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 inline 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 = ixgbe_rx_offset(rx_ring);
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 inline bool ixgbe_page_is_reserved(struct page *page)
1944 {
1945 	return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1946 }
1947 
1948 static bool ixgbe_can_reuse_rx_page(struct ixgbe_rx_buffer *rx_buffer)
1949 {
1950 	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1951 	struct page *page = rx_buffer->page;
1952 
1953 	/* avoid re-using remote pages */
1954 	if (unlikely(ixgbe_page_is_reserved(page)))
1955 		return false;
1956 
1957 #if (PAGE_SIZE < 8192)
1958 	/* if we are only owner of page we can reuse it */
1959 	if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
1960 		return false;
1961 #else
1962 	/* The last offset is a bit aggressive in that we assume the
1963 	 * worst case of FCoE being enabled and using a 3K buffer.
1964 	 * However this should have minimal impact as the 1K extra is
1965 	 * still less than one buffer in size.
1966 	 */
1967 #define IXGBE_LAST_OFFSET \
1968 	(SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBE_RXBUFFER_3K)
1969 	if (rx_buffer->page_offset > IXGBE_LAST_OFFSET)
1970 		return false;
1971 #endif
1972 
1973 	/* If we have drained the page fragment pool we need to update
1974 	 * the pagecnt_bias and page count so that we fully restock the
1975 	 * number of references the driver holds.
1976 	 */
1977 	if (unlikely(pagecnt_bias == 1)) {
1978 		page_ref_add(page, USHRT_MAX - 1);
1979 		rx_buffer->pagecnt_bias = USHRT_MAX;
1980 	}
1981 
1982 	return true;
1983 }
1984 
1985 /**
1986  * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1987  * @rx_ring: rx descriptor ring to transact packets on
1988  * @rx_buffer: buffer containing page to add
1989  * @skb: sk_buff to place the data into
1990  * @size: size of data in rx_buffer
1991  *
1992  * This function will add the data contained in rx_buffer->page to the skb.
1993  * This is done either through a direct copy if the data in the buffer is
1994  * less than the skb header size, otherwise it will just attach the page as
1995  * a frag to the skb.
1996  *
1997  * The function will then update the page offset if necessary and return
1998  * true if the buffer can be reused by the adapter.
1999  **/
2000 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
2001 			      struct ixgbe_rx_buffer *rx_buffer,
2002 			      struct sk_buff *skb,
2003 			      unsigned int size)
2004 {
2005 #if (PAGE_SIZE < 8192)
2006 	unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2007 #else
2008 	unsigned int truesize = ring_uses_build_skb(rx_ring) ?
2009 				SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) :
2010 				SKB_DATA_ALIGN(size);
2011 #endif
2012 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
2013 			rx_buffer->page_offset, size, truesize);
2014 #if (PAGE_SIZE < 8192)
2015 	rx_buffer->page_offset ^= truesize;
2016 #else
2017 	rx_buffer->page_offset += truesize;
2018 #endif
2019 }
2020 
2021 static struct ixgbe_rx_buffer *ixgbe_get_rx_buffer(struct ixgbe_ring *rx_ring,
2022 						   union ixgbe_adv_rx_desc *rx_desc,
2023 						   struct sk_buff **skb,
2024 						   const unsigned int size)
2025 {
2026 	struct ixgbe_rx_buffer *rx_buffer;
2027 
2028 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
2029 	prefetchw(rx_buffer->page);
2030 	*skb = rx_buffer->skb;
2031 
2032 	/* Delay unmapping of the first packet. It carries the header
2033 	 * information, HW may still access the header after the writeback.
2034 	 * Only unmap it when EOP is reached
2035 	 */
2036 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)) {
2037 		if (!*skb)
2038 			goto skip_sync;
2039 	} else {
2040 		if (*skb)
2041 			ixgbe_dma_sync_frag(rx_ring, *skb);
2042 	}
2043 
2044 	/* we are reusing so sync this buffer for CPU use */
2045 	dma_sync_single_range_for_cpu(rx_ring->dev,
2046 				      rx_buffer->dma,
2047 				      rx_buffer->page_offset,
2048 				      size,
2049 				      DMA_FROM_DEVICE);
2050 skip_sync:
2051 	rx_buffer->pagecnt_bias--;
2052 
2053 	return rx_buffer;
2054 }
2055 
2056 static void ixgbe_put_rx_buffer(struct ixgbe_ring *rx_ring,
2057 				struct ixgbe_rx_buffer *rx_buffer,
2058 				struct sk_buff *skb)
2059 {
2060 	if (ixgbe_can_reuse_rx_page(rx_buffer)) {
2061 		/* hand second half of page back to the ring */
2062 		ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2063 	} else {
2064 		if (!IS_ERR(skb) && IXGBE_CB(skb)->dma == rx_buffer->dma) {
2065 			/* the page has been released from the ring */
2066 			IXGBE_CB(skb)->page_released = true;
2067 		} else {
2068 			/* we are not reusing the buffer so unmap it */
2069 			dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2070 					     ixgbe_rx_pg_size(rx_ring),
2071 					     DMA_FROM_DEVICE,
2072 					     IXGBE_RX_DMA_ATTR);
2073 		}
2074 		__page_frag_cache_drain(rx_buffer->page,
2075 					rx_buffer->pagecnt_bias);
2076 	}
2077 
2078 	/* clear contents of rx_buffer */
2079 	rx_buffer->page = NULL;
2080 	rx_buffer->skb = NULL;
2081 }
2082 
2083 static struct sk_buff *ixgbe_construct_skb(struct ixgbe_ring *rx_ring,
2084 					   struct ixgbe_rx_buffer *rx_buffer,
2085 					   struct xdp_buff *xdp,
2086 					   union ixgbe_adv_rx_desc *rx_desc)
2087 {
2088 	unsigned int size = xdp->data_end - xdp->data;
2089 #if (PAGE_SIZE < 8192)
2090 	unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2091 #else
2092 	unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
2093 					       xdp->data_hard_start);
2094 #endif
2095 	struct sk_buff *skb;
2096 
2097 	/* prefetch first cache line of first page */
2098 	prefetch(xdp->data);
2099 #if L1_CACHE_BYTES < 128
2100 	prefetch(xdp->data + L1_CACHE_BYTES);
2101 #endif
2102 	/* Note, we get here by enabling legacy-rx via:
2103 	 *
2104 	 *    ethtool --set-priv-flags <dev> legacy-rx on
2105 	 *
2106 	 * In this mode, we currently get 0 extra XDP headroom as
2107 	 * opposed to having legacy-rx off, where we process XDP
2108 	 * packets going to stack via ixgbe_build_skb(). The latter
2109 	 * provides us currently with 192 bytes of headroom.
2110 	 *
2111 	 * For ixgbe_construct_skb() mode it means that the
2112 	 * xdp->data_meta will always point to xdp->data, since
2113 	 * the helper cannot expand the head. Should this ever
2114 	 * change in future for legacy-rx mode on, then lets also
2115 	 * add xdp->data_meta handling here.
2116 	 */
2117 
2118 	/* allocate a skb to store the frags */
2119 	skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBE_RX_HDR_SIZE);
2120 	if (unlikely(!skb))
2121 		return NULL;
2122 
2123 	if (size > IXGBE_RX_HDR_SIZE) {
2124 		if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2125 			IXGBE_CB(skb)->dma = rx_buffer->dma;
2126 
2127 		skb_add_rx_frag(skb, 0, rx_buffer->page,
2128 				xdp->data - page_address(rx_buffer->page),
2129 				size, truesize);
2130 #if (PAGE_SIZE < 8192)
2131 		rx_buffer->page_offset ^= truesize;
2132 #else
2133 		rx_buffer->page_offset += truesize;
2134 #endif
2135 	} else {
2136 		memcpy(__skb_put(skb, size),
2137 		       xdp->data, ALIGN(size, sizeof(long)));
2138 		rx_buffer->pagecnt_bias++;
2139 	}
2140 
2141 	return skb;
2142 }
2143 
2144 static struct sk_buff *ixgbe_build_skb(struct ixgbe_ring *rx_ring,
2145 				       struct ixgbe_rx_buffer *rx_buffer,
2146 				       struct xdp_buff *xdp,
2147 				       union ixgbe_adv_rx_desc *rx_desc)
2148 {
2149 	unsigned int metasize = xdp->data - xdp->data_meta;
2150 #if (PAGE_SIZE < 8192)
2151 	unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2152 #else
2153 	unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
2154 				SKB_DATA_ALIGN(xdp->data_end -
2155 					       xdp->data_hard_start);
2156 #endif
2157 	struct sk_buff *skb;
2158 
2159 	/* Prefetch first cache line of first page. If xdp->data_meta
2160 	 * is unused, this points extactly as xdp->data, otherwise we
2161 	 * likely have a consumer accessing first few bytes of meta
2162 	 * data, and then actual data.
2163 	 */
2164 	prefetch(xdp->data_meta);
2165 #if L1_CACHE_BYTES < 128
2166 	prefetch(xdp->data_meta + L1_CACHE_BYTES);
2167 #endif
2168 
2169 	/* build an skb to around the page buffer */
2170 	skb = build_skb(xdp->data_hard_start, truesize);
2171 	if (unlikely(!skb))
2172 		return NULL;
2173 
2174 	/* update pointers within the skb to store the data */
2175 	skb_reserve(skb, xdp->data - xdp->data_hard_start);
2176 	__skb_put(skb, xdp->data_end - xdp->data);
2177 	if (metasize)
2178 		skb_metadata_set(skb, metasize);
2179 
2180 	/* record DMA address if this is the start of a chain of buffers */
2181 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2182 		IXGBE_CB(skb)->dma = rx_buffer->dma;
2183 
2184 	/* update buffer offset */
2185 #if (PAGE_SIZE < 8192)
2186 	rx_buffer->page_offset ^= truesize;
2187 #else
2188 	rx_buffer->page_offset += truesize;
2189 #endif
2190 
2191 	return skb;
2192 }
2193 
2194 static struct sk_buff *ixgbe_run_xdp(struct ixgbe_adapter *adapter,
2195 				     struct ixgbe_ring *rx_ring,
2196 				     struct xdp_buff *xdp)
2197 {
2198 	int err, result = IXGBE_XDP_PASS;
2199 	struct bpf_prog *xdp_prog;
2200 	struct xdp_frame *xdpf;
2201 	u32 act;
2202 
2203 	rcu_read_lock();
2204 	xdp_prog = READ_ONCE(rx_ring->xdp_prog);
2205 
2206 	if (!xdp_prog)
2207 		goto xdp_out;
2208 
2209 	prefetchw(xdp->data_hard_start); /* xdp_frame write */
2210 
2211 	act = bpf_prog_run_xdp(xdp_prog, xdp);
2212 	switch (act) {
2213 	case XDP_PASS:
2214 		break;
2215 	case XDP_TX:
2216 		xdpf = xdp_convert_buff_to_frame(xdp);
2217 		if (unlikely(!xdpf)) {
2218 			result = IXGBE_XDP_CONSUMED;
2219 			break;
2220 		}
2221 		result = ixgbe_xmit_xdp_ring(adapter, xdpf);
2222 		break;
2223 	case XDP_REDIRECT:
2224 		err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
2225 		if (!err)
2226 			result = IXGBE_XDP_REDIR;
2227 		else
2228 			result = IXGBE_XDP_CONSUMED;
2229 		break;
2230 	default:
2231 		bpf_warn_invalid_xdp_action(act);
2232 		fallthrough;
2233 	case XDP_ABORTED:
2234 		trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
2235 		fallthrough; /* handle aborts by dropping packet */
2236 	case XDP_DROP:
2237 		result = IXGBE_XDP_CONSUMED;
2238 		break;
2239 	}
2240 xdp_out:
2241 	rcu_read_unlock();
2242 	return ERR_PTR(-result);
2243 }
2244 
2245 static unsigned int ixgbe_rx_frame_truesize(struct ixgbe_ring *rx_ring,
2246 					    unsigned int size)
2247 {
2248 	unsigned int truesize;
2249 
2250 #if (PAGE_SIZE < 8192)
2251 	truesize = ixgbe_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
2252 #else
2253 	truesize = ring_uses_build_skb(rx_ring) ?
2254 		SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) +
2255 		SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
2256 		SKB_DATA_ALIGN(size);
2257 #endif
2258 	return truesize;
2259 }
2260 
2261 static void ixgbe_rx_buffer_flip(struct ixgbe_ring *rx_ring,
2262 				 struct ixgbe_rx_buffer *rx_buffer,
2263 				 unsigned int size)
2264 {
2265 	unsigned int truesize = ixgbe_rx_frame_truesize(rx_ring, size);
2266 #if (PAGE_SIZE < 8192)
2267 	rx_buffer->page_offset ^= truesize;
2268 #else
2269 	rx_buffer->page_offset += truesize;
2270 #endif
2271 }
2272 
2273 /**
2274  * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2275  * @q_vector: structure containing interrupt and ring information
2276  * @rx_ring: rx descriptor ring to transact packets on
2277  * @budget: Total limit on number of packets to process
2278  *
2279  * This function provides a "bounce buffer" approach to Rx interrupt
2280  * processing.  The advantage to this is that on systems that have
2281  * expensive overhead for IOMMU access this provides a means of avoiding
2282  * it by maintaining the mapping of the page to the syste.
2283  *
2284  * Returns amount of work completed
2285  **/
2286 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2287 			       struct ixgbe_ring *rx_ring,
2288 			       const int budget)
2289 {
2290 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2291 	struct ixgbe_adapter *adapter = q_vector->adapter;
2292 #ifdef IXGBE_FCOE
2293 	int ddp_bytes;
2294 	unsigned int mss = 0;
2295 #endif /* IXGBE_FCOE */
2296 	u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2297 	unsigned int xdp_xmit = 0;
2298 	struct xdp_buff xdp;
2299 
2300 	xdp.rxq = &rx_ring->xdp_rxq;
2301 
2302 	/* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
2303 #if (PAGE_SIZE < 8192)
2304 	xdp.frame_sz = ixgbe_rx_frame_truesize(rx_ring, 0);
2305 #endif
2306 
2307 	while (likely(total_rx_packets < budget)) {
2308 		union ixgbe_adv_rx_desc *rx_desc;
2309 		struct ixgbe_rx_buffer *rx_buffer;
2310 		struct sk_buff *skb;
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);
2331 
2332 		/* retrieve a buffer from the ring */
2333 		if (!skb) {
2334 			xdp.data = page_address(rx_buffer->page) +
2335 				   rx_buffer->page_offset;
2336 			xdp.data_meta = xdp.data;
2337 			xdp.data_hard_start = xdp.data -
2338 					      ixgbe_rx_offset(rx_ring);
2339 			xdp.data_end = xdp.data + size;
2340 #if (PAGE_SIZE > 4096)
2341 			/* At larger PAGE_SIZE, frame_sz depend on len size */
2342 			xdp.frame_sz = ixgbe_rx_frame_truesize(rx_ring, size);
2343 #endif
2344 			skb = ixgbe_run_xdp(adapter, rx_ring, &xdp);
2345 		}
2346 
2347 		if (IS_ERR(skb)) {
2348 			unsigned int xdp_res = -PTR_ERR(skb);
2349 
2350 			if (xdp_res & (IXGBE_XDP_TX | IXGBE_XDP_REDIR)) {
2351 				xdp_xmit |= xdp_res;
2352 				ixgbe_rx_buffer_flip(rx_ring, rx_buffer, size);
2353 			} else {
2354 				rx_buffer->pagecnt_bias++;
2355 			}
2356 			total_rx_packets++;
2357 			total_rx_bytes += size;
2358 		} else if (skb) {
2359 			ixgbe_add_rx_frag(rx_ring, rx_buffer, skb, size);
2360 		} else if (ring_uses_build_skb(rx_ring)) {
2361 			skb = ixgbe_build_skb(rx_ring, rx_buffer,
2362 					      &xdp, rx_desc);
2363 		} else {
2364 			skb = ixgbe_construct_skb(rx_ring, rx_buffer,
2365 						  &xdp, rx_desc);
2366 		}
2367 
2368 		/* exit if we failed to retrieve a buffer */
2369 		if (!skb) {
2370 			rx_ring->rx_stats.alloc_rx_buff_failed++;
2371 			rx_buffer->pagecnt_bias++;
2372 			break;
2373 		}
2374 
2375 		ixgbe_put_rx_buffer(rx_ring, rx_buffer, skb);
2376 		cleaned_count++;
2377 
2378 		/* place incomplete frames back on ring for completion */
2379 		if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2380 			continue;
2381 
2382 		/* verify the packet layout is correct */
2383 		if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2384 			continue;
2385 
2386 		/* probably a little skewed due to removing CRC */
2387 		total_rx_bytes += skb->len;
2388 
2389 		/* populate checksum, timestamp, VLAN, and protocol */
2390 		ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2391 
2392 #ifdef IXGBE_FCOE
2393 		/* if ddp, not passing to ULD unless for FCP_RSP or error */
2394 		if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2395 			ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2396 			/* include DDPed FCoE data */
2397 			if (ddp_bytes > 0) {
2398 				if (!mss) {
2399 					mss = rx_ring->netdev->mtu -
2400 						sizeof(struct fcoe_hdr) -
2401 						sizeof(struct fc_frame_header) -
2402 						sizeof(struct fcoe_crc_eof);
2403 					if (mss > 512)
2404 						mss &= ~511;
2405 				}
2406 				total_rx_bytes += ddp_bytes;
2407 				total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2408 								 mss);
2409 			}
2410 			if (!ddp_bytes) {
2411 				dev_kfree_skb_any(skb);
2412 				continue;
2413 			}
2414 		}
2415 
2416 #endif /* IXGBE_FCOE */
2417 		ixgbe_rx_skb(q_vector, skb);
2418 
2419 		/* update budget accounting */
2420 		total_rx_packets++;
2421 	}
2422 
2423 	if (xdp_xmit & IXGBE_XDP_REDIR)
2424 		xdp_do_flush_map();
2425 
2426 	if (xdp_xmit & IXGBE_XDP_TX) {
2427 		struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
2428 
2429 		/* Force memory writes to complete before letting h/w
2430 		 * know there are new descriptors to fetch.
2431 		 */
2432 		wmb();
2433 		writel(ring->next_to_use, ring->tail);
2434 	}
2435 
2436 	u64_stats_update_begin(&rx_ring->syncp);
2437 	rx_ring->stats.packets += total_rx_packets;
2438 	rx_ring->stats.bytes += total_rx_bytes;
2439 	u64_stats_update_end(&rx_ring->syncp);
2440 	q_vector->rx.total_packets += total_rx_packets;
2441 	q_vector->rx.total_bytes += total_rx_bytes;
2442 
2443 	return total_rx_packets;
2444 }
2445 
2446 /**
2447  * ixgbe_configure_msix - Configure MSI-X hardware
2448  * @adapter: board private structure
2449  *
2450  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2451  * interrupts.
2452  **/
2453 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2454 {
2455 	struct ixgbe_q_vector *q_vector;
2456 	int v_idx;
2457 	u32 mask;
2458 
2459 	/* Populate MSIX to EITR Select */
2460 	if (adapter->num_vfs > 32) {
2461 		u32 eitrsel = BIT(adapter->num_vfs - 32) - 1;
2462 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2463 	}
2464 
2465 	/*
2466 	 * Populate the IVAR table and set the ITR values to the
2467 	 * corresponding register.
2468 	 */
2469 	for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2470 		struct ixgbe_ring *ring;
2471 		q_vector = adapter->q_vector[v_idx];
2472 
2473 		ixgbe_for_each_ring(ring, q_vector->rx)
2474 			ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2475 
2476 		ixgbe_for_each_ring(ring, q_vector->tx)
2477 			ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2478 
2479 		ixgbe_write_eitr(q_vector);
2480 	}
2481 
2482 	switch (adapter->hw.mac.type) {
2483 	case ixgbe_mac_82598EB:
2484 		ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2485 			       v_idx);
2486 		break;
2487 	case ixgbe_mac_82599EB:
2488 	case ixgbe_mac_X540:
2489 	case ixgbe_mac_X550:
2490 	case ixgbe_mac_X550EM_x:
2491 	case ixgbe_mac_x550em_a:
2492 		ixgbe_set_ivar(adapter, -1, 1, v_idx);
2493 		break;
2494 	default:
2495 		break;
2496 	}
2497 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2498 
2499 	/* set up to autoclear timer, and the vectors */
2500 	mask = IXGBE_EIMS_ENABLE_MASK;
2501 	mask &= ~(IXGBE_EIMS_OTHER |
2502 		  IXGBE_EIMS_MAILBOX |
2503 		  IXGBE_EIMS_LSC);
2504 
2505 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2506 }
2507 
2508 /**
2509  * ixgbe_update_itr - update the dynamic ITR value based on statistics
2510  * @q_vector: structure containing interrupt and ring information
2511  * @ring_container: structure containing ring performance data
2512  *
2513  *      Stores a new ITR value based on packets and byte
2514  *      counts during the last interrupt.  The advantage of per interrupt
2515  *      computation is faster updates and more accurate ITR for the current
2516  *      traffic pattern.  Constants in this function were computed
2517  *      based on theoretical maximum wire speed and thresholds were set based
2518  *      on testing data as well as attempting to minimize response time
2519  *      while increasing bulk throughput.
2520  **/
2521 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2522 			     struct ixgbe_ring_container *ring_container)
2523 {
2524 	unsigned int itr = IXGBE_ITR_ADAPTIVE_MIN_USECS |
2525 			   IXGBE_ITR_ADAPTIVE_LATENCY;
2526 	unsigned int avg_wire_size, packets, bytes;
2527 	unsigned long next_update = jiffies;
2528 
2529 	/* If we don't have any rings just leave ourselves set for maximum
2530 	 * possible latency so we take ourselves out of the equation.
2531 	 */
2532 	if (!ring_container->ring)
2533 		return;
2534 
2535 	/* If we didn't update within up to 1 - 2 jiffies we can assume
2536 	 * that either packets are coming in so slow there hasn't been
2537 	 * any work, or that there is so much work that NAPI is dealing
2538 	 * with interrupt moderation and we don't need to do anything.
2539 	 */
2540 	if (time_after(next_update, ring_container->next_update))
2541 		goto clear_counts;
2542 
2543 	packets = ring_container->total_packets;
2544 
2545 	/* We have no packets to actually measure against. This means
2546 	 * either one of the other queues on this vector is active or
2547 	 * we are a Tx queue doing TSO with too high of an interrupt rate.
2548 	 *
2549 	 * When this occurs just tick up our delay by the minimum value
2550 	 * and hope that this extra delay will prevent us from being called
2551 	 * without any work on our queue.
2552 	 */
2553 	if (!packets) {
2554 		itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2555 		if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2556 			itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2557 		itr += ring_container->itr & IXGBE_ITR_ADAPTIVE_LATENCY;
2558 		goto clear_counts;
2559 	}
2560 
2561 	bytes = ring_container->total_bytes;
2562 
2563 	/* If packets are less than 4 or bytes are less than 9000 assume
2564 	 * insufficient data to use bulk rate limiting approach. We are
2565 	 * likely latency driven.
2566 	 */
2567 	if (packets < 4 && bytes < 9000) {
2568 		itr = IXGBE_ITR_ADAPTIVE_LATENCY;
2569 		goto adjust_by_size;
2570 	}
2571 
2572 	/* Between 4 and 48 we can assume that our current interrupt delay
2573 	 * is only slightly too low. As such we should increase it by a small
2574 	 * fixed amount.
2575 	 */
2576 	if (packets < 48) {
2577 		itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2578 		if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2579 			itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2580 		goto clear_counts;
2581 	}
2582 
2583 	/* Between 48 and 96 is our "goldilocks" zone where we are working
2584 	 * out "just right". Just report that our current ITR is good for us.
2585 	 */
2586 	if (packets < 96) {
2587 		itr = q_vector->itr >> 2;
2588 		goto clear_counts;
2589 	}
2590 
2591 	/* If packet count is 96 or greater we are likely looking at a slight
2592 	 * overrun of the delay we want. Try halving our delay to see if that
2593 	 * will cut the number of packets in half per interrupt.
2594 	 */
2595 	if (packets < 256) {
2596 		itr = q_vector->itr >> 3;
2597 		if (itr < IXGBE_ITR_ADAPTIVE_MIN_USECS)
2598 			itr = IXGBE_ITR_ADAPTIVE_MIN_USECS;
2599 		goto clear_counts;
2600 	}
2601 
2602 	/* The paths below assume we are dealing with a bulk ITR since number
2603 	 * of packets is 256 or greater. We are just going to have to compute
2604 	 * a value and try to bring the count under control, though for smaller
2605 	 * packet sizes there isn't much we can do as NAPI polling will likely
2606 	 * be kicking in sooner rather than later.
2607 	 */
2608 	itr = IXGBE_ITR_ADAPTIVE_BULK;
2609 
2610 adjust_by_size:
2611 	/* If packet counts are 256 or greater we can assume we have a gross
2612 	 * overestimation of what the rate should be. Instead of trying to fine
2613 	 * tune it just use the formula below to try and dial in an exact value
2614 	 * give the current packet size of the frame.
2615 	 */
2616 	avg_wire_size = bytes / packets;
2617 
2618 	/* The following is a crude approximation of:
2619 	 *  wmem_default / (size + overhead) = desired_pkts_per_int
2620 	 *  rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
2621 	 *  (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
2622 	 *
2623 	 * Assuming wmem_default is 212992 and overhead is 640 bytes per
2624 	 * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
2625 	 * formula down to
2626 	 *
2627 	 *  (170 * (size + 24)) / (size + 640) = ITR
2628 	 *
2629 	 * We first do some math on the packet size and then finally bitshift
2630 	 * by 8 after rounding up. We also have to account for PCIe link speed
2631 	 * difference as ITR scales based on this.
2632 	 */
2633 	if (avg_wire_size <= 60) {
2634 		/* Start at 50k ints/sec */
2635 		avg_wire_size = 5120;
2636 	} else if (avg_wire_size <= 316) {
2637 		/* 50K ints/sec to 16K ints/sec */
2638 		avg_wire_size *= 40;
2639 		avg_wire_size += 2720;
2640 	} else if (avg_wire_size <= 1084) {
2641 		/* 16K ints/sec to 9.2K ints/sec */
2642 		avg_wire_size *= 15;
2643 		avg_wire_size += 11452;
2644 	} else if (avg_wire_size < 1968) {
2645 		/* 9.2K ints/sec to 8K ints/sec */
2646 		avg_wire_size *= 5;
2647 		avg_wire_size += 22420;
2648 	} else {
2649 		/* plateau at a limit of 8K ints/sec */
2650 		avg_wire_size = 32256;
2651 	}
2652 
2653 	/* If we are in low latency mode half our delay which doubles the rate
2654 	 * to somewhere between 100K to 16K ints/sec
2655 	 */
2656 	if (itr & IXGBE_ITR_ADAPTIVE_LATENCY)
2657 		avg_wire_size >>= 1;
2658 
2659 	/* Resultant value is 256 times larger than it needs to be. This
2660 	 * gives us room to adjust the value as needed to either increase
2661 	 * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
2662 	 *
2663 	 * Use addition as we have already recorded the new latency flag
2664 	 * for the ITR value.
2665 	 */
2666 	switch (q_vector->adapter->link_speed) {
2667 	case IXGBE_LINK_SPEED_10GB_FULL:
2668 	case IXGBE_LINK_SPEED_100_FULL:
2669 	default:
2670 		itr += DIV_ROUND_UP(avg_wire_size,
2671 				    IXGBE_ITR_ADAPTIVE_MIN_INC * 256) *
2672 		       IXGBE_ITR_ADAPTIVE_MIN_INC;
2673 		break;
2674 	case IXGBE_LINK_SPEED_2_5GB_FULL:
2675 	case IXGBE_LINK_SPEED_1GB_FULL:
2676 	case IXGBE_LINK_SPEED_10_FULL:
2677 		if (avg_wire_size > 8064)
2678 			avg_wire_size = 8064;
2679 		itr += DIV_ROUND_UP(avg_wire_size,
2680 				    IXGBE_ITR_ADAPTIVE_MIN_INC * 64) *
2681 		       IXGBE_ITR_ADAPTIVE_MIN_INC;
2682 		break;
2683 	}
2684 
2685 clear_counts:
2686 	/* write back value */
2687 	ring_container->itr = itr;
2688 
2689 	/* next update should occur within next jiffy */
2690 	ring_container->next_update = next_update + 1;
2691 
2692 	ring_container->total_bytes = 0;
2693 	ring_container->total_packets = 0;
2694 }
2695 
2696 /**
2697  * ixgbe_write_eitr - write EITR register in hardware specific way
2698  * @q_vector: structure containing interrupt and ring information
2699  *
2700  * This function is made to be called by ethtool and by the driver
2701  * when it needs to update EITR registers at runtime.  Hardware
2702  * specific quirks/differences are taken care of here.
2703  */
2704 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2705 {
2706 	struct ixgbe_adapter *adapter = q_vector->adapter;
2707 	struct ixgbe_hw *hw = &adapter->hw;
2708 	int v_idx = q_vector->v_idx;
2709 	u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2710 
2711 	switch (adapter->hw.mac.type) {
2712 	case ixgbe_mac_82598EB:
2713 		/* must write high and low 16 bits to reset counter */
2714 		itr_reg |= (itr_reg << 16);
2715 		break;
2716 	case ixgbe_mac_82599EB:
2717 	case ixgbe_mac_X540:
2718 	case ixgbe_mac_X550:
2719 	case ixgbe_mac_X550EM_x:
2720 	case ixgbe_mac_x550em_a:
2721 		/*
2722 		 * set the WDIS bit to not clear the timer bits and cause an
2723 		 * immediate assertion of the interrupt
2724 		 */
2725 		itr_reg |= IXGBE_EITR_CNT_WDIS;
2726 		break;
2727 	default:
2728 		break;
2729 	}
2730 	IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2731 }
2732 
2733 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2734 {
2735 	u32 new_itr;
2736 
2737 	ixgbe_update_itr(q_vector, &q_vector->tx);
2738 	ixgbe_update_itr(q_vector, &q_vector->rx);
2739 
2740 	/* use the smallest value of new ITR delay calculations */
2741 	new_itr = min(q_vector->rx.itr, q_vector->tx.itr);
2742 
2743 	/* Clear latency flag if set, shift into correct position */
2744 	new_itr &= ~IXGBE_ITR_ADAPTIVE_LATENCY;
2745 	new_itr <<= 2;
2746 
2747 	if (new_itr != q_vector->itr) {
2748 		/* save the algorithm value here */
2749 		q_vector->itr = new_itr;
2750 
2751 		ixgbe_write_eitr(q_vector);
2752 	}
2753 }
2754 
2755 /**
2756  * ixgbe_check_overtemp_subtask - check for over temperature
2757  * @adapter: pointer to adapter
2758  **/
2759 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2760 {
2761 	struct ixgbe_hw *hw = &adapter->hw;
2762 	u32 eicr = adapter->interrupt_event;
2763 	s32 rc;
2764 
2765 	if (test_bit(__IXGBE_DOWN, &adapter->state))
2766 		return;
2767 
2768 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2769 		return;
2770 
2771 	adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2772 
2773 	switch (hw->device_id) {
2774 	case IXGBE_DEV_ID_82599_T3_LOM:
2775 		/*
2776 		 * Since the warning interrupt is for both ports
2777 		 * we don't have to check if:
2778 		 *  - This interrupt wasn't for our port.
2779 		 *  - We may have missed the interrupt so always have to
2780 		 *    check if we  got a LSC
2781 		 */
2782 		if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2783 		    !(eicr & IXGBE_EICR_LSC))
2784 			return;
2785 
2786 		if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2787 			u32 speed;
2788 			bool link_up = false;
2789 
2790 			hw->mac.ops.check_link(hw, &speed, &link_up, false);
2791 
2792 			if (link_up)
2793 				return;
2794 		}
2795 
2796 		/* Check if this is not due to overtemp */
2797 		if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2798 			return;
2799 
2800 		break;
2801 	case IXGBE_DEV_ID_X550EM_A_1G_T:
2802 	case IXGBE_DEV_ID_X550EM_A_1G_T_L:
2803 		rc = hw->phy.ops.check_overtemp(hw);
2804 		if (rc != IXGBE_ERR_OVERTEMP)
2805 			return;
2806 		break;
2807 	default:
2808 		if (adapter->hw.mac.type >= ixgbe_mac_X540)
2809 			return;
2810 		if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2811 			return;
2812 		break;
2813 	}
2814 	e_crit(drv, "%s\n", ixgbe_overheat_msg);
2815 
2816 	adapter->interrupt_event = 0;
2817 }
2818 
2819 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2820 {
2821 	struct ixgbe_hw *hw = &adapter->hw;
2822 
2823 	if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2824 	    (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2825 		e_crit(probe, "Fan has stopped, replace the adapter\n");
2826 		/* write to clear the interrupt */
2827 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2828 	}
2829 }
2830 
2831 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2832 {
2833 	struct ixgbe_hw *hw = &adapter->hw;
2834 
2835 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2836 		return;
2837 
2838 	switch (adapter->hw.mac.type) {
2839 	case ixgbe_mac_82599EB:
2840 		/*
2841 		 * Need to check link state so complete overtemp check
2842 		 * on service task
2843 		 */
2844 		if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2845 		     (eicr & IXGBE_EICR_LSC)) &&
2846 		    (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2847 			adapter->interrupt_event = eicr;
2848 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2849 			ixgbe_service_event_schedule(adapter);
2850 			return;
2851 		}
2852 		return;
2853 	case ixgbe_mac_x550em_a:
2854 		if (eicr & IXGBE_EICR_GPI_SDP0_X550EM_a) {
2855 			adapter->interrupt_event = eicr;
2856 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2857 			ixgbe_service_event_schedule(adapter);
2858 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC,
2859 					IXGBE_EICR_GPI_SDP0_X550EM_a);
2860 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICR,
2861 					IXGBE_EICR_GPI_SDP0_X550EM_a);
2862 		}
2863 		return;
2864 	case ixgbe_mac_X550:
2865 	case ixgbe_mac_X540:
2866 		if (!(eicr & IXGBE_EICR_TS))
2867 			return;
2868 		break;
2869 	default:
2870 		return;
2871 	}
2872 
2873 	e_crit(drv, "%s\n", ixgbe_overheat_msg);
2874 }
2875 
2876 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2877 {
2878 	switch (hw->mac.type) {
2879 	case ixgbe_mac_82598EB:
2880 		if (hw->phy.type == ixgbe_phy_nl)
2881 			return true;
2882 		return false;
2883 	case ixgbe_mac_82599EB:
2884 	case ixgbe_mac_X550EM_x:
2885 	case ixgbe_mac_x550em_a:
2886 		switch (hw->mac.ops.get_media_type(hw)) {
2887 		case ixgbe_media_type_fiber:
2888 		case ixgbe_media_type_fiber_qsfp:
2889 			return true;
2890 		default:
2891 			return false;
2892 		}
2893 	default:
2894 		return false;
2895 	}
2896 }
2897 
2898 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2899 {
2900 	struct ixgbe_hw *hw = &adapter->hw;
2901 	u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw);
2902 
2903 	if (!ixgbe_is_sfp(hw))
2904 		return;
2905 
2906 	/* Later MAC's use different SDP */
2907 	if (hw->mac.type >= ixgbe_mac_X540)
2908 		eicr_mask = IXGBE_EICR_GPI_SDP0_X540;
2909 
2910 	if (eicr & eicr_mask) {
2911 		/* Clear the interrupt */
2912 		IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask);
2913 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2914 			adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2915 			adapter->sfp_poll_time = 0;
2916 			ixgbe_service_event_schedule(adapter);
2917 		}
2918 	}
2919 
2920 	if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
2921 	    (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2922 		/* Clear the interrupt */
2923 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2924 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2925 			adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2926 			ixgbe_service_event_schedule(adapter);
2927 		}
2928 	}
2929 }
2930 
2931 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2932 {
2933 	struct ixgbe_hw *hw = &adapter->hw;
2934 
2935 	adapter->lsc_int++;
2936 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2937 	adapter->link_check_timeout = jiffies;
2938 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2939 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2940 		IXGBE_WRITE_FLUSH(hw);
2941 		ixgbe_service_event_schedule(adapter);
2942 	}
2943 }
2944 
2945 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2946 					   u64 qmask)
2947 {
2948 	u32 mask;
2949 	struct ixgbe_hw *hw = &adapter->hw;
2950 
2951 	switch (hw->mac.type) {
2952 	case ixgbe_mac_82598EB:
2953 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2954 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2955 		break;
2956 	case ixgbe_mac_82599EB:
2957 	case ixgbe_mac_X540:
2958 	case ixgbe_mac_X550:
2959 	case ixgbe_mac_X550EM_x:
2960 	case ixgbe_mac_x550em_a:
2961 		mask = (qmask & 0xFFFFFFFF);
2962 		if (mask)
2963 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2964 		mask = (qmask >> 32);
2965 		if (mask)
2966 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2967 		break;
2968 	default:
2969 		break;
2970 	}
2971 	/* skip the flush */
2972 }
2973 
2974 /**
2975  * ixgbe_irq_enable - Enable default interrupt generation settings
2976  * @adapter: board private structure
2977  * @queues: enable irqs for queues
2978  * @flush: flush register write
2979  **/
2980 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2981 				    bool flush)
2982 {
2983 	struct ixgbe_hw *hw = &adapter->hw;
2984 	u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2985 
2986 	/* don't reenable LSC while waiting for link */
2987 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2988 		mask &= ~IXGBE_EIMS_LSC;
2989 
2990 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2991 		switch (adapter->hw.mac.type) {
2992 		case ixgbe_mac_82599EB:
2993 			mask |= IXGBE_EIMS_GPI_SDP0(hw);
2994 			break;
2995 		case ixgbe_mac_X540:
2996 		case ixgbe_mac_X550:
2997 		case ixgbe_mac_X550EM_x:
2998 		case ixgbe_mac_x550em_a:
2999 			mask |= IXGBE_EIMS_TS;
3000 			break;
3001 		default:
3002 			break;
3003 		}
3004 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3005 		mask |= IXGBE_EIMS_GPI_SDP1(hw);
3006 	switch (adapter->hw.mac.type) {
3007 	case ixgbe_mac_82599EB:
3008 		mask |= IXGBE_EIMS_GPI_SDP1(hw);
3009 		mask |= IXGBE_EIMS_GPI_SDP2(hw);
3010 		fallthrough;
3011 	case ixgbe_mac_X540:
3012 	case ixgbe_mac_X550:
3013 	case ixgbe_mac_X550EM_x:
3014 	case ixgbe_mac_x550em_a:
3015 		if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
3016 		    adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
3017 		    adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP_N)
3018 			mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw);
3019 		if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
3020 			mask |= IXGBE_EICR_GPI_SDP0_X540;
3021 		mask |= IXGBE_EIMS_ECC;
3022 		mask |= IXGBE_EIMS_MAILBOX;
3023 		break;
3024 	default:
3025 		break;
3026 	}
3027 
3028 	if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
3029 	    !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
3030 		mask |= IXGBE_EIMS_FLOW_DIR;
3031 
3032 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
3033 	if (queues)
3034 		ixgbe_irq_enable_queues(adapter, ~0);
3035 	if (flush)
3036 		IXGBE_WRITE_FLUSH(&adapter->hw);
3037 }
3038 
3039 static irqreturn_t ixgbe_msix_other(int irq, void *data)
3040 {
3041 	struct ixgbe_adapter *adapter = data;
3042 	struct ixgbe_hw *hw = &adapter->hw;
3043 	u32 eicr;
3044 
3045 	/*
3046 	 * Workaround for Silicon errata.  Use clear-by-write instead
3047 	 * of clear-by-read.  Reading with EICS will return the
3048 	 * interrupt causes without clearing, which later be done
3049 	 * with the write to EICR.
3050 	 */
3051 	eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
3052 
3053 	/* The lower 16bits of the EICR register are for the queue interrupts
3054 	 * which should be masked here in order to not accidentally clear them if
3055 	 * the bits are high when ixgbe_msix_other is called. There is a race
3056 	 * condition otherwise which results in possible performance loss
3057 	 * especially if the ixgbe_msix_other interrupt is triggering
3058 	 * consistently (as it would when PPS is turned on for the X540 device)
3059 	 */
3060 	eicr &= 0xFFFF0000;
3061 
3062 	IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
3063 
3064 	if (eicr & IXGBE_EICR_LSC)
3065 		ixgbe_check_lsc(adapter);
3066 
3067 	if (eicr & IXGBE_EICR_MAILBOX)
3068 		ixgbe_msg_task(adapter);
3069 
3070 	switch (hw->mac.type) {
3071 	case ixgbe_mac_82599EB:
3072 	case ixgbe_mac_X540:
3073 	case ixgbe_mac_X550:
3074 	case ixgbe_mac_X550EM_x:
3075 	case ixgbe_mac_x550em_a:
3076 		if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
3077 		    (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
3078 			adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
3079 			ixgbe_service_event_schedule(adapter);
3080 			IXGBE_WRITE_REG(hw, IXGBE_EICR,
3081 					IXGBE_EICR_GPI_SDP0_X540);
3082 		}
3083 		if (eicr & IXGBE_EICR_ECC) {
3084 			e_info(link, "Received ECC Err, initiating reset\n");
3085 			set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3086 			ixgbe_service_event_schedule(adapter);
3087 			IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3088 		}
3089 		/* Handle Flow Director Full threshold interrupt */
3090 		if (eicr & IXGBE_EICR_FLOW_DIR) {
3091 			int reinit_count = 0;
3092 			int i;
3093 			for (i = 0; i < adapter->num_tx_queues; i++) {
3094 				struct ixgbe_ring *ring = adapter->tx_ring[i];
3095 				if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
3096 						       &ring->state))
3097 					reinit_count++;
3098 			}
3099 			if (reinit_count) {
3100 				/* no more flow director interrupts until after init */
3101 				IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
3102 				adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
3103 				ixgbe_service_event_schedule(adapter);
3104 			}
3105 		}
3106 		ixgbe_check_sfp_event(adapter, eicr);
3107 		ixgbe_check_overtemp_event(adapter, eicr);
3108 		break;
3109 	default:
3110 		break;
3111 	}
3112 
3113 	ixgbe_check_fan_failure(adapter, eicr);
3114 
3115 	if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3116 		ixgbe_ptp_check_pps_event(adapter);
3117 
3118 	/* re-enable the original interrupt state, no lsc, no queues */
3119 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
3120 		ixgbe_irq_enable(adapter, false, false);
3121 
3122 	return IRQ_HANDLED;
3123 }
3124 
3125 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
3126 {
3127 	struct ixgbe_q_vector *q_vector = data;
3128 
3129 	/* EIAM disabled interrupts (on this vector) for us */
3130 
3131 	if (q_vector->rx.ring || q_vector->tx.ring)
3132 		napi_schedule_irqoff(&q_vector->napi);
3133 
3134 	return IRQ_HANDLED;
3135 }
3136 
3137 /**
3138  * ixgbe_poll - NAPI Rx polling callback
3139  * @napi: structure for representing this polling device
3140  * @budget: how many packets driver is allowed to clean
3141  *
3142  * This function is used for legacy and MSI, NAPI mode
3143  **/
3144 int ixgbe_poll(struct napi_struct *napi, int budget)
3145 {
3146 	struct ixgbe_q_vector *q_vector =
3147 				container_of(napi, struct ixgbe_q_vector, napi);
3148 	struct ixgbe_adapter *adapter = q_vector->adapter;
3149 	struct ixgbe_ring *ring;
3150 	int per_ring_budget, work_done = 0;
3151 	bool clean_complete = true;
3152 
3153 #ifdef CONFIG_IXGBE_DCA
3154 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3155 		ixgbe_update_dca(q_vector);
3156 #endif
3157 
3158 	ixgbe_for_each_ring(ring, q_vector->tx) {
3159 		bool wd = ring->xsk_umem ?
3160 			  ixgbe_clean_xdp_tx_irq(q_vector, ring, budget) :
3161 			  ixgbe_clean_tx_irq(q_vector, ring, budget);
3162 
3163 		if (!wd)
3164 			clean_complete = false;
3165 	}
3166 
3167 	/* Exit if we are called by netpoll */
3168 	if (budget <= 0)
3169 		return budget;
3170 
3171 	/* attempt to distribute budget to each queue fairly, but don't allow
3172 	 * the budget to go below 1 because we'll exit polling */
3173 	if (q_vector->rx.count > 1)
3174 		per_ring_budget = max(budget/q_vector->rx.count, 1);
3175 	else
3176 		per_ring_budget = budget;
3177 
3178 	ixgbe_for_each_ring(ring, q_vector->rx) {
3179 		int cleaned = ring->xsk_umem ?
3180 			      ixgbe_clean_rx_irq_zc(q_vector, ring,
3181 						    per_ring_budget) :
3182 			      ixgbe_clean_rx_irq(q_vector, ring,
3183 						 per_ring_budget);
3184 
3185 		work_done += cleaned;
3186 		if (cleaned >= per_ring_budget)
3187 			clean_complete = false;
3188 	}
3189 
3190 	/* If all work not completed, return budget and keep polling */
3191 	if (!clean_complete)
3192 		return budget;
3193 
3194 	/* all work done, exit the polling mode */
3195 	if (likely(napi_complete_done(napi, work_done))) {
3196 		if (adapter->rx_itr_setting & 1)
3197 			ixgbe_set_itr(q_vector);
3198 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
3199 			ixgbe_irq_enable_queues(adapter,
3200 						BIT_ULL(q_vector->v_idx));
3201 	}
3202 
3203 	return min(work_done, budget - 1);
3204 }
3205 
3206 /**
3207  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
3208  * @adapter: board private structure
3209  *
3210  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
3211  * interrupts from the kernel.
3212  **/
3213 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
3214 {
3215 	struct net_device *netdev = adapter->netdev;
3216 	unsigned int ri = 0, ti = 0;
3217 	int vector, err;
3218 
3219 	for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3220 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3221 		struct msix_entry *entry = &adapter->msix_entries[vector];
3222 
3223 		if (q_vector->tx.ring && q_vector->rx.ring) {
3224 			snprintf(q_vector->name, sizeof(q_vector->name),
3225 				 "%s-TxRx-%u", netdev->name, ri++);
3226 			ti++;
3227 		} else if (q_vector->rx.ring) {
3228 			snprintf(q_vector->name, sizeof(q_vector->name),
3229 				 "%s-rx-%u", netdev->name, ri++);
3230 		} else if (q_vector->tx.ring) {
3231 			snprintf(q_vector->name, sizeof(q_vector->name),
3232 				 "%s-tx-%u", netdev->name, ti++);
3233 		} else {
3234 			/* skip this unused q_vector */
3235 			continue;
3236 		}
3237 		err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
3238 				  q_vector->name, q_vector);
3239 		if (err) {
3240 			e_err(probe, "request_irq failed for MSIX interrupt "
3241 			      "Error: %d\n", err);
3242 			goto free_queue_irqs;
3243 		}
3244 		/* If Flow Director is enabled, set interrupt affinity */
3245 		if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3246 			/* assign the mask for this irq */
3247 			irq_set_affinity_hint(entry->vector,
3248 					      &q_vector->affinity_mask);
3249 		}
3250 	}
3251 
3252 	err = request_irq(adapter->msix_entries[vector].vector,
3253 			  ixgbe_msix_other, 0, netdev->name, adapter);
3254 	if (err) {
3255 		e_err(probe, "request_irq for msix_other failed: %d\n", err);
3256 		goto free_queue_irqs;
3257 	}
3258 
3259 	return 0;
3260 
3261 free_queue_irqs:
3262 	while (vector) {
3263 		vector--;
3264 		irq_set_affinity_hint(adapter->msix_entries[vector].vector,
3265 				      NULL);
3266 		free_irq(adapter->msix_entries[vector].vector,
3267 			 adapter->q_vector[vector]);
3268 	}
3269 	adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3270 	pci_disable_msix(adapter->pdev);
3271 	kfree(adapter->msix_entries);
3272 	adapter->msix_entries = NULL;
3273 	return err;
3274 }
3275 
3276 /**
3277  * ixgbe_intr - legacy mode Interrupt Handler
3278  * @irq: interrupt number
3279  * @data: pointer to a network interface device structure
3280  **/
3281 static irqreturn_t ixgbe_intr(int irq, void *data)
3282 {
3283 	struct ixgbe_adapter *adapter = data;
3284 	struct ixgbe_hw *hw = &adapter->hw;
3285 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3286 	u32 eicr;
3287 
3288 	/*
3289 	 * Workaround for silicon errata #26 on 82598.  Mask the interrupt
3290 	 * before the read of EICR.
3291 	 */
3292 	IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
3293 
3294 	/* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
3295 	 * therefore no explicit interrupt disable is necessary */
3296 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
3297 	if (!eicr) {
3298 		/*
3299 		 * shared interrupt alert!
3300 		 * make sure interrupts are enabled because the read will
3301 		 * have disabled interrupts due to EIAM
3302 		 * finish the workaround of silicon errata on 82598.  Unmask
3303 		 * the interrupt that we masked before the EICR read.
3304 		 */
3305 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
3306 			ixgbe_irq_enable(adapter, true, true);
3307 		return IRQ_NONE;	/* Not our interrupt */
3308 	}
3309 
3310 	if (eicr & IXGBE_EICR_LSC)
3311 		ixgbe_check_lsc(adapter);
3312 
3313 	switch (hw->mac.type) {
3314 	case ixgbe_mac_82599EB:
3315 		ixgbe_check_sfp_event(adapter, eicr);
3316 		fallthrough;
3317 	case ixgbe_mac_X540:
3318 	case ixgbe_mac_X550:
3319 	case ixgbe_mac_X550EM_x:
3320 	case ixgbe_mac_x550em_a:
3321 		if (eicr & IXGBE_EICR_ECC) {
3322 			e_info(link, "Received ECC Err, initiating reset\n");
3323 			set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3324 			ixgbe_service_event_schedule(adapter);
3325 			IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3326 		}
3327 		ixgbe_check_overtemp_event(adapter, eicr);
3328 		break;
3329 	default:
3330 		break;
3331 	}
3332 
3333 	ixgbe_check_fan_failure(adapter, eicr);
3334 	if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3335 		ixgbe_ptp_check_pps_event(adapter);
3336 
3337 	/* would disable interrupts here but EIAM disabled it */
3338 	napi_schedule_irqoff(&q_vector->napi);
3339 
3340 	/*
3341 	 * re-enable link(maybe) and non-queue interrupts, no flush.
3342 	 * ixgbe_poll will re-enable the queue interrupts
3343 	 */
3344 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
3345 		ixgbe_irq_enable(adapter, false, false);
3346 
3347 	return IRQ_HANDLED;
3348 }
3349 
3350 /**
3351  * ixgbe_request_irq - initialize interrupts
3352  * @adapter: board private structure
3353  *
3354  * Attempts to configure interrupts using the best available
3355  * capabilities of the hardware and kernel.
3356  **/
3357 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
3358 {
3359 	struct net_device *netdev = adapter->netdev;
3360 	int err;
3361 
3362 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3363 		err = ixgbe_request_msix_irqs(adapter);
3364 	else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
3365 		err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
3366 				  netdev->name, adapter);
3367 	else
3368 		err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
3369 				  netdev->name, adapter);
3370 
3371 	if (err)
3372 		e_err(probe, "request_irq failed, Error %d\n", err);
3373 
3374 	return err;
3375 }
3376 
3377 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
3378 {
3379 	int vector;
3380 
3381 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
3382 		free_irq(adapter->pdev->irq, adapter);
3383 		return;
3384 	}
3385 
3386 	if (!adapter->msix_entries)
3387 		return;
3388 
3389 	for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3390 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3391 		struct msix_entry *entry = &adapter->msix_entries[vector];
3392 
3393 		/* free only the irqs that were actually requested */
3394 		if (!q_vector->rx.ring && !q_vector->tx.ring)
3395 			continue;
3396 
3397 		/* clear the affinity_mask in the IRQ descriptor */
3398 		irq_set_affinity_hint(entry->vector, NULL);
3399 
3400 		free_irq(entry->vector, q_vector);
3401 	}
3402 
3403 	free_irq(adapter->msix_entries[vector].vector, adapter);
3404 }
3405 
3406 /**
3407  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3408  * @adapter: board private structure
3409  **/
3410 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3411 {
3412 	switch (adapter->hw.mac.type) {
3413 	case ixgbe_mac_82598EB:
3414 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3415 		break;
3416 	case ixgbe_mac_82599EB:
3417 	case ixgbe_mac_X540:
3418 	case ixgbe_mac_X550:
3419 	case ixgbe_mac_X550EM_x:
3420 	case ixgbe_mac_x550em_a:
3421 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3422 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3423 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3424 		break;
3425 	default:
3426 		break;
3427 	}
3428 	IXGBE_WRITE_FLUSH(&adapter->hw);
3429 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3430 		int vector;
3431 
3432 		for (vector = 0; vector < adapter->num_q_vectors; vector++)
3433 			synchronize_irq(adapter->msix_entries[vector].vector);
3434 
3435 		synchronize_irq(adapter->msix_entries[vector++].vector);
3436 	} else {
3437 		synchronize_irq(adapter->pdev->irq);
3438 	}
3439 }
3440 
3441 /**
3442  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3443  * @adapter: board private structure
3444  *
3445  **/
3446 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3447 {
3448 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3449 
3450 	ixgbe_write_eitr(q_vector);
3451 
3452 	ixgbe_set_ivar(adapter, 0, 0, 0);
3453 	ixgbe_set_ivar(adapter, 1, 0, 0);
3454 
3455 	e_info(hw, "Legacy interrupt IVAR setup done\n");
3456 }
3457 
3458 /**
3459  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3460  * @adapter: board private structure
3461  * @ring: structure containing ring specific data
3462  *
3463  * Configure the Tx descriptor ring after a reset.
3464  **/
3465 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3466 			     struct ixgbe_ring *ring)
3467 {
3468 	struct ixgbe_hw *hw = &adapter->hw;
3469 	u64 tdba = ring->dma;
3470 	int wait_loop = 10;
3471 	u32 txdctl = IXGBE_TXDCTL_ENABLE;
3472 	u8 reg_idx = ring->reg_idx;
3473 
3474 	ring->xsk_umem = NULL;
3475 	if (ring_is_xdp(ring))
3476 		ring->xsk_umem = ixgbe_xsk_umem(adapter, ring);
3477 
3478 	/* disable queue to avoid issues while updating state */
3479 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3480 	IXGBE_WRITE_FLUSH(hw);
3481 
3482 	IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3483 			(tdba & DMA_BIT_MASK(32)));
3484 	IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3485 	IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3486 			ring->count * sizeof(union ixgbe_adv_tx_desc));
3487 	IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3488 	IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3489 	ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3490 
3491 	/*
3492 	 * set WTHRESH to encourage burst writeback, it should not be set
3493 	 * higher than 1 when:
3494 	 * - ITR is 0 as it could cause false TX hangs
3495 	 * - ITR is set to > 100k int/sec and BQL is enabled
3496 	 *
3497 	 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3498 	 * to or less than the number of on chip descriptors, which is
3499 	 * currently 40.
3500 	 */
3501 	if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3502 		txdctl |= 1u << 16;	/* WTHRESH = 1 */
3503 	else
3504 		txdctl |= 8u << 16;	/* WTHRESH = 8 */
3505 
3506 	/*
3507 	 * Setting PTHRESH to 32 both improves performance
3508 	 * and avoids a TX hang with DFP enabled
3509 	 */
3510 	txdctl |= (1u << 8) |	/* HTHRESH = 1 */
3511 		   32;		/* PTHRESH = 32 */
3512 
3513 	/* reinitialize flowdirector state */
3514 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3515 		ring->atr_sample_rate = adapter->atr_sample_rate;
3516 		ring->atr_count = 0;
3517 		set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3518 	} else {
3519 		ring->atr_sample_rate = 0;
3520 	}
3521 
3522 	/* initialize XPS */
3523 	if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3524 		struct ixgbe_q_vector *q_vector = ring->q_vector;
3525 
3526 		if (q_vector)
3527 			netif_set_xps_queue(ring->netdev,
3528 					    &q_vector->affinity_mask,
3529 					    ring->queue_index);
3530 	}
3531 
3532 	clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3533 
3534 	/* reinitialize tx_buffer_info */
3535 	memset(ring->tx_buffer_info, 0,
3536 	       sizeof(struct ixgbe_tx_buffer) * ring->count);
3537 
3538 	/* enable queue */
3539 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3540 
3541 	/* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3542 	if (hw->mac.type == ixgbe_mac_82598EB &&
3543 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3544 		return;
3545 
3546 	/* poll to verify queue is enabled */
3547 	do {
3548 		usleep_range(1000, 2000);
3549 		txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3550 	} while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3551 	if (!wait_loop)
3552 		hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
3553 }
3554 
3555 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3556 {
3557 	struct ixgbe_hw *hw = &adapter->hw;
3558 	u32 rttdcs, mtqc;
3559 	u8 tcs = adapter->hw_tcs;
3560 
3561 	if (hw->mac.type == ixgbe_mac_82598EB)
3562 		return;
3563 
3564 	/* disable the arbiter while setting MTQC */
3565 	rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3566 	rttdcs |= IXGBE_RTTDCS_ARBDIS;
3567 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3568 
3569 	/* set transmit pool layout */
3570 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3571 		mtqc = IXGBE_MTQC_VT_ENA;
3572 		if (tcs > 4)
3573 			mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3574 		else if (tcs > 1)
3575 			mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3576 		else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3577 			 IXGBE_82599_VMDQ_4Q_MASK)
3578 			mtqc |= IXGBE_MTQC_32VF;
3579 		else
3580 			mtqc |= IXGBE_MTQC_64VF;
3581 	} else {
3582 		if (tcs > 4) {
3583 			mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3584 		} else if (tcs > 1) {
3585 			mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3586 		} else {
3587 			u8 max_txq = adapter->num_tx_queues +
3588 				adapter->num_xdp_queues;
3589 			if (max_txq > 63)
3590 				mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3591 			else
3592 				mtqc = IXGBE_MTQC_64Q_1PB;
3593 		}
3594 	}
3595 
3596 	IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3597 
3598 	/* Enable Security TX Buffer IFG for multiple pb */
3599 	if (tcs) {
3600 		u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3601 		sectx |= IXGBE_SECTX_DCB;
3602 		IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3603 	}
3604 
3605 	/* re-enable the arbiter */
3606 	rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3607 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3608 }
3609 
3610 /**
3611  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3612  * @adapter: board private structure
3613  *
3614  * Configure the Tx unit of the MAC after a reset.
3615  **/
3616 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3617 {
3618 	struct ixgbe_hw *hw = &adapter->hw;
3619 	u32 dmatxctl;
3620 	u32 i;
3621 
3622 	ixgbe_setup_mtqc(adapter);
3623 
3624 	if (hw->mac.type != ixgbe_mac_82598EB) {
3625 		/* DMATXCTL.EN must be before Tx queues are enabled */
3626 		dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3627 		dmatxctl |= IXGBE_DMATXCTL_TE;
3628 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3629 	}
3630 
3631 	/* Setup the HW Tx Head and Tail descriptor pointers */
3632 	for (i = 0; i < adapter->num_tx_queues; i++)
3633 		ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3634 	for (i = 0; i < adapter->num_xdp_queues; i++)
3635 		ixgbe_configure_tx_ring(adapter, adapter->xdp_ring[i]);
3636 }
3637 
3638 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3639 				 struct ixgbe_ring *ring)
3640 {
3641 	struct ixgbe_hw *hw = &adapter->hw;
3642 	u8 reg_idx = ring->reg_idx;
3643 	u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3644 
3645 	srrctl |= IXGBE_SRRCTL_DROP_EN;
3646 
3647 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3648 }
3649 
3650 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3651 				  struct ixgbe_ring *ring)
3652 {
3653 	struct ixgbe_hw *hw = &adapter->hw;
3654 	u8 reg_idx = ring->reg_idx;
3655 	u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3656 
3657 	srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3658 
3659 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3660 }
3661 
3662 #ifdef CONFIG_IXGBE_DCB
3663 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3664 #else
3665 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3666 #endif
3667 {
3668 	int i;
3669 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3670 
3671 	if (adapter->ixgbe_ieee_pfc)
3672 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3673 
3674 	/*
3675 	 * We should set the drop enable bit if:
3676 	 *  SR-IOV is enabled
3677 	 *   or
3678 	 *  Number of Rx queues > 1 and flow control is disabled
3679 	 *
3680 	 *  This allows us to avoid head of line blocking for security
3681 	 *  and performance reasons.
3682 	 */
3683 	if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3684 	    !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3685 		for (i = 0; i < adapter->num_rx_queues; i++)
3686 			ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3687 	} else {
3688 		for (i = 0; i < adapter->num_rx_queues; i++)
3689 			ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3690 	}
3691 }
3692 
3693 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3694 
3695 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3696 				   struct ixgbe_ring *rx_ring)
3697 {
3698 	struct ixgbe_hw *hw = &adapter->hw;
3699 	u32 srrctl;
3700 	u8 reg_idx = rx_ring->reg_idx;
3701 
3702 	if (hw->mac.type == ixgbe_mac_82598EB) {
3703 		u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3704 
3705 		/*
3706 		 * if VMDq is not active we must program one srrctl register
3707 		 * per RSS queue since we have enabled RDRXCTL.MVMEN
3708 		 */
3709 		reg_idx &= mask;
3710 	}
3711 
3712 	/* configure header buffer length, needed for RSC */
3713 	srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3714 
3715 	/* configure the packet buffer length */
3716 	if (rx_ring->xsk_umem) {
3717 		u32 xsk_buf_len = xsk_umem_get_rx_frame_size(rx_ring->xsk_umem);
3718 
3719 		/* If the MAC support setting RXDCTL.RLPML, the
3720 		 * SRRCTL[n].BSIZEPKT is set to PAGE_SIZE and
3721 		 * RXDCTL.RLPML is set to the actual UMEM buffer
3722 		 * size. If not, then we are stuck with a 1k buffer
3723 		 * size resolution. In this case frames larger than
3724 		 * the UMEM buffer size viewed in a 1k resolution will
3725 		 * be dropped.
3726 		 */
3727 		if (hw->mac.type != ixgbe_mac_82599EB)
3728 			srrctl |= PAGE_SIZE >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3729 		else
3730 			srrctl |= xsk_buf_len >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3731 	} else if (test_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state)) {
3732 		srrctl |= IXGBE_RXBUFFER_3K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3733 	} else {
3734 		srrctl |= IXGBE_RXBUFFER_2K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3735 	}
3736 
3737 	/* configure descriptor type */
3738 	srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3739 
3740 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3741 }
3742 
3743 /**
3744  * ixgbe_rss_indir_tbl_entries - Return RSS indirection table entries
3745  * @adapter: device handle
3746  *
3747  *  - 82598/82599/X540:     128
3748  *  - X550(non-SRIOV mode): 512
3749  *  - X550(SRIOV mode):     64
3750  */
3751 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3752 {
3753 	if (adapter->hw.mac.type < ixgbe_mac_X550)
3754 		return 128;
3755 	else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3756 		return 64;
3757 	else
3758 		return 512;
3759 }
3760 
3761 /**
3762  * ixgbe_store_key - Write the RSS key to HW
3763  * @adapter: device handle
3764  *
3765  * Write the RSS key stored in adapter.rss_key to HW.
3766  */
3767 void ixgbe_store_key(struct ixgbe_adapter *adapter)
3768 {
3769 	struct ixgbe_hw *hw = &adapter->hw;
3770 	int i;
3771 
3772 	for (i = 0; i < 10; i++)
3773 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3774 }
3775 
3776 /**
3777  * ixgbe_init_rss_key - Initialize adapter RSS key
3778  * @adapter: device handle
3779  *
3780  * Allocates and initializes the RSS key if it is not allocated.
3781  **/
3782 static inline int ixgbe_init_rss_key(struct ixgbe_adapter *adapter)
3783 {
3784 	u32 *rss_key;
3785 
3786 	if (!adapter->rss_key) {
3787 		rss_key = kzalloc(IXGBE_RSS_KEY_SIZE, GFP_KERNEL);
3788 		if (unlikely(!rss_key))
3789 			return -ENOMEM;
3790 
3791 		netdev_rss_key_fill(rss_key, IXGBE_RSS_KEY_SIZE);
3792 		adapter->rss_key = rss_key;
3793 	}
3794 
3795 	return 0;
3796 }
3797 
3798 /**
3799  * ixgbe_store_reta - Write the RETA table to HW
3800  * @adapter: device handle
3801  *
3802  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3803  */
3804 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3805 {
3806 	u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3807 	struct ixgbe_hw *hw = &adapter->hw;
3808 	u32 reta = 0;
3809 	u32 indices_multi;
3810 	u8 *indir_tbl = adapter->rss_indir_tbl;
3811 
3812 	/* Fill out the redirection table as follows:
3813 	 *  - 82598:      8 bit wide entries containing pair of 4 bit RSS
3814 	 *    indices.
3815 	 *  - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3816 	 *  - X550:       8 bit wide entries containing 6 bit RSS index
3817 	 */
3818 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3819 		indices_multi = 0x11;
3820 	else
3821 		indices_multi = 0x1;
3822 
3823 	/* Write redirection table to HW */
3824 	for (i = 0; i < reta_entries; i++) {
3825 		reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3826 		if ((i & 3) == 3) {
3827 			if (i < 128)
3828 				IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3829 			else
3830 				IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3831 						reta);
3832 			reta = 0;
3833 		}
3834 	}
3835 }
3836 
3837 /**
3838  * ixgbe_store_vfreta - Write the RETA table to HW (x550 devices in SRIOV mode)
3839  * @adapter: device handle
3840  *
3841  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3842  */
3843 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3844 {
3845 	u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3846 	struct ixgbe_hw *hw = &adapter->hw;
3847 	u32 vfreta = 0;
3848 
3849 	/* Write redirection table to HW */
3850 	for (i = 0; i < reta_entries; i++) {
3851 		u16 pool = adapter->num_rx_pools;
3852 
3853 		vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3854 		if ((i & 3) != 3)
3855 			continue;
3856 
3857 		while (pool--)
3858 			IXGBE_WRITE_REG(hw,
3859 					IXGBE_PFVFRETA(i >> 2, VMDQ_P(pool)),
3860 					vfreta);
3861 		vfreta = 0;
3862 	}
3863 }
3864 
3865 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3866 {
3867 	u32 i, j;
3868 	u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3869 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3870 
3871 	/* Program table for at least 4 queues w/ SR-IOV so that VFs can
3872 	 * make full use of any rings they may have.  We will use the
3873 	 * PSRTYPE register to control how many rings we use within the PF.
3874 	 */
3875 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 4))
3876 		rss_i = 4;
3877 
3878 	/* Fill out hash function seeds */
3879 	ixgbe_store_key(adapter);
3880 
3881 	/* Fill out redirection table */
3882 	memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3883 
3884 	for (i = 0, j = 0; i < reta_entries; i++, j++) {
3885 		if (j == rss_i)
3886 			j = 0;
3887 
3888 		adapter->rss_indir_tbl[i] = j;
3889 	}
3890 
3891 	ixgbe_store_reta(adapter);
3892 }
3893 
3894 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3895 {
3896 	struct ixgbe_hw *hw = &adapter->hw;
3897 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3898 	int i, j;
3899 
3900 	/* Fill out hash function seeds */
3901 	for (i = 0; i < 10; i++) {
3902 		u16 pool = adapter->num_rx_pools;
3903 
3904 		while (pool--)
3905 			IXGBE_WRITE_REG(hw,
3906 					IXGBE_PFVFRSSRK(i, VMDQ_P(pool)),
3907 					*(adapter->rss_key + i));
3908 	}
3909 
3910 	/* Fill out the redirection table */
3911 	for (i = 0, j = 0; i < 64; i++, j++) {
3912 		if (j == rss_i)
3913 			j = 0;
3914 
3915 		adapter->rss_indir_tbl[i] = j;
3916 	}
3917 
3918 	ixgbe_store_vfreta(adapter);
3919 }
3920 
3921 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3922 {
3923 	struct ixgbe_hw *hw = &adapter->hw;
3924 	u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3925 	u32 rxcsum;
3926 
3927 	/* Disable indicating checksum in descriptor, enables RSS hash */
3928 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3929 	rxcsum |= IXGBE_RXCSUM_PCSD;
3930 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3931 
3932 	if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3933 		if (adapter->ring_feature[RING_F_RSS].mask)
3934 			mrqc = IXGBE_MRQC_RSSEN;
3935 	} else {
3936 		u8 tcs = adapter->hw_tcs;
3937 
3938 		if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3939 			if (tcs > 4)
3940 				mrqc = IXGBE_MRQC_VMDQRT8TCEN;	/* 8 TCs */
3941 			else if (tcs > 1)
3942 				mrqc = IXGBE_MRQC_VMDQRT4TCEN;	/* 4 TCs */
3943 			else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3944 				 IXGBE_82599_VMDQ_4Q_MASK)
3945 				mrqc = IXGBE_MRQC_VMDQRSS32EN;
3946 			else
3947 				mrqc = IXGBE_MRQC_VMDQRSS64EN;
3948 
3949 			/* Enable L3/L4 for Tx Switched packets only for X550,
3950 			 * older devices do not support this feature
3951 			 */
3952 			if (hw->mac.type >= ixgbe_mac_X550)
3953 				mrqc |= IXGBE_MRQC_L3L4TXSWEN;
3954 		} else {
3955 			if (tcs > 4)
3956 				mrqc = IXGBE_MRQC_RTRSS8TCEN;
3957 			else if (tcs > 1)
3958 				mrqc = IXGBE_MRQC_RTRSS4TCEN;
3959 			else
3960 				mrqc = IXGBE_MRQC_RSSEN;
3961 		}
3962 	}
3963 
3964 	/* Perform hash on these packet types */
3965 	rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3966 		     IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3967 		     IXGBE_MRQC_RSS_FIELD_IPV6 |
3968 		     IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3969 
3970 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3971 		rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3972 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3973 		rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3974 
3975 	if ((hw->mac.type >= ixgbe_mac_X550) &&
3976 	    (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
3977 		u16 pool = adapter->num_rx_pools;
3978 
3979 		/* Enable VF RSS mode */
3980 		mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
3981 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3982 
3983 		/* Setup RSS through the VF registers */
3984 		ixgbe_setup_vfreta(adapter);
3985 		vfmrqc = IXGBE_MRQC_RSSEN;
3986 		vfmrqc |= rss_field;
3987 
3988 		while (pool--)
3989 			IXGBE_WRITE_REG(hw,
3990 					IXGBE_PFVFMRQC(VMDQ_P(pool)),
3991 					vfmrqc);
3992 	} else {
3993 		ixgbe_setup_reta(adapter);
3994 		mrqc |= rss_field;
3995 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3996 	}
3997 }
3998 
3999 /**
4000  * ixgbe_configure_rscctl - enable RSC for the indicated ring
4001  * @adapter: address of board private structure
4002  * @ring: structure containing ring specific data
4003  **/
4004 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
4005 				   struct ixgbe_ring *ring)
4006 {
4007 	struct ixgbe_hw *hw = &adapter->hw;
4008 	u32 rscctrl;
4009 	u8 reg_idx = ring->reg_idx;
4010 
4011 	if (!ring_is_rsc_enabled(ring))
4012 		return;
4013 
4014 	rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
4015 	rscctrl |= IXGBE_RSCCTL_RSCEN;
4016 	/*
4017 	 * we must limit the number of descriptors so that the
4018 	 * total size of max desc * buf_len is not greater
4019 	 * than 65536
4020 	 */
4021 	rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
4022 	IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
4023 }
4024 
4025 #define IXGBE_MAX_RX_DESC_POLL 10
4026 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
4027 				       struct ixgbe_ring *ring)
4028 {
4029 	struct ixgbe_hw *hw = &adapter->hw;
4030 	int wait_loop = IXGBE_MAX_RX_DESC_POLL;
4031 	u32 rxdctl;
4032 	u8 reg_idx = ring->reg_idx;
4033 
4034 	if (ixgbe_removed(hw->hw_addr))
4035 		return;
4036 	/* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
4037 	if (hw->mac.type == ixgbe_mac_82598EB &&
4038 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
4039 		return;
4040 
4041 	do {
4042 		usleep_range(1000, 2000);
4043 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4044 	} while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4045 
4046 	if (!wait_loop) {
4047 		e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
4048 		      "the polling period\n", reg_idx);
4049 	}
4050 }
4051 
4052 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
4053 			     struct ixgbe_ring *ring)
4054 {
4055 	struct ixgbe_hw *hw = &adapter->hw;
4056 	union ixgbe_adv_rx_desc *rx_desc;
4057 	u64 rdba = ring->dma;
4058 	u32 rxdctl;
4059 	u8 reg_idx = ring->reg_idx;
4060 
4061 	xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
4062 	ring->xsk_umem = ixgbe_xsk_umem(adapter, ring);
4063 	if (ring->xsk_umem) {
4064 		WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4065 						   MEM_TYPE_XSK_BUFF_POOL,
4066 						   NULL));
4067 		xsk_buff_set_rxq_info(ring->xsk_umem, &ring->xdp_rxq);
4068 	} else {
4069 		WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4070 						   MEM_TYPE_PAGE_SHARED, NULL));
4071 	}
4072 
4073 	/* disable queue to avoid use of these values while updating state */
4074 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4075 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4076 
4077 	/* write value back with RXDCTL.ENABLE bit cleared */
4078 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4079 	IXGBE_WRITE_FLUSH(hw);
4080 
4081 	IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
4082 	IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
4083 	IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
4084 			ring->count * sizeof(union ixgbe_adv_rx_desc));
4085 	/* Force flushing of IXGBE_RDLEN to prevent MDD */
4086 	IXGBE_WRITE_FLUSH(hw);
4087 
4088 	IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
4089 	IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
4090 	ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
4091 
4092 	ixgbe_configure_srrctl(adapter, ring);
4093 	ixgbe_configure_rscctl(adapter, ring);
4094 
4095 	if (hw->mac.type == ixgbe_mac_82598EB) {
4096 		/*
4097 		 * enable cache line friendly hardware writes:
4098 		 * PTHRESH=32 descriptors (half the internal cache),
4099 		 * this also removes ugly rx_no_buffer_count increment
4100 		 * HTHRESH=4 descriptors (to minimize latency on fetch)
4101 		 * WTHRESH=8 burst writeback up to two cache lines
4102 		 */
4103 		rxdctl &= ~0x3FFFFF;
4104 		rxdctl |=  0x080420;
4105 #if (PAGE_SIZE < 8192)
4106 	/* RXDCTL.RLPML does not work on 82599 */
4107 	} else if (hw->mac.type != ixgbe_mac_82599EB) {
4108 		rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4109 			    IXGBE_RXDCTL_RLPML_EN);
4110 
4111 		/* Limit the maximum frame size so we don't overrun the skb.
4112 		 * This can happen in SRIOV mode when the MTU of the VF is
4113 		 * higher than the MTU of the PF.
4114 		 */
4115 		if (ring_uses_build_skb(ring) &&
4116 		    !test_bit(__IXGBE_RX_3K_BUFFER, &ring->state))
4117 			rxdctl |= IXGBE_MAX_2K_FRAME_BUILD_SKB |
4118 				  IXGBE_RXDCTL_RLPML_EN;
4119 #endif
4120 	}
4121 
4122 	if (ring->xsk_umem && hw->mac.type != ixgbe_mac_82599EB) {
4123 		u32 xsk_buf_len = xsk_umem_get_rx_frame_size(ring->xsk_umem);
4124 
4125 		rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4126 			    IXGBE_RXDCTL_RLPML_EN);
4127 		rxdctl |= xsk_buf_len | IXGBE_RXDCTL_RLPML_EN;
4128 
4129 		ring->rx_buf_len = xsk_buf_len;
4130 	}
4131 
4132 	/* initialize rx_buffer_info */
4133 	memset(ring->rx_buffer_info, 0,
4134 	       sizeof(struct ixgbe_rx_buffer) * ring->count);
4135 
4136 	/* initialize Rx descriptor 0 */
4137 	rx_desc = IXGBE_RX_DESC(ring, 0);
4138 	rx_desc->wb.upper.length = 0;
4139 
4140 	/* enable receive descriptor ring */
4141 	rxdctl |= IXGBE_RXDCTL_ENABLE;
4142 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4143 
4144 	ixgbe_rx_desc_queue_enable(adapter, ring);
4145 	if (ring->xsk_umem)
4146 		ixgbe_alloc_rx_buffers_zc(ring, ixgbe_desc_unused(ring));
4147 	else
4148 		ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
4149 }
4150 
4151 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
4152 {
4153 	struct ixgbe_hw *hw = &adapter->hw;
4154 	int rss_i = adapter->ring_feature[RING_F_RSS].indices;
4155 	u16 pool = adapter->num_rx_pools;
4156 
4157 	/* PSRTYPE must be initialized in non 82598 adapters */
4158 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4159 		      IXGBE_PSRTYPE_UDPHDR |
4160 		      IXGBE_PSRTYPE_IPV4HDR |
4161 		      IXGBE_PSRTYPE_L2HDR |
4162 		      IXGBE_PSRTYPE_IPV6HDR;
4163 
4164 	if (hw->mac.type == ixgbe_mac_82598EB)
4165 		return;
4166 
4167 	if (rss_i > 3)
4168 		psrtype |= 2u << 29;
4169 	else if (rss_i > 1)
4170 		psrtype |= 1u << 29;
4171 
4172 	while (pool--)
4173 		IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4174 }
4175 
4176 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
4177 {
4178 	struct ixgbe_hw *hw = &adapter->hw;
4179 	u16 pool = adapter->num_rx_pools;
4180 	u32 reg_offset, vf_shift, vmolr;
4181 	u32 gcr_ext, vmdctl;
4182 	int i;
4183 
4184 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
4185 		return;
4186 
4187 	vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
4188 	vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
4189 	vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
4190 	vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
4191 	vmdctl |= IXGBE_VT_CTL_REPLEN;
4192 	IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
4193 
4194 	/* accept untagged packets until a vlan tag is
4195 	 * specifically set for the VMDQ queue/pool
4196 	 */
4197 	vmolr = IXGBE_VMOLR_AUPE;
4198 	while (pool--)
4199 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(pool)), vmolr);
4200 
4201 	vf_shift = VMDQ_P(0) % 32;
4202 	reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
4203 
4204 	/* Enable only the PF's pool for Tx/Rx */
4205 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), GENMASK(31, vf_shift));
4206 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
4207 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), GENMASK(31, vf_shift));
4208 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
4209 	if (adapter->bridge_mode == BRIDGE_MODE_VEB)
4210 		IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
4211 
4212 	/* Map PF MAC address in RAR Entry 0 to first pool following VFs */
4213 	hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
4214 
4215 	/* clear VLAN promisc flag so VFTA will be updated if necessary */
4216 	adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4217 
4218 	/*
4219 	 * Set up VF register offsets for selected VT Mode,
4220 	 * i.e. 32 or 64 VFs for SR-IOV
4221 	 */
4222 	switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4223 	case IXGBE_82599_VMDQ_8Q_MASK:
4224 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
4225 		break;
4226 	case IXGBE_82599_VMDQ_4Q_MASK:
4227 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
4228 		break;
4229 	default:
4230 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
4231 		break;
4232 	}
4233 
4234 	IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
4235 
4236 	for (i = 0; i < adapter->num_vfs; i++) {
4237 		/* configure spoof checking */
4238 		ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i,
4239 					  adapter->vfinfo[i].spoofchk_enabled);
4240 
4241 		/* Enable/Disable RSS query feature  */
4242 		ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
4243 					  adapter->vfinfo[i].rss_query_enabled);
4244 	}
4245 }
4246 
4247 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
4248 {
4249 	struct ixgbe_hw *hw = &adapter->hw;
4250 	struct net_device *netdev = adapter->netdev;
4251 	int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4252 	struct ixgbe_ring *rx_ring;
4253 	int i;
4254 	u32 mhadd, hlreg0;
4255 
4256 #ifdef IXGBE_FCOE
4257 	/* adjust max frame to be able to do baby jumbo for FCoE */
4258 	if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
4259 	    (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
4260 		max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4261 
4262 #endif /* IXGBE_FCOE */
4263 
4264 	/* adjust max frame to be at least the size of a standard frame */
4265 	if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
4266 		max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
4267 
4268 	mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
4269 	if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
4270 		mhadd &= ~IXGBE_MHADD_MFS_MASK;
4271 		mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
4272 
4273 		IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
4274 	}
4275 
4276 	hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4277 	/* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
4278 	hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4279 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4280 
4281 	/*
4282 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
4283 	 * the Base and Length of the Rx Descriptor Ring
4284 	 */
4285 	for (i = 0; i < adapter->num_rx_queues; i++) {
4286 		rx_ring = adapter->rx_ring[i];
4287 
4288 		clear_ring_rsc_enabled(rx_ring);
4289 		clear_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4290 		clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4291 
4292 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4293 			set_ring_rsc_enabled(rx_ring);
4294 
4295 		if (test_bit(__IXGBE_RX_FCOE, &rx_ring->state))
4296 			set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4297 
4298 		if (adapter->flags2 & IXGBE_FLAG2_RX_LEGACY)
4299 			continue;
4300 
4301 		set_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4302 
4303 #if (PAGE_SIZE < 8192)
4304 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4305 			set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4306 
4307 		if (IXGBE_2K_TOO_SMALL_WITH_PADDING ||
4308 		    (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
4309 			set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4310 #endif
4311 	}
4312 }
4313 
4314 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
4315 {
4316 	struct ixgbe_hw *hw = &adapter->hw;
4317 	u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4318 
4319 	switch (hw->mac.type) {
4320 	case ixgbe_mac_82598EB:
4321 		/*
4322 		 * For VMDq support of different descriptor types or
4323 		 * buffer sizes through the use of multiple SRRCTL
4324 		 * registers, RDRXCTL.MVMEN must be set to 1
4325 		 *
4326 		 * also, the manual doesn't mention it clearly but DCA hints
4327 		 * will only use queue 0's tags unless this bit is set.  Side
4328 		 * effects of setting this bit are only that SRRCTL must be
4329 		 * fully programmed [0..15]
4330 		 */
4331 		rdrxctl |= IXGBE_RDRXCTL_MVMEN;
4332 		break;
4333 	case ixgbe_mac_X550:
4334 	case ixgbe_mac_X550EM_x:
4335 	case ixgbe_mac_x550em_a:
4336 		if (adapter->num_vfs)
4337 			rdrxctl |= IXGBE_RDRXCTL_PSP;
4338 		fallthrough;
4339 	case ixgbe_mac_82599EB:
4340 	case ixgbe_mac_X540:
4341 		/* Disable RSC for ACK packets */
4342 		IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
4343 		   (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
4344 		rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4345 		/* hardware requires some bits to be set by default */
4346 		rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
4347 		rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4348 		break;
4349 	default:
4350 		/* We should do nothing since we don't know this hardware */
4351 		return;
4352 	}
4353 
4354 	IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4355 }
4356 
4357 /**
4358  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
4359  * @adapter: board private structure
4360  *
4361  * Configure the Rx unit of the MAC after a reset.
4362  **/
4363 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
4364 {
4365 	struct ixgbe_hw *hw = &adapter->hw;
4366 	int i;
4367 	u32 rxctrl, rfctl;
4368 
4369 	/* disable receives while setting up the descriptors */
4370 	hw->mac.ops.disable_rx(hw);
4371 
4372 	ixgbe_setup_psrtype(adapter);
4373 	ixgbe_setup_rdrxctl(adapter);
4374 
4375 	/* RSC Setup */
4376 	rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4377 	rfctl &= ~IXGBE_RFCTL_RSC_DIS;
4378 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
4379 		rfctl |= IXGBE_RFCTL_RSC_DIS;
4380 
4381 	/* disable NFS filtering */
4382 	rfctl |= (IXGBE_RFCTL_NFSW_DIS | IXGBE_RFCTL_NFSR_DIS);
4383 	IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4384 
4385 	/* Program registers for the distribution of queues */
4386 	ixgbe_setup_mrqc(adapter);
4387 
4388 	/* set_rx_buffer_len must be called before ring initialization */
4389 	ixgbe_set_rx_buffer_len(adapter);
4390 
4391 	/*
4392 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
4393 	 * the Base and Length of the Rx Descriptor Ring
4394 	 */
4395 	for (i = 0; i < adapter->num_rx_queues; i++)
4396 		ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
4397 
4398 	rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4399 	/* disable drop enable for 82598 parts */
4400 	if (hw->mac.type == ixgbe_mac_82598EB)
4401 		rxctrl |= IXGBE_RXCTRL_DMBYPS;
4402 
4403 	/* enable all receives */
4404 	rxctrl |= IXGBE_RXCTRL_RXEN;
4405 	hw->mac.ops.enable_rx_dma(hw, rxctrl);
4406 }
4407 
4408 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
4409 				 __be16 proto, u16 vid)
4410 {
4411 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4412 	struct ixgbe_hw *hw = &adapter->hw;
4413 
4414 	/* add VID to filter table */
4415 	if (!vid || !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4416 		hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true, !!vid);
4417 
4418 	set_bit(vid, adapter->active_vlans);
4419 
4420 	return 0;
4421 }
4422 
4423 static int ixgbe_find_vlvf_entry(struct ixgbe_hw *hw, u32 vlan)
4424 {
4425 	u32 vlvf;
4426 	int idx;
4427 
4428 	/* short cut the special case */
4429 	if (vlan == 0)
4430 		return 0;
4431 
4432 	/* Search for the vlan id in the VLVF entries */
4433 	for (idx = IXGBE_VLVF_ENTRIES; --idx;) {
4434 		vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(idx));
4435 		if ((vlvf & VLAN_VID_MASK) == vlan)
4436 			break;
4437 	}
4438 
4439 	return idx;
4440 }
4441 
4442 void ixgbe_update_pf_promisc_vlvf(struct ixgbe_adapter *adapter, u32 vid)
4443 {
4444 	struct ixgbe_hw *hw = &adapter->hw;
4445 	u32 bits, word;
4446 	int idx;
4447 
4448 	idx = ixgbe_find_vlvf_entry(hw, vid);
4449 	if (!idx)
4450 		return;
4451 
4452 	/* See if any other pools are set for this VLAN filter
4453 	 * entry other than the PF.
4454 	 */
4455 	word = idx * 2 + (VMDQ_P(0) / 32);
4456 	bits = ~BIT(VMDQ_P(0) % 32);
4457 	bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4458 
4459 	/* Disable the filter so this falls into the default pool. */
4460 	if (!bits && !IXGBE_READ_REG(hw, IXGBE_VLVFB(word ^ 1))) {
4461 		if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4462 			IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), 0);
4463 		IXGBE_WRITE_REG(hw, IXGBE_VLVF(idx), 0);
4464 	}
4465 }
4466 
4467 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
4468 				  __be16 proto, u16 vid)
4469 {
4470 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4471 	struct ixgbe_hw *hw = &adapter->hw;
4472 
4473 	/* remove VID from filter table */
4474 	if (vid && !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4475 		hw->mac.ops.set_vfta(hw, vid, VMDQ_P(0), false, true);
4476 
4477 	clear_bit(vid, adapter->active_vlans);
4478 
4479 	return 0;
4480 }
4481 
4482 /**
4483  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
4484  * @adapter: driver data
4485  */
4486 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
4487 {
4488 	struct ixgbe_hw *hw = &adapter->hw;
4489 	u32 vlnctrl;
4490 	int i, j;
4491 
4492 	switch (hw->mac.type) {
4493 	case ixgbe_mac_82598EB:
4494 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4495 		vlnctrl &= ~IXGBE_VLNCTRL_VME;
4496 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4497 		break;
4498 	case ixgbe_mac_82599EB:
4499 	case ixgbe_mac_X540:
4500 	case ixgbe_mac_X550:
4501 	case ixgbe_mac_X550EM_x:
4502 	case ixgbe_mac_x550em_a:
4503 		for (i = 0; i < adapter->num_rx_queues; i++) {
4504 			struct ixgbe_ring *ring = adapter->rx_ring[i];
4505 
4506 			if (!netif_is_ixgbe(ring->netdev))
4507 				continue;
4508 
4509 			j = ring->reg_idx;
4510 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4511 			vlnctrl &= ~IXGBE_RXDCTL_VME;
4512 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4513 		}
4514 		break;
4515 	default:
4516 		break;
4517 	}
4518 }
4519 
4520 /**
4521  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
4522  * @adapter: driver data
4523  */
4524 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
4525 {
4526 	struct ixgbe_hw *hw = &adapter->hw;
4527 	u32 vlnctrl;
4528 	int i, j;
4529 
4530 	switch (hw->mac.type) {
4531 	case ixgbe_mac_82598EB:
4532 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4533 		vlnctrl |= IXGBE_VLNCTRL_VME;
4534 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4535 		break;
4536 	case ixgbe_mac_82599EB:
4537 	case ixgbe_mac_X540:
4538 	case ixgbe_mac_X550:
4539 	case ixgbe_mac_X550EM_x:
4540 	case ixgbe_mac_x550em_a:
4541 		for (i = 0; i < adapter->num_rx_queues; i++) {
4542 			struct ixgbe_ring *ring = adapter->rx_ring[i];
4543 
4544 			if (!netif_is_ixgbe(ring->netdev))
4545 				continue;
4546 
4547 			j = ring->reg_idx;
4548 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4549 			vlnctrl |= IXGBE_RXDCTL_VME;
4550 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4551 		}
4552 		break;
4553 	default:
4554 		break;
4555 	}
4556 }
4557 
4558 static void ixgbe_vlan_promisc_enable(struct ixgbe_adapter *adapter)
4559 {
4560 	struct ixgbe_hw *hw = &adapter->hw;
4561 	u32 vlnctrl, i;
4562 
4563 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4564 
4565 	if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
4566 	/* For VMDq and SR-IOV we must leave VLAN filtering enabled */
4567 		vlnctrl |= IXGBE_VLNCTRL_VFE;
4568 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4569 	} else {
4570 		vlnctrl &= ~IXGBE_VLNCTRL_VFE;
4571 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4572 		return;
4573 	}
4574 
4575 	/* Nothing to do for 82598 */
4576 	if (hw->mac.type == ixgbe_mac_82598EB)
4577 		return;
4578 
4579 	/* We are already in VLAN promisc, nothing to do */
4580 	if (adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)
4581 		return;
4582 
4583 	/* Set flag so we don't redo unnecessary work */
4584 	adapter->flags2 |= IXGBE_FLAG2_VLAN_PROMISC;
4585 
4586 	/* Add PF to all active pools */
4587 	for (i = IXGBE_VLVF_ENTRIES; --i;) {
4588 		u32 reg_offset = IXGBE_VLVFB(i * 2 + VMDQ_P(0) / 32);
4589 		u32 vlvfb = IXGBE_READ_REG(hw, reg_offset);
4590 
4591 		vlvfb |= BIT(VMDQ_P(0) % 32);
4592 		IXGBE_WRITE_REG(hw, reg_offset, vlvfb);
4593 	}
4594 
4595 	/* Set all bits in the VLAN filter table array */
4596 	for (i = hw->mac.vft_size; i--;)
4597 		IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), ~0U);
4598 }
4599 
4600 #define VFTA_BLOCK_SIZE 8
4601 static void ixgbe_scrub_vfta(struct ixgbe_adapter *adapter, u32 vfta_offset)
4602 {
4603 	struct ixgbe_hw *hw = &adapter->hw;
4604 	u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4605 	u32 vid_start = vfta_offset * 32;
4606 	u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4607 	u32 i, vid, word, bits;
4608 
4609 	for (i = IXGBE_VLVF_ENTRIES; --i;) {
4610 		u32 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(i));
4611 
4612 		/* pull VLAN ID from VLVF */
4613 		vid = vlvf & VLAN_VID_MASK;
4614 
4615 		/* only concern outselves with a certain range */
4616 		if (vid < vid_start || vid >= vid_end)
4617 			continue;
4618 
4619 		if (vlvf) {
4620 			/* record VLAN ID in VFTA */
4621 			vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4622 
4623 			/* if PF is part of this then continue */
4624 			if (test_bit(vid, adapter->active_vlans))
4625 				continue;
4626 		}
4627 
4628 		/* remove PF from the pool */
4629 		word = i * 2 + VMDQ_P(0) / 32;
4630 		bits = ~BIT(VMDQ_P(0) % 32);
4631 		bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4632 		IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), bits);
4633 	}
4634 
4635 	/* extract values from active_vlans and write back to VFTA */
4636 	for (i = VFTA_BLOCK_SIZE; i--;) {
4637 		vid = (vfta_offset + i) * 32;
4638 		word = vid / BITS_PER_LONG;
4639 		bits = vid % BITS_PER_LONG;
4640 
4641 		vfta[i] |= adapter->active_vlans[word] >> bits;
4642 
4643 		IXGBE_WRITE_REG(hw, IXGBE_VFTA(vfta_offset + i), vfta[i]);
4644 	}
4645 }
4646 
4647 static void ixgbe_vlan_promisc_disable(struct ixgbe_adapter *adapter)
4648 {
4649 	struct ixgbe_hw *hw = &adapter->hw;
4650 	u32 vlnctrl, i;
4651 
4652 	/* Set VLAN filtering to enabled */
4653 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4654 	vlnctrl |= IXGBE_VLNCTRL_VFE;
4655 	IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4656 
4657 	if (!(adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) ||
4658 	    hw->mac.type == ixgbe_mac_82598EB)
4659 		return;
4660 
4661 	/* We are not in VLAN promisc, nothing to do */
4662 	if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4663 		return;
4664 
4665 	/* Set flag so we don't redo unnecessary work */
4666 	adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4667 
4668 	for (i = 0; i < hw->mac.vft_size; i += VFTA_BLOCK_SIZE)
4669 		ixgbe_scrub_vfta(adapter, i);
4670 }
4671 
4672 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
4673 {
4674 	u16 vid = 1;
4675 
4676 	ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
4677 
4678 	for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
4679 		ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
4680 }
4681 
4682 /**
4683  * ixgbe_write_mc_addr_list - write multicast addresses to MTA
4684  * @netdev: network interface device structure
4685  *
4686  * Writes multicast address list to the MTA hash table.
4687  * Returns: -ENOMEM on failure
4688  *                0 on no addresses written
4689  *                X on writing X addresses to MTA
4690  **/
4691 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
4692 {
4693 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4694 	struct ixgbe_hw *hw = &adapter->hw;
4695 
4696 	if (!netif_running(netdev))
4697 		return 0;
4698 
4699 	if (hw->mac.ops.update_mc_addr_list)
4700 		hw->mac.ops.update_mc_addr_list(hw, netdev);
4701 	else
4702 		return -ENOMEM;
4703 
4704 #ifdef CONFIG_PCI_IOV
4705 	ixgbe_restore_vf_multicasts(adapter);
4706 #endif
4707 
4708 	return netdev_mc_count(netdev);
4709 }
4710 
4711 #ifdef CONFIG_PCI_IOV
4712 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4713 {
4714 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4715 	struct ixgbe_hw *hw = &adapter->hw;
4716 	int i;
4717 
4718 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4719 		mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4720 
4721 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4722 			hw->mac.ops.set_rar(hw, i,
4723 					    mac_table->addr,
4724 					    mac_table->pool,
4725 					    IXGBE_RAH_AV);
4726 		else
4727 			hw->mac.ops.clear_rar(hw, i);
4728 	}
4729 }
4730 
4731 #endif
4732 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4733 {
4734 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4735 	struct ixgbe_hw *hw = &adapter->hw;
4736 	int i;
4737 
4738 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4739 		if (!(mac_table->state & IXGBE_MAC_STATE_MODIFIED))
4740 			continue;
4741 
4742 		mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4743 
4744 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4745 			hw->mac.ops.set_rar(hw, i,
4746 					    mac_table->addr,
4747 					    mac_table->pool,
4748 					    IXGBE_RAH_AV);
4749 		else
4750 			hw->mac.ops.clear_rar(hw, i);
4751 	}
4752 }
4753 
4754 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4755 {
4756 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4757 	struct ixgbe_hw *hw = &adapter->hw;
4758 	int i;
4759 
4760 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4761 		mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4762 		mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4763 	}
4764 
4765 	ixgbe_sync_mac_table(adapter);
4766 }
4767 
4768 static int ixgbe_available_rars(struct ixgbe_adapter *adapter, u16 pool)
4769 {
4770 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4771 	struct ixgbe_hw *hw = &adapter->hw;
4772 	int i, count = 0;
4773 
4774 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4775 		/* do not count default RAR as available */
4776 		if (mac_table->state & IXGBE_MAC_STATE_DEFAULT)
4777 			continue;
4778 
4779 		/* only count unused and addresses that belong to us */
4780 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE) {
4781 			if (mac_table->pool != pool)
4782 				continue;
4783 		}
4784 
4785 		count++;
4786 	}
4787 
4788 	return count;
4789 }
4790 
4791 /* this function destroys the first RAR entry */
4792 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter)
4793 {
4794 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4795 	struct ixgbe_hw *hw = &adapter->hw;
4796 
4797 	memcpy(&mac_table->addr, hw->mac.addr, ETH_ALEN);
4798 	mac_table->pool = VMDQ_P(0);
4799 
4800 	mac_table->state = IXGBE_MAC_STATE_DEFAULT | IXGBE_MAC_STATE_IN_USE;
4801 
4802 	hw->mac.ops.set_rar(hw, 0, mac_table->addr, mac_table->pool,
4803 			    IXGBE_RAH_AV);
4804 }
4805 
4806 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter,
4807 			 const u8 *addr, u16 pool)
4808 {
4809 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4810 	struct ixgbe_hw *hw = &adapter->hw;
4811 	int i;
4812 
4813 	if (is_zero_ether_addr(addr))
4814 		return -EINVAL;
4815 
4816 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4817 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4818 			continue;
4819 
4820 		ether_addr_copy(mac_table->addr, addr);
4821 		mac_table->pool = pool;
4822 
4823 		mac_table->state |= IXGBE_MAC_STATE_MODIFIED |
4824 				    IXGBE_MAC_STATE_IN_USE;
4825 
4826 		ixgbe_sync_mac_table(adapter);
4827 
4828 		return i;
4829 	}
4830 
4831 	return -ENOMEM;
4832 }
4833 
4834 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter,
4835 			 const u8 *addr, u16 pool)
4836 {
4837 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4838 	struct ixgbe_hw *hw = &adapter->hw;
4839 	int i;
4840 
4841 	if (is_zero_ether_addr(addr))
4842 		return -EINVAL;
4843 
4844 	/* search table for addr, if found clear IN_USE flag and sync */
4845 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4846 		/* we can only delete an entry if it is in use */
4847 		if (!(mac_table->state & IXGBE_MAC_STATE_IN_USE))
4848 			continue;
4849 		/* we only care about entries that belong to the given pool */
4850 		if (mac_table->pool != pool)
4851 			continue;
4852 		/* we only care about a specific MAC address */
4853 		if (!ether_addr_equal(addr, mac_table->addr))
4854 			continue;
4855 
4856 		mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4857 		mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4858 
4859 		ixgbe_sync_mac_table(adapter);
4860 
4861 		return 0;
4862 	}
4863 
4864 	return -ENOMEM;
4865 }
4866 
4867 static int ixgbe_uc_sync(struct net_device *netdev, const unsigned char *addr)
4868 {
4869 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4870 	int ret;
4871 
4872 	ret = ixgbe_add_mac_filter(adapter, addr, VMDQ_P(0));
4873 
4874 	return min_t(int, ret, 0);
4875 }
4876 
4877 static int ixgbe_uc_unsync(struct net_device *netdev, const unsigned char *addr)
4878 {
4879 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4880 
4881 	ixgbe_del_mac_filter(adapter, addr, VMDQ_P(0));
4882 
4883 	return 0;
4884 }
4885 
4886 /**
4887  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4888  * @netdev: network interface device structure
4889  *
4890  * The set_rx_method entry point is called whenever the unicast/multicast
4891  * address list or the network interface flags are updated.  This routine is
4892  * responsible for configuring the hardware for proper unicast, multicast and
4893  * promiscuous mode.
4894  **/
4895 void ixgbe_set_rx_mode(struct net_device *netdev)
4896 {
4897 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4898 	struct ixgbe_hw *hw = &adapter->hw;
4899 	u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4900 	netdev_features_t features = netdev->features;
4901 	int count;
4902 
4903 	/* Check for Promiscuous and All Multicast modes */
4904 	fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4905 
4906 	/* set all bits that we expect to always be set */
4907 	fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4908 	fctrl |= IXGBE_FCTRL_BAM;
4909 	fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4910 	fctrl |= IXGBE_FCTRL_PMCF;
4911 
4912 	/* clear the bits we are changing the status of */
4913 	fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4914 	if (netdev->flags & IFF_PROMISC) {
4915 		hw->addr_ctrl.user_set_promisc = true;
4916 		fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4917 		vmolr |= IXGBE_VMOLR_MPE;
4918 		features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4919 	} else {
4920 		if (netdev->flags & IFF_ALLMULTI) {
4921 			fctrl |= IXGBE_FCTRL_MPE;
4922 			vmolr |= IXGBE_VMOLR_MPE;
4923 		}
4924 		hw->addr_ctrl.user_set_promisc = false;
4925 	}
4926 
4927 	/*
4928 	 * Write addresses to available RAR registers, if there is not
4929 	 * sufficient space to store all the addresses then enable
4930 	 * unicast promiscuous mode
4931 	 */
4932 	if (__dev_uc_sync(netdev, ixgbe_uc_sync, ixgbe_uc_unsync)) {
4933 		fctrl |= IXGBE_FCTRL_UPE;
4934 		vmolr |= IXGBE_VMOLR_ROPE;
4935 	}
4936 
4937 	/* Write addresses to the MTA, if the attempt fails
4938 	 * then we should just turn on promiscuous mode so
4939 	 * that we can at least receive multicast traffic
4940 	 */
4941 	count = ixgbe_write_mc_addr_list(netdev);
4942 	if (count < 0) {
4943 		fctrl |= IXGBE_FCTRL_MPE;
4944 		vmolr |= IXGBE_VMOLR_MPE;
4945 	} else if (count) {
4946 		vmolr |= IXGBE_VMOLR_ROMPE;
4947 	}
4948 
4949 	if (hw->mac.type != ixgbe_mac_82598EB) {
4950 		vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4951 			 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4952 			   IXGBE_VMOLR_ROPE);
4953 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4954 	}
4955 
4956 	/* This is useful for sniffing bad packets. */
4957 	if (features & NETIF_F_RXALL) {
4958 		/* UPE and MPE will be handled by normal PROMISC logic
4959 		 * in e1000e_set_rx_mode */
4960 		fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4961 			  IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4962 			  IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4963 
4964 		fctrl &= ~(IXGBE_FCTRL_DPF);
4965 		/* NOTE:  VLAN filtering is disabled by setting PROMISC */
4966 	}
4967 
4968 	IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4969 
4970 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
4971 		ixgbe_vlan_strip_enable(adapter);
4972 	else
4973 		ixgbe_vlan_strip_disable(adapter);
4974 
4975 	if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
4976 		ixgbe_vlan_promisc_disable(adapter);
4977 	else
4978 		ixgbe_vlan_promisc_enable(adapter);
4979 }
4980 
4981 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
4982 {
4983 	int q_idx;
4984 
4985 	for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4986 		napi_enable(&adapter->q_vector[q_idx]->napi);
4987 }
4988 
4989 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
4990 {
4991 	int q_idx;
4992 
4993 	for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4994 		napi_disable(&adapter->q_vector[q_idx]->napi);
4995 }
4996 
4997 static int ixgbe_udp_tunnel_sync(struct net_device *dev, unsigned int table)
4998 {
4999 	struct ixgbe_adapter *adapter = netdev_priv(dev);
5000 	struct ixgbe_hw *hw = &adapter->hw;
5001 	struct udp_tunnel_info ti;
5002 
5003 	udp_tunnel_nic_get_port(dev, table, 0, &ti);
5004 	if (ti.type == UDP_TUNNEL_TYPE_VXLAN)
5005 		adapter->vxlan_port = ti.port;
5006 	else
5007 		adapter->geneve_port = ti.port;
5008 
5009 	IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL,
5010 			ntohs(adapter->vxlan_port) |
5011 			ntohs(adapter->geneve_port) <<
5012 				IXGBE_VXLANCTRL_GENEVE_UDPPORT_SHIFT);
5013 	return 0;
5014 }
5015 
5016 static const struct udp_tunnel_nic_info ixgbe_udp_tunnels_x550 = {
5017 	.sync_table	= ixgbe_udp_tunnel_sync,
5018 	.flags		= UDP_TUNNEL_NIC_INFO_IPV4_ONLY,
5019 	.tables		= {
5020 		{ .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN,  },
5021 	},
5022 };
5023 
5024 static const struct udp_tunnel_nic_info ixgbe_udp_tunnels_x550em_a = {
5025 	.sync_table	= ixgbe_udp_tunnel_sync,
5026 	.flags		= UDP_TUNNEL_NIC_INFO_IPV4_ONLY,
5027 	.tables		= {
5028 		{ .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN,  },
5029 		{ .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_GENEVE, },
5030 	},
5031 };
5032 
5033 #ifdef CONFIG_IXGBE_DCB
5034 /**
5035  * ixgbe_configure_dcb - Configure DCB hardware
5036  * @adapter: ixgbe adapter struct
5037  *
5038  * This is called by the driver on open to configure the DCB hardware.
5039  * This is also called by the gennetlink interface when reconfiguring
5040  * the DCB state.
5041  */
5042 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
5043 {
5044 	struct ixgbe_hw *hw = &adapter->hw;
5045 	int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
5046 
5047 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
5048 		if (hw->mac.type == ixgbe_mac_82598EB)
5049 			netif_set_gso_max_size(adapter->netdev, 65536);
5050 		return;
5051 	}
5052 
5053 	if (hw->mac.type == ixgbe_mac_82598EB)
5054 		netif_set_gso_max_size(adapter->netdev, 32768);
5055 
5056 #ifdef IXGBE_FCOE
5057 	if (adapter->netdev->features & NETIF_F_FCOE_MTU)
5058 		max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
5059 #endif
5060 
5061 	/* reconfigure the hardware */
5062 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
5063 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5064 						DCB_TX_CONFIG);
5065 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5066 						DCB_RX_CONFIG);
5067 		ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
5068 	} else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
5069 		ixgbe_dcb_hw_ets(&adapter->hw,
5070 				 adapter->ixgbe_ieee_ets,
5071 				 max_frame);
5072 		ixgbe_dcb_hw_pfc_config(&adapter->hw,
5073 					adapter->ixgbe_ieee_pfc->pfc_en,
5074 					adapter->ixgbe_ieee_ets->prio_tc);
5075 	}
5076 
5077 	/* Enable RSS Hash per TC */
5078 	if (hw->mac.type != ixgbe_mac_82598EB) {
5079 		u32 msb = 0;
5080 		u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
5081 
5082 		while (rss_i) {
5083 			msb++;
5084 			rss_i >>= 1;
5085 		}
5086 
5087 		/* write msb to all 8 TCs in one write */
5088 		IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
5089 	}
5090 }
5091 #endif
5092 
5093 /* Additional bittime to account for IXGBE framing */
5094 #define IXGBE_ETH_FRAMING 20
5095 
5096 /**
5097  * ixgbe_hpbthresh - calculate high water mark for flow control
5098  *
5099  * @adapter: board private structure to calculate for
5100  * @pb: packet buffer to calculate
5101  */
5102 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
5103 {
5104 	struct ixgbe_hw *hw = &adapter->hw;
5105 	struct net_device *dev = adapter->netdev;
5106 	int link, tc, kb, marker;
5107 	u32 dv_id, rx_pba;
5108 
5109 	/* Calculate max LAN frame size */
5110 	tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
5111 
5112 #ifdef IXGBE_FCOE
5113 	/* FCoE traffic class uses FCOE jumbo frames */
5114 	if ((dev->features & NETIF_F_FCOE_MTU) &&
5115 	    (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5116 	    (pb == ixgbe_fcoe_get_tc(adapter)))
5117 		tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5118 #endif
5119 
5120 	/* Calculate delay value for device */
5121 	switch (hw->mac.type) {
5122 	case ixgbe_mac_X540:
5123 	case ixgbe_mac_X550:
5124 	case ixgbe_mac_X550EM_x:
5125 	case ixgbe_mac_x550em_a:
5126 		dv_id = IXGBE_DV_X540(link, tc);
5127 		break;
5128 	default:
5129 		dv_id = IXGBE_DV(link, tc);
5130 		break;
5131 	}
5132 
5133 	/* Loopback switch introduces additional latency */
5134 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5135 		dv_id += IXGBE_B2BT(tc);
5136 
5137 	/* Delay value is calculated in bit times convert to KB */
5138 	kb = IXGBE_BT2KB(dv_id);
5139 	rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
5140 
5141 	marker = rx_pba - kb;
5142 
5143 	/* It is possible that the packet buffer is not large enough
5144 	 * to provide required headroom. In this case throw an error
5145 	 * to user and a do the best we can.
5146 	 */
5147 	if (marker < 0) {
5148 		e_warn(drv, "Packet Buffer(%i) can not provide enough"
5149 			    "headroom to support flow control."
5150 			    "Decrease MTU or number of traffic classes\n", pb);
5151 		marker = tc + 1;
5152 	}
5153 
5154 	return marker;
5155 }
5156 
5157 /**
5158  * ixgbe_lpbthresh - calculate low water mark for for flow control
5159  *
5160  * @adapter: board private structure to calculate for
5161  * @pb: packet buffer to calculate
5162  */
5163 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
5164 {
5165 	struct ixgbe_hw *hw = &adapter->hw;
5166 	struct net_device *dev = adapter->netdev;
5167 	int tc;
5168 	u32 dv_id;
5169 
5170 	/* Calculate max LAN frame size */
5171 	tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5172 
5173 #ifdef IXGBE_FCOE
5174 	/* FCoE traffic class uses FCOE jumbo frames */
5175 	if ((dev->features & NETIF_F_FCOE_MTU) &&
5176 	    (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5177 	    (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
5178 		tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5179 #endif
5180 
5181 	/* Calculate delay value for device */
5182 	switch (hw->mac.type) {
5183 	case ixgbe_mac_X540:
5184 	case ixgbe_mac_X550:
5185 	case ixgbe_mac_X550EM_x:
5186 	case ixgbe_mac_x550em_a:
5187 		dv_id = IXGBE_LOW_DV_X540(tc);
5188 		break;
5189 	default:
5190 		dv_id = IXGBE_LOW_DV(tc);
5191 		break;
5192 	}
5193 
5194 	/* Delay value is calculated in bit times convert to KB */
5195 	return IXGBE_BT2KB(dv_id);
5196 }
5197 
5198 /*
5199  * ixgbe_pbthresh_setup - calculate and setup high low water marks
5200  */
5201 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
5202 {
5203 	struct ixgbe_hw *hw = &adapter->hw;
5204 	int num_tc = adapter->hw_tcs;
5205 	int i;
5206 
5207 	if (!num_tc)
5208 		num_tc = 1;
5209 
5210 	for (i = 0; i < num_tc; i++) {
5211 		hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
5212 		hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
5213 
5214 		/* Low water marks must not be larger than high water marks */
5215 		if (hw->fc.low_water[i] > hw->fc.high_water[i])
5216 			hw->fc.low_water[i] = 0;
5217 	}
5218 
5219 	for (; i < MAX_TRAFFIC_CLASS; i++)
5220 		hw->fc.high_water[i] = 0;
5221 }
5222 
5223 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
5224 {
5225 	struct ixgbe_hw *hw = &adapter->hw;
5226 	int hdrm;
5227 	u8 tc = adapter->hw_tcs;
5228 
5229 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5230 	    adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5231 		hdrm = 32 << adapter->fdir_pballoc;
5232 	else
5233 		hdrm = 0;
5234 
5235 	hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
5236 	ixgbe_pbthresh_setup(adapter);
5237 }
5238 
5239 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
5240 {
5241 	struct ixgbe_hw *hw = &adapter->hw;
5242 	struct hlist_node *node2;
5243 	struct ixgbe_fdir_filter *filter;
5244 	u8 queue;
5245 
5246 	spin_lock(&adapter->fdir_perfect_lock);
5247 
5248 	if (!hlist_empty(&adapter->fdir_filter_list))
5249 		ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
5250 
5251 	hlist_for_each_entry_safe(filter, node2,
5252 				  &adapter->fdir_filter_list, fdir_node) {
5253 		if (filter->action == IXGBE_FDIR_DROP_QUEUE) {
5254 			queue = IXGBE_FDIR_DROP_QUEUE;
5255 		} else {
5256 			u32 ring = ethtool_get_flow_spec_ring(filter->action);
5257 			u8 vf = ethtool_get_flow_spec_ring_vf(filter->action);
5258 
5259 			if (!vf && (ring >= adapter->num_rx_queues)) {
5260 				e_err(drv, "FDIR restore failed without VF, ring: %u\n",
5261 				      ring);
5262 				continue;
5263 			} else if (vf &&
5264 				   ((vf > adapter->num_vfs) ||
5265 				     ring >= adapter->num_rx_queues_per_pool)) {
5266 				e_err(drv, "FDIR restore failed with VF, vf: %hhu, ring: %u\n",
5267 				      vf, ring);
5268 				continue;
5269 			}
5270 
5271 			/* Map the ring onto the absolute queue index */
5272 			if (!vf)
5273 				queue = adapter->rx_ring[ring]->reg_idx;
5274 			else
5275 				queue = ((vf - 1) *
5276 					adapter->num_rx_queues_per_pool) + ring;
5277 		}
5278 
5279 		ixgbe_fdir_write_perfect_filter_82599(hw,
5280 				&filter->filter, filter->sw_idx, queue);
5281 	}
5282 
5283 	spin_unlock(&adapter->fdir_perfect_lock);
5284 }
5285 
5286 /**
5287  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
5288  * @rx_ring: ring to free buffers from
5289  **/
5290 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
5291 {
5292 	u16 i = rx_ring->next_to_clean;
5293 	struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
5294 
5295 	if (rx_ring->xsk_umem) {
5296 		ixgbe_xsk_clean_rx_ring(rx_ring);
5297 		goto skip_free;
5298 	}
5299 
5300 	/* Free all the Rx ring sk_buffs */
5301 	while (i != rx_ring->next_to_alloc) {
5302 		if (rx_buffer->skb) {
5303 			struct sk_buff *skb = rx_buffer->skb;
5304 			if (IXGBE_CB(skb)->page_released)
5305 				dma_unmap_page_attrs(rx_ring->dev,
5306 						     IXGBE_CB(skb)->dma,
5307 						     ixgbe_rx_pg_size(rx_ring),
5308 						     DMA_FROM_DEVICE,
5309 						     IXGBE_RX_DMA_ATTR);
5310 			dev_kfree_skb(skb);
5311 		}
5312 
5313 		/* Invalidate cache lines that may have been written to by
5314 		 * device so that we avoid corrupting memory.
5315 		 */
5316 		dma_sync_single_range_for_cpu(rx_ring->dev,
5317 					      rx_buffer->dma,
5318 					      rx_buffer->page_offset,
5319 					      ixgbe_rx_bufsz(rx_ring),
5320 					      DMA_FROM_DEVICE);
5321 
5322 		/* free resources associated with mapping */
5323 		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
5324 				     ixgbe_rx_pg_size(rx_ring),
5325 				     DMA_FROM_DEVICE,
5326 				     IXGBE_RX_DMA_ATTR);
5327 		__page_frag_cache_drain(rx_buffer->page,
5328 					rx_buffer->pagecnt_bias);
5329 
5330 		i++;
5331 		rx_buffer++;
5332 		if (i == rx_ring->count) {
5333 			i = 0;
5334 			rx_buffer = rx_ring->rx_buffer_info;
5335 		}
5336 	}
5337 
5338 skip_free:
5339 	rx_ring->next_to_alloc = 0;
5340 	rx_ring->next_to_clean = 0;
5341 	rx_ring->next_to_use = 0;
5342 }
5343 
5344 static int ixgbe_fwd_ring_up(struct ixgbe_adapter *adapter,
5345 			     struct ixgbe_fwd_adapter *accel)
5346 {
5347 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
5348 	int num_tc = netdev_get_num_tc(adapter->netdev);
5349 	struct net_device *vdev = accel->netdev;
5350 	int i, baseq, err;
5351 
5352 	baseq = accel->pool * adapter->num_rx_queues_per_pool;
5353 	netdev_dbg(vdev, "pool %i:%i queues %i:%i\n",
5354 		   accel->pool, adapter->num_rx_pools,
5355 		   baseq, baseq + adapter->num_rx_queues_per_pool);
5356 
5357 	accel->rx_base_queue = baseq;
5358 	accel->tx_base_queue = baseq;
5359 
5360 	/* record configuration for macvlan interface in vdev */
5361 	for (i = 0; i < num_tc; i++)
5362 		netdev_bind_sb_channel_queue(adapter->netdev, vdev,
5363 					     i, rss_i, baseq + (rss_i * i));
5364 
5365 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5366 		adapter->rx_ring[baseq + i]->netdev = vdev;
5367 
5368 	/* Guarantee all rings are updated before we update the
5369 	 * MAC address filter.
5370 	 */
5371 	wmb();
5372 
5373 	/* ixgbe_add_mac_filter will return an index if it succeeds, so we
5374 	 * need to only treat it as an error value if it is negative.
5375 	 */
5376 	err = ixgbe_add_mac_filter(adapter, vdev->dev_addr,
5377 				   VMDQ_P(accel->pool));
5378 	if (err >= 0)
5379 		return 0;
5380 
5381 	/* if we cannot add the MAC rule then disable the offload */
5382 	macvlan_release_l2fw_offload(vdev);
5383 
5384 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5385 		adapter->rx_ring[baseq + i]->netdev = NULL;
5386 
5387 	netdev_err(vdev, "L2FW offload disabled due to L2 filter error\n");
5388 
5389 	/* unbind the queues and drop the subordinate channel config */
5390 	netdev_unbind_sb_channel(adapter->netdev, vdev);
5391 	netdev_set_sb_channel(vdev, 0);
5392 
5393 	clear_bit(accel->pool, adapter->fwd_bitmask);
5394 	kfree(accel);
5395 
5396 	return err;
5397 }
5398 
5399 static int ixgbe_macvlan_up(struct net_device *vdev,
5400 			    struct netdev_nested_priv *priv)
5401 {
5402 	struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)priv->data;
5403 	struct ixgbe_fwd_adapter *accel;
5404 
5405 	if (!netif_is_macvlan(vdev))
5406 		return 0;
5407 
5408 	accel = macvlan_accel_priv(vdev);
5409 	if (!accel)
5410 		return 0;
5411 
5412 	ixgbe_fwd_ring_up(adapter, accel);
5413 
5414 	return 0;
5415 }
5416 
5417 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
5418 {
5419 	struct netdev_nested_priv priv = {
5420 		.data = (void *)adapter,
5421 	};
5422 
5423 	netdev_walk_all_upper_dev_rcu(adapter->netdev,
5424 				      ixgbe_macvlan_up, &priv);
5425 }
5426 
5427 static void ixgbe_configure(struct ixgbe_adapter *adapter)
5428 {
5429 	struct ixgbe_hw *hw = &adapter->hw;
5430 
5431 	ixgbe_configure_pb(adapter);
5432 #ifdef CONFIG_IXGBE_DCB
5433 	ixgbe_configure_dcb(adapter);
5434 #endif
5435 	/*
5436 	 * We must restore virtualization before VLANs or else
5437 	 * the VLVF registers will not be populated
5438 	 */
5439 	ixgbe_configure_virtualization(adapter);
5440 
5441 	ixgbe_set_rx_mode(adapter->netdev);
5442 	ixgbe_restore_vlan(adapter);
5443 	ixgbe_ipsec_restore(adapter);
5444 
5445 	switch (hw->mac.type) {
5446 	case ixgbe_mac_82599EB:
5447 	case ixgbe_mac_X540:
5448 		hw->mac.ops.disable_rx_buff(hw);
5449 		break;
5450 	default:
5451 		break;
5452 	}
5453 
5454 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5455 		ixgbe_init_fdir_signature_82599(&adapter->hw,
5456 						adapter->fdir_pballoc);
5457 	} else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
5458 		ixgbe_init_fdir_perfect_82599(&adapter->hw,
5459 					      adapter->fdir_pballoc);
5460 		ixgbe_fdir_filter_restore(adapter);
5461 	}
5462 
5463 	switch (hw->mac.type) {
5464 	case ixgbe_mac_82599EB:
5465 	case ixgbe_mac_X540:
5466 		hw->mac.ops.enable_rx_buff(hw);
5467 		break;
5468 	default:
5469 		break;
5470 	}
5471 
5472 #ifdef CONFIG_IXGBE_DCA
5473 	/* configure DCA */
5474 	if (adapter->flags & IXGBE_FLAG_DCA_CAPABLE)
5475 		ixgbe_setup_dca(adapter);
5476 #endif /* CONFIG_IXGBE_DCA */
5477 
5478 #ifdef IXGBE_FCOE
5479 	/* configure FCoE L2 filters, redirection table, and Rx control */
5480 	ixgbe_configure_fcoe(adapter);
5481 
5482 #endif /* IXGBE_FCOE */
5483 	ixgbe_configure_tx(adapter);
5484 	ixgbe_configure_rx(adapter);
5485 	ixgbe_configure_dfwd(adapter);
5486 }
5487 
5488 /**
5489  * ixgbe_sfp_link_config - set up SFP+ link
5490  * @adapter: pointer to private adapter struct
5491  **/
5492 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
5493 {
5494 	/*
5495 	 * We are assuming the worst case scenario here, and that
5496 	 * is that an SFP was inserted/removed after the reset
5497 	 * but before SFP detection was enabled.  As such the best
5498 	 * solution is to just start searching as soon as we start
5499 	 */
5500 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5501 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5502 
5503 	adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5504 	adapter->sfp_poll_time = 0;
5505 }
5506 
5507 /**
5508  * ixgbe_non_sfp_link_config - set up non-SFP+ link
5509  * @hw: pointer to private hardware struct
5510  *
5511  * Returns 0 on success, negative on failure
5512  **/
5513 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
5514 {
5515 	u32 speed;
5516 	bool autoneg, link_up = false;
5517 	int ret = IXGBE_ERR_LINK_SETUP;
5518 
5519 	if (hw->mac.ops.check_link)
5520 		ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
5521 
5522 	if (ret)
5523 		return ret;
5524 
5525 	speed = hw->phy.autoneg_advertised;
5526 	if (!speed && hw->mac.ops.get_link_capabilities) {
5527 		ret = hw->mac.ops.get_link_capabilities(hw, &speed,
5528 							&autoneg);
5529 		speed &= ~(IXGBE_LINK_SPEED_5GB_FULL |
5530 			   IXGBE_LINK_SPEED_2_5GB_FULL);
5531 	}
5532 
5533 	if (ret)
5534 		return ret;
5535 
5536 	if (hw->mac.ops.setup_link)
5537 		ret = hw->mac.ops.setup_link(hw, speed, link_up);
5538 
5539 	return ret;
5540 }
5541 
5542 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
5543 {
5544 	struct ixgbe_hw *hw = &adapter->hw;
5545 	u32 gpie = 0;
5546 
5547 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5548 		gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
5549 		       IXGBE_GPIE_OCD;
5550 		gpie |= IXGBE_GPIE_EIAME;
5551 		/*
5552 		 * use EIAM to auto-mask when MSI-X interrupt is asserted
5553 		 * this saves a register write for every interrupt
5554 		 */
5555 		switch (hw->mac.type) {
5556 		case ixgbe_mac_82598EB:
5557 			IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5558 			break;
5559 		case ixgbe_mac_82599EB:
5560 		case ixgbe_mac_X540:
5561 		case ixgbe_mac_X550:
5562 		case ixgbe_mac_X550EM_x:
5563 		case ixgbe_mac_x550em_a:
5564 		default:
5565 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
5566 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
5567 			break;
5568 		}
5569 	} else {
5570 		/* legacy interrupts, use EIAM to auto-mask when reading EICR,
5571 		 * specifically only auto mask tx and rx interrupts */
5572 		IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5573 	}
5574 
5575 	/* XXX: to interrupt immediately for EICS writes, enable this */
5576 	/* gpie |= IXGBE_GPIE_EIMEN; */
5577 
5578 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
5579 		gpie &= ~IXGBE_GPIE_VTMODE_MASK;
5580 
5581 		switch (adapter->ring_feature[RING_F_VMDQ].mask) {
5582 		case IXGBE_82599_VMDQ_8Q_MASK:
5583 			gpie |= IXGBE_GPIE_VTMODE_16;
5584 			break;
5585 		case IXGBE_82599_VMDQ_4Q_MASK:
5586 			gpie |= IXGBE_GPIE_VTMODE_32;
5587 			break;
5588 		default:
5589 			gpie |= IXGBE_GPIE_VTMODE_64;
5590 			break;
5591 		}
5592 	}
5593 
5594 	/* Enable Thermal over heat sensor interrupt */
5595 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
5596 		switch (adapter->hw.mac.type) {
5597 		case ixgbe_mac_82599EB:
5598 			gpie |= IXGBE_SDP0_GPIEN_8259X;
5599 			break;
5600 		default:
5601 			break;
5602 		}
5603 	}
5604 
5605 	/* Enable fan failure interrupt */
5606 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
5607 		gpie |= IXGBE_SDP1_GPIEN(hw);
5608 
5609 	switch (hw->mac.type) {
5610 	case ixgbe_mac_82599EB:
5611 		gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X;
5612 		break;
5613 	case ixgbe_mac_X550EM_x:
5614 	case ixgbe_mac_x550em_a:
5615 		gpie |= IXGBE_SDP0_GPIEN_X540;
5616 		break;
5617 	default:
5618 		break;
5619 	}
5620 
5621 	IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
5622 }
5623 
5624 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
5625 {
5626 	struct ixgbe_hw *hw = &adapter->hw;
5627 	int err;
5628 	u32 ctrl_ext;
5629 
5630 	ixgbe_get_hw_control(adapter);
5631 	ixgbe_setup_gpie(adapter);
5632 
5633 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5634 		ixgbe_configure_msix(adapter);
5635 	else
5636 		ixgbe_configure_msi_and_legacy(adapter);
5637 
5638 	/* enable the optics for 82599 SFP+ fiber */
5639 	if (hw->mac.ops.enable_tx_laser)
5640 		hw->mac.ops.enable_tx_laser(hw);
5641 
5642 	if (hw->phy.ops.set_phy_power)
5643 		hw->phy.ops.set_phy_power(hw, true);
5644 
5645 	smp_mb__before_atomic();
5646 	clear_bit(__IXGBE_DOWN, &adapter->state);
5647 	ixgbe_napi_enable_all(adapter);
5648 
5649 	if (ixgbe_is_sfp(hw)) {
5650 		ixgbe_sfp_link_config(adapter);
5651 	} else {
5652 		err = ixgbe_non_sfp_link_config(hw);
5653 		if (err)
5654 			e_err(probe, "link_config FAILED %d\n", err);
5655 	}
5656 
5657 	/* clear any pending interrupts, may auto mask */
5658 	IXGBE_READ_REG(hw, IXGBE_EICR);
5659 	ixgbe_irq_enable(adapter, true, true);
5660 
5661 	/*
5662 	 * If this adapter has a fan, check to see if we had a failure
5663 	 * before we enabled the interrupt.
5664 	 */
5665 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5666 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5667 		if (esdp & IXGBE_ESDP_SDP1)
5668 			e_crit(drv, "Fan has stopped, replace the adapter\n");
5669 	}
5670 
5671 	/* bring the link up in the watchdog, this could race with our first
5672 	 * link up interrupt but shouldn't be a problem */
5673 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5674 	adapter->link_check_timeout = jiffies;
5675 	mod_timer(&adapter->service_timer, jiffies);
5676 
5677 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
5678 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5679 	ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
5680 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5681 }
5682 
5683 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
5684 {
5685 	WARN_ON(in_interrupt());
5686 	/* put off any impending NetWatchDogTimeout */
5687 	netif_trans_update(adapter->netdev);
5688 
5689 	while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
5690 		usleep_range(1000, 2000);
5691 	if (adapter->hw.phy.type == ixgbe_phy_fw)
5692 		ixgbe_watchdog_link_is_down(adapter);
5693 	ixgbe_down(adapter);
5694 	/*
5695 	 * If SR-IOV enabled then wait a bit before bringing the adapter
5696 	 * back up to give the VFs time to respond to the reset.  The
5697 	 * two second wait is based upon the watchdog timer cycle in
5698 	 * the VF driver.
5699 	 */
5700 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5701 		msleep(2000);
5702 	ixgbe_up(adapter);
5703 	clear_bit(__IXGBE_RESETTING, &adapter->state);
5704 }
5705 
5706 void ixgbe_up(struct ixgbe_adapter *adapter)
5707 {
5708 	/* hardware has been reset, we need to reload some things */
5709 	ixgbe_configure(adapter);
5710 
5711 	ixgbe_up_complete(adapter);
5712 }
5713 
5714 static unsigned long ixgbe_get_completion_timeout(struct ixgbe_adapter *adapter)
5715 {
5716 	u16 devctl2;
5717 
5718 	pcie_capability_read_word(adapter->pdev, PCI_EXP_DEVCTL2, &devctl2);
5719 
5720 	switch (devctl2 & IXGBE_PCIDEVCTRL2_TIMEO_MASK) {
5721 	case IXGBE_PCIDEVCTRL2_17_34s:
5722 	case IXGBE_PCIDEVCTRL2_4_8s:
5723 		/* For now we cap the upper limit on delay to 2 seconds
5724 		 * as we end up going up to 34 seconds of delay in worst
5725 		 * case timeout value.
5726 		 */
5727 	case IXGBE_PCIDEVCTRL2_1_2s:
5728 		return 2000000ul;	/* 2.0 s */
5729 	case IXGBE_PCIDEVCTRL2_260_520ms:
5730 		return 520000ul;	/* 520 ms */
5731 	case IXGBE_PCIDEVCTRL2_65_130ms:
5732 		return 130000ul;	/* 130 ms */
5733 	case IXGBE_PCIDEVCTRL2_16_32ms:
5734 		return 32000ul;		/* 32 ms */
5735 	case IXGBE_PCIDEVCTRL2_1_2ms:
5736 		return 2000ul;		/* 2 ms */
5737 	case IXGBE_PCIDEVCTRL2_50_100us:
5738 		return 100ul;		/* 100 us */
5739 	case IXGBE_PCIDEVCTRL2_16_32ms_def:
5740 		return 32000ul;		/* 32 ms */
5741 	default:
5742 		break;
5743 	}
5744 
5745 	/* We shouldn't need to hit this path, but just in case default as
5746 	 * though completion timeout is not supported and support 32ms.
5747 	 */
5748 	return 32000ul;
5749 }
5750 
5751 void ixgbe_disable_rx(struct ixgbe_adapter *adapter)
5752 {
5753 	unsigned long wait_delay, delay_interval;
5754 	struct ixgbe_hw *hw = &adapter->hw;
5755 	int i, wait_loop;
5756 	u32 rxdctl;
5757 
5758 	/* disable receives */
5759 	hw->mac.ops.disable_rx(hw);
5760 
5761 	if (ixgbe_removed(hw->hw_addr))
5762 		return;
5763 
5764 	/* disable all enabled Rx queues */
5765 	for (i = 0; i < adapter->num_rx_queues; i++) {
5766 		struct ixgbe_ring *ring = adapter->rx_ring[i];
5767 		u8 reg_idx = ring->reg_idx;
5768 
5769 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5770 		rxdctl &= ~IXGBE_RXDCTL_ENABLE;
5771 		rxdctl |= IXGBE_RXDCTL_SWFLSH;
5772 
5773 		/* write value back with RXDCTL.ENABLE bit cleared */
5774 		IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
5775 	}
5776 
5777 	/* RXDCTL.EN may not change on 82598 if link is down, so skip it */
5778 	if (hw->mac.type == ixgbe_mac_82598EB &&
5779 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5780 		return;
5781 
5782 	/* Determine our minimum delay interval. We will increase this value
5783 	 * with each subsequent test. This way if the device returns quickly
5784 	 * we should spend as little time as possible waiting, however as
5785 	 * the time increases we will wait for larger periods of time.
5786 	 *
5787 	 * The trick here is that we increase the interval using the
5788 	 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5789 	 * of that wait is that it totals up to 100x whatever interval we
5790 	 * choose. Since our minimum wait is 100us we can just divide the
5791 	 * total timeout by 100 to get our minimum delay interval.
5792 	 */
5793 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5794 
5795 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
5796 	wait_delay = delay_interval;
5797 
5798 	while (wait_loop--) {
5799 		usleep_range(wait_delay, wait_delay + 10);
5800 		wait_delay += delay_interval * 2;
5801 		rxdctl = 0;
5802 
5803 		/* OR together the reading of all the active RXDCTL registers,
5804 		 * and then test the result. We need the disable to complete
5805 		 * before we start freeing the memory and invalidating the
5806 		 * DMA mappings.
5807 		 */
5808 		for (i = 0; i < adapter->num_rx_queues; i++) {
5809 			struct ixgbe_ring *ring = adapter->rx_ring[i];
5810 			u8 reg_idx = ring->reg_idx;
5811 
5812 			rxdctl |= IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5813 		}
5814 
5815 		if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
5816 			return;
5817 	}
5818 
5819 	e_err(drv,
5820 	      "RXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5821 }
5822 
5823 void ixgbe_disable_tx(struct ixgbe_adapter *adapter)
5824 {
5825 	unsigned long wait_delay, delay_interval;
5826 	struct ixgbe_hw *hw = &adapter->hw;
5827 	int i, wait_loop;
5828 	u32 txdctl;
5829 
5830 	if (ixgbe_removed(hw->hw_addr))
5831 		return;
5832 
5833 	/* disable all enabled Tx queues */
5834 	for (i = 0; i < adapter->num_tx_queues; i++) {
5835 		struct ixgbe_ring *ring = adapter->tx_ring[i];
5836 		u8 reg_idx = ring->reg_idx;
5837 
5838 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5839 	}
5840 
5841 	/* disable all enabled XDP Tx queues */
5842 	for (i = 0; i < adapter->num_xdp_queues; i++) {
5843 		struct ixgbe_ring *ring = adapter->xdp_ring[i];
5844 		u8 reg_idx = ring->reg_idx;
5845 
5846 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5847 	}
5848 
5849 	/* If the link is not up there shouldn't be much in the way of
5850 	 * pending transactions. Those that are left will be flushed out
5851 	 * when the reset logic goes through the flush sequence to clean out
5852 	 * the pending Tx transactions.
5853 	 */
5854 	if (!(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5855 		goto dma_engine_disable;
5856 
5857 	/* Determine our minimum delay interval. We will increase this value
5858 	 * with each subsequent test. This way if the device returns quickly
5859 	 * we should spend as little time as possible waiting, however as
5860 	 * the time increases we will wait for larger periods of time.
5861 	 *
5862 	 * The trick here is that we increase the interval using the
5863 	 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5864 	 * of that wait is that it totals up to 100x whatever interval we
5865 	 * choose. Since our minimum wait is 100us we can just divide the
5866 	 * total timeout by 100 to get our minimum delay interval.
5867 	 */
5868 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5869 
5870 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
5871 	wait_delay = delay_interval;
5872 
5873 	while (wait_loop--) {
5874 		usleep_range(wait_delay, wait_delay + 10);
5875 		wait_delay += delay_interval * 2;
5876 		txdctl = 0;
5877 
5878 		/* OR together the reading of all the active TXDCTL registers,
5879 		 * and then test the result. We need the disable to complete
5880 		 * before we start freeing the memory and invalidating the
5881 		 * DMA mappings.
5882 		 */
5883 		for (i = 0; i < adapter->num_tx_queues; i++) {
5884 			struct ixgbe_ring *ring = adapter->tx_ring[i];
5885 			u8 reg_idx = ring->reg_idx;
5886 
5887 			txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5888 		}
5889 		for (i = 0; i < adapter->num_xdp_queues; i++) {
5890 			struct ixgbe_ring *ring = adapter->xdp_ring[i];
5891 			u8 reg_idx = ring->reg_idx;
5892 
5893 			txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5894 		}
5895 
5896 		if (!(txdctl & IXGBE_TXDCTL_ENABLE))
5897 			goto dma_engine_disable;
5898 	}
5899 
5900 	e_err(drv,
5901 	      "TXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5902 
5903 dma_engine_disable:
5904 	/* Disable the Tx DMA engine on 82599 and later MAC */
5905 	switch (hw->mac.type) {
5906 	case ixgbe_mac_82599EB:
5907 	case ixgbe_mac_X540:
5908 	case ixgbe_mac_X550:
5909 	case ixgbe_mac_X550EM_x:
5910 	case ixgbe_mac_x550em_a:
5911 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5912 				(IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5913 				 ~IXGBE_DMATXCTL_TE));
5914 		fallthrough;
5915 	default:
5916 		break;
5917 	}
5918 }
5919 
5920 void ixgbe_reset(struct ixgbe_adapter *adapter)
5921 {
5922 	struct ixgbe_hw *hw = &adapter->hw;
5923 	struct net_device *netdev = adapter->netdev;
5924 	int err;
5925 
5926 	if (ixgbe_removed(hw->hw_addr))
5927 		return;
5928 	/* lock SFP init bit to prevent race conditions with the watchdog */
5929 	while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5930 		usleep_range(1000, 2000);
5931 
5932 	/* clear all SFP and link config related flags while holding SFP_INIT */
5933 	adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
5934 			     IXGBE_FLAG2_SFP_NEEDS_RESET);
5935 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5936 
5937 	err = hw->mac.ops.init_hw(hw);
5938 	switch (err) {
5939 	case 0:
5940 	case IXGBE_ERR_SFP_NOT_PRESENT:
5941 	case IXGBE_ERR_SFP_NOT_SUPPORTED:
5942 		break;
5943 	case IXGBE_ERR_MASTER_REQUESTS_PENDING:
5944 		e_dev_err("master disable timed out\n");
5945 		break;
5946 	case IXGBE_ERR_EEPROM_VERSION:
5947 		/* We are running on a pre-production device, log a warning */
5948 		e_dev_warn("This device is a pre-production adapter/LOM. "
5949 			   "Please be aware there may be issues associated with "
5950 			   "your hardware.  If you are experiencing problems "
5951 			   "please contact your Intel or hardware "
5952 			   "representative who provided you with this "
5953 			   "hardware.\n");
5954 		break;
5955 	default:
5956 		e_dev_err("Hardware Error: %d\n", err);
5957 	}
5958 
5959 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5960 
5961 	/* flush entries out of MAC table */
5962 	ixgbe_flush_sw_mac_table(adapter);
5963 	__dev_uc_unsync(netdev, NULL);
5964 
5965 	/* do not flush user set addresses */
5966 	ixgbe_mac_set_default_filter(adapter);
5967 
5968 	/* update SAN MAC vmdq pool selection */
5969 	if (hw->mac.san_mac_rar_index)
5970 		hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
5971 
5972 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5973 		ixgbe_ptp_reset(adapter);
5974 
5975 	if (hw->phy.ops.set_phy_power) {
5976 		if (!netif_running(adapter->netdev) && !adapter->wol)
5977 			hw->phy.ops.set_phy_power(hw, false);
5978 		else
5979 			hw->phy.ops.set_phy_power(hw, true);
5980 	}
5981 }
5982 
5983 /**
5984  * ixgbe_clean_tx_ring - Free Tx Buffers
5985  * @tx_ring: ring to be cleaned
5986  **/
5987 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
5988 {
5989 	u16 i = tx_ring->next_to_clean;
5990 	struct ixgbe_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
5991 
5992 	if (tx_ring->xsk_umem) {
5993 		ixgbe_xsk_clean_tx_ring(tx_ring);
5994 		goto out;
5995 	}
5996 
5997 	while (i != tx_ring->next_to_use) {
5998 		union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
5999 
6000 		/* Free all the Tx ring sk_buffs */
6001 		if (ring_is_xdp(tx_ring))
6002 			xdp_return_frame(tx_buffer->xdpf);
6003 		else
6004 			dev_kfree_skb_any(tx_buffer->skb);
6005 
6006 		/* unmap skb header data */
6007 		dma_unmap_single(tx_ring->dev,
6008 				 dma_unmap_addr(tx_buffer, dma),
6009 				 dma_unmap_len(tx_buffer, len),
6010 				 DMA_TO_DEVICE);
6011 
6012 		/* check for eop_desc to determine the end of the packet */
6013 		eop_desc = tx_buffer->next_to_watch;
6014 		tx_desc = IXGBE_TX_DESC(tx_ring, i);
6015 
6016 		/* unmap remaining buffers */
6017 		while (tx_desc != eop_desc) {
6018 			tx_buffer++;
6019 			tx_desc++;
6020 			i++;
6021 			if (unlikely(i == tx_ring->count)) {
6022 				i = 0;
6023 				tx_buffer = tx_ring->tx_buffer_info;
6024 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6025 			}
6026 
6027 			/* unmap any remaining paged data */
6028 			if (dma_unmap_len(tx_buffer, len))
6029 				dma_unmap_page(tx_ring->dev,
6030 					       dma_unmap_addr(tx_buffer, dma),
6031 					       dma_unmap_len(tx_buffer, len),
6032 					       DMA_TO_DEVICE);
6033 		}
6034 
6035 		/* move us one more past the eop_desc for start of next pkt */
6036 		tx_buffer++;
6037 		i++;
6038 		if (unlikely(i == tx_ring->count)) {
6039 			i = 0;
6040 			tx_buffer = tx_ring->tx_buffer_info;
6041 		}
6042 	}
6043 
6044 	/* reset BQL for queue */
6045 	if (!ring_is_xdp(tx_ring))
6046 		netdev_tx_reset_queue(txring_txq(tx_ring));
6047 
6048 out:
6049 	/* reset next_to_use and next_to_clean */
6050 	tx_ring->next_to_use = 0;
6051 	tx_ring->next_to_clean = 0;
6052 }
6053 
6054 /**
6055  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
6056  * @adapter: board private structure
6057  **/
6058 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
6059 {
6060 	int i;
6061 
6062 	for (i = 0; i < adapter->num_rx_queues; i++)
6063 		ixgbe_clean_rx_ring(adapter->rx_ring[i]);
6064 }
6065 
6066 /**
6067  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
6068  * @adapter: board private structure
6069  **/
6070 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
6071 {
6072 	int i;
6073 
6074 	for (i = 0; i < adapter->num_tx_queues; i++)
6075 		ixgbe_clean_tx_ring(adapter->tx_ring[i]);
6076 	for (i = 0; i < adapter->num_xdp_queues; i++)
6077 		ixgbe_clean_tx_ring(adapter->xdp_ring[i]);
6078 }
6079 
6080 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
6081 {
6082 	struct hlist_node *node2;
6083 	struct ixgbe_fdir_filter *filter;
6084 
6085 	spin_lock(&adapter->fdir_perfect_lock);
6086 
6087 	hlist_for_each_entry_safe(filter, node2,
6088 				  &adapter->fdir_filter_list, fdir_node) {
6089 		hlist_del(&filter->fdir_node);
6090 		kfree(filter);
6091 	}
6092 	adapter->fdir_filter_count = 0;
6093 
6094 	spin_unlock(&adapter->fdir_perfect_lock);
6095 }
6096 
6097 void ixgbe_down(struct ixgbe_adapter *adapter)
6098 {
6099 	struct net_device *netdev = adapter->netdev;
6100 	struct ixgbe_hw *hw = &adapter->hw;
6101 	int i;
6102 
6103 	/* signal that we are down to the interrupt handler */
6104 	if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
6105 		return; /* do nothing if already down */
6106 
6107 	/* Shut off incoming Tx traffic */
6108 	netif_tx_stop_all_queues(netdev);
6109 
6110 	/* call carrier off first to avoid false dev_watchdog timeouts */
6111 	netif_carrier_off(netdev);
6112 	netif_tx_disable(netdev);
6113 
6114 	/* Disable Rx */
6115 	ixgbe_disable_rx(adapter);
6116 
6117 	/* synchronize_rcu() needed for pending XDP buffers to drain */
6118 	if (adapter->xdp_ring[0])
6119 		synchronize_rcu();
6120 
6121 	ixgbe_irq_disable(adapter);
6122 
6123 	ixgbe_napi_disable_all(adapter);
6124 
6125 	clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
6126 	adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6127 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6128 
6129 	del_timer_sync(&adapter->service_timer);
6130 
6131 	if (adapter->num_vfs) {
6132 		/* Clear EITR Select mapping */
6133 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
6134 
6135 		/* Mark all the VFs as inactive */
6136 		for (i = 0 ; i < adapter->num_vfs; i++)
6137 			adapter->vfinfo[i].clear_to_send = false;
6138 
6139 		/* ping all the active vfs to let them know we are going down */
6140 		ixgbe_ping_all_vfs(adapter);
6141 
6142 		/* Disable all VFTE/VFRE TX/RX */
6143 		ixgbe_disable_tx_rx(adapter);
6144 	}
6145 
6146 	/* disable transmits in the hardware now that interrupts are off */
6147 	ixgbe_disable_tx(adapter);
6148 
6149 	if (!pci_channel_offline(adapter->pdev))
6150 		ixgbe_reset(adapter);
6151 
6152 	/* power down the optics for 82599 SFP+ fiber */
6153 	if (hw->mac.ops.disable_tx_laser)
6154 		hw->mac.ops.disable_tx_laser(hw);
6155 
6156 	ixgbe_clean_all_tx_rings(adapter);
6157 	ixgbe_clean_all_rx_rings(adapter);
6158 }
6159 
6160 /**
6161  * ixgbe_eee_capable - helper function to determine EEE support on X550
6162  * @adapter: board private structure
6163  */
6164 static void ixgbe_set_eee_capable(struct ixgbe_adapter *adapter)
6165 {
6166 	struct ixgbe_hw *hw = &adapter->hw;
6167 
6168 	switch (hw->device_id) {
6169 	case IXGBE_DEV_ID_X550EM_A_1G_T:
6170 	case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6171 		if (!hw->phy.eee_speeds_supported)
6172 			break;
6173 		adapter->flags2 |= IXGBE_FLAG2_EEE_CAPABLE;
6174 		if (!hw->phy.eee_speeds_advertised)
6175 			break;
6176 		adapter->flags2 |= IXGBE_FLAG2_EEE_ENABLED;
6177 		break;
6178 	default:
6179 		adapter->flags2 &= ~IXGBE_FLAG2_EEE_CAPABLE;
6180 		adapter->flags2 &= ~IXGBE_FLAG2_EEE_ENABLED;
6181 		break;
6182 	}
6183 }
6184 
6185 /**
6186  * ixgbe_tx_timeout - Respond to a Tx Hang
6187  * @netdev: network interface device structure
6188  **/
6189 static void ixgbe_tx_timeout(struct net_device *netdev, unsigned int txqueue)
6190 {
6191 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6192 
6193 	/* Do the reset outside of interrupt context */
6194 	ixgbe_tx_timeout_reset(adapter);
6195 }
6196 
6197 #ifdef CONFIG_IXGBE_DCB
6198 static void ixgbe_init_dcb(struct ixgbe_adapter *adapter)
6199 {
6200 	struct ixgbe_hw *hw = &adapter->hw;
6201 	struct tc_configuration *tc;
6202 	int j;
6203 
6204 	switch (hw->mac.type) {
6205 	case ixgbe_mac_82598EB:
6206 	case ixgbe_mac_82599EB:
6207 		adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
6208 		adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
6209 		break;
6210 	case ixgbe_mac_X540:
6211 	case ixgbe_mac_X550:
6212 		adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
6213 		adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
6214 		break;
6215 	case ixgbe_mac_X550EM_x:
6216 	case ixgbe_mac_x550em_a:
6217 	default:
6218 		adapter->dcb_cfg.num_tcs.pg_tcs = DEF_TRAFFIC_CLASS;
6219 		adapter->dcb_cfg.num_tcs.pfc_tcs = DEF_TRAFFIC_CLASS;
6220 		break;
6221 	}
6222 
6223 	/* Configure DCB traffic classes */
6224 	for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
6225 		tc = &adapter->dcb_cfg.tc_config[j];
6226 		tc->path[DCB_TX_CONFIG].bwg_id = 0;
6227 		tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
6228 		tc->path[DCB_RX_CONFIG].bwg_id = 0;
6229 		tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
6230 		tc->dcb_pfc = pfc_disabled;
6231 	}
6232 
6233 	/* Initialize default user to priority mapping, UPx->TC0 */
6234 	tc = &adapter->dcb_cfg.tc_config[0];
6235 	tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
6236 	tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
6237 
6238 	adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
6239 	adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
6240 	adapter->dcb_cfg.pfc_mode_enable = false;
6241 	adapter->dcb_set_bitmap = 0x00;
6242 	if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
6243 		adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
6244 	memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
6245 	       sizeof(adapter->temp_dcb_cfg));
6246 }
6247 #endif
6248 
6249 /**
6250  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
6251  * @adapter: board private structure to initialize
6252  * @ii: pointer to ixgbe_info for device
6253  *
6254  * ixgbe_sw_init initializes the Adapter private data structure.
6255  * Fields are initialized based on PCI device information and
6256  * OS network device settings (MTU size).
6257  **/
6258 static int ixgbe_sw_init(struct ixgbe_adapter *adapter,
6259 			 const struct ixgbe_info *ii)
6260 {
6261 	struct ixgbe_hw *hw = &adapter->hw;
6262 	struct pci_dev *pdev = adapter->pdev;
6263 	unsigned int rss, fdir;
6264 	u32 fwsm;
6265 	int i;
6266 
6267 	/* PCI config space info */
6268 
6269 	hw->vendor_id = pdev->vendor;
6270 	hw->device_id = pdev->device;
6271 	hw->revision_id = pdev->revision;
6272 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
6273 	hw->subsystem_device_id = pdev->subsystem_device;
6274 
6275 	/* get_invariants needs the device IDs */
6276 	ii->get_invariants(hw);
6277 
6278 	/* Set common capability flags and settings */
6279 	rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
6280 	adapter->ring_feature[RING_F_RSS].limit = rss;
6281 	adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
6282 	adapter->max_q_vectors = MAX_Q_VECTORS_82599;
6283 	adapter->atr_sample_rate = 20;
6284 	fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
6285 	adapter->ring_feature[RING_F_FDIR].limit = fdir;
6286 	adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
6287 	adapter->ring_feature[RING_F_VMDQ].limit = 1;
6288 #ifdef CONFIG_IXGBE_DCA
6289 	adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
6290 #endif
6291 #ifdef CONFIG_IXGBE_DCB
6292 	adapter->flags |= IXGBE_FLAG_DCB_CAPABLE;
6293 	adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6294 #endif
6295 #ifdef IXGBE_FCOE
6296 	adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
6297 	adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6298 #ifdef CONFIG_IXGBE_DCB
6299 	/* Default traffic class to use for FCoE */
6300 	adapter->fcoe.up = IXGBE_FCOE_DEFTC;
6301 #endif /* CONFIG_IXGBE_DCB */
6302 #endif /* IXGBE_FCOE */
6303 
6304 	/* initialize static ixgbe jump table entries */
6305 	adapter->jump_tables[0] = kzalloc(sizeof(*adapter->jump_tables[0]),
6306 					  GFP_KERNEL);
6307 	if (!adapter->jump_tables[0])
6308 		return -ENOMEM;
6309 	adapter->jump_tables[0]->mat = ixgbe_ipv4_fields;
6310 
6311 	for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++)
6312 		adapter->jump_tables[i] = NULL;
6313 
6314 	adapter->mac_table = kcalloc(hw->mac.num_rar_entries,
6315 				     sizeof(struct ixgbe_mac_addr),
6316 				     GFP_KERNEL);
6317 	if (!adapter->mac_table)
6318 		return -ENOMEM;
6319 
6320 	if (ixgbe_init_rss_key(adapter))
6321 		return -ENOMEM;
6322 
6323 	adapter->af_xdp_zc_qps = bitmap_zalloc(MAX_XDP_QUEUES, GFP_KERNEL);
6324 	if (!adapter->af_xdp_zc_qps)
6325 		return -ENOMEM;
6326 
6327 	/* Set MAC specific capability flags and exceptions */
6328 	switch (hw->mac.type) {
6329 	case ixgbe_mac_82598EB:
6330 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
6331 
6332 		if (hw->device_id == IXGBE_DEV_ID_82598AT)
6333 			adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
6334 
6335 		adapter->max_q_vectors = MAX_Q_VECTORS_82598;
6336 		adapter->ring_feature[RING_F_FDIR].limit = 0;
6337 		adapter->atr_sample_rate = 0;
6338 		adapter->fdir_pballoc = 0;
6339 #ifdef IXGBE_FCOE
6340 		adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6341 		adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6342 #ifdef CONFIG_IXGBE_DCB
6343 		adapter->fcoe.up = 0;
6344 #endif /* IXGBE_DCB */
6345 #endif /* IXGBE_FCOE */
6346 		break;
6347 	case ixgbe_mac_82599EB:
6348 		if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
6349 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6350 		break;
6351 	case ixgbe_mac_X540:
6352 		fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
6353 		if (fwsm & IXGBE_FWSM_TS_ENABLED)
6354 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6355 		break;
6356 	case ixgbe_mac_x550em_a:
6357 		switch (hw->device_id) {
6358 		case IXGBE_DEV_ID_X550EM_A_1G_T:
6359 		case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6360 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6361 			break;
6362 		default:
6363 			break;
6364 		}
6365 		fallthrough;
6366 	case ixgbe_mac_X550EM_x:
6367 #ifdef CONFIG_IXGBE_DCB
6368 		adapter->flags &= ~IXGBE_FLAG_DCB_CAPABLE;
6369 #endif
6370 #ifdef IXGBE_FCOE
6371 		adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6372 #ifdef CONFIG_IXGBE_DCB
6373 		adapter->fcoe.up = 0;
6374 #endif /* IXGBE_DCB */
6375 #endif /* IXGBE_FCOE */
6376 		fallthrough;
6377 	case ixgbe_mac_X550:
6378 		if (hw->mac.type == ixgbe_mac_X550)
6379 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6380 #ifdef CONFIG_IXGBE_DCA
6381 		adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
6382 #endif
6383 		break;
6384 	default:
6385 		break;
6386 	}
6387 
6388 #ifdef IXGBE_FCOE
6389 	/* FCoE support exists, always init the FCoE lock */
6390 	spin_lock_init(&adapter->fcoe.lock);
6391 
6392 #endif
6393 	/* n-tuple support exists, always init our spinlock */
6394 	spin_lock_init(&adapter->fdir_perfect_lock);
6395 
6396 #ifdef CONFIG_IXGBE_DCB
6397 	ixgbe_init_dcb(adapter);
6398 #endif
6399 	ixgbe_init_ipsec_offload(adapter);
6400 
6401 	/* default flow control settings */
6402 	hw->fc.requested_mode = ixgbe_fc_full;
6403 	hw->fc.current_mode = ixgbe_fc_full;	/* init for ethtool output */
6404 	ixgbe_pbthresh_setup(adapter);
6405 	hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
6406 	hw->fc.send_xon = true;
6407 	hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
6408 
6409 #ifdef CONFIG_PCI_IOV
6410 	if (max_vfs > 0)
6411 		e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
6412 
6413 	/* assign number of SR-IOV VFs */
6414 	if (hw->mac.type != ixgbe_mac_82598EB) {
6415 		if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
6416 			max_vfs = 0;
6417 			e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
6418 		}
6419 	}
6420 #endif /* CONFIG_PCI_IOV */
6421 
6422 	/* enable itr by default in dynamic mode */
6423 	adapter->rx_itr_setting = 1;
6424 	adapter->tx_itr_setting = 1;
6425 
6426 	/* set default ring sizes */
6427 	adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
6428 	adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
6429 
6430 	/* set default work limits */
6431 	adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
6432 
6433 	/* initialize eeprom parameters */
6434 	if (ixgbe_init_eeprom_params_generic(hw)) {
6435 		e_dev_err("EEPROM initialization failed\n");
6436 		return -EIO;
6437 	}
6438 
6439 	/* PF holds first pool slot */
6440 	set_bit(0, adapter->fwd_bitmask);
6441 	set_bit(__IXGBE_DOWN, &adapter->state);
6442 
6443 	return 0;
6444 }
6445 
6446 /**
6447  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
6448  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
6449  *
6450  * Return 0 on success, negative on failure
6451  **/
6452 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
6453 {
6454 	struct device *dev = tx_ring->dev;
6455 	int orig_node = dev_to_node(dev);
6456 	int ring_node = NUMA_NO_NODE;
6457 	int size;
6458 
6459 	size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
6460 
6461 	if (tx_ring->q_vector)
6462 		ring_node = tx_ring->q_vector->numa_node;
6463 
6464 	tx_ring->tx_buffer_info = vmalloc_node(size, ring_node);
6465 	if (!tx_ring->tx_buffer_info)
6466 		tx_ring->tx_buffer_info = vmalloc(size);
6467 	if (!tx_ring->tx_buffer_info)
6468 		goto err;
6469 
6470 	/* round up to nearest 4K */
6471 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
6472 	tx_ring->size = ALIGN(tx_ring->size, 4096);
6473 
6474 	set_dev_node(dev, ring_node);
6475 	tx_ring->desc = dma_alloc_coherent(dev,
6476 					   tx_ring->size,
6477 					   &tx_ring->dma,
6478 					   GFP_KERNEL);
6479 	set_dev_node(dev, orig_node);
6480 	if (!tx_ring->desc)
6481 		tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
6482 						   &tx_ring->dma, GFP_KERNEL);
6483 	if (!tx_ring->desc)
6484 		goto err;
6485 
6486 	tx_ring->next_to_use = 0;
6487 	tx_ring->next_to_clean = 0;
6488 	return 0;
6489 
6490 err:
6491 	vfree(tx_ring->tx_buffer_info);
6492 	tx_ring->tx_buffer_info = NULL;
6493 	dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
6494 	return -ENOMEM;
6495 }
6496 
6497 /**
6498  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
6499  * @adapter: board private structure
6500  *
6501  * If this function returns with an error, then it's possible one or
6502  * more of the rings is populated (while the rest are not).  It is the
6503  * callers duty to clean those orphaned rings.
6504  *
6505  * Return 0 on success, negative on failure
6506  **/
6507 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
6508 {
6509 	int i, j = 0, err = 0;
6510 
6511 	for (i = 0; i < adapter->num_tx_queues; i++) {
6512 		err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
6513 		if (!err)
6514 			continue;
6515 
6516 		e_err(probe, "Allocation for Tx Queue %u failed\n", i);
6517 		goto err_setup_tx;
6518 	}
6519 	for (j = 0; j < adapter->num_xdp_queues; j++) {
6520 		err = ixgbe_setup_tx_resources(adapter->xdp_ring[j]);
6521 		if (!err)
6522 			continue;
6523 
6524 		e_err(probe, "Allocation for Tx Queue %u failed\n", j);
6525 		goto err_setup_tx;
6526 	}
6527 
6528 	return 0;
6529 err_setup_tx:
6530 	/* rewind the index freeing the rings as we go */
6531 	while (j--)
6532 		ixgbe_free_tx_resources(adapter->xdp_ring[j]);
6533 	while (i--)
6534 		ixgbe_free_tx_resources(adapter->tx_ring[i]);
6535 	return err;
6536 }
6537 
6538 /**
6539  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
6540  * @adapter: pointer to ixgbe_adapter
6541  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
6542  *
6543  * Returns 0 on success, negative on failure
6544  **/
6545 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
6546 			     struct ixgbe_ring *rx_ring)
6547 {
6548 	struct device *dev = rx_ring->dev;
6549 	int orig_node = dev_to_node(dev);
6550 	int ring_node = NUMA_NO_NODE;
6551 	int size;
6552 
6553 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
6554 
6555 	if (rx_ring->q_vector)
6556 		ring_node = rx_ring->q_vector->numa_node;
6557 
6558 	rx_ring->rx_buffer_info = vmalloc_node(size, ring_node);
6559 	if (!rx_ring->rx_buffer_info)
6560 		rx_ring->rx_buffer_info = vmalloc(size);
6561 	if (!rx_ring->rx_buffer_info)
6562 		goto err;
6563 
6564 	/* Round up to nearest 4K */
6565 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
6566 	rx_ring->size = ALIGN(rx_ring->size, 4096);
6567 
6568 	set_dev_node(dev, ring_node);
6569 	rx_ring->desc = dma_alloc_coherent(dev,
6570 					   rx_ring->size,
6571 					   &rx_ring->dma,
6572 					   GFP_KERNEL);
6573 	set_dev_node(dev, orig_node);
6574 	if (!rx_ring->desc)
6575 		rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
6576 						   &rx_ring->dma, GFP_KERNEL);
6577 	if (!rx_ring->desc)
6578 		goto err;
6579 
6580 	rx_ring->next_to_clean = 0;
6581 	rx_ring->next_to_use = 0;
6582 
6583 	/* XDP RX-queue info */
6584 	if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
6585 			     rx_ring->queue_index) < 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 inline bool ixgbe_ipv6_csum_is_sctp(struct sk_buff *skb)
8039 {
8040 	unsigned int offset = 0;
8041 
8042 	ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
8043 
8044 	return offset == skb_checksum_start_offset(skb);
8045 }
8046 
8047 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
8048 			  struct ixgbe_tx_buffer *first,
8049 			  struct ixgbe_ipsec_tx_data *itd)
8050 {
8051 	struct sk_buff *skb = first->skb;
8052 	u32 vlan_macip_lens = 0;
8053 	u32 fceof_saidx = 0;
8054 	u32 type_tucmd = 0;
8055 
8056 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
8057 csum_failed:
8058 		if (!(first->tx_flags & (IXGBE_TX_FLAGS_HW_VLAN |
8059 					 IXGBE_TX_FLAGS_CC)))
8060 			return;
8061 		goto no_csum;
8062 	}
8063 
8064 	switch (skb->csum_offset) {
8065 	case offsetof(struct tcphdr, check):
8066 		type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
8067 		fallthrough;
8068 	case offsetof(struct udphdr, check):
8069 		break;
8070 	case offsetof(struct sctphdr, checksum):
8071 		/* validate that this is actually an SCTP request */
8072 		if (((first->protocol == htons(ETH_P_IP)) &&
8073 		     (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
8074 		    ((first->protocol == htons(ETH_P_IPV6)) &&
8075 		     ixgbe_ipv6_csum_is_sctp(skb))) {
8076 			type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
8077 			break;
8078 		}
8079 		fallthrough;
8080 	default:
8081 		skb_checksum_help(skb);
8082 		goto csum_failed;
8083 	}
8084 
8085 	/* update TX checksum flag */
8086 	first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
8087 	vlan_macip_lens = skb_checksum_start_offset(skb) -
8088 			  skb_network_offset(skb);
8089 no_csum:
8090 	/* vlan_macip_lens: MACLEN, VLAN tag */
8091 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
8092 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8093 
8094 	fceof_saidx |= itd->sa_idx;
8095 	type_tucmd |= itd->flags | itd->trailer_len;
8096 
8097 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd, 0);
8098 }
8099 
8100 #define IXGBE_SET_FLAG(_input, _flag, _result) \
8101 	((_flag <= _result) ? \
8102 	 ((u32)(_input & _flag) * (_result / _flag)) : \
8103 	 ((u32)(_input & _flag) / (_flag / _result)))
8104 
8105 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
8106 {
8107 	/* set type for advanced descriptor with frame checksum insertion */
8108 	u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8109 		       IXGBE_ADVTXD_DCMD_DEXT |
8110 		       IXGBE_ADVTXD_DCMD_IFCS;
8111 
8112 	/* set HW vlan bit if vlan is present */
8113 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
8114 				   IXGBE_ADVTXD_DCMD_VLE);
8115 
8116 	/* set segmentation enable bits for TSO/FSO */
8117 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
8118 				   IXGBE_ADVTXD_DCMD_TSE);
8119 
8120 	/* set timestamp bit if present */
8121 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
8122 				   IXGBE_ADVTXD_MAC_TSTAMP);
8123 
8124 	/* insert frame checksum */
8125 	cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
8126 
8127 	return cmd_type;
8128 }
8129 
8130 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
8131 				   u32 tx_flags, unsigned int paylen)
8132 {
8133 	u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
8134 
8135 	/* enable L4 checksum for TSO and TX checksum offload */
8136 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8137 					IXGBE_TX_FLAGS_CSUM,
8138 					IXGBE_ADVTXD_POPTS_TXSM);
8139 
8140 	/* enable IPv4 checksum for TSO */
8141 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8142 					IXGBE_TX_FLAGS_IPV4,
8143 					IXGBE_ADVTXD_POPTS_IXSM);
8144 
8145 	/* enable IPsec */
8146 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8147 					IXGBE_TX_FLAGS_IPSEC,
8148 					IXGBE_ADVTXD_POPTS_IPSEC);
8149 
8150 	/*
8151 	 * Check Context must be set if Tx switch is enabled, which it
8152 	 * always is for case where virtual functions are running
8153 	 */
8154 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8155 					IXGBE_TX_FLAGS_CC,
8156 					IXGBE_ADVTXD_CC);
8157 
8158 	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
8159 }
8160 
8161 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8162 {
8163 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
8164 
8165 	/* Herbert's original patch had:
8166 	 *  smp_mb__after_netif_stop_queue();
8167 	 * but since that doesn't exist yet, just open code it.
8168 	 */
8169 	smp_mb();
8170 
8171 	/* We need to check again in a case another CPU has just
8172 	 * made room available.
8173 	 */
8174 	if (likely(ixgbe_desc_unused(tx_ring) < size))
8175 		return -EBUSY;
8176 
8177 	/* A reprieve! - use start_queue because it doesn't call schedule */
8178 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
8179 	++tx_ring->tx_stats.restart_queue;
8180 	return 0;
8181 }
8182 
8183 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8184 {
8185 	if (likely(ixgbe_desc_unused(tx_ring) >= size))
8186 		return 0;
8187 
8188 	return __ixgbe_maybe_stop_tx(tx_ring, size);
8189 }
8190 
8191 static int ixgbe_tx_map(struct ixgbe_ring *tx_ring,
8192 			struct ixgbe_tx_buffer *first,
8193 			const u8 hdr_len)
8194 {
8195 	struct sk_buff *skb = first->skb;
8196 	struct ixgbe_tx_buffer *tx_buffer;
8197 	union ixgbe_adv_tx_desc *tx_desc;
8198 	skb_frag_t *frag;
8199 	dma_addr_t dma;
8200 	unsigned int data_len, size;
8201 	u32 tx_flags = first->tx_flags;
8202 	u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
8203 	u16 i = tx_ring->next_to_use;
8204 
8205 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
8206 
8207 	ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
8208 
8209 	size = skb_headlen(skb);
8210 	data_len = skb->data_len;
8211 
8212 #ifdef IXGBE_FCOE
8213 	if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
8214 		if (data_len < sizeof(struct fcoe_crc_eof)) {
8215 			size -= sizeof(struct fcoe_crc_eof) - data_len;
8216 			data_len = 0;
8217 		} else {
8218 			data_len -= sizeof(struct fcoe_crc_eof);
8219 		}
8220 	}
8221 
8222 #endif
8223 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
8224 
8225 	tx_buffer = first;
8226 
8227 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
8228 		if (dma_mapping_error(tx_ring->dev, dma))
8229 			goto dma_error;
8230 
8231 		/* record length, and DMA address */
8232 		dma_unmap_len_set(tx_buffer, len, size);
8233 		dma_unmap_addr_set(tx_buffer, dma, dma);
8234 
8235 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
8236 
8237 		while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
8238 			tx_desc->read.cmd_type_len =
8239 				cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
8240 
8241 			i++;
8242 			tx_desc++;
8243 			if (i == tx_ring->count) {
8244 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8245 				i = 0;
8246 			}
8247 			tx_desc->read.olinfo_status = 0;
8248 
8249 			dma += IXGBE_MAX_DATA_PER_TXD;
8250 			size -= IXGBE_MAX_DATA_PER_TXD;
8251 
8252 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
8253 		}
8254 
8255 		if (likely(!data_len))
8256 			break;
8257 
8258 		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
8259 
8260 		i++;
8261 		tx_desc++;
8262 		if (i == tx_ring->count) {
8263 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8264 			i = 0;
8265 		}
8266 		tx_desc->read.olinfo_status = 0;
8267 
8268 #ifdef IXGBE_FCOE
8269 		size = min_t(unsigned int, data_len, skb_frag_size(frag));
8270 #else
8271 		size = skb_frag_size(frag);
8272 #endif
8273 		data_len -= size;
8274 
8275 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
8276 				       DMA_TO_DEVICE);
8277 
8278 		tx_buffer = &tx_ring->tx_buffer_info[i];
8279 	}
8280 
8281 	/* write last descriptor with RS and EOP bits */
8282 	cmd_type |= size | IXGBE_TXD_CMD;
8283 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8284 
8285 	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
8286 
8287 	/* set the timestamp */
8288 	first->time_stamp = jiffies;
8289 
8290 	skb_tx_timestamp(skb);
8291 
8292 	/*
8293 	 * Force memory writes to complete before letting h/w know there
8294 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
8295 	 * memory model archs, such as IA-64).
8296 	 *
8297 	 * We also need this memory barrier to make certain all of the
8298 	 * status bits have been updated before next_to_watch is written.
8299 	 */
8300 	wmb();
8301 
8302 	/* set next_to_watch value indicating a packet is present */
8303 	first->next_to_watch = tx_desc;
8304 
8305 	i++;
8306 	if (i == tx_ring->count)
8307 		i = 0;
8308 
8309 	tx_ring->next_to_use = i;
8310 
8311 	ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
8312 
8313 	if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
8314 		writel(i, tx_ring->tail);
8315 	}
8316 
8317 	return 0;
8318 dma_error:
8319 	dev_err(tx_ring->dev, "TX DMA map failed\n");
8320 
8321 	/* clear dma mappings for failed tx_buffer_info map */
8322 	for (;;) {
8323 		tx_buffer = &tx_ring->tx_buffer_info[i];
8324 		if (dma_unmap_len(tx_buffer, len))
8325 			dma_unmap_page(tx_ring->dev,
8326 				       dma_unmap_addr(tx_buffer, dma),
8327 				       dma_unmap_len(tx_buffer, len),
8328 				       DMA_TO_DEVICE);
8329 		dma_unmap_len_set(tx_buffer, len, 0);
8330 		if (tx_buffer == first)
8331 			break;
8332 		if (i == 0)
8333 			i += tx_ring->count;
8334 		i--;
8335 	}
8336 
8337 	dev_kfree_skb_any(first->skb);
8338 	first->skb = NULL;
8339 
8340 	tx_ring->next_to_use = i;
8341 
8342 	return -1;
8343 }
8344 
8345 static void ixgbe_atr(struct ixgbe_ring *ring,
8346 		      struct ixgbe_tx_buffer *first)
8347 {
8348 	struct ixgbe_q_vector *q_vector = ring->q_vector;
8349 	union ixgbe_atr_hash_dword input = { .dword = 0 };
8350 	union ixgbe_atr_hash_dword common = { .dword = 0 };
8351 	union {
8352 		unsigned char *network;
8353 		struct iphdr *ipv4;
8354 		struct ipv6hdr *ipv6;
8355 	} hdr;
8356 	struct tcphdr *th;
8357 	unsigned int hlen;
8358 	struct sk_buff *skb;
8359 	__be16 vlan_id;
8360 	int l4_proto;
8361 
8362 	/* if ring doesn't have a interrupt vector, cannot perform ATR */
8363 	if (!q_vector)
8364 		return;
8365 
8366 	/* do nothing if sampling is disabled */
8367 	if (!ring->atr_sample_rate)
8368 		return;
8369 
8370 	ring->atr_count++;
8371 
8372 	/* currently only IPv4/IPv6 with TCP is supported */
8373 	if ((first->protocol != htons(ETH_P_IP)) &&
8374 	    (first->protocol != htons(ETH_P_IPV6)))
8375 		return;
8376 
8377 	/* snag network header to get L4 type and address */
8378 	skb = first->skb;
8379 	hdr.network = skb_network_header(skb);
8380 	if (unlikely(hdr.network <= skb->data))
8381 		return;
8382 	if (skb->encapsulation &&
8383 	    first->protocol == htons(ETH_P_IP) &&
8384 	    hdr.ipv4->protocol == IPPROTO_UDP) {
8385 		struct ixgbe_adapter *adapter = q_vector->adapter;
8386 
8387 		if (unlikely(skb_tail_pointer(skb) < hdr.network +
8388 			     VXLAN_HEADROOM))
8389 			return;
8390 
8391 		/* verify the port is recognized as VXLAN */
8392 		if (adapter->vxlan_port &&
8393 		    udp_hdr(skb)->dest == adapter->vxlan_port)
8394 			hdr.network = skb_inner_network_header(skb);
8395 
8396 		if (adapter->geneve_port &&
8397 		    udp_hdr(skb)->dest == adapter->geneve_port)
8398 			hdr.network = skb_inner_network_header(skb);
8399 	}
8400 
8401 	/* Make sure we have at least [minimum IPv4 header + TCP]
8402 	 * or [IPv6 header] bytes
8403 	 */
8404 	if (unlikely(skb_tail_pointer(skb) < hdr.network + 40))
8405 		return;
8406 
8407 	/* Currently only IPv4/IPv6 with TCP is supported */
8408 	switch (hdr.ipv4->version) {
8409 	case IPVERSION:
8410 		/* access ihl as u8 to avoid unaligned access on ia64 */
8411 		hlen = (hdr.network[0] & 0x0F) << 2;
8412 		l4_proto = hdr.ipv4->protocol;
8413 		break;
8414 	case 6:
8415 		hlen = hdr.network - skb->data;
8416 		l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
8417 		hlen -= hdr.network - skb->data;
8418 		break;
8419 	default:
8420 		return;
8421 	}
8422 
8423 	if (l4_proto != IPPROTO_TCP)
8424 		return;
8425 
8426 	if (unlikely(skb_tail_pointer(skb) < hdr.network +
8427 		     hlen + sizeof(struct tcphdr)))
8428 		return;
8429 
8430 	th = (struct tcphdr *)(hdr.network + hlen);
8431 
8432 	/* skip this packet since the socket is closing */
8433 	if (th->fin)
8434 		return;
8435 
8436 	/* sample on all syn packets or once every atr sample count */
8437 	if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
8438 		return;
8439 
8440 	/* reset sample count */
8441 	ring->atr_count = 0;
8442 
8443 	vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
8444 
8445 	/*
8446 	 * src and dst are inverted, think how the receiver sees them
8447 	 *
8448 	 * The input is broken into two sections, a non-compressed section
8449 	 * containing vm_pool, vlan_id, and flow_type.  The rest of the data
8450 	 * is XORed together and stored in the compressed dword.
8451 	 */
8452 	input.formatted.vlan_id = vlan_id;
8453 
8454 	/*
8455 	 * since src port and flex bytes occupy the same word XOR them together
8456 	 * and write the value to source port portion of compressed dword
8457 	 */
8458 	if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
8459 		common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
8460 	else
8461 		common.port.src ^= th->dest ^ first->protocol;
8462 	common.port.dst ^= th->source;
8463 
8464 	switch (hdr.ipv4->version) {
8465 	case IPVERSION:
8466 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
8467 		common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
8468 		break;
8469 	case 6:
8470 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
8471 		common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
8472 			     hdr.ipv6->saddr.s6_addr32[1] ^
8473 			     hdr.ipv6->saddr.s6_addr32[2] ^
8474 			     hdr.ipv6->saddr.s6_addr32[3] ^
8475 			     hdr.ipv6->daddr.s6_addr32[0] ^
8476 			     hdr.ipv6->daddr.s6_addr32[1] ^
8477 			     hdr.ipv6->daddr.s6_addr32[2] ^
8478 			     hdr.ipv6->daddr.s6_addr32[3];
8479 		break;
8480 	default:
8481 		break;
8482 	}
8483 
8484 	if (hdr.network != skb_network_header(skb))
8485 		input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
8486 
8487 	/* This assumes the Rx queue and Tx queue are bound to the same CPU */
8488 	ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
8489 					      input, common, ring->queue_index);
8490 }
8491 
8492 #ifdef IXGBE_FCOE
8493 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
8494 			      struct net_device *sb_dev)
8495 {
8496 	struct ixgbe_adapter *adapter;
8497 	struct ixgbe_ring_feature *f;
8498 	int txq;
8499 
8500 	if (sb_dev) {
8501 		u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
8502 		struct net_device *vdev = sb_dev;
8503 
8504 		txq = vdev->tc_to_txq[tc].offset;
8505 		txq += reciprocal_scale(skb_get_hash(skb),
8506 					vdev->tc_to_txq[tc].count);
8507 
8508 		return txq;
8509 	}
8510 
8511 	/*
8512 	 * only execute the code below if protocol is FCoE
8513 	 * or FIP and we have FCoE enabled on the adapter
8514 	 */
8515 	switch (vlan_get_protocol(skb)) {
8516 	case htons(ETH_P_FCOE):
8517 	case htons(ETH_P_FIP):
8518 		adapter = netdev_priv(dev);
8519 
8520 		if (!sb_dev && (adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
8521 			break;
8522 		fallthrough;
8523 	default:
8524 		return netdev_pick_tx(dev, skb, sb_dev);
8525 	}
8526 
8527 	f = &adapter->ring_feature[RING_F_FCOE];
8528 
8529 	txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
8530 					   smp_processor_id();
8531 
8532 	while (txq >= f->indices)
8533 		txq -= f->indices;
8534 
8535 	return txq + f->offset;
8536 }
8537 
8538 #endif
8539 int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
8540 			struct xdp_frame *xdpf)
8541 {
8542 	struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
8543 	struct ixgbe_tx_buffer *tx_buffer;
8544 	union ixgbe_adv_tx_desc *tx_desc;
8545 	u32 len, cmd_type;
8546 	dma_addr_t dma;
8547 	u16 i;
8548 
8549 	len = xdpf->len;
8550 
8551 	if (unlikely(!ixgbe_desc_unused(ring)))
8552 		return IXGBE_XDP_CONSUMED;
8553 
8554 	dma = dma_map_single(ring->dev, xdpf->data, len, DMA_TO_DEVICE);
8555 	if (dma_mapping_error(ring->dev, dma))
8556 		return IXGBE_XDP_CONSUMED;
8557 
8558 	/* record the location of the first descriptor for this packet */
8559 	tx_buffer = &ring->tx_buffer_info[ring->next_to_use];
8560 	tx_buffer->bytecount = len;
8561 	tx_buffer->gso_segs = 1;
8562 	tx_buffer->protocol = 0;
8563 
8564 	i = ring->next_to_use;
8565 	tx_desc = IXGBE_TX_DESC(ring, i);
8566 
8567 	dma_unmap_len_set(tx_buffer, len, len);
8568 	dma_unmap_addr_set(tx_buffer, dma, dma);
8569 	tx_buffer->xdpf = xdpf;
8570 
8571 	tx_desc->read.buffer_addr = cpu_to_le64(dma);
8572 
8573 	/* put descriptor type bits */
8574 	cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8575 		   IXGBE_ADVTXD_DCMD_DEXT |
8576 		   IXGBE_ADVTXD_DCMD_IFCS;
8577 	cmd_type |= len | IXGBE_TXD_CMD;
8578 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8579 	tx_desc->read.olinfo_status =
8580 		cpu_to_le32(len << IXGBE_ADVTXD_PAYLEN_SHIFT);
8581 
8582 	/* Avoid any potential race with xdp_xmit and cleanup */
8583 	smp_wmb();
8584 
8585 	/* set next_to_watch value indicating a packet is present */
8586 	i++;
8587 	if (i == ring->count)
8588 		i = 0;
8589 
8590 	tx_buffer->next_to_watch = tx_desc;
8591 	ring->next_to_use = i;
8592 
8593 	return IXGBE_XDP_TX;
8594 }
8595 
8596 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
8597 			  struct ixgbe_adapter *adapter,
8598 			  struct ixgbe_ring *tx_ring)
8599 {
8600 	struct ixgbe_tx_buffer *first;
8601 	int tso;
8602 	u32 tx_flags = 0;
8603 	unsigned short f;
8604 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
8605 	struct ixgbe_ipsec_tx_data ipsec_tx = { 0 };
8606 	__be16 protocol = skb->protocol;
8607 	u8 hdr_len = 0;
8608 
8609 	/*
8610 	 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
8611 	 *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
8612 	 *       + 2 desc gap to keep tail from touching head,
8613 	 *       + 1 desc for context descriptor,
8614 	 * otherwise try next time
8615 	 */
8616 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
8617 		count += TXD_USE_COUNT(skb_frag_size(
8618 						&skb_shinfo(skb)->frags[f]));
8619 
8620 	if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
8621 		tx_ring->tx_stats.tx_busy++;
8622 		return NETDEV_TX_BUSY;
8623 	}
8624 
8625 	/* record the location of the first descriptor for this packet */
8626 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
8627 	first->skb = skb;
8628 	first->bytecount = skb->len;
8629 	first->gso_segs = 1;
8630 
8631 	/* if we have a HW VLAN tag being added default to the HW one */
8632 	if (skb_vlan_tag_present(skb)) {
8633 		tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
8634 		tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8635 	/* else if it is a SW VLAN check the next protocol and store the tag */
8636 	} else if (protocol == htons(ETH_P_8021Q)) {
8637 		struct vlan_hdr *vhdr, _vhdr;
8638 		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
8639 		if (!vhdr)
8640 			goto out_drop;
8641 
8642 		tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
8643 				  IXGBE_TX_FLAGS_VLAN_SHIFT;
8644 		tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
8645 	}
8646 	protocol = vlan_get_protocol(skb);
8647 
8648 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
8649 	    adapter->ptp_clock) {
8650 		if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
8651 		    !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
8652 					   &adapter->state)) {
8653 			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
8654 			tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
8655 
8656 			/* schedule check for Tx timestamp */
8657 			adapter->ptp_tx_skb = skb_get(skb);
8658 			adapter->ptp_tx_start = jiffies;
8659 			schedule_work(&adapter->ptp_tx_work);
8660 		} else {
8661 			adapter->tx_hwtstamp_skipped++;
8662 		}
8663 	}
8664 
8665 #ifdef CONFIG_PCI_IOV
8666 	/*
8667 	 * Use the l2switch_enable flag - would be false if the DMA
8668 	 * Tx switch had been disabled.
8669 	 */
8670 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
8671 		tx_flags |= IXGBE_TX_FLAGS_CC;
8672 
8673 #endif
8674 	/* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
8675 	if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8676 	    ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
8677 	     (skb->priority != TC_PRIO_CONTROL))) {
8678 		tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
8679 		tx_flags |= (skb->priority & 0x7) <<
8680 					IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
8681 		if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
8682 			struct vlan_ethhdr *vhdr;
8683 
8684 			if (skb_cow_head(skb, 0))
8685 				goto out_drop;
8686 			vhdr = (struct vlan_ethhdr *)skb->data;
8687 			vhdr->h_vlan_TCI = htons(tx_flags >>
8688 						 IXGBE_TX_FLAGS_VLAN_SHIFT);
8689 		} else {
8690 			tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8691 		}
8692 	}
8693 
8694 	/* record initial flags and protocol */
8695 	first->tx_flags = tx_flags;
8696 	first->protocol = protocol;
8697 
8698 #ifdef IXGBE_FCOE
8699 	/* setup tx offload for FCoE */
8700 	if ((protocol == htons(ETH_P_FCOE)) &&
8701 	    (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
8702 		tso = ixgbe_fso(tx_ring, first, &hdr_len);
8703 		if (tso < 0)
8704 			goto out_drop;
8705 
8706 		goto xmit_fcoe;
8707 	}
8708 
8709 #endif /* IXGBE_FCOE */
8710 
8711 #ifdef CONFIG_IXGBE_IPSEC
8712 	if (xfrm_offload(skb) &&
8713 	    !ixgbe_ipsec_tx(tx_ring, first, &ipsec_tx))
8714 		goto out_drop;
8715 #endif
8716 	tso = ixgbe_tso(tx_ring, first, &hdr_len, &ipsec_tx);
8717 	if (tso < 0)
8718 		goto out_drop;
8719 	else if (!tso)
8720 		ixgbe_tx_csum(tx_ring, first, &ipsec_tx);
8721 
8722 	/* add the ATR filter if ATR is on */
8723 	if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
8724 		ixgbe_atr(tx_ring, first);
8725 
8726 #ifdef IXGBE_FCOE
8727 xmit_fcoe:
8728 #endif /* IXGBE_FCOE */
8729 	if (ixgbe_tx_map(tx_ring, first, hdr_len))
8730 		goto cleanup_tx_timestamp;
8731 
8732 	return NETDEV_TX_OK;
8733 
8734 out_drop:
8735 	dev_kfree_skb_any(first->skb);
8736 	first->skb = NULL;
8737 cleanup_tx_timestamp:
8738 	if (unlikely(tx_flags & IXGBE_TX_FLAGS_TSTAMP)) {
8739 		dev_kfree_skb_any(adapter->ptp_tx_skb);
8740 		adapter->ptp_tx_skb = NULL;
8741 		cancel_work_sync(&adapter->ptp_tx_work);
8742 		clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS, &adapter->state);
8743 	}
8744 
8745 	return NETDEV_TX_OK;
8746 }
8747 
8748 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
8749 				      struct net_device *netdev,
8750 				      struct ixgbe_ring *ring)
8751 {
8752 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8753 	struct ixgbe_ring *tx_ring;
8754 
8755 	/*
8756 	 * The minimum packet size for olinfo paylen is 17 so pad the skb
8757 	 * in order to meet this minimum size requirement.
8758 	 */
8759 	if (skb_put_padto(skb, 17))
8760 		return NETDEV_TX_OK;
8761 
8762 	tx_ring = ring ? ring : adapter->tx_ring[skb_get_queue_mapping(skb)];
8763 	if (unlikely(test_bit(__IXGBE_TX_DISABLED, &tx_ring->state)))
8764 		return NETDEV_TX_BUSY;
8765 
8766 	return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
8767 }
8768 
8769 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
8770 				    struct net_device *netdev)
8771 {
8772 	return __ixgbe_xmit_frame(skb, netdev, NULL);
8773 }
8774 
8775 /**
8776  * ixgbe_set_mac - Change the Ethernet Address of the NIC
8777  * @netdev: network interface device structure
8778  * @p: pointer to an address structure
8779  *
8780  * Returns 0 on success, negative on failure
8781  **/
8782 static int ixgbe_set_mac(struct net_device *netdev, void *p)
8783 {
8784 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8785 	struct ixgbe_hw *hw = &adapter->hw;
8786 	struct sockaddr *addr = p;
8787 
8788 	if (!is_valid_ether_addr(addr->sa_data))
8789 		return -EADDRNOTAVAIL;
8790 
8791 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
8792 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
8793 
8794 	ixgbe_mac_set_default_filter(adapter);
8795 
8796 	return 0;
8797 }
8798 
8799 static int
8800 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
8801 {
8802 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8803 	struct ixgbe_hw *hw = &adapter->hw;
8804 	u16 value;
8805 	int rc;
8806 
8807 	if (adapter->mii_bus) {
8808 		int regnum = addr;
8809 
8810 		if (devad != MDIO_DEVAD_NONE)
8811 			regnum |= (devad << 16) | MII_ADDR_C45;
8812 
8813 		return mdiobus_read(adapter->mii_bus, prtad, regnum);
8814 	}
8815 
8816 	if (prtad != hw->phy.mdio.prtad)
8817 		return -EINVAL;
8818 	rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
8819 	if (!rc)
8820 		rc = value;
8821 	return rc;
8822 }
8823 
8824 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
8825 			    u16 addr, u16 value)
8826 {
8827 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8828 	struct ixgbe_hw *hw = &adapter->hw;
8829 
8830 	if (adapter->mii_bus) {
8831 		int regnum = addr;
8832 
8833 		if (devad != MDIO_DEVAD_NONE)
8834 			regnum |= (devad << 16) | MII_ADDR_C45;
8835 
8836 		return mdiobus_write(adapter->mii_bus, prtad, regnum, value);
8837 	}
8838 
8839 	if (prtad != hw->phy.mdio.prtad)
8840 		return -EINVAL;
8841 	return hw->phy.ops.write_reg(hw, addr, devad, value);
8842 }
8843 
8844 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
8845 {
8846 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8847 
8848 	switch (cmd) {
8849 	case SIOCSHWTSTAMP:
8850 		return ixgbe_ptp_set_ts_config(adapter, req);
8851 	case SIOCGHWTSTAMP:
8852 		return ixgbe_ptp_get_ts_config(adapter, req);
8853 	case SIOCGMIIPHY:
8854 		if (!adapter->hw.phy.ops.read_reg)
8855 			return -EOPNOTSUPP;
8856 		fallthrough;
8857 	default:
8858 		return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
8859 	}
8860 }
8861 
8862 /**
8863  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
8864  * netdev->dev_addrs
8865  * @dev: network interface device structure
8866  *
8867  * Returns non-zero on failure
8868  **/
8869 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
8870 {
8871 	int err = 0;
8872 	struct ixgbe_adapter *adapter = netdev_priv(dev);
8873 	struct ixgbe_hw *hw = &adapter->hw;
8874 
8875 	if (is_valid_ether_addr(hw->mac.san_addr)) {
8876 		rtnl_lock();
8877 		err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
8878 		rtnl_unlock();
8879 
8880 		/* update SAN MAC vmdq pool selection */
8881 		hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
8882 	}
8883 	return err;
8884 }
8885 
8886 /**
8887  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
8888  * netdev->dev_addrs
8889  * @dev: network interface device structure
8890  *
8891  * Returns non-zero on failure
8892  **/
8893 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
8894 {
8895 	int err = 0;
8896 	struct ixgbe_adapter *adapter = netdev_priv(dev);
8897 	struct ixgbe_mac_info *mac = &adapter->hw.mac;
8898 
8899 	if (is_valid_ether_addr(mac->san_addr)) {
8900 		rtnl_lock();
8901 		err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
8902 		rtnl_unlock();
8903 	}
8904 	return err;
8905 }
8906 
8907 static void ixgbe_get_ring_stats64(struct rtnl_link_stats64 *stats,
8908 				   struct ixgbe_ring *ring)
8909 {
8910 	u64 bytes, packets;
8911 	unsigned int start;
8912 
8913 	if (ring) {
8914 		do {
8915 			start = u64_stats_fetch_begin_irq(&ring->syncp);
8916 			packets = ring->stats.packets;
8917 			bytes   = ring->stats.bytes;
8918 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8919 		stats->tx_packets += packets;
8920 		stats->tx_bytes   += bytes;
8921 	}
8922 }
8923 
8924 static void ixgbe_get_stats64(struct net_device *netdev,
8925 			      struct rtnl_link_stats64 *stats)
8926 {
8927 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8928 	int i;
8929 
8930 	rcu_read_lock();
8931 	for (i = 0; i < adapter->num_rx_queues; i++) {
8932 		struct ixgbe_ring *ring = READ_ONCE(adapter->rx_ring[i]);
8933 		u64 bytes, packets;
8934 		unsigned int start;
8935 
8936 		if (ring) {
8937 			do {
8938 				start = u64_stats_fetch_begin_irq(&ring->syncp);
8939 				packets = ring->stats.packets;
8940 				bytes   = ring->stats.bytes;
8941 			} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8942 			stats->rx_packets += packets;
8943 			stats->rx_bytes   += bytes;
8944 		}
8945 	}
8946 
8947 	for (i = 0; i < adapter->num_tx_queues; i++) {
8948 		struct ixgbe_ring *ring = READ_ONCE(adapter->tx_ring[i]);
8949 
8950 		ixgbe_get_ring_stats64(stats, ring);
8951 	}
8952 	for (i = 0; i < adapter->num_xdp_queues; i++) {
8953 		struct ixgbe_ring *ring = READ_ONCE(adapter->xdp_ring[i]);
8954 
8955 		ixgbe_get_ring_stats64(stats, ring);
8956 	}
8957 	rcu_read_unlock();
8958 
8959 	/* following stats updated by ixgbe_watchdog_task() */
8960 	stats->multicast	= netdev->stats.multicast;
8961 	stats->rx_errors	= netdev->stats.rx_errors;
8962 	stats->rx_length_errors	= netdev->stats.rx_length_errors;
8963 	stats->rx_crc_errors	= netdev->stats.rx_crc_errors;
8964 	stats->rx_missed_errors	= netdev->stats.rx_missed_errors;
8965 }
8966 
8967 #ifdef CONFIG_IXGBE_DCB
8968 /**
8969  * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
8970  * @adapter: pointer to ixgbe_adapter
8971  * @tc: number of traffic classes currently enabled
8972  *
8973  * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
8974  * 802.1Q priority maps to a packet buffer that exists.
8975  */
8976 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
8977 {
8978 	struct ixgbe_hw *hw = &adapter->hw;
8979 	u32 reg, rsave;
8980 	int i;
8981 
8982 	/* 82598 have a static priority to TC mapping that can not
8983 	 * be changed so no validation is needed.
8984 	 */
8985 	if (hw->mac.type == ixgbe_mac_82598EB)
8986 		return;
8987 
8988 	reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
8989 	rsave = reg;
8990 
8991 	for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
8992 		u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
8993 
8994 		/* If up2tc is out of bounds default to zero */
8995 		if (up2tc > tc)
8996 			reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
8997 	}
8998 
8999 	if (reg != rsave)
9000 		IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
9001 
9002 	return;
9003 }
9004 
9005 /**
9006  * ixgbe_set_prio_tc_map - Configure netdev prio tc map
9007  * @adapter: Pointer to adapter struct
9008  *
9009  * Populate the netdev user priority to tc map
9010  */
9011 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
9012 {
9013 	struct net_device *dev = adapter->netdev;
9014 	struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
9015 	struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
9016 	u8 prio;
9017 
9018 	for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
9019 		u8 tc = 0;
9020 
9021 		if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
9022 			tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
9023 		else if (ets)
9024 			tc = ets->prio_tc[prio];
9025 
9026 		netdev_set_prio_tc_map(dev, prio, tc);
9027 	}
9028 }
9029 
9030 #endif /* CONFIG_IXGBE_DCB */
9031 static int ixgbe_reassign_macvlan_pool(struct net_device *vdev,
9032 				       struct netdev_nested_priv *priv)
9033 {
9034 	struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)priv->data;
9035 	struct ixgbe_fwd_adapter *accel;
9036 	int pool;
9037 
9038 	/* we only care about macvlans... */
9039 	if (!netif_is_macvlan(vdev))
9040 		return 0;
9041 
9042 	/* that have hardware offload enabled... */
9043 	accel = macvlan_accel_priv(vdev);
9044 	if (!accel)
9045 		return 0;
9046 
9047 	/* If we can relocate to a different bit do so */
9048 	pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
9049 	if (pool < adapter->num_rx_pools) {
9050 		set_bit(pool, adapter->fwd_bitmask);
9051 		accel->pool = pool;
9052 		return 0;
9053 	}
9054 
9055 	/* if we cannot find a free pool then disable the offload */
9056 	netdev_err(vdev, "L2FW offload disabled due to lack of queue resources\n");
9057 	macvlan_release_l2fw_offload(vdev);
9058 
9059 	/* unbind the queues and drop the subordinate channel config */
9060 	netdev_unbind_sb_channel(adapter->netdev, vdev);
9061 	netdev_set_sb_channel(vdev, 0);
9062 
9063 	kfree(accel);
9064 
9065 	return 0;
9066 }
9067 
9068 static void ixgbe_defrag_macvlan_pools(struct net_device *dev)
9069 {
9070 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9071 	struct netdev_nested_priv priv = {
9072 		.data = (void *)adapter,
9073 	};
9074 
9075 	/* flush any stale bits out of the fwd bitmask */
9076 	bitmap_clear(adapter->fwd_bitmask, 1, 63);
9077 
9078 	/* walk through upper devices reassigning pools */
9079 	netdev_walk_all_upper_dev_rcu(dev, ixgbe_reassign_macvlan_pool,
9080 				      &priv);
9081 }
9082 
9083 /**
9084  * ixgbe_setup_tc - configure net_device for multiple traffic classes
9085  *
9086  * @dev: net device to configure
9087  * @tc: number of traffic classes to enable
9088  */
9089 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
9090 {
9091 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9092 	struct ixgbe_hw *hw = &adapter->hw;
9093 
9094 	/* Hardware supports up to 8 traffic classes */
9095 	if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
9096 		return -EINVAL;
9097 
9098 	if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
9099 		return -EINVAL;
9100 
9101 	/* Hardware has to reinitialize queues and interrupts to
9102 	 * match packet buffer alignment. Unfortunately, the
9103 	 * hardware is not flexible enough to do this dynamically.
9104 	 */
9105 	if (netif_running(dev))
9106 		ixgbe_close(dev);
9107 	else
9108 		ixgbe_reset(adapter);
9109 
9110 	ixgbe_clear_interrupt_scheme(adapter);
9111 
9112 #ifdef CONFIG_IXGBE_DCB
9113 	if (tc) {
9114 		if (adapter->xdp_prog) {
9115 			e_warn(probe, "DCB is not supported with XDP\n");
9116 
9117 			ixgbe_init_interrupt_scheme(adapter);
9118 			if (netif_running(dev))
9119 				ixgbe_open(dev);
9120 			return -EINVAL;
9121 		}
9122 
9123 		netdev_set_num_tc(dev, tc);
9124 		ixgbe_set_prio_tc_map(adapter);
9125 
9126 		adapter->hw_tcs = tc;
9127 		adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
9128 
9129 		if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
9130 			adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
9131 			adapter->hw.fc.requested_mode = ixgbe_fc_none;
9132 		}
9133 	} else {
9134 		netdev_reset_tc(dev);
9135 
9136 		if (adapter->hw.mac.type == ixgbe_mac_82598EB)
9137 			adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
9138 
9139 		adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
9140 		adapter->hw_tcs = tc;
9141 
9142 		adapter->temp_dcb_cfg.pfc_mode_enable = false;
9143 		adapter->dcb_cfg.pfc_mode_enable = false;
9144 	}
9145 
9146 	ixgbe_validate_rtr(adapter, tc);
9147 
9148 #endif /* CONFIG_IXGBE_DCB */
9149 	ixgbe_init_interrupt_scheme(adapter);
9150 
9151 	ixgbe_defrag_macvlan_pools(dev);
9152 
9153 	if (netif_running(dev))
9154 		return ixgbe_open(dev);
9155 
9156 	return 0;
9157 }
9158 
9159 static int ixgbe_delete_clsu32(struct ixgbe_adapter *adapter,
9160 			       struct tc_cls_u32_offload *cls)
9161 {
9162 	u32 hdl = cls->knode.handle;
9163 	u32 uhtid = TC_U32_USERHTID(cls->knode.handle);
9164 	u32 loc = cls->knode.handle & 0xfffff;
9165 	int err = 0, i, j;
9166 	struct ixgbe_jump_table *jump = NULL;
9167 
9168 	if (loc > IXGBE_MAX_HW_ENTRIES)
9169 		return -EINVAL;
9170 
9171 	if ((uhtid != 0x800) && (uhtid >= IXGBE_MAX_LINK_HANDLE))
9172 		return -EINVAL;
9173 
9174 	/* Clear this filter in the link data it is associated with */
9175 	if (uhtid != 0x800) {
9176 		jump = adapter->jump_tables[uhtid];
9177 		if (!jump)
9178 			return -EINVAL;
9179 		if (!test_bit(loc - 1, jump->child_loc_map))
9180 			return -EINVAL;
9181 		clear_bit(loc - 1, jump->child_loc_map);
9182 	}
9183 
9184 	/* Check if the filter being deleted is a link */
9185 	for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9186 		jump = adapter->jump_tables[i];
9187 		if (jump && jump->link_hdl == hdl) {
9188 			/* Delete filters in the hardware in the child hash
9189 			 * table associated with this link
9190 			 */
9191 			for (j = 0; j < IXGBE_MAX_HW_ENTRIES; j++) {
9192 				if (!test_bit(j, jump->child_loc_map))
9193 					continue;
9194 				spin_lock(&adapter->fdir_perfect_lock);
9195 				err = ixgbe_update_ethtool_fdir_entry(adapter,
9196 								      NULL,
9197 								      j + 1);
9198 				spin_unlock(&adapter->fdir_perfect_lock);
9199 				clear_bit(j, jump->child_loc_map);
9200 			}
9201 			/* Remove resources for this link */
9202 			kfree(jump->input);
9203 			kfree(jump->mask);
9204 			kfree(jump);
9205 			adapter->jump_tables[i] = NULL;
9206 			return err;
9207 		}
9208 	}
9209 
9210 	spin_lock(&adapter->fdir_perfect_lock);
9211 	err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, loc);
9212 	spin_unlock(&adapter->fdir_perfect_lock);
9213 	return err;
9214 }
9215 
9216 static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter *adapter,
9217 					    struct tc_cls_u32_offload *cls)
9218 {
9219 	u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9220 
9221 	if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9222 		return -EINVAL;
9223 
9224 	/* This ixgbe devices do not support hash tables at the moment
9225 	 * so abort when given hash tables.
9226 	 */
9227 	if (cls->hnode.divisor > 0)
9228 		return -EINVAL;
9229 
9230 	set_bit(uhtid - 1, &adapter->tables);
9231 	return 0;
9232 }
9233 
9234 static int ixgbe_configure_clsu32_del_hnode(struct ixgbe_adapter *adapter,
9235 					    struct tc_cls_u32_offload *cls)
9236 {
9237 	u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9238 
9239 	if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9240 		return -EINVAL;
9241 
9242 	clear_bit(uhtid - 1, &adapter->tables);
9243 	return 0;
9244 }
9245 
9246 #ifdef CONFIG_NET_CLS_ACT
9247 struct upper_walk_data {
9248 	struct ixgbe_adapter *adapter;
9249 	u64 action;
9250 	int ifindex;
9251 	u8 queue;
9252 };
9253 
9254 static int get_macvlan_queue(struct net_device *upper,
9255 			     struct netdev_nested_priv *priv)
9256 {
9257 	if (netif_is_macvlan(upper)) {
9258 		struct ixgbe_fwd_adapter *vadapter = macvlan_accel_priv(upper);
9259 		struct ixgbe_adapter *adapter;
9260 		struct upper_walk_data *data;
9261 		int ifindex;
9262 
9263 		data = (struct upper_walk_data *)priv->data;
9264 		ifindex = data->ifindex;
9265 		adapter = data->adapter;
9266 		if (vadapter && upper->ifindex == ifindex) {
9267 			data->queue = adapter->rx_ring[vadapter->rx_base_queue]->reg_idx;
9268 			data->action = data->queue;
9269 			return 1;
9270 		}
9271 	}
9272 
9273 	return 0;
9274 }
9275 
9276 static int handle_redirect_action(struct ixgbe_adapter *adapter, int ifindex,
9277 				  u8 *queue, u64 *action)
9278 {
9279 	struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
9280 	unsigned int num_vfs = adapter->num_vfs, vf;
9281 	struct netdev_nested_priv priv;
9282 	struct upper_walk_data data;
9283 	struct net_device *upper;
9284 
9285 	/* redirect to a SRIOV VF */
9286 	for (vf = 0; vf < num_vfs; ++vf) {
9287 		upper = pci_get_drvdata(adapter->vfinfo[vf].vfdev);
9288 		if (upper->ifindex == ifindex) {
9289 			*queue = vf * __ALIGN_MASK(1, ~vmdq->mask);
9290 			*action = vf + 1;
9291 			*action <<= ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF;
9292 			return 0;
9293 		}
9294 	}
9295 
9296 	/* redirect to a offloaded macvlan netdev */
9297 	data.adapter = adapter;
9298 	data.ifindex = ifindex;
9299 	data.action = 0;
9300 	data.queue = 0;
9301 	priv.data = (void *)&data;
9302 	if (netdev_walk_all_upper_dev_rcu(adapter->netdev,
9303 					  get_macvlan_queue, &priv)) {
9304 		*action = data.action;
9305 		*queue = data.queue;
9306 
9307 		return 0;
9308 	}
9309 
9310 	return -EINVAL;
9311 }
9312 
9313 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9314 			    struct tcf_exts *exts, u64 *action, u8 *queue)
9315 {
9316 	const struct tc_action *a;
9317 	int i;
9318 
9319 	if (!tcf_exts_has_actions(exts))
9320 		return -EINVAL;
9321 
9322 	tcf_exts_for_each_action(i, a, exts) {
9323 		/* Drop action */
9324 		if (is_tcf_gact_shot(a)) {
9325 			*action = IXGBE_FDIR_DROP_QUEUE;
9326 			*queue = IXGBE_FDIR_DROP_QUEUE;
9327 			return 0;
9328 		}
9329 
9330 		/* Redirect to a VF or a offloaded macvlan */
9331 		if (is_tcf_mirred_egress_redirect(a)) {
9332 			struct net_device *dev = tcf_mirred_dev(a);
9333 
9334 			if (!dev)
9335 				return -EINVAL;
9336 			return handle_redirect_action(adapter, dev->ifindex,
9337 						      queue, action);
9338 		}
9339 
9340 		return -EINVAL;
9341 	}
9342 
9343 	return -EINVAL;
9344 }
9345 #else
9346 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9347 			    struct tcf_exts *exts, u64 *action, u8 *queue)
9348 {
9349 	return -EINVAL;
9350 }
9351 #endif /* CONFIG_NET_CLS_ACT */
9352 
9353 static int ixgbe_clsu32_build_input(struct ixgbe_fdir_filter *input,
9354 				    union ixgbe_atr_input *mask,
9355 				    struct tc_cls_u32_offload *cls,
9356 				    struct ixgbe_mat_field *field_ptr,
9357 				    struct ixgbe_nexthdr *nexthdr)
9358 {
9359 	int i, j, off;
9360 	__be32 val, m;
9361 	bool found_entry = false, found_jump_field = false;
9362 
9363 	for (i = 0; i < cls->knode.sel->nkeys; i++) {
9364 		off = cls->knode.sel->keys[i].off;
9365 		val = cls->knode.sel->keys[i].val;
9366 		m = cls->knode.sel->keys[i].mask;
9367 
9368 		for (j = 0; field_ptr[j].val; j++) {
9369 			if (field_ptr[j].off == off) {
9370 				field_ptr[j].val(input, mask, (__force u32)val,
9371 						 (__force u32)m);
9372 				input->filter.formatted.flow_type |=
9373 					field_ptr[j].type;
9374 				found_entry = true;
9375 				break;
9376 			}
9377 		}
9378 		if (nexthdr) {
9379 			if (nexthdr->off == cls->knode.sel->keys[i].off &&
9380 			    nexthdr->val ==
9381 			    (__force u32)cls->knode.sel->keys[i].val &&
9382 			    nexthdr->mask ==
9383 			    (__force u32)cls->knode.sel->keys[i].mask)
9384 				found_jump_field = true;
9385 			else
9386 				continue;
9387 		}
9388 	}
9389 
9390 	if (nexthdr && !found_jump_field)
9391 		return -EINVAL;
9392 
9393 	if (!found_entry)
9394 		return 0;
9395 
9396 	mask->formatted.flow_type = IXGBE_ATR_L4TYPE_IPV6_MASK |
9397 				    IXGBE_ATR_L4TYPE_MASK;
9398 
9399 	if (input->filter.formatted.flow_type == IXGBE_ATR_FLOW_TYPE_IPV4)
9400 		mask->formatted.flow_type &= IXGBE_ATR_L4TYPE_IPV6_MASK;
9401 
9402 	return 0;
9403 }
9404 
9405 static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
9406 				  struct tc_cls_u32_offload *cls)
9407 {
9408 	__be16 protocol = cls->common.protocol;
9409 	u32 loc = cls->knode.handle & 0xfffff;
9410 	struct ixgbe_hw *hw = &adapter->hw;
9411 	struct ixgbe_mat_field *field_ptr;
9412 	struct ixgbe_fdir_filter *input = NULL;
9413 	union ixgbe_atr_input *mask = NULL;
9414 	struct ixgbe_jump_table *jump = NULL;
9415 	int i, err = -EINVAL;
9416 	u8 queue;
9417 	u32 uhtid, link_uhtid;
9418 
9419 	uhtid = TC_U32_USERHTID(cls->knode.handle);
9420 	link_uhtid = TC_U32_USERHTID(cls->knode.link_handle);
9421 
9422 	/* At the moment cls_u32 jumps to network layer and skips past
9423 	 * L2 headers. The canonical method to match L2 frames is to use
9424 	 * negative values. However this is error prone at best but really
9425 	 * just broken because there is no way to "know" what sort of hdr
9426 	 * is in front of the network layer. Fix cls_u32 to support L2
9427 	 * headers when needed.
9428 	 */
9429 	if (protocol != htons(ETH_P_IP))
9430 		return err;
9431 
9432 	if (loc >= ((1024 << adapter->fdir_pballoc) - 2)) {
9433 		e_err(drv, "Location out of range\n");
9434 		return err;
9435 	}
9436 
9437 	/* cls u32 is a graph starting at root node 0x800. The driver tracks
9438 	 * links and also the fields used to advance the parser across each
9439 	 * link (e.g. nexthdr/eat parameters from 'tc'). This way we can map
9440 	 * the u32 graph onto the hardware parse graph denoted in ixgbe_model.h
9441 	 * To add support for new nodes update ixgbe_model.h parse structures
9442 	 * this function _should_ be generic try not to hardcode values here.
9443 	 */
9444 	if (uhtid == 0x800) {
9445 		field_ptr = (adapter->jump_tables[0])->mat;
9446 	} else {
9447 		if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9448 			return err;
9449 		if (!adapter->jump_tables[uhtid])
9450 			return err;
9451 		field_ptr = (adapter->jump_tables[uhtid])->mat;
9452 	}
9453 
9454 	if (!field_ptr)
9455 		return err;
9456 
9457 	/* At this point we know the field_ptr is valid and need to either
9458 	 * build cls_u32 link or attach filter. Because adding a link to
9459 	 * a handle that does not exist is invalid and the same for adding
9460 	 * rules to handles that don't exist.
9461 	 */
9462 
9463 	if (link_uhtid) {
9464 		struct ixgbe_nexthdr *nexthdr = ixgbe_ipv4_jumps;
9465 
9466 		if (link_uhtid >= IXGBE_MAX_LINK_HANDLE)
9467 			return err;
9468 
9469 		if (!test_bit(link_uhtid - 1, &adapter->tables))
9470 			return err;
9471 
9472 		/* Multiple filters as links to the same hash table are not
9473 		 * supported. To add a new filter with the same next header
9474 		 * but different match/jump conditions, create a new hash table
9475 		 * and link to it.
9476 		 */
9477 		if (adapter->jump_tables[link_uhtid] &&
9478 		    (adapter->jump_tables[link_uhtid])->link_hdl) {
9479 			e_err(drv, "Link filter exists for link: %x\n",
9480 			      link_uhtid);
9481 			return err;
9482 		}
9483 
9484 		for (i = 0; nexthdr[i].jump; i++) {
9485 			if (nexthdr[i].o != cls->knode.sel->offoff ||
9486 			    nexthdr[i].s != cls->knode.sel->offshift ||
9487 			    nexthdr[i].m !=
9488 			    (__force u32)cls->knode.sel->offmask)
9489 				return err;
9490 
9491 			jump = kzalloc(sizeof(*jump), GFP_KERNEL);
9492 			if (!jump)
9493 				return -ENOMEM;
9494 			input = kzalloc(sizeof(*input), GFP_KERNEL);
9495 			if (!input) {
9496 				err = -ENOMEM;
9497 				goto free_jump;
9498 			}
9499 			mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9500 			if (!mask) {
9501 				err = -ENOMEM;
9502 				goto free_input;
9503 			}
9504 			jump->input = input;
9505 			jump->mask = mask;
9506 			jump->link_hdl = cls->knode.handle;
9507 
9508 			err = ixgbe_clsu32_build_input(input, mask, cls,
9509 						       field_ptr, &nexthdr[i]);
9510 			if (!err) {
9511 				jump->mat = nexthdr[i].jump;
9512 				adapter->jump_tables[link_uhtid] = jump;
9513 				break;
9514 			} else {
9515 				kfree(mask);
9516 				kfree(input);
9517 				kfree(jump);
9518 			}
9519 		}
9520 		return 0;
9521 	}
9522 
9523 	input = kzalloc(sizeof(*input), GFP_KERNEL);
9524 	if (!input)
9525 		return -ENOMEM;
9526 	mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9527 	if (!mask) {
9528 		err = -ENOMEM;
9529 		goto free_input;
9530 	}
9531 
9532 	if ((uhtid != 0x800) && (adapter->jump_tables[uhtid])) {
9533 		if ((adapter->jump_tables[uhtid])->input)
9534 			memcpy(input, (adapter->jump_tables[uhtid])->input,
9535 			       sizeof(*input));
9536 		if ((adapter->jump_tables[uhtid])->mask)
9537 			memcpy(mask, (adapter->jump_tables[uhtid])->mask,
9538 			       sizeof(*mask));
9539 
9540 		/* Lookup in all child hash tables if this location is already
9541 		 * filled with a filter
9542 		 */
9543 		for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9544 			struct ixgbe_jump_table *link = adapter->jump_tables[i];
9545 
9546 			if (link && (test_bit(loc - 1, link->child_loc_map))) {
9547 				e_err(drv, "Filter exists in location: %x\n",
9548 				      loc);
9549 				err = -EINVAL;
9550 				goto err_out;
9551 			}
9552 		}
9553 	}
9554 	err = ixgbe_clsu32_build_input(input, mask, cls, field_ptr, NULL);
9555 	if (err)
9556 		goto err_out;
9557 
9558 	err = parse_tc_actions(adapter, cls->knode.exts, &input->action,
9559 			       &queue);
9560 	if (err < 0)
9561 		goto err_out;
9562 
9563 	input->sw_idx = loc;
9564 
9565 	spin_lock(&adapter->fdir_perfect_lock);
9566 
9567 	if (hlist_empty(&adapter->fdir_filter_list)) {
9568 		memcpy(&adapter->fdir_mask, mask, sizeof(*mask));
9569 		err = ixgbe_fdir_set_input_mask_82599(hw, mask);
9570 		if (err)
9571 			goto err_out_w_lock;
9572 	} else if (memcmp(&adapter->fdir_mask, mask, sizeof(*mask))) {
9573 		err = -EINVAL;
9574 		goto err_out_w_lock;
9575 	}
9576 
9577 	ixgbe_atr_compute_perfect_hash_82599(&input->filter, mask);
9578 	err = ixgbe_fdir_write_perfect_filter_82599(hw, &input->filter,
9579 						    input->sw_idx, queue);
9580 	if (!err)
9581 		ixgbe_update_ethtool_fdir_entry(adapter, input, input->sw_idx);
9582 	spin_unlock(&adapter->fdir_perfect_lock);
9583 
9584 	if ((uhtid != 0x800) && (adapter->jump_tables[uhtid]))
9585 		set_bit(loc - 1, (adapter->jump_tables[uhtid])->child_loc_map);
9586 
9587 	kfree(mask);
9588 	return err;
9589 err_out_w_lock:
9590 	spin_unlock(&adapter->fdir_perfect_lock);
9591 err_out:
9592 	kfree(mask);
9593 free_input:
9594 	kfree(input);
9595 free_jump:
9596 	kfree(jump);
9597 	return err;
9598 }
9599 
9600 static int ixgbe_setup_tc_cls_u32(struct ixgbe_adapter *adapter,
9601 				  struct tc_cls_u32_offload *cls_u32)
9602 {
9603 	switch (cls_u32->command) {
9604 	case TC_CLSU32_NEW_KNODE:
9605 	case TC_CLSU32_REPLACE_KNODE:
9606 		return ixgbe_configure_clsu32(adapter, cls_u32);
9607 	case TC_CLSU32_DELETE_KNODE:
9608 		return ixgbe_delete_clsu32(adapter, cls_u32);
9609 	case TC_CLSU32_NEW_HNODE:
9610 	case TC_CLSU32_REPLACE_HNODE:
9611 		return ixgbe_configure_clsu32_add_hnode(adapter, cls_u32);
9612 	case TC_CLSU32_DELETE_HNODE:
9613 		return ixgbe_configure_clsu32_del_hnode(adapter, cls_u32);
9614 	default:
9615 		return -EOPNOTSUPP;
9616 	}
9617 }
9618 
9619 static int ixgbe_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
9620 				   void *cb_priv)
9621 {
9622 	struct ixgbe_adapter *adapter = cb_priv;
9623 
9624 	if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
9625 		return -EOPNOTSUPP;
9626 
9627 	switch (type) {
9628 	case TC_SETUP_CLSU32:
9629 		return ixgbe_setup_tc_cls_u32(adapter, type_data);
9630 	default:
9631 		return -EOPNOTSUPP;
9632 	}
9633 }
9634 
9635 static int ixgbe_setup_tc_mqprio(struct net_device *dev,
9636 				 struct tc_mqprio_qopt *mqprio)
9637 {
9638 	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
9639 	return ixgbe_setup_tc(dev, mqprio->num_tc);
9640 }
9641 
9642 static LIST_HEAD(ixgbe_block_cb_list);
9643 
9644 static int __ixgbe_setup_tc(struct net_device *dev, enum tc_setup_type type,
9645 			    void *type_data)
9646 {
9647 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9648 
9649 	switch (type) {
9650 	case TC_SETUP_BLOCK:
9651 		return flow_block_cb_setup_simple(type_data,
9652 						  &ixgbe_block_cb_list,
9653 						  ixgbe_setup_tc_block_cb,
9654 						  adapter, adapter, true);
9655 	case TC_SETUP_QDISC_MQPRIO:
9656 		return ixgbe_setup_tc_mqprio(dev, type_data);
9657 	default:
9658 		return -EOPNOTSUPP;
9659 	}
9660 }
9661 
9662 #ifdef CONFIG_PCI_IOV
9663 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
9664 {
9665 	struct net_device *netdev = adapter->netdev;
9666 
9667 	rtnl_lock();
9668 	ixgbe_setup_tc(netdev, adapter->hw_tcs);
9669 	rtnl_unlock();
9670 }
9671 
9672 #endif
9673 void ixgbe_do_reset(struct net_device *netdev)
9674 {
9675 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
9676 
9677 	if (netif_running(netdev))
9678 		ixgbe_reinit_locked(adapter);
9679 	else
9680 		ixgbe_reset(adapter);
9681 }
9682 
9683 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
9684 					    netdev_features_t features)
9685 {
9686 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
9687 
9688 	/* If Rx checksum is disabled, then RSC/LRO should also be disabled */
9689 	if (!(features & NETIF_F_RXCSUM))
9690 		features &= ~NETIF_F_LRO;
9691 
9692 	/* Turn off LRO if not RSC capable */
9693 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
9694 		features &= ~NETIF_F_LRO;
9695 
9696 	if (adapter->xdp_prog && (features & NETIF_F_LRO)) {
9697 		e_dev_err("LRO is not supported with XDP\n");
9698 		features &= ~NETIF_F_LRO;
9699 	}
9700 
9701 	return features;
9702 }
9703 
9704 static void ixgbe_reset_l2fw_offload(struct ixgbe_adapter *adapter)
9705 {
9706 	int rss = min_t(int, ixgbe_max_rss_indices(adapter),
9707 			num_online_cpus());
9708 
9709 	/* go back to full RSS if we're not running SR-IOV */
9710 	if (!adapter->ring_feature[RING_F_VMDQ].offset)
9711 		adapter->flags &= ~(IXGBE_FLAG_VMDQ_ENABLED |
9712 				    IXGBE_FLAG_SRIOV_ENABLED);
9713 
9714 	adapter->ring_feature[RING_F_RSS].limit = rss;
9715 	adapter->ring_feature[RING_F_VMDQ].limit = 1;
9716 
9717 	ixgbe_setup_tc(adapter->netdev, adapter->hw_tcs);
9718 }
9719 
9720 static int ixgbe_set_features(struct net_device *netdev,
9721 			      netdev_features_t features)
9722 {
9723 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
9724 	netdev_features_t changed = netdev->features ^ features;
9725 	bool need_reset = false;
9726 
9727 	/* Make sure RSC matches LRO, reset if change */
9728 	if (!(features & NETIF_F_LRO)) {
9729 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
9730 			need_reset = true;
9731 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
9732 	} else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
9733 		   !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
9734 		if (adapter->rx_itr_setting == 1 ||
9735 		    adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
9736 			adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
9737 			need_reset = true;
9738 		} else if ((changed ^ features) & NETIF_F_LRO) {
9739 			e_info(probe, "rx-usecs set too low, "
9740 			       "disabling RSC\n");
9741 		}
9742 	}
9743 
9744 	/*
9745 	 * Check if Flow Director n-tuple support or hw_tc support was
9746 	 * enabled or disabled.  If the state changed, we need to reset.
9747 	 */
9748 	if ((features & NETIF_F_NTUPLE) || (features & NETIF_F_HW_TC)) {
9749 		/* turn off ATR, enable perfect filters and reset */
9750 		if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
9751 			need_reset = true;
9752 
9753 		adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
9754 		adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9755 	} else {
9756 		/* turn off perfect filters, enable ATR and reset */
9757 		if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
9758 			need_reset = true;
9759 
9760 		adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9761 
9762 		/* We cannot enable ATR if SR-IOV is enabled */
9763 		if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED ||
9764 		    /* We cannot enable ATR if we have 2 or more tcs */
9765 		    (adapter->hw_tcs > 1) ||
9766 		    /* We cannot enable ATR if RSS is disabled */
9767 		    (adapter->ring_feature[RING_F_RSS].limit <= 1) ||
9768 		    /* A sample rate of 0 indicates ATR disabled */
9769 		    (!adapter->atr_sample_rate))
9770 			; /* do nothing not supported */
9771 		else /* otherwise supported and set the flag */
9772 			adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
9773 	}
9774 
9775 	if (changed & NETIF_F_RXALL)
9776 		need_reset = true;
9777 
9778 	netdev->features = features;
9779 
9780 	if ((changed & NETIF_F_HW_L2FW_DOFFLOAD) && adapter->num_rx_pools > 1)
9781 		ixgbe_reset_l2fw_offload(adapter);
9782 	else if (need_reset)
9783 		ixgbe_do_reset(netdev);
9784 	else if (changed & (NETIF_F_HW_VLAN_CTAG_RX |
9785 			    NETIF_F_HW_VLAN_CTAG_FILTER))
9786 		ixgbe_set_rx_mode(netdev);
9787 
9788 	return 1;
9789 }
9790 
9791 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
9792 			     struct net_device *dev,
9793 			     const unsigned char *addr, u16 vid,
9794 			     u16 flags,
9795 			     struct netlink_ext_ack *extack)
9796 {
9797 	/* guarantee we can provide a unique filter for the unicast address */
9798 	if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
9799 		struct ixgbe_adapter *adapter = netdev_priv(dev);
9800 		u16 pool = VMDQ_P(0);
9801 
9802 		if (netdev_uc_count(dev) >= ixgbe_available_rars(adapter, pool))
9803 			return -ENOMEM;
9804 	}
9805 
9806 	return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
9807 }
9808 
9809 /**
9810  * ixgbe_configure_bridge_mode - set various bridge modes
9811  * @adapter: the private structure
9812  * @mode: requested bridge mode
9813  *
9814  * Configure some settings require for various bridge modes.
9815  **/
9816 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
9817 				       __u16 mode)
9818 {
9819 	struct ixgbe_hw *hw = &adapter->hw;
9820 	unsigned int p, num_pools;
9821 	u32 vmdctl;
9822 
9823 	switch (mode) {
9824 	case BRIDGE_MODE_VEPA:
9825 		/* disable Tx loopback, rely on switch hairpin mode */
9826 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
9827 
9828 		/* must enable Rx switching replication to allow multicast
9829 		 * packet reception on all VFs, and to enable source address
9830 		 * pruning.
9831 		 */
9832 		vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9833 		vmdctl |= IXGBE_VT_CTL_REPLEN;
9834 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9835 
9836 		/* enable Rx source address pruning. Note, this requires
9837 		 * replication to be enabled or else it does nothing.
9838 		 */
9839 		num_pools = adapter->num_vfs + adapter->num_rx_pools;
9840 		for (p = 0; p < num_pools; p++) {
9841 			if (hw->mac.ops.set_source_address_pruning)
9842 				hw->mac.ops.set_source_address_pruning(hw,
9843 								       true,
9844 								       p);
9845 		}
9846 		break;
9847 	case BRIDGE_MODE_VEB:
9848 		/* enable Tx loopback for internal VF/PF communication */
9849 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
9850 				IXGBE_PFDTXGSWC_VT_LBEN);
9851 
9852 		/* disable Rx switching replication unless we have SR-IOV
9853 		 * virtual functions
9854 		 */
9855 		vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9856 		if (!adapter->num_vfs)
9857 			vmdctl &= ~IXGBE_VT_CTL_REPLEN;
9858 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9859 
9860 		/* disable Rx source address pruning, since we don't expect to
9861 		 * be receiving external loopback of our transmitted frames.
9862 		 */
9863 		num_pools = adapter->num_vfs + adapter->num_rx_pools;
9864 		for (p = 0; p < num_pools; p++) {
9865 			if (hw->mac.ops.set_source_address_pruning)
9866 				hw->mac.ops.set_source_address_pruning(hw,
9867 								       false,
9868 								       p);
9869 		}
9870 		break;
9871 	default:
9872 		return -EINVAL;
9873 	}
9874 
9875 	adapter->bridge_mode = mode;
9876 
9877 	e_info(drv, "enabling bridge mode: %s\n",
9878 	       mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
9879 
9880 	return 0;
9881 }
9882 
9883 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
9884 				    struct nlmsghdr *nlh, u16 flags,
9885 				    struct netlink_ext_ack *extack)
9886 {
9887 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9888 	struct nlattr *attr, *br_spec;
9889 	int rem;
9890 
9891 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
9892 		return -EOPNOTSUPP;
9893 
9894 	br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
9895 	if (!br_spec)
9896 		return -EINVAL;
9897 
9898 	nla_for_each_nested(attr, br_spec, rem) {
9899 		int status;
9900 		__u16 mode;
9901 
9902 		if (nla_type(attr) != IFLA_BRIDGE_MODE)
9903 			continue;
9904 
9905 		if (nla_len(attr) < sizeof(mode))
9906 			return -EINVAL;
9907 
9908 		mode = nla_get_u16(attr);
9909 		status = ixgbe_configure_bridge_mode(adapter, mode);
9910 		if (status)
9911 			return status;
9912 
9913 		break;
9914 	}
9915 
9916 	return 0;
9917 }
9918 
9919 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
9920 				    struct net_device *dev,
9921 				    u32 filter_mask, int nlflags)
9922 {
9923 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9924 
9925 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
9926 		return 0;
9927 
9928 	return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
9929 				       adapter->bridge_mode, 0, 0, nlflags,
9930 				       filter_mask, NULL);
9931 }
9932 
9933 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
9934 {
9935 	struct ixgbe_adapter *adapter = netdev_priv(pdev);
9936 	struct ixgbe_fwd_adapter *accel;
9937 	int tcs = adapter->hw_tcs ? : 1;
9938 	int pool, err;
9939 
9940 	if (adapter->xdp_prog) {
9941 		e_warn(probe, "L2FW offload is not supported with XDP\n");
9942 		return ERR_PTR(-EINVAL);
9943 	}
9944 
9945 	/* The hardware supported by ixgbe only filters on the destination MAC
9946 	 * address. In order to avoid issues we only support offloading modes
9947 	 * where the hardware can actually provide the functionality.
9948 	 */
9949 	if (!macvlan_supports_dest_filter(vdev))
9950 		return ERR_PTR(-EMEDIUMTYPE);
9951 
9952 	/* We need to lock down the macvlan to be a single queue device so that
9953 	 * we can reuse the tc_to_txq field in the macvlan netdev to represent
9954 	 * the queue mapping to our netdev.
9955 	 */
9956 	if (netif_is_multiqueue(vdev))
9957 		return ERR_PTR(-ERANGE);
9958 
9959 	pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
9960 	if (pool == adapter->num_rx_pools) {
9961 		u16 used_pools = adapter->num_vfs + adapter->num_rx_pools;
9962 		u16 reserved_pools;
9963 
9964 		if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
9965 		     adapter->num_rx_pools >= (MAX_TX_QUEUES / tcs)) ||
9966 		    adapter->num_rx_pools > IXGBE_MAX_MACVLANS)
9967 			return ERR_PTR(-EBUSY);
9968 
9969 		/* Hardware has a limited number of available pools. Each VF,
9970 		 * and the PF require a pool. Check to ensure we don't
9971 		 * attempt to use more then the available number of pools.
9972 		 */
9973 		if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
9974 			return ERR_PTR(-EBUSY);
9975 
9976 		/* Enable VMDq flag so device will be set in VM mode */
9977 		adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED |
9978 				  IXGBE_FLAG_SRIOV_ENABLED;
9979 
9980 		/* Try to reserve as many queues per pool as possible,
9981 		 * we start with the configurations that support 4 queues
9982 		 * per pools, followed by 2, and then by just 1 per pool.
9983 		 */
9984 		if (used_pools < 32 && adapter->num_rx_pools < 16)
9985 			reserved_pools = min_t(u16,
9986 					       32 - used_pools,
9987 					       16 - adapter->num_rx_pools);
9988 		else if (adapter->num_rx_pools < 32)
9989 			reserved_pools = min_t(u16,
9990 					       64 - used_pools,
9991 					       32 - adapter->num_rx_pools);
9992 		else
9993 			reserved_pools = 64 - used_pools;
9994 
9995 
9996 		if (!reserved_pools)
9997 			return ERR_PTR(-EBUSY);
9998 
9999 		adapter->ring_feature[RING_F_VMDQ].limit += reserved_pools;
10000 
10001 		/* Force reinit of ring allocation with VMDQ enabled */
10002 		err = ixgbe_setup_tc(pdev, adapter->hw_tcs);
10003 		if (err)
10004 			return ERR_PTR(err);
10005 
10006 		if (pool >= adapter->num_rx_pools)
10007 			return ERR_PTR(-ENOMEM);
10008 	}
10009 
10010 	accel = kzalloc(sizeof(*accel), GFP_KERNEL);
10011 	if (!accel)
10012 		return ERR_PTR(-ENOMEM);
10013 
10014 	set_bit(pool, adapter->fwd_bitmask);
10015 	netdev_set_sb_channel(vdev, pool);
10016 	accel->pool = pool;
10017 	accel->netdev = vdev;
10018 
10019 	if (!netif_running(pdev))
10020 		return accel;
10021 
10022 	err = ixgbe_fwd_ring_up(adapter, accel);
10023 	if (err)
10024 		return ERR_PTR(err);
10025 
10026 	return accel;
10027 }
10028 
10029 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
10030 {
10031 	struct ixgbe_fwd_adapter *accel = priv;
10032 	struct ixgbe_adapter *adapter = netdev_priv(pdev);
10033 	unsigned int rxbase = accel->rx_base_queue;
10034 	unsigned int i;
10035 
10036 	/* delete unicast filter associated with offloaded interface */
10037 	ixgbe_del_mac_filter(adapter, accel->netdev->dev_addr,
10038 			     VMDQ_P(accel->pool));
10039 
10040 	/* Allow remaining Rx packets to get flushed out of the
10041 	 * Rx FIFO before we drop the netdev for the ring.
10042 	 */
10043 	usleep_range(10000, 20000);
10044 
10045 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
10046 		struct ixgbe_ring *ring = adapter->rx_ring[rxbase + i];
10047 		struct ixgbe_q_vector *qv = ring->q_vector;
10048 
10049 		/* Make sure we aren't processing any packets and clear
10050 		 * netdev to shut down the ring.
10051 		 */
10052 		if (netif_running(adapter->netdev))
10053 			napi_synchronize(&qv->napi);
10054 		ring->netdev = NULL;
10055 	}
10056 
10057 	/* unbind the queues and drop the subordinate channel config */
10058 	netdev_unbind_sb_channel(pdev, accel->netdev);
10059 	netdev_set_sb_channel(accel->netdev, 0);
10060 
10061 	clear_bit(accel->pool, adapter->fwd_bitmask);
10062 	kfree(accel);
10063 }
10064 
10065 #define IXGBE_MAX_MAC_HDR_LEN		127
10066 #define IXGBE_MAX_NETWORK_HDR_LEN	511
10067 
10068 static netdev_features_t
10069 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
10070 		     netdev_features_t features)
10071 {
10072 	unsigned int network_hdr_len, mac_hdr_len;
10073 
10074 	/* Make certain the headers can be described by a context descriptor */
10075 	mac_hdr_len = skb_network_header(skb) - skb->data;
10076 	if (unlikely(mac_hdr_len > IXGBE_MAX_MAC_HDR_LEN))
10077 		return features & ~(NETIF_F_HW_CSUM |
10078 				    NETIF_F_SCTP_CRC |
10079 				    NETIF_F_GSO_UDP_L4 |
10080 				    NETIF_F_HW_VLAN_CTAG_TX |
10081 				    NETIF_F_TSO |
10082 				    NETIF_F_TSO6);
10083 
10084 	network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
10085 	if (unlikely(network_hdr_len >  IXGBE_MAX_NETWORK_HDR_LEN))
10086 		return features & ~(NETIF_F_HW_CSUM |
10087 				    NETIF_F_SCTP_CRC |
10088 				    NETIF_F_GSO_UDP_L4 |
10089 				    NETIF_F_TSO |
10090 				    NETIF_F_TSO6);
10091 
10092 	/* We can only support IPV4 TSO in tunnels if we can mangle the
10093 	 * inner IP ID field, so strip TSO if MANGLEID is not supported.
10094 	 * IPsec offoad sets skb->encapsulation but still can handle
10095 	 * the TSO, so it's the exception.
10096 	 */
10097 	if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) {
10098 #ifdef CONFIG_IXGBE_IPSEC
10099 		if (!secpath_exists(skb))
10100 #endif
10101 			features &= ~NETIF_F_TSO;
10102 	}
10103 
10104 	return features;
10105 }
10106 
10107 static int ixgbe_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
10108 {
10109 	int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
10110 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10111 	struct bpf_prog *old_prog;
10112 	bool need_reset;
10113 
10114 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
10115 		return -EINVAL;
10116 
10117 	if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
10118 		return -EINVAL;
10119 
10120 	/* verify ixgbe ring attributes are sufficient for XDP */
10121 	for (i = 0; i < adapter->num_rx_queues; i++) {
10122 		struct ixgbe_ring *ring = adapter->rx_ring[i];
10123 
10124 		if (ring_is_rsc_enabled(ring))
10125 			return -EINVAL;
10126 
10127 		if (frame_size > ixgbe_rx_bufsz(ring))
10128 			return -EINVAL;
10129 	}
10130 
10131 	if (nr_cpu_ids > MAX_XDP_QUEUES)
10132 		return -ENOMEM;
10133 
10134 	old_prog = xchg(&adapter->xdp_prog, prog);
10135 	need_reset = (!!prog != !!old_prog);
10136 
10137 	/* If transitioning XDP modes reconfigure rings */
10138 	if (need_reset) {
10139 		int err;
10140 
10141 		if (!prog)
10142 			/* Wait until ndo_xsk_wakeup completes. */
10143 			synchronize_rcu();
10144 		err = ixgbe_setup_tc(dev, adapter->hw_tcs);
10145 
10146 		if (err) {
10147 			rcu_assign_pointer(adapter->xdp_prog, old_prog);
10148 			return -EINVAL;
10149 		}
10150 	} else {
10151 		for (i = 0; i < adapter->num_rx_queues; i++)
10152 			(void)xchg(&adapter->rx_ring[i]->xdp_prog,
10153 			    adapter->xdp_prog);
10154 	}
10155 
10156 	if (old_prog)
10157 		bpf_prog_put(old_prog);
10158 
10159 	/* Kick start the NAPI context if there is an AF_XDP socket open
10160 	 * on that queue id. This so that receiving will start.
10161 	 */
10162 	if (need_reset && prog)
10163 		for (i = 0; i < adapter->num_rx_queues; i++)
10164 			if (adapter->xdp_ring[i]->xsk_umem)
10165 				(void)ixgbe_xsk_wakeup(adapter->netdev, i,
10166 						       XDP_WAKEUP_RX);
10167 
10168 	return 0;
10169 }
10170 
10171 static int ixgbe_xdp(struct net_device *dev, struct netdev_bpf *xdp)
10172 {
10173 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10174 
10175 	switch (xdp->command) {
10176 	case XDP_SETUP_PROG:
10177 		return ixgbe_xdp_setup(dev, xdp->prog);
10178 	case XDP_SETUP_XSK_UMEM:
10179 		return ixgbe_xsk_umem_setup(adapter, xdp->xsk.umem,
10180 					    xdp->xsk.queue_id);
10181 
10182 	default:
10183 		return -EINVAL;
10184 	}
10185 }
10186 
10187 void ixgbe_xdp_ring_update_tail(struct ixgbe_ring *ring)
10188 {
10189 	/* Force memory writes to complete before letting h/w know there
10190 	 * are new descriptors to fetch.
10191 	 */
10192 	wmb();
10193 	writel(ring->next_to_use, ring->tail);
10194 }
10195 
10196 static int ixgbe_xdp_xmit(struct net_device *dev, int n,
10197 			  struct xdp_frame **frames, u32 flags)
10198 {
10199 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10200 	struct ixgbe_ring *ring;
10201 	int drops = 0;
10202 	int i;
10203 
10204 	if (unlikely(test_bit(__IXGBE_DOWN, &adapter->state)))
10205 		return -ENETDOWN;
10206 
10207 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
10208 		return -EINVAL;
10209 
10210 	/* During program transitions its possible adapter->xdp_prog is assigned
10211 	 * but ring has not been configured yet. In this case simply abort xmit.
10212 	 */
10213 	ring = adapter->xdp_prog ? adapter->xdp_ring[smp_processor_id()] : NULL;
10214 	if (unlikely(!ring))
10215 		return -ENXIO;
10216 
10217 	if (unlikely(test_bit(__IXGBE_TX_DISABLED, &ring->state)))
10218 		return -ENXIO;
10219 
10220 	for (i = 0; i < n; i++) {
10221 		struct xdp_frame *xdpf = frames[i];
10222 		int err;
10223 
10224 		err = ixgbe_xmit_xdp_ring(adapter, xdpf);
10225 		if (err != IXGBE_XDP_TX) {
10226 			xdp_return_frame_rx_napi(xdpf);
10227 			drops++;
10228 		}
10229 	}
10230 
10231 	if (unlikely(flags & XDP_XMIT_FLUSH))
10232 		ixgbe_xdp_ring_update_tail(ring);
10233 
10234 	return n - drops;
10235 }
10236 
10237 static const struct net_device_ops ixgbe_netdev_ops = {
10238 	.ndo_open		= ixgbe_open,
10239 	.ndo_stop		= ixgbe_close,
10240 	.ndo_start_xmit		= ixgbe_xmit_frame,
10241 	.ndo_set_rx_mode	= ixgbe_set_rx_mode,
10242 	.ndo_validate_addr	= eth_validate_addr,
10243 	.ndo_set_mac_address	= ixgbe_set_mac,
10244 	.ndo_change_mtu		= ixgbe_change_mtu,
10245 	.ndo_tx_timeout		= ixgbe_tx_timeout,
10246 	.ndo_set_tx_maxrate	= ixgbe_tx_maxrate,
10247 	.ndo_vlan_rx_add_vid	= ixgbe_vlan_rx_add_vid,
10248 	.ndo_vlan_rx_kill_vid	= ixgbe_vlan_rx_kill_vid,
10249 	.ndo_do_ioctl		= ixgbe_ioctl,
10250 	.ndo_set_vf_mac		= ixgbe_ndo_set_vf_mac,
10251 	.ndo_set_vf_vlan	= ixgbe_ndo_set_vf_vlan,
10252 	.ndo_set_vf_rate	= ixgbe_ndo_set_vf_bw,
10253 	.ndo_set_vf_spoofchk	= ixgbe_ndo_set_vf_spoofchk,
10254 	.ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
10255 	.ndo_set_vf_trust	= ixgbe_ndo_set_vf_trust,
10256 	.ndo_get_vf_config	= ixgbe_ndo_get_vf_config,
10257 	.ndo_get_stats64	= ixgbe_get_stats64,
10258 	.ndo_setup_tc		= __ixgbe_setup_tc,
10259 #ifdef IXGBE_FCOE
10260 	.ndo_select_queue	= ixgbe_select_queue,
10261 	.ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
10262 	.ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
10263 	.ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
10264 	.ndo_fcoe_enable = ixgbe_fcoe_enable,
10265 	.ndo_fcoe_disable = ixgbe_fcoe_disable,
10266 	.ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
10267 	.ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
10268 #endif /* IXGBE_FCOE */
10269 	.ndo_set_features = ixgbe_set_features,
10270 	.ndo_fix_features = ixgbe_fix_features,
10271 	.ndo_fdb_add		= ixgbe_ndo_fdb_add,
10272 	.ndo_bridge_setlink	= ixgbe_ndo_bridge_setlink,
10273 	.ndo_bridge_getlink	= ixgbe_ndo_bridge_getlink,
10274 	.ndo_dfwd_add_station	= ixgbe_fwd_add,
10275 	.ndo_dfwd_del_station	= ixgbe_fwd_del,
10276 	.ndo_udp_tunnel_add	= udp_tunnel_nic_add_port,
10277 	.ndo_udp_tunnel_del	= udp_tunnel_nic_del_port,
10278 	.ndo_features_check	= ixgbe_features_check,
10279 	.ndo_bpf		= ixgbe_xdp,
10280 	.ndo_xdp_xmit		= ixgbe_xdp_xmit,
10281 	.ndo_xsk_wakeup         = ixgbe_xsk_wakeup,
10282 };
10283 
10284 static void ixgbe_disable_txr_hw(struct ixgbe_adapter *adapter,
10285 				 struct ixgbe_ring *tx_ring)
10286 {
10287 	unsigned long wait_delay, delay_interval;
10288 	struct ixgbe_hw *hw = &adapter->hw;
10289 	u8 reg_idx = tx_ring->reg_idx;
10290 	int wait_loop;
10291 	u32 txdctl;
10292 
10293 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
10294 
10295 	/* delay mechanism from ixgbe_disable_tx */
10296 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10297 
10298 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
10299 	wait_delay = delay_interval;
10300 
10301 	while (wait_loop--) {
10302 		usleep_range(wait_delay, wait_delay + 10);
10303 		wait_delay += delay_interval * 2;
10304 		txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
10305 
10306 		if (!(txdctl & IXGBE_TXDCTL_ENABLE))
10307 			return;
10308 	}
10309 
10310 	e_err(drv, "TXDCTL.ENABLE not cleared within the polling period\n");
10311 }
10312 
10313 static void ixgbe_disable_txr(struct ixgbe_adapter *adapter,
10314 			      struct ixgbe_ring *tx_ring)
10315 {
10316 	set_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10317 	ixgbe_disable_txr_hw(adapter, tx_ring);
10318 }
10319 
10320 static void ixgbe_disable_rxr_hw(struct ixgbe_adapter *adapter,
10321 				 struct ixgbe_ring *rx_ring)
10322 {
10323 	unsigned long wait_delay, delay_interval;
10324 	struct ixgbe_hw *hw = &adapter->hw;
10325 	u8 reg_idx = rx_ring->reg_idx;
10326 	int wait_loop;
10327 	u32 rxdctl;
10328 
10329 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10330 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
10331 	rxdctl |= IXGBE_RXDCTL_SWFLSH;
10332 
10333 	/* write value back with RXDCTL.ENABLE bit cleared */
10334 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
10335 
10336 	/* RXDCTL.EN may not change on 82598 if link is down, so skip it */
10337 	if (hw->mac.type == ixgbe_mac_82598EB &&
10338 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
10339 		return;
10340 
10341 	/* delay mechanism from ixgbe_disable_rx */
10342 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10343 
10344 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
10345 	wait_delay = delay_interval;
10346 
10347 	while (wait_loop--) {
10348 		usleep_range(wait_delay, wait_delay + 10);
10349 		wait_delay += delay_interval * 2;
10350 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10351 
10352 		if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
10353 			return;
10354 	}
10355 
10356 	e_err(drv, "RXDCTL.ENABLE not cleared within the polling period\n");
10357 }
10358 
10359 static void ixgbe_reset_txr_stats(struct ixgbe_ring *tx_ring)
10360 {
10361 	memset(&tx_ring->stats, 0, sizeof(tx_ring->stats));
10362 	memset(&tx_ring->tx_stats, 0, sizeof(tx_ring->tx_stats));
10363 }
10364 
10365 static void ixgbe_reset_rxr_stats(struct ixgbe_ring *rx_ring)
10366 {
10367 	memset(&rx_ring->stats, 0, sizeof(rx_ring->stats));
10368 	memset(&rx_ring->rx_stats, 0, sizeof(rx_ring->rx_stats));
10369 }
10370 
10371 /**
10372  * ixgbe_txrx_ring_disable - Disable Rx/Tx/XDP Tx rings
10373  * @adapter: adapter structure
10374  * @ring: ring index
10375  *
10376  * This function disables a certain Rx/Tx/XDP Tx ring. The function
10377  * assumes that the netdev is running.
10378  **/
10379 void ixgbe_txrx_ring_disable(struct ixgbe_adapter *adapter, int ring)
10380 {
10381 	struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10382 
10383 	rx_ring = adapter->rx_ring[ring];
10384 	tx_ring = adapter->tx_ring[ring];
10385 	xdp_ring = adapter->xdp_ring[ring];
10386 
10387 	ixgbe_disable_txr(adapter, tx_ring);
10388 	if (xdp_ring)
10389 		ixgbe_disable_txr(adapter, xdp_ring);
10390 	ixgbe_disable_rxr_hw(adapter, rx_ring);
10391 
10392 	if (xdp_ring)
10393 		synchronize_rcu();
10394 
10395 	/* Rx/Tx/XDP Tx share the same napi context. */
10396 	napi_disable(&rx_ring->q_vector->napi);
10397 
10398 	ixgbe_clean_tx_ring(tx_ring);
10399 	if (xdp_ring)
10400 		ixgbe_clean_tx_ring(xdp_ring);
10401 	ixgbe_clean_rx_ring(rx_ring);
10402 
10403 	ixgbe_reset_txr_stats(tx_ring);
10404 	if (xdp_ring)
10405 		ixgbe_reset_txr_stats(xdp_ring);
10406 	ixgbe_reset_rxr_stats(rx_ring);
10407 }
10408 
10409 /**
10410  * ixgbe_txrx_ring_enable - Enable Rx/Tx/XDP Tx rings
10411  * @adapter: adapter structure
10412  * @ring: ring index
10413  *
10414  * This function enables a certain Rx/Tx/XDP Tx ring. The function
10415  * assumes that the netdev is running.
10416  **/
10417 void ixgbe_txrx_ring_enable(struct ixgbe_adapter *adapter, int ring)
10418 {
10419 	struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10420 
10421 	rx_ring = adapter->rx_ring[ring];
10422 	tx_ring = adapter->tx_ring[ring];
10423 	xdp_ring = adapter->xdp_ring[ring];
10424 
10425 	/* Rx/Tx/XDP Tx share the same napi context. */
10426 	napi_enable(&rx_ring->q_vector->napi);
10427 
10428 	ixgbe_configure_tx_ring(adapter, tx_ring);
10429 	if (xdp_ring)
10430 		ixgbe_configure_tx_ring(adapter, xdp_ring);
10431 	ixgbe_configure_rx_ring(adapter, rx_ring);
10432 
10433 	clear_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10434 	if (xdp_ring)
10435 		clear_bit(__IXGBE_TX_DISABLED, &xdp_ring->state);
10436 }
10437 
10438 /**
10439  * ixgbe_enumerate_functions - Get the number of ports this device has
10440  * @adapter: adapter structure
10441  *
10442  * This function enumerates the phsyical functions co-located on a single slot,
10443  * in order to determine how many ports a device has. This is most useful in
10444  * determining the required GT/s of PCIe bandwidth necessary for optimal
10445  * performance.
10446  **/
10447 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
10448 {
10449 	struct pci_dev *entry, *pdev = adapter->pdev;
10450 	int physfns = 0;
10451 
10452 	/* Some cards can not use the generic count PCIe functions method,
10453 	 * because they are behind a parent switch, so we hardcode these with
10454 	 * the correct number of functions.
10455 	 */
10456 	if (ixgbe_pcie_from_parent(&adapter->hw))
10457 		physfns = 4;
10458 
10459 	list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
10460 		/* don't count virtual functions */
10461 		if (entry->is_virtfn)
10462 			continue;
10463 
10464 		/* When the devices on the bus don't all match our device ID,
10465 		 * we can't reliably determine the correct number of
10466 		 * functions. This can occur if a function has been direct
10467 		 * attached to a virtual machine using VT-d, for example. In
10468 		 * this case, simply return -1 to indicate this.
10469 		 */
10470 		if ((entry->vendor != pdev->vendor) ||
10471 		    (entry->device != pdev->device))
10472 			return -1;
10473 
10474 		physfns++;
10475 	}
10476 
10477 	return physfns;
10478 }
10479 
10480 /**
10481  * ixgbe_wol_supported - Check whether device supports WoL
10482  * @adapter: the adapter private structure
10483  * @device_id: the device ID
10484  * @subdevice_id: the subsystem device ID
10485  *
10486  * This function is used by probe and ethtool to determine
10487  * which devices have WoL support
10488  *
10489  **/
10490 bool ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
10491 			 u16 subdevice_id)
10492 {
10493 	struct ixgbe_hw *hw = &adapter->hw;
10494 	u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
10495 
10496 	/* WOL not supported on 82598 */
10497 	if (hw->mac.type == ixgbe_mac_82598EB)
10498 		return false;
10499 
10500 	/* check eeprom to see if WOL is enabled for X540 and newer */
10501 	if (hw->mac.type >= ixgbe_mac_X540) {
10502 		if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
10503 		    ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
10504 		     (hw->bus.func == 0)))
10505 			return true;
10506 	}
10507 
10508 	/* WOL is determined based on device IDs for 82599 MACs */
10509 	switch (device_id) {
10510 	case IXGBE_DEV_ID_82599_SFP:
10511 		/* Only these subdevices could supports WOL */
10512 		switch (subdevice_id) {
10513 		case IXGBE_SUBDEV_ID_82599_560FLR:
10514 		case IXGBE_SUBDEV_ID_82599_LOM_SNAP6:
10515 		case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
10516 		case IXGBE_SUBDEV_ID_82599_SFP_2OCP:
10517 			/* only support first port */
10518 			if (hw->bus.func != 0)
10519 				break;
10520 			fallthrough;
10521 		case IXGBE_SUBDEV_ID_82599_SP_560FLR:
10522 		case IXGBE_SUBDEV_ID_82599_SFP:
10523 		case IXGBE_SUBDEV_ID_82599_RNDC:
10524 		case IXGBE_SUBDEV_ID_82599_ECNA_DP:
10525 		case IXGBE_SUBDEV_ID_82599_SFP_1OCP:
10526 		case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM1:
10527 		case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM2:
10528 			return true;
10529 		}
10530 		break;
10531 	case IXGBE_DEV_ID_82599EN_SFP:
10532 		/* Only these subdevices support WOL */
10533 		switch (subdevice_id) {
10534 		case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
10535 			return true;
10536 		}
10537 		break;
10538 	case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
10539 		/* All except this subdevice support WOL */
10540 		if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
10541 			return true;
10542 		break;
10543 	case IXGBE_DEV_ID_82599_KX4:
10544 		return  true;
10545 	default:
10546 		break;
10547 	}
10548 
10549 	return false;
10550 }
10551 
10552 /**
10553  * ixgbe_set_fw_version - Set FW version
10554  * @adapter: the adapter private structure
10555  *
10556  * This function is used by probe and ethtool to determine the FW version to
10557  * format to display. The FW version is taken from the EEPROM/NVM.
10558  */
10559 static void ixgbe_set_fw_version(struct ixgbe_adapter *adapter)
10560 {
10561 	struct ixgbe_hw *hw = &adapter->hw;
10562 	struct ixgbe_nvm_version nvm_ver;
10563 
10564 	ixgbe_get_oem_prod_version(hw, &nvm_ver);
10565 	if (nvm_ver.oem_valid) {
10566 		snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10567 			 "%x.%x.%x", nvm_ver.oem_major, nvm_ver.oem_minor,
10568 			 nvm_ver.oem_release);
10569 		return;
10570 	}
10571 
10572 	ixgbe_get_etk_id(hw, &nvm_ver);
10573 	ixgbe_get_orom_version(hw, &nvm_ver);
10574 
10575 	if (nvm_ver.or_valid) {
10576 		snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10577 			 "0x%08x, %d.%d.%d", nvm_ver.etk_id, nvm_ver.or_major,
10578 			 nvm_ver.or_build, nvm_ver.or_patch);
10579 		return;
10580 	}
10581 
10582 	/* Set ETrack ID format */
10583 	snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10584 		 "0x%08x", nvm_ver.etk_id);
10585 }
10586 
10587 /**
10588  * ixgbe_probe - Device Initialization Routine
10589  * @pdev: PCI device information struct
10590  * @ent: entry in ixgbe_pci_tbl
10591  *
10592  * Returns 0 on success, negative on failure
10593  *
10594  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
10595  * The OS initialization, configuring of the adapter private structure,
10596  * and a hardware reset occur.
10597  **/
10598 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
10599 {
10600 	struct net_device *netdev;
10601 	struct ixgbe_adapter *adapter = NULL;
10602 	struct ixgbe_hw *hw;
10603 	const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
10604 	int i, err, pci_using_dac, expected_gts;
10605 	unsigned int indices = MAX_TX_QUEUES;
10606 	u8 part_str[IXGBE_PBANUM_LENGTH];
10607 	bool disable_dev = false;
10608 #ifdef IXGBE_FCOE
10609 	u16 device_caps;
10610 #endif
10611 	u32 eec;
10612 
10613 	/* Catch broken hardware that put the wrong VF device ID in
10614 	 * the PCIe SR-IOV capability.
10615 	 */
10616 	if (pdev->is_virtfn) {
10617 		WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
10618 		     pci_name(pdev), pdev->vendor, pdev->device);
10619 		return -EINVAL;
10620 	}
10621 
10622 	err = pci_enable_device_mem(pdev);
10623 	if (err)
10624 		return err;
10625 
10626 	if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
10627 		pci_using_dac = 1;
10628 	} else {
10629 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
10630 		if (err) {
10631 			dev_err(&pdev->dev,
10632 				"No usable DMA configuration, aborting\n");
10633 			goto err_dma;
10634 		}
10635 		pci_using_dac = 0;
10636 	}
10637 
10638 	err = pci_request_mem_regions(pdev, ixgbe_driver_name);
10639 	if (err) {
10640 		dev_err(&pdev->dev,
10641 			"pci_request_selected_regions failed 0x%x\n", err);
10642 		goto err_pci_reg;
10643 	}
10644 
10645 	pci_enable_pcie_error_reporting(pdev);
10646 
10647 	pci_set_master(pdev);
10648 	pci_save_state(pdev);
10649 
10650 	if (ii->mac == ixgbe_mac_82598EB) {
10651 #ifdef CONFIG_IXGBE_DCB
10652 		/* 8 TC w/ 4 queues per TC */
10653 		indices = 4 * MAX_TRAFFIC_CLASS;
10654 #else
10655 		indices = IXGBE_MAX_RSS_INDICES;
10656 #endif
10657 	}
10658 
10659 	netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
10660 	if (!netdev) {
10661 		err = -ENOMEM;
10662 		goto err_alloc_etherdev;
10663 	}
10664 
10665 	SET_NETDEV_DEV(netdev, &pdev->dev);
10666 
10667 	adapter = netdev_priv(netdev);
10668 
10669 	adapter->netdev = netdev;
10670 	adapter->pdev = pdev;
10671 	hw = &adapter->hw;
10672 	hw->back = adapter;
10673 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
10674 
10675 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
10676 			      pci_resource_len(pdev, 0));
10677 	adapter->io_addr = hw->hw_addr;
10678 	if (!hw->hw_addr) {
10679 		err = -EIO;
10680 		goto err_ioremap;
10681 	}
10682 
10683 	netdev->netdev_ops = &ixgbe_netdev_ops;
10684 	ixgbe_set_ethtool_ops(netdev);
10685 	netdev->watchdog_timeo = 5 * HZ;
10686 	strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
10687 
10688 	/* Setup hw api */
10689 	hw->mac.ops   = *ii->mac_ops;
10690 	hw->mac.type  = ii->mac;
10691 	hw->mvals     = ii->mvals;
10692 	if (ii->link_ops)
10693 		hw->link.ops  = *ii->link_ops;
10694 
10695 	/* EEPROM */
10696 	hw->eeprom.ops = *ii->eeprom_ops;
10697 	eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
10698 	if (ixgbe_removed(hw->hw_addr)) {
10699 		err = -EIO;
10700 		goto err_ioremap;
10701 	}
10702 	/* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
10703 	if (!(eec & BIT(8)))
10704 		hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
10705 
10706 	/* PHY */
10707 	hw->phy.ops = *ii->phy_ops;
10708 	hw->phy.sfp_type = ixgbe_sfp_type_unknown;
10709 	/* ixgbe_identify_phy_generic will set prtad and mmds properly */
10710 	hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
10711 	hw->phy.mdio.mmds = 0;
10712 	hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
10713 	hw->phy.mdio.dev = netdev;
10714 	hw->phy.mdio.mdio_read = ixgbe_mdio_read;
10715 	hw->phy.mdio.mdio_write = ixgbe_mdio_write;
10716 
10717 	/* setup the private structure */
10718 	err = ixgbe_sw_init(adapter, ii);
10719 	if (err)
10720 		goto err_sw_init;
10721 
10722 	switch (adapter->hw.mac.type) {
10723 	case ixgbe_mac_X550:
10724 	case ixgbe_mac_X550EM_x:
10725 		netdev->udp_tunnel_nic_info = &ixgbe_udp_tunnels_x550;
10726 		break;
10727 	case ixgbe_mac_x550em_a:
10728 		netdev->udp_tunnel_nic_info = &ixgbe_udp_tunnels_x550em_a;
10729 		break;
10730 	default:
10731 		break;
10732 	}
10733 
10734 	/* Make sure the SWFW semaphore is in a valid state */
10735 	if (hw->mac.ops.init_swfw_sync)
10736 		hw->mac.ops.init_swfw_sync(hw);
10737 
10738 	/* Make it possible the adapter to be woken up via WOL */
10739 	switch (adapter->hw.mac.type) {
10740 	case ixgbe_mac_82599EB:
10741 	case ixgbe_mac_X540:
10742 	case ixgbe_mac_X550:
10743 	case ixgbe_mac_X550EM_x:
10744 	case ixgbe_mac_x550em_a:
10745 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
10746 		break;
10747 	default:
10748 		break;
10749 	}
10750 
10751 	/*
10752 	 * If there is a fan on this device and it has failed log the
10753 	 * failure.
10754 	 */
10755 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
10756 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
10757 		if (esdp & IXGBE_ESDP_SDP1)
10758 			e_crit(probe, "Fan has stopped, replace the adapter\n");
10759 	}
10760 
10761 	if (allow_unsupported_sfp)
10762 		hw->allow_unsupported_sfp = allow_unsupported_sfp;
10763 
10764 	/* reset_hw fills in the perm_addr as well */
10765 	hw->phy.reset_if_overtemp = true;
10766 	err = hw->mac.ops.reset_hw(hw);
10767 	hw->phy.reset_if_overtemp = false;
10768 	ixgbe_set_eee_capable(adapter);
10769 	if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
10770 		err = 0;
10771 	} else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
10772 		e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
10773 		e_dev_err("Reload the driver after installing a supported module.\n");
10774 		goto err_sw_init;
10775 	} else if (err) {
10776 		e_dev_err("HW Init failed: %d\n", err);
10777 		goto err_sw_init;
10778 	}
10779 
10780 #ifdef CONFIG_PCI_IOV
10781 	/* SR-IOV not supported on the 82598 */
10782 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
10783 		goto skip_sriov;
10784 	/* Mailbox */
10785 	ixgbe_init_mbx_params_pf(hw);
10786 	hw->mbx.ops = ii->mbx_ops;
10787 	pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
10788 	ixgbe_enable_sriov(adapter, max_vfs);
10789 skip_sriov:
10790 
10791 #endif
10792 	netdev->features = NETIF_F_SG |
10793 			   NETIF_F_TSO |
10794 			   NETIF_F_TSO6 |
10795 			   NETIF_F_RXHASH |
10796 			   NETIF_F_RXCSUM |
10797 			   NETIF_F_HW_CSUM;
10798 
10799 #define IXGBE_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
10800 				    NETIF_F_GSO_GRE_CSUM | \
10801 				    NETIF_F_GSO_IPXIP4 | \
10802 				    NETIF_F_GSO_IPXIP6 | \
10803 				    NETIF_F_GSO_UDP_TUNNEL | \
10804 				    NETIF_F_GSO_UDP_TUNNEL_CSUM)
10805 
10806 	netdev->gso_partial_features = IXGBE_GSO_PARTIAL_FEATURES;
10807 	netdev->features |= NETIF_F_GSO_PARTIAL |
10808 			    IXGBE_GSO_PARTIAL_FEATURES;
10809 
10810 	if (hw->mac.type >= ixgbe_mac_82599EB)
10811 		netdev->features |= NETIF_F_SCTP_CRC | NETIF_F_GSO_UDP_L4;
10812 
10813 #ifdef CONFIG_IXGBE_IPSEC
10814 #define IXGBE_ESP_FEATURES	(NETIF_F_HW_ESP | \
10815 				 NETIF_F_HW_ESP_TX_CSUM | \
10816 				 NETIF_F_GSO_ESP)
10817 
10818 	if (adapter->ipsec)
10819 		netdev->features |= IXGBE_ESP_FEATURES;
10820 #endif
10821 	/* copy netdev features into list of user selectable features */
10822 	netdev->hw_features |= netdev->features |
10823 			       NETIF_F_HW_VLAN_CTAG_FILTER |
10824 			       NETIF_F_HW_VLAN_CTAG_RX |
10825 			       NETIF_F_HW_VLAN_CTAG_TX |
10826 			       NETIF_F_RXALL |
10827 			       NETIF_F_HW_L2FW_DOFFLOAD;
10828 
10829 	if (hw->mac.type >= ixgbe_mac_82599EB)
10830 		netdev->hw_features |= NETIF_F_NTUPLE |
10831 				       NETIF_F_HW_TC;
10832 
10833 	if (pci_using_dac)
10834 		netdev->features |= NETIF_F_HIGHDMA;
10835 
10836 	netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
10837 	netdev->hw_enc_features |= netdev->vlan_features;
10838 	netdev->mpls_features |= NETIF_F_SG |
10839 				 NETIF_F_TSO |
10840 				 NETIF_F_TSO6 |
10841 				 NETIF_F_HW_CSUM;
10842 	netdev->mpls_features |= IXGBE_GSO_PARTIAL_FEATURES;
10843 
10844 	/* set this bit last since it cannot be part of vlan_features */
10845 	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
10846 			    NETIF_F_HW_VLAN_CTAG_RX |
10847 			    NETIF_F_HW_VLAN_CTAG_TX;
10848 
10849 	netdev->priv_flags |= IFF_UNICAST_FLT;
10850 	netdev->priv_flags |= IFF_SUPP_NOFCS;
10851 
10852 	/* MTU range: 68 - 9710 */
10853 	netdev->min_mtu = ETH_MIN_MTU;
10854 	netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE - (ETH_HLEN + ETH_FCS_LEN);
10855 
10856 #ifdef CONFIG_IXGBE_DCB
10857 	if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
10858 		netdev->dcbnl_ops = &ixgbe_dcbnl_ops;
10859 #endif
10860 
10861 #ifdef IXGBE_FCOE
10862 	if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
10863 		unsigned int fcoe_l;
10864 
10865 		if (hw->mac.ops.get_device_caps) {
10866 			hw->mac.ops.get_device_caps(hw, &device_caps);
10867 			if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
10868 				adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
10869 		}
10870 
10871 
10872 		fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
10873 		adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
10874 
10875 		netdev->features |= NETIF_F_FSO |
10876 				    NETIF_F_FCOE_CRC;
10877 
10878 		netdev->vlan_features |= NETIF_F_FSO |
10879 					 NETIF_F_FCOE_CRC |
10880 					 NETIF_F_FCOE_MTU;
10881 	}
10882 #endif /* IXGBE_FCOE */
10883 	if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
10884 		netdev->hw_features |= NETIF_F_LRO;
10885 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
10886 		netdev->features |= NETIF_F_LRO;
10887 
10888 	if (ixgbe_check_fw_error(adapter)) {
10889 		err = -EIO;
10890 		goto err_sw_init;
10891 	}
10892 
10893 	/* make sure the EEPROM is good */
10894 	if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
10895 		e_dev_err("The EEPROM Checksum Is Not Valid\n");
10896 		err = -EIO;
10897 		goto err_sw_init;
10898 	}
10899 
10900 	eth_platform_get_mac_address(&adapter->pdev->dev,
10901 				     adapter->hw.mac.perm_addr);
10902 
10903 	memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
10904 
10905 	if (!is_valid_ether_addr(netdev->dev_addr)) {
10906 		e_dev_err("invalid MAC address\n");
10907 		err = -EIO;
10908 		goto err_sw_init;
10909 	}
10910 
10911 	/* Set hw->mac.addr to permanent MAC address */
10912 	ether_addr_copy(hw->mac.addr, hw->mac.perm_addr);
10913 	ixgbe_mac_set_default_filter(adapter);
10914 
10915 	timer_setup(&adapter->service_timer, ixgbe_service_timer, 0);
10916 
10917 	if (ixgbe_removed(hw->hw_addr)) {
10918 		err = -EIO;
10919 		goto err_sw_init;
10920 	}
10921 	INIT_WORK(&adapter->service_task, ixgbe_service_task);
10922 	set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
10923 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
10924 
10925 	err = ixgbe_init_interrupt_scheme(adapter);
10926 	if (err)
10927 		goto err_sw_init;
10928 
10929 	for (i = 0; i < adapter->num_rx_queues; i++)
10930 		u64_stats_init(&adapter->rx_ring[i]->syncp);
10931 	for (i = 0; i < adapter->num_tx_queues; i++)
10932 		u64_stats_init(&adapter->tx_ring[i]->syncp);
10933 	for (i = 0; i < adapter->num_xdp_queues; i++)
10934 		u64_stats_init(&adapter->xdp_ring[i]->syncp);
10935 
10936 	/* WOL not supported for all devices */
10937 	adapter->wol = 0;
10938 	hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
10939 	hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
10940 						pdev->subsystem_device);
10941 	if (hw->wol_enabled)
10942 		adapter->wol = IXGBE_WUFC_MAG;
10943 
10944 	device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
10945 
10946 	/* save off EEPROM version number */
10947 	ixgbe_set_fw_version(adapter);
10948 
10949 	/* pick up the PCI bus settings for reporting later */
10950 	if (ixgbe_pcie_from_parent(hw))
10951 		ixgbe_get_parent_bus_info(adapter);
10952 	else
10953 		 hw->mac.ops.get_bus_info(hw);
10954 
10955 	/* calculate the expected PCIe bandwidth required for optimal
10956 	 * performance. Note that some older parts will never have enough
10957 	 * bandwidth due to being older generation PCIe parts. We clamp these
10958 	 * parts to ensure no warning is displayed if it can't be fixed.
10959 	 */
10960 	switch (hw->mac.type) {
10961 	case ixgbe_mac_82598EB:
10962 		expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
10963 		break;
10964 	default:
10965 		expected_gts = ixgbe_enumerate_functions(adapter) * 10;
10966 		break;
10967 	}
10968 
10969 	/* don't check link if we failed to enumerate functions */
10970 	if (expected_gts > 0)
10971 		ixgbe_check_minimum_link(adapter, expected_gts);
10972 
10973 	err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
10974 	if (err)
10975 		strlcpy(part_str, "Unknown", sizeof(part_str));
10976 	if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
10977 		e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
10978 			   hw->mac.type, hw->phy.type, hw->phy.sfp_type,
10979 			   part_str);
10980 	else
10981 		e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
10982 			   hw->mac.type, hw->phy.type, part_str);
10983 
10984 	e_dev_info("%pM\n", netdev->dev_addr);
10985 
10986 	/* reset the hardware with the new settings */
10987 	err = hw->mac.ops.start_hw(hw);
10988 	if (err == IXGBE_ERR_EEPROM_VERSION) {
10989 		/* We are running on a pre-production device, log a warning */
10990 		e_dev_warn("This device is a pre-production adapter/LOM. "
10991 			   "Please be aware there may be issues associated "
10992 			   "with your hardware.  If you are experiencing "
10993 			   "problems please contact your Intel or hardware "
10994 			   "representative who provided you with this "
10995 			   "hardware.\n");
10996 	}
10997 	strcpy(netdev->name, "eth%d");
10998 	pci_set_drvdata(pdev, adapter);
10999 	err = register_netdev(netdev);
11000 	if (err)
11001 		goto err_register;
11002 
11003 
11004 	/* power down the optics for 82599 SFP+ fiber */
11005 	if (hw->mac.ops.disable_tx_laser)
11006 		hw->mac.ops.disable_tx_laser(hw);
11007 
11008 	/* carrier off reporting is important to ethtool even BEFORE open */
11009 	netif_carrier_off(netdev);
11010 
11011 #ifdef CONFIG_IXGBE_DCA
11012 	if (dca_add_requester(&pdev->dev) == 0) {
11013 		adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
11014 		ixgbe_setup_dca(adapter);
11015 	}
11016 #endif
11017 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
11018 		e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
11019 		for (i = 0; i < adapter->num_vfs; i++)
11020 			ixgbe_vf_configuration(pdev, (i | 0x10000000));
11021 	}
11022 
11023 	/* firmware requires driver version to be 0xFFFFFFFF
11024 	 * since os does not support feature
11025 	 */
11026 	if (hw->mac.ops.set_fw_drv_ver)
11027 		hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF, 0xFF,
11028 					   sizeof(UTS_RELEASE) - 1,
11029 					   UTS_RELEASE);
11030 
11031 	/* add san mac addr to netdev */
11032 	ixgbe_add_sanmac_netdev(netdev);
11033 
11034 	e_dev_info("%s\n", ixgbe_default_device_descr);
11035 
11036 #ifdef CONFIG_IXGBE_HWMON
11037 	if (ixgbe_sysfs_init(adapter))
11038 		e_err(probe, "failed to allocate sysfs resources\n");
11039 #endif /* CONFIG_IXGBE_HWMON */
11040 
11041 	ixgbe_dbg_adapter_init(adapter);
11042 
11043 	/* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
11044 	if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
11045 		hw->mac.ops.setup_link(hw,
11046 			IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
11047 			true);
11048 
11049 	err = ixgbe_mii_bus_init(hw);
11050 	if (err)
11051 		goto err_netdev;
11052 
11053 	return 0;
11054 
11055 err_netdev:
11056 	unregister_netdev(netdev);
11057 err_register:
11058 	ixgbe_release_hw_control(adapter);
11059 	ixgbe_clear_interrupt_scheme(adapter);
11060 err_sw_init:
11061 	ixgbe_disable_sriov(adapter);
11062 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
11063 	iounmap(adapter->io_addr);
11064 	kfree(adapter->jump_tables[0]);
11065 	kfree(adapter->mac_table);
11066 	kfree(adapter->rss_key);
11067 	bitmap_free(adapter->af_xdp_zc_qps);
11068 err_ioremap:
11069 	disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11070 	free_netdev(netdev);
11071 err_alloc_etherdev:
11072 	pci_release_mem_regions(pdev);
11073 err_pci_reg:
11074 err_dma:
11075 	if (!adapter || disable_dev)
11076 		pci_disable_device(pdev);
11077 	return err;
11078 }
11079 
11080 /**
11081  * ixgbe_remove - Device Removal Routine
11082  * @pdev: PCI device information struct
11083  *
11084  * ixgbe_remove is called by the PCI subsystem to alert the driver
11085  * that it should release a PCI device.  The could be caused by a
11086  * Hot-Plug event, or because the driver is going to be removed from
11087  * memory.
11088  **/
11089 static void ixgbe_remove(struct pci_dev *pdev)
11090 {
11091 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11092 	struct net_device *netdev;
11093 	bool disable_dev;
11094 	int i;
11095 
11096 	/* if !adapter then we already cleaned up in probe */
11097 	if (!adapter)
11098 		return;
11099 
11100 	netdev  = adapter->netdev;
11101 	ixgbe_dbg_adapter_exit(adapter);
11102 
11103 	set_bit(__IXGBE_REMOVING, &adapter->state);
11104 	cancel_work_sync(&adapter->service_task);
11105 
11106 	if (adapter->mii_bus)
11107 		mdiobus_unregister(adapter->mii_bus);
11108 
11109 #ifdef CONFIG_IXGBE_DCA
11110 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
11111 		adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
11112 		dca_remove_requester(&pdev->dev);
11113 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
11114 				IXGBE_DCA_CTRL_DCA_DISABLE);
11115 	}
11116 
11117 #endif
11118 #ifdef CONFIG_IXGBE_HWMON
11119 	ixgbe_sysfs_exit(adapter);
11120 #endif /* CONFIG_IXGBE_HWMON */
11121 
11122 	/* remove the added san mac */
11123 	ixgbe_del_sanmac_netdev(netdev);
11124 
11125 #ifdef CONFIG_PCI_IOV
11126 	ixgbe_disable_sriov(adapter);
11127 #endif
11128 	if (netdev->reg_state == NETREG_REGISTERED)
11129 		unregister_netdev(netdev);
11130 
11131 	ixgbe_stop_ipsec_offload(adapter);
11132 	ixgbe_clear_interrupt_scheme(adapter);
11133 
11134 	ixgbe_release_hw_control(adapter);
11135 
11136 #ifdef CONFIG_DCB
11137 	kfree(adapter->ixgbe_ieee_pfc);
11138 	kfree(adapter->ixgbe_ieee_ets);
11139 
11140 #endif
11141 	iounmap(adapter->io_addr);
11142 	pci_release_mem_regions(pdev);
11143 
11144 	e_dev_info("complete\n");
11145 
11146 	for (i = 0; i < IXGBE_MAX_LINK_HANDLE; i++) {
11147 		if (adapter->jump_tables[i]) {
11148 			kfree(adapter->jump_tables[i]->input);
11149 			kfree(adapter->jump_tables[i]->mask);
11150 		}
11151 		kfree(adapter->jump_tables[i]);
11152 	}
11153 
11154 	kfree(adapter->mac_table);
11155 	kfree(adapter->rss_key);
11156 	bitmap_free(adapter->af_xdp_zc_qps);
11157 	disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11158 	free_netdev(netdev);
11159 
11160 	pci_disable_pcie_error_reporting(pdev);
11161 
11162 	if (disable_dev)
11163 		pci_disable_device(pdev);
11164 }
11165 
11166 /**
11167  * ixgbe_io_error_detected - called when PCI error is detected
11168  * @pdev: Pointer to PCI device
11169  * @state: The current pci connection state
11170  *
11171  * This function is called after a PCI bus error affecting
11172  * this device has been detected.
11173  */
11174 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
11175 						pci_channel_state_t state)
11176 {
11177 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11178 	struct net_device *netdev = adapter->netdev;
11179 
11180 #ifdef CONFIG_PCI_IOV
11181 	struct ixgbe_hw *hw = &adapter->hw;
11182 	struct pci_dev *bdev, *vfdev;
11183 	u32 dw0, dw1, dw2, dw3;
11184 	int vf, pos;
11185 	u16 req_id, pf_func;
11186 
11187 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
11188 	    adapter->num_vfs == 0)
11189 		goto skip_bad_vf_detection;
11190 
11191 	bdev = pdev->bus->self;
11192 	while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
11193 		bdev = bdev->bus->self;
11194 
11195 	if (!bdev)
11196 		goto skip_bad_vf_detection;
11197 
11198 	pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
11199 	if (!pos)
11200 		goto skip_bad_vf_detection;
11201 
11202 	dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
11203 	dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
11204 	dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
11205 	dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
11206 	if (ixgbe_removed(hw->hw_addr))
11207 		goto skip_bad_vf_detection;
11208 
11209 	req_id = dw1 >> 16;
11210 	/* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
11211 	if (!(req_id & 0x0080))
11212 		goto skip_bad_vf_detection;
11213 
11214 	pf_func = req_id & 0x01;
11215 	if ((pf_func & 1) == (pdev->devfn & 1)) {
11216 		unsigned int device_id;
11217 
11218 		vf = (req_id & 0x7F) >> 1;
11219 		e_dev_err("VF %d has caused a PCIe error\n", vf);
11220 		e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
11221 				"%8.8x\tdw3: %8.8x\n",
11222 		dw0, dw1, dw2, dw3);
11223 		switch (adapter->hw.mac.type) {
11224 		case ixgbe_mac_82599EB:
11225 			device_id = IXGBE_82599_VF_DEVICE_ID;
11226 			break;
11227 		case ixgbe_mac_X540:
11228 			device_id = IXGBE_X540_VF_DEVICE_ID;
11229 			break;
11230 		case ixgbe_mac_X550:
11231 			device_id = IXGBE_DEV_ID_X550_VF;
11232 			break;
11233 		case ixgbe_mac_X550EM_x:
11234 			device_id = IXGBE_DEV_ID_X550EM_X_VF;
11235 			break;
11236 		case ixgbe_mac_x550em_a:
11237 			device_id = IXGBE_DEV_ID_X550EM_A_VF;
11238 			break;
11239 		default:
11240 			device_id = 0;
11241 			break;
11242 		}
11243 
11244 		/* Find the pci device of the offending VF */
11245 		vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
11246 		while (vfdev) {
11247 			if (vfdev->devfn == (req_id & 0xFF))
11248 				break;
11249 			vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
11250 					       device_id, vfdev);
11251 		}
11252 		/*
11253 		 * There's a slim chance the VF could have been hot plugged,
11254 		 * so if it is no longer present we don't need to issue the
11255 		 * VFLR.  Just clean up the AER in that case.
11256 		 */
11257 		if (vfdev) {
11258 			pcie_flr(vfdev);
11259 			/* Free device reference count */
11260 			pci_dev_put(vfdev);
11261 		}
11262 	}
11263 
11264 	/*
11265 	 * Even though the error may have occurred on the other port
11266 	 * we still need to increment the vf error reference count for
11267 	 * both ports because the I/O resume function will be called
11268 	 * for both of them.
11269 	 */
11270 	adapter->vferr_refcount++;
11271 
11272 	return PCI_ERS_RESULT_RECOVERED;
11273 
11274 skip_bad_vf_detection:
11275 #endif /* CONFIG_PCI_IOV */
11276 	if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
11277 		return PCI_ERS_RESULT_DISCONNECT;
11278 
11279 	if (!netif_device_present(netdev))
11280 		return PCI_ERS_RESULT_DISCONNECT;
11281 
11282 	rtnl_lock();
11283 	netif_device_detach(netdev);
11284 
11285 	if (netif_running(netdev))
11286 		ixgbe_close_suspend(adapter);
11287 
11288 	if (state == pci_channel_io_perm_failure) {
11289 		rtnl_unlock();
11290 		return PCI_ERS_RESULT_DISCONNECT;
11291 	}
11292 
11293 	if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
11294 		pci_disable_device(pdev);
11295 	rtnl_unlock();
11296 
11297 	/* Request a slot reset. */
11298 	return PCI_ERS_RESULT_NEED_RESET;
11299 }
11300 
11301 /**
11302  * ixgbe_io_slot_reset - called after the pci bus has been reset.
11303  * @pdev: Pointer to PCI device
11304  *
11305  * Restart the card from scratch, as if from a cold-boot.
11306  */
11307 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
11308 {
11309 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11310 	pci_ers_result_t result;
11311 
11312 	if (pci_enable_device_mem(pdev)) {
11313 		e_err(probe, "Cannot re-enable PCI device after reset.\n");
11314 		result = PCI_ERS_RESULT_DISCONNECT;
11315 	} else {
11316 		smp_mb__before_atomic();
11317 		clear_bit(__IXGBE_DISABLED, &adapter->state);
11318 		adapter->hw.hw_addr = adapter->io_addr;
11319 		pci_set_master(pdev);
11320 		pci_restore_state(pdev);
11321 		pci_save_state(pdev);
11322 
11323 		pci_wake_from_d3(pdev, false);
11324 
11325 		ixgbe_reset(adapter);
11326 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
11327 		result = PCI_ERS_RESULT_RECOVERED;
11328 	}
11329 
11330 	return result;
11331 }
11332 
11333 /**
11334  * ixgbe_io_resume - called when traffic can start flowing again.
11335  * @pdev: Pointer to PCI device
11336  *
11337  * This callback is called when the error recovery driver tells us that
11338  * its OK to resume normal operation.
11339  */
11340 static void ixgbe_io_resume(struct pci_dev *pdev)
11341 {
11342 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11343 	struct net_device *netdev = adapter->netdev;
11344 
11345 #ifdef CONFIG_PCI_IOV
11346 	if (adapter->vferr_refcount) {
11347 		e_info(drv, "Resuming after VF err\n");
11348 		adapter->vferr_refcount--;
11349 		return;
11350 	}
11351 
11352 #endif
11353 	rtnl_lock();
11354 	if (netif_running(netdev))
11355 		ixgbe_open(netdev);
11356 
11357 	netif_device_attach(netdev);
11358 	rtnl_unlock();
11359 }
11360 
11361 static const struct pci_error_handlers ixgbe_err_handler = {
11362 	.error_detected = ixgbe_io_error_detected,
11363 	.slot_reset = ixgbe_io_slot_reset,
11364 	.resume = ixgbe_io_resume,
11365 };
11366 
11367 static SIMPLE_DEV_PM_OPS(ixgbe_pm_ops, ixgbe_suspend, ixgbe_resume);
11368 
11369 static struct pci_driver ixgbe_driver = {
11370 	.name      = ixgbe_driver_name,
11371 	.id_table  = ixgbe_pci_tbl,
11372 	.probe     = ixgbe_probe,
11373 	.remove    = ixgbe_remove,
11374 	.driver.pm = &ixgbe_pm_ops,
11375 	.shutdown  = ixgbe_shutdown,
11376 	.sriov_configure = ixgbe_pci_sriov_configure,
11377 	.err_handler = &ixgbe_err_handler
11378 };
11379 
11380 /**
11381  * ixgbe_init_module - Driver Registration Routine
11382  *
11383  * ixgbe_init_module is the first routine called when the driver is
11384  * loaded. All it does is register with the PCI subsystem.
11385  **/
11386 static int __init ixgbe_init_module(void)
11387 {
11388 	int ret;
11389 	pr_info("%s\n", ixgbe_driver_string);
11390 	pr_info("%s\n", ixgbe_copyright);
11391 
11392 	ixgbe_wq = create_singlethread_workqueue(ixgbe_driver_name);
11393 	if (!ixgbe_wq) {
11394 		pr_err("%s: Failed to create workqueue\n", ixgbe_driver_name);
11395 		return -ENOMEM;
11396 	}
11397 
11398 	ixgbe_dbg_init();
11399 
11400 	ret = pci_register_driver(&ixgbe_driver);
11401 	if (ret) {
11402 		destroy_workqueue(ixgbe_wq);
11403 		ixgbe_dbg_exit();
11404 		return ret;
11405 	}
11406 
11407 #ifdef CONFIG_IXGBE_DCA
11408 	dca_register_notify(&dca_notifier);
11409 #endif
11410 
11411 	return 0;
11412 }
11413 
11414 module_init(ixgbe_init_module);
11415 
11416 /**
11417  * ixgbe_exit_module - Driver Exit Cleanup Routine
11418  *
11419  * ixgbe_exit_module is called just before the driver is removed
11420  * from memory.
11421  **/
11422 static void __exit ixgbe_exit_module(void)
11423 {
11424 #ifdef CONFIG_IXGBE_DCA
11425 	dca_unregister_notify(&dca_notifier);
11426 #endif
11427 	pci_unregister_driver(&ixgbe_driver);
11428 
11429 	ixgbe_dbg_exit();
11430 	if (ixgbe_wq) {
11431 		destroy_workqueue(ixgbe_wq);
11432 		ixgbe_wq = NULL;
11433 	}
11434 }
11435 
11436 #ifdef CONFIG_IXGBE_DCA
11437 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
11438 			    void *p)
11439 {
11440 	int ret_val;
11441 
11442 	ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
11443 					 __ixgbe_notify_dca);
11444 
11445 	return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
11446 }
11447 
11448 #endif /* CONFIG_IXGBE_DCA */
11449 
11450 module_exit(ixgbe_exit_module);
11451 
11452 /* ixgbe_main.c */
11453