accel), GFP_KERNEL);
9999 	if (!accel)
10000 		return ERR_PTR(-ENOMEM);
10001 
10002 	set_bit(pool, adapter->fwd_bitmask);
10003 	netdev_set_sb_channel(vdev, pool);
10004 	accel->pool = pool;
10005 	accel->netdev = vdev;
10006 
10007 	if (!netif_running(pdev))
10008 		return accel;
10009 
10010 	err = ixgbe_fwd_ring_up(adapter, accel);
10011 	if (err)
10012 		return ERR_PTR(err);
10013 
10014 	return accel;
10015 }
10016 
10017 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
10018 {
10019 	struct ixgbe_fwd_adapter *accel = priv;
10020 	struct ixgbe_adapter *adapter = netdev_priv(pdev);
10021 	unsigned int rxbase = accel->rx_base_queue;
10022 	unsigned int i;
10023 
10024 	/* delete unicast filter associated with offloaded interface */
10025 	ixgbe_del_mac_filter(adapter, accel->netdev->dev_addr,
10026 			     VMDQ_P(accel->pool));
10027 
10028 	/* Allow remaining Rx packets to get flushed out of the
10029 	 * Rx FIFO before we drop the netdev for the ring.
10030 	 */
10031 	usleep_range(10000, 20000);
10032 
10033 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
10034 		struct ixgbe_ring *ring = adapter->rx_ring[rxbase + i];
10035 		struct ixgbe_q_vector *qv = ring->q_vector;
10036 
10037 		/* Make sure we aren't processing any packets and clear
10038 		 * netdev to shut down the ring.
10039 		 */
10040 		if (netif_running(adapter->netdev))
10041 			napi_synchronize(&qv->napi);
10042 		ring->netdev = NULL;
10043 	}
10044 
10045 	/* unbind the queues and drop the subordinate channel config */
10046 	netdev_unbind_sb_channel(pdev, accel->netdev);
10047 	netdev_set_sb_channel(accel->netdev, 0);
10048 
10049 	clear_bit(accel->pool, adapter->fwd_bitmask);
10050 	kfree(accel);
10051 }
10052 
10053 #define IXGBE_MAX_MAC_HDR_LEN		127
10054 #define IXGBE_MAX_NETWORK_HDR_LEN	511
10055 
10056 static netdev_features_t
10057 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
10058 		     netdev_features_t features)
10059 {
10060 	unsigned int network_hdr_len, mac_hdr_len;
10061 
10062 	/* Make certain the headers can be described by a context descriptor */
10063 	mac_hdr_len = skb_network_header(skb) - skb->data;
10064 	if (unlikely(mac_hdr_len > IXGBE_MAX_MAC_HDR_LEN))
10065 		return features & ~(NETIF_F_HW_CSUM |
10066 				    NETIF_F_SCTP_CRC |
10067 				    NETIF_F_GSO_UDP_L4 |
10068 				    NETIF_F_HW_VLAN_CTAG_TX |
10069 				    NETIF_F_TSO |
10070 				    NETIF_F_TSO6);
10071 
10072 	network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
10073 	if (unlikely(network_hdr_len >  IXGBE_MAX_NETWORK_HDR_LEN))
10074 		return features & ~(NETIF_F_HW_CSUM |
10075 				    NETIF_F_SCTP_CRC |
10076 				    NETIF_F_GSO_UDP_L4 |
10077 				    NETIF_F_TSO |
10078 				    NETIF_F_TSO6);
10079 
10080 	/* We can only support IPV4 TSO in tunnels if we can mangle the
10081 	 * inner IP ID field, so strip TSO if MANGLEID is not supported.
10082 	 * IPsec offoad sets skb->encapsulation but still can handle
10083 	 * the TSO, so it's the exception.
10084 	 */
10085 	if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) {
10086 #ifdef CONFIG_IXGBE_IPSEC
10087 		if (!secpath_exists(skb))
10088 #endif
10089 			features &= ~NETIF_F_TSO;
10090 	}
10091 
10092 	return features;
10093 }
10094 
10095 static int ixgbe_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
10096 {
10097 	int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
10098 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10099 	struct bpf_prog *old_prog;
10100 	bool need_reset;
10101 
10102 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
10103 		return -EINVAL;
10104 
10105 	if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
10106 		return -EINVAL;
10107 
10108 	/* verify ixgbe ring attributes are sufficient for XDP */
10109 	for (i = 0; i < adapter->num_rx_queues; i++) {
10110 		struct ixgbe_ring *ring = adapter->rx_ring[i];
10111 
10112 		if (ring_is_rsc_enabled(ring))
10113 			return -EINVAL;
10114 
10115 		if (frame_size > ixgbe_rx_bufsz(ring))
10116 			return -EINVAL;
10117 	}
10118 
10119 	if (nr_cpu_ids > MAX_XDP_QUEUES)
10120 		return -ENOMEM;
10121 
10122 	old_prog = xchg(&adapter->xdp_prog, prog);
10123 	need_reset = (!!prog != !!old_prog);
10124 
10125 	/* If transitioning XDP modes reconfigure rings */
10126 	if (need_reset) {
10127 		int err;
10128 
10129 		if (!prog)
10130 			/* Wait until ndo_xsk_wakeup completes. */
10131 			synchronize_rcu();
10132 		err = ixgbe_setup_tc(dev, adapter->hw_tcs);
10133 
10134 		if (err) {
10135 			rcu_assign_pointer(adapter->xdp_prog, old_prog);
10136 			return -EINVAL;
10137 		}
10138 	} else {
10139 		for (i = 0; i < adapter->num_rx_queues; i++)
10140 			(void)xchg(&adapter->rx_ring[i]->xdp_prog,
10141 			    adapter->xdp_prog);
10142 	}
10143 
10144 	if (old_prog)
10145 		bpf_prog_put(old_prog);
10146 
10147 	/* Kick start the NAPI context if there is an AF_XDP socket open
10148 	 * on that queue id. This so that receiving will start.
10149 	 */
10150 	if (need_reset && prog)
10151 		for (i = 0; i < adapter->num_rx_queues; i++)
10152 			if (adapter->xdp_ring[i]->xsk_pool)
10153 				(void)ixgbe_xsk_wakeup(adapter->netdev, i,
10154 						       XDP_WAKEUP_RX);
10155 
10156 	return 0;
10157 }
10158 
10159 static int ixgbe_xdp(struct net_device *dev, struct netdev_bpf *xdp)
10160 {
10161 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10162 
10163 	switch (xdp->command) {
10164 	case XDP_SETUP_PROG:
10165 		return ixgbe_xdp_setup(dev, xdp->prog);
10166 	case XDP_SETUP_XSK_POOL:
10167 		return ixgbe_xsk_pool_setup(adapter, xdp->xsk.pool,
10168 					    xdp->xsk.queue_id);
10169 
10170 	default:
10171 		return -EINVAL;
10172 	}
10173 }
10174 
10175 void ixgbe_xdp_ring_update_tail(struct ixgbe_ring *ring)
10176 {
10177 	/* Force memory writes to complete before letting h/w know there
10178 	 * are new descriptors to fetch.
10179 	 */
10180 	wmb();
10181 	writel(ring->next_to_use, ring->tail);
10182 }
10183 
10184 static int ixgbe_xdp_xmit(struct net_device *dev, int n,
10185 			  struct xdp_frame **frames, u32 flags)
10186 {
10187 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10188 	struct ixgbe_ring *ring;
10189 	int drops = 0;
10190 	int i;
10191 
10192 	if (unlikely(test_bit(__IXGBE_DOWN, &adapter->state)))
10193 		return -ENETDOWN;
10194 
10195 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
10196 		return -EINVAL;
10197 
10198 	/* During program transitions its possible adapter->xdp_prog is assigned
10199 	 * but ring has not been configured yet. In this case simply abort xmit.
10200 	 */
10201 	ring = adapter->xdp_prog ? adapter->xdp_ring[smp_processor_id()] : NULL;
10202 	if (unlikely(!ring))
10203 		return -ENXIO;
10204 
10205 	if (unlikely(test_bit(__IXGBE_TX_DISABLED, &ring->state)))
10206 		return -ENXIO;
10207 
10208 	for (i = 0; i < n; i++) {
10209 		struct xdp_frame *xdpf = frames[i];
10210 		int err;
10211 
10212 		err = ixgbe_xmit_xdp_ring(adapter, xdpf);
10213 		if (err != IXGBE_XDP_TX) {
10214 			xdp_return_frame_rx_napi(xdpf);
10215 			drops++;
10216 		}
10217 	}
10218 
10219 	if (unlikely(flags & XDP_XMIT_FLUSH))
10220 		ixgbe_xdp_ring_update_tail(ring);
10221 
10222 	return n - drops;
10223 }
10224 
10225 static const struct net_device_ops ixgbe_netdev_ops = {
10226 	.ndo_open		= ixgbe_open,
10227 	.ndo_stop		= ixgbe_close,
10228 	.ndo_start_xmit		= ixgbe_xmit_frame,
10229 	.ndo_set_rx_mode	= ixgbe_set_rx_mode,
10230 	.ndo_validate_addr	= eth_validate_addr,
10231 	.ndo_set_mac_address	= ixgbe_set_mac,
10232 	.ndo_change_mtu		= ixgbe_change_mtu,
10233 	.ndo_tx_timeout		= ixgbe_tx_timeout,
10234 	.ndo_set_tx_maxrate	= ixgbe_tx_maxrate,
10235 	.ndo_vlan_rx_add_vid	= ixgbe_vlan_rx_add_vid,
10236 	.ndo_vlan_rx_kill_vid	= ixgbe_vlan_rx_kill_vid,
10237 	.ndo_do_ioctl		= ixgbe_ioctl,
10238 	.ndo_set_vf_mac		= ixgbe_ndo_set_vf_mac,
10239 	.ndo_set_vf_vlan	= ixgbe_ndo_set_vf_vlan,
10240 	.ndo_set_vf_rate	= ixgbe_ndo_set_vf_bw,
10241 	.ndo_set_vf_spoofchk	= ixgbe_ndo_set_vf_spoofchk,
10242 	.ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
10243 	.ndo_set_vf_trust	= ixgbe_ndo_set_vf_trust,
10244 	.ndo_get_vf_config	= ixgbe_ndo_get_vf_config,
10245 	.ndo_get_stats64	= ixgbe_get_stats64,
10246 	.ndo_setup_tc		= __ixgbe_setup_tc,
10247 #ifdef IXGBE_FCOE
10248 	.ndo_select_queue	= ixgbe_select_queue,
10249 	.ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
10250 	.ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
10251 	.ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
10252 	.ndo_fcoe_enable = ixgbe_fcoe_enable,
10253 	.ndo_fcoe_disable = ixgbe_fcoe_disable,
10254 	.ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
10255 	.ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
10256 #endif /* IXGBE_FCOE */
10257 	.ndo_set_features = ixgbe_set_features,
10258 	.ndo_fix_features = ixgbe_fix_features,
10259 	.ndo_fdb_add		= ixgbe_ndo_fdb_add,
10260 	.ndo_bridge_setlink	= ixgbe_ndo_bridge_setlink,
10261 	.ndo_bridge_getlink	= ixgbe_ndo_bridge_getlink,
10262 	.ndo_dfwd_add_station	= ixgbe_fwd_add,
10263 	.ndo_dfwd_del_station	= ixgbe_fwd_del,
10264 	.ndo_features_check	= ixgbe_features_check,
10265 	.ndo_bpf		= ixgbe_xdp,
10266 	.ndo_xdp_xmit		= ixgbe_xdp_xmit,
10267 	.ndo_xsk_wakeup         = ixgbe_xsk_wakeup,
10268 };
10269 
10270 static void ixgbe_disable_txr_hw(struct ixgbe_adapter *adapter,
10271 				 struct ixgbe_ring *tx_ring)
10272 {
10273 	unsigned long wait_delay, delay_interval;
10274 	struct ixgbe_hw *hw = &adapter->hw;
10275 	u8 reg_idx = tx_ring->reg_idx;
10276 	int wait_loop;
10277 	u32 txdctl;
10278 
10279 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
10280 
10281 	/* delay mechanism from ixgbe_disable_tx */
10282 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10283 
10284 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
10285 	wait_delay = delay_interval;
10286 
10287 	while (wait_loop--) {
10288 		usleep_range(wait_delay, wait_delay + 10);
10289 		wait_delay += delay_interval * 2;
10290 		txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
10291 
10292 		if (!(txdctl & IXGBE_TXDCTL_ENABLE))
10293 			return;
10294 	}
10295 
10296 	e_err(drv, "TXDCTL.ENABLE not cleared within the polling period\n");
10297 }
10298 
10299 static void ixgbe_disable_txr(struct ixgbe_adapter *adapter,
10300 			      struct ixgbe_ring *tx_ring)
10301 {
10302 	set_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10303 	ixgbe_disable_txr_hw(adapter, tx_ring);
10304 }
10305 
10306 static void ixgbe_disable_rxr_hw(struct ixgbe_adapter *adapter,
10307 				 struct ixgbe_ring *rx_ring)
10308 {
10309 	unsigned long wait_delay, delay_interval;
10310 	struct ixgbe_hw *hw = &adapter->hw;
10311 	u8 reg_idx = rx_ring->reg_idx;
10312 	int wait_loop;
10313 	u32 rxdctl;
10314 
10315 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10316 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
10317 	rxdctl |= IXGBE_RXDCTL_SWFLSH;
10318 
10319 	/* write value back with RXDCTL.ENABLE bit cleared */
10320 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
10321 
10322 	/* RXDCTL.EN may not change on 82598 if link is down, so skip it */
10323 	if (hw->mac.type == ixgbe_mac_82598EB &&
10324 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
10325 		return;
10326 
10327 	/* delay mechanism from ixgbe_disable_rx */
10328 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10329 
10330 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
10331 	wait_delay = delay_interval;
10332 
10333 	while (wait_loop--) {
10334 		usleep_range(wait_delay, wait_delay + 10);
10335 		wait_delay += delay_interval * 2;
10336 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10337 
10338 		if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
10339 			return;
10340 	}
10341 
10342 	e_err(drv, "RXDCTL.ENABLE not cleared within the polling period\n");
10343 }
10344 
10345 static void ixgbe_reset_txr_stats(struct ixgbe_ring *tx_ring)
10346 {
10347 	memset(&tx_ring->stats, 0, sizeof(tx_ring->stats));
10348 	memset(&tx_ring->tx_stats, 0, sizeof(tx_ring->tx_stats));
10349 }
10350 
10351 static void ixgbe_reset_rxr_stats(struct ixgbe_ring *rx_ring)
10352 {
10353 	memset(&rx_ring->stats, 0, sizeof(rx_ring->stats));
10354 	memset(&rx_ring->rx_stats, 0, sizeof(rx_ring->rx_stats));
10355 }
10356 
10357 /**
10358  * ixgbe_txrx_ring_disable - Disable Rx/Tx/XDP Tx rings
10359  * @adapter: adapter structure
10360  * @ring: ring index
10361  *
10362  * This function disables a certain Rx/Tx/XDP Tx ring. The function
10363  * assumes that the netdev is running.
10364  **/
10365 void ixgbe_txrx_ring_disable(struct ixgbe_adapter *adapter, int ring)
10366 {
10367 	struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10368 
10369 	rx_ring = adapter->rx_ring[ring];
10370 	tx_ring = adapter->tx_ring[ring];
10371 	xdp_ring = adapter->xdp_ring[ring];
10372 
10373 	ixgbe_disable_txr(adapter, tx_ring);
10374 	if (xdp_ring)
10375 		ixgbe_disable_txr(adapter, xdp_ring);
10376 	ixgbe_disable_rxr_hw(adapter, rx_ring);
10377 
10378 	if (xdp_ring)
10379 		synchronize_rcu();
10380 
10381 	/* Rx/Tx/XDP Tx share the same napi context. */
10382 	napi_disable(&rx_ring->q_vector->napi);
10383 
10384 	ixgbe_clean_tx_ring(tx_ring);
10385 	if (xdp_ring)
10386 		ixgbe_clean_tx_ring(xdp_ring);
10387 	ixgbe_clean_rx_ring(rx_ring);
10388 
10389 	ixgbe_reset_txr_stats(tx_ring);
10390 	if (xdp_ring)
10391 		ixgbe_reset_txr_stats(xdp_ring);
10392 	ixgbe_reset_rxr_stats(rx_ring);
10393 }
10394 
10395 /**
10396  * ixgbe_txrx_ring_enable - Enable Rx/Tx/XDP Tx rings
10397  * @adapter: adapter structure
10398  * @ring: ring index
10399  *
10400  * This function enables a certain Rx/Tx/XDP Tx ring. The function
10401  * assumes that the netdev is running.
10402  **/
10403 void ixgbe_txrx_ring_enable(struct ixgbe_adapter *adapter, int ring)
10404 {
10405 	struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10406 
10407 	rx_ring = adapter->rx_ring[ring];
10408 	tx_ring = adapter->tx_ring[ring];
10409 	xdp_ring = adapter->xdp_ring[ring];
10410 
10411 	/* Rx/Tx/XDP Tx share the same napi context. */
10412 	napi_enable(&rx_ring->q_vector->napi);
10413 
10414 	ixgbe_configure_tx_ring(adapter, tx_ring);
10415 	if (xdp_ring)
10416 		ixgbe_configure_tx_ring(adapter, xdp_ring);
10417 	ixgbe_configure_rx_ring(adapter, rx_ring);
10418 
10419 	clear_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10420 	if (xdp_ring)
10421 		clear_bit(__IXGBE_TX_DISABLED, &xdp_ring->state);
10422 }
10423 
10424 /**
10425  * ixgbe_enumerate_functions - Get the number of ports this device has
10426  * @adapter: adapter structure
10427  *
10428  * This function enumerates the phsyical functions co-located on a single slot,
10429  * in order to determine how many ports a device has. This is most useful in
10430  * determining the required GT/s of PCIe bandwidth necessary for optimal
10431  * performance.
10432  **/
10433 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
10434 {
10435 	struct pci_dev *entry, *pdev = adapter->pdev;
10436 	int physfns = 0;
10437 
10438 	/* Some cards can not use the generic count PCIe functions method,
10439 	 * because they are behind a parent switch, so we hardcode these with
10440 	 * the correct number of functions.
10441 	 */
10442 	if (ixgbe_pcie_from_parent(&adapter->hw))
10443 		physfns = 4;
10444 
10445 	list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
10446 		/* don't count virtual functions */
10447 		if (entry->is_virtfn)
10448 			continue;
10449 
10450 		/* When the devices on the bus don't all match our device ID,
10451 		 * we can't reliably determine the correct number of
10452 		 * functions. This can occur if a function has been direct
10453 		 * attached to a virtual machine using VT-d, for example. In
10454 		 * this case, simply return -1 to indicate this.
10455 		 */
10456 		if ((entry->vendor != pdev->vendor) ||
10457 		    (entry->device != pdev->device))
10458 			return -1;
10459 
10460 		physfns++;
10461 	}
10462 
10463 	return physfns;
10464 }
10465 
10466 /**
10467  * ixgbe_wol_supported - Check whether device supports WoL
10468  * @adapter: the adapter private structure
10469  * @device_id: the device ID
10470  * @subdevice_id: the subsystem device ID
10471  *
10472  * This function is used by probe and ethtool to determine
10473  * which devices have WoL support
10474  *
10475  **/
10476 bool ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
10477 			 u16 subdevice_id)
10478 {
10479 	struct ixgbe_hw *hw = &adapter->hw;
10480 	u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
10481 
10482 	/* WOL not supported on 82598 */
10483 	if (hw->mac.type == ixgbe_mac_82598EB)
10484 		return false;
10485 
10486 	/* check eeprom to see if WOL is enabled for X540 and newer */
10487 	if (hw->mac.type >= ixgbe_mac_X540) {
10488 		if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
10489 		    ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
10490 		     (hw->bus.func == 0)))
10491 			return true;
10492 	}
10493 
10494 	/* WOL is determined based on device IDs for 82599 MACs */
10495 	switch (device_id) {
10496 	case IXGBE_DEV_ID_82599_SFP:
10497 		/* Only these subdevices could supports WOL */
10498 		switch (subdevice_id) {
10499 		case IXGBE_SUBDEV_ID_82599_560FLR:
10500 		case IXGBE_SUBDEV_ID_82599_LOM_SNAP6:
10501 		case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
10502 		case IXGBE_SUBDEV_ID_82599_SFP_2OCP:
10503 			/* only support first port */
10504 			if (hw->bus.func != 0)
10505 				break;
10506 			fallthrough;
10507 		case IXGBE_SUBDEV_ID_82599_SP_560FLR:
10508 		case IXGBE_SUBDEV_ID_82599_SFP:
10509 		case IXGBE_SUBDEV_ID_82599_RNDC:
10510 		case IXGBE_SUBDEV_ID_82599_ECNA_DP:
10511 		case IXGBE_SUBDEV_ID_82599_SFP_1OCP:
10512 		case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM1:
10513 		case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM2:
10514 			return true;
10515 		}
10516 		break;
10517 	case IXGBE_DEV_ID_82599EN_SFP:
10518 		/* Only these subdevices support WOL */
10519 		switch (subdevice_id) {
10520 		case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
10521 			return true;
10522 		}
10523 		break;
10524 	case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
10525 		/* All except this subdevice support WOL */
10526 		if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
10527 			return true;
10528 		break;
10529 	case IXGBE_DEV_ID_82599_KX4:
10530 		return  true;
10531 	default:
10532 		break;
10533 	}
10534 
10535 	return false;
10536 }
10537 
10538 /**
10539  * ixgbe_set_fw_version - Set FW version
10540  * @adapter: the adapter private structure
10541  *
10542  * This function is used by probe and ethtool to determine the FW version to
10543  * format to display. The FW version is taken from the EEPROM/NVM.
10544  */
10545 static void ixgbe_set_fw_version(struct ixgbe_adapter *adapter)
10546 {
10547 	struct ixgbe_hw *hw = &adapter->hw;
10548 	struct ixgbe_nvm_version nvm_ver;
10549 
10550 	ixgbe_get_oem_prod_version(hw, &nvm_ver);
10551 	if (nvm_ver.oem_valid) {
10552 		snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10553 			 "%x.%x.%x", nvm_ver.oem_major, nvm_ver.oem_minor,
10554 			 nvm_ver.oem_release);
10555 		return;
10556 	}
10557 
10558 	ixgbe_get_etk_id(hw, &nvm_ver);
10559 	ixgbe_get_orom_version(hw, &nvm_ver);
10560 
10561 	if (nvm_ver.or_valid) {
10562 		snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10563 			 "0x%08x, %d.%d.%d", nvm_ver.etk_id, nvm_ver.or_major,
10564 			 nvm_ver.or_build, nvm_ver.or_patch);
10565 		return;
10566 	}
10567 
10568 	/* Set ETrack ID format */
10569 	snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10570 		 "0x%08x", nvm_ver.etk_id);
10571 }
10572 
10573 /**
10574  * ixgbe_probe - Device Initialization Routine
10575  * @pdev: PCI device information struct
10576  * @ent: entry in ixgbe_pci_tbl
10577  *
10578  * Returns 0 on success, negative on failure
10579  *
10580  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
10581  * The OS initialization, configuring of the adapter private structure,
10582  * and a hardware reset occur.
10583  **/
10584 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
10585 {
10586 	struct net_device *netdev;
10587 	struct ixgbe_adapter *adapter = NULL;
10588 	struct ixgbe_hw *hw;
10589 	const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
10590 	int i, err, pci_using_dac, expected_gts;
10591 	unsigned int indices = MAX_TX_QUEUES;
10592 	u8 part_str[IXGBE_PBANUM_LENGTH];
10593 	bool disable_dev = false;
10594 #ifdef IXGBE_FCOE
10595 	u16 device_caps;
10596 #endif
10597 	u32 eec;
10598 
10599 	/* Catch broken hardware that put the wrong VF device ID in
10600 	 * the PCIe SR-IOV capability.
10601 	 */
10602 	if (pdev->is_virtfn) {
10603 		WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
10604 		     pci_name(pdev), pdev->vendor, pdev->device);
10605 		return -EINVAL;
10606 	}
10607 
10608 	err = pci_enable_device_mem(pdev);
10609 	if (err)
10610 		return err;
10611 
10612 	if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
10613 		pci_using_dac = 1;
10614 	} else {
10615 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
10616 		if (err) {
10617 			dev_err(&pdev->dev,
10618 				"No usable DMA configuration, aborting\n");
10619 			goto err_dma;
10620 		}
10621 		pci_using_dac = 0;
10622 	}
10623 
10624 	err = pci_request_mem_regions(pdev, ixgbe_driver_name);
10625 	if (err) {
10626 		dev_err(&pdev->dev,
10627 			"pci_request_selected_regions failed 0x%x\n", err);
10628 		goto err_pci_reg;
10629 	}
10630 
10631 	pci_enable_pcie_error_reporting(pdev);
10632 
10633 	pci_set_master(pdev);
10634 	pci_save_state(pdev);
10635 
10636 	if (ii->mac == ixgbe_mac_82598EB) {
10637 #ifdef CONFIG_IXGBE_DCB
10638 		/* 8 TC w/ 4 queues per TC */
10639 		indices = 4 * MAX_TRAFFIC_CLASS;
10640 #else
10641 		indices = IXGBE_MAX_RSS_INDICES;
10642 #endif
10643 	}
10644 
10645 	netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
10646 	if (!netdev) {
10647 		err = -ENOMEM;
10648 		goto err_alloc_etherdev;
10649 	}
10650 
10651 	SET_NETDEV_DEV(netdev, &pdev->dev);
10652 
10653 	adapter = netdev_priv(netdev);
10654 
10655 	adapter->netdev = netdev;
10656 	adapter->pdev = pdev;
10657 	hw = &adapter->hw;
10658 	hw->back = adapter;
10659 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
10660 
10661 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
10662 			      pci_resource_len(pdev, 0));
10663 	adapter->io_addr = hw->hw_addr;
10664 	if (!hw->hw_addr) {
10665 		err = -EIO;
10666 		goto err_ioremap;
10667 	}
10668 
10669 	netdev->netdev_ops = &ixgbe_netdev_ops;
10670 	ixgbe_set_ethtool_ops(netdev);
10671 	netdev->watchdog_timeo = 5 * HZ;
10672 	strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
10673 
10674 	/* Setup hw api */
10675 	hw->mac.ops   = *ii->mac_ops;
10676 	hw->mac.type  = ii->mac;
10677 	hw->mvals     = ii->mvals;
10678 	if (ii->link_ops)
10679 		hw->link.ops  = *ii->link_ops;
10680 
10681 	/* EEPROM */
10682 	hw->eeprom.ops = *ii->eeprom_ops;
10683 	eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
10684 	if (ixgbe_removed(hw->hw_addr)) {
10685 		err = -EIO;
10686 		goto err_ioremap;
10687 	}
10688 	/* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
10689 	if (!(eec & BIT(8)))
10690 		hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
10691 
10692 	/* PHY */
10693 	hw->phy.ops = *ii->phy_ops;
10694 	hw->phy.sfp_type = ixgbe_sfp_type_unknown;
10695 	/* ixgbe_identify_phy_generic will set prtad and mmds properly */
10696 	hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
10697 	hw->phy.mdio.mmds = 0;
10698 	hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
10699 	hw->phy.mdio.dev = netdev;
10700 	hw->phy.mdio.mdio_read = ixgbe_mdio_read;
10701 	hw->phy.mdio.mdio_write = ixgbe_mdio_write;
10702 
10703 	/* setup the private structure */
10704 	err = ixgbe_sw_init(adapter, ii);
10705 	if (err)
10706 		goto err_sw_init;
10707 
10708 	switch (adapter->hw.mac.type) {
10709 	case ixgbe_mac_X550:
10710 	case ixgbe_mac_X550EM_x:
10711 		netdev->udp_tunnel_nic_info = &ixgbe_udp_tunnels_x550;
10712 		break;
10713 	case ixgbe_mac_x550em_a:
10714 		netdev->udp_tunnel_nic_info = &ixgbe_udp_tunnels_x550em_a;
10715 		break;
10716 	default:
10717 		break;
10718 	}
10719 
10720 	/* Make sure the SWFW semaphore is in a valid state */
10721 	if (hw->mac.ops.init_swfw_sync)
10722 		hw->mac.ops.init_swfw_sync(hw);
10723 
10724 	/* Make it possible the adapter to be woken up via WOL */
10725 	switch (adapter->hw.mac.type) {
10726 	case ixgbe_mac_82599EB:
10727 	case ixgbe_mac_X540:
10728 	case ixgbe_mac_X550:
10729 	case ixgbe_mac_X550EM_x:
10730 	case ixgbe_mac_x550em_a:
10731 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
10732 		break;
10733 	default:
10734 		break;
10735 	}
10736 
10737 	/*
10738 	 * If there is a fan on this device and it has failed log the
10739 	 * failure.
10740 	 */
10741 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
10742 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
10743 		if (esdp & IXGBE_ESDP_SDP1)
10744 			e_crit(probe, "Fan has stopped, replace the adapter\n");
10745 	}
10746 
10747 	if (allow_unsupported_sfp)
10748 		hw->allow_unsupported_sfp = allow_unsupported_sfp;
10749 
10750 	/* reset_hw fills in the perm_addr as well */
10751 	hw->phy.reset_if_overtemp = true;
10752 	err = hw->mac.ops.reset_hw(hw);
10753 	hw->phy.reset_if_overtemp = false;
10754 	ixgbe_set_eee_capable(adapter);
10755 	if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
10756 		err = 0;
10757 	} else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
10758 		e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
10759 		e_dev_err("Reload the driver after installing a supported module.\n");
10760 		goto err_sw_init;
10761 	} else if (err) {
10762 		e_dev_err("HW Init failed: %d\n", err);
10763 		goto err_sw_init;
10764 	}
10765 
10766 #ifdef CONFIG_PCI_IOV
10767 	/* SR-IOV not supported on the 82598 */
10768 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
10769 		goto skip_sriov;
10770 	/* Mailbox */
10771 	ixgbe_init_mbx_params_pf(hw);
10772 	hw->mbx.ops = ii->mbx_ops;
10773 	pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
10774 	ixgbe_enable_sriov(adapter, max_vfs);
10775 skip_sriov:
10776 
10777 #endif
10778 	netdev->features = NETIF_F_SG |
10779 			   NETIF_F_TSO |
10780 			   NETIF_F_TSO6 |
10781 			   NETIF_F_RXHASH |
10782 			   NETIF_F_RXCSUM |
10783 			   NETIF_F_HW_CSUM;
10784 
10785 #define IXGBE_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
10786 				    NETIF_F_GSO_GRE_CSUM | \
10787 				    NETIF_F_GSO_IPXIP4 | \
10788 				    NETIF_F_GSO_IPXIP6 | \
10789 				    NETIF_F_GSO_UDP_TUNNEL | \
10790 				    NETIF_F_GSO_UDP_TUNNEL_CSUM)
10791 
10792 	netdev->gso_partial_features = IXGBE_GSO_PARTIAL_FEATURES;
10793 	netdev->features |= NETIF_F_GSO_PARTIAL |
10794 			    IXGBE_GSO_PARTIAL_FEATURES;
10795 
10796 	if (hw->mac.type >= ixgbe_mac_82599EB)
10797 		netdev->features |= NETIF_F_SCTP_CRC | NETIF_F_GSO_UDP_L4;
10798 
10799 #ifdef CONFIG_IXGBE_IPSEC
10800 #define IXGBE_ESP_FEATURES	(NETIF_F_HW_ESP | \
10801 				 NETIF_F_HW_ESP_TX_CSUM | \
10802 				 NETIF_F_GSO_ESP)
10803 
10804 	if (adapter->ipsec)
10805 		netdev->features |= IXGBE_ESP_FEATURES;
10806 #endif
10807 	/* copy netdev features into list of user selectable features */
10808 	netdev->hw_features |= netdev->features |
10809 			       NETIF_F_HW_VLAN_CTAG_FILTER |
10810 			       NETIF_F_HW_VLAN_CTAG_RX |
10811 			       NETIF_F_HW_VLAN_CTAG_TX |
10812 			       NETIF_F_RXALL |
10813 			       NETIF_F_HW_L2FW_DOFFLOAD;
10814 
10815 	if (hw->mac.type >= ixgbe_mac_82599EB)
10816 		netdev->hw_features |= NETIF_F_NTUPLE |
10817 				       NETIF_F_HW_TC;
10818 
10819 	if (pci_using_dac)
10820 		netdev->features |= NETIF_F_HIGHDMA;
10821 
10822 	netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
10823 	netdev->hw_enc_features |= netdev->vlan_features;
10824 	netdev->mpls_features |= NETIF_F_SG |
10825 				 NETIF_F_TSO |
10826 				 NETIF_F_TSO6 |
10827 				 NETIF_F_HW_CSUM;
10828 	netdev->mpls_features |= IXGBE_GSO_PARTIAL_FEATURES;
10829 
10830 	/* set this bit last since it cannot be part of vlan_features */
10831 	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
10832 			    NETIF_F_HW_VLAN_CTAG_RX |
10833 			    NETIF_F_HW_VLAN_CTAG_TX;
10834 
10835 	netdev->priv_flags |= IFF_UNICAST_FLT;
10836 	netdev->priv_flags |= IFF_SUPP_NOFCS;
10837 
10838 	/* MTU range: 68 - 9710 */
10839 	netdev->min_mtu = ETH_MIN_MTU;
10840 	netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE - (ETH_HLEN + ETH_FCS_LEN);
10841 
10842 #ifdef CONFIG_IXGBE_DCB
10843 	if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
10844 		netdev->dcbnl_ops = &ixgbe_dcbnl_ops;
10845 #endif
10846 
10847 #ifdef IXGBE_FCOE
10848 	if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
10849 		unsigned int fcoe_l;
10850 
10851 		if (hw->mac.ops.get_device_caps) {
10852 			hw->mac.ops.get_device_caps(hw, &device_caps);
10853 			if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
10854 				adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
10855 		}
10856 
10857 
10858 		fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
10859 		adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
10860 
10861 		netdev->features |= NETIF_F_FSO |
10862 				    NETIF_F_FCOE_CRC;
10863 
10864 		netdev->vlan_features |= NETIF_F_FSO |
10865 					 NETIF_F_FCOE_CRC |
10866 					 NETIF_F_FCOE_MTU;
10867 	}
10868 #endif /* IXGBE_FCOE */
10869 	if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
10870 		netdev->hw_features |= NETIF_F_LRO;
10871 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
10872 		netdev->features |= NETIF_F_LRO;
10873 
10874 	if (ixgbe_check_fw_error(adapter)) {
10875 		err = -EIO;
10876 		goto err_sw_init;
10877 	}
10878 
10879 	/* make sure the EEPROM is good */
10880 	if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
10881 		e_dev_err("The EEPROM Checksum Is Not Valid\n");
10882 		err = -EIO;
10883 		goto err_sw_init;
10884 	}
10885 
10886 	eth_platform_get_mac_address(&adapter->pdev->dev,
10887 				     adapter->hw.mac.perm_addr);
10888 
10889 	memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
10890 
10891 	if (!is_valid_ether_addr(netdev->dev_addr)) {
10892 		e_dev_err("invalid MAC address\n");
10893 		err = -EIO;
10894 		goto err_sw_init;
10895 	}
10896 
10897 	/* Set hw->mac.addr to permanent MAC address */
10898 	ether_addr_copy(hw->mac.addr, hw->mac.perm_addr);
10899 	ixgbe_mac_set_default_filter(adapter);
10900 
10901 	timer_setup(&adapter->service_timer, ixgbe_service_timer, 0);
10902 
10903 	if (ixgbe_removed(hw->hw_addr)) {
10904 		err = -EIO;
10905 		goto err_sw_init;
10906 	}
10907 	INIT_WORK(&adapter->service_task, ixgbe_service_task);
10908 	set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
10909 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
10910 
10911 	err = ixgbe_init_interrupt_scheme(adapter);
10912 	if (err)
10913 		goto err_sw_init;
10914 
10915 	for (i = 0; i < adapter->num_rx_queues; i++)
10916 		u64_stats_init(&adapter->rx_ring[i]->syncp);
10917 	for (i = 0; i < adapter->num_tx_queues; i++)
10918 		u64_stats_init(&adapter->tx_ring[i]->syncp);
10919 	for (i = 0; i < adapter->num_xdp_queues; i++)
10920 		u64_stats_init(&adapter->xdp_ring[i]->syncp);
10921 
10922 	/* WOL not supported for all devices */
10923 	adapter->wol = 0;
10924 	hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
10925 	hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
10926 						pdev->subsystem_device);
10927 	if (hw->wol_enabled)
10928 		adapter->wol = IXGBE_WUFC_MAG;
10929 
10930 	device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
10931 
10932 	/* save off EEPROM version number */
10933 	ixgbe_set_fw_version(adapter);
10934 
10935 	/* pick up the PCI bus settings for reporting later */
10936 	if (ixgbe_pcie_from_parent(hw))
10937 		ixgbe_get_parent_bus_info(adapter);
10938 	else
10939 		 hw->mac.ops.get_bus_info(hw);
10940 
10941 	/* calculate the expected PCIe bandwidth required for optimal
10942 	 * performance. Note that some older parts will never have enough
10943 	 * bandwidth due to being older generation PCIe parts. We clamp these
10944 	 * parts to ensure no warning is displayed if it can't be fixed.
10945 	 */
10946 	switch (hw->mac.type) {
10947 	case ixgbe_mac_82598EB:
10948 		expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
10949 		break;
10950 	default:
10951 		expected_gts = ixgbe_enumerate_functions(adapter) * 10;
10952 		break;
10953 	}
10954 
10955 	/* don't check link if we failed to enumerate functions */
10956 	if (expected_gts > 0)
10957 		ixgbe_check_minimum_link(adapter, expected_gts);
10958 
10959 	err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
10960 	if (err)
10961 		strlcpy(part_str, "Unknown", sizeof(part_str));
10962 	if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
10963 		e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
10964 			   hw->mac.type, hw->phy.type, hw->phy.sfp_type,
10965 			   part_str);
10966 	else
10967 		e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
10968 			   hw->mac.type, hw->phy.type, part_str);
10969 
10970 	e_dev_info("%pM\n", netdev->dev_addr);
10971 
10972 	/* reset the hardware with the new settings */
10973 	err = hw->mac.ops.start_hw(hw);
10974 	if (err == IXGBE_ERR_EEPROM_VERSION) {
10975 		/* We are running on a pre-production device, log a warning */
10976 		e_dev_warn("This device is a pre-production adapter/LOM. "
10977 			   "Please be aware there may be issues associated "
10978 			   "with your hardware.  If you are experiencing "
10979 			   "problems please contact your Intel or hardware "
10980 			   "representative who provided you with this "
10981 			   "hardware.\n");
10982 	}
10983 	strcpy(netdev->name, "eth%d");
10984 	pci_set_drvdata(pdev, adapter);
10985 	err = register_netdev(netdev);
10986 	if (err)
10987 		goto err_register;
10988 
10989 
10990 	/* power down the optics for 82599 SFP+ fiber */
10991 	if (hw->mac.ops.disable_tx_laser)
10992 		hw->mac.ops.disable_tx_laser(hw);
10993 
10994 	/* carrier off reporting is important to ethtool even BEFORE open */
10995 	netif_carrier_off(netdev);
10996 
10997 #ifdef CONFIG_IXGBE_DCA
10998 	if (dca_add_requester(&pdev->dev) == 0) {
10999 		adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
11000 		ixgbe_setup_dca(adapter);
11001 	}
11002 #endif
11003 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
11004 		e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
11005 		for (i = 0; i < adapter->num_vfs; i++)
11006 			ixgbe_vf_configuration(pdev, (i | 0x10000000));
11007 	}
11008 
11009 	/* firmware requires driver version to be 0xFFFFFFFF
11010 	 * since os does not support feature
11011 	 */
11012 	if (hw->mac.ops.set_fw_drv_ver)
11013 		hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF, 0xFF,
11014 					   sizeof(UTS_RELEASE) - 1,
11015 					   UTS_RELEASE);
11016 
11017 	/* add san mac addr to netdev */
11018 	ixgbe_add_sanmac_netdev(netdev);
11019 
11020 	e_dev_info("%s\n", ixgbe_default_device_descr);
11021 
11022 #ifdef CONFIG_IXGBE_HWMON
11023 	if (ixgbe_sysfs_init(adapter))
11024 		e_err(probe, "failed to allocate sysfs resources\n");
11025 #endif /* CONFIG_IXGBE_HWMON */
11026 
11027 	ixgbe_dbg_adapter_init(adapter);
11028 
11029 	/* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
11030 	if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
11031 		hw->mac.ops.setup_link(hw,
11032 			IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
11033 			true);
11034 
11035 	err = ixgbe_mii_bus_init(hw);
11036 	if (err)
11037 		goto err_netdev;
11038 
11039 	return 0;
11040 
11041 err_netdev:
11042 	unregister_netdev(netdev);
11043 err_register:
11044 	ixgbe_release_hw_control(adapter);
11045 	ixgbe_clear_interrupt_scheme(adapter);
11046 err_sw_init:
11047 	ixgbe_disable_sriov(adapter);
11048 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
11049 	iounmap(adapter->io_addr);
11050 	kfree(adapter->jump_tables[0]);
11051 	kfree(adapter->mac_table);
11052 	kfree(adapter->rss_key);
11053 	bitmap_free(adapter->af_xdp_zc_qps);
11054 err_ioremap:
11055 	disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11056 	free_netdev(netdev);
11057 err_alloc_etherdev:
11058 	pci_release_mem_regions(pdev);
11059 err_pci_reg:
11060 err_dma:
11061 	if (!adapter || disable_dev)
11062 		pci_disable_device(pdev);
11063 	return err;
11064 }
11065 
11066 /**
11067  * ixgbe_remove - Device Removal Routine
11068  * @pdev: PCI device information struct
11069  *
11070  * ixgbe_remove is called by the PCI subsystem to alert the driver
11071  * that it should release a PCI device.  The could be caused by a
11072  * Hot-Plug event, or because the driver is going to be removed from
11073  * memory.
11074  **/
11075 static void ixgbe_remove(struct pci_dev *pdev)
11076 {
11077 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11078 	struct net_device *netdev;
11079 	bool disable_dev;
11080 	int i;
11081 
11082 	/* if !adapter then we already cleaned up in probe */
11083 	if (!adapter)
11084 		return;
11085 
11086 	netdev  = adapter->netdev;
11087 	ixgbe_dbg_adapter_exit(adapter);
11088 
11089 	set_bit(__IXGBE_REMOVING, &adapter->state);
11090 	cancel_work_sync(&adapter->service_task);
11091 
11092 	if (adapter->mii_bus)
11093 		mdiobus_unregister(adapter->mii_bus);
11094 
11095 #ifdef CONFIG_IXGBE_DCA
11096 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
11097 		adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
11098 		dca_remove_requester(&pdev->dev);
11099 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
11100 				IXGBE_DCA_CTRL_DCA_DISABLE);
11101 	}
11102 
11103 #endif
11104 #ifdef CONFIG_IXGBE_HWMON
11105 	ixgbe_sysfs_exit(adapter);
11106 #endif /* CONFIG_IXGBE_HWMON */
11107 
11108 	/* remove the added san mac */
11109 	ixgbe_del_sanmac_netdev(netdev);
11110 
11111 #ifdef CONFIG_PCI_IOV
11112 	ixgbe_disable_sriov(adapter);
11113 #endif
11114 	if (netdev->reg_state == NETREG_REGISTERED)
11115 		unregister_netdev(netdev);
11116 
11117 	ixgbe_stop_ipsec_offload(adapter);
11118 	ixgbe_clear_interrupt_scheme(adapter);
11119 
11120 	ixgbe_release_hw_control(adapter);
11121 
11122 #ifdef CONFIG_DCB
11123 	kfree(adapter->ixgbe_ieee_pfc);
11124 	kfree(adapter->ixgbe_ieee_ets);
11125 
11126 #endif
11127 	iounmap(adapter->io_addr);
11128 	pci_release_mem_regions(pdev);
11129 
11130 	e_dev_info("complete\n");
11131 
11132 	for (i = 0; i < IXGBE_MAX_LINK_HANDLE; i++) {
11133 		if (adapter->jump_tables[i]) {
11134 			kfree(adapter->jump_tables[i]->input);
11135 			kfree(adapter->jump_tables[i]->mask);
11136 		}
11137 		kfree(adapter->jump_tables[i]);
11138 	}
11139 
11140 	kfree(adapter->mac_table);
11141 	kfree(adapter->rss_key);
11142 	bitmap_free(adapter->af_xdp_zc_qps);
11143 	disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11144 	free_netdev(netdev);
11145 
11146 	pci_disable_pcie_error_reporting(pdev);
11147 
11148 	if (disable_dev)
11149 		pci_disable_device(pdev);
11150 }
11151 
11152 /**
11153  * ixgbe_io_error_detected - called when PCI error is detected
11154  * @pdev: Pointer to PCI device
11155  * @state: The current pci connection state
11156  *
11157  * This function is called after a PCI bus error affecting
11158  * this device has been detected.
11159  */
11160 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
11161 						pci_channel_state_t state)
11162 {
11163 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11164 	struct net_device *netdev = adapter->netdev;
11165 
11166 #ifdef CONFIG_PCI_IOV
11167 	struct ixgbe_hw *hw = &adapter->hw;
11168 	struct pci_dev *bdev, *vfdev;
11169 	u32 dw0, dw1, dw2, dw3;
11170 	int vf, pos;
11171 	u16 req_id, pf_func;
11172 
11173 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
11174 	    adapter->num_vfs == 0)
11175 		goto skip_bad_vf_detection;
11176 
11177 	bdev = pdev->bus->self;
11178 	while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
11179 		bdev = bdev->bus->self;
11180 
11181 	if (!bdev)
11182 		goto skip_bad_vf_detection;
11183 
11184 	pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
11185 	if (!pos)
11186 		goto skip_bad_vf_detection;
11187 
11188 	dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
11189 	dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
11190 	dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
11191 	dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
11192 	if (ixgbe_removed(hw->hw_addr))
11193 		goto skip_bad_vf_detection;
11194 
11195 	req_id = dw1 >> 16;
11196 	/* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
11197 	if (!(req_id & 0x0080))
11198 		goto skip_bad_vf_detection;
11199 
11200 	pf_func = req_id & 0x01;
11201 	if ((pf_func & 1) == (pdev->devfn & 1)) {
11202 		unsigned int device_id;
11203 
11204 		vf = (req_id & 0x7F) >> 1;
11205 		e_dev_err("VF %d has caused a PCIe error\n", vf);
11206 		e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
11207 				"%8.8x\tdw3: %8.8x\n",
11208 		dw0, dw1, dw2, dw3);
11209 		switch (adapter->hw.mac.type) {
11210 		case ixgbe_mac_82599EB:
11211 			device_id = IXGBE_82599_VF_DEVICE_ID;
11212 			break;
11213 		case ixgbe_mac_X540:
11214 			device_id = IXGBE_X540_VF_DEVICE_ID;
11215 			break;
11216 		case ixgbe_mac_X550:
11217 			device_id = IXGBE_DEV_ID_X550_VF;
11218 			break;
11219 		case ixgbe_mac_X550EM_x:
11220 			device_id = IXGBE_DEV_ID_X550EM_X_VF;
11221 			break;
11222 		case ixgbe_mac_x550em_a:
11223 			device_id = IXGBE_DEV_ID_X550EM_A_VF;
11224 			break;
11225 		default:
11226 			device_id = 0;
11227 			break;
11228 		}
11229 
11230 		/* Find the pci device of the offending VF */
11231 		vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
11232 		while (vfdev) {
11233 			if (vfdev->devfn == (req_id & 0xFF))
11234 				break;
11235 			vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
11236 					       device_id, vfdev);
11237 		}
11238 		/*
11239 		 * There's a slim chance the VF could have been hot plugged,
11240 		 * so if it is no longer present we don't need to issue the
11241 		 * VFLR.  Just clean up the AER in that case.
11242 		 */
11243 		if (vfdev) {
11244 			pcie_flr(vfdev);
11245 			/* Free device reference count */
11246 			pci_dev_put(vfdev);
11247 		}
11248 	}
11249 
11250 	/*
11251 	 * Even though the error may have occurred on the other port
11252 	 * we still need to increment the vf error reference count for
11253 	 * both ports because the I/O resume function will be called
11254 	 * for both of them.
11255 	 */
11256 	adapter->vferr_refcount++;
11257 
11258 	return PCI_ERS_RESULT_RECOVERED;
11259 
11260 skip_bad_vf_detection:
11261 #endif /* CONFIG_PCI_IOV */
11262 	if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
11263 		return PCI_ERS_RESULT_DISCONNECT;
11264 
11265 	if (!netif_device_present(netdev))
11266 		return PCI_ERS_RESULT_DISCONNECT;
11267 
11268 	rtnl_lock();
11269 	netif_device_detach(netdev);
11270 
11271 	if (netif_running(netdev))
11272 		ixgbe_close_suspend(adapter);
11273 
11274 	if (state == pci_channel_io_perm_failure) {
11275 		rtnl_unlock();
11276 		return PCI_ERS_RESULT_DISCONNECT;
11277 	}
11278 
11279 	if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
11280 		pci_disable_device(pdev);
11281 	rtnl_unlock();
11282 
11283 	/* Request a slot reset. */
11284 	return PCI_ERS_RESULT_NEED_RESET;
11285 }
11286 
11287 /**
11288  * ixgbe_io_slot_reset - called after the pci bus has been reset.
11289  * @pdev: Pointer to PCI device
11290  *
11291  * Restart the card from scratch, as if from a cold-boot.
11292  */
11293 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
11294 {
11295 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11296 	pci_ers_result_t result;
11297 
11298 	if (pci_enable_device_mem(pdev)) {
11299 		e_err(probe, "Cannot re-enable PCI device after reset.\n");
11300 		result = PCI_ERS_RESULT_DISCONNECT;
11301 	} else {
11302 		smp_mb__before_atomic();
11303 		clear_bit(__IXGBE_DISABLED, &adapter->state);
11304 		adapter->hw.hw_addr = adapter->io_addr;
11305 		pci_set_master(pdev);
11306 		pci_restore_state(pdev);
11307 		pci_save_state(pdev);
11308 
11309 		pci_wake_from_d3(pdev, false);
11310 
11311 		ixgbe_reset(adapter);
11312 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
11313 		result = PCI_ERS_RESULT_RECOVERED;
11314 	}
11315 
11316 	return result;
11317 }
11318 
11319 /**
11320  * ixgbe_io_resume - called when traffic can start flowing again.
11321  * @pdev: Pointer to PCI device
11322  *
11323  * This callback is called when the error recovery driver tells us that
11324  * its OK to resume normal operation.
11325  */
11326 static void ixgbe_io_resume(struct pci_dev *pdev)
11327 {
11328 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11329 	struct net_device *netdev = adapter->netdev;
11330 
11331 #ifdef CONFIG_PCI_IOV
11332 	if (adapter->vferr_refcount) {
11333 		e_info(drv, "Resuming after VF err\n");
11334 		adapter->vferr_refcount--;
11335 		return;
11336 	}
11337 
11338 #endif
11339 	rtnl_lock();
11340 	if (netif_running(netdev))
11341 		ixgbe_open(netdev);
11342 
11343 	netif_device_attach(netdev);
11344 	rtnl_unlock();
11345 }
11346 
11347 static const struct pci_error_handlers ixgbe_err_handler = {
11348 	.error_detected = ixgbe_io_error_detected,
11349 	.slot_reset = ixgbe_io_slot_reset,
11350 	.resume = ixgbe_io_resume,
11351 };
11352 
11353 static SIMPLE_DEV_PM_OPS(ixgbe_pm_ops, ixgbe_suspend, ixgbe_resume);
11354 
11355 static struct pci_driver ixgbe_driver = {
11356 	.name      = ixgbe_driver_name,
11357 	.id_table  = ixgbe_pci_tbl,
11358 	.probe     = ixgbe_probe,
11359 	.remove    = ixgbe_remove,
11360 	.driver.pm = &ixgbe_pm_ops,
11361 	.shutdown  = ixgbe_shutdown,
11362 	.sriov_configure = ixgbe_pci_sriov_configure,
11363 	.err_handler = &ixgbe_err_handler
11364 };
11365 
11366 /**
11367  * ixgbe_init_module - Driver Registration Routine
11368  *
11369  * ixgbe_init_module is the first routine called when the driver is
11370  * loaded. All it does is register with the PCI subsystem.
11371  **/
11372 static int __init ixgbe_init_module(void)
11373 {
11374 	int ret;
11375 	pr_info("%s\n", ixgbe_driver_string);
11376 	pr_info("%s\n", ixgbe_copyright);
11377 
11378 	ixgbe_wq = create_singlethread_workqueue(ixgbe_driver_name);
11379 	if (!ixgbe_wq) {
11380 		pr_err("%s: Failed to create workqueue\n", ixgbe_driver_name);
11381 		return -ENOMEM;
11382 	}
11383 
11384 	ixgbe_dbg_init();
11385 
11386 	ret = pci_register_driver(&ixgbe_driver);
11387 	if (ret) {
11388 		destroy_workqueue(ixgbe_wq);
11389 		ixgbe_dbg_exit();
11390 		return ret;
11391 	}
11392 
11393 #ifdef CONFIG_IXGBE_DCA
11394 	dca_register_notify(&dca_notifier);
11395 #endif
11396 
11397 	return 0;
11398 }
11399 
11400 module_init(ixgbe_init_module);
11401 
11402 /**
11403  * ixgbe_exit_module - Driver Exit Cleanup Routine
11404  *
11405  * ixgbe_exit_module is called just before the driver is removed
11406  * from memory.
11407  **/
11408 static void __exit ixgbe_exit_module(void)
11409 {
11410 #ifdef CONFIG_IXGBE_DCA
11411 	dca_unregister_notify(&dca_notifier);
11412 #endif
11413 	pci_unregister_driver(&ixgbe_driver);
11414 
11415 	ixgbe_dbg_exit();
11416 	if (ixgbe_wq) {
11417 		destroy_workqueue(ixgbe_wq);
11418 		ixgbe_wq = NULL;
11419 	}
11420 }
11421 
11422 #ifdef CONFIG_IXGBE_DCA
11423 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
11424 			    void *p)
11425 {
11426 	int ret_val;
11427 
11428 	ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
11429 					 __ixgbe_notify_dca);
11430 
11431 	return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
11432 }
11433 
11434 #endif /* CONFIG_IXGBE_DCA */
11435 
11436 module_exit(ixgbe_exit_module);
11437 
11438 /* ixgbe_main.c */
11439