xref: /openbmc/linux/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c (revision b0e55fef624e511e060fa05e4ca96cae6d902f04)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 1999 - 2018 Intel Corporation. */
3 
4 #include <linux/types.h>
5 #include <linux/module.h>
6 #include <linux/pci.h>
7 #include <linux/netdevice.h>
8 #include <linux/vmalloc.h>
9 #include <linux/string.h>
10 #include <linux/in.h>
11 #include <linux/interrupt.h>
12 #include <linux/ip.h>
13 #include <linux/tcp.h>
14 #include <linux/sctp.h>
15 #include <linux/pkt_sched.h>
16 #include <linux/ipv6.h>
17 #include <linux/slab.h>
18 #include <net/checksum.h>
19 #include <net/ip6_checksum.h>
20 #include <linux/etherdevice.h>
21 #include <linux/ethtool.h>
22 #include <linux/if.h>
23 #include <linux/if_vlan.h>
24 #include <linux/if_macvlan.h>
25 #include <linux/if_bridge.h>
26 #include <linux/prefetch.h>
27 #include <linux/bpf.h>
28 #include <linux/bpf_trace.h>
29 #include <linux/atomic.h>
30 #include <linux/numa.h>
31 #include <scsi/fc/fc_fcoe.h>
32 #include <net/udp_tunnel.h>
33 #include <net/pkt_cls.h>
34 #include <net/tc_act/tc_gact.h>
35 #include <net/tc_act/tc_mirred.h>
36 #include <net/vxlan.h>
37 #include <net/mpls.h>
38 #include <net/xdp_sock.h>
39 #include <net/xfrm.h>
40 
41 #include "ixgbe.h"
42 #include "ixgbe_common.h"
43 #include "ixgbe_dcb_82599.h"
44 #include "ixgbe_phy.h"
45 #include "ixgbe_sriov.h"
46 #include "ixgbe_model.h"
47 #include "ixgbe_txrx_common.h"
48 
49 char ixgbe_driver_name[] = "ixgbe";
50 static const char ixgbe_driver_string[] =
51 			      "Intel(R) 10 Gigabit PCI Express Network Driver";
52 #ifdef IXGBE_FCOE
53 char ixgbe_default_device_descr[] =
54 			      "Intel(R) 10 Gigabit Network Connection";
55 #else
56 static char ixgbe_default_device_descr[] =
57 			      "Intel(R) 10 Gigabit Network Connection";
58 #endif
59 #define DRV_VERSION "5.1.0-k"
60 const char ixgbe_driver_version[] = DRV_VERSION;
61 static const char ixgbe_copyright[] =
62 				"Copyright (c) 1999-2016 Intel Corporation.";
63 
64 static const char ixgbe_overheat_msg[] = "Network adapter has been stopped because it has over heated. Restart the computer. If the problem persists, power off the system and replace the adapter";
65 
66 static const struct ixgbe_info *ixgbe_info_tbl[] = {
67 	[board_82598]		= &ixgbe_82598_info,
68 	[board_82599]		= &ixgbe_82599_info,
69 	[board_X540]		= &ixgbe_X540_info,
70 	[board_X550]		= &ixgbe_X550_info,
71 	[board_X550EM_x]	= &ixgbe_X550EM_x_info,
72 	[board_x550em_x_fw]	= &ixgbe_x550em_x_fw_info,
73 	[board_x550em_a]	= &ixgbe_x550em_a_info,
74 	[board_x550em_a_fw]	= &ixgbe_x550em_a_fw_info,
75 };
76 
77 /* ixgbe_pci_tbl - PCI Device ID Table
78  *
79  * Wildcard entries (PCI_ANY_ID) should come last
80  * Last entry must be all 0s
81  *
82  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
83  *   Class, Class Mask, private data (not used) }
84  */
85 static const struct pci_device_id ixgbe_pci_tbl[] = {
86 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
87 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
88 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
89 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
90 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
91 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
92 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
93 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
94 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
95 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
96 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
97 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
98 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
99 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
100 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
101 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
102 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
103 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
104 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
105 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
106 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
107 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
108 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
109 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
110 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
111 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
112 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_QSFP_SF_QP), board_82599 },
113 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
114 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
115 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
116 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T), board_X550},
117 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T1), board_X550},
118 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KX4), board_X550EM_x},
119 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_XFI), board_X550EM_x},
120 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KR), board_X550EM_x},
121 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_10G_T), board_X550EM_x},
122 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_SFP), board_X550EM_x},
123 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_1G_T), board_x550em_x_fw},
124 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR), board_x550em_a },
125 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR_L), board_x550em_a },
126 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP_N), board_x550em_a },
127 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII), board_x550em_a },
128 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII_L), board_x550em_a },
129 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_10G_T), board_x550em_a},
130 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP), board_x550em_a },
131 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T), board_x550em_a_fw },
132 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T_L), board_x550em_a_fw },
133 	/* required last entry */
134 	{0, }
135 };
136 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
137 
138 #ifdef CONFIG_IXGBE_DCA
139 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
140 			    void *p);
141 static struct notifier_block dca_notifier = {
142 	.notifier_call = ixgbe_notify_dca,
143 	.next          = NULL,
144 	.priority      = 0
145 };
146 #endif
147 
148 #ifdef CONFIG_PCI_IOV
149 static unsigned int max_vfs;
150 module_param(max_vfs, uint, 0);
151 MODULE_PARM_DESC(max_vfs,
152 		 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63. (Deprecated)");
153 #endif /* CONFIG_PCI_IOV */
154 
155 static unsigned int allow_unsupported_sfp;
156 module_param(allow_unsupported_sfp, uint, 0);
157 MODULE_PARM_DESC(allow_unsupported_sfp,
158 		 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
159 
160 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
161 static int debug = -1;
162 module_param(debug, int, 0);
163 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
164 
165 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
166 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
167 MODULE_LICENSE("GPL v2");
168 MODULE_VERSION(DRV_VERSION);
169 
170 static struct workqueue_struct *ixgbe_wq;
171 
172 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev);
173 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *);
174 
175 static const struct net_device_ops ixgbe_netdev_ops;
176 
177 static bool netif_is_ixgbe(struct net_device *dev)
178 {
179 	return dev && (dev->netdev_ops == &ixgbe_netdev_ops);
180 }
181 
182 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
183 					  u32 reg, u16 *value)
184 {
185 	struct pci_dev *parent_dev;
186 	struct pci_bus *parent_bus;
187 
188 	parent_bus = adapter->pdev->bus->parent;
189 	if (!parent_bus)
190 		return -1;
191 
192 	parent_dev = parent_bus->self;
193 	if (!parent_dev)
194 		return -1;
195 
196 	if (!pci_is_pcie(parent_dev))
197 		return -1;
198 
199 	pcie_capability_read_word(parent_dev, reg, value);
200 	if (*value == IXGBE_FAILED_READ_CFG_WORD &&
201 	    ixgbe_check_cfg_remove(&adapter->hw, parent_dev))
202 		return -1;
203 	return 0;
204 }
205 
206 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
207 {
208 	struct ixgbe_hw *hw = &adapter->hw;
209 	u16 link_status = 0;
210 	int err;
211 
212 	hw->bus.type = ixgbe_bus_type_pci_express;
213 
214 	/* Get the negotiated link width and speed from PCI config space of the
215 	 * parent, as this device is behind a switch
216 	 */
217 	err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
218 
219 	/* assume caller will handle error case */
220 	if (err)
221 		return err;
222 
223 	hw->bus.width = ixgbe_convert_bus_width(link_status);
224 	hw->bus.speed = ixgbe_convert_bus_speed(link_status);
225 
226 	return 0;
227 }
228 
229 /**
230  * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
231  * @hw: hw specific details
232  *
233  * This function is used by probe to determine whether a device's PCI-Express
234  * bandwidth details should be gathered from the parent bus instead of from the
235  * device. Used to ensure that various locations all have the correct device ID
236  * checks.
237  */
238 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
239 {
240 	switch (hw->device_id) {
241 	case IXGBE_DEV_ID_82599_SFP_SF_QP:
242 	case IXGBE_DEV_ID_82599_QSFP_SF_QP:
243 		return true;
244 	default:
245 		return false;
246 	}
247 }
248 
249 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
250 				     int expected_gts)
251 {
252 	struct ixgbe_hw *hw = &adapter->hw;
253 	struct pci_dev *pdev;
254 
255 	/* Some devices are not connected over PCIe and thus do not negotiate
256 	 * speed. These devices do not have valid bus info, and thus any report
257 	 * we generate may not be correct.
258 	 */
259 	if (hw->bus.type == ixgbe_bus_type_internal)
260 		return;
261 
262 	/* determine whether to use the parent device */
263 	if (ixgbe_pcie_from_parent(&adapter->hw))
264 		pdev = adapter->pdev->bus->parent->self;
265 	else
266 		pdev = adapter->pdev;
267 
268 	pcie_print_link_status(pdev);
269 }
270 
271 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
272 {
273 	if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
274 	    !test_bit(__IXGBE_REMOVING, &adapter->state) &&
275 	    !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
276 		queue_work(ixgbe_wq, &adapter->service_task);
277 }
278 
279 static void ixgbe_remove_adapter(struct ixgbe_hw *hw)
280 {
281 	struct ixgbe_adapter *adapter = hw->back;
282 
283 	if (!hw->hw_addr)
284 		return;
285 	hw->hw_addr = NULL;
286 	e_dev_err("Adapter removed\n");
287 	if (test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
288 		ixgbe_service_event_schedule(adapter);
289 }
290 
291 static u32 ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg)
292 {
293 	u8 __iomem *reg_addr;
294 	u32 value;
295 	int i;
296 
297 	reg_addr = READ_ONCE(hw->hw_addr);
298 	if (ixgbe_removed(reg_addr))
299 		return IXGBE_FAILED_READ_REG;
300 
301 	/* Register read of 0xFFFFFFF can indicate the adapter has been removed,
302 	 * so perform several status register reads to determine if the adapter
303 	 * has been removed.
304 	 */
305 	for (i = 0; i < IXGBE_FAILED_READ_RETRIES; i++) {
306 		value = readl(reg_addr + IXGBE_STATUS);
307 		if (value != IXGBE_FAILED_READ_REG)
308 			break;
309 		mdelay(3);
310 	}
311 
312 	if (value == IXGBE_FAILED_READ_REG)
313 		ixgbe_remove_adapter(hw);
314 	else
315 		value = readl(reg_addr + reg);
316 	return value;
317 }
318 
319 /**
320  * ixgbe_read_reg - Read from device register
321  * @hw: hw specific details
322  * @reg: offset of register to read
323  *
324  * Returns : value read or IXGBE_FAILED_READ_REG if removed
325  *
326  * This function is used to read device registers. It checks for device
327  * removal by confirming any read that returns all ones by checking the
328  * status register value for all ones. This function avoids reading from
329  * the hardware if a removal was previously detected in which case it
330  * returns IXGBE_FAILED_READ_REG (all ones).
331  */
332 u32 ixgbe_read_reg(struct ixgbe_hw *hw, u32 reg)
333 {
334 	u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
335 	u32 value;
336 
337 	if (ixgbe_removed(reg_addr))
338 		return IXGBE_FAILED_READ_REG;
339 	if (unlikely(hw->phy.nw_mng_if_sel &
340 		     IXGBE_NW_MNG_IF_SEL_SGMII_ENABLE)) {
341 		struct ixgbe_adapter *adapter;
342 		int i;
343 
344 		for (i = 0; i < 200; ++i) {
345 			value = readl(reg_addr + IXGBE_MAC_SGMII_BUSY);
346 			if (likely(!value))
347 				goto writes_completed;
348 			if (value == IXGBE_FAILED_READ_REG) {
349 				ixgbe_remove_adapter(hw);
350 				return IXGBE_FAILED_READ_REG;
351 			}
352 			udelay(5);
353 		}
354 
355 		adapter = hw->back;
356 		e_warn(hw, "register writes incomplete %08x\n", value);
357 	}
358 
359 writes_completed:
360 	value = readl(reg_addr + reg);
361 	if (unlikely(value == IXGBE_FAILED_READ_REG))
362 		value = ixgbe_check_remove(hw, reg);
363 	return value;
364 }
365 
366 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev)
367 {
368 	u16 value;
369 
370 	pci_read_config_word(pdev, PCI_VENDOR_ID, &value);
371 	if (value == IXGBE_FAILED_READ_CFG_WORD) {
372 		ixgbe_remove_adapter(hw);
373 		return true;
374 	}
375 	return false;
376 }
377 
378 u16 ixgbe_read_pci_cfg_word(struct ixgbe_hw *hw, u32 reg)
379 {
380 	struct ixgbe_adapter *adapter = hw->back;
381 	u16 value;
382 
383 	if (ixgbe_removed(hw->hw_addr))
384 		return IXGBE_FAILED_READ_CFG_WORD;
385 	pci_read_config_word(adapter->pdev, reg, &value);
386 	if (value == IXGBE_FAILED_READ_CFG_WORD &&
387 	    ixgbe_check_cfg_remove(hw, adapter->pdev))
388 		return IXGBE_FAILED_READ_CFG_WORD;
389 	return value;
390 }
391 
392 #ifdef CONFIG_PCI_IOV
393 static u32 ixgbe_read_pci_cfg_dword(struct ixgbe_hw *hw, u32 reg)
394 {
395 	struct ixgbe_adapter *adapter = hw->back;
396 	u32 value;
397 
398 	if (ixgbe_removed(hw->hw_addr))
399 		return IXGBE_FAILED_READ_CFG_DWORD;
400 	pci_read_config_dword(adapter->pdev, reg, &value);
401 	if (value == IXGBE_FAILED_READ_CFG_DWORD &&
402 	    ixgbe_check_cfg_remove(hw, adapter->pdev))
403 		return IXGBE_FAILED_READ_CFG_DWORD;
404 	return value;
405 }
406 #endif /* CONFIG_PCI_IOV */
407 
408 void ixgbe_write_pci_cfg_word(struct ixgbe_hw *hw, u32 reg, u16 value)
409 {
410 	struct ixgbe_adapter *adapter = hw->back;
411 
412 	if (ixgbe_removed(hw->hw_addr))
413 		return;
414 	pci_write_config_word(adapter->pdev, reg, value);
415 }
416 
417 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
418 {
419 	BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
420 
421 	/* flush memory to make sure state is correct before next watchdog */
422 	smp_mb__before_atomic();
423 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
424 }
425 
426 struct ixgbe_reg_info {
427 	u32 ofs;
428 	char *name;
429 };
430 
431 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
432 
433 	/* General Registers */
434 	{IXGBE_CTRL, "CTRL"},
435 	{IXGBE_STATUS, "STATUS"},
436 	{IXGBE_CTRL_EXT, "CTRL_EXT"},
437 
438 	/* Interrupt Registers */
439 	{IXGBE_EICR, "EICR"},
440 
441 	/* RX Registers */
442 	{IXGBE_SRRCTL(0), "SRRCTL"},
443 	{IXGBE_DCA_RXCTRL(0), "DRXCTL"},
444 	{IXGBE_RDLEN(0), "RDLEN"},
445 	{IXGBE_RDH(0), "RDH"},
446 	{IXGBE_RDT(0), "RDT"},
447 	{IXGBE_RXDCTL(0), "RXDCTL"},
448 	{IXGBE_RDBAL(0), "RDBAL"},
449 	{IXGBE_RDBAH(0), "RDBAH"},
450 
451 	/* TX Registers */
452 	{IXGBE_TDBAL(0), "TDBAL"},
453 	{IXGBE_TDBAH(0), "TDBAH"},
454 	{IXGBE_TDLEN(0), "TDLEN"},
455 	{IXGBE_TDH(0), "TDH"},
456 	{IXGBE_TDT(0), "TDT"},
457 	{IXGBE_TXDCTL(0), "TXDCTL"},
458 
459 	/* List Terminator */
460 	{ .name = NULL }
461 };
462 
463 
464 /*
465  * ixgbe_regdump - register printout routine
466  */
467 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
468 {
469 	int i;
470 	char rname[16];
471 	u32 regs[64];
472 
473 	switch (reginfo->ofs) {
474 	case IXGBE_SRRCTL(0):
475 		for (i = 0; i < 64; i++)
476 			regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
477 		break;
478 	case IXGBE_DCA_RXCTRL(0):
479 		for (i = 0; i < 64; i++)
480 			regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
481 		break;
482 	case IXGBE_RDLEN(0):
483 		for (i = 0; i < 64; i++)
484 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
485 		break;
486 	case IXGBE_RDH(0):
487 		for (i = 0; i < 64; i++)
488 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
489 		break;
490 	case IXGBE_RDT(0):
491 		for (i = 0; i < 64; i++)
492 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
493 		break;
494 	case IXGBE_RXDCTL(0):
495 		for (i = 0; i < 64; i++)
496 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
497 		break;
498 	case IXGBE_RDBAL(0):
499 		for (i = 0; i < 64; i++)
500 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
501 		break;
502 	case IXGBE_RDBAH(0):
503 		for (i = 0; i < 64; i++)
504 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
505 		break;
506 	case IXGBE_TDBAL(0):
507 		for (i = 0; i < 64; i++)
508 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
509 		break;
510 	case IXGBE_TDBAH(0):
511 		for (i = 0; i < 64; i++)
512 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
513 		break;
514 	case IXGBE_TDLEN(0):
515 		for (i = 0; i < 64; i++)
516 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
517 		break;
518 	case IXGBE_TDH(0):
519 		for (i = 0; i < 64; i++)
520 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
521 		break;
522 	case IXGBE_TDT(0):
523 		for (i = 0; i < 64; i++)
524 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
525 		break;
526 	case IXGBE_TXDCTL(0):
527 		for (i = 0; i < 64; i++)
528 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
529 		break;
530 	default:
531 		pr_info("%-15s %08x\n",
532 			reginfo->name, IXGBE_READ_REG(hw, reginfo->ofs));
533 		return;
534 	}
535 
536 	i = 0;
537 	while (i < 64) {
538 		int j;
539 		char buf[9 * 8 + 1];
540 		char *p = buf;
541 
542 		snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i, i + 7);
543 		for (j = 0; j < 8; j++)
544 			p += sprintf(p, " %08x", regs[i++]);
545 		pr_err("%-15s%s\n", rname, buf);
546 	}
547 
548 }
549 
550 static void ixgbe_print_buffer(struct ixgbe_ring *ring, int n)
551 {
552 	struct ixgbe_tx_buffer *tx_buffer;
553 
554 	tx_buffer = &ring->tx_buffer_info[ring->next_to_clean];
555 	pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
556 		n, ring->next_to_use, ring->next_to_clean,
557 		(u64)dma_unmap_addr(tx_buffer, dma),
558 		dma_unmap_len(tx_buffer, len),
559 		tx_buffer->next_to_watch,
560 		(u64)tx_buffer->time_stamp);
561 }
562 
563 /*
564  * ixgbe_dump - Print registers, tx-rings and rx-rings
565  */
566 static void ixgbe_dump(struct ixgbe_adapter *adapter)
567 {
568 	struct net_device *netdev = adapter->netdev;
569 	struct ixgbe_hw *hw = &adapter->hw;
570 	struct ixgbe_reg_info *reginfo;
571 	int n = 0;
572 	struct ixgbe_ring *ring;
573 	struct ixgbe_tx_buffer *tx_buffer;
574 	union ixgbe_adv_tx_desc *tx_desc;
575 	struct my_u0 { u64 a; u64 b; } *u0;
576 	struct ixgbe_ring *rx_ring;
577 	union ixgbe_adv_rx_desc *rx_desc;
578 	struct ixgbe_rx_buffer *rx_buffer_info;
579 	int i = 0;
580 
581 	if (!netif_msg_hw(adapter))
582 		return;
583 
584 	/* Print netdevice Info */
585 	if (netdev) {
586 		dev_info(&adapter->pdev->dev, "Net device Info\n");
587 		pr_info("Device Name     state            "
588 			"trans_start\n");
589 		pr_info("%-15s %016lX %016lX\n",
590 			netdev->name,
591 			netdev->state,
592 			dev_trans_start(netdev));
593 	}
594 
595 	/* Print Registers */
596 	dev_info(&adapter->pdev->dev, "Register Dump\n");
597 	pr_info(" Register Name   Value\n");
598 	for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
599 	     reginfo->name; reginfo++) {
600 		ixgbe_regdump(hw, reginfo);
601 	}
602 
603 	/* Print TX Ring Summary */
604 	if (!netdev || !netif_running(netdev))
605 		return;
606 
607 	dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
608 	pr_info(" %s     %s              %s        %s\n",
609 		"Queue [NTU] [NTC] [bi(ntc)->dma  ]",
610 		"leng", "ntw", "timestamp");
611 	for (n = 0; n < adapter->num_tx_queues; n++) {
612 		ring = adapter->tx_ring[n];
613 		ixgbe_print_buffer(ring, n);
614 	}
615 
616 	for (n = 0; n < adapter->num_xdp_queues; n++) {
617 		ring = adapter->xdp_ring[n];
618 		ixgbe_print_buffer(ring, n);
619 	}
620 
621 	/* Print TX Rings */
622 	if (!netif_msg_tx_done(adapter))
623 		goto rx_ring_summary;
624 
625 	dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
626 
627 	/* Transmit Descriptor Formats
628 	 *
629 	 * 82598 Advanced Transmit Descriptor
630 	 *   +--------------------------------------------------------------+
631 	 * 0 |         Buffer Address [63:0]                                |
632 	 *   +--------------------------------------------------------------+
633 	 * 8 |  PAYLEN  | POPTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
634 	 *   +--------------------------------------------------------------+
635 	 *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
636 	 *
637 	 * 82598 Advanced Transmit Descriptor (Write-Back Format)
638 	 *   +--------------------------------------------------------------+
639 	 * 0 |                          RSV [63:0]                          |
640 	 *   +--------------------------------------------------------------+
641 	 * 8 |            RSV           |  STA  |          NXTSEQ           |
642 	 *   +--------------------------------------------------------------+
643 	 *   63                       36 35   32 31                         0
644 	 *
645 	 * 82599+ Advanced Transmit Descriptor
646 	 *   +--------------------------------------------------------------+
647 	 * 0 |         Buffer Address [63:0]                                |
648 	 *   +--------------------------------------------------------------+
649 	 * 8 |PAYLEN  |POPTS|CC|IDX  |STA  |DCMD  |DTYP |MAC  |RSV  |DTALEN |
650 	 *   +--------------------------------------------------------------+
651 	 *   63     46 45 40 39 38 36 35 32 31  24 23 20 19 18 17 16 15     0
652 	 *
653 	 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
654 	 *   +--------------------------------------------------------------+
655 	 * 0 |                          RSV [63:0]                          |
656 	 *   +--------------------------------------------------------------+
657 	 * 8 |            RSV           |  STA  |           RSV             |
658 	 *   +--------------------------------------------------------------+
659 	 *   63                       36 35   32 31                         0
660 	 */
661 
662 	for (n = 0; n < adapter->num_tx_queues; n++) {
663 		ring = adapter->tx_ring[n];
664 		pr_info("------------------------------------\n");
665 		pr_info("TX QUEUE INDEX = %d\n", ring->queue_index);
666 		pr_info("------------------------------------\n");
667 		pr_info("%s%s    %s              %s        %s          %s\n",
668 			"T [desc]     [address 63:0  ] ",
669 			"[PlPOIdStDDt Ln] [bi->dma       ] ",
670 			"leng", "ntw", "timestamp", "bi->skb");
671 
672 		for (i = 0; ring->desc && (i < ring->count); i++) {
673 			tx_desc = IXGBE_TX_DESC(ring, i);
674 			tx_buffer = &ring->tx_buffer_info[i];
675 			u0 = (struct my_u0 *)tx_desc;
676 			if (dma_unmap_len(tx_buffer, len) > 0) {
677 				const char *ring_desc;
678 
679 				if (i == ring->next_to_use &&
680 				    i == ring->next_to_clean)
681 					ring_desc = " NTC/U";
682 				else if (i == ring->next_to_use)
683 					ring_desc = " NTU";
684 				else if (i == ring->next_to_clean)
685 					ring_desc = " NTC";
686 				else
687 					ring_desc = "";
688 				pr_info("T [0x%03X]    %016llX %016llX %016llX %08X %p %016llX %p%s",
689 					i,
690 					le64_to_cpu((__force __le64)u0->a),
691 					le64_to_cpu((__force __le64)u0->b),
692 					(u64)dma_unmap_addr(tx_buffer, dma),
693 					dma_unmap_len(tx_buffer, len),
694 					tx_buffer->next_to_watch,
695 					(u64)tx_buffer->time_stamp,
696 					tx_buffer->skb,
697 					ring_desc);
698 
699 				if (netif_msg_pktdata(adapter) &&
700 				    tx_buffer->skb)
701 					print_hex_dump(KERN_INFO, "",
702 						DUMP_PREFIX_ADDRESS, 16, 1,
703 						tx_buffer->skb->data,
704 						dma_unmap_len(tx_buffer, len),
705 						true);
706 			}
707 		}
708 	}
709 
710 	/* Print RX Rings Summary */
711 rx_ring_summary:
712 	dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
713 	pr_info("Queue [NTU] [NTC]\n");
714 	for (n = 0; n < adapter->num_rx_queues; n++) {
715 		rx_ring = adapter->rx_ring[n];
716 		pr_info("%5d %5X %5X\n",
717 			n, rx_ring->next_to_use, rx_ring->next_to_clean);
718 	}
719 
720 	/* Print RX Rings */
721 	if (!netif_msg_rx_status(adapter))
722 		return;
723 
724 	dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
725 
726 	/* Receive Descriptor Formats
727 	 *
728 	 * 82598 Advanced Receive Descriptor (Read) Format
729 	 *    63                                           1        0
730 	 *    +-----------------------------------------------------+
731 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
732 	 *    +----------------------------------------------+------+
733 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
734 	 *    +-----------------------------------------------------+
735 	 *
736 	 *
737 	 * 82598 Advanced Receive Descriptor (Write-Back) Format
738 	 *
739 	 *   63       48 47    32 31  30      21 20 16 15   4 3     0
740 	 *   +------------------------------------------------------+
741 	 * 0 |       RSS Hash /  |SPH| HDR_LEN  | RSV |Packet|  RSS |
742 	 *   | Packet   | IP     |   |          |     | Type | Type |
743 	 *   | Checksum | Ident  |   |          |     |      |      |
744 	 *   +------------------------------------------------------+
745 	 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
746 	 *   +------------------------------------------------------+
747 	 *   63       48 47    32 31            20 19               0
748 	 *
749 	 * 82599+ Advanced Receive Descriptor (Read) Format
750 	 *    63                                           1        0
751 	 *    +-----------------------------------------------------+
752 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
753 	 *    +----------------------------------------------+------+
754 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
755 	 *    +-----------------------------------------------------+
756 	 *
757 	 *
758 	 * 82599+ Advanced Receive Descriptor (Write-Back) Format
759 	 *
760 	 *   63       48 47    32 31  30      21 20 17 16   4 3     0
761 	 *   +------------------------------------------------------+
762 	 * 0 |RSS / Frag Checksum|SPH| HDR_LEN  |RSC- |Packet|  RSS |
763 	 *   |/ RTT / PCoE_PARAM |   |          | CNT | Type | Type |
764 	 *   |/ Flow Dir Flt ID  |   |          |     |      |      |
765 	 *   +------------------------------------------------------+
766 	 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
767 	 *   +------------------------------------------------------+
768 	 *   63       48 47    32 31          20 19                 0
769 	 */
770 
771 	for (n = 0; n < adapter->num_rx_queues; n++) {
772 		rx_ring = adapter->rx_ring[n];
773 		pr_info("------------------------------------\n");
774 		pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
775 		pr_info("------------------------------------\n");
776 		pr_info("%s%s%s\n",
777 			"R  [desc]      [ PktBuf     A0] ",
778 			"[  HeadBuf   DD] [bi->dma       ] [bi->skb       ] ",
779 			"<-- Adv Rx Read format");
780 		pr_info("%s%s%s\n",
781 			"RWB[desc]      [PcsmIpSHl PtRs] ",
782 			"[vl er S cks ln] ---------------- [bi->skb       ] ",
783 			"<-- Adv Rx Write-Back format");
784 
785 		for (i = 0; i < rx_ring->count; i++) {
786 			const char *ring_desc;
787 
788 			if (i == rx_ring->next_to_use)
789 				ring_desc = " NTU";
790 			else if (i == rx_ring->next_to_clean)
791 				ring_desc = " NTC";
792 			else
793 				ring_desc = "";
794 
795 			rx_buffer_info = &rx_ring->rx_buffer_info[i];
796 			rx_desc = IXGBE_RX_DESC(rx_ring, i);
797 			u0 = (struct my_u0 *)rx_desc;
798 			if (rx_desc->wb.upper.length) {
799 				/* Descriptor Done */
800 				pr_info("RWB[0x%03X]     %016llX %016llX ---------------- %p%s\n",
801 					i,
802 					le64_to_cpu((__force __le64)u0->a),
803 					le64_to_cpu((__force __le64)u0->b),
804 					rx_buffer_info->skb,
805 					ring_desc);
806 			} else {
807 				pr_info("R  [0x%03X]     %016llX %016llX %016llX %p%s\n",
808 					i,
809 					le64_to_cpu((__force __le64)u0->a),
810 					le64_to_cpu((__force __le64)u0->b),
811 					(u64)rx_buffer_info->dma,
812 					rx_buffer_info->skb,
813 					ring_desc);
814 
815 				if (netif_msg_pktdata(adapter) &&
816 				    rx_buffer_info->dma) {
817 					print_hex_dump(KERN_INFO, "",
818 					   DUMP_PREFIX_ADDRESS, 16, 1,
819 					   page_address(rx_buffer_info->page) +
820 						    rx_buffer_info->page_offset,
821 					   ixgbe_rx_bufsz(rx_ring), true);
822 				}
823 			}
824 		}
825 	}
826 }
827 
828 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
829 {
830 	u32 ctrl_ext;
831 
832 	/* Let firmware take over control of h/w */
833 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
834 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
835 			ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
836 }
837 
838 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
839 {
840 	u32 ctrl_ext;
841 
842 	/* Let firmware know the driver has taken over */
843 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
844 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
845 			ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
846 }
847 
848 /**
849  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
850  * @adapter: pointer to adapter struct
851  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
852  * @queue: queue to map the corresponding interrupt to
853  * @msix_vector: the vector to map to the corresponding queue
854  *
855  */
856 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
857 			   u8 queue, u8 msix_vector)
858 {
859 	u32 ivar, index;
860 	struct ixgbe_hw *hw = &adapter->hw;
861 	switch (hw->mac.type) {
862 	case ixgbe_mac_82598EB:
863 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
864 		if (direction == -1)
865 			direction = 0;
866 		index = (((direction * 64) + queue) >> 2) & 0x1F;
867 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
868 		ivar &= ~(0xFF << (8 * (queue & 0x3)));
869 		ivar |= (msix_vector << (8 * (queue & 0x3)));
870 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
871 		break;
872 	case ixgbe_mac_82599EB:
873 	case ixgbe_mac_X540:
874 	case ixgbe_mac_X550:
875 	case ixgbe_mac_X550EM_x:
876 	case ixgbe_mac_x550em_a:
877 		if (direction == -1) {
878 			/* other causes */
879 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
880 			index = ((queue & 1) * 8);
881 			ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
882 			ivar &= ~(0xFF << index);
883 			ivar |= (msix_vector << index);
884 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
885 			break;
886 		} else {
887 			/* tx or rx causes */
888 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
889 			index = ((16 * (queue & 1)) + (8 * direction));
890 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
891 			ivar &= ~(0xFF << index);
892 			ivar |= (msix_vector << index);
893 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
894 			break;
895 		}
896 	default:
897 		break;
898 	}
899 }
900 
901 void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
902 			    u64 qmask)
903 {
904 	u32 mask;
905 
906 	switch (adapter->hw.mac.type) {
907 	case ixgbe_mac_82598EB:
908 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
909 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
910 		break;
911 	case ixgbe_mac_82599EB:
912 	case ixgbe_mac_X540:
913 	case ixgbe_mac_X550:
914 	case ixgbe_mac_X550EM_x:
915 	case ixgbe_mac_x550em_a:
916 		mask = (qmask & 0xFFFFFFFF);
917 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
918 		mask = (qmask >> 32);
919 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
920 		break;
921 	default:
922 		break;
923 	}
924 }
925 
926 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
927 {
928 	struct ixgbe_hw *hw = &adapter->hw;
929 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
930 	int i;
931 	u32 data;
932 
933 	if ((hw->fc.current_mode != ixgbe_fc_full) &&
934 	    (hw->fc.current_mode != ixgbe_fc_rx_pause))
935 		return;
936 
937 	switch (hw->mac.type) {
938 	case ixgbe_mac_82598EB:
939 		data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
940 		break;
941 	default:
942 		data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
943 	}
944 	hwstats->lxoffrxc += data;
945 
946 	/* refill credits (no tx hang) if we received xoff */
947 	if (!data)
948 		return;
949 
950 	for (i = 0; i < adapter->num_tx_queues; i++)
951 		clear_bit(__IXGBE_HANG_CHECK_ARMED,
952 			  &adapter->tx_ring[i]->state);
953 
954 	for (i = 0; i < adapter->num_xdp_queues; i++)
955 		clear_bit(__IXGBE_HANG_CHECK_ARMED,
956 			  &adapter->xdp_ring[i]->state);
957 }
958 
959 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
960 {
961 	struct ixgbe_hw *hw = &adapter->hw;
962 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
963 	u32 xoff[8] = {0};
964 	u8 tc;
965 	int i;
966 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
967 
968 	if (adapter->ixgbe_ieee_pfc)
969 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
970 
971 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
972 		ixgbe_update_xoff_rx_lfc(adapter);
973 		return;
974 	}
975 
976 	/* update stats for each tc, only valid with PFC enabled */
977 	for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
978 		u32 pxoffrxc;
979 
980 		switch (hw->mac.type) {
981 		case ixgbe_mac_82598EB:
982 			pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
983 			break;
984 		default:
985 			pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
986 		}
987 		hwstats->pxoffrxc[i] += pxoffrxc;
988 		/* Get the TC for given UP */
989 		tc = netdev_get_prio_tc_map(adapter->netdev, i);
990 		xoff[tc] += pxoffrxc;
991 	}
992 
993 	/* disarm tx queues that have received xoff frames */
994 	for (i = 0; i < adapter->num_tx_queues; i++) {
995 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
996 
997 		tc = tx_ring->dcb_tc;
998 		if (xoff[tc])
999 			clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1000 	}
1001 
1002 	for (i = 0; i < adapter->num_xdp_queues; i++) {
1003 		struct ixgbe_ring *xdp_ring = adapter->xdp_ring[i];
1004 
1005 		tc = xdp_ring->dcb_tc;
1006 		if (xoff[tc])
1007 			clear_bit(__IXGBE_HANG_CHECK_ARMED, &xdp_ring->state);
1008 	}
1009 }
1010 
1011 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
1012 {
1013 	return ring->stats.packets;
1014 }
1015 
1016 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
1017 {
1018 	unsigned int head, tail;
1019 
1020 	head = ring->next_to_clean;
1021 	tail = ring->next_to_use;
1022 
1023 	return ((head <= tail) ? tail : tail + ring->count) - head;
1024 }
1025 
1026 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
1027 {
1028 	u32 tx_done = ixgbe_get_tx_completed(tx_ring);
1029 	u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
1030 	u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
1031 
1032 	clear_check_for_tx_hang(tx_ring);
1033 
1034 	/*
1035 	 * Check for a hung queue, but be thorough. This verifies
1036 	 * that a transmit has been completed since the previous
1037 	 * check AND there is at least one packet pending. The
1038 	 * ARMED bit is set to indicate a potential hang. The
1039 	 * bit is cleared if a pause frame is received to remove
1040 	 * false hang detection due to PFC or 802.3x frames. By
1041 	 * requiring this to fail twice we avoid races with
1042 	 * pfc clearing the ARMED bit and conditions where we
1043 	 * run the check_tx_hang logic with a transmit completion
1044 	 * pending but without time to complete it yet.
1045 	 */
1046 	if (tx_done_old == tx_done && tx_pending)
1047 		/* make sure it is true for two checks in a row */
1048 		return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
1049 					&tx_ring->state);
1050 	/* update completed stats and continue */
1051 	tx_ring->tx_stats.tx_done_old = tx_done;
1052 	/* reset the countdown */
1053 	clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1054 
1055 	return false;
1056 }
1057 
1058 /**
1059  * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1060  * @adapter: driver private struct
1061  **/
1062 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
1063 {
1064 
1065 	/* Do the reset outside of interrupt context */
1066 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1067 		set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
1068 		e_warn(drv, "initiating reset due to tx timeout\n");
1069 		ixgbe_service_event_schedule(adapter);
1070 	}
1071 }
1072 
1073 /**
1074  * ixgbe_tx_maxrate - callback to set the maximum per-queue bitrate
1075  * @netdev: network interface device structure
1076  * @queue_index: Tx queue to set
1077  * @maxrate: desired maximum transmit bitrate
1078  **/
1079 static int ixgbe_tx_maxrate(struct net_device *netdev,
1080 			    int queue_index, u32 maxrate)
1081 {
1082 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
1083 	struct ixgbe_hw *hw = &adapter->hw;
1084 	u32 bcnrc_val = ixgbe_link_mbps(adapter);
1085 
1086 	if (!maxrate)
1087 		return 0;
1088 
1089 	/* Calculate the rate factor values to set */
1090 	bcnrc_val <<= IXGBE_RTTBCNRC_RF_INT_SHIFT;
1091 	bcnrc_val /= maxrate;
1092 
1093 	/* clear everything but the rate factor */
1094 	bcnrc_val &= IXGBE_RTTBCNRC_RF_INT_MASK |
1095 	IXGBE_RTTBCNRC_RF_DEC_MASK;
1096 
1097 	/* enable the rate scheduler */
1098 	bcnrc_val |= IXGBE_RTTBCNRC_RS_ENA;
1099 
1100 	IXGBE_WRITE_REG(hw, IXGBE_RTTDQSEL, queue_index);
1101 	IXGBE_WRITE_REG(hw, IXGBE_RTTBCNRC, bcnrc_val);
1102 
1103 	return 0;
1104 }
1105 
1106 /**
1107  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1108  * @q_vector: structure containing interrupt and ring information
1109  * @tx_ring: tx ring to clean
1110  * @napi_budget: Used to determine if we are in netpoll
1111  **/
1112 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
1113 			       struct ixgbe_ring *tx_ring, int napi_budget)
1114 {
1115 	struct ixgbe_adapter *adapter = q_vector->adapter;
1116 	struct ixgbe_tx_buffer *tx_buffer;
1117 	union ixgbe_adv_tx_desc *tx_desc;
1118 	unsigned int total_bytes = 0, total_packets = 0, total_ipsec = 0;
1119 	unsigned int budget = q_vector->tx.work_limit;
1120 	unsigned int i = tx_ring->next_to_clean;
1121 
1122 	if (test_bit(__IXGBE_DOWN, &adapter->state))
1123 		return true;
1124 
1125 	tx_buffer = &tx_ring->tx_buffer_info[i];
1126 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
1127 	i -= tx_ring->count;
1128 
1129 	do {
1130 		union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1131 
1132 		/* if next_to_watch is not set then there is no work pending */
1133 		if (!eop_desc)
1134 			break;
1135 
1136 		/* prevent any other reads prior to eop_desc */
1137 		smp_rmb();
1138 
1139 		/* if DD is not set pending work has not been completed */
1140 		if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
1141 			break;
1142 
1143 		/* clear next_to_watch to prevent false hangs */
1144 		tx_buffer->next_to_watch = NULL;
1145 
1146 		/* update the statistics for this packet */
1147 		total_bytes += tx_buffer->bytecount;
1148 		total_packets += tx_buffer->gso_segs;
1149 		if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_IPSEC)
1150 			total_ipsec++;
1151 
1152 		/* free the skb */
1153 		if (ring_is_xdp(tx_ring))
1154 			xdp_return_frame(tx_buffer->xdpf);
1155 		else
1156 			napi_consume_skb(tx_buffer->skb, napi_budget);
1157 
1158 		/* unmap skb header data */
1159 		dma_unmap_single(tx_ring->dev,
1160 				 dma_unmap_addr(tx_buffer, dma),
1161 				 dma_unmap_len(tx_buffer, len),
1162 				 DMA_TO_DEVICE);
1163 
1164 		/* clear tx_buffer data */
1165 		dma_unmap_len_set(tx_buffer, len, 0);
1166 
1167 		/* unmap remaining buffers */
1168 		while (tx_desc != eop_desc) {
1169 			tx_buffer++;
1170 			tx_desc++;
1171 			i++;
1172 			if (unlikely(!i)) {
1173 				i -= tx_ring->count;
1174 				tx_buffer = tx_ring->tx_buffer_info;
1175 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1176 			}
1177 
1178 			/* unmap any remaining paged data */
1179 			if (dma_unmap_len(tx_buffer, len)) {
1180 				dma_unmap_page(tx_ring->dev,
1181 					       dma_unmap_addr(tx_buffer, dma),
1182 					       dma_unmap_len(tx_buffer, len),
1183 					       DMA_TO_DEVICE);
1184 				dma_unmap_len_set(tx_buffer, len, 0);
1185 			}
1186 		}
1187 
1188 		/* move us one more past the eop_desc for start of next pkt */
1189 		tx_buffer++;
1190 		tx_desc++;
1191 		i++;
1192 		if (unlikely(!i)) {
1193 			i -= tx_ring->count;
1194 			tx_buffer = tx_ring->tx_buffer_info;
1195 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1196 		}
1197 
1198 		/* issue prefetch for next Tx descriptor */
1199 		prefetch(tx_desc);
1200 
1201 		/* update budget accounting */
1202 		budget--;
1203 	} while (likely(budget));
1204 
1205 	i += tx_ring->count;
1206 	tx_ring->next_to_clean = i;
1207 	u64_stats_update_begin(&tx_ring->syncp);
1208 	tx_ring->stats.bytes += total_bytes;
1209 	tx_ring->stats.packets += total_packets;
1210 	u64_stats_update_end(&tx_ring->syncp);
1211 	q_vector->tx.total_bytes += total_bytes;
1212 	q_vector->tx.total_packets += total_packets;
1213 	adapter->tx_ipsec += total_ipsec;
1214 
1215 	if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1216 		/* schedule immediate reset if we believe we hung */
1217 		struct ixgbe_hw *hw = &adapter->hw;
1218 		e_err(drv, "Detected Tx Unit Hang %s\n"
1219 			"  Tx Queue             <%d>\n"
1220 			"  TDH, TDT             <%x>, <%x>\n"
1221 			"  next_to_use          <%x>\n"
1222 			"  next_to_clean        <%x>\n"
1223 			"tx_buffer_info[next_to_clean]\n"
1224 			"  time_stamp           <%lx>\n"
1225 			"  jiffies              <%lx>\n",
1226 			ring_is_xdp(tx_ring) ? "(XDP)" : "",
1227 			tx_ring->queue_index,
1228 			IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1229 			IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1230 			tx_ring->next_to_use, i,
1231 			tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1232 
1233 		if (!ring_is_xdp(tx_ring))
1234 			netif_stop_subqueue(tx_ring->netdev,
1235 					    tx_ring->queue_index);
1236 
1237 		e_info(probe,
1238 		       "tx hang %d detected on queue %d, resetting adapter\n",
1239 			adapter->tx_timeout_count + 1, tx_ring->queue_index);
1240 
1241 		/* schedule immediate reset if we believe we hung */
1242 		ixgbe_tx_timeout_reset(adapter);
1243 
1244 		/* the adapter is about to reset, no point in enabling stuff */
1245 		return true;
1246 	}
1247 
1248 	if (ring_is_xdp(tx_ring))
1249 		return !!budget;
1250 
1251 	netdev_tx_completed_queue(txring_txq(tx_ring),
1252 				  total_packets, total_bytes);
1253 
1254 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1255 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1256 		     (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1257 		/* Make sure that anybody stopping the queue after this
1258 		 * sees the new next_to_clean.
1259 		 */
1260 		smp_mb();
1261 		if (__netif_subqueue_stopped(tx_ring->netdev,
1262 					     tx_ring->queue_index)
1263 		    && !test_bit(__IXGBE_DOWN, &adapter->state)) {
1264 			netif_wake_subqueue(tx_ring->netdev,
1265 					    tx_ring->queue_index);
1266 			++tx_ring->tx_stats.restart_queue;
1267 		}
1268 	}
1269 
1270 	return !!budget;
1271 }
1272 
1273 #ifdef CONFIG_IXGBE_DCA
1274 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1275 				struct ixgbe_ring *tx_ring,
1276 				int cpu)
1277 {
1278 	struct ixgbe_hw *hw = &adapter->hw;
1279 	u32 txctrl = 0;
1280 	u16 reg_offset;
1281 
1282 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1283 		txctrl = dca3_get_tag(tx_ring->dev, cpu);
1284 
1285 	switch (hw->mac.type) {
1286 	case ixgbe_mac_82598EB:
1287 		reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1288 		break;
1289 	case ixgbe_mac_82599EB:
1290 	case ixgbe_mac_X540:
1291 		reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1292 		txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1293 		break;
1294 	default:
1295 		/* for unknown hardware do not write register */
1296 		return;
1297 	}
1298 
1299 	/*
1300 	 * We can enable relaxed ordering for reads, but not writes when
1301 	 * DCA is enabled.  This is due to a known issue in some chipsets
1302 	 * which will cause the DCA tag to be cleared.
1303 	 */
1304 	txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1305 		  IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1306 		  IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1307 
1308 	IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1309 }
1310 
1311 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1312 				struct ixgbe_ring *rx_ring,
1313 				int cpu)
1314 {
1315 	struct ixgbe_hw *hw = &adapter->hw;
1316 	u32 rxctrl = 0;
1317 	u8 reg_idx = rx_ring->reg_idx;
1318 
1319 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1320 		rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1321 
1322 	switch (hw->mac.type) {
1323 	case ixgbe_mac_82599EB:
1324 	case ixgbe_mac_X540:
1325 		rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1326 		break;
1327 	default:
1328 		break;
1329 	}
1330 
1331 	/*
1332 	 * We can enable relaxed ordering for reads, but not writes when
1333 	 * DCA is enabled.  This is due to a known issue in some chipsets
1334 	 * which will cause the DCA tag to be cleared.
1335 	 */
1336 	rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1337 		  IXGBE_DCA_RXCTRL_DATA_DCA_EN |
1338 		  IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1339 
1340 	IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1341 }
1342 
1343 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1344 {
1345 	struct ixgbe_adapter *adapter = q_vector->adapter;
1346 	struct ixgbe_ring *ring;
1347 	int cpu = get_cpu();
1348 
1349 	if (q_vector->cpu == cpu)
1350 		goto out_no_update;
1351 
1352 	ixgbe_for_each_ring(ring, q_vector->tx)
1353 		ixgbe_update_tx_dca(adapter, ring, cpu);
1354 
1355 	ixgbe_for_each_ring(ring, q_vector->rx)
1356 		ixgbe_update_rx_dca(adapter, ring, cpu);
1357 
1358 	q_vector->cpu = cpu;
1359 out_no_update:
1360 	put_cpu();
1361 }
1362 
1363 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1364 {
1365 	int i;
1366 
1367 	/* always use CB2 mode, difference is masked in the CB driver */
1368 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1369 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1370 				IXGBE_DCA_CTRL_DCA_MODE_CB2);
1371 	else
1372 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1373 				IXGBE_DCA_CTRL_DCA_DISABLE);
1374 
1375 	for (i = 0; i < adapter->num_q_vectors; i++) {
1376 		adapter->q_vector[i]->cpu = -1;
1377 		ixgbe_update_dca(adapter->q_vector[i]);
1378 	}
1379 }
1380 
1381 static int __ixgbe_notify_dca(struct device *dev, void *data)
1382 {
1383 	struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1384 	unsigned long event = *(unsigned long *)data;
1385 
1386 	if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1387 		return 0;
1388 
1389 	switch (event) {
1390 	case DCA_PROVIDER_ADD:
1391 		/* if we're already enabled, don't do it again */
1392 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1393 			break;
1394 		if (dca_add_requester(dev) == 0) {
1395 			adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1396 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1397 					IXGBE_DCA_CTRL_DCA_MODE_CB2);
1398 			break;
1399 		}
1400 		/* fall through - DCA is disabled. */
1401 	case DCA_PROVIDER_REMOVE:
1402 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1403 			dca_remove_requester(dev);
1404 			adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1405 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1406 					IXGBE_DCA_CTRL_DCA_DISABLE);
1407 		}
1408 		break;
1409 	}
1410 
1411 	return 0;
1412 }
1413 
1414 #endif /* CONFIG_IXGBE_DCA */
1415 
1416 #define IXGBE_RSS_L4_TYPES_MASK \
1417 	((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
1418 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
1419 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
1420 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
1421 
1422 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1423 				 union ixgbe_adv_rx_desc *rx_desc,
1424 				 struct sk_buff *skb)
1425 {
1426 	u16 rss_type;
1427 
1428 	if (!(ring->netdev->features & NETIF_F_RXHASH))
1429 		return;
1430 
1431 	rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
1432 		   IXGBE_RXDADV_RSSTYPE_MASK;
1433 
1434 	if (!rss_type)
1435 		return;
1436 
1437 	skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1438 		     (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
1439 		     PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1440 }
1441 
1442 #ifdef IXGBE_FCOE
1443 /**
1444  * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1445  * @ring: structure containing ring specific data
1446  * @rx_desc: advanced rx descriptor
1447  *
1448  * Returns : true if it is FCoE pkt
1449  */
1450 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1451 				    union ixgbe_adv_rx_desc *rx_desc)
1452 {
1453 	__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1454 
1455 	return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1456 	       ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1457 		(cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1458 			     IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1459 }
1460 
1461 #endif /* IXGBE_FCOE */
1462 /**
1463  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1464  * @ring: structure containing ring specific data
1465  * @rx_desc: current Rx descriptor being processed
1466  * @skb: skb currently being received and modified
1467  **/
1468 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1469 				     union ixgbe_adv_rx_desc *rx_desc,
1470 				     struct sk_buff *skb)
1471 {
1472 	__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1473 	bool encap_pkt = false;
1474 
1475 	skb_checksum_none_assert(skb);
1476 
1477 	/* Rx csum disabled */
1478 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
1479 		return;
1480 
1481 	/* check for VXLAN and Geneve packets */
1482 	if (pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_VXLAN)) {
1483 		encap_pkt = true;
1484 		skb->encapsulation = 1;
1485 	}
1486 
1487 	/* if IP and error */
1488 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1489 	    ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1490 		ring->rx_stats.csum_err++;
1491 		return;
1492 	}
1493 
1494 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1495 		return;
1496 
1497 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1498 		/*
1499 		 * 82599 errata, UDP frames with a 0 checksum can be marked as
1500 		 * checksum errors.
1501 		 */
1502 		if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1503 		    test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1504 			return;
1505 
1506 		ring->rx_stats.csum_err++;
1507 		return;
1508 	}
1509 
1510 	/* It must be a TCP or UDP packet with a valid checksum */
1511 	skb->ip_summed = CHECKSUM_UNNECESSARY;
1512 	if (encap_pkt) {
1513 		if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_OUTERIPCS))
1514 			return;
1515 
1516 		if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_OUTERIPER)) {
1517 			skb->ip_summed = CHECKSUM_NONE;
1518 			return;
1519 		}
1520 		/* If we checked the outer header let the stack know */
1521 		skb->csum_level = 1;
1522 	}
1523 }
1524 
1525 static inline unsigned int ixgbe_rx_offset(struct ixgbe_ring *rx_ring)
1526 {
1527 	return ring_uses_build_skb(rx_ring) ? IXGBE_SKB_PAD : 0;
1528 }
1529 
1530 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1531 				    struct ixgbe_rx_buffer *bi)
1532 {
1533 	struct page *page = bi->page;
1534 	dma_addr_t dma;
1535 
1536 	/* since we are recycling buffers we should seldom need to alloc */
1537 	if (likely(page))
1538 		return true;
1539 
1540 	/* alloc new page for storage */
1541 	page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
1542 	if (unlikely(!page)) {
1543 		rx_ring->rx_stats.alloc_rx_page_failed++;
1544 		return false;
1545 	}
1546 
1547 	/* map page for use */
1548 	dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1549 				 ixgbe_rx_pg_size(rx_ring),
1550 				 DMA_FROM_DEVICE,
1551 				 IXGBE_RX_DMA_ATTR);
1552 
1553 	/*
1554 	 * if mapping failed free memory back to system since
1555 	 * there isn't much point in holding memory we can't use
1556 	 */
1557 	if (dma_mapping_error(rx_ring->dev, dma)) {
1558 		__free_pages(page, ixgbe_rx_pg_order(rx_ring));
1559 
1560 		rx_ring->rx_stats.alloc_rx_page_failed++;
1561 		return false;
1562 	}
1563 
1564 	bi->dma = dma;
1565 	bi->page = page;
1566 	bi->page_offset = ixgbe_rx_offset(rx_ring);
1567 	page_ref_add(page, USHRT_MAX - 1);
1568 	bi->pagecnt_bias = USHRT_MAX;
1569 	rx_ring->rx_stats.alloc_rx_page++;
1570 
1571 	return true;
1572 }
1573 
1574 /**
1575  * ixgbe_alloc_rx_buffers - Replace used receive buffers
1576  * @rx_ring: ring to place buffers on
1577  * @cleaned_count: number of buffers to replace
1578  **/
1579 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1580 {
1581 	union ixgbe_adv_rx_desc *rx_desc;
1582 	struct ixgbe_rx_buffer *bi;
1583 	u16 i = rx_ring->next_to_use;
1584 	u16 bufsz;
1585 
1586 	/* nothing to do */
1587 	if (!cleaned_count)
1588 		return;
1589 
1590 	rx_desc = IXGBE_RX_DESC(rx_ring, i);
1591 	bi = &rx_ring->rx_buffer_info[i];
1592 	i -= rx_ring->count;
1593 
1594 	bufsz = ixgbe_rx_bufsz(rx_ring);
1595 
1596 	do {
1597 		if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1598 			break;
1599 
1600 		/* sync the buffer for use by the device */
1601 		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1602 						 bi->page_offset, bufsz,
1603 						 DMA_FROM_DEVICE);
1604 
1605 		/*
1606 		 * Refresh the desc even if buffer_addrs didn't change
1607 		 * because each write-back erases this info.
1608 		 */
1609 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1610 
1611 		rx_desc++;
1612 		bi++;
1613 		i++;
1614 		if (unlikely(!i)) {
1615 			rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1616 			bi = rx_ring->rx_buffer_info;
1617 			i -= rx_ring->count;
1618 		}
1619 
1620 		/* clear the length for the next_to_use descriptor */
1621 		rx_desc->wb.upper.length = 0;
1622 
1623 		cleaned_count--;
1624 	} while (cleaned_count);
1625 
1626 	i += rx_ring->count;
1627 
1628 	if (rx_ring->next_to_use != i) {
1629 		rx_ring->next_to_use = i;
1630 
1631 		/* update next to alloc since we have filled the ring */
1632 		rx_ring->next_to_alloc = i;
1633 
1634 		/* Force memory writes to complete before letting h/w
1635 		 * know there are new descriptors to fetch.  (Only
1636 		 * applicable for weak-ordered memory model archs,
1637 		 * such as IA-64).
1638 		 */
1639 		wmb();
1640 		writel(i, rx_ring->tail);
1641 	}
1642 }
1643 
1644 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1645 				   struct sk_buff *skb)
1646 {
1647 	u16 hdr_len = skb_headlen(skb);
1648 
1649 	/* set gso_size to avoid messing up TCP MSS */
1650 	skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1651 						 IXGBE_CB(skb)->append_cnt);
1652 	skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1653 }
1654 
1655 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1656 				   struct sk_buff *skb)
1657 {
1658 	/* if append_cnt is 0 then frame is not RSC */
1659 	if (!IXGBE_CB(skb)->append_cnt)
1660 		return;
1661 
1662 	rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1663 	rx_ring->rx_stats.rsc_flush++;
1664 
1665 	ixgbe_set_rsc_gso_size(rx_ring, skb);
1666 
1667 	/* gso_size is computed using append_cnt so always clear it last */
1668 	IXGBE_CB(skb)->append_cnt = 0;
1669 }
1670 
1671 /**
1672  * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1673  * @rx_ring: rx descriptor ring packet is being transacted on
1674  * @rx_desc: pointer to the EOP Rx descriptor
1675  * @skb: pointer to current skb being populated
1676  *
1677  * This function checks the ring, descriptor, and packet information in
1678  * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1679  * other fields within the skb.
1680  **/
1681 void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1682 			      union ixgbe_adv_rx_desc *rx_desc,
1683 			      struct sk_buff *skb)
1684 {
1685 	struct net_device *dev = rx_ring->netdev;
1686 	u32 flags = rx_ring->q_vector->adapter->flags;
1687 
1688 	ixgbe_update_rsc_stats(rx_ring, skb);
1689 
1690 	ixgbe_rx_hash(rx_ring, rx_desc, skb);
1691 
1692 	ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1693 
1694 	if (unlikely(flags & IXGBE_FLAG_RX_HWTSTAMP_ENABLED))
1695 		ixgbe_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
1696 
1697 	if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1698 	    ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1699 		u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1700 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1701 	}
1702 
1703 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_STAT_SECP))
1704 		ixgbe_ipsec_rx(rx_ring, rx_desc, skb);
1705 
1706 	/* record Rx queue, or update MACVLAN statistics */
1707 	if (netif_is_ixgbe(dev))
1708 		skb_record_rx_queue(skb, rx_ring->queue_index);
1709 	else
1710 		macvlan_count_rx(netdev_priv(dev), skb->len + ETH_HLEN, true,
1711 				 false);
1712 
1713 	skb->protocol = eth_type_trans(skb, dev);
1714 }
1715 
1716 void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1717 		  struct sk_buff *skb)
1718 {
1719 	napi_gro_receive(&q_vector->napi, skb);
1720 }
1721 
1722 /**
1723  * ixgbe_is_non_eop - process handling of non-EOP buffers
1724  * @rx_ring: Rx ring being processed
1725  * @rx_desc: Rx descriptor for current buffer
1726  * @skb: Current socket buffer containing buffer in progress
1727  *
1728  * This function updates next to clean.  If the buffer is an EOP buffer
1729  * this function exits returning false, otherwise it will place the
1730  * sk_buff in the next buffer to be chained and return true indicating
1731  * that this is in fact a non-EOP buffer.
1732  **/
1733 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1734 			     union ixgbe_adv_rx_desc *rx_desc,
1735 			     struct sk_buff *skb)
1736 {
1737 	u32 ntc = rx_ring->next_to_clean + 1;
1738 
1739 	/* fetch, update, and store next to clean */
1740 	ntc = (ntc < rx_ring->count) ? ntc : 0;
1741 	rx_ring->next_to_clean = ntc;
1742 
1743 	prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1744 
1745 	/* update RSC append count if present */
1746 	if (ring_is_rsc_enabled(rx_ring)) {
1747 		__le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1748 				     cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1749 
1750 		if (unlikely(rsc_enabled)) {
1751 			u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1752 
1753 			rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1754 			IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1755 
1756 			/* update ntc based on RSC value */
1757 			ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1758 			ntc &= IXGBE_RXDADV_NEXTP_MASK;
1759 			ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1760 		}
1761 	}
1762 
1763 	/* if we are the last buffer then there is nothing else to do */
1764 	if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1765 		return false;
1766 
1767 	/* place skb in next buffer to be received */
1768 	rx_ring->rx_buffer_info[ntc].skb = skb;
1769 	rx_ring->rx_stats.non_eop_descs++;
1770 
1771 	return true;
1772 }
1773 
1774 /**
1775  * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1776  * @rx_ring: rx descriptor ring packet is being transacted on
1777  * @skb: pointer to current skb being adjusted
1778  *
1779  * This function is an ixgbe specific version of __pskb_pull_tail.  The
1780  * main difference between this version and the original function is that
1781  * this function can make several assumptions about the state of things
1782  * that allow for significant optimizations versus the standard function.
1783  * As a result we can do things like drop a frag and maintain an accurate
1784  * truesize for the skb.
1785  */
1786 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1787 			    struct sk_buff *skb)
1788 {
1789 	skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
1790 	unsigned char *va;
1791 	unsigned int pull_len;
1792 
1793 	/*
1794 	 * it is valid to use page_address instead of kmap since we are
1795 	 * working with pages allocated out of the lomem pool per
1796 	 * alloc_page(GFP_ATOMIC)
1797 	 */
1798 	va = skb_frag_address(frag);
1799 
1800 	/*
1801 	 * we need the header to contain the greater of either ETH_HLEN or
1802 	 * 60 bytes if the skb->len is less than 60 for skb_pad.
1803 	 */
1804 	pull_len = eth_get_headlen(skb->dev, va, IXGBE_RX_HDR_SIZE);
1805 
1806 	/* align pull length to size of long to optimize memcpy performance */
1807 	skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1808 
1809 	/* update all of the pointers */
1810 	skb_frag_size_sub(frag, pull_len);
1811 	skb_frag_off_add(frag, pull_len);
1812 	skb->data_len -= pull_len;
1813 	skb->tail += pull_len;
1814 }
1815 
1816 /**
1817  * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1818  * @rx_ring: rx descriptor ring packet is being transacted on
1819  * @skb: pointer to current skb being updated
1820  *
1821  * This function provides a basic DMA sync up for the first fragment of an
1822  * skb.  The reason for doing this is that the first fragment cannot be
1823  * unmapped until we have reached the end of packet descriptor for a buffer
1824  * chain.
1825  */
1826 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1827 				struct sk_buff *skb)
1828 {
1829 	if (ring_uses_build_skb(rx_ring)) {
1830 		unsigned long offset = (unsigned long)(skb->data) & ~PAGE_MASK;
1831 
1832 		dma_sync_single_range_for_cpu(rx_ring->dev,
1833 					      IXGBE_CB(skb)->dma,
1834 					      offset,
1835 					      skb_headlen(skb),
1836 					      DMA_FROM_DEVICE);
1837 	} else {
1838 		skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
1839 
1840 		dma_sync_single_range_for_cpu(rx_ring->dev,
1841 					      IXGBE_CB(skb)->dma,
1842 					      skb_frag_off(frag),
1843 					      skb_frag_size(frag),
1844 					      DMA_FROM_DEVICE);
1845 	}
1846 
1847 	/* If the page was released, just unmap it. */
1848 	if (unlikely(IXGBE_CB(skb)->page_released)) {
1849 		dma_unmap_page_attrs(rx_ring->dev, IXGBE_CB(skb)->dma,
1850 				     ixgbe_rx_pg_size(rx_ring),
1851 				     DMA_FROM_DEVICE,
1852 				     IXGBE_RX_DMA_ATTR);
1853 	}
1854 }
1855 
1856 /**
1857  * ixgbe_cleanup_headers - Correct corrupted or empty headers
1858  * @rx_ring: rx descriptor ring packet is being transacted on
1859  * @rx_desc: pointer to the EOP Rx descriptor
1860  * @skb: pointer to current skb being fixed
1861  *
1862  * Check if the skb is valid in the XDP case it will be an error pointer.
1863  * Return true in this case to abort processing and advance to next
1864  * descriptor.
1865  *
1866  * Check for corrupted packet headers caused by senders on the local L2
1867  * embedded NIC switch not setting up their Tx Descriptors right.  These
1868  * should be very rare.
1869  *
1870  * Also address the case where we are pulling data in on pages only
1871  * and as such no data is present in the skb header.
1872  *
1873  * In addition if skb is not at least 60 bytes we need to pad it so that
1874  * it is large enough to qualify as a valid Ethernet frame.
1875  *
1876  * Returns true if an error was encountered and skb was freed.
1877  **/
1878 bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1879 			   union ixgbe_adv_rx_desc *rx_desc,
1880 			   struct sk_buff *skb)
1881 {
1882 	struct net_device *netdev = rx_ring->netdev;
1883 
1884 	/* XDP packets use error pointer so abort at this point */
1885 	if (IS_ERR(skb))
1886 		return true;
1887 
1888 	/* Verify netdev is present, and that packet does not have any
1889 	 * errors that would be unacceptable to the netdev.
1890 	 */
1891 	if (!netdev ||
1892 	    (unlikely(ixgbe_test_staterr(rx_desc,
1893 					 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1894 	     !(netdev->features & NETIF_F_RXALL)))) {
1895 		dev_kfree_skb_any(skb);
1896 		return true;
1897 	}
1898 
1899 	/* place header in linear portion of buffer */
1900 	if (!skb_headlen(skb))
1901 		ixgbe_pull_tail(rx_ring, skb);
1902 
1903 #ifdef IXGBE_FCOE
1904 	/* do not attempt to pad FCoE Frames as this will disrupt DDP */
1905 	if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1906 		return false;
1907 
1908 #endif
1909 	/* if eth_skb_pad returns an error the skb was freed */
1910 	if (eth_skb_pad(skb))
1911 		return true;
1912 
1913 	return false;
1914 }
1915 
1916 /**
1917  * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1918  * @rx_ring: rx descriptor ring to store buffers on
1919  * @old_buff: donor buffer to have page reused
1920  *
1921  * Synchronizes page for reuse by the adapter
1922  **/
1923 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1924 				struct ixgbe_rx_buffer *old_buff)
1925 {
1926 	struct ixgbe_rx_buffer *new_buff;
1927 	u16 nta = rx_ring->next_to_alloc;
1928 
1929 	new_buff = &rx_ring->rx_buffer_info[nta];
1930 
1931 	/* update, and store next to alloc */
1932 	nta++;
1933 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1934 
1935 	/* Transfer page from old buffer to new buffer.
1936 	 * Move each member individually to avoid possible store
1937 	 * forwarding stalls and unnecessary copy of skb.
1938 	 */
1939 	new_buff->dma		= old_buff->dma;
1940 	new_buff->page		= old_buff->page;
1941 	new_buff->page_offset	= old_buff->page_offset;
1942 	new_buff->pagecnt_bias	= old_buff->pagecnt_bias;
1943 }
1944 
1945 static inline bool ixgbe_page_is_reserved(struct page *page)
1946 {
1947 	return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1948 }
1949 
1950 static bool ixgbe_can_reuse_rx_page(struct ixgbe_rx_buffer *rx_buffer)
1951 {
1952 	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1953 	struct page *page = rx_buffer->page;
1954 
1955 	/* avoid re-using remote pages */
1956 	if (unlikely(ixgbe_page_is_reserved(page)))
1957 		return false;
1958 
1959 #if (PAGE_SIZE < 8192)
1960 	/* if we are only owner of page we can reuse it */
1961 	if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
1962 		return false;
1963 #else
1964 	/* The last offset is a bit aggressive in that we assume the
1965 	 * worst case of FCoE being enabled and using a 3K buffer.
1966 	 * However this should have minimal impact as the 1K extra is
1967 	 * still less than one buffer in size.
1968 	 */
1969 #define IXGBE_LAST_OFFSET \
1970 	(SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBE_RXBUFFER_3K)
1971 	if (rx_buffer->page_offset > IXGBE_LAST_OFFSET)
1972 		return false;
1973 #endif
1974 
1975 	/* If we have drained the page fragment pool we need to update
1976 	 * the pagecnt_bias and page count so that we fully restock the
1977 	 * number of references the driver holds.
1978 	 */
1979 	if (unlikely(pagecnt_bias == 1)) {
1980 		page_ref_add(page, USHRT_MAX - 1);
1981 		rx_buffer->pagecnt_bias = USHRT_MAX;
1982 	}
1983 
1984 	return true;
1985 }
1986 
1987 /**
1988  * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1989  * @rx_ring: rx descriptor ring to transact packets on
1990  * @rx_buffer: buffer containing page to add
1991  * @skb: sk_buff to place the data into
1992  * @size: size of data in rx_buffer
1993  *
1994  * This function will add the data contained in rx_buffer->page to the skb.
1995  * This is done either through a direct copy if the data in the buffer is
1996  * less than the skb header size, otherwise it will just attach the page as
1997  * a frag to the skb.
1998  *
1999  * The function will then update the page offset if necessary and return
2000  * true if the buffer can be reused by the adapter.
2001  **/
2002 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
2003 			      struct ixgbe_rx_buffer *rx_buffer,
2004 			      struct sk_buff *skb,
2005 			      unsigned int size)
2006 {
2007 #if (PAGE_SIZE < 8192)
2008 	unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2009 #else
2010 	unsigned int truesize = ring_uses_build_skb(rx_ring) ?
2011 				SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) :
2012 				SKB_DATA_ALIGN(size);
2013 #endif
2014 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
2015 			rx_buffer->page_offset, size, truesize);
2016 #if (PAGE_SIZE < 8192)
2017 	rx_buffer->page_offset ^= truesize;
2018 #else
2019 	rx_buffer->page_offset += truesize;
2020 #endif
2021 }
2022 
2023 static struct ixgbe_rx_buffer *ixgbe_get_rx_buffer(struct ixgbe_ring *rx_ring,
2024 						   union ixgbe_adv_rx_desc *rx_desc,
2025 						   struct sk_buff **skb,
2026 						   const unsigned int size)
2027 {
2028 	struct ixgbe_rx_buffer *rx_buffer;
2029 
2030 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
2031 	prefetchw(rx_buffer->page);
2032 	*skb = rx_buffer->skb;
2033 
2034 	/* Delay unmapping of the first packet. It carries the header
2035 	 * information, HW may still access the header after the writeback.
2036 	 * Only unmap it when EOP is reached
2037 	 */
2038 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)) {
2039 		if (!*skb)
2040 			goto skip_sync;
2041 	} else {
2042 		if (*skb)
2043 			ixgbe_dma_sync_frag(rx_ring, *skb);
2044 	}
2045 
2046 	/* we are reusing so sync this buffer for CPU use */
2047 	dma_sync_single_range_for_cpu(rx_ring->dev,
2048 				      rx_buffer->dma,
2049 				      rx_buffer->page_offset,
2050 				      size,
2051 				      DMA_FROM_DEVICE);
2052 skip_sync:
2053 	rx_buffer->pagecnt_bias--;
2054 
2055 	return rx_buffer;
2056 }
2057 
2058 static void ixgbe_put_rx_buffer(struct ixgbe_ring *rx_ring,
2059 				struct ixgbe_rx_buffer *rx_buffer,
2060 				struct sk_buff *skb)
2061 {
2062 	if (ixgbe_can_reuse_rx_page(rx_buffer)) {
2063 		/* hand second half of page back to the ring */
2064 		ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2065 	} else {
2066 		if (!IS_ERR(skb) && IXGBE_CB(skb)->dma == rx_buffer->dma) {
2067 			/* the page has been released from the ring */
2068 			IXGBE_CB(skb)->page_released = true;
2069 		} else {
2070 			/* we are not reusing the buffer so unmap it */
2071 			dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2072 					     ixgbe_rx_pg_size(rx_ring),
2073 					     DMA_FROM_DEVICE,
2074 					     IXGBE_RX_DMA_ATTR);
2075 		}
2076 		__page_frag_cache_drain(rx_buffer->page,
2077 					rx_buffer->pagecnt_bias);
2078 	}
2079 
2080 	/* clear contents of rx_buffer */
2081 	rx_buffer->page = NULL;
2082 	rx_buffer->skb = NULL;
2083 }
2084 
2085 static struct sk_buff *ixgbe_construct_skb(struct ixgbe_ring *rx_ring,
2086 					   struct ixgbe_rx_buffer *rx_buffer,
2087 					   struct xdp_buff *xdp,
2088 					   union ixgbe_adv_rx_desc *rx_desc)
2089 {
2090 	unsigned int size = xdp->data_end - xdp->data;
2091 #if (PAGE_SIZE < 8192)
2092 	unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2093 #else
2094 	unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
2095 					       xdp->data_hard_start);
2096 #endif
2097 	struct sk_buff *skb;
2098 
2099 	/* prefetch first cache line of first page */
2100 	prefetch(xdp->data);
2101 #if L1_CACHE_BYTES < 128
2102 	prefetch(xdp->data + L1_CACHE_BYTES);
2103 #endif
2104 	/* Note, we get here by enabling legacy-rx via:
2105 	 *
2106 	 *    ethtool --set-priv-flags <dev> legacy-rx on
2107 	 *
2108 	 * In this mode, we currently get 0 extra XDP headroom as
2109 	 * opposed to having legacy-rx off, where we process XDP
2110 	 * packets going to stack via ixgbe_build_skb(). The latter
2111 	 * provides us currently with 192 bytes of headroom.
2112 	 *
2113 	 * For ixgbe_construct_skb() mode it means that the
2114 	 * xdp->data_meta will always point to xdp->data, since
2115 	 * the helper cannot expand the head. Should this ever
2116 	 * change in future for legacy-rx mode on, then lets also
2117 	 * add xdp->data_meta handling here.
2118 	 */
2119 
2120 	/* allocate a skb to store the frags */
2121 	skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBE_RX_HDR_SIZE);
2122 	if (unlikely(!skb))
2123 		return NULL;
2124 
2125 	if (size > IXGBE_RX_HDR_SIZE) {
2126 		if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2127 			IXGBE_CB(skb)->dma = rx_buffer->dma;
2128 
2129 		skb_add_rx_frag(skb, 0, rx_buffer->page,
2130 				xdp->data - page_address(rx_buffer->page),
2131 				size, truesize);
2132 #if (PAGE_SIZE < 8192)
2133 		rx_buffer->page_offset ^= truesize;
2134 #else
2135 		rx_buffer->page_offset += truesize;
2136 #endif
2137 	} else {
2138 		memcpy(__skb_put(skb, size),
2139 		       xdp->data, ALIGN(size, sizeof(long)));
2140 		rx_buffer->pagecnt_bias++;
2141 	}
2142 
2143 	return skb;
2144 }
2145 
2146 static struct sk_buff *ixgbe_build_skb(struct ixgbe_ring *rx_ring,
2147 				       struct ixgbe_rx_buffer *rx_buffer,
2148 				       struct xdp_buff *xdp,
2149 				       union ixgbe_adv_rx_desc *rx_desc)
2150 {
2151 	unsigned int metasize = xdp->data - xdp->data_meta;
2152 #if (PAGE_SIZE < 8192)
2153 	unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2154 #else
2155 	unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
2156 				SKB_DATA_ALIGN(xdp->data_end -
2157 					       xdp->data_hard_start);
2158 #endif
2159 	struct sk_buff *skb;
2160 
2161 	/* Prefetch first cache line of first page. If xdp->data_meta
2162 	 * is unused, this points extactly as xdp->data, otherwise we
2163 	 * likely have a consumer accessing first few bytes of meta
2164 	 * data, and then actual data.
2165 	 */
2166 	prefetch(xdp->data_meta);
2167 #if L1_CACHE_BYTES < 128
2168 	prefetch(xdp->data_meta + L1_CACHE_BYTES);
2169 #endif
2170 
2171 	/* build an skb to around the page buffer */
2172 	skb = build_skb(xdp->data_hard_start, truesize);
2173 	if (unlikely(!skb))
2174 		return NULL;
2175 
2176 	/* update pointers within the skb to store the data */
2177 	skb_reserve(skb, xdp->data - xdp->data_hard_start);
2178 	__skb_put(skb, xdp->data_end - xdp->data);
2179 	if (metasize)
2180 		skb_metadata_set(skb, metasize);
2181 
2182 	/* record DMA address if this is the start of a chain of buffers */
2183 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2184 		IXGBE_CB(skb)->dma = rx_buffer->dma;
2185 
2186 	/* update buffer offset */
2187 #if (PAGE_SIZE < 8192)
2188 	rx_buffer->page_offset ^= truesize;
2189 #else
2190 	rx_buffer->page_offset += truesize;
2191 #endif
2192 
2193 	return skb;
2194 }
2195 
2196 static struct sk_buff *ixgbe_run_xdp(struct ixgbe_adapter *adapter,
2197 				     struct ixgbe_ring *rx_ring,
2198 				     struct xdp_buff *xdp)
2199 {
2200 	int err, result = IXGBE_XDP_PASS;
2201 	struct bpf_prog *xdp_prog;
2202 	struct xdp_frame *xdpf;
2203 	u32 act;
2204 
2205 	rcu_read_lock();
2206 	xdp_prog = READ_ONCE(rx_ring->xdp_prog);
2207 
2208 	if (!xdp_prog)
2209 		goto xdp_out;
2210 
2211 	prefetchw(xdp->data_hard_start); /* xdp_frame write */
2212 
2213 	act = bpf_prog_run_xdp(xdp_prog, xdp);
2214 	switch (act) {
2215 	case XDP_PASS:
2216 		break;
2217 	case XDP_TX:
2218 		xdpf = convert_to_xdp_frame(xdp);
2219 		if (unlikely(!xdpf)) {
2220 			result = IXGBE_XDP_CONSUMED;
2221 			break;
2222 		}
2223 		result = ixgbe_xmit_xdp_ring(adapter, xdpf);
2224 		break;
2225 	case XDP_REDIRECT:
2226 		err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
2227 		if (!err)
2228 			result = IXGBE_XDP_REDIR;
2229 		else
2230 			result = IXGBE_XDP_CONSUMED;
2231 		break;
2232 	default:
2233 		bpf_warn_invalid_xdp_action(act);
2234 		/* fallthrough */
2235 	case XDP_ABORTED:
2236 		trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
2237 		/* fallthrough -- handle aborts by dropping packet */
2238 	case XDP_DROP:
2239 		result = IXGBE_XDP_CONSUMED;
2240 		break;
2241 	}
2242 xdp_out:
2243 	rcu_read_unlock();
2244 	return ERR_PTR(-result);
2245 }
2246 
2247 static void ixgbe_rx_buffer_flip(struct ixgbe_ring *rx_ring,
2248 				 struct ixgbe_rx_buffer *rx_buffer,
2249 				 unsigned int size)
2250 {
2251 #if (PAGE_SIZE < 8192)
2252 	unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2253 
2254 	rx_buffer->page_offset ^= truesize;
2255 #else
2256 	unsigned int truesize = ring_uses_build_skb(rx_ring) ?
2257 				SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) :
2258 				SKB_DATA_ALIGN(size);
2259 
2260 	rx_buffer->page_offset += truesize;
2261 #endif
2262 }
2263 
2264 /**
2265  * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2266  * @q_vector: structure containing interrupt and ring information
2267  * @rx_ring: rx descriptor ring to transact packets on
2268  * @budget: Total limit on number of packets to process
2269  *
2270  * This function provides a "bounce buffer" approach to Rx interrupt
2271  * processing.  The advantage to this is that on systems that have
2272  * expensive overhead for IOMMU access this provides a means of avoiding
2273  * it by maintaining the mapping of the page to the syste.
2274  *
2275  * Returns amount of work completed
2276  **/
2277 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2278 			       struct ixgbe_ring *rx_ring,
2279 			       const int budget)
2280 {
2281 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2282 	struct ixgbe_adapter *adapter = q_vector->adapter;
2283 #ifdef IXGBE_FCOE
2284 	int ddp_bytes;
2285 	unsigned int mss = 0;
2286 #endif /* IXGBE_FCOE */
2287 	u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2288 	unsigned int xdp_xmit = 0;
2289 	struct xdp_buff xdp;
2290 
2291 	xdp.rxq = &rx_ring->xdp_rxq;
2292 
2293 	while (likely(total_rx_packets < budget)) {
2294 		union ixgbe_adv_rx_desc *rx_desc;
2295 		struct ixgbe_rx_buffer *rx_buffer;
2296 		struct sk_buff *skb;
2297 		unsigned int size;
2298 
2299 		/* return some buffers to hardware, one at a time is too slow */
2300 		if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2301 			ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2302 			cleaned_count = 0;
2303 		}
2304 
2305 		rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2306 		size = le16_to_cpu(rx_desc->wb.upper.length);
2307 		if (!size)
2308 			break;
2309 
2310 		/* This memory barrier is needed to keep us from reading
2311 		 * any other fields out of the rx_desc until we know the
2312 		 * descriptor has been written back
2313 		 */
2314 		dma_rmb();
2315 
2316 		rx_buffer = ixgbe_get_rx_buffer(rx_ring, rx_desc, &skb, size);
2317 
2318 		/* retrieve a buffer from the ring */
2319 		if (!skb) {
2320 			xdp.data = page_address(rx_buffer->page) +
2321 				   rx_buffer->page_offset;
2322 			xdp.data_meta = xdp.data;
2323 			xdp.data_hard_start = xdp.data -
2324 					      ixgbe_rx_offset(rx_ring);
2325 			xdp.data_end = xdp.data + size;
2326 
2327 			skb = ixgbe_run_xdp(adapter, rx_ring, &xdp);
2328 		}
2329 
2330 		if (IS_ERR(skb)) {
2331 			unsigned int xdp_res = -PTR_ERR(skb);
2332 
2333 			if (xdp_res & (IXGBE_XDP_TX | IXGBE_XDP_REDIR)) {
2334 				xdp_xmit |= xdp_res;
2335 				ixgbe_rx_buffer_flip(rx_ring, rx_buffer, size);
2336 			} else {
2337 				rx_buffer->pagecnt_bias++;
2338 			}
2339 			total_rx_packets++;
2340 			total_rx_bytes += size;
2341 		} else if (skb) {
2342 			ixgbe_add_rx_frag(rx_ring, rx_buffer, skb, size);
2343 		} else if (ring_uses_build_skb(rx_ring)) {
2344 			skb = ixgbe_build_skb(rx_ring, rx_buffer,
2345 					      &xdp, rx_desc);
2346 		} else {
2347 			skb = ixgbe_construct_skb(rx_ring, rx_buffer,
2348 						  &xdp, rx_desc);
2349 		}
2350 
2351 		/* exit if we failed to retrieve a buffer */
2352 		if (!skb) {
2353 			rx_ring->rx_stats.alloc_rx_buff_failed++;
2354 			rx_buffer->pagecnt_bias++;
2355 			break;
2356 		}
2357 
2358 		ixgbe_put_rx_buffer(rx_ring, rx_buffer, skb);
2359 		cleaned_count++;
2360 
2361 		/* place incomplete frames back on ring for completion */
2362 		if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2363 			continue;
2364 
2365 		/* verify the packet layout is correct */
2366 		if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2367 			continue;
2368 
2369 		/* probably a little skewed due to removing CRC */
2370 		total_rx_bytes += skb->len;
2371 
2372 		/* populate checksum, timestamp, VLAN, and protocol */
2373 		ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2374 
2375 #ifdef IXGBE_FCOE
2376 		/* if ddp, not passing to ULD unless for FCP_RSP or error */
2377 		if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2378 			ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2379 			/* include DDPed FCoE data */
2380 			if (ddp_bytes > 0) {
2381 				if (!mss) {
2382 					mss = rx_ring->netdev->mtu -
2383 						sizeof(struct fcoe_hdr) -
2384 						sizeof(struct fc_frame_header) -
2385 						sizeof(struct fcoe_crc_eof);
2386 					if (mss > 512)
2387 						mss &= ~511;
2388 				}
2389 				total_rx_bytes += ddp_bytes;
2390 				total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2391 								 mss);
2392 			}
2393 			if (!ddp_bytes) {
2394 				dev_kfree_skb_any(skb);
2395 				continue;
2396 			}
2397 		}
2398 
2399 #endif /* IXGBE_FCOE */
2400 		ixgbe_rx_skb(q_vector, skb);
2401 
2402 		/* update budget accounting */
2403 		total_rx_packets++;
2404 	}
2405 
2406 	if (xdp_xmit & IXGBE_XDP_REDIR)
2407 		xdp_do_flush_map();
2408 
2409 	if (xdp_xmit & IXGBE_XDP_TX) {
2410 		struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
2411 
2412 		/* Force memory writes to complete before letting h/w
2413 		 * know there are new descriptors to fetch.
2414 		 */
2415 		wmb();
2416 		writel(ring->next_to_use, ring->tail);
2417 	}
2418 
2419 	u64_stats_update_begin(&rx_ring->syncp);
2420 	rx_ring->stats.packets += total_rx_packets;
2421 	rx_ring->stats.bytes += total_rx_bytes;
2422 	u64_stats_update_end(&rx_ring->syncp);
2423 	q_vector->rx.total_packets += total_rx_packets;
2424 	q_vector->rx.total_bytes += total_rx_bytes;
2425 
2426 	return total_rx_packets;
2427 }
2428 
2429 /**
2430  * ixgbe_configure_msix - Configure MSI-X hardware
2431  * @adapter: board private structure
2432  *
2433  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2434  * interrupts.
2435  **/
2436 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2437 {
2438 	struct ixgbe_q_vector *q_vector;
2439 	int v_idx;
2440 	u32 mask;
2441 
2442 	/* Populate MSIX to EITR Select */
2443 	if (adapter->num_vfs > 32) {
2444 		u32 eitrsel = BIT(adapter->num_vfs - 32) - 1;
2445 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2446 	}
2447 
2448 	/*
2449 	 * Populate the IVAR table and set the ITR values to the
2450 	 * corresponding register.
2451 	 */
2452 	for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2453 		struct ixgbe_ring *ring;
2454 		q_vector = adapter->q_vector[v_idx];
2455 
2456 		ixgbe_for_each_ring(ring, q_vector->rx)
2457 			ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2458 
2459 		ixgbe_for_each_ring(ring, q_vector->tx)
2460 			ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2461 
2462 		ixgbe_write_eitr(q_vector);
2463 	}
2464 
2465 	switch (adapter->hw.mac.type) {
2466 	case ixgbe_mac_82598EB:
2467 		ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2468 			       v_idx);
2469 		break;
2470 	case ixgbe_mac_82599EB:
2471 	case ixgbe_mac_X540:
2472 	case ixgbe_mac_X550:
2473 	case ixgbe_mac_X550EM_x:
2474 	case ixgbe_mac_x550em_a:
2475 		ixgbe_set_ivar(adapter, -1, 1, v_idx);
2476 		break;
2477 	default:
2478 		break;
2479 	}
2480 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2481 
2482 	/* set up to autoclear timer, and the vectors */
2483 	mask = IXGBE_EIMS_ENABLE_MASK;
2484 	mask &= ~(IXGBE_EIMS_OTHER |
2485 		  IXGBE_EIMS_MAILBOX |
2486 		  IXGBE_EIMS_LSC);
2487 
2488 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2489 }
2490 
2491 /**
2492  * ixgbe_update_itr - update the dynamic ITR value based on statistics
2493  * @q_vector: structure containing interrupt and ring information
2494  * @ring_container: structure containing ring performance data
2495  *
2496  *      Stores a new ITR value based on packets and byte
2497  *      counts during the last interrupt.  The advantage of per interrupt
2498  *      computation is faster updates and more accurate ITR for the current
2499  *      traffic pattern.  Constants in this function were computed
2500  *      based on theoretical maximum wire speed and thresholds were set based
2501  *      on testing data as well as attempting to minimize response time
2502  *      while increasing bulk throughput.
2503  **/
2504 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2505 			     struct ixgbe_ring_container *ring_container)
2506 {
2507 	unsigned int itr = IXGBE_ITR_ADAPTIVE_MIN_USECS |
2508 			   IXGBE_ITR_ADAPTIVE_LATENCY;
2509 	unsigned int avg_wire_size, packets, bytes;
2510 	unsigned long next_update = jiffies;
2511 
2512 	/* If we don't have any rings just leave ourselves set for maximum
2513 	 * possible latency so we take ourselves out of the equation.
2514 	 */
2515 	if (!ring_container->ring)
2516 		return;
2517 
2518 	/* If we didn't update within up to 1 - 2 jiffies we can assume
2519 	 * that either packets are coming in so slow there hasn't been
2520 	 * any work, or that there is so much work that NAPI is dealing
2521 	 * with interrupt moderation and we don't need to do anything.
2522 	 */
2523 	if (time_after(next_update, ring_container->next_update))
2524 		goto clear_counts;
2525 
2526 	packets = ring_container->total_packets;
2527 
2528 	/* We have no packets to actually measure against. This means
2529 	 * either one of the other queues on this vector is active or
2530 	 * we are a Tx queue doing TSO with too high of an interrupt rate.
2531 	 *
2532 	 * When this occurs just tick up our delay by the minimum value
2533 	 * and hope that this extra delay will prevent us from being called
2534 	 * without any work on our queue.
2535 	 */
2536 	if (!packets) {
2537 		itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2538 		if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2539 			itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2540 		itr += ring_container->itr & IXGBE_ITR_ADAPTIVE_LATENCY;
2541 		goto clear_counts;
2542 	}
2543 
2544 	bytes = ring_container->total_bytes;
2545 
2546 	/* If packets are less than 4 or bytes are less than 9000 assume
2547 	 * insufficient data to use bulk rate limiting approach. We are
2548 	 * likely latency driven.
2549 	 */
2550 	if (packets < 4 && bytes < 9000) {
2551 		itr = IXGBE_ITR_ADAPTIVE_LATENCY;
2552 		goto adjust_by_size;
2553 	}
2554 
2555 	/* Between 4 and 48 we can assume that our current interrupt delay
2556 	 * is only slightly too low. As such we should increase it by a small
2557 	 * fixed amount.
2558 	 */
2559 	if (packets < 48) {
2560 		itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2561 		if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2562 			itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2563 		goto clear_counts;
2564 	}
2565 
2566 	/* Between 48 and 96 is our "goldilocks" zone where we are working
2567 	 * out "just right". Just report that our current ITR is good for us.
2568 	 */
2569 	if (packets < 96) {
2570 		itr = q_vector->itr >> 2;
2571 		goto clear_counts;
2572 	}
2573 
2574 	/* If packet count is 96 or greater we are likely looking at a slight
2575 	 * overrun of the delay we want. Try halving our delay to see if that
2576 	 * will cut the number of packets in half per interrupt.
2577 	 */
2578 	if (packets < 256) {
2579 		itr = q_vector->itr >> 3;
2580 		if (itr < IXGBE_ITR_ADAPTIVE_MIN_USECS)
2581 			itr = IXGBE_ITR_ADAPTIVE_MIN_USECS;
2582 		goto clear_counts;
2583 	}
2584 
2585 	/* The paths below assume we are dealing with a bulk ITR since number
2586 	 * of packets is 256 or greater. We are just going to have to compute
2587 	 * a value and try to bring the count under control, though for smaller
2588 	 * packet sizes there isn't much we can do as NAPI polling will likely
2589 	 * be kicking in sooner rather than later.
2590 	 */
2591 	itr = IXGBE_ITR_ADAPTIVE_BULK;
2592 
2593 adjust_by_size:
2594 	/* If packet counts are 256 or greater we can assume we have a gross
2595 	 * overestimation of what the rate should be. Instead of trying to fine
2596 	 * tune it just use the formula below to try and dial in an exact value
2597 	 * give the current packet size of the frame.
2598 	 */
2599 	avg_wire_size = bytes / packets;
2600 
2601 	/* The following is a crude approximation of:
2602 	 *  wmem_default / (size + overhead) = desired_pkts_per_int
2603 	 *  rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
2604 	 *  (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
2605 	 *
2606 	 * Assuming wmem_default is 212992 and overhead is 640 bytes per
2607 	 * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
2608 	 * formula down to
2609 	 *
2610 	 *  (170 * (size + 24)) / (size + 640) = ITR
2611 	 *
2612 	 * We first do some math on the packet size and then finally bitshift
2613 	 * by 8 after rounding up. We also have to account for PCIe link speed
2614 	 * difference as ITR scales based on this.
2615 	 */
2616 	if (avg_wire_size <= 60) {
2617 		/* Start at 50k ints/sec */
2618 		avg_wire_size = 5120;
2619 	} else if (avg_wire_size <= 316) {
2620 		/* 50K ints/sec to 16K ints/sec */
2621 		avg_wire_size *= 40;
2622 		avg_wire_size += 2720;
2623 	} else if (avg_wire_size <= 1084) {
2624 		/* 16K ints/sec to 9.2K ints/sec */
2625 		avg_wire_size *= 15;
2626 		avg_wire_size += 11452;
2627 	} else if (avg_wire_size < 1968) {
2628 		/* 9.2K ints/sec to 8K ints/sec */
2629 		avg_wire_size *= 5;
2630 		avg_wire_size += 22420;
2631 	} else {
2632 		/* plateau at a limit of 8K ints/sec */
2633 		avg_wire_size = 32256;
2634 	}
2635 
2636 	/* If we are in low latency mode half our delay which doubles the rate
2637 	 * to somewhere between 100K to 16K ints/sec
2638 	 */
2639 	if (itr & IXGBE_ITR_ADAPTIVE_LATENCY)
2640 		avg_wire_size >>= 1;
2641 
2642 	/* Resultant value is 256 times larger than it needs to be. This
2643 	 * gives us room to adjust the value as needed to either increase
2644 	 * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
2645 	 *
2646 	 * Use addition as we have already recorded the new latency flag
2647 	 * for the ITR value.
2648 	 */
2649 	switch (q_vector->adapter->link_speed) {
2650 	case IXGBE_LINK_SPEED_10GB_FULL:
2651 	case IXGBE_LINK_SPEED_100_FULL:
2652 	default:
2653 		itr += DIV_ROUND_UP(avg_wire_size,
2654 				    IXGBE_ITR_ADAPTIVE_MIN_INC * 256) *
2655 		       IXGBE_ITR_ADAPTIVE_MIN_INC;
2656 		break;
2657 	case IXGBE_LINK_SPEED_2_5GB_FULL:
2658 	case IXGBE_LINK_SPEED_1GB_FULL:
2659 	case IXGBE_LINK_SPEED_10_FULL:
2660 		if (avg_wire_size > 8064)
2661 			avg_wire_size = 8064;
2662 		itr += DIV_ROUND_UP(avg_wire_size,
2663 				    IXGBE_ITR_ADAPTIVE_MIN_INC * 64) *
2664 		       IXGBE_ITR_ADAPTIVE_MIN_INC;
2665 		break;
2666 	}
2667 
2668 clear_counts:
2669 	/* write back value */
2670 	ring_container->itr = itr;
2671 
2672 	/* next update should occur within next jiffy */
2673 	ring_container->next_update = next_update + 1;
2674 
2675 	ring_container->total_bytes = 0;
2676 	ring_container->total_packets = 0;
2677 }
2678 
2679 /**
2680  * ixgbe_write_eitr - write EITR register in hardware specific way
2681  * @q_vector: structure containing interrupt and ring information
2682  *
2683  * This function is made to be called by ethtool and by the driver
2684  * when it needs to update EITR registers at runtime.  Hardware
2685  * specific quirks/differences are taken care of here.
2686  */
2687 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2688 {
2689 	struct ixgbe_adapter *adapter = q_vector->adapter;
2690 	struct ixgbe_hw *hw = &adapter->hw;
2691 	int v_idx = q_vector->v_idx;
2692 	u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2693 
2694 	switch (adapter->hw.mac.type) {
2695 	case ixgbe_mac_82598EB:
2696 		/* must write high and low 16 bits to reset counter */
2697 		itr_reg |= (itr_reg << 16);
2698 		break;
2699 	case ixgbe_mac_82599EB:
2700 	case ixgbe_mac_X540:
2701 	case ixgbe_mac_X550:
2702 	case ixgbe_mac_X550EM_x:
2703 	case ixgbe_mac_x550em_a:
2704 		/*
2705 		 * set the WDIS bit to not clear the timer bits and cause an
2706 		 * immediate assertion of the interrupt
2707 		 */
2708 		itr_reg |= IXGBE_EITR_CNT_WDIS;
2709 		break;
2710 	default:
2711 		break;
2712 	}
2713 	IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2714 }
2715 
2716 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2717 {
2718 	u32 new_itr;
2719 
2720 	ixgbe_update_itr(q_vector, &q_vector->tx);
2721 	ixgbe_update_itr(q_vector, &q_vector->rx);
2722 
2723 	/* use the smallest value of new ITR delay calculations */
2724 	new_itr = min(q_vector->rx.itr, q_vector->tx.itr);
2725 
2726 	/* Clear latency flag if set, shift into correct position */
2727 	new_itr &= ~IXGBE_ITR_ADAPTIVE_LATENCY;
2728 	new_itr <<= 2;
2729 
2730 	if (new_itr != q_vector->itr) {
2731 		/* save the algorithm value here */
2732 		q_vector->itr = new_itr;
2733 
2734 		ixgbe_write_eitr(q_vector);
2735 	}
2736 }
2737 
2738 /**
2739  * ixgbe_check_overtemp_subtask - check for over temperature
2740  * @adapter: pointer to adapter
2741  **/
2742 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2743 {
2744 	struct ixgbe_hw *hw = &adapter->hw;
2745 	u32 eicr = adapter->interrupt_event;
2746 	s32 rc;
2747 
2748 	if (test_bit(__IXGBE_DOWN, &adapter->state))
2749 		return;
2750 
2751 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2752 		return;
2753 
2754 	adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2755 
2756 	switch (hw->device_id) {
2757 	case IXGBE_DEV_ID_82599_T3_LOM:
2758 		/*
2759 		 * Since the warning interrupt is for both ports
2760 		 * we don't have to check if:
2761 		 *  - This interrupt wasn't for our port.
2762 		 *  - We may have missed the interrupt so always have to
2763 		 *    check if we  got a LSC
2764 		 */
2765 		if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2766 		    !(eicr & IXGBE_EICR_LSC))
2767 			return;
2768 
2769 		if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2770 			u32 speed;
2771 			bool link_up = false;
2772 
2773 			hw->mac.ops.check_link(hw, &speed, &link_up, false);
2774 
2775 			if (link_up)
2776 				return;
2777 		}
2778 
2779 		/* Check if this is not due to overtemp */
2780 		if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2781 			return;
2782 
2783 		break;
2784 	case IXGBE_DEV_ID_X550EM_A_1G_T:
2785 	case IXGBE_DEV_ID_X550EM_A_1G_T_L:
2786 		rc = hw->phy.ops.check_overtemp(hw);
2787 		if (rc != IXGBE_ERR_OVERTEMP)
2788 			return;
2789 		break;
2790 	default:
2791 		if (adapter->hw.mac.type >= ixgbe_mac_X540)
2792 			return;
2793 		if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2794 			return;
2795 		break;
2796 	}
2797 	e_crit(drv, "%s\n", ixgbe_overheat_msg);
2798 
2799 	adapter->interrupt_event = 0;
2800 }
2801 
2802 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2803 {
2804 	struct ixgbe_hw *hw = &adapter->hw;
2805 
2806 	if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2807 	    (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2808 		e_crit(probe, "Fan has stopped, replace the adapter\n");
2809 		/* write to clear the interrupt */
2810 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2811 	}
2812 }
2813 
2814 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2815 {
2816 	struct ixgbe_hw *hw = &adapter->hw;
2817 
2818 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2819 		return;
2820 
2821 	switch (adapter->hw.mac.type) {
2822 	case ixgbe_mac_82599EB:
2823 		/*
2824 		 * Need to check link state so complete overtemp check
2825 		 * on service task
2826 		 */
2827 		if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2828 		     (eicr & IXGBE_EICR_LSC)) &&
2829 		    (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2830 			adapter->interrupt_event = eicr;
2831 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2832 			ixgbe_service_event_schedule(adapter);
2833 			return;
2834 		}
2835 		return;
2836 	case ixgbe_mac_x550em_a:
2837 		if (eicr & IXGBE_EICR_GPI_SDP0_X550EM_a) {
2838 			adapter->interrupt_event = eicr;
2839 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2840 			ixgbe_service_event_schedule(adapter);
2841 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC,
2842 					IXGBE_EICR_GPI_SDP0_X550EM_a);
2843 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICR,
2844 					IXGBE_EICR_GPI_SDP0_X550EM_a);
2845 		}
2846 		return;
2847 	case ixgbe_mac_X550:
2848 	case ixgbe_mac_X540:
2849 		if (!(eicr & IXGBE_EICR_TS))
2850 			return;
2851 		break;
2852 	default:
2853 		return;
2854 	}
2855 
2856 	e_crit(drv, "%s\n", ixgbe_overheat_msg);
2857 }
2858 
2859 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2860 {
2861 	switch (hw->mac.type) {
2862 	case ixgbe_mac_82598EB:
2863 		if (hw->phy.type == ixgbe_phy_nl)
2864 			return true;
2865 		return false;
2866 	case ixgbe_mac_82599EB:
2867 	case ixgbe_mac_X550EM_x:
2868 	case ixgbe_mac_x550em_a:
2869 		switch (hw->mac.ops.get_media_type(hw)) {
2870 		case ixgbe_media_type_fiber:
2871 		case ixgbe_media_type_fiber_qsfp:
2872 			return true;
2873 		default:
2874 			return false;
2875 		}
2876 	default:
2877 		return false;
2878 	}
2879 }
2880 
2881 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2882 {
2883 	struct ixgbe_hw *hw = &adapter->hw;
2884 	u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw);
2885 
2886 	if (!ixgbe_is_sfp(hw))
2887 		return;
2888 
2889 	/* Later MAC's use different SDP */
2890 	if (hw->mac.type >= ixgbe_mac_X540)
2891 		eicr_mask = IXGBE_EICR_GPI_SDP0_X540;
2892 
2893 	if (eicr & eicr_mask) {
2894 		/* Clear the interrupt */
2895 		IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask);
2896 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2897 			adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2898 			adapter->sfp_poll_time = 0;
2899 			ixgbe_service_event_schedule(adapter);
2900 		}
2901 	}
2902 
2903 	if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
2904 	    (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2905 		/* Clear the interrupt */
2906 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2907 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2908 			adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2909 			ixgbe_service_event_schedule(adapter);
2910 		}
2911 	}
2912 }
2913 
2914 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2915 {
2916 	struct ixgbe_hw *hw = &adapter->hw;
2917 
2918 	adapter->lsc_int++;
2919 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2920 	adapter->link_check_timeout = jiffies;
2921 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2922 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2923 		IXGBE_WRITE_FLUSH(hw);
2924 		ixgbe_service_event_schedule(adapter);
2925 	}
2926 }
2927 
2928 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2929 					   u64 qmask)
2930 {
2931 	u32 mask;
2932 	struct ixgbe_hw *hw = &adapter->hw;
2933 
2934 	switch (hw->mac.type) {
2935 	case ixgbe_mac_82598EB:
2936 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2937 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2938 		break;
2939 	case ixgbe_mac_82599EB:
2940 	case ixgbe_mac_X540:
2941 	case ixgbe_mac_X550:
2942 	case ixgbe_mac_X550EM_x:
2943 	case ixgbe_mac_x550em_a:
2944 		mask = (qmask & 0xFFFFFFFF);
2945 		if (mask)
2946 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2947 		mask = (qmask >> 32);
2948 		if (mask)
2949 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2950 		break;
2951 	default:
2952 		break;
2953 	}
2954 	/* skip the flush */
2955 }
2956 
2957 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2958 					    u64 qmask)
2959 {
2960 	u32 mask;
2961 	struct ixgbe_hw *hw = &adapter->hw;
2962 
2963 	switch (hw->mac.type) {
2964 	case ixgbe_mac_82598EB:
2965 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2966 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2967 		break;
2968 	case ixgbe_mac_82599EB:
2969 	case ixgbe_mac_X540:
2970 	case ixgbe_mac_X550:
2971 	case ixgbe_mac_X550EM_x:
2972 	case ixgbe_mac_x550em_a:
2973 		mask = (qmask & 0xFFFFFFFF);
2974 		if (mask)
2975 			IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2976 		mask = (qmask >> 32);
2977 		if (mask)
2978 			IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2979 		break;
2980 	default:
2981 		break;
2982 	}
2983 	/* skip the flush */
2984 }
2985 
2986 /**
2987  * ixgbe_irq_enable - Enable default interrupt generation settings
2988  * @adapter: board private structure
2989  * @queues: enable irqs for queues
2990  * @flush: flush register write
2991  **/
2992 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2993 				    bool flush)
2994 {
2995 	struct ixgbe_hw *hw = &adapter->hw;
2996 	u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2997 
2998 	/* don't reenable LSC while waiting for link */
2999 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
3000 		mask &= ~IXGBE_EIMS_LSC;
3001 
3002 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
3003 		switch (adapter->hw.mac.type) {
3004 		case ixgbe_mac_82599EB:
3005 			mask |= IXGBE_EIMS_GPI_SDP0(hw);
3006 			break;
3007 		case ixgbe_mac_X540:
3008 		case ixgbe_mac_X550:
3009 		case ixgbe_mac_X550EM_x:
3010 		case ixgbe_mac_x550em_a:
3011 			mask |= IXGBE_EIMS_TS;
3012 			break;
3013 		default:
3014 			break;
3015 		}
3016 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3017 		mask |= IXGBE_EIMS_GPI_SDP1(hw);
3018 	switch (adapter->hw.mac.type) {
3019 	case ixgbe_mac_82599EB:
3020 		mask |= IXGBE_EIMS_GPI_SDP1(hw);
3021 		mask |= IXGBE_EIMS_GPI_SDP2(hw);
3022 		/* fall through */
3023 	case ixgbe_mac_X540:
3024 	case ixgbe_mac_X550:
3025 	case ixgbe_mac_X550EM_x:
3026 	case ixgbe_mac_x550em_a:
3027 		if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
3028 		    adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
3029 		    adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP_N)
3030 			mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw);
3031 		if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
3032 			mask |= IXGBE_EICR_GPI_SDP0_X540;
3033 		mask |= IXGBE_EIMS_ECC;
3034 		mask |= IXGBE_EIMS_MAILBOX;
3035 		break;
3036 	default:
3037 		break;
3038 	}
3039 
3040 	if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
3041 	    !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
3042 		mask |= IXGBE_EIMS_FLOW_DIR;
3043 
3044 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
3045 	if (queues)
3046 		ixgbe_irq_enable_queues(adapter, ~0);
3047 	if (flush)
3048 		IXGBE_WRITE_FLUSH(&adapter->hw);
3049 }
3050 
3051 static irqreturn_t ixgbe_msix_other(int irq, void *data)
3052 {
3053 	struct ixgbe_adapter *adapter = data;
3054 	struct ixgbe_hw *hw = &adapter->hw;
3055 	u32 eicr;
3056 
3057 	/*
3058 	 * Workaround for Silicon errata.  Use clear-by-write instead
3059 	 * of clear-by-read.  Reading with EICS will return the
3060 	 * interrupt causes without clearing, which later be done
3061 	 * with the write to EICR.
3062 	 */
3063 	eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
3064 
3065 	/* The lower 16bits of the EICR register are for the queue interrupts
3066 	 * which should be masked here in order to not accidentally clear them if
3067 	 * the bits are high when ixgbe_msix_other is called. There is a race
3068 	 * condition otherwise which results in possible performance loss
3069 	 * especially if the ixgbe_msix_other interrupt is triggering
3070 	 * consistently (as it would when PPS is turned on for the X540 device)
3071 	 */
3072 	eicr &= 0xFFFF0000;
3073 
3074 	IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
3075 
3076 	if (eicr & IXGBE_EICR_LSC)
3077 		ixgbe_check_lsc(adapter);
3078 
3079 	if (eicr & IXGBE_EICR_MAILBOX)
3080 		ixgbe_msg_task(adapter);
3081 
3082 	switch (hw->mac.type) {
3083 	case ixgbe_mac_82599EB:
3084 	case ixgbe_mac_X540:
3085 	case ixgbe_mac_X550:
3086 	case ixgbe_mac_X550EM_x:
3087 	case ixgbe_mac_x550em_a:
3088 		if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
3089 		    (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
3090 			adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
3091 			ixgbe_service_event_schedule(adapter);
3092 			IXGBE_WRITE_REG(hw, IXGBE_EICR,
3093 					IXGBE_EICR_GPI_SDP0_X540);
3094 		}
3095 		if (eicr & IXGBE_EICR_ECC) {
3096 			e_info(link, "Received ECC Err, initiating reset\n");
3097 			set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3098 			ixgbe_service_event_schedule(adapter);
3099 			IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3100 		}
3101 		/* Handle Flow Director Full threshold interrupt */
3102 		if (eicr & IXGBE_EICR_FLOW_DIR) {
3103 			int reinit_count = 0;
3104 			int i;
3105 			for (i = 0; i < adapter->num_tx_queues; i++) {
3106 				struct ixgbe_ring *ring = adapter->tx_ring[i];
3107 				if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
3108 						       &ring->state))
3109 					reinit_count++;
3110 			}
3111 			if (reinit_count) {
3112 				/* no more flow director interrupts until after init */
3113 				IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
3114 				adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
3115 				ixgbe_service_event_schedule(adapter);
3116 			}
3117 		}
3118 		ixgbe_check_sfp_event(adapter, eicr);
3119 		ixgbe_check_overtemp_event(adapter, eicr);
3120 		break;
3121 	default:
3122 		break;
3123 	}
3124 
3125 	ixgbe_check_fan_failure(adapter, eicr);
3126 
3127 	if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3128 		ixgbe_ptp_check_pps_event(adapter);
3129 
3130 	/* re-enable the original interrupt state, no lsc, no queues */
3131 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
3132 		ixgbe_irq_enable(adapter, false, false);
3133 
3134 	return IRQ_HANDLED;
3135 }
3136 
3137 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
3138 {
3139 	struct ixgbe_q_vector *q_vector = data;
3140 
3141 	/* EIAM disabled interrupts (on this vector) for us */
3142 
3143 	if (q_vector->rx.ring || q_vector->tx.ring)
3144 		napi_schedule_irqoff(&q_vector->napi);
3145 
3146 	return IRQ_HANDLED;
3147 }
3148 
3149 /**
3150  * ixgbe_poll - NAPI Rx polling callback
3151  * @napi: structure for representing this polling device
3152  * @budget: how many packets driver is allowed to clean
3153  *
3154  * This function is used for legacy and MSI, NAPI mode
3155  **/
3156 int ixgbe_poll(struct napi_struct *napi, int budget)
3157 {
3158 	struct ixgbe_q_vector *q_vector =
3159 				container_of(napi, struct ixgbe_q_vector, napi);
3160 	struct ixgbe_adapter *adapter = q_vector->adapter;
3161 	struct ixgbe_ring *ring;
3162 	int per_ring_budget, work_done = 0;
3163 	bool clean_complete = true;
3164 
3165 #ifdef CONFIG_IXGBE_DCA
3166 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3167 		ixgbe_update_dca(q_vector);
3168 #endif
3169 
3170 	ixgbe_for_each_ring(ring, q_vector->tx) {
3171 		bool wd = ring->xsk_umem ?
3172 			  ixgbe_clean_xdp_tx_irq(q_vector, ring, budget) :
3173 			  ixgbe_clean_tx_irq(q_vector, ring, budget);
3174 
3175 		if (!wd)
3176 			clean_complete = false;
3177 	}
3178 
3179 	/* Exit if we are called by netpoll */
3180 	if (budget <= 0)
3181 		return budget;
3182 
3183 	/* attempt to distribute budget to each queue fairly, but don't allow
3184 	 * the budget to go below 1 because we'll exit polling */
3185 	if (q_vector->rx.count > 1)
3186 		per_ring_budget = max(budget/q_vector->rx.count, 1);
3187 	else
3188 		per_ring_budget = budget;
3189 
3190 	ixgbe_for_each_ring(ring, q_vector->rx) {
3191 		int cleaned = ring->xsk_umem ?
3192 			      ixgbe_clean_rx_irq_zc(q_vector, ring,
3193 						    per_ring_budget) :
3194 			      ixgbe_clean_rx_irq(q_vector, ring,
3195 						 per_ring_budget);
3196 
3197 		work_done += cleaned;
3198 		if (cleaned >= per_ring_budget)
3199 			clean_complete = false;
3200 	}
3201 
3202 	/* If all work not completed, return budget and keep polling */
3203 	if (!clean_complete)
3204 		return budget;
3205 
3206 	/* all work done, exit the polling mode */
3207 	if (likely(napi_complete_done(napi, work_done))) {
3208 		if (adapter->rx_itr_setting & 1)
3209 			ixgbe_set_itr(q_vector);
3210 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
3211 			ixgbe_irq_enable_queues(adapter,
3212 						BIT_ULL(q_vector->v_idx));
3213 	}
3214 
3215 	return min(work_done, budget - 1);
3216 }
3217 
3218 /**
3219  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
3220  * @adapter: board private structure
3221  *
3222  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
3223  * interrupts from the kernel.
3224  **/
3225 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
3226 {
3227 	struct net_device *netdev = adapter->netdev;
3228 	unsigned int ri = 0, ti = 0;
3229 	int vector, err;
3230 
3231 	for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3232 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3233 		struct msix_entry *entry = &adapter->msix_entries[vector];
3234 
3235 		if (q_vector->tx.ring && q_vector->rx.ring) {
3236 			snprintf(q_vector->name, sizeof(q_vector->name),
3237 				 "%s-TxRx-%u", netdev->name, ri++);
3238 			ti++;
3239 		} else if (q_vector->rx.ring) {
3240 			snprintf(q_vector->name, sizeof(q_vector->name),
3241 				 "%s-rx-%u", netdev->name, ri++);
3242 		} else if (q_vector->tx.ring) {
3243 			snprintf(q_vector->name, sizeof(q_vector->name),
3244 				 "%s-tx-%u", netdev->name, ti++);
3245 		} else {
3246 			/* skip this unused q_vector */
3247 			continue;
3248 		}
3249 		err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
3250 				  q_vector->name, q_vector);
3251 		if (err) {
3252 			e_err(probe, "request_irq failed for MSIX interrupt "
3253 			      "Error: %d\n", err);
3254 			goto free_queue_irqs;
3255 		}
3256 		/* If Flow Director is enabled, set interrupt affinity */
3257 		if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3258 			/* assign the mask for this irq */
3259 			irq_set_affinity_hint(entry->vector,
3260 					      &q_vector->affinity_mask);
3261 		}
3262 	}
3263 
3264 	err = request_irq(adapter->msix_entries[vector].vector,
3265 			  ixgbe_msix_other, 0, netdev->name, adapter);
3266 	if (err) {
3267 		e_err(probe, "request_irq for msix_other failed: %d\n", err);
3268 		goto free_queue_irqs;
3269 	}
3270 
3271 	return 0;
3272 
3273 free_queue_irqs:
3274 	while (vector) {
3275 		vector--;
3276 		irq_set_affinity_hint(adapter->msix_entries[vector].vector,
3277 				      NULL);
3278 		free_irq(adapter->msix_entries[vector].vector,
3279 			 adapter->q_vector[vector]);
3280 	}
3281 	adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3282 	pci_disable_msix(adapter->pdev);
3283 	kfree(adapter->msix_entries);
3284 	adapter->msix_entries = NULL;
3285 	return err;
3286 }
3287 
3288 /**
3289  * ixgbe_intr - legacy mode Interrupt Handler
3290  * @irq: interrupt number
3291  * @data: pointer to a network interface device structure
3292  **/
3293 static irqreturn_t ixgbe_intr(int irq, void *data)
3294 {
3295 	struct ixgbe_adapter *adapter = data;
3296 	struct ixgbe_hw *hw = &adapter->hw;
3297 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3298 	u32 eicr;
3299 
3300 	/*
3301 	 * Workaround for silicon errata #26 on 82598.  Mask the interrupt
3302 	 * before the read of EICR.
3303 	 */
3304 	IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
3305 
3306 	/* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
3307 	 * therefore no explicit interrupt disable is necessary */
3308 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
3309 	if (!eicr) {
3310 		/*
3311 		 * shared interrupt alert!
3312 		 * make sure interrupts are enabled because the read will
3313 		 * have disabled interrupts due to EIAM
3314 		 * finish the workaround of silicon errata on 82598.  Unmask
3315 		 * the interrupt that we masked before the EICR read.
3316 		 */
3317 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
3318 			ixgbe_irq_enable(adapter, true, true);
3319 		return IRQ_NONE;	/* Not our interrupt */
3320 	}
3321 
3322 	if (eicr & IXGBE_EICR_LSC)
3323 		ixgbe_check_lsc(adapter);
3324 
3325 	switch (hw->mac.type) {
3326 	case ixgbe_mac_82599EB:
3327 		ixgbe_check_sfp_event(adapter, eicr);
3328 		/* Fall through */
3329 	case ixgbe_mac_X540:
3330 	case ixgbe_mac_X550:
3331 	case ixgbe_mac_X550EM_x:
3332 	case ixgbe_mac_x550em_a:
3333 		if (eicr & IXGBE_EICR_ECC) {
3334 			e_info(link, "Received ECC Err, initiating reset\n");
3335 			set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3336 			ixgbe_service_event_schedule(adapter);
3337 			IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3338 		}
3339 		ixgbe_check_overtemp_event(adapter, eicr);
3340 		break;
3341 	default:
3342 		break;
3343 	}
3344 
3345 	ixgbe_check_fan_failure(adapter, eicr);
3346 	if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3347 		ixgbe_ptp_check_pps_event(adapter);
3348 
3349 	/* would disable interrupts here but EIAM disabled it */
3350 	napi_schedule_irqoff(&q_vector->napi);
3351 
3352 	/*
3353 	 * re-enable link(maybe) and non-queue interrupts, no flush.
3354 	 * ixgbe_poll will re-enable the queue interrupts
3355 	 */
3356 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
3357 		ixgbe_irq_enable(adapter, false, false);
3358 
3359 	return IRQ_HANDLED;
3360 }
3361 
3362 /**
3363  * ixgbe_request_irq - initialize interrupts
3364  * @adapter: board private structure
3365  *
3366  * Attempts to configure interrupts using the best available
3367  * capabilities of the hardware and kernel.
3368  **/
3369 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
3370 {
3371 	struct net_device *netdev = adapter->netdev;
3372 	int err;
3373 
3374 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3375 		err = ixgbe_request_msix_irqs(adapter);
3376 	else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
3377 		err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
3378 				  netdev->name, adapter);
3379 	else
3380 		err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
3381 				  netdev->name, adapter);
3382 
3383 	if (err)
3384 		e_err(probe, "request_irq failed, Error %d\n", err);
3385 
3386 	return err;
3387 }
3388 
3389 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
3390 {
3391 	int vector;
3392 
3393 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
3394 		free_irq(adapter->pdev->irq, adapter);
3395 		return;
3396 	}
3397 
3398 	if (!adapter->msix_entries)
3399 		return;
3400 
3401 	for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3402 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3403 		struct msix_entry *entry = &adapter->msix_entries[vector];
3404 
3405 		/* free only the irqs that were actually requested */
3406 		if (!q_vector->rx.ring && !q_vector->tx.ring)
3407 			continue;
3408 
3409 		/* clear the affinity_mask in the IRQ descriptor */
3410 		irq_set_affinity_hint(entry->vector, NULL);
3411 
3412 		free_irq(entry->vector, q_vector);
3413 	}
3414 
3415 	free_irq(adapter->msix_entries[vector].vector, adapter);
3416 }
3417 
3418 /**
3419  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3420  * @adapter: board private structure
3421  **/
3422 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3423 {
3424 	switch (adapter->hw.mac.type) {
3425 	case ixgbe_mac_82598EB:
3426 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3427 		break;
3428 	case ixgbe_mac_82599EB:
3429 	case ixgbe_mac_X540:
3430 	case ixgbe_mac_X550:
3431 	case ixgbe_mac_X550EM_x:
3432 	case ixgbe_mac_x550em_a:
3433 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3434 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3435 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3436 		break;
3437 	default:
3438 		break;
3439 	}
3440 	IXGBE_WRITE_FLUSH(&adapter->hw);
3441 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3442 		int vector;
3443 
3444 		for (vector = 0; vector < adapter->num_q_vectors; vector++)
3445 			synchronize_irq(adapter->msix_entries[vector].vector);
3446 
3447 		synchronize_irq(adapter->msix_entries[vector++].vector);
3448 	} else {
3449 		synchronize_irq(adapter->pdev->irq);
3450 	}
3451 }
3452 
3453 /**
3454  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3455  * @adapter: board private structure
3456  *
3457  **/
3458 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3459 {
3460 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3461 
3462 	ixgbe_write_eitr(q_vector);
3463 
3464 	ixgbe_set_ivar(adapter, 0, 0, 0);
3465 	ixgbe_set_ivar(adapter, 1, 0, 0);
3466 
3467 	e_info(hw, "Legacy interrupt IVAR setup done\n");
3468 }
3469 
3470 /**
3471  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3472  * @adapter: board private structure
3473  * @ring: structure containing ring specific data
3474  *
3475  * Configure the Tx descriptor ring after a reset.
3476  **/
3477 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3478 			     struct ixgbe_ring *ring)
3479 {
3480 	struct ixgbe_hw *hw = &adapter->hw;
3481 	u64 tdba = ring->dma;
3482 	int wait_loop = 10;
3483 	u32 txdctl = IXGBE_TXDCTL_ENABLE;
3484 	u8 reg_idx = ring->reg_idx;
3485 
3486 	ring->xsk_umem = NULL;
3487 	if (ring_is_xdp(ring))
3488 		ring->xsk_umem = ixgbe_xsk_umem(adapter, ring);
3489 
3490 	/* disable queue to avoid issues while updating state */
3491 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3492 	IXGBE_WRITE_FLUSH(hw);
3493 
3494 	IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3495 			(tdba & DMA_BIT_MASK(32)));
3496 	IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3497 	IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3498 			ring->count * sizeof(union ixgbe_adv_tx_desc));
3499 	IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3500 	IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3501 	ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3502 
3503 	/*
3504 	 * set WTHRESH to encourage burst writeback, it should not be set
3505 	 * higher than 1 when:
3506 	 * - ITR is 0 as it could cause false TX hangs
3507 	 * - ITR is set to > 100k int/sec and BQL is enabled
3508 	 *
3509 	 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3510 	 * to or less than the number of on chip descriptors, which is
3511 	 * currently 40.
3512 	 */
3513 	if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3514 		txdctl |= 1u << 16;	/* WTHRESH = 1 */
3515 	else
3516 		txdctl |= 8u << 16;	/* WTHRESH = 8 */
3517 
3518 	/*
3519 	 * Setting PTHRESH to 32 both improves performance
3520 	 * and avoids a TX hang with DFP enabled
3521 	 */
3522 	txdctl |= (1u << 8) |	/* HTHRESH = 1 */
3523 		   32;		/* PTHRESH = 32 */
3524 
3525 	/* reinitialize flowdirector state */
3526 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3527 		ring->atr_sample_rate = adapter->atr_sample_rate;
3528 		ring->atr_count = 0;
3529 		set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3530 	} else {
3531 		ring->atr_sample_rate = 0;
3532 	}
3533 
3534 	/* initialize XPS */
3535 	if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3536 		struct ixgbe_q_vector *q_vector = ring->q_vector;
3537 
3538 		if (q_vector)
3539 			netif_set_xps_queue(ring->netdev,
3540 					    &q_vector->affinity_mask,
3541 					    ring->queue_index);
3542 	}
3543 
3544 	clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3545 
3546 	/* reinitialize tx_buffer_info */
3547 	memset(ring->tx_buffer_info, 0,
3548 	       sizeof(struct ixgbe_tx_buffer) * ring->count);
3549 
3550 	/* enable queue */
3551 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3552 
3553 	/* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3554 	if (hw->mac.type == ixgbe_mac_82598EB &&
3555 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3556 		return;
3557 
3558 	/* poll to verify queue is enabled */
3559 	do {
3560 		usleep_range(1000, 2000);
3561 		txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3562 	} while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3563 	if (!wait_loop)
3564 		hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
3565 }
3566 
3567 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3568 {
3569 	struct ixgbe_hw *hw = &adapter->hw;
3570 	u32 rttdcs, mtqc;
3571 	u8 tcs = adapter->hw_tcs;
3572 
3573 	if (hw->mac.type == ixgbe_mac_82598EB)
3574 		return;
3575 
3576 	/* disable the arbiter while setting MTQC */
3577 	rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3578 	rttdcs |= IXGBE_RTTDCS_ARBDIS;
3579 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3580 
3581 	/* set transmit pool layout */
3582 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3583 		mtqc = IXGBE_MTQC_VT_ENA;
3584 		if (tcs > 4)
3585 			mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3586 		else if (tcs > 1)
3587 			mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3588 		else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3589 			 IXGBE_82599_VMDQ_4Q_MASK)
3590 			mtqc |= IXGBE_MTQC_32VF;
3591 		else
3592 			mtqc |= IXGBE_MTQC_64VF;
3593 	} else {
3594 		if (tcs > 4) {
3595 			mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3596 		} else if (tcs > 1) {
3597 			mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3598 		} else {
3599 			u8 max_txq = adapter->num_tx_queues +
3600 				adapter->num_xdp_queues;
3601 			if (max_txq > 63)
3602 				mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3603 			else
3604 				mtqc = IXGBE_MTQC_64Q_1PB;
3605 		}
3606 	}
3607 
3608 	IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3609 
3610 	/* Enable Security TX Buffer IFG for multiple pb */
3611 	if (tcs) {
3612 		u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3613 		sectx |= IXGBE_SECTX_DCB;
3614 		IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3615 	}
3616 
3617 	/* re-enable the arbiter */
3618 	rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3619 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3620 }
3621 
3622 /**
3623  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3624  * @adapter: board private structure
3625  *
3626  * Configure the Tx unit of the MAC after a reset.
3627  **/
3628 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3629 {
3630 	struct ixgbe_hw *hw = &adapter->hw;
3631 	u32 dmatxctl;
3632 	u32 i;
3633 
3634 	ixgbe_setup_mtqc(adapter);
3635 
3636 	if (hw->mac.type != ixgbe_mac_82598EB) {
3637 		/* DMATXCTL.EN must be before Tx queues are enabled */
3638 		dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3639 		dmatxctl |= IXGBE_DMATXCTL_TE;
3640 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3641 	}
3642 
3643 	/* Setup the HW Tx Head and Tail descriptor pointers */
3644 	for (i = 0; i < adapter->num_tx_queues; i++)
3645 		ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3646 	for (i = 0; i < adapter->num_xdp_queues; i++)
3647 		ixgbe_configure_tx_ring(adapter, adapter->xdp_ring[i]);
3648 }
3649 
3650 static void ixgbe_enable_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 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3663 				  struct ixgbe_ring *ring)
3664 {
3665 	struct ixgbe_hw *hw = &adapter->hw;
3666 	u8 reg_idx = ring->reg_idx;
3667 	u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3668 
3669 	srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3670 
3671 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3672 }
3673 
3674 #ifdef CONFIG_IXGBE_DCB
3675 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3676 #else
3677 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3678 #endif
3679 {
3680 	int i;
3681 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3682 
3683 	if (adapter->ixgbe_ieee_pfc)
3684 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3685 
3686 	/*
3687 	 * We should set the drop enable bit if:
3688 	 *  SR-IOV is enabled
3689 	 *   or
3690 	 *  Number of Rx queues > 1 and flow control is disabled
3691 	 *
3692 	 *  This allows us to avoid head of line blocking for security
3693 	 *  and performance reasons.
3694 	 */
3695 	if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3696 	    !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3697 		for (i = 0; i < adapter->num_rx_queues; i++)
3698 			ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3699 	} else {
3700 		for (i = 0; i < adapter->num_rx_queues; i++)
3701 			ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3702 	}
3703 }
3704 
3705 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3706 
3707 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3708 				   struct ixgbe_ring *rx_ring)
3709 {
3710 	struct ixgbe_hw *hw = &adapter->hw;
3711 	u32 srrctl;
3712 	u8 reg_idx = rx_ring->reg_idx;
3713 
3714 	if (hw->mac.type == ixgbe_mac_82598EB) {
3715 		u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3716 
3717 		/*
3718 		 * if VMDq is not active we must program one srrctl register
3719 		 * per RSS queue since we have enabled RDRXCTL.MVMEN
3720 		 */
3721 		reg_idx &= mask;
3722 	}
3723 
3724 	/* configure header buffer length, needed for RSC */
3725 	srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3726 
3727 	/* configure the packet buffer length */
3728 	if (rx_ring->xsk_umem) {
3729 		u32 xsk_buf_len = rx_ring->xsk_umem->chunk_size_nohr -
3730 				  XDP_PACKET_HEADROOM;
3731 
3732 		/* If the MAC support setting RXDCTL.RLPML, the
3733 		 * SRRCTL[n].BSIZEPKT is set to PAGE_SIZE and
3734 		 * RXDCTL.RLPML is set to the actual UMEM buffer
3735 		 * size. If not, then we are stuck with a 1k buffer
3736 		 * size resolution. In this case frames larger than
3737 		 * the UMEM buffer size viewed in a 1k resolution will
3738 		 * be dropped.
3739 		 */
3740 		if (hw->mac.type != ixgbe_mac_82599EB)
3741 			srrctl |= PAGE_SIZE >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3742 		else
3743 			srrctl |= xsk_buf_len >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3744 	} else if (test_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state)) {
3745 		srrctl |= IXGBE_RXBUFFER_3K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3746 	} else {
3747 		srrctl |= IXGBE_RXBUFFER_2K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3748 	}
3749 
3750 	/* configure descriptor type */
3751 	srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3752 
3753 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3754 }
3755 
3756 /**
3757  * ixgbe_rss_indir_tbl_entries - Return RSS indirection table entries
3758  * @adapter: device handle
3759  *
3760  *  - 82598/82599/X540:     128
3761  *  - X550(non-SRIOV mode): 512
3762  *  - X550(SRIOV mode):     64
3763  */
3764 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3765 {
3766 	if (adapter->hw.mac.type < ixgbe_mac_X550)
3767 		return 128;
3768 	else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3769 		return 64;
3770 	else
3771 		return 512;
3772 }
3773 
3774 /**
3775  * ixgbe_store_key - Write the RSS key to HW
3776  * @adapter: device handle
3777  *
3778  * Write the RSS key stored in adapter.rss_key to HW.
3779  */
3780 void ixgbe_store_key(struct ixgbe_adapter *adapter)
3781 {
3782 	struct ixgbe_hw *hw = &adapter->hw;
3783 	int i;
3784 
3785 	for (i = 0; i < 10; i++)
3786 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3787 }
3788 
3789 /**
3790  * ixgbe_init_rss_key - Initialize adapter RSS key
3791  * @adapter: device handle
3792  *
3793  * Allocates and initializes the RSS key if it is not allocated.
3794  **/
3795 static inline int ixgbe_init_rss_key(struct ixgbe_adapter *adapter)
3796 {
3797 	u32 *rss_key;
3798 
3799 	if (!adapter->rss_key) {
3800 		rss_key = kzalloc(IXGBE_RSS_KEY_SIZE, GFP_KERNEL);
3801 		if (unlikely(!rss_key))
3802 			return -ENOMEM;
3803 
3804 		netdev_rss_key_fill(rss_key, IXGBE_RSS_KEY_SIZE);
3805 		adapter->rss_key = rss_key;
3806 	}
3807 
3808 	return 0;
3809 }
3810 
3811 /**
3812  * ixgbe_store_reta - Write the RETA table to HW
3813  * @adapter: device handle
3814  *
3815  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3816  */
3817 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3818 {
3819 	u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3820 	struct ixgbe_hw *hw = &adapter->hw;
3821 	u32 reta = 0;
3822 	u32 indices_multi;
3823 	u8 *indir_tbl = adapter->rss_indir_tbl;
3824 
3825 	/* Fill out the redirection table as follows:
3826 	 *  - 82598:      8 bit wide entries containing pair of 4 bit RSS
3827 	 *    indices.
3828 	 *  - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3829 	 *  - X550:       8 bit wide entries containing 6 bit RSS index
3830 	 */
3831 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3832 		indices_multi = 0x11;
3833 	else
3834 		indices_multi = 0x1;
3835 
3836 	/* Write redirection table to HW */
3837 	for (i = 0; i < reta_entries; i++) {
3838 		reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3839 		if ((i & 3) == 3) {
3840 			if (i < 128)
3841 				IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3842 			else
3843 				IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3844 						reta);
3845 			reta = 0;
3846 		}
3847 	}
3848 }
3849 
3850 /**
3851  * ixgbe_store_vfreta - Write the RETA table to HW (x550 devices in SRIOV mode)
3852  * @adapter: device handle
3853  *
3854  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3855  */
3856 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3857 {
3858 	u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3859 	struct ixgbe_hw *hw = &adapter->hw;
3860 	u32 vfreta = 0;
3861 
3862 	/* Write redirection table to HW */
3863 	for (i = 0; i < reta_entries; i++) {
3864 		u16 pool = adapter->num_rx_pools;
3865 
3866 		vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3867 		if ((i & 3) != 3)
3868 			continue;
3869 
3870 		while (pool--)
3871 			IXGBE_WRITE_REG(hw,
3872 					IXGBE_PFVFRETA(i >> 2, VMDQ_P(pool)),
3873 					vfreta);
3874 		vfreta = 0;
3875 	}
3876 }
3877 
3878 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3879 {
3880 	u32 i, j;
3881 	u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3882 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3883 
3884 	/* Program table for at least 4 queues w/ SR-IOV so that VFs can
3885 	 * make full use of any rings they may have.  We will use the
3886 	 * PSRTYPE register to control how many rings we use within the PF.
3887 	 */
3888 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 4))
3889 		rss_i = 4;
3890 
3891 	/* Fill out hash function seeds */
3892 	ixgbe_store_key(adapter);
3893 
3894 	/* Fill out redirection table */
3895 	memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3896 
3897 	for (i = 0, j = 0; i < reta_entries; i++, j++) {
3898 		if (j == rss_i)
3899 			j = 0;
3900 
3901 		adapter->rss_indir_tbl[i] = j;
3902 	}
3903 
3904 	ixgbe_store_reta(adapter);
3905 }
3906 
3907 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3908 {
3909 	struct ixgbe_hw *hw = &adapter->hw;
3910 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3911 	int i, j;
3912 
3913 	/* Fill out hash function seeds */
3914 	for (i = 0; i < 10; i++) {
3915 		u16 pool = adapter->num_rx_pools;
3916 
3917 		while (pool--)
3918 			IXGBE_WRITE_REG(hw,
3919 					IXGBE_PFVFRSSRK(i, VMDQ_P(pool)),
3920 					*(adapter->rss_key + i));
3921 	}
3922 
3923 	/* Fill out the redirection table */
3924 	for (i = 0, j = 0; i < 64; i++, j++) {
3925 		if (j == rss_i)
3926 			j = 0;
3927 
3928 		adapter->rss_indir_tbl[i] = j;
3929 	}
3930 
3931 	ixgbe_store_vfreta(adapter);
3932 }
3933 
3934 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3935 {
3936 	struct ixgbe_hw *hw = &adapter->hw;
3937 	u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3938 	u32 rxcsum;
3939 
3940 	/* Disable indicating checksum in descriptor, enables RSS hash */
3941 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3942 	rxcsum |= IXGBE_RXCSUM_PCSD;
3943 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3944 
3945 	if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3946 		if (adapter->ring_feature[RING_F_RSS].mask)
3947 			mrqc = IXGBE_MRQC_RSSEN;
3948 	} else {
3949 		u8 tcs = adapter->hw_tcs;
3950 
3951 		if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3952 			if (tcs > 4)
3953 				mrqc = IXGBE_MRQC_VMDQRT8TCEN;	/* 8 TCs */
3954 			else if (tcs > 1)
3955 				mrqc = IXGBE_MRQC_VMDQRT4TCEN;	/* 4 TCs */
3956 			else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3957 				 IXGBE_82599_VMDQ_4Q_MASK)
3958 				mrqc = IXGBE_MRQC_VMDQRSS32EN;
3959 			else
3960 				mrqc = IXGBE_MRQC_VMDQRSS64EN;
3961 
3962 			/* Enable L3/L4 for Tx Switched packets only for X550,
3963 			 * older devices do not support this feature
3964 			 */
3965 			if (hw->mac.type >= ixgbe_mac_X550)
3966 				mrqc |= IXGBE_MRQC_L3L4TXSWEN;
3967 		} else {
3968 			if (tcs > 4)
3969 				mrqc = IXGBE_MRQC_RTRSS8TCEN;
3970 			else if (tcs > 1)
3971 				mrqc = IXGBE_MRQC_RTRSS4TCEN;
3972 			else
3973 				mrqc = IXGBE_MRQC_RSSEN;
3974 		}
3975 	}
3976 
3977 	/* Perform hash on these packet types */
3978 	rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3979 		     IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3980 		     IXGBE_MRQC_RSS_FIELD_IPV6 |
3981 		     IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3982 
3983 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3984 		rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3985 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3986 		rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3987 
3988 	if ((hw->mac.type >= ixgbe_mac_X550) &&
3989 	    (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
3990 		u16 pool = adapter->num_rx_pools;
3991 
3992 		/* Enable VF RSS mode */
3993 		mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
3994 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3995 
3996 		/* Setup RSS through the VF registers */
3997 		ixgbe_setup_vfreta(adapter);
3998 		vfmrqc = IXGBE_MRQC_RSSEN;
3999 		vfmrqc |= rss_field;
4000 
4001 		while (pool--)
4002 			IXGBE_WRITE_REG(hw,
4003 					IXGBE_PFVFMRQC(VMDQ_P(pool)),
4004 					vfmrqc);
4005 	} else {
4006 		ixgbe_setup_reta(adapter);
4007 		mrqc |= rss_field;
4008 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
4009 	}
4010 }
4011 
4012 /**
4013  * ixgbe_configure_rscctl - enable RSC for the indicated ring
4014  * @adapter: address of board private structure
4015  * @ring: structure containing ring specific data
4016  **/
4017 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
4018 				   struct ixgbe_ring *ring)
4019 {
4020 	struct ixgbe_hw *hw = &adapter->hw;
4021 	u32 rscctrl;
4022 	u8 reg_idx = ring->reg_idx;
4023 
4024 	if (!ring_is_rsc_enabled(ring))
4025 		return;
4026 
4027 	rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
4028 	rscctrl |= IXGBE_RSCCTL_RSCEN;
4029 	/*
4030 	 * we must limit the number of descriptors so that the
4031 	 * total size of max desc * buf_len is not greater
4032 	 * than 65536
4033 	 */
4034 	rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
4035 	IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
4036 }
4037 
4038 #define IXGBE_MAX_RX_DESC_POLL 10
4039 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
4040 				       struct ixgbe_ring *ring)
4041 {
4042 	struct ixgbe_hw *hw = &adapter->hw;
4043 	int wait_loop = IXGBE_MAX_RX_DESC_POLL;
4044 	u32 rxdctl;
4045 	u8 reg_idx = ring->reg_idx;
4046 
4047 	if (ixgbe_removed(hw->hw_addr))
4048 		return;
4049 	/* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
4050 	if (hw->mac.type == ixgbe_mac_82598EB &&
4051 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
4052 		return;
4053 
4054 	do {
4055 		usleep_range(1000, 2000);
4056 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4057 	} while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4058 
4059 	if (!wait_loop) {
4060 		e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
4061 		      "the polling period\n", reg_idx);
4062 	}
4063 }
4064 
4065 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
4066 			     struct ixgbe_ring *ring)
4067 {
4068 	struct ixgbe_hw *hw = &adapter->hw;
4069 	union ixgbe_adv_rx_desc *rx_desc;
4070 	u64 rdba = ring->dma;
4071 	u32 rxdctl;
4072 	u8 reg_idx = ring->reg_idx;
4073 
4074 	xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
4075 	ring->xsk_umem = ixgbe_xsk_umem(adapter, ring);
4076 	if (ring->xsk_umem) {
4077 		ring->zca.free = ixgbe_zca_free;
4078 		WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4079 						   MEM_TYPE_ZERO_COPY,
4080 						   &ring->zca));
4081 
4082 	} else {
4083 		WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4084 						   MEM_TYPE_PAGE_SHARED, NULL));
4085 	}
4086 
4087 	/* disable queue to avoid use of these values while updating state */
4088 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4089 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4090 
4091 	/* write value back with RXDCTL.ENABLE bit cleared */
4092 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4093 	IXGBE_WRITE_FLUSH(hw);
4094 
4095 	IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
4096 	IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
4097 	IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
4098 			ring->count * sizeof(union ixgbe_adv_rx_desc));
4099 	/* Force flushing of IXGBE_RDLEN to prevent MDD */
4100 	IXGBE_WRITE_FLUSH(hw);
4101 
4102 	IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
4103 	IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
4104 	ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
4105 
4106 	ixgbe_configure_srrctl(adapter, ring);
4107 	ixgbe_configure_rscctl(adapter, ring);
4108 
4109 	if (hw->mac.type == ixgbe_mac_82598EB) {
4110 		/*
4111 		 * enable cache line friendly hardware writes:
4112 		 * PTHRESH=32 descriptors (half the internal cache),
4113 		 * this also removes ugly rx_no_buffer_count increment
4114 		 * HTHRESH=4 descriptors (to minimize latency on fetch)
4115 		 * WTHRESH=8 burst writeback up to two cache lines
4116 		 */
4117 		rxdctl &= ~0x3FFFFF;
4118 		rxdctl |=  0x080420;
4119 #if (PAGE_SIZE < 8192)
4120 	/* RXDCTL.RLPML does not work on 82599 */
4121 	} else if (hw->mac.type != ixgbe_mac_82599EB) {
4122 		rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4123 			    IXGBE_RXDCTL_RLPML_EN);
4124 
4125 		/* Limit the maximum frame size so we don't overrun the skb.
4126 		 * This can happen in SRIOV mode when the MTU of the VF is
4127 		 * higher than the MTU of the PF.
4128 		 */
4129 		if (ring_uses_build_skb(ring) &&
4130 		    !test_bit(__IXGBE_RX_3K_BUFFER, &ring->state))
4131 			rxdctl |= IXGBE_MAX_2K_FRAME_BUILD_SKB |
4132 				  IXGBE_RXDCTL_RLPML_EN;
4133 #endif
4134 	}
4135 
4136 	if (ring->xsk_umem && hw->mac.type != ixgbe_mac_82599EB) {
4137 		u32 xsk_buf_len = ring->xsk_umem->chunk_size_nohr -
4138 				  XDP_PACKET_HEADROOM;
4139 
4140 		rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4141 			    IXGBE_RXDCTL_RLPML_EN);
4142 		rxdctl |= xsk_buf_len | IXGBE_RXDCTL_RLPML_EN;
4143 
4144 		ring->rx_buf_len = xsk_buf_len;
4145 	}
4146 
4147 	/* initialize rx_buffer_info */
4148 	memset(ring->rx_buffer_info, 0,
4149 	       sizeof(struct ixgbe_rx_buffer) * ring->count);
4150 
4151 	/* initialize Rx descriptor 0 */
4152 	rx_desc = IXGBE_RX_DESC(ring, 0);
4153 	rx_desc->wb.upper.length = 0;
4154 
4155 	/* enable receive descriptor ring */
4156 	rxdctl |= IXGBE_RXDCTL_ENABLE;
4157 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4158 
4159 	ixgbe_rx_desc_queue_enable(adapter, ring);
4160 	if (ring->xsk_umem)
4161 		ixgbe_alloc_rx_buffers_zc(ring, ixgbe_desc_unused(ring));
4162 	else
4163 		ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
4164 }
4165 
4166 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
4167 {
4168 	struct ixgbe_hw *hw = &adapter->hw;
4169 	int rss_i = adapter->ring_feature[RING_F_RSS].indices;
4170 	u16 pool = adapter->num_rx_pools;
4171 
4172 	/* PSRTYPE must be initialized in non 82598 adapters */
4173 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4174 		      IXGBE_PSRTYPE_UDPHDR |
4175 		      IXGBE_PSRTYPE_IPV4HDR |
4176 		      IXGBE_PSRTYPE_L2HDR |
4177 		      IXGBE_PSRTYPE_IPV6HDR;
4178 
4179 	if (hw->mac.type == ixgbe_mac_82598EB)
4180 		return;
4181 
4182 	if (rss_i > 3)
4183 		psrtype |= 2u << 29;
4184 	else if (rss_i > 1)
4185 		psrtype |= 1u << 29;
4186 
4187 	while (pool--)
4188 		IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4189 }
4190 
4191 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
4192 {
4193 	struct ixgbe_hw *hw = &adapter->hw;
4194 	u16 pool = adapter->num_rx_pools;
4195 	u32 reg_offset, vf_shift, vmolr;
4196 	u32 gcr_ext, vmdctl;
4197 	int i;
4198 
4199 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
4200 		return;
4201 
4202 	vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
4203 	vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
4204 	vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
4205 	vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
4206 	vmdctl |= IXGBE_VT_CTL_REPLEN;
4207 	IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
4208 
4209 	/* accept untagged packets until a vlan tag is
4210 	 * specifically set for the VMDQ queue/pool
4211 	 */
4212 	vmolr = IXGBE_VMOLR_AUPE;
4213 	while (pool--)
4214 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(pool)), vmolr);
4215 
4216 	vf_shift = VMDQ_P(0) % 32;
4217 	reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
4218 
4219 	/* Enable only the PF's pool for Tx/Rx */
4220 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), GENMASK(31, vf_shift));
4221 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
4222 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), GENMASK(31, vf_shift));
4223 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
4224 	if (adapter->bridge_mode == BRIDGE_MODE_VEB)
4225 		IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
4226 
4227 	/* Map PF MAC address in RAR Entry 0 to first pool following VFs */
4228 	hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
4229 
4230 	/* clear VLAN promisc flag so VFTA will be updated if necessary */
4231 	adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4232 
4233 	/*
4234 	 * Set up VF register offsets for selected VT Mode,
4235 	 * i.e. 32 or 64 VFs for SR-IOV
4236 	 */
4237 	switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4238 	case IXGBE_82599_VMDQ_8Q_MASK:
4239 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
4240 		break;
4241 	case IXGBE_82599_VMDQ_4Q_MASK:
4242 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
4243 		break;
4244 	default:
4245 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
4246 		break;
4247 	}
4248 
4249 	IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
4250 
4251 	for (i = 0; i < adapter->num_vfs; i++) {
4252 		/* configure spoof checking */
4253 		ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i,
4254 					  adapter->vfinfo[i].spoofchk_enabled);
4255 
4256 		/* Enable/Disable RSS query feature  */
4257 		ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
4258 					  adapter->vfinfo[i].rss_query_enabled);
4259 	}
4260 }
4261 
4262 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
4263 {
4264 	struct ixgbe_hw *hw = &adapter->hw;
4265 	struct net_device *netdev = adapter->netdev;
4266 	int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4267 	struct ixgbe_ring *rx_ring;
4268 	int i;
4269 	u32 mhadd, hlreg0;
4270 
4271 #ifdef IXGBE_FCOE
4272 	/* adjust max frame to be able to do baby jumbo for FCoE */
4273 	if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
4274 	    (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
4275 		max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4276 
4277 #endif /* IXGBE_FCOE */
4278 
4279 	/* adjust max frame to be at least the size of a standard frame */
4280 	if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
4281 		max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
4282 
4283 	mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
4284 	if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
4285 		mhadd &= ~IXGBE_MHADD_MFS_MASK;
4286 		mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
4287 
4288 		IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
4289 	}
4290 
4291 	hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4292 	/* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
4293 	hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4294 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4295 
4296 	/*
4297 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
4298 	 * the Base and Length of the Rx Descriptor Ring
4299 	 */
4300 	for (i = 0; i < adapter->num_rx_queues; i++) {
4301 		rx_ring = adapter->rx_ring[i];
4302 
4303 		clear_ring_rsc_enabled(rx_ring);
4304 		clear_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4305 		clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4306 
4307 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4308 			set_ring_rsc_enabled(rx_ring);
4309 
4310 		if (test_bit(__IXGBE_RX_FCOE, &rx_ring->state))
4311 			set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4312 
4313 		if (adapter->flags2 & IXGBE_FLAG2_RX_LEGACY)
4314 			continue;
4315 
4316 		set_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4317 
4318 #if (PAGE_SIZE < 8192)
4319 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4320 			set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4321 
4322 		if (IXGBE_2K_TOO_SMALL_WITH_PADDING ||
4323 		    (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
4324 			set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4325 #endif
4326 	}
4327 }
4328 
4329 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
4330 {
4331 	struct ixgbe_hw *hw = &adapter->hw;
4332 	u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4333 
4334 	switch (hw->mac.type) {
4335 	case ixgbe_mac_82598EB:
4336 		/*
4337 		 * For VMDq support of different descriptor types or
4338 		 * buffer sizes through the use of multiple SRRCTL
4339 		 * registers, RDRXCTL.MVMEN must be set to 1
4340 		 *
4341 		 * also, the manual doesn't mention it clearly but DCA hints
4342 		 * will only use queue 0's tags unless this bit is set.  Side
4343 		 * effects of setting this bit are only that SRRCTL must be
4344 		 * fully programmed [0..15]
4345 		 */
4346 		rdrxctl |= IXGBE_RDRXCTL_MVMEN;
4347 		break;
4348 	case ixgbe_mac_X550:
4349 	case ixgbe_mac_X550EM_x:
4350 	case ixgbe_mac_x550em_a:
4351 		if (adapter->num_vfs)
4352 			rdrxctl |= IXGBE_RDRXCTL_PSP;
4353 		/* fall through */
4354 	case ixgbe_mac_82599EB:
4355 	case ixgbe_mac_X540:
4356 		/* Disable RSC for ACK packets */
4357 		IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
4358 		   (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
4359 		rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4360 		/* hardware requires some bits to be set by default */
4361 		rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
4362 		rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4363 		break;
4364 	default:
4365 		/* We should do nothing since we don't know this hardware */
4366 		return;
4367 	}
4368 
4369 	IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4370 }
4371 
4372 /**
4373  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
4374  * @adapter: board private structure
4375  *
4376  * Configure the Rx unit of the MAC after a reset.
4377  **/
4378 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
4379 {
4380 	struct ixgbe_hw *hw = &adapter->hw;
4381 	int i;
4382 	u32 rxctrl, rfctl;
4383 
4384 	/* disable receives while setting up the descriptors */
4385 	hw->mac.ops.disable_rx(hw);
4386 
4387 	ixgbe_setup_psrtype(adapter);
4388 	ixgbe_setup_rdrxctl(adapter);
4389 
4390 	/* RSC Setup */
4391 	rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4392 	rfctl &= ~IXGBE_RFCTL_RSC_DIS;
4393 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
4394 		rfctl |= IXGBE_RFCTL_RSC_DIS;
4395 
4396 	/* disable NFS filtering */
4397 	rfctl |= (IXGBE_RFCTL_NFSW_DIS | IXGBE_RFCTL_NFSR_DIS);
4398 	IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4399 
4400 	/* Program registers for the distribution of queues */
4401 	ixgbe_setup_mrqc(adapter);
4402 
4403 	/* set_rx_buffer_len must be called before ring initialization */
4404 	ixgbe_set_rx_buffer_len(adapter);
4405 
4406 	/*
4407 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
4408 	 * the Base and Length of the Rx Descriptor Ring
4409 	 */
4410 	for (i = 0; i < adapter->num_rx_queues; i++)
4411 		ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
4412 
4413 	rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4414 	/* disable drop enable for 82598 parts */
4415 	if (hw->mac.type == ixgbe_mac_82598EB)
4416 		rxctrl |= IXGBE_RXCTRL_DMBYPS;
4417 
4418 	/* enable all receives */
4419 	rxctrl |= IXGBE_RXCTRL_RXEN;
4420 	hw->mac.ops.enable_rx_dma(hw, rxctrl);
4421 }
4422 
4423 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
4424 				 __be16 proto, u16 vid)
4425 {
4426 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4427 	struct ixgbe_hw *hw = &adapter->hw;
4428 
4429 	/* add VID to filter table */
4430 	if (!vid || !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4431 		hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true, !!vid);
4432 
4433 	set_bit(vid, adapter->active_vlans);
4434 
4435 	return 0;
4436 }
4437 
4438 static int ixgbe_find_vlvf_entry(struct ixgbe_hw *hw, u32 vlan)
4439 {
4440 	u32 vlvf;
4441 	int idx;
4442 
4443 	/* short cut the special case */
4444 	if (vlan == 0)
4445 		return 0;
4446 
4447 	/* Search for the vlan id in the VLVF entries */
4448 	for (idx = IXGBE_VLVF_ENTRIES; --idx;) {
4449 		vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(idx));
4450 		if ((vlvf & VLAN_VID_MASK) == vlan)
4451 			break;
4452 	}
4453 
4454 	return idx;
4455 }
4456 
4457 void ixgbe_update_pf_promisc_vlvf(struct ixgbe_adapter *adapter, u32 vid)
4458 {
4459 	struct ixgbe_hw *hw = &adapter->hw;
4460 	u32 bits, word;
4461 	int idx;
4462 
4463 	idx = ixgbe_find_vlvf_entry(hw, vid);
4464 	if (!idx)
4465 		return;
4466 
4467 	/* See if any other pools are set for this VLAN filter
4468 	 * entry other than the PF.
4469 	 */
4470 	word = idx * 2 + (VMDQ_P(0) / 32);
4471 	bits = ~BIT(VMDQ_P(0) % 32);
4472 	bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4473 
4474 	/* Disable the filter so this falls into the default pool. */
4475 	if (!bits && !IXGBE_READ_REG(hw, IXGBE_VLVFB(word ^ 1))) {
4476 		if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4477 			IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), 0);
4478 		IXGBE_WRITE_REG(hw, IXGBE_VLVF(idx), 0);
4479 	}
4480 }
4481 
4482 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
4483 				  __be16 proto, u16 vid)
4484 {
4485 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4486 	struct ixgbe_hw *hw = &adapter->hw;
4487 
4488 	/* remove VID from filter table */
4489 	if (vid && !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4490 		hw->mac.ops.set_vfta(hw, vid, VMDQ_P(0), false, true);
4491 
4492 	clear_bit(vid, adapter->active_vlans);
4493 
4494 	return 0;
4495 }
4496 
4497 /**
4498  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
4499  * @adapter: driver data
4500  */
4501 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
4502 {
4503 	struct ixgbe_hw *hw = &adapter->hw;
4504 	u32 vlnctrl;
4505 	int i, j;
4506 
4507 	switch (hw->mac.type) {
4508 	case ixgbe_mac_82598EB:
4509 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4510 		vlnctrl &= ~IXGBE_VLNCTRL_VME;
4511 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4512 		break;
4513 	case ixgbe_mac_82599EB:
4514 	case ixgbe_mac_X540:
4515 	case ixgbe_mac_X550:
4516 	case ixgbe_mac_X550EM_x:
4517 	case ixgbe_mac_x550em_a:
4518 		for (i = 0; i < adapter->num_rx_queues; i++) {
4519 			struct ixgbe_ring *ring = adapter->rx_ring[i];
4520 
4521 			if (!netif_is_ixgbe(ring->netdev))
4522 				continue;
4523 
4524 			j = ring->reg_idx;
4525 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4526 			vlnctrl &= ~IXGBE_RXDCTL_VME;
4527 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4528 		}
4529 		break;
4530 	default:
4531 		break;
4532 	}
4533 }
4534 
4535 /**
4536  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
4537  * @adapter: driver data
4538  */
4539 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
4540 {
4541 	struct ixgbe_hw *hw = &adapter->hw;
4542 	u32 vlnctrl;
4543 	int i, j;
4544 
4545 	switch (hw->mac.type) {
4546 	case ixgbe_mac_82598EB:
4547 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4548 		vlnctrl |= IXGBE_VLNCTRL_VME;
4549 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4550 		break;
4551 	case ixgbe_mac_82599EB:
4552 	case ixgbe_mac_X540:
4553 	case ixgbe_mac_X550:
4554 	case ixgbe_mac_X550EM_x:
4555 	case ixgbe_mac_x550em_a:
4556 		for (i = 0; i < adapter->num_rx_queues; i++) {
4557 			struct ixgbe_ring *ring = adapter->rx_ring[i];
4558 
4559 			if (!netif_is_ixgbe(ring->netdev))
4560 				continue;
4561 
4562 			j = ring->reg_idx;
4563 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4564 			vlnctrl |= IXGBE_RXDCTL_VME;
4565 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4566 		}
4567 		break;
4568 	default:
4569 		break;
4570 	}
4571 }
4572 
4573 static void ixgbe_vlan_promisc_enable(struct ixgbe_adapter *adapter)
4574 {
4575 	struct ixgbe_hw *hw = &adapter->hw;
4576 	u32 vlnctrl, i;
4577 
4578 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4579 
4580 	if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
4581 	/* For VMDq and SR-IOV we must leave VLAN filtering enabled */
4582 		vlnctrl |= IXGBE_VLNCTRL_VFE;
4583 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4584 	} else {
4585 		vlnctrl &= ~IXGBE_VLNCTRL_VFE;
4586 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4587 		return;
4588 	}
4589 
4590 	/* Nothing to do for 82598 */
4591 	if (hw->mac.type == ixgbe_mac_82598EB)
4592 		return;
4593 
4594 	/* We are already in VLAN promisc, nothing to do */
4595 	if (adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)
4596 		return;
4597 
4598 	/* Set flag so we don't redo unnecessary work */
4599 	adapter->flags2 |= IXGBE_FLAG2_VLAN_PROMISC;
4600 
4601 	/* Add PF to all active pools */
4602 	for (i = IXGBE_VLVF_ENTRIES; --i;) {
4603 		u32 reg_offset = IXGBE_VLVFB(i * 2 + VMDQ_P(0) / 32);
4604 		u32 vlvfb = IXGBE_READ_REG(hw, reg_offset);
4605 
4606 		vlvfb |= BIT(VMDQ_P(0) % 32);
4607 		IXGBE_WRITE_REG(hw, reg_offset, vlvfb);
4608 	}
4609 
4610 	/* Set all bits in the VLAN filter table array */
4611 	for (i = hw->mac.vft_size; i--;)
4612 		IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), ~0U);
4613 }
4614 
4615 #define VFTA_BLOCK_SIZE 8
4616 static void ixgbe_scrub_vfta(struct ixgbe_adapter *adapter, u32 vfta_offset)
4617 {
4618 	struct ixgbe_hw *hw = &adapter->hw;
4619 	u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4620 	u32 vid_start = vfta_offset * 32;
4621 	u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4622 	u32 i, vid, word, bits;
4623 
4624 	for (i = IXGBE_VLVF_ENTRIES; --i;) {
4625 		u32 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(i));
4626 
4627 		/* pull VLAN ID from VLVF */
4628 		vid = vlvf & VLAN_VID_MASK;
4629 
4630 		/* only concern outselves with a certain range */
4631 		if (vid < vid_start || vid >= vid_end)
4632 			continue;
4633 
4634 		if (vlvf) {
4635 			/* record VLAN ID in VFTA */
4636 			vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4637 
4638 			/* if PF is part of this then continue */
4639 			if (test_bit(vid, adapter->active_vlans))
4640 				continue;
4641 		}
4642 
4643 		/* remove PF from the pool */
4644 		word = i * 2 + VMDQ_P(0) / 32;
4645 		bits = ~BIT(VMDQ_P(0) % 32);
4646 		bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4647 		IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), bits);
4648 	}
4649 
4650 	/* extract values from active_vlans and write back to VFTA */
4651 	for (i = VFTA_BLOCK_SIZE; i--;) {
4652 		vid = (vfta_offset + i) * 32;
4653 		word = vid / BITS_PER_LONG;
4654 		bits = vid % BITS_PER_LONG;
4655 
4656 		vfta[i] |= adapter->active_vlans[word] >> bits;
4657 
4658 		IXGBE_WRITE_REG(hw, IXGBE_VFTA(vfta_offset + i), vfta[i]);
4659 	}
4660 }
4661 
4662 static void ixgbe_vlan_promisc_disable(struct ixgbe_adapter *adapter)
4663 {
4664 	struct ixgbe_hw *hw = &adapter->hw;
4665 	u32 vlnctrl, i;
4666 
4667 	/* Set VLAN filtering to enabled */
4668 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4669 	vlnctrl |= IXGBE_VLNCTRL_VFE;
4670 	IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4671 
4672 	if (!(adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) ||
4673 	    hw->mac.type == ixgbe_mac_82598EB)
4674 		return;
4675 
4676 	/* We are not in VLAN promisc, nothing to do */
4677 	if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4678 		return;
4679 
4680 	/* Set flag so we don't redo unnecessary work */
4681 	adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4682 
4683 	for (i = 0; i < hw->mac.vft_size; i += VFTA_BLOCK_SIZE)
4684 		ixgbe_scrub_vfta(adapter, i);
4685 }
4686 
4687 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
4688 {
4689 	u16 vid = 1;
4690 
4691 	ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
4692 
4693 	for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
4694 		ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
4695 }
4696 
4697 /**
4698  * ixgbe_write_mc_addr_list - write multicast addresses to MTA
4699  * @netdev: network interface device structure
4700  *
4701  * Writes multicast address list to the MTA hash table.
4702  * Returns: -ENOMEM on failure
4703  *                0 on no addresses written
4704  *                X on writing X addresses to MTA
4705  **/
4706 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
4707 {
4708 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4709 	struct ixgbe_hw *hw = &adapter->hw;
4710 
4711 	if (!netif_running(netdev))
4712 		return 0;
4713 
4714 	if (hw->mac.ops.update_mc_addr_list)
4715 		hw->mac.ops.update_mc_addr_list(hw, netdev);
4716 	else
4717 		return -ENOMEM;
4718 
4719 #ifdef CONFIG_PCI_IOV
4720 	ixgbe_restore_vf_multicasts(adapter);
4721 #endif
4722 
4723 	return netdev_mc_count(netdev);
4724 }
4725 
4726 #ifdef CONFIG_PCI_IOV
4727 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4728 {
4729 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4730 	struct ixgbe_hw *hw = &adapter->hw;
4731 	int i;
4732 
4733 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4734 		mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4735 
4736 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4737 			hw->mac.ops.set_rar(hw, i,
4738 					    mac_table->addr,
4739 					    mac_table->pool,
4740 					    IXGBE_RAH_AV);
4741 		else
4742 			hw->mac.ops.clear_rar(hw, i);
4743 	}
4744 }
4745 
4746 #endif
4747 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4748 {
4749 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4750 	struct ixgbe_hw *hw = &adapter->hw;
4751 	int i;
4752 
4753 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4754 		if (!(mac_table->state & IXGBE_MAC_STATE_MODIFIED))
4755 			continue;
4756 
4757 		mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4758 
4759 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4760 			hw->mac.ops.set_rar(hw, i,
4761 					    mac_table->addr,
4762 					    mac_table->pool,
4763 					    IXGBE_RAH_AV);
4764 		else
4765 			hw->mac.ops.clear_rar(hw, i);
4766 	}
4767 }
4768 
4769 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4770 {
4771 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4772 	struct ixgbe_hw *hw = &adapter->hw;
4773 	int i;
4774 
4775 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4776 		mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4777 		mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4778 	}
4779 
4780 	ixgbe_sync_mac_table(adapter);
4781 }
4782 
4783 static int ixgbe_available_rars(struct ixgbe_adapter *adapter, u16 pool)
4784 {
4785 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4786 	struct ixgbe_hw *hw = &adapter->hw;
4787 	int i, count = 0;
4788 
4789 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4790 		/* do not count default RAR as available */
4791 		if (mac_table->state & IXGBE_MAC_STATE_DEFAULT)
4792 			continue;
4793 
4794 		/* only count unused and addresses that belong to us */
4795 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE) {
4796 			if (mac_table->pool != pool)
4797 				continue;
4798 		}
4799 
4800 		count++;
4801 	}
4802 
4803 	return count;
4804 }
4805 
4806 /* this function destroys the first RAR entry */
4807 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter)
4808 {
4809 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4810 	struct ixgbe_hw *hw = &adapter->hw;
4811 
4812 	memcpy(&mac_table->addr, hw->mac.addr, ETH_ALEN);
4813 	mac_table->pool = VMDQ_P(0);
4814 
4815 	mac_table->state = IXGBE_MAC_STATE_DEFAULT | IXGBE_MAC_STATE_IN_USE;
4816 
4817 	hw->mac.ops.set_rar(hw, 0, mac_table->addr, mac_table->pool,
4818 			    IXGBE_RAH_AV);
4819 }
4820 
4821 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter,
4822 			 const u8 *addr, u16 pool)
4823 {
4824 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4825 	struct ixgbe_hw *hw = &adapter->hw;
4826 	int i;
4827 
4828 	if (is_zero_ether_addr(addr))
4829 		return -EINVAL;
4830 
4831 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4832 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4833 			continue;
4834 
4835 		ether_addr_copy(mac_table->addr, addr);
4836 		mac_table->pool = pool;
4837 
4838 		mac_table->state |= IXGBE_MAC_STATE_MODIFIED |
4839 				    IXGBE_MAC_STATE_IN_USE;
4840 
4841 		ixgbe_sync_mac_table(adapter);
4842 
4843 		return i;
4844 	}
4845 
4846 	return -ENOMEM;
4847 }
4848 
4849 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter,
4850 			 const u8 *addr, u16 pool)
4851 {
4852 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4853 	struct ixgbe_hw *hw = &adapter->hw;
4854 	int i;
4855 
4856 	if (is_zero_ether_addr(addr))
4857 		return -EINVAL;
4858 
4859 	/* search table for addr, if found clear IN_USE flag and sync */
4860 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4861 		/* we can only delete an entry if it is in use */
4862 		if (!(mac_table->state & IXGBE_MAC_STATE_IN_USE))
4863 			continue;
4864 		/* we only care about entries that belong to the given pool */
4865 		if (mac_table->pool != pool)
4866 			continue;
4867 		/* we only care about a specific MAC address */
4868 		if (!ether_addr_equal(addr, mac_table->addr))
4869 			continue;
4870 
4871 		mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4872 		mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4873 
4874 		ixgbe_sync_mac_table(adapter);
4875 
4876 		return 0;
4877 	}
4878 
4879 	return -ENOMEM;
4880 }
4881 
4882 static int ixgbe_uc_sync(struct net_device *netdev, const unsigned char *addr)
4883 {
4884 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4885 	int ret;
4886 
4887 	ret = ixgbe_add_mac_filter(adapter, addr, VMDQ_P(0));
4888 
4889 	return min_t(int, ret, 0);
4890 }
4891 
4892 static int ixgbe_uc_unsync(struct net_device *netdev, const unsigned char *addr)
4893 {
4894 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4895 
4896 	ixgbe_del_mac_filter(adapter, addr, VMDQ_P(0));
4897 
4898 	return 0;
4899 }
4900 
4901 /**
4902  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4903  * @netdev: network interface device structure
4904  *
4905  * The set_rx_method entry point is called whenever the unicast/multicast
4906  * address list or the network interface flags are updated.  This routine is
4907  * responsible for configuring the hardware for proper unicast, multicast and
4908  * promiscuous mode.
4909  **/
4910 void ixgbe_set_rx_mode(struct net_device *netdev)
4911 {
4912 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4913 	struct ixgbe_hw *hw = &adapter->hw;
4914 	u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4915 	netdev_features_t features = netdev->features;
4916 	int count;
4917 
4918 	/* Check for Promiscuous and All Multicast modes */
4919 	fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4920 
4921 	/* set all bits that we expect to always be set */
4922 	fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4923 	fctrl |= IXGBE_FCTRL_BAM;
4924 	fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4925 	fctrl |= IXGBE_FCTRL_PMCF;
4926 
4927 	/* clear the bits we are changing the status of */
4928 	fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4929 	if (netdev->flags & IFF_PROMISC) {
4930 		hw->addr_ctrl.user_set_promisc = true;
4931 		fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4932 		vmolr |= IXGBE_VMOLR_MPE;
4933 		features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4934 	} else {
4935 		if (netdev->flags & IFF_ALLMULTI) {
4936 			fctrl |= IXGBE_FCTRL_MPE;
4937 			vmolr |= IXGBE_VMOLR_MPE;
4938 		}
4939 		hw->addr_ctrl.user_set_promisc = false;
4940 	}
4941 
4942 	/*
4943 	 * Write addresses to available RAR registers, if there is not
4944 	 * sufficient space to store all the addresses then enable
4945 	 * unicast promiscuous mode
4946 	 */
4947 	if (__dev_uc_sync(netdev, ixgbe_uc_sync, ixgbe_uc_unsync)) {
4948 		fctrl |= IXGBE_FCTRL_UPE;
4949 		vmolr |= IXGBE_VMOLR_ROPE;
4950 	}
4951 
4952 	/* Write addresses to the MTA, if the attempt fails
4953 	 * then we should just turn on promiscuous mode so
4954 	 * that we can at least receive multicast traffic
4955 	 */
4956 	count = ixgbe_write_mc_addr_list(netdev);
4957 	if (count < 0) {
4958 		fctrl |= IXGBE_FCTRL_MPE;
4959 		vmolr |= IXGBE_VMOLR_MPE;
4960 	} else if (count) {
4961 		vmolr |= IXGBE_VMOLR_ROMPE;
4962 	}
4963 
4964 	if (hw->mac.type != ixgbe_mac_82598EB) {
4965 		vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4966 			 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4967 			   IXGBE_VMOLR_ROPE);
4968 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4969 	}
4970 
4971 	/* This is useful for sniffing bad packets. */
4972 	if (features & NETIF_F_RXALL) {
4973 		/* UPE and MPE will be handled by normal PROMISC logic
4974 		 * in e1000e_set_rx_mode */
4975 		fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4976 			  IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4977 			  IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4978 
4979 		fctrl &= ~(IXGBE_FCTRL_DPF);
4980 		/* NOTE:  VLAN filtering is disabled by setting PROMISC */
4981 	}
4982 
4983 	IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4984 
4985 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
4986 		ixgbe_vlan_strip_enable(adapter);
4987 	else
4988 		ixgbe_vlan_strip_disable(adapter);
4989 
4990 	if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
4991 		ixgbe_vlan_promisc_disable(adapter);
4992 	else
4993 		ixgbe_vlan_promisc_enable(adapter);
4994 }
4995 
4996 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
4997 {
4998 	int q_idx;
4999 
5000 	for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
5001 		napi_enable(&adapter->q_vector[q_idx]->napi);
5002 }
5003 
5004 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
5005 {
5006 	int q_idx;
5007 
5008 	for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
5009 		napi_disable(&adapter->q_vector[q_idx]->napi);
5010 }
5011 
5012 static void ixgbe_clear_udp_tunnel_port(struct ixgbe_adapter *adapter, u32 mask)
5013 {
5014 	struct ixgbe_hw *hw = &adapter->hw;
5015 	u32 vxlanctrl;
5016 
5017 	if (!(adapter->flags & (IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE |
5018 				IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE)))
5019 		return;
5020 
5021 	vxlanctrl = IXGBE_READ_REG(hw, IXGBE_VXLANCTRL) & ~mask;
5022 	IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, vxlanctrl);
5023 
5024 	if (mask & IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK)
5025 		adapter->vxlan_port = 0;
5026 
5027 	if (mask & IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK)
5028 		adapter->geneve_port = 0;
5029 }
5030 
5031 #ifdef CONFIG_IXGBE_DCB
5032 /**
5033  * ixgbe_configure_dcb - Configure DCB hardware
5034  * @adapter: ixgbe adapter struct
5035  *
5036  * This is called by the driver on open to configure the DCB hardware.
5037  * This is also called by the gennetlink interface when reconfiguring
5038  * the DCB state.
5039  */
5040 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
5041 {
5042 	struct ixgbe_hw *hw = &adapter->hw;
5043 	int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
5044 
5045 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
5046 		if (hw->mac.type == ixgbe_mac_82598EB)
5047 			netif_set_gso_max_size(adapter->netdev, 65536);
5048 		return;
5049 	}
5050 
5051 	if (hw->mac.type == ixgbe_mac_82598EB)
5052 		netif_set_gso_max_size(adapter->netdev, 32768);
5053 
5054 #ifdef IXGBE_FCOE
5055 	if (adapter->netdev->features & NETIF_F_FCOE_MTU)
5056 		max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
5057 #endif
5058 
5059 	/* reconfigure the hardware */
5060 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
5061 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5062 						DCB_TX_CONFIG);
5063 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5064 						DCB_RX_CONFIG);
5065 		ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
5066 	} else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
5067 		ixgbe_dcb_hw_ets(&adapter->hw,
5068 				 adapter->ixgbe_ieee_ets,
5069 				 max_frame);
5070 		ixgbe_dcb_hw_pfc_config(&adapter->hw,
5071 					adapter->ixgbe_ieee_pfc->pfc_en,
5072 					adapter->ixgbe_ieee_ets->prio_tc);
5073 	}
5074 
5075 	/* Enable RSS Hash per TC */
5076 	if (hw->mac.type != ixgbe_mac_82598EB) {
5077 		u32 msb = 0;
5078 		u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
5079 
5080 		while (rss_i) {
5081 			msb++;
5082 			rss_i >>= 1;
5083 		}
5084 
5085 		/* write msb to all 8 TCs in one write */
5086 		IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
5087 	}
5088 }
5089 #endif
5090 
5091 /* Additional bittime to account for IXGBE framing */
5092 #define IXGBE_ETH_FRAMING 20
5093 
5094 /**
5095  * ixgbe_hpbthresh - calculate high water mark for flow control
5096  *
5097  * @adapter: board private structure to calculate for
5098  * @pb: packet buffer to calculate
5099  */
5100 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
5101 {
5102 	struct ixgbe_hw *hw = &adapter->hw;
5103 	struct net_device *dev = adapter->netdev;
5104 	int link, tc, kb, marker;
5105 	u32 dv_id, rx_pba;
5106 
5107 	/* Calculate max LAN frame size */
5108 	tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
5109 
5110 #ifdef IXGBE_FCOE
5111 	/* FCoE traffic class uses FCOE jumbo frames */
5112 	if ((dev->features & NETIF_F_FCOE_MTU) &&
5113 	    (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5114 	    (pb == ixgbe_fcoe_get_tc(adapter)))
5115 		tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5116 #endif
5117 
5118 	/* Calculate delay value for device */
5119 	switch (hw->mac.type) {
5120 	case ixgbe_mac_X540:
5121 	case ixgbe_mac_X550:
5122 	case ixgbe_mac_X550EM_x:
5123 	case ixgbe_mac_x550em_a:
5124 		dv_id = IXGBE_DV_X540(link, tc);
5125 		break;
5126 	default:
5127 		dv_id = IXGBE_DV(link, tc);
5128 		break;
5129 	}
5130 
5131 	/* Loopback switch introduces additional latency */
5132 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5133 		dv_id += IXGBE_B2BT(tc);
5134 
5135 	/* Delay value is calculated in bit times convert to KB */
5136 	kb = IXGBE_BT2KB(dv_id);
5137 	rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
5138 
5139 	marker = rx_pba - kb;
5140 
5141 	/* It is possible that the packet buffer is not large enough
5142 	 * to provide required headroom. In this case throw an error
5143 	 * to user and a do the best we can.
5144 	 */
5145 	if (marker < 0) {
5146 		e_warn(drv, "Packet Buffer(%i) can not provide enough"
5147 			    "headroom to support flow control."
5148 			    "Decrease MTU or number of traffic classes\n", pb);
5149 		marker = tc + 1;
5150 	}
5151 
5152 	return marker;
5153 }
5154 
5155 /**
5156  * ixgbe_lpbthresh - calculate low water mark for for flow control
5157  *
5158  * @adapter: board private structure to calculate for
5159  * @pb: packet buffer to calculate
5160  */
5161 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
5162 {
5163 	struct ixgbe_hw *hw = &adapter->hw;
5164 	struct net_device *dev = adapter->netdev;
5165 	int tc;
5166 	u32 dv_id;
5167 
5168 	/* Calculate max LAN frame size */
5169 	tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5170 
5171 #ifdef IXGBE_FCOE
5172 	/* FCoE traffic class uses FCOE jumbo frames */
5173 	if ((dev->features & NETIF_F_FCOE_MTU) &&
5174 	    (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5175 	    (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
5176 		tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5177 #endif
5178 
5179 	/* Calculate delay value for device */
5180 	switch (hw->mac.type) {
5181 	case ixgbe_mac_X540:
5182 	case ixgbe_mac_X550:
5183 	case ixgbe_mac_X550EM_x:
5184 	case ixgbe_mac_x550em_a:
5185 		dv_id = IXGBE_LOW_DV_X540(tc);
5186 		break;
5187 	default:
5188 		dv_id = IXGBE_LOW_DV(tc);
5189 		break;
5190 	}
5191 
5192 	/* Delay value is calculated in bit times convert to KB */
5193 	return IXGBE_BT2KB(dv_id);
5194 }
5195 
5196 /*
5197  * ixgbe_pbthresh_setup - calculate and setup high low water marks
5198  */
5199 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
5200 {
5201 	struct ixgbe_hw *hw = &adapter->hw;
5202 	int num_tc = adapter->hw_tcs;
5203 	int i;
5204 
5205 	if (!num_tc)
5206 		num_tc = 1;
5207 
5208 	for (i = 0; i < num_tc; i++) {
5209 		hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
5210 		hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
5211 
5212 		/* Low water marks must not be larger than high water marks */
5213 		if (hw->fc.low_water[i] > hw->fc.high_water[i])
5214 			hw->fc.low_water[i] = 0;
5215 	}
5216 
5217 	for (; i < MAX_TRAFFIC_CLASS; i++)
5218 		hw->fc.high_water[i] = 0;
5219 }
5220 
5221 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
5222 {
5223 	struct ixgbe_hw *hw = &adapter->hw;
5224 	int hdrm;
5225 	u8 tc = adapter->hw_tcs;
5226 
5227 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5228 	    adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5229 		hdrm = 32 << adapter->fdir_pballoc;
5230 	else
5231 		hdrm = 0;
5232 
5233 	hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
5234 	ixgbe_pbthresh_setup(adapter);
5235 }
5236 
5237 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
5238 {
5239 	struct ixgbe_hw *hw = &adapter->hw;
5240 	struct hlist_node *node2;
5241 	struct ixgbe_fdir_filter *filter;
5242 	u64 action;
5243 
5244 	spin_lock(&adapter->fdir_perfect_lock);
5245 
5246 	if (!hlist_empty(&adapter->fdir_filter_list))
5247 		ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
5248 
5249 	hlist_for_each_entry_safe(filter, node2,
5250 				  &adapter->fdir_filter_list, fdir_node) {
5251 		action = filter->action;
5252 		if (action != IXGBE_FDIR_DROP_QUEUE && action != 0)
5253 			action =
5254 			(action >> ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF) - 1;
5255 
5256 		ixgbe_fdir_write_perfect_filter_82599(hw,
5257 				&filter->filter,
5258 				filter->sw_idx,
5259 				(action == IXGBE_FDIR_DROP_QUEUE) ?
5260 				IXGBE_FDIR_DROP_QUEUE :
5261 				adapter->rx_ring[action]->reg_idx);
5262 	}
5263 
5264 	spin_unlock(&adapter->fdir_perfect_lock);
5265 }
5266 
5267 /**
5268  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
5269  * @rx_ring: ring to free buffers from
5270  **/
5271 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
5272 {
5273 	u16 i = rx_ring->next_to_clean;
5274 	struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
5275 
5276 	if (rx_ring->xsk_umem) {
5277 		ixgbe_xsk_clean_rx_ring(rx_ring);
5278 		goto skip_free;
5279 	}
5280 
5281 	/* Free all the Rx ring sk_buffs */
5282 	while (i != rx_ring->next_to_alloc) {
5283 		if (rx_buffer->skb) {
5284 			struct sk_buff *skb = rx_buffer->skb;
5285 			if (IXGBE_CB(skb)->page_released)
5286 				dma_unmap_page_attrs(rx_ring->dev,
5287 						     IXGBE_CB(skb)->dma,
5288 						     ixgbe_rx_pg_size(rx_ring),
5289 						     DMA_FROM_DEVICE,
5290 						     IXGBE_RX_DMA_ATTR);
5291 			dev_kfree_skb(skb);
5292 		}
5293 
5294 		/* Invalidate cache lines that may have been written to by
5295 		 * device so that we avoid corrupting memory.
5296 		 */
5297 		dma_sync_single_range_for_cpu(rx_ring->dev,
5298 					      rx_buffer->dma,
5299 					      rx_buffer->page_offset,
5300 					      ixgbe_rx_bufsz(rx_ring),
5301 					      DMA_FROM_DEVICE);
5302 
5303 		/* free resources associated with mapping */
5304 		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
5305 				     ixgbe_rx_pg_size(rx_ring),
5306 				     DMA_FROM_DEVICE,
5307 				     IXGBE_RX_DMA_ATTR);
5308 		__page_frag_cache_drain(rx_buffer->page,
5309 					rx_buffer->pagecnt_bias);
5310 
5311 		i++;
5312 		rx_buffer++;
5313 		if (i == rx_ring->count) {
5314 			i = 0;
5315 			rx_buffer = rx_ring->rx_buffer_info;
5316 		}
5317 	}
5318 
5319 skip_free:
5320 	rx_ring->next_to_alloc = 0;
5321 	rx_ring->next_to_clean = 0;
5322 	rx_ring->next_to_use = 0;
5323 }
5324 
5325 static int ixgbe_fwd_ring_up(struct ixgbe_adapter *adapter,
5326 			     struct ixgbe_fwd_adapter *accel)
5327 {
5328 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
5329 	int num_tc = netdev_get_num_tc(adapter->netdev);
5330 	struct net_device *vdev = accel->netdev;
5331 	int i, baseq, err;
5332 
5333 	baseq = accel->pool * adapter->num_rx_queues_per_pool;
5334 	netdev_dbg(vdev, "pool %i:%i queues %i:%i\n",
5335 		   accel->pool, adapter->num_rx_pools,
5336 		   baseq, baseq + adapter->num_rx_queues_per_pool);
5337 
5338 	accel->rx_base_queue = baseq;
5339 	accel->tx_base_queue = baseq;
5340 
5341 	/* record configuration for macvlan interface in vdev */
5342 	for (i = 0; i < num_tc; i++)
5343 		netdev_bind_sb_channel_queue(adapter->netdev, vdev,
5344 					     i, rss_i, baseq + (rss_i * i));
5345 
5346 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5347 		adapter->rx_ring[baseq + i]->netdev = vdev;
5348 
5349 	/* Guarantee all rings are updated before we update the
5350 	 * MAC address filter.
5351 	 */
5352 	wmb();
5353 
5354 	/* ixgbe_add_mac_filter will return an index if it succeeds, so we
5355 	 * need to only treat it as an error value if it is negative.
5356 	 */
5357 	err = ixgbe_add_mac_filter(adapter, vdev->dev_addr,
5358 				   VMDQ_P(accel->pool));
5359 	if (err >= 0)
5360 		return 0;
5361 
5362 	/* if we cannot add the MAC rule then disable the offload */
5363 	macvlan_release_l2fw_offload(vdev);
5364 
5365 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5366 		adapter->rx_ring[baseq + i]->netdev = NULL;
5367 
5368 	netdev_err(vdev, "L2FW offload disabled due to L2 filter error\n");
5369 
5370 	/* unbind the queues and drop the subordinate channel config */
5371 	netdev_unbind_sb_channel(adapter->netdev, vdev);
5372 	netdev_set_sb_channel(vdev, 0);
5373 
5374 	clear_bit(accel->pool, adapter->fwd_bitmask);
5375 	kfree(accel);
5376 
5377 	return err;
5378 }
5379 
5380 static int ixgbe_macvlan_up(struct net_device *vdev, void *data)
5381 {
5382 	struct ixgbe_adapter *adapter = data;
5383 	struct ixgbe_fwd_adapter *accel;
5384 
5385 	if (!netif_is_macvlan(vdev))
5386 		return 0;
5387 
5388 	accel = macvlan_accel_priv(vdev);
5389 	if (!accel)
5390 		return 0;
5391 
5392 	ixgbe_fwd_ring_up(adapter, accel);
5393 
5394 	return 0;
5395 }
5396 
5397 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
5398 {
5399 	netdev_walk_all_upper_dev_rcu(adapter->netdev,
5400 				      ixgbe_macvlan_up, adapter);
5401 }
5402 
5403 static void ixgbe_configure(struct ixgbe_adapter *adapter)
5404 {
5405 	struct ixgbe_hw *hw = &adapter->hw;
5406 
5407 	ixgbe_configure_pb(adapter);
5408 #ifdef CONFIG_IXGBE_DCB
5409 	ixgbe_configure_dcb(adapter);
5410 #endif
5411 	/*
5412 	 * We must restore virtualization before VLANs or else
5413 	 * the VLVF registers will not be populated
5414 	 */
5415 	ixgbe_configure_virtualization(adapter);
5416 
5417 	ixgbe_set_rx_mode(adapter->netdev);
5418 	ixgbe_restore_vlan(adapter);
5419 	ixgbe_ipsec_restore(adapter);
5420 
5421 	switch (hw->mac.type) {
5422 	case ixgbe_mac_82599EB:
5423 	case ixgbe_mac_X540:
5424 		hw->mac.ops.disable_rx_buff(hw);
5425 		break;
5426 	default:
5427 		break;
5428 	}
5429 
5430 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5431 		ixgbe_init_fdir_signature_82599(&adapter->hw,
5432 						adapter->fdir_pballoc);
5433 	} else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
5434 		ixgbe_init_fdir_perfect_82599(&adapter->hw,
5435 					      adapter->fdir_pballoc);
5436 		ixgbe_fdir_filter_restore(adapter);
5437 	}
5438 
5439 	switch (hw->mac.type) {
5440 	case ixgbe_mac_82599EB:
5441 	case ixgbe_mac_X540:
5442 		hw->mac.ops.enable_rx_buff(hw);
5443 		break;
5444 	default:
5445 		break;
5446 	}
5447 
5448 #ifdef CONFIG_IXGBE_DCA
5449 	/* configure DCA */
5450 	if (adapter->flags & IXGBE_FLAG_DCA_CAPABLE)
5451 		ixgbe_setup_dca(adapter);
5452 #endif /* CONFIG_IXGBE_DCA */
5453 
5454 #ifdef IXGBE_FCOE
5455 	/* configure FCoE L2 filters, redirection table, and Rx control */
5456 	ixgbe_configure_fcoe(adapter);
5457 
5458 #endif /* IXGBE_FCOE */
5459 	ixgbe_configure_tx(adapter);
5460 	ixgbe_configure_rx(adapter);
5461 	ixgbe_configure_dfwd(adapter);
5462 }
5463 
5464 /**
5465  * ixgbe_sfp_link_config - set up SFP+ link
5466  * @adapter: pointer to private adapter struct
5467  **/
5468 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
5469 {
5470 	/*
5471 	 * We are assuming the worst case scenario here, and that
5472 	 * is that an SFP was inserted/removed after the reset
5473 	 * but before SFP detection was enabled.  As such the best
5474 	 * solution is to just start searching as soon as we start
5475 	 */
5476 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5477 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5478 
5479 	adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5480 	adapter->sfp_poll_time = 0;
5481 }
5482 
5483 /**
5484  * ixgbe_non_sfp_link_config - set up non-SFP+ link
5485  * @hw: pointer to private hardware struct
5486  *
5487  * Returns 0 on success, negative on failure
5488  **/
5489 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
5490 {
5491 	u32 speed;
5492 	bool autoneg, link_up = false;
5493 	int ret = IXGBE_ERR_LINK_SETUP;
5494 
5495 	if (hw->mac.ops.check_link)
5496 		ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
5497 
5498 	if (ret)
5499 		return ret;
5500 
5501 	speed = hw->phy.autoneg_advertised;
5502 	if ((!speed) && (hw->mac.ops.get_link_capabilities))
5503 		ret = hw->mac.ops.get_link_capabilities(hw, &speed,
5504 							&autoneg);
5505 	if (ret)
5506 		return ret;
5507 
5508 	if (hw->mac.ops.setup_link)
5509 		ret = hw->mac.ops.setup_link(hw, speed, link_up);
5510 
5511 	return ret;
5512 }
5513 
5514 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
5515 {
5516 	struct ixgbe_hw *hw = &adapter->hw;
5517 	u32 gpie = 0;
5518 
5519 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5520 		gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
5521 		       IXGBE_GPIE_OCD;
5522 		gpie |= IXGBE_GPIE_EIAME;
5523 		/*
5524 		 * use EIAM to auto-mask when MSI-X interrupt is asserted
5525 		 * this saves a register write for every interrupt
5526 		 */
5527 		switch (hw->mac.type) {
5528 		case ixgbe_mac_82598EB:
5529 			IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5530 			break;
5531 		case ixgbe_mac_82599EB:
5532 		case ixgbe_mac_X540:
5533 		case ixgbe_mac_X550:
5534 		case ixgbe_mac_X550EM_x:
5535 		case ixgbe_mac_x550em_a:
5536 		default:
5537 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
5538 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
5539 			break;
5540 		}
5541 	} else {
5542 		/* legacy interrupts, use EIAM to auto-mask when reading EICR,
5543 		 * specifically only auto mask tx and rx interrupts */
5544 		IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5545 	}
5546 
5547 	/* XXX: to interrupt immediately for EICS writes, enable this */
5548 	/* gpie |= IXGBE_GPIE_EIMEN; */
5549 
5550 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
5551 		gpie &= ~IXGBE_GPIE_VTMODE_MASK;
5552 
5553 		switch (adapter->ring_feature[RING_F_VMDQ].mask) {
5554 		case IXGBE_82599_VMDQ_8Q_MASK:
5555 			gpie |= IXGBE_GPIE_VTMODE_16;
5556 			break;
5557 		case IXGBE_82599_VMDQ_4Q_MASK:
5558 			gpie |= IXGBE_GPIE_VTMODE_32;
5559 			break;
5560 		default:
5561 			gpie |= IXGBE_GPIE_VTMODE_64;
5562 			break;
5563 		}
5564 	}
5565 
5566 	/* Enable Thermal over heat sensor interrupt */
5567 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
5568 		switch (adapter->hw.mac.type) {
5569 		case ixgbe_mac_82599EB:
5570 			gpie |= IXGBE_SDP0_GPIEN_8259X;
5571 			break;
5572 		default:
5573 			break;
5574 		}
5575 	}
5576 
5577 	/* Enable fan failure interrupt */
5578 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
5579 		gpie |= IXGBE_SDP1_GPIEN(hw);
5580 
5581 	switch (hw->mac.type) {
5582 	case ixgbe_mac_82599EB:
5583 		gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X;
5584 		break;
5585 	case ixgbe_mac_X550EM_x:
5586 	case ixgbe_mac_x550em_a:
5587 		gpie |= IXGBE_SDP0_GPIEN_X540;
5588 		break;
5589 	default:
5590 		break;
5591 	}
5592 
5593 	IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
5594 }
5595 
5596 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
5597 {
5598 	struct ixgbe_hw *hw = &adapter->hw;
5599 	int err;
5600 	u32 ctrl_ext;
5601 
5602 	ixgbe_get_hw_control(adapter);
5603 	ixgbe_setup_gpie(adapter);
5604 
5605 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5606 		ixgbe_configure_msix(adapter);
5607 	else
5608 		ixgbe_configure_msi_and_legacy(adapter);
5609 
5610 	/* enable the optics for 82599 SFP+ fiber */
5611 	if (hw->mac.ops.enable_tx_laser)
5612 		hw->mac.ops.enable_tx_laser(hw);
5613 
5614 	if (hw->phy.ops.set_phy_power)
5615 		hw->phy.ops.set_phy_power(hw, true);
5616 
5617 	smp_mb__before_atomic();
5618 	clear_bit(__IXGBE_DOWN, &adapter->state);
5619 	ixgbe_napi_enable_all(adapter);
5620 
5621 	if (ixgbe_is_sfp(hw)) {
5622 		ixgbe_sfp_link_config(adapter);
5623 	} else {
5624 		err = ixgbe_non_sfp_link_config(hw);
5625 		if (err)
5626 			e_err(probe, "link_config FAILED %d\n", err);
5627 	}
5628 
5629 	/* clear any pending interrupts, may auto mask */
5630 	IXGBE_READ_REG(hw, IXGBE_EICR);
5631 	ixgbe_irq_enable(adapter, true, true);
5632 
5633 	/*
5634 	 * If this adapter has a fan, check to see if we had a failure
5635 	 * before we enabled the interrupt.
5636 	 */
5637 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5638 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5639 		if (esdp & IXGBE_ESDP_SDP1)
5640 			e_crit(drv, "Fan has stopped, replace the adapter\n");
5641 	}
5642 
5643 	/* bring the link up in the watchdog, this could race with our first
5644 	 * link up interrupt but shouldn't be a problem */
5645 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5646 	adapter->link_check_timeout = jiffies;
5647 	mod_timer(&adapter->service_timer, jiffies);
5648 
5649 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
5650 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5651 	ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
5652 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5653 }
5654 
5655 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
5656 {
5657 	WARN_ON(in_interrupt());
5658 	/* put off any impending NetWatchDogTimeout */
5659 	netif_trans_update(adapter->netdev);
5660 
5661 	while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
5662 		usleep_range(1000, 2000);
5663 	if (adapter->hw.phy.type == ixgbe_phy_fw)
5664 		ixgbe_watchdog_link_is_down(adapter);
5665 	ixgbe_down(adapter);
5666 	/*
5667 	 * If SR-IOV enabled then wait a bit before bringing the adapter
5668 	 * back up to give the VFs time to respond to the reset.  The
5669 	 * two second wait is based upon the watchdog timer cycle in
5670 	 * the VF driver.
5671 	 */
5672 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5673 		msleep(2000);
5674 	ixgbe_up(adapter);
5675 	clear_bit(__IXGBE_RESETTING, &adapter->state);
5676 }
5677 
5678 void ixgbe_up(struct ixgbe_adapter *adapter)
5679 {
5680 	/* hardware has been reset, we need to reload some things */
5681 	ixgbe_configure(adapter);
5682 
5683 	ixgbe_up_complete(adapter);
5684 }
5685 
5686 static unsigned long ixgbe_get_completion_timeout(struct ixgbe_adapter *adapter)
5687 {
5688 	u16 devctl2;
5689 
5690 	pcie_capability_read_word(adapter->pdev, PCI_EXP_DEVCTL2, &devctl2);
5691 
5692 	switch (devctl2 & IXGBE_PCIDEVCTRL2_TIMEO_MASK) {
5693 	case IXGBE_PCIDEVCTRL2_17_34s:
5694 	case IXGBE_PCIDEVCTRL2_4_8s:
5695 		/* For now we cap the upper limit on delay to 2 seconds
5696 		 * as we end up going up to 34 seconds of delay in worst
5697 		 * case timeout value.
5698 		 */
5699 	case IXGBE_PCIDEVCTRL2_1_2s:
5700 		return 2000000ul;	/* 2.0 s */
5701 	case IXGBE_PCIDEVCTRL2_260_520ms:
5702 		return 520000ul;	/* 520 ms */
5703 	case IXGBE_PCIDEVCTRL2_65_130ms:
5704 		return 130000ul;	/* 130 ms */
5705 	case IXGBE_PCIDEVCTRL2_16_32ms:
5706 		return 32000ul;		/* 32 ms */
5707 	case IXGBE_PCIDEVCTRL2_1_2ms:
5708 		return 2000ul;		/* 2 ms */
5709 	case IXGBE_PCIDEVCTRL2_50_100us:
5710 		return 100ul;		/* 100 us */
5711 	case IXGBE_PCIDEVCTRL2_16_32ms_def:
5712 		return 32000ul;		/* 32 ms */
5713 	default:
5714 		break;
5715 	}
5716 
5717 	/* We shouldn't need to hit this path, but just in case default as
5718 	 * though completion timeout is not supported and support 32ms.
5719 	 */
5720 	return 32000ul;
5721 }
5722 
5723 void ixgbe_disable_rx(struct ixgbe_adapter *adapter)
5724 {
5725 	unsigned long wait_delay, delay_interval;
5726 	struct ixgbe_hw *hw = &adapter->hw;
5727 	int i, wait_loop;
5728 	u32 rxdctl;
5729 
5730 	/* disable receives */
5731 	hw->mac.ops.disable_rx(hw);
5732 
5733 	if (ixgbe_removed(hw->hw_addr))
5734 		return;
5735 
5736 	/* disable all enabled Rx queues */
5737 	for (i = 0; i < adapter->num_rx_queues; i++) {
5738 		struct ixgbe_ring *ring = adapter->rx_ring[i];
5739 		u8 reg_idx = ring->reg_idx;
5740 
5741 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5742 		rxdctl &= ~IXGBE_RXDCTL_ENABLE;
5743 		rxdctl |= IXGBE_RXDCTL_SWFLSH;
5744 
5745 		/* write value back with RXDCTL.ENABLE bit cleared */
5746 		IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
5747 	}
5748 
5749 	/* RXDCTL.EN may not change on 82598 if link is down, so skip it */
5750 	if (hw->mac.type == ixgbe_mac_82598EB &&
5751 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5752 		return;
5753 
5754 	/* Determine our minimum delay interval. We will increase this value
5755 	 * with each subsequent test. This way if the device returns quickly
5756 	 * we should spend as little time as possible waiting, however as
5757 	 * the time increases we will wait for larger periods of time.
5758 	 *
5759 	 * The trick here is that we increase the interval using the
5760 	 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5761 	 * of that wait is that it totals up to 100x whatever interval we
5762 	 * choose. Since our minimum wait is 100us we can just divide the
5763 	 * total timeout by 100 to get our minimum delay interval.
5764 	 */
5765 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5766 
5767 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
5768 	wait_delay = delay_interval;
5769 
5770 	while (wait_loop--) {
5771 		usleep_range(wait_delay, wait_delay + 10);
5772 		wait_delay += delay_interval * 2;
5773 		rxdctl = 0;
5774 
5775 		/* OR together the reading of all the active RXDCTL registers,
5776 		 * and then test the result. We need the disable to complete
5777 		 * before we start freeing the memory and invalidating the
5778 		 * DMA mappings.
5779 		 */
5780 		for (i = 0; i < adapter->num_rx_queues; i++) {
5781 			struct ixgbe_ring *ring = adapter->rx_ring[i];
5782 			u8 reg_idx = ring->reg_idx;
5783 
5784 			rxdctl |= IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5785 		}
5786 
5787 		if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
5788 			return;
5789 	}
5790 
5791 	e_err(drv,
5792 	      "RXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5793 }
5794 
5795 void ixgbe_disable_tx(struct ixgbe_adapter *adapter)
5796 {
5797 	unsigned long wait_delay, delay_interval;
5798 	struct ixgbe_hw *hw = &adapter->hw;
5799 	int i, wait_loop;
5800 	u32 txdctl;
5801 
5802 	if (ixgbe_removed(hw->hw_addr))
5803 		return;
5804 
5805 	/* disable all enabled Tx queues */
5806 	for (i = 0; i < adapter->num_tx_queues; i++) {
5807 		struct ixgbe_ring *ring = adapter->tx_ring[i];
5808 		u8 reg_idx = ring->reg_idx;
5809 
5810 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5811 	}
5812 
5813 	/* disable all enabled XDP Tx queues */
5814 	for (i = 0; i < adapter->num_xdp_queues; i++) {
5815 		struct ixgbe_ring *ring = adapter->xdp_ring[i];
5816 		u8 reg_idx = ring->reg_idx;
5817 
5818 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5819 	}
5820 
5821 	/* If the link is not up there shouldn't be much in the way of
5822 	 * pending transactions. Those that are left will be flushed out
5823 	 * when the reset logic goes through the flush sequence to clean out
5824 	 * the pending Tx transactions.
5825 	 */
5826 	if (!(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5827 		goto dma_engine_disable;
5828 
5829 	/* Determine our minimum delay interval. We will increase this value
5830 	 * with each subsequent test. This way if the device returns quickly
5831 	 * we should spend as little time as possible waiting, however as
5832 	 * the time increases we will wait for larger periods of time.
5833 	 *
5834 	 * The trick here is that we increase the interval using the
5835 	 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5836 	 * of that wait is that it totals up to 100x whatever interval we
5837 	 * choose. Since our minimum wait is 100us we can just divide the
5838 	 * total timeout by 100 to get our minimum delay interval.
5839 	 */
5840 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5841 
5842 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
5843 	wait_delay = delay_interval;
5844 
5845 	while (wait_loop--) {
5846 		usleep_range(wait_delay, wait_delay + 10);
5847 		wait_delay += delay_interval * 2;
5848 		txdctl = 0;
5849 
5850 		/* OR together the reading of all the active TXDCTL registers,
5851 		 * and then test the result. We need the disable to complete
5852 		 * before we start freeing the memory and invalidating the
5853 		 * DMA mappings.
5854 		 */
5855 		for (i = 0; i < adapter->num_tx_queues; i++) {
5856 			struct ixgbe_ring *ring = adapter->tx_ring[i];
5857 			u8 reg_idx = ring->reg_idx;
5858 
5859 			txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5860 		}
5861 		for (i = 0; i < adapter->num_xdp_queues; i++) {
5862 			struct ixgbe_ring *ring = adapter->xdp_ring[i];
5863 			u8 reg_idx = ring->reg_idx;
5864 
5865 			txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5866 		}
5867 
5868 		if (!(txdctl & IXGBE_TXDCTL_ENABLE))
5869 			goto dma_engine_disable;
5870 	}
5871 
5872 	e_err(drv,
5873 	      "TXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5874 
5875 dma_engine_disable:
5876 	/* Disable the Tx DMA engine on 82599 and later MAC */
5877 	switch (hw->mac.type) {
5878 	case ixgbe_mac_82599EB:
5879 	case ixgbe_mac_X540:
5880 	case ixgbe_mac_X550:
5881 	case ixgbe_mac_X550EM_x:
5882 	case ixgbe_mac_x550em_a:
5883 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5884 				(IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5885 				 ~IXGBE_DMATXCTL_TE));
5886 		/* fall through */
5887 	default:
5888 		break;
5889 	}
5890 }
5891 
5892 void ixgbe_reset(struct ixgbe_adapter *adapter)
5893 {
5894 	struct ixgbe_hw *hw = &adapter->hw;
5895 	struct net_device *netdev = adapter->netdev;
5896 	int err;
5897 
5898 	if (ixgbe_removed(hw->hw_addr))
5899 		return;
5900 	/* lock SFP init bit to prevent race conditions with the watchdog */
5901 	while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5902 		usleep_range(1000, 2000);
5903 
5904 	/* clear all SFP and link config related flags while holding SFP_INIT */
5905 	adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
5906 			     IXGBE_FLAG2_SFP_NEEDS_RESET);
5907 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5908 
5909 	err = hw->mac.ops.init_hw(hw);
5910 	switch (err) {
5911 	case 0:
5912 	case IXGBE_ERR_SFP_NOT_PRESENT:
5913 	case IXGBE_ERR_SFP_NOT_SUPPORTED:
5914 		break;
5915 	case IXGBE_ERR_MASTER_REQUESTS_PENDING:
5916 		e_dev_err("master disable timed out\n");
5917 		break;
5918 	case IXGBE_ERR_EEPROM_VERSION:
5919 		/* We are running on a pre-production device, log a warning */
5920 		e_dev_warn("This device is a pre-production adapter/LOM. "
5921 			   "Please be aware there may be issues associated with "
5922 			   "your hardware.  If you are experiencing problems "
5923 			   "please contact your Intel or hardware "
5924 			   "representative who provided you with this "
5925 			   "hardware.\n");
5926 		break;
5927 	default:
5928 		e_dev_err("Hardware Error: %d\n", err);
5929 	}
5930 
5931 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5932 
5933 	/* flush entries out of MAC table */
5934 	ixgbe_flush_sw_mac_table(adapter);
5935 	__dev_uc_unsync(netdev, NULL);
5936 
5937 	/* do not flush user set addresses */
5938 	ixgbe_mac_set_default_filter(adapter);
5939 
5940 	/* update SAN MAC vmdq pool selection */
5941 	if (hw->mac.san_mac_rar_index)
5942 		hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
5943 
5944 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5945 		ixgbe_ptp_reset(adapter);
5946 
5947 	if (hw->phy.ops.set_phy_power) {
5948 		if (!netif_running(adapter->netdev) && !adapter->wol)
5949 			hw->phy.ops.set_phy_power(hw, false);
5950 		else
5951 			hw->phy.ops.set_phy_power(hw, true);
5952 	}
5953 }
5954 
5955 /**
5956  * ixgbe_clean_tx_ring - Free Tx Buffers
5957  * @tx_ring: ring to be cleaned
5958  **/
5959 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
5960 {
5961 	u16 i = tx_ring->next_to_clean;
5962 	struct ixgbe_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
5963 
5964 	if (tx_ring->xsk_umem) {
5965 		ixgbe_xsk_clean_tx_ring(tx_ring);
5966 		goto out;
5967 	}
5968 
5969 	while (i != tx_ring->next_to_use) {
5970 		union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
5971 
5972 		/* Free all the Tx ring sk_buffs */
5973 		if (ring_is_xdp(tx_ring))
5974 			xdp_return_frame(tx_buffer->xdpf);
5975 		else
5976 			dev_kfree_skb_any(tx_buffer->skb);
5977 
5978 		/* unmap skb header data */
5979 		dma_unmap_single(tx_ring->dev,
5980 				 dma_unmap_addr(tx_buffer, dma),
5981 				 dma_unmap_len(tx_buffer, len),
5982 				 DMA_TO_DEVICE);
5983 
5984 		/* check for eop_desc to determine the end of the packet */
5985 		eop_desc = tx_buffer->next_to_watch;
5986 		tx_desc = IXGBE_TX_DESC(tx_ring, i);
5987 
5988 		/* unmap remaining buffers */
5989 		while (tx_desc != eop_desc) {
5990 			tx_buffer++;
5991 			tx_desc++;
5992 			i++;
5993 			if (unlikely(i == tx_ring->count)) {
5994 				i = 0;
5995 				tx_buffer = tx_ring->tx_buffer_info;
5996 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
5997 			}
5998 
5999 			/* unmap any remaining paged data */
6000 			if (dma_unmap_len(tx_buffer, len))
6001 				dma_unmap_page(tx_ring->dev,
6002 					       dma_unmap_addr(tx_buffer, dma),
6003 					       dma_unmap_len(tx_buffer, len),
6004 					       DMA_TO_DEVICE);
6005 		}
6006 
6007 		/* move us one more past the eop_desc for start of next pkt */
6008 		tx_buffer++;
6009 		i++;
6010 		if (unlikely(i == tx_ring->count)) {
6011 			i = 0;
6012 			tx_buffer = tx_ring->tx_buffer_info;
6013 		}
6014 	}
6015 
6016 	/* reset BQL for queue */
6017 	if (!ring_is_xdp(tx_ring))
6018 		netdev_tx_reset_queue(txring_txq(tx_ring));
6019 
6020 out:
6021 	/* reset next_to_use and next_to_clean */
6022 	tx_ring->next_to_use = 0;
6023 	tx_ring->next_to_clean = 0;
6024 }
6025 
6026 /**
6027  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
6028  * @adapter: board private structure
6029  **/
6030 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
6031 {
6032 	int i;
6033 
6034 	for (i = 0; i < adapter->num_rx_queues; i++)
6035 		ixgbe_clean_rx_ring(adapter->rx_ring[i]);
6036 }
6037 
6038 /**
6039  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
6040  * @adapter: board private structure
6041  **/
6042 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
6043 {
6044 	int i;
6045 
6046 	for (i = 0; i < adapter->num_tx_queues; i++)
6047 		ixgbe_clean_tx_ring(adapter->tx_ring[i]);
6048 	for (i = 0; i < adapter->num_xdp_queues; i++)
6049 		ixgbe_clean_tx_ring(adapter->xdp_ring[i]);
6050 }
6051 
6052 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
6053 {
6054 	struct hlist_node *node2;
6055 	struct ixgbe_fdir_filter *filter;
6056 
6057 	spin_lock(&adapter->fdir_perfect_lock);
6058 
6059 	hlist_for_each_entry_safe(filter, node2,
6060 				  &adapter->fdir_filter_list, fdir_node) {
6061 		hlist_del(&filter->fdir_node);
6062 		kfree(filter);
6063 	}
6064 	adapter->fdir_filter_count = 0;
6065 
6066 	spin_unlock(&adapter->fdir_perfect_lock);
6067 }
6068 
6069 void ixgbe_down(struct ixgbe_adapter *adapter)
6070 {
6071 	struct net_device *netdev = adapter->netdev;
6072 	struct ixgbe_hw *hw = &adapter->hw;
6073 	int i;
6074 
6075 	/* signal that we are down to the interrupt handler */
6076 	if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
6077 		return; /* do nothing if already down */
6078 
6079 	/* Shut off incoming Tx traffic */
6080 	netif_tx_stop_all_queues(netdev);
6081 
6082 	/* call carrier off first to avoid false dev_watchdog timeouts */
6083 	netif_carrier_off(netdev);
6084 	netif_tx_disable(netdev);
6085 
6086 	/* Disable Rx */
6087 	ixgbe_disable_rx(adapter);
6088 
6089 	/* synchronize_rcu() needed for pending XDP buffers to drain */
6090 	if (adapter->xdp_ring[0])
6091 		synchronize_rcu();
6092 
6093 	ixgbe_irq_disable(adapter);
6094 
6095 	ixgbe_napi_disable_all(adapter);
6096 
6097 	clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
6098 	adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6099 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6100 
6101 	del_timer_sync(&adapter->service_timer);
6102 
6103 	if (adapter->num_vfs) {
6104 		/* Clear EITR Select mapping */
6105 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
6106 
6107 		/* Mark all the VFs as inactive */
6108 		for (i = 0 ; i < adapter->num_vfs; i++)
6109 			adapter->vfinfo[i].clear_to_send = false;
6110 
6111 		/* ping all the active vfs to let them know we are going down */
6112 		ixgbe_ping_all_vfs(adapter);
6113 
6114 		/* Disable all VFTE/VFRE TX/RX */
6115 		ixgbe_disable_tx_rx(adapter);
6116 	}
6117 
6118 	/* disable transmits in the hardware now that interrupts are off */
6119 	ixgbe_disable_tx(adapter);
6120 
6121 	if (!pci_channel_offline(adapter->pdev))
6122 		ixgbe_reset(adapter);
6123 
6124 	/* power down the optics for 82599 SFP+ fiber */
6125 	if (hw->mac.ops.disable_tx_laser)
6126 		hw->mac.ops.disable_tx_laser(hw);
6127 
6128 	ixgbe_clean_all_tx_rings(adapter);
6129 	ixgbe_clean_all_rx_rings(adapter);
6130 }
6131 
6132 /**
6133  * ixgbe_eee_capable - helper function to determine EEE support on X550
6134  * @adapter: board private structure
6135  */
6136 static void ixgbe_set_eee_capable(struct ixgbe_adapter *adapter)
6137 {
6138 	struct ixgbe_hw *hw = &adapter->hw;
6139 
6140 	switch (hw->device_id) {
6141 	case IXGBE_DEV_ID_X550EM_A_1G_T:
6142 	case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6143 		if (!hw->phy.eee_speeds_supported)
6144 			break;
6145 		adapter->flags2 |= IXGBE_FLAG2_EEE_CAPABLE;
6146 		if (!hw->phy.eee_speeds_advertised)
6147 			break;
6148 		adapter->flags2 |= IXGBE_FLAG2_EEE_ENABLED;
6149 		break;
6150 	default:
6151 		adapter->flags2 &= ~IXGBE_FLAG2_EEE_CAPABLE;
6152 		adapter->flags2 &= ~IXGBE_FLAG2_EEE_ENABLED;
6153 		break;
6154 	}
6155 }
6156 
6157 /**
6158  * ixgbe_tx_timeout - Respond to a Tx Hang
6159  * @netdev: network interface device structure
6160  **/
6161 static void ixgbe_tx_timeout(struct net_device *netdev)
6162 {
6163 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6164 
6165 	/* Do the reset outside of interrupt context */
6166 	ixgbe_tx_timeout_reset(adapter);
6167 }
6168 
6169 #ifdef CONFIG_IXGBE_DCB
6170 static void ixgbe_init_dcb(struct ixgbe_adapter *adapter)
6171 {
6172 	struct ixgbe_hw *hw = &adapter->hw;
6173 	struct tc_configuration *tc;
6174 	int j;
6175 
6176 	switch (hw->mac.type) {
6177 	case ixgbe_mac_82598EB:
6178 	case ixgbe_mac_82599EB:
6179 		adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
6180 		adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
6181 		break;
6182 	case ixgbe_mac_X540:
6183 	case ixgbe_mac_X550:
6184 		adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
6185 		adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
6186 		break;
6187 	case ixgbe_mac_X550EM_x:
6188 	case ixgbe_mac_x550em_a:
6189 	default:
6190 		adapter->dcb_cfg.num_tcs.pg_tcs = DEF_TRAFFIC_CLASS;
6191 		adapter->dcb_cfg.num_tcs.pfc_tcs = DEF_TRAFFIC_CLASS;
6192 		break;
6193 	}
6194 
6195 	/* Configure DCB traffic classes */
6196 	for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
6197 		tc = &adapter->dcb_cfg.tc_config[j];
6198 		tc->path[DCB_TX_CONFIG].bwg_id = 0;
6199 		tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
6200 		tc->path[DCB_RX_CONFIG].bwg_id = 0;
6201 		tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
6202 		tc->dcb_pfc = pfc_disabled;
6203 	}
6204 
6205 	/* Initialize default user to priority mapping, UPx->TC0 */
6206 	tc = &adapter->dcb_cfg.tc_config[0];
6207 	tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
6208 	tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
6209 
6210 	adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
6211 	adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
6212 	adapter->dcb_cfg.pfc_mode_enable = false;
6213 	adapter->dcb_set_bitmap = 0x00;
6214 	if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
6215 		adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
6216 	memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
6217 	       sizeof(adapter->temp_dcb_cfg));
6218 }
6219 #endif
6220 
6221 /**
6222  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
6223  * @adapter: board private structure to initialize
6224  * @ii: pointer to ixgbe_info for device
6225  *
6226  * ixgbe_sw_init initializes the Adapter private data structure.
6227  * Fields are initialized based on PCI device information and
6228  * OS network device settings (MTU size).
6229  **/
6230 static int ixgbe_sw_init(struct ixgbe_adapter *adapter,
6231 			 const struct ixgbe_info *ii)
6232 {
6233 	struct ixgbe_hw *hw = &adapter->hw;
6234 	struct pci_dev *pdev = adapter->pdev;
6235 	unsigned int rss, fdir;
6236 	u32 fwsm;
6237 	int i;
6238 
6239 	/* PCI config space info */
6240 
6241 	hw->vendor_id = pdev->vendor;
6242 	hw->device_id = pdev->device;
6243 	hw->revision_id = pdev->revision;
6244 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
6245 	hw->subsystem_device_id = pdev->subsystem_device;
6246 
6247 	/* get_invariants needs the device IDs */
6248 	ii->get_invariants(hw);
6249 
6250 	/* Set common capability flags and settings */
6251 	rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
6252 	adapter->ring_feature[RING_F_RSS].limit = rss;
6253 	adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
6254 	adapter->max_q_vectors = MAX_Q_VECTORS_82599;
6255 	adapter->atr_sample_rate = 20;
6256 	fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
6257 	adapter->ring_feature[RING_F_FDIR].limit = fdir;
6258 	adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
6259 	adapter->ring_feature[RING_F_VMDQ].limit = 1;
6260 #ifdef CONFIG_IXGBE_DCA
6261 	adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
6262 #endif
6263 #ifdef CONFIG_IXGBE_DCB
6264 	adapter->flags |= IXGBE_FLAG_DCB_CAPABLE;
6265 	adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6266 #endif
6267 #ifdef IXGBE_FCOE
6268 	adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
6269 	adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6270 #ifdef CONFIG_IXGBE_DCB
6271 	/* Default traffic class to use for FCoE */
6272 	adapter->fcoe.up = IXGBE_FCOE_DEFTC;
6273 #endif /* CONFIG_IXGBE_DCB */
6274 #endif /* IXGBE_FCOE */
6275 
6276 	/* initialize static ixgbe jump table entries */
6277 	adapter->jump_tables[0] = kzalloc(sizeof(*adapter->jump_tables[0]),
6278 					  GFP_KERNEL);
6279 	if (!adapter->jump_tables[0])
6280 		return -ENOMEM;
6281 	adapter->jump_tables[0]->mat = ixgbe_ipv4_fields;
6282 
6283 	for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++)
6284 		adapter->jump_tables[i] = NULL;
6285 
6286 	adapter->mac_table = kcalloc(hw->mac.num_rar_entries,
6287 				     sizeof(struct ixgbe_mac_addr),
6288 				     GFP_KERNEL);
6289 	if (!adapter->mac_table)
6290 		return -ENOMEM;
6291 
6292 	if (ixgbe_init_rss_key(adapter))
6293 		return -ENOMEM;
6294 
6295 	adapter->af_xdp_zc_qps = bitmap_zalloc(MAX_XDP_QUEUES, GFP_KERNEL);
6296 	if (!adapter->af_xdp_zc_qps)
6297 		return -ENOMEM;
6298 
6299 	/* Set MAC specific capability flags and exceptions */
6300 	switch (hw->mac.type) {
6301 	case ixgbe_mac_82598EB:
6302 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
6303 
6304 		if (hw->device_id == IXGBE_DEV_ID_82598AT)
6305 			adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
6306 
6307 		adapter->max_q_vectors = MAX_Q_VECTORS_82598;
6308 		adapter->ring_feature[RING_F_FDIR].limit = 0;
6309 		adapter->atr_sample_rate = 0;
6310 		adapter->fdir_pballoc = 0;
6311 #ifdef IXGBE_FCOE
6312 		adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6313 		adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6314 #ifdef CONFIG_IXGBE_DCB
6315 		adapter->fcoe.up = 0;
6316 #endif /* IXGBE_DCB */
6317 #endif /* IXGBE_FCOE */
6318 		break;
6319 	case ixgbe_mac_82599EB:
6320 		if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
6321 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6322 		break;
6323 	case ixgbe_mac_X540:
6324 		fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
6325 		if (fwsm & IXGBE_FWSM_TS_ENABLED)
6326 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6327 		break;
6328 	case ixgbe_mac_x550em_a:
6329 		adapter->flags |= IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE;
6330 		switch (hw->device_id) {
6331 		case IXGBE_DEV_ID_X550EM_A_1G_T:
6332 		case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6333 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6334 			break;
6335 		default:
6336 			break;
6337 		}
6338 	/* fall through */
6339 	case ixgbe_mac_X550EM_x:
6340 #ifdef CONFIG_IXGBE_DCB
6341 		adapter->flags &= ~IXGBE_FLAG_DCB_CAPABLE;
6342 #endif
6343 #ifdef IXGBE_FCOE
6344 		adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6345 #ifdef CONFIG_IXGBE_DCB
6346 		adapter->fcoe.up = 0;
6347 #endif /* IXGBE_DCB */
6348 #endif /* IXGBE_FCOE */
6349 	/* Fall Through */
6350 	case ixgbe_mac_X550:
6351 		if (hw->mac.type == ixgbe_mac_X550)
6352 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6353 #ifdef CONFIG_IXGBE_DCA
6354 		adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
6355 #endif
6356 		adapter->flags |= IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE;
6357 		break;
6358 	default:
6359 		break;
6360 	}
6361 
6362 #ifdef IXGBE_FCOE
6363 	/* FCoE support exists, always init the FCoE lock */
6364 	spin_lock_init(&adapter->fcoe.lock);
6365 
6366 #endif
6367 	/* n-tuple support exists, always init our spinlock */
6368 	spin_lock_init(&adapter->fdir_perfect_lock);
6369 
6370 #ifdef CONFIG_IXGBE_DCB
6371 	ixgbe_init_dcb(adapter);
6372 #endif
6373 	ixgbe_init_ipsec_offload(adapter);
6374 
6375 	/* default flow control settings */
6376 	hw->fc.requested_mode = ixgbe_fc_full;
6377 	hw->fc.current_mode = ixgbe_fc_full;	/* init for ethtool output */
6378 	ixgbe_pbthresh_setup(adapter);
6379 	hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
6380 	hw->fc.send_xon = true;
6381 	hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
6382 
6383 #ifdef CONFIG_PCI_IOV
6384 	if (max_vfs > 0)
6385 		e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
6386 
6387 	/* assign number of SR-IOV VFs */
6388 	if (hw->mac.type != ixgbe_mac_82598EB) {
6389 		if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
6390 			max_vfs = 0;
6391 			e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
6392 		}
6393 	}
6394 #endif /* CONFIG_PCI_IOV */
6395 
6396 	/* enable itr by default in dynamic mode */
6397 	adapter->rx_itr_setting = 1;
6398 	adapter->tx_itr_setting = 1;
6399 
6400 	/* set default ring sizes */
6401 	adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
6402 	adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
6403 
6404 	/* set default work limits */
6405 	adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
6406 
6407 	/* initialize eeprom parameters */
6408 	if (ixgbe_init_eeprom_params_generic(hw)) {
6409 		e_dev_err("EEPROM initialization failed\n");
6410 		return -EIO;
6411 	}
6412 
6413 	/* PF holds first pool slot */
6414 	set_bit(0, adapter->fwd_bitmask);
6415 	set_bit(__IXGBE_DOWN, &adapter->state);
6416 
6417 	return 0;
6418 }
6419 
6420 /**
6421  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
6422  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
6423  *
6424  * Return 0 on success, negative on failure
6425  **/
6426 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
6427 {
6428 	struct device *dev = tx_ring->dev;
6429 	int orig_node = dev_to_node(dev);
6430 	int ring_node = NUMA_NO_NODE;
6431 	int size;
6432 
6433 	size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
6434 
6435 	if (tx_ring->q_vector)
6436 		ring_node = tx_ring->q_vector->numa_node;
6437 
6438 	tx_ring->tx_buffer_info = vmalloc_node(size, ring_node);
6439 	if (!tx_ring->tx_buffer_info)
6440 		tx_ring->tx_buffer_info = vmalloc(size);
6441 	if (!tx_ring->tx_buffer_info)
6442 		goto err;
6443 
6444 	/* round up to nearest 4K */
6445 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
6446 	tx_ring->size = ALIGN(tx_ring->size, 4096);
6447 
6448 	set_dev_node(dev, ring_node);
6449 	tx_ring->desc = dma_alloc_coherent(dev,
6450 					   tx_ring->size,
6451 					   &tx_ring->dma,
6452 					   GFP_KERNEL);
6453 	set_dev_node(dev, orig_node);
6454 	if (!tx_ring->desc)
6455 		tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
6456 						   &tx_ring->dma, GFP_KERNEL);
6457 	if (!tx_ring->desc)
6458 		goto err;
6459 
6460 	tx_ring->next_to_use = 0;
6461 	tx_ring->next_to_clean = 0;
6462 	return 0;
6463 
6464 err:
6465 	vfree(tx_ring->tx_buffer_info);
6466 	tx_ring->tx_buffer_info = NULL;
6467 	dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
6468 	return -ENOMEM;
6469 }
6470 
6471 /**
6472  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
6473  * @adapter: board private structure
6474  *
6475  * If this function returns with an error, then it's possible one or
6476  * more of the rings is populated (while the rest are not).  It is the
6477  * callers duty to clean those orphaned rings.
6478  *
6479  * Return 0 on success, negative on failure
6480  **/
6481 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
6482 {
6483 	int i, j = 0, err = 0;
6484 
6485 	for (i = 0; i < adapter->num_tx_queues; i++) {
6486 		err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
6487 		if (!err)
6488 			continue;
6489 
6490 		e_err(probe, "Allocation for Tx Queue %u failed\n", i);
6491 		goto err_setup_tx;
6492 	}
6493 	for (j = 0; j < adapter->num_xdp_queues; j++) {
6494 		err = ixgbe_setup_tx_resources(adapter->xdp_ring[j]);
6495 		if (!err)
6496 			continue;
6497 
6498 		e_err(probe, "Allocation for Tx Queue %u failed\n", j);
6499 		goto err_setup_tx;
6500 	}
6501 
6502 	return 0;
6503 err_setup_tx:
6504 	/* rewind the index freeing the rings as we go */
6505 	while (j--)
6506 		ixgbe_free_tx_resources(adapter->xdp_ring[j]);
6507 	while (i--)
6508 		ixgbe_free_tx_resources(adapter->tx_ring[i]);
6509 	return err;
6510 }
6511 
6512 /**
6513  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
6514  * @adapter: pointer to ixgbe_adapter
6515  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
6516  *
6517  * Returns 0 on success, negative on failure
6518  **/
6519 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
6520 			     struct ixgbe_ring *rx_ring)
6521 {
6522 	struct device *dev = rx_ring->dev;
6523 	int orig_node = dev_to_node(dev);
6524 	int ring_node = NUMA_NO_NODE;
6525 	int size;
6526 
6527 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
6528 
6529 	if (rx_ring->q_vector)
6530 		ring_node = rx_ring->q_vector->numa_node;
6531 
6532 	rx_ring->rx_buffer_info = vmalloc_node(size, ring_node);
6533 	if (!rx_ring->rx_buffer_info)
6534 		rx_ring->rx_buffer_info = vmalloc(size);
6535 	if (!rx_ring->rx_buffer_info)
6536 		goto err;
6537 
6538 	/* Round up to nearest 4K */
6539 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
6540 	rx_ring->size = ALIGN(rx_ring->size, 4096);
6541 
6542 	set_dev_node(dev, ring_node);
6543 	rx_ring->desc = dma_alloc_coherent(dev,
6544 					   rx_ring->size,
6545 					   &rx_ring->dma,
6546 					   GFP_KERNEL);
6547 	set_dev_node(dev, orig_node);
6548 	if (!rx_ring->desc)
6549 		rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
6550 						   &rx_ring->dma, GFP_KERNEL);
6551 	if (!rx_ring->desc)
6552 		goto err;
6553 
6554 	rx_ring->next_to_clean = 0;
6555 	rx_ring->next_to_use = 0;
6556 
6557 	/* XDP RX-queue info */
6558 	if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
6559 			     rx_ring->queue_index) < 0)
6560 		goto err;
6561 
6562 	rx_ring->xdp_prog = adapter->xdp_prog;
6563 
6564 	return 0;
6565 err:
6566 	vfree(rx_ring->rx_buffer_info);
6567 	rx_ring->rx_buffer_info = NULL;
6568 	dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
6569 	return -ENOMEM;
6570 }
6571 
6572 /**
6573  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
6574  * @adapter: board private structure
6575  *
6576  * If this function returns with an error, then it's possible one or
6577  * more of the rings is populated (while the rest are not).  It is the
6578  * callers duty to clean those orphaned rings.
6579  *
6580  * Return 0 on success, negative on failure
6581  **/
6582 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
6583 {
6584 	int i, err = 0;
6585 
6586 	for (i = 0; i < adapter->num_rx_queues; i++) {
6587 		err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
6588 		if (!err)
6589 			continue;
6590 
6591 		e_err(probe, "Allocation for Rx Queue %u failed\n", i);
6592 		goto err_setup_rx;
6593 	}
6594 
6595 #ifdef IXGBE_FCOE
6596 	err = ixgbe_setup_fcoe_ddp_resources(adapter);
6597 	if (!err)
6598 #endif
6599 		return 0;
6600 err_setup_rx:
6601 	/* rewind the index freeing the rings as we go */
6602 	while (i--)
6603 		ixgbe_free_rx_resources(adapter->rx_ring[i]);
6604 	return err;
6605 }
6606 
6607 /**
6608  * ixgbe_free_tx_resources - Free Tx Resources per Queue
6609  * @tx_ring: Tx descriptor ring for a specific queue
6610  *
6611  * Free all transmit software resources
6612  **/
6613 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
6614 {
6615 	ixgbe_clean_tx_ring(tx_ring);
6616 
6617 	vfree(tx_ring->tx_buffer_info);
6618 	tx_ring->tx_buffer_info = NULL;
6619 
6620 	/* if not set, then don't free */
6621 	if (!tx_ring->desc)
6622 		return;
6623 
6624 	dma_free_coherent(tx_ring->dev, tx_ring->size,
6625 			  tx_ring->desc, tx_ring->dma);
6626 
6627 	tx_ring->desc = NULL;
6628 }
6629 
6630 /**
6631  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
6632  * @adapter: board private structure
6633  *
6634  * Free all transmit software resources
6635  **/
6636 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
6637 {
6638 	int i;
6639 
6640 	for (i = 0; i < adapter->num_tx_queues; i++)
6641 		if (adapter->tx_ring[i]->desc)
6642 			ixgbe_free_tx_resources(adapter->tx_ring[i]);
6643 	for (i = 0; i < adapter->num_xdp_queues; i++)
6644 		if (adapter->xdp_ring[i]->desc)
6645 			ixgbe_free_tx_resources(adapter->xdp_ring[i]);
6646 }
6647 
6648 /**
6649  * ixgbe_free_rx_resources - Free Rx Resources
6650  * @rx_ring: ring to clean the resources from
6651  *
6652  * Free all receive software resources
6653  **/
6654 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
6655 {
6656 	ixgbe_clean_rx_ring(rx_ring);
6657 
6658 	rx_ring->xdp_prog = NULL;
6659 	xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
6660 	vfree(rx_ring->rx_buffer_info);
6661 	rx_ring->rx_buffer_info = NULL;
6662 
6663 	/* if not set, then don't free */
6664 	if (!rx_ring->desc)
6665 		return;
6666 
6667 	dma_free_coherent(rx_ring->dev, rx_ring->size,
6668 			  rx_ring->desc, rx_ring->dma);
6669 
6670 	rx_ring->desc = NULL;
6671 }
6672 
6673 /**
6674  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
6675  * @adapter: board private structure
6676  *
6677  * Free all receive software resources
6678  **/
6679 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
6680 {
6681 	int i;
6682 
6683 #ifdef IXGBE_FCOE
6684 	ixgbe_free_fcoe_ddp_resources(adapter);
6685 
6686 #endif
6687 	for (i = 0; i < adapter->num_rx_queues; i++)
6688 		if (adapter->rx_ring[i]->desc)
6689 			ixgbe_free_rx_resources(adapter->rx_ring[i]);
6690 }
6691 
6692 /**
6693  * ixgbe_change_mtu - Change the Maximum Transfer Unit
6694  * @netdev: network interface device structure
6695  * @new_mtu: new value for maximum frame size
6696  *
6697  * Returns 0 on success, negative on failure
6698  **/
6699 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
6700 {
6701 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6702 
6703 	if (adapter->xdp_prog) {
6704 		int new_frame_size = new_mtu + ETH_HLEN + ETH_FCS_LEN +
6705 				     VLAN_HLEN;
6706 		int i;
6707 
6708 		for (i = 0; i < adapter->num_rx_queues; i++) {
6709 			struct ixgbe_ring *ring = adapter->rx_ring[i];
6710 
6711 			if (new_frame_size > ixgbe_rx_bufsz(ring)) {
6712 				e_warn(probe, "Requested MTU size is not supported with XDP\n");
6713 				return -EINVAL;
6714 			}
6715 		}
6716 	}
6717 
6718 	/*
6719 	 * For 82599EB we cannot allow legacy VFs to enable their receive
6720 	 * paths when MTU greater than 1500 is configured.  So display a
6721 	 * warning that legacy VFs will be disabled.
6722 	 */
6723 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
6724 	    (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
6725 	    (new_mtu > ETH_DATA_LEN))
6726 		e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
6727 
6728 	netdev_dbg(netdev, "changing MTU from %d to %d\n",
6729 		   netdev->mtu, new_mtu);
6730 
6731 	/* must set new MTU before calling down or up */
6732 	netdev->mtu = new_mtu;
6733 
6734 	if (netif_running(netdev))
6735 		ixgbe_reinit_locked(adapter);
6736 
6737 	return 0;
6738 }
6739 
6740 /**
6741  * ixgbe_open - Called when a network interface is made active
6742  * @netdev: network interface device structure
6743  *
6744  * Returns 0 on success, negative value on failure
6745  *
6746  * The open entry point is called when a network interface is made
6747  * active by the system (IFF_UP).  At this point all resources needed
6748  * for transmit and receive operations are allocated, the interrupt
6749  * handler is registered with the OS, the watchdog timer is started,
6750  * and the stack is notified that the interface is ready.
6751  **/
6752 int ixgbe_open(struct net_device *netdev)
6753 {
6754 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6755 	struct ixgbe_hw *hw = &adapter->hw;
6756 	int err, queues;
6757 
6758 	/* disallow open during test */
6759 	if (test_bit(__IXGBE_TESTING, &adapter->state))
6760 		return -EBUSY;
6761 
6762 	netif_carrier_off(netdev);
6763 
6764 	/* allocate transmit descriptors */
6765 	err = ixgbe_setup_all_tx_resources(adapter);
6766 	if (err)
6767 		goto err_setup_tx;
6768 
6769 	/* allocate receive descriptors */
6770 	err = ixgbe_setup_all_rx_resources(adapter);
6771 	if (err)
6772 		goto err_setup_rx;
6773 
6774 	ixgbe_configure(adapter);
6775 
6776 	err = ixgbe_request_irq(adapter);
6777 	if (err)
6778 		goto err_req_irq;
6779 
6780 	/* Notify the stack of the actual queue counts. */
6781 	queues = adapter->num_tx_queues;
6782 	err = netif_set_real_num_tx_queues(netdev, queues);
6783 	if (err)
6784 		goto err_set_queues;
6785 
6786 	queues = adapter->num_rx_queues;
6787 	err = netif_set_real_num_rx_queues(netdev, queues);
6788 	if (err)
6789 		goto err_set_queues;
6790 
6791 	ixgbe_ptp_init(adapter);
6792 
6793 	ixgbe_up_complete(adapter);
6794 
6795 	ixgbe_clear_udp_tunnel_port(adapter, IXGBE_VXLANCTRL_ALL_UDPPORT_MASK);
6796 	udp_tunnel_get_rx_info(netdev);
6797 
6798 	return 0;
6799 
6800 err_set_queues:
6801 	ixgbe_free_irq(adapter);
6802 err_req_irq:
6803 	ixgbe_free_all_rx_resources(adapter);
6804 	if (hw->phy.ops.set_phy_power && !adapter->wol)
6805 		hw->phy.ops.set_phy_power(&adapter->hw, false);
6806 err_setup_rx:
6807 	ixgbe_free_all_tx_resources(adapter);
6808 err_setup_tx:
6809 	ixgbe_reset(adapter);
6810 
6811 	return err;
6812 }
6813 
6814 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
6815 {
6816 	ixgbe_ptp_suspend(adapter);
6817 
6818 	if (adapter->hw.phy.ops.enter_lplu) {
6819 		adapter->hw.phy.reset_disable = true;
6820 		ixgbe_down(adapter);
6821 		adapter->hw.phy.ops.enter_lplu(&adapter->hw);
6822 		adapter->hw.phy.reset_disable = false;
6823 	} else {
6824 		ixgbe_down(adapter);
6825 	}
6826 
6827 	ixgbe_free_irq(adapter);
6828 
6829 	ixgbe_free_all_tx_resources(adapter);
6830 	ixgbe_free_all_rx_resources(adapter);
6831 }
6832 
6833 /**
6834  * ixgbe_close - Disables a network interface
6835  * @netdev: network interface device structure
6836  *
6837  * Returns 0, this is not allowed to fail
6838  *
6839  * The close entry point is called when an interface is de-activated
6840  * by the OS.  The hardware is still under the drivers control, but
6841  * needs to be disabled.  A global MAC reset is issued to stop the
6842  * hardware, and all transmit and receive resources are freed.
6843  **/
6844 int ixgbe_close(struct net_device *netdev)
6845 {
6846 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6847 
6848 	ixgbe_ptp_stop(adapter);
6849 
6850 	if (netif_device_present(netdev))
6851 		ixgbe_close_suspend(adapter);
6852 
6853 	ixgbe_fdir_filter_exit(adapter);
6854 
6855 	ixgbe_release_hw_control(adapter);
6856 
6857 	return 0;
6858 }
6859 
6860 #ifdef CONFIG_PM
6861 static int ixgbe_resume(struct pci_dev *pdev)
6862 {
6863 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6864 	struct net_device *netdev = adapter->netdev;
6865 	u32 err;
6866 
6867 	adapter->hw.hw_addr = adapter->io_addr;
6868 	pci_set_power_state(pdev, PCI_D0);
6869 	pci_restore_state(pdev);
6870 	/*
6871 	 * pci_restore_state clears dev->state_saved so call
6872 	 * pci_save_state to restore it.
6873 	 */
6874 	pci_save_state(pdev);
6875 
6876 	err = pci_enable_device_mem(pdev);
6877 	if (err) {
6878 		e_dev_err("Cannot enable PCI device from suspend\n");
6879 		return err;
6880 	}
6881 	smp_mb__before_atomic();
6882 	clear_bit(__IXGBE_DISABLED, &adapter->state);
6883 	pci_set_master(pdev);
6884 
6885 	pci_wake_from_d3(pdev, false);
6886 
6887 	ixgbe_reset(adapter);
6888 
6889 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6890 
6891 	rtnl_lock();
6892 	err = ixgbe_init_interrupt_scheme(adapter);
6893 	if (!err && netif_running(netdev))
6894 		err = ixgbe_open(netdev);
6895 
6896 
6897 	if (!err)
6898 		netif_device_attach(netdev);
6899 	rtnl_unlock();
6900 
6901 	return err;
6902 }
6903 #endif /* CONFIG_PM */
6904 
6905 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
6906 {
6907 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6908 	struct net_device *netdev = adapter->netdev;
6909 	struct ixgbe_hw *hw = &adapter->hw;
6910 	u32 ctrl;
6911 	u32 wufc = adapter->wol;
6912 #ifdef CONFIG_PM
6913 	int retval = 0;
6914 #endif
6915 
6916 	rtnl_lock();
6917 	netif_device_detach(netdev);
6918 
6919 	if (netif_running(netdev))
6920 		ixgbe_close_suspend(adapter);
6921 
6922 	ixgbe_clear_interrupt_scheme(adapter);
6923 	rtnl_unlock();
6924 
6925 #ifdef CONFIG_PM
6926 	retval = pci_save_state(pdev);
6927 	if (retval)
6928 		return retval;
6929 
6930 #endif
6931 	if (hw->mac.ops.stop_link_on_d3)
6932 		hw->mac.ops.stop_link_on_d3(hw);
6933 
6934 	if (wufc) {
6935 		u32 fctrl;
6936 
6937 		ixgbe_set_rx_mode(netdev);
6938 
6939 		/* enable the optics for 82599 SFP+ fiber as we can WoL */
6940 		if (hw->mac.ops.enable_tx_laser)
6941 			hw->mac.ops.enable_tx_laser(hw);
6942 
6943 		/* enable the reception of multicast packets */
6944 		fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6945 		fctrl |= IXGBE_FCTRL_MPE;
6946 		IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
6947 
6948 		ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
6949 		ctrl |= IXGBE_CTRL_GIO_DIS;
6950 		IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
6951 
6952 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
6953 	} else {
6954 		IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
6955 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
6956 	}
6957 
6958 	switch (hw->mac.type) {
6959 	case ixgbe_mac_82598EB:
6960 		pci_wake_from_d3(pdev, false);
6961 		break;
6962 	case ixgbe_mac_82599EB:
6963 	case ixgbe_mac_X540:
6964 	case ixgbe_mac_X550:
6965 	case ixgbe_mac_X550EM_x:
6966 	case ixgbe_mac_x550em_a:
6967 		pci_wake_from_d3(pdev, !!wufc);
6968 		break;
6969 	default:
6970 		break;
6971 	}
6972 
6973 	*enable_wake = !!wufc;
6974 	if (hw->phy.ops.set_phy_power && !*enable_wake)
6975 		hw->phy.ops.set_phy_power(hw, false);
6976 
6977 	ixgbe_release_hw_control(adapter);
6978 
6979 	if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
6980 		pci_disable_device(pdev);
6981 
6982 	return 0;
6983 }
6984 
6985 #ifdef CONFIG_PM
6986 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
6987 {
6988 	int retval;
6989 	bool wake;
6990 
6991 	retval = __ixgbe_shutdown(pdev, &wake);
6992 	if (retval)
6993 		return retval;
6994 
6995 	if (wake) {
6996 		pci_prepare_to_sleep(pdev);
6997 	} else {
6998 		pci_wake_from_d3(pdev, false);
6999 		pci_set_power_state(pdev, PCI_D3hot);
7000 	}
7001 
7002 	return 0;
7003 }
7004 #endif /* CONFIG_PM */
7005 
7006 static void ixgbe_shutdown(struct pci_dev *pdev)
7007 {
7008 	bool wake;
7009 
7010 	__ixgbe_shutdown(pdev, &wake);
7011 
7012 	if (system_state == SYSTEM_POWER_OFF) {
7013 		pci_wake_from_d3(pdev, wake);
7014 		pci_set_power_state(pdev, PCI_D3hot);
7015 	}
7016 }
7017 
7018 /**
7019  * ixgbe_update_stats - Update the board statistics counters.
7020  * @adapter: board private structure
7021  **/
7022 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
7023 {
7024 	struct net_device *netdev = adapter->netdev;
7025 	struct ixgbe_hw *hw = &adapter->hw;
7026 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
7027 	u64 total_mpc = 0;
7028 	u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
7029 	u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
7030 	u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
7031 	u64 alloc_rx_page = 0;
7032 	u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
7033 
7034 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7035 	    test_bit(__IXGBE_RESETTING, &adapter->state))
7036 		return;
7037 
7038 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
7039 		u64 rsc_count = 0;
7040 		u64 rsc_flush = 0;
7041 		for (i = 0; i < adapter->num_rx_queues; i++) {
7042 			rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
7043 			rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
7044 		}
7045 		adapter->rsc_total_count = rsc_count;
7046 		adapter->rsc_total_flush = rsc_flush;
7047 	}
7048 
7049 	for (i = 0; i < adapter->num_rx_queues; i++) {
7050 		struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
7051 		non_eop_descs += rx_ring->rx_stats.non_eop_descs;
7052 		alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
7053 		alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
7054 		alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
7055 		hw_csum_rx_error += rx_ring->rx_stats.csum_err;
7056 		bytes += rx_ring->stats.bytes;
7057 		packets += rx_ring->stats.packets;
7058 	}
7059 	adapter->non_eop_descs = non_eop_descs;
7060 	adapter->alloc_rx_page = alloc_rx_page;
7061 	adapter->alloc_rx_page_failed = alloc_rx_page_failed;
7062 	adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
7063 	adapter->hw_csum_rx_error = hw_csum_rx_error;
7064 	netdev->stats.rx_bytes = bytes;
7065 	netdev->stats.rx_packets = packets;
7066 
7067 	bytes = 0;
7068 	packets = 0;
7069 	/* gather some stats to the adapter struct that are per queue */
7070 	for (i = 0; i < adapter->num_tx_queues; i++) {
7071 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7072 		restart_queue += tx_ring->tx_stats.restart_queue;
7073 		tx_busy += tx_ring->tx_stats.tx_busy;
7074 		bytes += tx_ring->stats.bytes;
7075 		packets += tx_ring->stats.packets;
7076 	}
7077 	for (i = 0; i < adapter->num_xdp_queues; i++) {
7078 		struct ixgbe_ring *xdp_ring = adapter->xdp_ring[i];
7079 
7080 		restart_queue += xdp_ring->tx_stats.restart_queue;
7081 		tx_busy += xdp_ring->tx_stats.tx_busy;
7082 		bytes += xdp_ring->stats.bytes;
7083 		packets += xdp_ring->stats.packets;
7084 	}
7085 	adapter->restart_queue = restart_queue;
7086 	adapter->tx_busy = tx_busy;
7087 	netdev->stats.tx_bytes = bytes;
7088 	netdev->stats.tx_packets = packets;
7089 
7090 	hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
7091 
7092 	/* 8 register reads */
7093 	for (i = 0; i < 8; i++) {
7094 		/* for packet buffers not used, the register should read 0 */
7095 		mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
7096 		missed_rx += mpc;
7097 		hwstats->mpc[i] += mpc;
7098 		total_mpc += hwstats->mpc[i];
7099 		hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
7100 		hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
7101 		switch (hw->mac.type) {
7102 		case ixgbe_mac_82598EB:
7103 			hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
7104 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
7105 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
7106 			hwstats->pxonrxc[i] +=
7107 				IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
7108 			break;
7109 		case ixgbe_mac_82599EB:
7110 		case ixgbe_mac_X540:
7111 		case ixgbe_mac_X550:
7112 		case ixgbe_mac_X550EM_x:
7113 		case ixgbe_mac_x550em_a:
7114 			hwstats->pxonrxc[i] +=
7115 				IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
7116 			break;
7117 		default:
7118 			break;
7119 		}
7120 	}
7121 
7122 	/*16 register reads */
7123 	for (i = 0; i < 16; i++) {
7124 		hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
7125 		hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
7126 		if ((hw->mac.type == ixgbe_mac_82599EB) ||
7127 		    (hw->mac.type == ixgbe_mac_X540) ||
7128 		    (hw->mac.type == ixgbe_mac_X550) ||
7129 		    (hw->mac.type == ixgbe_mac_X550EM_x) ||
7130 		    (hw->mac.type == ixgbe_mac_x550em_a)) {
7131 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
7132 			IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
7133 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
7134 			IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
7135 		}
7136 	}
7137 
7138 	hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
7139 	/* work around hardware counting issue */
7140 	hwstats->gprc -= missed_rx;
7141 
7142 	ixgbe_update_xoff_received(adapter);
7143 
7144 	/* 82598 hardware only has a 32 bit counter in the high register */
7145 	switch (hw->mac.type) {
7146 	case ixgbe_mac_82598EB:
7147 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
7148 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
7149 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
7150 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
7151 		break;
7152 	case ixgbe_mac_X540:
7153 	case ixgbe_mac_X550:
7154 	case ixgbe_mac_X550EM_x:
7155 	case ixgbe_mac_x550em_a:
7156 		/* OS2BMC stats are X540 and later */
7157 		hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
7158 		hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
7159 		hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
7160 		hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
7161 		/* fall through */
7162 	case ixgbe_mac_82599EB:
7163 		for (i = 0; i < 16; i++)
7164 			adapter->hw_rx_no_dma_resources +=
7165 					     IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
7166 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
7167 		IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
7168 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
7169 		IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
7170 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
7171 		IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
7172 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
7173 		hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
7174 		hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
7175 #ifdef IXGBE_FCOE
7176 		hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
7177 		hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
7178 		hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
7179 		hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
7180 		hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
7181 		hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
7182 		/* Add up per cpu counters for total ddp aloc fail */
7183 		if (adapter->fcoe.ddp_pool) {
7184 			struct ixgbe_fcoe *fcoe = &adapter->fcoe;
7185 			struct ixgbe_fcoe_ddp_pool *ddp_pool;
7186 			unsigned int cpu;
7187 			u64 noddp = 0, noddp_ext_buff = 0;
7188 			for_each_possible_cpu(cpu) {
7189 				ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
7190 				noddp += ddp_pool->noddp;
7191 				noddp_ext_buff += ddp_pool->noddp_ext_buff;
7192 			}
7193 			hwstats->fcoe_noddp = noddp;
7194 			hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
7195 		}
7196 #endif /* IXGBE_FCOE */
7197 		break;
7198 	default:
7199 		break;
7200 	}
7201 	bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
7202 	hwstats->bprc += bprc;
7203 	hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
7204 	if (hw->mac.type == ixgbe_mac_82598EB)
7205 		hwstats->mprc -= bprc;
7206 	hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
7207 	hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
7208 	hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
7209 	hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
7210 	hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
7211 	hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
7212 	hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
7213 	hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
7214 	lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
7215 	hwstats->lxontxc += lxon;
7216 	lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
7217 	hwstats->lxofftxc += lxoff;
7218 	hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
7219 	hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
7220 	/*
7221 	 * 82598 errata - tx of flow control packets is included in tx counters
7222 	 */
7223 	xon_off_tot = lxon + lxoff;
7224 	hwstats->gptc -= xon_off_tot;
7225 	hwstats->mptc -= xon_off_tot;
7226 	hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
7227 	hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
7228 	hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
7229 	hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
7230 	hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
7231 	hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
7232 	hwstats->ptc64 -= xon_off_tot;
7233 	hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
7234 	hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
7235 	hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
7236 	hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
7237 	hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
7238 	hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
7239 
7240 	/* Fill out the OS statistics structure */
7241 	netdev->stats.multicast = hwstats->mprc;
7242 
7243 	/* Rx Errors */
7244 	netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
7245 	netdev->stats.rx_dropped = 0;
7246 	netdev->stats.rx_length_errors = hwstats->rlec;
7247 	netdev->stats.rx_crc_errors = hwstats->crcerrs;
7248 	netdev->stats.rx_missed_errors = total_mpc;
7249 }
7250 
7251 /**
7252  * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
7253  * @adapter: pointer to the device adapter structure
7254  **/
7255 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
7256 {
7257 	struct ixgbe_hw *hw = &adapter->hw;
7258 	int i;
7259 
7260 	if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
7261 		return;
7262 
7263 	adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
7264 
7265 	/* if interface is down do nothing */
7266 	if (test_bit(__IXGBE_DOWN, &adapter->state))
7267 		return;
7268 
7269 	/* do nothing if we are not using signature filters */
7270 	if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
7271 		return;
7272 
7273 	adapter->fdir_overflow++;
7274 
7275 	if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
7276 		for (i = 0; i < adapter->num_tx_queues; i++)
7277 			set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7278 				&(adapter->tx_ring[i]->state));
7279 		for (i = 0; i < adapter->num_xdp_queues; i++)
7280 			set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7281 				&adapter->xdp_ring[i]->state);
7282 		/* re-enable flow director interrupts */
7283 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
7284 	} else {
7285 		e_err(probe, "failed to finish FDIR re-initialization, "
7286 		      "ignored adding FDIR ATR filters\n");
7287 	}
7288 }
7289 
7290 /**
7291  * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
7292  * @adapter: pointer to the device adapter structure
7293  *
7294  * This function serves two purposes.  First it strobes the interrupt lines
7295  * in order to make certain interrupts are occurring.  Secondly it sets the
7296  * bits needed to check for TX hangs.  As a result we should immediately
7297  * determine if a hang has occurred.
7298  */
7299 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
7300 {
7301 	struct ixgbe_hw *hw = &adapter->hw;
7302 	u64 eics = 0;
7303 	int i;
7304 
7305 	/* If we're down, removing or resetting, just bail */
7306 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7307 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
7308 	    test_bit(__IXGBE_RESETTING, &adapter->state))
7309 		return;
7310 
7311 	/* Force detection of hung controller */
7312 	if (netif_carrier_ok(adapter->netdev)) {
7313 		for (i = 0; i < adapter->num_tx_queues; i++)
7314 			set_check_for_tx_hang(adapter->tx_ring[i]);
7315 		for (i = 0; i < adapter->num_xdp_queues; i++)
7316 			set_check_for_tx_hang(adapter->xdp_ring[i]);
7317 	}
7318 
7319 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
7320 		/*
7321 		 * for legacy and MSI interrupts don't set any bits
7322 		 * that are enabled for EIAM, because this operation
7323 		 * would set *both* EIMS and EICS for any bit in EIAM
7324 		 */
7325 		IXGBE_WRITE_REG(hw, IXGBE_EICS,
7326 			(IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
7327 	} else {
7328 		/* get one bit for every active tx/rx interrupt vector */
7329 		for (i = 0; i < adapter->num_q_vectors; i++) {
7330 			struct ixgbe_q_vector *qv = adapter->q_vector[i];
7331 			if (qv->rx.ring || qv->tx.ring)
7332 				eics |= BIT_ULL(i);
7333 		}
7334 	}
7335 
7336 	/* Cause software interrupt to ensure rings are cleaned */
7337 	ixgbe_irq_rearm_queues(adapter, eics);
7338 }
7339 
7340 /**
7341  * ixgbe_watchdog_update_link - update the link status
7342  * @adapter: pointer to the device adapter structure
7343  **/
7344 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
7345 {
7346 	struct ixgbe_hw *hw = &adapter->hw;
7347 	u32 link_speed = adapter->link_speed;
7348 	bool link_up = adapter->link_up;
7349 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
7350 
7351 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
7352 		return;
7353 
7354 	if (hw->mac.ops.check_link) {
7355 		hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
7356 	} else {
7357 		/* always assume link is up, if no check link function */
7358 		link_speed = IXGBE_LINK_SPEED_10GB_FULL;
7359 		link_up = true;
7360 	}
7361 
7362 	if (adapter->ixgbe_ieee_pfc)
7363 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
7364 
7365 	if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
7366 		hw->mac.ops.fc_enable(hw);
7367 		ixgbe_set_rx_drop_en(adapter);
7368 	}
7369 
7370 	if (link_up ||
7371 	    time_after(jiffies, (adapter->link_check_timeout +
7372 				 IXGBE_TRY_LINK_TIMEOUT))) {
7373 		adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
7374 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
7375 		IXGBE_WRITE_FLUSH(hw);
7376 	}
7377 
7378 	adapter->link_up = link_up;
7379 	adapter->link_speed = link_speed;
7380 }
7381 
7382 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
7383 {
7384 #ifdef CONFIG_IXGBE_DCB
7385 	struct net_device *netdev = adapter->netdev;
7386 	struct dcb_app app = {
7387 			      .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
7388 			      .protocol = 0,
7389 			     };
7390 	u8 up = 0;
7391 
7392 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
7393 		up = dcb_ieee_getapp_mask(netdev, &app);
7394 
7395 	adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
7396 #endif
7397 }
7398 
7399 /**
7400  * ixgbe_watchdog_link_is_up - update netif_carrier status and
7401  *                             print link up message
7402  * @adapter: pointer to the device adapter structure
7403  **/
7404 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
7405 {
7406 	struct net_device *netdev = adapter->netdev;
7407 	struct ixgbe_hw *hw = &adapter->hw;
7408 	u32 link_speed = adapter->link_speed;
7409 	const char *speed_str;
7410 	bool flow_rx, flow_tx;
7411 
7412 	/* only continue if link was previously down */
7413 	if (netif_carrier_ok(netdev))
7414 		return;
7415 
7416 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7417 
7418 	switch (hw->mac.type) {
7419 	case ixgbe_mac_82598EB: {
7420 		u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
7421 		u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
7422 		flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
7423 		flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
7424 	}
7425 		break;
7426 	case ixgbe_mac_X540:
7427 	case ixgbe_mac_X550:
7428 	case ixgbe_mac_X550EM_x:
7429 	case ixgbe_mac_x550em_a:
7430 	case ixgbe_mac_82599EB: {
7431 		u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
7432 		u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
7433 		flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
7434 		flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
7435 	}
7436 		break;
7437 	default:
7438 		flow_tx = false;
7439 		flow_rx = false;
7440 		break;
7441 	}
7442 
7443 	adapter->last_rx_ptp_check = jiffies;
7444 
7445 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7446 		ixgbe_ptp_start_cyclecounter(adapter);
7447 
7448 	switch (link_speed) {
7449 	case IXGBE_LINK_SPEED_10GB_FULL:
7450 		speed_str = "10 Gbps";
7451 		break;
7452 	case IXGBE_LINK_SPEED_5GB_FULL:
7453 		speed_str = "5 Gbps";
7454 		break;
7455 	case IXGBE_LINK_SPEED_2_5GB_FULL:
7456 		speed_str = "2.5 Gbps";
7457 		break;
7458 	case IXGBE_LINK_SPEED_1GB_FULL:
7459 		speed_str = "1 Gbps";
7460 		break;
7461 	case IXGBE_LINK_SPEED_100_FULL:
7462 		speed_str = "100 Mbps";
7463 		break;
7464 	case IXGBE_LINK_SPEED_10_FULL:
7465 		speed_str = "10 Mbps";
7466 		break;
7467 	default:
7468 		speed_str = "unknown speed";
7469 		break;
7470 	}
7471 	e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str,
7472 	       ((flow_rx && flow_tx) ? "RX/TX" :
7473 	       (flow_rx ? "RX" :
7474 	       (flow_tx ? "TX" : "None"))));
7475 
7476 	netif_carrier_on(netdev);
7477 	ixgbe_check_vf_rate_limit(adapter);
7478 
7479 	/* enable transmits */
7480 	netif_tx_wake_all_queues(adapter->netdev);
7481 
7482 	/* update the default user priority for VFs */
7483 	ixgbe_update_default_up(adapter);
7484 
7485 	/* ping all the active vfs to let them know link has changed */
7486 	ixgbe_ping_all_vfs(adapter);
7487 }
7488 
7489 /**
7490  * ixgbe_watchdog_link_is_down - update netif_carrier status and
7491  *                               print link down message
7492  * @adapter: pointer to the adapter structure
7493  **/
7494 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
7495 {
7496 	struct net_device *netdev = adapter->netdev;
7497 	struct ixgbe_hw *hw = &adapter->hw;
7498 
7499 	adapter->link_up = false;
7500 	adapter->link_speed = 0;
7501 
7502 	/* only continue if link was up previously */
7503 	if (!netif_carrier_ok(netdev))
7504 		return;
7505 
7506 	/* poll for SFP+ cable when link is down */
7507 	if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
7508 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
7509 
7510 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7511 		ixgbe_ptp_start_cyclecounter(adapter);
7512 
7513 	e_info(drv, "NIC Link is Down\n");
7514 	netif_carrier_off(netdev);
7515 
7516 	/* ping all the active vfs to let them know link has changed */
7517 	ixgbe_ping_all_vfs(adapter);
7518 }
7519 
7520 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
7521 {
7522 	int i;
7523 
7524 	for (i = 0; i < adapter->num_tx_queues; i++) {
7525 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7526 
7527 		if (tx_ring->next_to_use != tx_ring->next_to_clean)
7528 			return true;
7529 	}
7530 
7531 	for (i = 0; i < adapter->num_xdp_queues; i++) {
7532 		struct ixgbe_ring *ring = adapter->xdp_ring[i];
7533 
7534 		if (ring->next_to_use != ring->next_to_clean)
7535 			return true;
7536 	}
7537 
7538 	return false;
7539 }
7540 
7541 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
7542 {
7543 	struct ixgbe_hw *hw = &adapter->hw;
7544 	struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
7545 	u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
7546 
7547 	int i, j;
7548 
7549 	if (!adapter->num_vfs)
7550 		return false;
7551 
7552 	/* resetting the PF is only needed for MAC before X550 */
7553 	if (hw->mac.type >= ixgbe_mac_X550)
7554 		return false;
7555 
7556 	for (i = 0; i < adapter->num_vfs; i++) {
7557 		for (j = 0; j < q_per_pool; j++) {
7558 			u32 h, t;
7559 
7560 			h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
7561 			t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
7562 
7563 			if (h != t)
7564 				return true;
7565 		}
7566 	}
7567 
7568 	return false;
7569 }
7570 
7571 /**
7572  * ixgbe_watchdog_flush_tx - flush queues on link down
7573  * @adapter: pointer to the device adapter structure
7574  **/
7575 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
7576 {
7577 	if (!netif_carrier_ok(adapter->netdev)) {
7578 		if (ixgbe_ring_tx_pending(adapter) ||
7579 		    ixgbe_vf_tx_pending(adapter)) {
7580 			/* We've lost link, so the controller stops DMA,
7581 			 * but we've got queued Tx work that's never going
7582 			 * to get done, so reset controller to flush Tx.
7583 			 * (Do the reset outside of interrupt context).
7584 			 */
7585 			e_warn(drv, "initiating reset to clear Tx work after link loss\n");
7586 			set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
7587 		}
7588 	}
7589 }
7590 
7591 #ifdef CONFIG_PCI_IOV
7592 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
7593 {
7594 	struct ixgbe_hw *hw = &adapter->hw;
7595 	struct pci_dev *pdev = adapter->pdev;
7596 	unsigned int vf;
7597 	u32 gpc;
7598 
7599 	if (!(netif_carrier_ok(adapter->netdev)))
7600 		return;
7601 
7602 	gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
7603 	if (gpc) /* If incrementing then no need for the check below */
7604 		return;
7605 	/* Check to see if a bad DMA write target from an errant or
7606 	 * malicious VF has caused a PCIe error.  If so then we can
7607 	 * issue a VFLR to the offending VF(s) and then resume without
7608 	 * requesting a full slot reset.
7609 	 */
7610 
7611 	if (!pdev)
7612 		return;
7613 
7614 	/* check status reg for all VFs owned by this PF */
7615 	for (vf = 0; vf < adapter->num_vfs; ++vf) {
7616 		struct pci_dev *vfdev = adapter->vfinfo[vf].vfdev;
7617 		u16 status_reg;
7618 
7619 		if (!vfdev)
7620 			continue;
7621 		pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
7622 		if (status_reg != IXGBE_FAILED_READ_CFG_WORD &&
7623 		    status_reg & PCI_STATUS_REC_MASTER_ABORT)
7624 			pcie_flr(vfdev);
7625 	}
7626 }
7627 
7628 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
7629 {
7630 	u32 ssvpc;
7631 
7632 	/* Do not perform spoof check for 82598 or if not in IOV mode */
7633 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7634 	    adapter->num_vfs == 0)
7635 		return;
7636 
7637 	ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
7638 
7639 	/*
7640 	 * ssvpc register is cleared on read, if zero then no
7641 	 * spoofed packets in the last interval.
7642 	 */
7643 	if (!ssvpc)
7644 		return;
7645 
7646 	e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
7647 }
7648 #else
7649 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
7650 {
7651 }
7652 
7653 static void
7654 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
7655 {
7656 }
7657 #endif /* CONFIG_PCI_IOV */
7658 
7659 
7660 /**
7661  * ixgbe_watchdog_subtask - check and bring link up
7662  * @adapter: pointer to the device adapter structure
7663  **/
7664 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
7665 {
7666 	/* if interface is down, removing or resetting, do nothing */
7667 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7668 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
7669 	    test_bit(__IXGBE_RESETTING, &adapter->state))
7670 		return;
7671 
7672 	ixgbe_watchdog_update_link(adapter);
7673 
7674 	if (adapter->link_up)
7675 		ixgbe_watchdog_link_is_up(adapter);
7676 	else
7677 		ixgbe_watchdog_link_is_down(adapter);
7678 
7679 	ixgbe_check_for_bad_vf(adapter);
7680 	ixgbe_spoof_check(adapter);
7681 	ixgbe_update_stats(adapter);
7682 
7683 	ixgbe_watchdog_flush_tx(adapter);
7684 }
7685 
7686 /**
7687  * ixgbe_sfp_detection_subtask - poll for SFP+ cable
7688  * @adapter: the ixgbe adapter structure
7689  **/
7690 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
7691 {
7692 	struct ixgbe_hw *hw = &adapter->hw;
7693 	s32 err;
7694 
7695 	/* not searching for SFP so there is nothing to do here */
7696 	if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
7697 	    !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7698 		return;
7699 
7700 	if (adapter->sfp_poll_time &&
7701 	    time_after(adapter->sfp_poll_time, jiffies))
7702 		return; /* If not yet time to poll for SFP */
7703 
7704 	/* someone else is in init, wait until next service event */
7705 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7706 		return;
7707 
7708 	adapter->sfp_poll_time = jiffies + IXGBE_SFP_POLL_JIFFIES - 1;
7709 
7710 	err = hw->phy.ops.identify_sfp(hw);
7711 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7712 		goto sfp_out;
7713 
7714 	if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
7715 		/* If no cable is present, then we need to reset
7716 		 * the next time we find a good cable. */
7717 		adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
7718 	}
7719 
7720 	/* exit on error */
7721 	if (err)
7722 		goto sfp_out;
7723 
7724 	/* exit if reset not needed */
7725 	if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7726 		goto sfp_out;
7727 
7728 	adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
7729 
7730 	/*
7731 	 * A module may be identified correctly, but the EEPROM may not have
7732 	 * support for that module.  setup_sfp() will fail in that case, so
7733 	 * we should not allow that module to load.
7734 	 */
7735 	if (hw->mac.type == ixgbe_mac_82598EB)
7736 		err = hw->phy.ops.reset(hw);
7737 	else
7738 		err = hw->mac.ops.setup_sfp(hw);
7739 
7740 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7741 		goto sfp_out;
7742 
7743 	adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
7744 	e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
7745 
7746 sfp_out:
7747 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7748 
7749 	if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
7750 	    (adapter->netdev->reg_state == NETREG_REGISTERED)) {
7751 		e_dev_err("failed to initialize because an unsupported "
7752 			  "SFP+ module type was detected.\n");
7753 		e_dev_err("Reload the driver after installing a "
7754 			  "supported module.\n");
7755 		unregister_netdev(adapter->netdev);
7756 	}
7757 }
7758 
7759 /**
7760  * ixgbe_sfp_link_config_subtask - set up link SFP after module install
7761  * @adapter: the ixgbe adapter structure
7762  **/
7763 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
7764 {
7765 	struct ixgbe_hw *hw = &adapter->hw;
7766 	u32 cap_speed;
7767 	u32 speed;
7768 	bool autoneg = false;
7769 
7770 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
7771 		return;
7772 
7773 	/* someone else is in init, wait until next service event */
7774 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7775 		return;
7776 
7777 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
7778 
7779 	hw->mac.ops.get_link_capabilities(hw, &cap_speed, &autoneg);
7780 
7781 	/* advertise highest capable link speed */
7782 	if (!autoneg && (cap_speed & IXGBE_LINK_SPEED_10GB_FULL))
7783 		speed = IXGBE_LINK_SPEED_10GB_FULL;
7784 	else
7785 		speed = cap_speed & (IXGBE_LINK_SPEED_10GB_FULL |
7786 				     IXGBE_LINK_SPEED_1GB_FULL);
7787 
7788 	if (hw->mac.ops.setup_link)
7789 		hw->mac.ops.setup_link(hw, speed, true);
7790 
7791 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
7792 	adapter->link_check_timeout = jiffies;
7793 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7794 }
7795 
7796 /**
7797  * ixgbe_service_timer - Timer Call-back
7798  * @t: pointer to timer_list structure
7799  **/
7800 static void ixgbe_service_timer(struct timer_list *t)
7801 {
7802 	struct ixgbe_adapter *adapter = from_timer(adapter, t, service_timer);
7803 	unsigned long next_event_offset;
7804 
7805 	/* poll faster when waiting for link */
7806 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
7807 		next_event_offset = HZ / 10;
7808 	else
7809 		next_event_offset = HZ * 2;
7810 
7811 	/* Reset the timer */
7812 	mod_timer(&adapter->service_timer, next_event_offset + jiffies);
7813 
7814 	ixgbe_service_event_schedule(adapter);
7815 }
7816 
7817 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter)
7818 {
7819 	struct ixgbe_hw *hw = &adapter->hw;
7820 	u32 status;
7821 
7822 	if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT))
7823 		return;
7824 
7825 	adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT;
7826 
7827 	if (!hw->phy.ops.handle_lasi)
7828 		return;
7829 
7830 	status = hw->phy.ops.handle_lasi(&adapter->hw);
7831 	if (status != IXGBE_ERR_OVERTEMP)
7832 		return;
7833 
7834 	e_crit(drv, "%s\n", ixgbe_overheat_msg);
7835 }
7836 
7837 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
7838 {
7839 	if (!test_and_clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state))
7840 		return;
7841 
7842 	rtnl_lock();
7843 	/* If we're already down, removing or resetting, just bail */
7844 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7845 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
7846 	    test_bit(__IXGBE_RESETTING, &adapter->state)) {
7847 		rtnl_unlock();
7848 		return;
7849 	}
7850 
7851 	ixgbe_dump(adapter);
7852 	netdev_err(adapter->netdev, "Reset adapter\n");
7853 	adapter->tx_timeout_count++;
7854 
7855 	ixgbe_reinit_locked(adapter);
7856 	rtnl_unlock();
7857 }
7858 
7859 /**
7860  * ixgbe_check_fw_error - Check firmware for errors
7861  * @adapter: the adapter private structure
7862  *
7863  * Check firmware errors in register FWSM
7864  */
7865 static bool ixgbe_check_fw_error(struct ixgbe_adapter *adapter)
7866 {
7867 	struct ixgbe_hw *hw = &adapter->hw;
7868 	u32 fwsm;
7869 
7870 	/* read fwsm.ext_err_ind register and log errors */
7871 	fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
7872 
7873 	if (fwsm & IXGBE_FWSM_EXT_ERR_IND_MASK ||
7874 	    !(fwsm & IXGBE_FWSM_FW_VAL_BIT))
7875 		e_dev_warn("Warning firmware error detected FWSM: 0x%08X\n",
7876 			   fwsm);
7877 
7878 	if (hw->mac.ops.fw_recovery_mode && hw->mac.ops.fw_recovery_mode(hw)) {
7879 		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");
7880 		return true;
7881 	}
7882 
7883 	return false;
7884 }
7885 
7886 /**
7887  * ixgbe_service_task - manages and runs subtasks
7888  * @work: pointer to work_struct containing our data
7889  **/
7890 static void ixgbe_service_task(struct work_struct *work)
7891 {
7892 	struct ixgbe_adapter *adapter = container_of(work,
7893 						     struct ixgbe_adapter,
7894 						     service_task);
7895 	if (ixgbe_removed(adapter->hw.hw_addr)) {
7896 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
7897 			rtnl_lock();
7898 			ixgbe_down(adapter);
7899 			rtnl_unlock();
7900 		}
7901 		ixgbe_service_event_complete(adapter);
7902 		return;
7903 	}
7904 	if (ixgbe_check_fw_error(adapter)) {
7905 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
7906 			unregister_netdev(adapter->netdev);
7907 		ixgbe_service_event_complete(adapter);
7908 		return;
7909 	}
7910 	if (adapter->flags2 & IXGBE_FLAG2_UDP_TUN_REREG_NEEDED) {
7911 		rtnl_lock();
7912 		adapter->flags2 &= ~IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
7913 		udp_tunnel_get_rx_info(adapter->netdev);
7914 		rtnl_unlock();
7915 	}
7916 	ixgbe_reset_subtask(adapter);
7917 	ixgbe_phy_interrupt_subtask(adapter);
7918 	ixgbe_sfp_detection_subtask(adapter);
7919 	ixgbe_sfp_link_config_subtask(adapter);
7920 	ixgbe_check_overtemp_subtask(adapter);
7921 	ixgbe_watchdog_subtask(adapter);
7922 	ixgbe_fdir_reinit_subtask(adapter);
7923 	ixgbe_check_hang_subtask(adapter);
7924 
7925 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
7926 		ixgbe_ptp_overflow_check(adapter);
7927 		if (adapter->flags & IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER)
7928 			ixgbe_ptp_rx_hang(adapter);
7929 		ixgbe_ptp_tx_hang(adapter);
7930 	}
7931 
7932 	ixgbe_service_event_complete(adapter);
7933 }
7934 
7935 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
7936 		     struct ixgbe_tx_buffer *first,
7937 		     u8 *hdr_len,
7938 		     struct ixgbe_ipsec_tx_data *itd)
7939 {
7940 	u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
7941 	struct sk_buff *skb = first->skb;
7942 	union {
7943 		struct iphdr *v4;
7944 		struct ipv6hdr *v6;
7945 		unsigned char *hdr;
7946 	} ip;
7947 	union {
7948 		struct tcphdr *tcp;
7949 		struct udphdr *udp;
7950 		unsigned char *hdr;
7951 	} l4;
7952 	u32 paylen, l4_offset;
7953 	u32 fceof_saidx = 0;
7954 	int err;
7955 
7956 	if (skb->ip_summed != CHECKSUM_PARTIAL)
7957 		return 0;
7958 
7959 	if (!skb_is_gso(skb))
7960 		return 0;
7961 
7962 	err = skb_cow_head(skb, 0);
7963 	if (err < 0)
7964 		return err;
7965 
7966 	if (eth_p_mpls(first->protocol))
7967 		ip.hdr = skb_inner_network_header(skb);
7968 	else
7969 		ip.hdr = skb_network_header(skb);
7970 	l4.hdr = skb_checksum_start(skb);
7971 
7972 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
7973 	type_tucmd = (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ?
7974 		      IXGBE_ADVTXD_TUCMD_L4T_UDP : IXGBE_ADVTXD_TUCMD_L4T_TCP;
7975 
7976 	/* initialize outer IP header fields */
7977 	if (ip.v4->version == 4) {
7978 		unsigned char *csum_start = skb_checksum_start(skb);
7979 		unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
7980 		int len = csum_start - trans_start;
7981 
7982 		/* IP header will have to cancel out any data that
7983 		 * is not a part of the outer IP header, so set to
7984 		 * a reverse csum if needed, else init check to 0.
7985 		 */
7986 		ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
7987 					   csum_fold(csum_partial(trans_start,
7988 								  len, 0)) : 0;
7989 		type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
7990 
7991 		ip.v4->tot_len = 0;
7992 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
7993 				   IXGBE_TX_FLAGS_CSUM |
7994 				   IXGBE_TX_FLAGS_IPV4;
7995 	} else {
7996 		ip.v6->payload_len = 0;
7997 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
7998 				   IXGBE_TX_FLAGS_CSUM;
7999 	}
8000 
8001 	/* determine offset of inner transport header */
8002 	l4_offset = l4.hdr - skb->data;
8003 
8004 	/* remove payload length from inner checksum */
8005 	paylen = skb->len - l4_offset;
8006 
8007 	if (type_tucmd & IXGBE_ADVTXD_TUCMD_L4T_TCP) {
8008 		/* compute length of segmentation header */
8009 		*hdr_len = (l4.tcp->doff * 4) + l4_offset;
8010 		csum_replace_by_diff(&l4.tcp->check,
8011 				     (__force __wsum)htonl(paylen));
8012 	} else {
8013 		/* compute length of segmentation header */
8014 		*hdr_len = sizeof(*l4.udp) + l4_offset;
8015 		csum_replace_by_diff(&l4.udp->check,
8016 				     (__force __wsum)htonl(paylen));
8017 	}
8018 
8019 	/* update gso size and bytecount with header size */
8020 	first->gso_segs = skb_shinfo(skb)->gso_segs;
8021 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
8022 
8023 	/* mss_l4len_id: use 0 as index for TSO */
8024 	mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
8025 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
8026 
8027 	fceof_saidx |= itd->sa_idx;
8028 	type_tucmd |= itd->flags | itd->trailer_len;
8029 
8030 	/* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
8031 	vlan_macip_lens = l4.hdr - ip.hdr;
8032 	vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
8033 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8034 
8035 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
8036 			  mss_l4len_idx);
8037 
8038 	return 1;
8039 }
8040 
8041 static inline bool ixgbe_ipv6_csum_is_sctp(struct sk_buff *skb)
8042 {
8043 	unsigned int offset = 0;
8044 
8045 	ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
8046 
8047 	return offset == skb_checksum_start_offset(skb);
8048 }
8049 
8050 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
8051 			  struct ixgbe_tx_buffer *first,
8052 			  struct ixgbe_ipsec_tx_data *itd)
8053 {
8054 	struct sk_buff *skb = first->skb;
8055 	u32 vlan_macip_lens = 0;
8056 	u32 fceof_saidx = 0;
8057 	u32 type_tucmd = 0;
8058 
8059 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
8060 csum_failed:
8061 		if (!(first->tx_flags & (IXGBE_TX_FLAGS_HW_VLAN |
8062 					 IXGBE_TX_FLAGS_CC)))
8063 			return;
8064 		goto no_csum;
8065 	}
8066 
8067 	switch (skb->csum_offset) {
8068 	case offsetof(struct tcphdr, check):
8069 		type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
8070 		/* fall through */
8071 	case offsetof(struct udphdr, check):
8072 		break;
8073 	case offsetof(struct sctphdr, checksum):
8074 		/* validate that this is actually an SCTP request */
8075 		if (((first->protocol == htons(ETH_P_IP)) &&
8076 		     (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
8077 		    ((first->protocol == htons(ETH_P_IPV6)) &&
8078 		     ixgbe_ipv6_csum_is_sctp(skb))) {
8079 			type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
8080 			break;
8081 		}
8082 		/* fall through */
8083 	default:
8084 		skb_checksum_help(skb);
8085 		goto csum_failed;
8086 	}
8087 
8088 	/* update TX checksum flag */
8089 	first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
8090 	vlan_macip_lens = skb_checksum_start_offset(skb) -
8091 			  skb_network_offset(skb);
8092 no_csum:
8093 	/* vlan_macip_lens: MACLEN, VLAN tag */
8094 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
8095 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8096 
8097 	fceof_saidx |= itd->sa_idx;
8098 	type_tucmd |= itd->flags | itd->trailer_len;
8099 
8100 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd, 0);
8101 }
8102 
8103 #define IXGBE_SET_FLAG(_input, _flag, _result) \
8104 	((_flag <= _result) ? \
8105 	 ((u32)(_input & _flag) * (_result / _flag)) : \
8106 	 ((u32)(_input & _flag) / (_flag / _result)))
8107 
8108 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
8109 {
8110 	/* set type for advanced descriptor with frame checksum insertion */
8111 	u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8112 		       IXGBE_ADVTXD_DCMD_DEXT |
8113 		       IXGBE_ADVTXD_DCMD_IFCS;
8114 
8115 	/* set HW vlan bit if vlan is present */
8116 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
8117 				   IXGBE_ADVTXD_DCMD_VLE);
8118 
8119 	/* set segmentation enable bits for TSO/FSO */
8120 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
8121 				   IXGBE_ADVTXD_DCMD_TSE);
8122 
8123 	/* set timestamp bit if present */
8124 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
8125 				   IXGBE_ADVTXD_MAC_TSTAMP);
8126 
8127 	/* insert frame checksum */
8128 	cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
8129 
8130 	return cmd_type;
8131 }
8132 
8133 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
8134 				   u32 tx_flags, unsigned int paylen)
8135 {
8136 	u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
8137 
8138 	/* enable L4 checksum for TSO and TX checksum offload */
8139 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8140 					IXGBE_TX_FLAGS_CSUM,
8141 					IXGBE_ADVTXD_POPTS_TXSM);
8142 
8143 	/* enable IPv4 checksum for TSO */
8144 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8145 					IXGBE_TX_FLAGS_IPV4,
8146 					IXGBE_ADVTXD_POPTS_IXSM);
8147 
8148 	/* enable IPsec */
8149 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8150 					IXGBE_TX_FLAGS_IPSEC,
8151 					IXGBE_ADVTXD_POPTS_IPSEC);
8152 
8153 	/*
8154 	 * Check Context must be set if Tx switch is enabled, which it
8155 	 * always is for case where virtual functions are running
8156 	 */
8157 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8158 					IXGBE_TX_FLAGS_CC,
8159 					IXGBE_ADVTXD_CC);
8160 
8161 	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
8162 }
8163 
8164 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8165 {
8166 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
8167 
8168 	/* Herbert's original patch had:
8169 	 *  smp_mb__after_netif_stop_queue();
8170 	 * but since that doesn't exist yet, just open code it.
8171 	 */
8172 	smp_mb();
8173 
8174 	/* We need to check again in a case another CPU has just
8175 	 * made room available.
8176 	 */
8177 	if (likely(ixgbe_desc_unused(tx_ring) < size))
8178 		return -EBUSY;
8179 
8180 	/* A reprieve! - use start_queue because it doesn't call schedule */
8181 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
8182 	++tx_ring->tx_stats.restart_queue;
8183 	return 0;
8184 }
8185 
8186 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8187 {
8188 	if (likely(ixgbe_desc_unused(tx_ring) >= size))
8189 		return 0;
8190 
8191 	return __ixgbe_maybe_stop_tx(tx_ring, size);
8192 }
8193 
8194 static int ixgbe_tx_map(struct ixgbe_ring *tx_ring,
8195 			struct ixgbe_tx_buffer *first,
8196 			const u8 hdr_len)
8197 {
8198 	struct sk_buff *skb = first->skb;
8199 	struct ixgbe_tx_buffer *tx_buffer;
8200 	union ixgbe_adv_tx_desc *tx_desc;
8201 	skb_frag_t *frag;
8202 	dma_addr_t dma;
8203 	unsigned int data_len, size;
8204 	u32 tx_flags = first->tx_flags;
8205 	u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
8206 	u16 i = tx_ring->next_to_use;
8207 
8208 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
8209 
8210 	ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
8211 
8212 	size = skb_headlen(skb);
8213 	data_len = skb->data_len;
8214 
8215 #ifdef IXGBE_FCOE
8216 	if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
8217 		if (data_len < sizeof(struct fcoe_crc_eof)) {
8218 			size -= sizeof(struct fcoe_crc_eof) - data_len;
8219 			data_len = 0;
8220 		} else {
8221 			data_len -= sizeof(struct fcoe_crc_eof);
8222 		}
8223 	}
8224 
8225 #endif
8226 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
8227 
8228 	tx_buffer = first;
8229 
8230 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
8231 		if (dma_mapping_error(tx_ring->dev, dma))
8232 			goto dma_error;
8233 
8234 		/* record length, and DMA address */
8235 		dma_unmap_len_set(tx_buffer, len, size);
8236 		dma_unmap_addr_set(tx_buffer, dma, dma);
8237 
8238 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
8239 
8240 		while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
8241 			tx_desc->read.cmd_type_len =
8242 				cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
8243 
8244 			i++;
8245 			tx_desc++;
8246 			if (i == tx_ring->count) {
8247 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8248 				i = 0;
8249 			}
8250 			tx_desc->read.olinfo_status = 0;
8251 
8252 			dma += IXGBE_MAX_DATA_PER_TXD;
8253 			size -= IXGBE_MAX_DATA_PER_TXD;
8254 
8255 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
8256 		}
8257 
8258 		if (likely(!data_len))
8259 			break;
8260 
8261 		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
8262 
8263 		i++;
8264 		tx_desc++;
8265 		if (i == tx_ring->count) {
8266 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8267 			i = 0;
8268 		}
8269 		tx_desc->read.olinfo_status = 0;
8270 
8271 #ifdef IXGBE_FCOE
8272 		size = min_t(unsigned int, data_len, skb_frag_size(frag));
8273 #else
8274 		size = skb_frag_size(frag);
8275 #endif
8276 		data_len -= size;
8277 
8278 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
8279 				       DMA_TO_DEVICE);
8280 
8281 		tx_buffer = &tx_ring->tx_buffer_info[i];
8282 	}
8283 
8284 	/* write last descriptor with RS and EOP bits */
8285 	cmd_type |= size | IXGBE_TXD_CMD;
8286 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8287 
8288 	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
8289 
8290 	/* set the timestamp */
8291 	first->time_stamp = jiffies;
8292 
8293 	skb_tx_timestamp(skb);
8294 
8295 	/*
8296 	 * Force memory writes to complete before letting h/w know there
8297 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
8298 	 * memory model archs, such as IA-64).
8299 	 *
8300 	 * We also need this memory barrier to make certain all of the
8301 	 * status bits have been updated before next_to_watch is written.
8302 	 */
8303 	wmb();
8304 
8305 	/* set next_to_watch value indicating a packet is present */
8306 	first->next_to_watch = tx_desc;
8307 
8308 	i++;
8309 	if (i == tx_ring->count)
8310 		i = 0;
8311 
8312 	tx_ring->next_to_use = i;
8313 
8314 	ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
8315 
8316 	if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
8317 		writel(i, tx_ring->tail);
8318 	}
8319 
8320 	return 0;
8321 dma_error:
8322 	dev_err(tx_ring->dev, "TX DMA map failed\n");
8323 
8324 	/* clear dma mappings for failed tx_buffer_info map */
8325 	for (;;) {
8326 		tx_buffer = &tx_ring->tx_buffer_info[i];
8327 		if (dma_unmap_len(tx_buffer, len))
8328 			dma_unmap_page(tx_ring->dev,
8329 				       dma_unmap_addr(tx_buffer, dma),
8330 				       dma_unmap_len(tx_buffer, len),
8331 				       DMA_TO_DEVICE);
8332 		dma_unmap_len_set(tx_buffer, len, 0);
8333 		if (tx_buffer == first)
8334 			break;
8335 		if (i == 0)
8336 			i += tx_ring->count;
8337 		i--;
8338 	}
8339 
8340 	dev_kfree_skb_any(first->skb);
8341 	first->skb = NULL;
8342 
8343 	tx_ring->next_to_use = i;
8344 
8345 	return -1;
8346 }
8347 
8348 static void ixgbe_atr(struct ixgbe_ring *ring,
8349 		      struct ixgbe_tx_buffer *first)
8350 {
8351 	struct ixgbe_q_vector *q_vector = ring->q_vector;
8352 	union ixgbe_atr_hash_dword input = { .dword = 0 };
8353 	union ixgbe_atr_hash_dword common = { .dword = 0 };
8354 	union {
8355 		unsigned char *network;
8356 		struct iphdr *ipv4;
8357 		struct ipv6hdr *ipv6;
8358 	} hdr;
8359 	struct tcphdr *th;
8360 	unsigned int hlen;
8361 	struct sk_buff *skb;
8362 	__be16 vlan_id;
8363 	int l4_proto;
8364 
8365 	/* if ring doesn't have a interrupt vector, cannot perform ATR */
8366 	if (!q_vector)
8367 		return;
8368 
8369 	/* do nothing if sampling is disabled */
8370 	if (!ring->atr_sample_rate)
8371 		return;
8372 
8373 	ring->atr_count++;
8374 
8375 	/* currently only IPv4/IPv6 with TCP is supported */
8376 	if ((first->protocol != htons(ETH_P_IP)) &&
8377 	    (first->protocol != htons(ETH_P_IPV6)))
8378 		return;
8379 
8380 	/* snag network header to get L4 type and address */
8381 	skb = first->skb;
8382 	hdr.network = skb_network_header(skb);
8383 	if (unlikely(hdr.network <= skb->data))
8384 		return;
8385 	if (skb->encapsulation &&
8386 	    first->protocol == htons(ETH_P_IP) &&
8387 	    hdr.ipv4->protocol == IPPROTO_UDP) {
8388 		struct ixgbe_adapter *adapter = q_vector->adapter;
8389 
8390 		if (unlikely(skb_tail_pointer(skb) < hdr.network +
8391 			     VXLAN_HEADROOM))
8392 			return;
8393 
8394 		/* verify the port is recognized as VXLAN */
8395 		if (adapter->vxlan_port &&
8396 		    udp_hdr(skb)->dest == adapter->vxlan_port)
8397 			hdr.network = skb_inner_network_header(skb);
8398 
8399 		if (adapter->geneve_port &&
8400 		    udp_hdr(skb)->dest == adapter->geneve_port)
8401 			hdr.network = skb_inner_network_header(skb);
8402 	}
8403 
8404 	/* Make sure we have at least [minimum IPv4 header + TCP]
8405 	 * or [IPv6 header] bytes
8406 	 */
8407 	if (unlikely(skb_tail_pointer(skb) < hdr.network + 40))
8408 		return;
8409 
8410 	/* Currently only IPv4/IPv6 with TCP is supported */
8411 	switch (hdr.ipv4->version) {
8412 	case IPVERSION:
8413 		/* access ihl as u8 to avoid unaligned access on ia64 */
8414 		hlen = (hdr.network[0] & 0x0F) << 2;
8415 		l4_proto = hdr.ipv4->protocol;
8416 		break;
8417 	case 6:
8418 		hlen = hdr.network - skb->data;
8419 		l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
8420 		hlen -= hdr.network - skb->data;
8421 		break;
8422 	default:
8423 		return;
8424 	}
8425 
8426 	if (l4_proto != IPPROTO_TCP)
8427 		return;
8428 
8429 	if (unlikely(skb_tail_pointer(skb) < hdr.network +
8430 		     hlen + sizeof(struct tcphdr)))
8431 		return;
8432 
8433 	th = (struct tcphdr *)(hdr.network + hlen);
8434 
8435 	/* skip this packet since the socket is closing */
8436 	if (th->fin)
8437 		return;
8438 
8439 	/* sample on all syn packets or once every atr sample count */
8440 	if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
8441 		return;
8442 
8443 	/* reset sample count */
8444 	ring->atr_count = 0;
8445 
8446 	vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
8447 
8448 	/*
8449 	 * src and dst are inverted, think how the receiver sees them
8450 	 *
8451 	 * The input is broken into two sections, a non-compressed section
8452 	 * containing vm_pool, vlan_id, and flow_type.  The rest of the data
8453 	 * is XORed together and stored in the compressed dword.
8454 	 */
8455 	input.formatted.vlan_id = vlan_id;
8456 
8457 	/*
8458 	 * since src port and flex bytes occupy the same word XOR them together
8459 	 * and write the value to source port portion of compressed dword
8460 	 */
8461 	if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
8462 		common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
8463 	else
8464 		common.port.src ^= th->dest ^ first->protocol;
8465 	common.port.dst ^= th->source;
8466 
8467 	switch (hdr.ipv4->version) {
8468 	case IPVERSION:
8469 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
8470 		common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
8471 		break;
8472 	case 6:
8473 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
8474 		common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
8475 			     hdr.ipv6->saddr.s6_addr32[1] ^
8476 			     hdr.ipv6->saddr.s6_addr32[2] ^
8477 			     hdr.ipv6->saddr.s6_addr32[3] ^
8478 			     hdr.ipv6->daddr.s6_addr32[0] ^
8479 			     hdr.ipv6->daddr.s6_addr32[1] ^
8480 			     hdr.ipv6->daddr.s6_addr32[2] ^
8481 			     hdr.ipv6->daddr.s6_addr32[3];
8482 		break;
8483 	default:
8484 		break;
8485 	}
8486 
8487 	if (hdr.network != skb_network_header(skb))
8488 		input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
8489 
8490 	/* This assumes the Rx queue and Tx queue are bound to the same CPU */
8491 	ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
8492 					      input, common, ring->queue_index);
8493 }
8494 
8495 #ifdef IXGBE_FCOE
8496 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
8497 			      struct net_device *sb_dev)
8498 {
8499 	struct ixgbe_adapter *adapter;
8500 	struct ixgbe_ring_feature *f;
8501 	int txq;
8502 
8503 	if (sb_dev) {
8504 		u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
8505 		struct net_device *vdev = sb_dev;
8506 
8507 		txq = vdev->tc_to_txq[tc].offset;
8508 		txq += reciprocal_scale(skb_get_hash(skb),
8509 					vdev->tc_to_txq[tc].count);
8510 
8511 		return txq;
8512 	}
8513 
8514 	/*
8515 	 * only execute the code below if protocol is FCoE
8516 	 * or FIP and we have FCoE enabled on the adapter
8517 	 */
8518 	switch (vlan_get_protocol(skb)) {
8519 	case htons(ETH_P_FCOE):
8520 	case htons(ETH_P_FIP):
8521 		adapter = netdev_priv(dev);
8522 
8523 		if (!sb_dev && (adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
8524 			break;
8525 		/* fall through */
8526 	default:
8527 		return netdev_pick_tx(dev, skb, sb_dev);
8528 	}
8529 
8530 	f = &adapter->ring_feature[RING_F_FCOE];
8531 
8532 	txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
8533 					   smp_processor_id();
8534 
8535 	while (txq >= f->indices)
8536 		txq -= f->indices;
8537 
8538 	return txq + f->offset;
8539 }
8540 
8541 #endif
8542 int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
8543 			struct xdp_frame *xdpf)
8544 {
8545 	struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
8546 	struct ixgbe_tx_buffer *tx_buffer;
8547 	union ixgbe_adv_tx_desc *tx_desc;
8548 	u32 len, cmd_type;
8549 	dma_addr_t dma;
8550 	u16 i;
8551 
8552 	len = xdpf->len;
8553 
8554 	if (unlikely(!ixgbe_desc_unused(ring)))
8555 		return IXGBE_XDP_CONSUMED;
8556 
8557 	dma = dma_map_single(ring->dev, xdpf->data, len, DMA_TO_DEVICE);
8558 	if (dma_mapping_error(ring->dev, dma))
8559 		return IXGBE_XDP_CONSUMED;
8560 
8561 	/* record the location of the first descriptor for this packet */
8562 	tx_buffer = &ring->tx_buffer_info[ring->next_to_use];
8563 	tx_buffer->bytecount = len;
8564 	tx_buffer->gso_segs = 1;
8565 	tx_buffer->protocol = 0;
8566 
8567 	i = ring->next_to_use;
8568 	tx_desc = IXGBE_TX_DESC(ring, i);
8569 
8570 	dma_unmap_len_set(tx_buffer, len, len);
8571 	dma_unmap_addr_set(tx_buffer, dma, dma);
8572 	tx_buffer->xdpf = xdpf;
8573 
8574 	tx_desc->read.buffer_addr = cpu_to_le64(dma);
8575 
8576 	/* put descriptor type bits */
8577 	cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8578 		   IXGBE_ADVTXD_DCMD_DEXT |
8579 		   IXGBE_ADVTXD_DCMD_IFCS;
8580 	cmd_type |= len | IXGBE_TXD_CMD;
8581 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8582 	tx_desc->read.olinfo_status =
8583 		cpu_to_le32(len << IXGBE_ADVTXD_PAYLEN_SHIFT);
8584 
8585 	/* Avoid any potential race with xdp_xmit and cleanup */
8586 	smp_wmb();
8587 
8588 	/* set next_to_watch value indicating a packet is present */
8589 	i++;
8590 	if (i == ring->count)
8591 		i = 0;
8592 
8593 	tx_buffer->next_to_watch = tx_desc;
8594 	ring->next_to_use = i;
8595 
8596 	return IXGBE_XDP_TX;
8597 }
8598 
8599 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
8600 			  struct ixgbe_adapter *adapter,
8601 			  struct ixgbe_ring *tx_ring)
8602 {
8603 	struct ixgbe_tx_buffer *first;
8604 	int tso;
8605 	u32 tx_flags = 0;
8606 	unsigned short f;
8607 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
8608 	struct ixgbe_ipsec_tx_data ipsec_tx = { 0 };
8609 	__be16 protocol = skb->protocol;
8610 	u8 hdr_len = 0;
8611 
8612 	/*
8613 	 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
8614 	 *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
8615 	 *       + 2 desc gap to keep tail from touching head,
8616 	 *       + 1 desc for context descriptor,
8617 	 * otherwise try next time
8618 	 */
8619 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
8620 		count += TXD_USE_COUNT(skb_frag_size(
8621 						&skb_shinfo(skb)->frags[f]));
8622 
8623 	if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
8624 		tx_ring->tx_stats.tx_busy++;
8625 		return NETDEV_TX_BUSY;
8626 	}
8627 
8628 	/* record the location of the first descriptor for this packet */
8629 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
8630 	first->skb = skb;
8631 	first->bytecount = skb->len;
8632 	first->gso_segs = 1;
8633 
8634 	/* if we have a HW VLAN tag being added default to the HW one */
8635 	if (skb_vlan_tag_present(skb)) {
8636 		tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
8637 		tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8638 	/* else if it is a SW VLAN check the next protocol and store the tag */
8639 	} else if (protocol == htons(ETH_P_8021Q)) {
8640 		struct vlan_hdr *vhdr, _vhdr;
8641 		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
8642 		if (!vhdr)
8643 			goto out_drop;
8644 
8645 		tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
8646 				  IXGBE_TX_FLAGS_VLAN_SHIFT;
8647 		tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
8648 	}
8649 	protocol = vlan_get_protocol(skb);
8650 
8651 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
8652 	    adapter->ptp_clock) {
8653 		if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
8654 		    !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
8655 					   &adapter->state)) {
8656 			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
8657 			tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
8658 
8659 			/* schedule check for Tx timestamp */
8660 			adapter->ptp_tx_skb = skb_get(skb);
8661 			adapter->ptp_tx_start = jiffies;
8662 			schedule_work(&adapter->ptp_tx_work);
8663 		} else {
8664 			adapter->tx_hwtstamp_skipped++;
8665 		}
8666 	}
8667 
8668 #ifdef CONFIG_PCI_IOV
8669 	/*
8670 	 * Use the l2switch_enable flag - would be false if the DMA
8671 	 * Tx switch had been disabled.
8672 	 */
8673 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
8674 		tx_flags |= IXGBE_TX_FLAGS_CC;
8675 
8676 #endif
8677 	/* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
8678 	if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8679 	    ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
8680 	     (skb->priority != TC_PRIO_CONTROL))) {
8681 		tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
8682 		tx_flags |= (skb->priority & 0x7) <<
8683 					IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
8684 		if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
8685 			struct vlan_ethhdr *vhdr;
8686 
8687 			if (skb_cow_head(skb, 0))
8688 				goto out_drop;
8689 			vhdr = (struct vlan_ethhdr *)skb->data;
8690 			vhdr->h_vlan_TCI = htons(tx_flags >>
8691 						 IXGBE_TX_FLAGS_VLAN_SHIFT);
8692 		} else {
8693 			tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8694 		}
8695 	}
8696 
8697 	/* record initial flags and protocol */
8698 	first->tx_flags = tx_flags;
8699 	first->protocol = protocol;
8700 
8701 #ifdef IXGBE_FCOE
8702 	/* setup tx offload for FCoE */
8703 	if ((protocol == htons(ETH_P_FCOE)) &&
8704 	    (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
8705 		tso = ixgbe_fso(tx_ring, first, &hdr_len);
8706 		if (tso < 0)
8707 			goto out_drop;
8708 
8709 		goto xmit_fcoe;
8710 	}
8711 
8712 #endif /* IXGBE_FCOE */
8713 
8714 #ifdef CONFIG_IXGBE_IPSEC
8715 	if (xfrm_offload(skb) &&
8716 	    !ixgbe_ipsec_tx(tx_ring, first, &ipsec_tx))
8717 		goto out_drop;
8718 #endif
8719 	tso = ixgbe_tso(tx_ring, first, &hdr_len, &ipsec_tx);
8720 	if (tso < 0)
8721 		goto out_drop;
8722 	else if (!tso)
8723 		ixgbe_tx_csum(tx_ring, first, &ipsec_tx);
8724 
8725 	/* add the ATR filter if ATR is on */
8726 	if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
8727 		ixgbe_atr(tx_ring, first);
8728 
8729 #ifdef IXGBE_FCOE
8730 xmit_fcoe:
8731 #endif /* IXGBE_FCOE */
8732 	if (ixgbe_tx_map(tx_ring, first, hdr_len))
8733 		goto cleanup_tx_timestamp;
8734 
8735 	return NETDEV_TX_OK;
8736 
8737 out_drop:
8738 	dev_kfree_skb_any(first->skb);
8739 	first->skb = NULL;
8740 cleanup_tx_timestamp:
8741 	if (unlikely(tx_flags & IXGBE_TX_FLAGS_TSTAMP)) {
8742 		dev_kfree_skb_any(adapter->ptp_tx_skb);
8743 		adapter->ptp_tx_skb = NULL;
8744 		cancel_work_sync(&adapter->ptp_tx_work);
8745 		clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS, &adapter->state);
8746 	}
8747 
8748 	return NETDEV_TX_OK;
8749 }
8750 
8751 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
8752 				      struct net_device *netdev,
8753 				      struct ixgbe_ring *ring)
8754 {
8755 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8756 	struct ixgbe_ring *tx_ring;
8757 
8758 	/*
8759 	 * The minimum packet size for olinfo paylen is 17 so pad the skb
8760 	 * in order to meet this minimum size requirement.
8761 	 */
8762 	if (skb_put_padto(skb, 17))
8763 		return NETDEV_TX_OK;
8764 
8765 	tx_ring = ring ? ring : adapter->tx_ring[skb_get_queue_mapping(skb)];
8766 	if (unlikely(test_bit(__IXGBE_TX_DISABLED, &tx_ring->state)))
8767 		return NETDEV_TX_BUSY;
8768 
8769 	return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
8770 }
8771 
8772 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
8773 				    struct net_device *netdev)
8774 {
8775 	return __ixgbe_xmit_frame(skb, netdev, NULL);
8776 }
8777 
8778 /**
8779  * ixgbe_set_mac - Change the Ethernet Address of the NIC
8780  * @netdev: network interface device structure
8781  * @p: pointer to an address structure
8782  *
8783  * Returns 0 on success, negative on failure
8784  **/
8785 static int ixgbe_set_mac(struct net_device *netdev, void *p)
8786 {
8787 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8788 	struct ixgbe_hw *hw = &adapter->hw;
8789 	struct sockaddr *addr = p;
8790 
8791 	if (!is_valid_ether_addr(addr->sa_data))
8792 		return -EADDRNOTAVAIL;
8793 
8794 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
8795 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
8796 
8797 	ixgbe_mac_set_default_filter(adapter);
8798 
8799 	return 0;
8800 }
8801 
8802 static int
8803 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
8804 {
8805 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8806 	struct ixgbe_hw *hw = &adapter->hw;
8807 	u16 value;
8808 	int rc;
8809 
8810 	if (adapter->mii_bus) {
8811 		int regnum = addr;
8812 
8813 		if (devad != MDIO_DEVAD_NONE)
8814 			regnum |= (devad << 16) | MII_ADDR_C45;
8815 
8816 		return mdiobus_read(adapter->mii_bus, prtad, regnum);
8817 	}
8818 
8819 	if (prtad != hw->phy.mdio.prtad)
8820 		return -EINVAL;
8821 	rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
8822 	if (!rc)
8823 		rc = value;
8824 	return rc;
8825 }
8826 
8827 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
8828 			    u16 addr, u16 value)
8829 {
8830 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8831 	struct ixgbe_hw *hw = &adapter->hw;
8832 
8833 	if (adapter->mii_bus) {
8834 		int regnum = addr;
8835 
8836 		if (devad != MDIO_DEVAD_NONE)
8837 			regnum |= (devad << 16) | MII_ADDR_C45;
8838 
8839 		return mdiobus_write(adapter->mii_bus, prtad, regnum, value);
8840 	}
8841 
8842 	if (prtad != hw->phy.mdio.prtad)
8843 		return -EINVAL;
8844 	return hw->phy.ops.write_reg(hw, addr, devad, value);
8845 }
8846 
8847 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
8848 {
8849 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8850 
8851 	switch (cmd) {
8852 	case SIOCSHWTSTAMP:
8853 		return ixgbe_ptp_set_ts_config(adapter, req);
8854 	case SIOCGHWTSTAMP:
8855 		return ixgbe_ptp_get_ts_config(adapter, req);
8856 	case SIOCGMIIPHY:
8857 		if (!adapter->hw.phy.ops.read_reg)
8858 			return -EOPNOTSUPP;
8859 		/* fall through */
8860 	default:
8861 		return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
8862 	}
8863 }
8864 
8865 /**
8866  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
8867  * netdev->dev_addrs
8868  * @dev: network interface device structure
8869  *
8870  * Returns non-zero on failure
8871  **/
8872 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
8873 {
8874 	int err = 0;
8875 	struct ixgbe_adapter *adapter = netdev_priv(dev);
8876 	struct ixgbe_hw *hw = &adapter->hw;
8877 
8878 	if (is_valid_ether_addr(hw->mac.san_addr)) {
8879 		rtnl_lock();
8880 		err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
8881 		rtnl_unlock();
8882 
8883 		/* update SAN MAC vmdq pool selection */
8884 		hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
8885 	}
8886 	return err;
8887 }
8888 
8889 /**
8890  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
8891  * netdev->dev_addrs
8892  * @dev: network interface device structure
8893  *
8894  * Returns non-zero on failure
8895  **/
8896 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
8897 {
8898 	int err = 0;
8899 	struct ixgbe_adapter *adapter = netdev_priv(dev);
8900 	struct ixgbe_mac_info *mac = &adapter->hw.mac;
8901 
8902 	if (is_valid_ether_addr(mac->san_addr)) {
8903 		rtnl_lock();
8904 		err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
8905 		rtnl_unlock();
8906 	}
8907 	return err;
8908 }
8909 
8910 static void ixgbe_get_ring_stats64(struct rtnl_link_stats64 *stats,
8911 				   struct ixgbe_ring *ring)
8912 {
8913 	u64 bytes, packets;
8914 	unsigned int start;
8915 
8916 	if (ring) {
8917 		do {
8918 			start = u64_stats_fetch_begin_irq(&ring->syncp);
8919 			packets = ring->stats.packets;
8920 			bytes   = ring->stats.bytes;
8921 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8922 		stats->tx_packets += packets;
8923 		stats->tx_bytes   += bytes;
8924 	}
8925 }
8926 
8927 static void ixgbe_get_stats64(struct net_device *netdev,
8928 			      struct rtnl_link_stats64 *stats)
8929 {
8930 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8931 	int i;
8932 
8933 	rcu_read_lock();
8934 	for (i = 0; i < adapter->num_rx_queues; i++) {
8935 		struct ixgbe_ring *ring = READ_ONCE(adapter->rx_ring[i]);
8936 		u64 bytes, packets;
8937 		unsigned int start;
8938 
8939 		if (ring) {
8940 			do {
8941 				start = u64_stats_fetch_begin_irq(&ring->syncp);
8942 				packets = ring->stats.packets;
8943 				bytes   = ring->stats.bytes;
8944 			} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8945 			stats->rx_packets += packets;
8946 			stats->rx_bytes   += bytes;
8947 		}
8948 	}
8949 
8950 	for (i = 0; i < adapter->num_tx_queues; i++) {
8951 		struct ixgbe_ring *ring = READ_ONCE(adapter->tx_ring[i]);
8952 
8953 		ixgbe_get_ring_stats64(stats, ring);
8954 	}
8955 	for (i = 0; i < adapter->num_xdp_queues; i++) {
8956 		struct ixgbe_ring *ring = READ_ONCE(adapter->xdp_ring[i]);
8957 
8958 		ixgbe_get_ring_stats64(stats, ring);
8959 	}
8960 	rcu_read_unlock();
8961 
8962 	/* following stats updated by ixgbe_watchdog_task() */
8963 	stats->multicast	= netdev->stats.multicast;
8964 	stats->rx_errors	= netdev->stats.rx_errors;
8965 	stats->rx_length_errors	= netdev->stats.rx_length_errors;
8966 	stats->rx_crc_errors	= netdev->stats.rx_crc_errors;
8967 	stats->rx_missed_errors	= netdev->stats.rx_missed_errors;
8968 }
8969 
8970 #ifdef CONFIG_IXGBE_DCB
8971 /**
8972  * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
8973  * @adapter: pointer to ixgbe_adapter
8974  * @tc: number of traffic classes currently enabled
8975  *
8976  * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
8977  * 802.1Q priority maps to a packet buffer that exists.
8978  */
8979 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
8980 {
8981 	struct ixgbe_hw *hw = &adapter->hw;
8982 	u32 reg, rsave;
8983 	int i;
8984 
8985 	/* 82598 have a static priority to TC mapping that can not
8986 	 * be changed so no validation is needed.
8987 	 */
8988 	if (hw->mac.type == ixgbe_mac_82598EB)
8989 		return;
8990 
8991 	reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
8992 	rsave = reg;
8993 
8994 	for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
8995 		u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
8996 
8997 		/* If up2tc is out of bounds default to zero */
8998 		if (up2tc > tc)
8999 			reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
9000 	}
9001 
9002 	if (reg != rsave)
9003 		IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
9004 
9005 	return;
9006 }
9007 
9008 /**
9009  * ixgbe_set_prio_tc_map - Configure netdev prio tc map
9010  * @adapter: Pointer to adapter struct
9011  *
9012  * Populate the netdev user priority to tc map
9013  */
9014 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
9015 {
9016 	struct net_device *dev = adapter->netdev;
9017 	struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
9018 	struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
9019 	u8 prio;
9020 
9021 	for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
9022 		u8 tc = 0;
9023 
9024 		if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
9025 			tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
9026 		else if (ets)
9027 			tc = ets->prio_tc[prio];
9028 
9029 		netdev_set_prio_tc_map(dev, prio, tc);
9030 	}
9031 }
9032 
9033 #endif /* CONFIG_IXGBE_DCB */
9034 static int ixgbe_reassign_macvlan_pool(struct net_device *vdev, void *data)
9035 {
9036 	struct ixgbe_adapter *adapter = data;
9037 	struct ixgbe_fwd_adapter *accel;
9038 	int pool;
9039 
9040 	/* we only care about macvlans... */
9041 	if (!netif_is_macvlan(vdev))
9042 		return 0;
9043 
9044 	/* that have hardware offload enabled... */
9045 	accel = macvlan_accel_priv(vdev);
9046 	if (!accel)
9047 		return 0;
9048 
9049 	/* If we can relocate to a different bit do so */
9050 	pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
9051 	if (pool < adapter->num_rx_pools) {
9052 		set_bit(pool, adapter->fwd_bitmask);
9053 		accel->pool = pool;
9054 		return 0;
9055 	}
9056 
9057 	/* if we cannot find a free pool then disable the offload */
9058 	netdev_err(vdev, "L2FW offload disabled due to lack of queue resources\n");
9059 	macvlan_release_l2fw_offload(vdev);
9060 
9061 	/* unbind the queues and drop the subordinate channel config */
9062 	netdev_unbind_sb_channel(adapter->netdev, vdev);
9063 	netdev_set_sb_channel(vdev, 0);
9064 
9065 	kfree(accel);
9066 
9067 	return 0;
9068 }
9069 
9070 static void ixgbe_defrag_macvlan_pools(struct net_device *dev)
9071 {
9072 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9073 
9074 	/* flush any stale bits out of the fwd bitmask */
9075 	bitmap_clear(adapter->fwd_bitmask, 1, 63);
9076 
9077 	/* walk through upper devices reassigning pools */
9078 	netdev_walk_all_upper_dev_rcu(dev, ixgbe_reassign_macvlan_pool,
9079 				      adapter);
9080 }
9081 
9082 /**
9083  * ixgbe_setup_tc - configure net_device for multiple traffic classes
9084  *
9085  * @dev: net device to configure
9086  * @tc: number of traffic classes to enable
9087  */
9088 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
9089 {
9090 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9091 	struct ixgbe_hw *hw = &adapter->hw;
9092 
9093 	/* Hardware supports up to 8 traffic classes */
9094 	if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
9095 		return -EINVAL;
9096 
9097 	if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
9098 		return -EINVAL;
9099 
9100 	/* Hardware has to reinitialize queues and interrupts to
9101 	 * match packet buffer alignment. Unfortunately, the
9102 	 * hardware is not flexible enough to do this dynamically.
9103 	 */
9104 	if (netif_running(dev))
9105 		ixgbe_close(dev);
9106 	else
9107 		ixgbe_reset(adapter);
9108 
9109 	ixgbe_clear_interrupt_scheme(adapter);
9110 
9111 #ifdef CONFIG_IXGBE_DCB
9112 	if (tc) {
9113 		if (adapter->xdp_prog) {
9114 			e_warn(probe, "DCB is not supported with XDP\n");
9115 
9116 			ixgbe_init_interrupt_scheme(adapter);
9117 			if (netif_running(dev))
9118 				ixgbe_open(dev);
9119 			return -EINVAL;
9120 		}
9121 
9122 		netdev_set_num_tc(dev, tc);
9123 		ixgbe_set_prio_tc_map(adapter);
9124 
9125 		adapter->hw_tcs = tc;
9126 		adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
9127 
9128 		if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
9129 			adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
9130 			adapter->hw.fc.requested_mode = ixgbe_fc_none;
9131 		}
9132 	} else {
9133 		netdev_reset_tc(dev);
9134 
9135 		if (adapter->hw.mac.type == ixgbe_mac_82598EB)
9136 			adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
9137 
9138 		adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
9139 		adapter->hw_tcs = tc;
9140 
9141 		adapter->temp_dcb_cfg.pfc_mode_enable = false;
9142 		adapter->dcb_cfg.pfc_mode_enable = false;
9143 	}
9144 
9145 	ixgbe_validate_rtr(adapter, tc);
9146 
9147 #endif /* CONFIG_IXGBE_DCB */
9148 	ixgbe_init_interrupt_scheme(adapter);
9149 
9150 	ixgbe_defrag_macvlan_pools(dev);
9151 
9152 	if (netif_running(dev))
9153 		return ixgbe_open(dev);
9154 
9155 	return 0;
9156 }
9157 
9158 static int ixgbe_delete_clsu32(struct ixgbe_adapter *adapter,
9159 			       struct tc_cls_u32_offload *cls)
9160 {
9161 	u32 hdl = cls->knode.handle;
9162 	u32 uhtid = TC_U32_USERHTID(cls->knode.handle);
9163 	u32 loc = cls->knode.handle & 0xfffff;
9164 	int err = 0, i, j;
9165 	struct ixgbe_jump_table *jump = NULL;
9166 
9167 	if (loc > IXGBE_MAX_HW_ENTRIES)
9168 		return -EINVAL;
9169 
9170 	if ((uhtid != 0x800) && (uhtid >= IXGBE_MAX_LINK_HANDLE))
9171 		return -EINVAL;
9172 
9173 	/* Clear this filter in the link data it is associated with */
9174 	if (uhtid != 0x800) {
9175 		jump = adapter->jump_tables[uhtid];
9176 		if (!jump)
9177 			return -EINVAL;
9178 		if (!test_bit(loc - 1, jump->child_loc_map))
9179 			return -EINVAL;
9180 		clear_bit(loc - 1, jump->child_loc_map);
9181 	}
9182 
9183 	/* Check if the filter being deleted is a link */
9184 	for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9185 		jump = adapter->jump_tables[i];
9186 		if (jump && jump->link_hdl == hdl) {
9187 			/* Delete filters in the hardware in the child hash
9188 			 * table associated with this link
9189 			 */
9190 			for (j = 0; j < IXGBE_MAX_HW_ENTRIES; j++) {
9191 				if (!test_bit(j, jump->child_loc_map))
9192 					continue;
9193 				spin_lock(&adapter->fdir_perfect_lock);
9194 				err = ixgbe_update_ethtool_fdir_entry(adapter,
9195 								      NULL,
9196 								      j + 1);
9197 				spin_unlock(&adapter->fdir_perfect_lock);
9198 				clear_bit(j, jump->child_loc_map);
9199 			}
9200 			/* Remove resources for this link */
9201 			kfree(jump->input);
9202 			kfree(jump->mask);
9203 			kfree(jump);
9204 			adapter->jump_tables[i] = NULL;
9205 			return err;
9206 		}
9207 	}
9208 
9209 	spin_lock(&adapter->fdir_perfect_lock);
9210 	err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, loc);
9211 	spin_unlock(&adapter->fdir_perfect_lock);
9212 	return err;
9213 }
9214 
9215 static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter *adapter,
9216 					    struct tc_cls_u32_offload *cls)
9217 {
9218 	u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9219 
9220 	if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9221 		return -EINVAL;
9222 
9223 	/* This ixgbe devices do not support hash tables at the moment
9224 	 * so abort when given hash tables.
9225 	 */
9226 	if (cls->hnode.divisor > 0)
9227 		return -EINVAL;
9228 
9229 	set_bit(uhtid - 1, &adapter->tables);
9230 	return 0;
9231 }
9232 
9233 static int ixgbe_configure_clsu32_del_hnode(struct ixgbe_adapter *adapter,
9234 					    struct tc_cls_u32_offload *cls)
9235 {
9236 	u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9237 
9238 	if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9239 		return -EINVAL;
9240 
9241 	clear_bit(uhtid - 1, &adapter->tables);
9242 	return 0;
9243 }
9244 
9245 #ifdef CONFIG_NET_CLS_ACT
9246 struct upper_walk_data {
9247 	struct ixgbe_adapter *adapter;
9248 	u64 action;
9249 	int ifindex;
9250 	u8 queue;
9251 };
9252 
9253 static int get_macvlan_queue(struct net_device *upper, void *_data)
9254 {
9255 	if (netif_is_macvlan(upper)) {
9256 		struct ixgbe_fwd_adapter *vadapter = macvlan_accel_priv(upper);
9257 		struct upper_walk_data *data = _data;
9258 		struct ixgbe_adapter *adapter = data->adapter;
9259 		int ifindex = data->ifindex;
9260 
9261 		if (vadapter && upper->ifindex == ifindex) {
9262 			data->queue = adapter->rx_ring[vadapter->rx_base_queue]->reg_idx;
9263 			data->action = data->queue;
9264 			return 1;
9265 		}
9266 	}
9267 
9268 	return 0;
9269 }
9270 
9271 static int handle_redirect_action(struct ixgbe_adapter *adapter, int ifindex,
9272 				  u8 *queue, u64 *action)
9273 {
9274 	struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
9275 	unsigned int num_vfs = adapter->num_vfs, vf;
9276 	struct upper_walk_data data;
9277 	struct net_device *upper;
9278 
9279 	/* redirect to a SRIOV VF */
9280 	for (vf = 0; vf < num_vfs; ++vf) {
9281 		upper = pci_get_drvdata(adapter->vfinfo[vf].vfdev);
9282 		if (upper->ifindex == ifindex) {
9283 			*queue = vf * __ALIGN_MASK(1, ~vmdq->mask);
9284 			*action = vf + 1;
9285 			*action <<= ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF;
9286 			return 0;
9287 		}
9288 	}
9289 
9290 	/* redirect to a offloaded macvlan netdev */
9291 	data.adapter = adapter;
9292 	data.ifindex = ifindex;
9293 	data.action = 0;
9294 	data.queue = 0;
9295 	if (netdev_walk_all_upper_dev_rcu(adapter->netdev,
9296 					  get_macvlan_queue, &data)) {
9297 		*action = data.action;
9298 		*queue = data.queue;
9299 
9300 		return 0;
9301 	}
9302 
9303 	return -EINVAL;
9304 }
9305 
9306 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9307 			    struct tcf_exts *exts, u64 *action, u8 *queue)
9308 {
9309 	const struct tc_action *a;
9310 	int i;
9311 
9312 	if (!tcf_exts_has_actions(exts))
9313 		return -EINVAL;
9314 
9315 	tcf_exts_for_each_action(i, a, exts) {
9316 		/* Drop action */
9317 		if (is_tcf_gact_shot(a)) {
9318 			*action = IXGBE_FDIR_DROP_QUEUE;
9319 			*queue = IXGBE_FDIR_DROP_QUEUE;
9320 			return 0;
9321 		}
9322 
9323 		/* Redirect to a VF or a offloaded macvlan */
9324 		if (is_tcf_mirred_egress_redirect(a)) {
9325 			struct net_device *dev = tcf_mirred_dev(a);
9326 
9327 			if (!dev)
9328 				return -EINVAL;
9329 			return handle_redirect_action(adapter, dev->ifindex,
9330 						      queue, action);
9331 		}
9332 
9333 		return -EINVAL;
9334 	}
9335 
9336 	return -EINVAL;
9337 }
9338 #else
9339 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9340 			    struct tcf_exts *exts, u64 *action, u8 *queue)
9341 {
9342 	return -EINVAL;
9343 }
9344 #endif /* CONFIG_NET_CLS_ACT */
9345 
9346 static int ixgbe_clsu32_build_input(struct ixgbe_fdir_filter *input,
9347 				    union ixgbe_atr_input *mask,
9348 				    struct tc_cls_u32_offload *cls,
9349 				    struct ixgbe_mat_field *field_ptr,
9350 				    struct ixgbe_nexthdr *nexthdr)
9351 {
9352 	int i, j, off;
9353 	__be32 val, m;
9354 	bool found_entry = false, found_jump_field = false;
9355 
9356 	for (i = 0; i < cls->knode.sel->nkeys; i++) {
9357 		off = cls->knode.sel->keys[i].off;
9358 		val = cls->knode.sel->keys[i].val;
9359 		m = cls->knode.sel->keys[i].mask;
9360 
9361 		for (j = 0; field_ptr[j].val; j++) {
9362 			if (field_ptr[j].off == off) {
9363 				field_ptr[j].val(input, mask, (__force u32)val,
9364 						 (__force u32)m);
9365 				input->filter.formatted.flow_type |=
9366 					field_ptr[j].type;
9367 				found_entry = true;
9368 				break;
9369 			}
9370 		}
9371 		if (nexthdr) {
9372 			if (nexthdr->off == cls->knode.sel->keys[i].off &&
9373 			    nexthdr->val ==
9374 			    (__force u32)cls->knode.sel->keys[i].val &&
9375 			    nexthdr->mask ==
9376 			    (__force u32)cls->knode.sel->keys[i].mask)
9377 				found_jump_field = true;
9378 			else
9379 				continue;
9380 		}
9381 	}
9382 
9383 	if (nexthdr && !found_jump_field)
9384 		return -EINVAL;
9385 
9386 	if (!found_entry)
9387 		return 0;
9388 
9389 	mask->formatted.flow_type = IXGBE_ATR_L4TYPE_IPV6_MASK |
9390 				    IXGBE_ATR_L4TYPE_MASK;
9391 
9392 	if (input->filter.formatted.flow_type == IXGBE_ATR_FLOW_TYPE_IPV4)
9393 		mask->formatted.flow_type &= IXGBE_ATR_L4TYPE_IPV6_MASK;
9394 
9395 	return 0;
9396 }
9397 
9398 static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
9399 				  struct tc_cls_u32_offload *cls)
9400 {
9401 	__be16 protocol = cls->common.protocol;
9402 	u32 loc = cls->knode.handle & 0xfffff;
9403 	struct ixgbe_hw *hw = &adapter->hw;
9404 	struct ixgbe_mat_field *field_ptr;
9405 	struct ixgbe_fdir_filter *input = NULL;
9406 	union ixgbe_atr_input *mask = NULL;
9407 	struct ixgbe_jump_table *jump = NULL;
9408 	int i, err = -EINVAL;
9409 	u8 queue;
9410 	u32 uhtid, link_uhtid;
9411 
9412 	uhtid = TC_U32_USERHTID(cls->knode.handle);
9413 	link_uhtid = TC_U32_USERHTID(cls->knode.link_handle);
9414 
9415 	/* At the moment cls_u32 jumps to network layer and skips past
9416 	 * L2 headers. The canonical method to match L2 frames is to use
9417 	 * negative values. However this is error prone at best but really
9418 	 * just broken because there is no way to "know" what sort of hdr
9419 	 * is in front of the network layer. Fix cls_u32 to support L2
9420 	 * headers when needed.
9421 	 */
9422 	if (protocol != htons(ETH_P_IP))
9423 		return err;
9424 
9425 	if (loc >= ((1024 << adapter->fdir_pballoc) - 2)) {
9426 		e_err(drv, "Location out of range\n");
9427 		return err;
9428 	}
9429 
9430 	/* cls u32 is a graph starting at root node 0x800. The driver tracks
9431 	 * links and also the fields used to advance the parser across each
9432 	 * link (e.g. nexthdr/eat parameters from 'tc'). This way we can map
9433 	 * the u32 graph onto the hardware parse graph denoted in ixgbe_model.h
9434 	 * To add support for new nodes update ixgbe_model.h parse structures
9435 	 * this function _should_ be generic try not to hardcode values here.
9436 	 */
9437 	if (uhtid == 0x800) {
9438 		field_ptr = (adapter->jump_tables[0])->mat;
9439 	} else {
9440 		if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9441 			return err;
9442 		if (!adapter->jump_tables[uhtid])
9443 			return err;
9444 		field_ptr = (adapter->jump_tables[uhtid])->mat;
9445 	}
9446 
9447 	if (!field_ptr)
9448 		return err;
9449 
9450 	/* At this point we know the field_ptr is valid and need to either
9451 	 * build cls_u32 link or attach filter. Because adding a link to
9452 	 * a handle that does not exist is invalid and the same for adding
9453 	 * rules to handles that don't exist.
9454 	 */
9455 
9456 	if (link_uhtid) {
9457 		struct ixgbe_nexthdr *nexthdr = ixgbe_ipv4_jumps;
9458 
9459 		if (link_uhtid >= IXGBE_MAX_LINK_HANDLE)
9460 			return err;
9461 
9462 		if (!test_bit(link_uhtid - 1, &adapter->tables))
9463 			return err;
9464 
9465 		/* Multiple filters as links to the same hash table are not
9466 		 * supported. To add a new filter with the same next header
9467 		 * but different match/jump conditions, create a new hash table
9468 		 * and link to it.
9469 		 */
9470 		if (adapter->jump_tables[link_uhtid] &&
9471 		    (adapter->jump_tables[link_uhtid])->link_hdl) {
9472 			e_err(drv, "Link filter exists for link: %x\n",
9473 			      link_uhtid);
9474 			return err;
9475 		}
9476 
9477 		for (i = 0; nexthdr[i].jump; i++) {
9478 			if (nexthdr[i].o != cls->knode.sel->offoff ||
9479 			    nexthdr[i].s != cls->knode.sel->offshift ||
9480 			    nexthdr[i].m !=
9481 			    (__force u32)cls->knode.sel->offmask)
9482 				return err;
9483 
9484 			jump = kzalloc(sizeof(*jump), GFP_KERNEL);
9485 			if (!jump)
9486 				return -ENOMEM;
9487 			input = kzalloc(sizeof(*input), GFP_KERNEL);
9488 			if (!input) {
9489 				err = -ENOMEM;
9490 				goto free_jump;
9491 			}
9492 			mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9493 			if (!mask) {
9494 				err = -ENOMEM;
9495 				goto free_input;
9496 			}
9497 			jump->input = input;
9498 			jump->mask = mask;
9499 			jump->link_hdl = cls->knode.handle;
9500 
9501 			err = ixgbe_clsu32_build_input(input, mask, cls,
9502 						       field_ptr, &nexthdr[i]);
9503 			if (!err) {
9504 				jump->mat = nexthdr[i].jump;
9505 				adapter->jump_tables[link_uhtid] = jump;
9506 				break;
9507 			} else {
9508 				kfree(mask);
9509 				kfree(input);
9510 				kfree(jump);
9511 			}
9512 		}
9513 		return 0;
9514 	}
9515 
9516 	input = kzalloc(sizeof(*input), GFP_KERNEL);
9517 	if (!input)
9518 		return -ENOMEM;
9519 	mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9520 	if (!mask) {
9521 		err = -ENOMEM;
9522 		goto free_input;
9523 	}
9524 
9525 	if ((uhtid != 0x800) && (adapter->jump_tables[uhtid])) {
9526 		if ((adapter->jump_tables[uhtid])->input)
9527 			memcpy(input, (adapter->jump_tables[uhtid])->input,
9528 			       sizeof(*input));
9529 		if ((adapter->jump_tables[uhtid])->mask)
9530 			memcpy(mask, (adapter->jump_tables[uhtid])->mask,
9531 			       sizeof(*mask));
9532 
9533 		/* Lookup in all child hash tables if this location is already
9534 		 * filled with a filter
9535 		 */
9536 		for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9537 			struct ixgbe_jump_table *link = adapter->jump_tables[i];
9538 
9539 			if (link && (test_bit(loc - 1, link->child_loc_map))) {
9540 				e_err(drv, "Filter exists in location: %x\n",
9541 				      loc);
9542 				err = -EINVAL;
9543 				goto err_out;
9544 			}
9545 		}
9546 	}
9547 	err = ixgbe_clsu32_build_input(input, mask, cls, field_ptr, NULL);
9548 	if (err)
9549 		goto err_out;
9550 
9551 	err = parse_tc_actions(adapter, cls->knode.exts, &input->action,
9552 			       &queue);
9553 	if (err < 0)
9554 		goto err_out;
9555 
9556 	input->sw_idx = loc;
9557 
9558 	spin_lock(&adapter->fdir_perfect_lock);
9559 
9560 	if (hlist_empty(&adapter->fdir_filter_list)) {
9561 		memcpy(&adapter->fdir_mask, mask, sizeof(*mask));
9562 		err = ixgbe_fdir_set_input_mask_82599(hw, mask);
9563 		if (err)
9564 			goto err_out_w_lock;
9565 	} else if (memcmp(&adapter->fdir_mask, mask, sizeof(*mask))) {
9566 		err = -EINVAL;
9567 		goto err_out_w_lock;
9568 	}
9569 
9570 	ixgbe_atr_compute_perfect_hash_82599(&input->filter, mask);
9571 	err = ixgbe_fdir_write_perfect_filter_82599(hw, &input->filter,
9572 						    input->sw_idx, queue);
9573 	if (!err)
9574 		ixgbe_update_ethtool_fdir_entry(adapter, input, input->sw_idx);
9575 	spin_unlock(&adapter->fdir_perfect_lock);
9576 
9577 	if ((uhtid != 0x800) && (adapter->jump_tables[uhtid]))
9578 		set_bit(loc - 1, (adapter->jump_tables[uhtid])->child_loc_map);
9579 
9580 	kfree(mask);
9581 	return err;
9582 err_out_w_lock:
9583 	spin_unlock(&adapter->fdir_perfect_lock);
9584 err_out:
9585 	kfree(mask);
9586 free_input:
9587 	kfree(input);
9588 free_jump:
9589 	kfree(jump);
9590 	return err;
9591 }
9592 
9593 static int ixgbe_setup_tc_cls_u32(struct ixgbe_adapter *adapter,
9594 				  struct tc_cls_u32_offload *cls_u32)
9595 {
9596 	switch (cls_u32->command) {
9597 	case TC_CLSU32_NEW_KNODE:
9598 	case TC_CLSU32_REPLACE_KNODE:
9599 		return ixgbe_configure_clsu32(adapter, cls_u32);
9600 	case TC_CLSU32_DELETE_KNODE:
9601 		return ixgbe_delete_clsu32(adapter, cls_u32);
9602 	case TC_CLSU32_NEW_HNODE:
9603 	case TC_CLSU32_REPLACE_HNODE:
9604 		return ixgbe_configure_clsu32_add_hnode(adapter, cls_u32);
9605 	case TC_CLSU32_DELETE_HNODE:
9606 		return ixgbe_configure_clsu32_del_hnode(adapter, cls_u32);
9607 	default:
9608 		return -EOPNOTSUPP;
9609 	}
9610 }
9611 
9612 static int ixgbe_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
9613 				   void *cb_priv)
9614 {
9615 	struct ixgbe_adapter *adapter = cb_priv;
9616 
9617 	if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
9618 		return -EOPNOTSUPP;
9619 
9620 	switch (type) {
9621 	case TC_SETUP_CLSU32:
9622 		return ixgbe_setup_tc_cls_u32(adapter, type_data);
9623 	default:
9624 		return -EOPNOTSUPP;
9625 	}
9626 }
9627 
9628 static int ixgbe_setup_tc_mqprio(struct net_device *dev,
9629 				 struct tc_mqprio_qopt *mqprio)
9630 {
9631 	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
9632 	return ixgbe_setup_tc(dev, mqprio->num_tc);
9633 }
9634 
9635 static LIST_HEAD(ixgbe_block_cb_list);
9636 
9637 static int __ixgbe_setup_tc(struct net_device *dev, enum tc_setup_type type,
9638 			    void *type_data)
9639 {
9640 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9641 
9642 	switch (type) {
9643 	case TC_SETUP_BLOCK:
9644 		return flow_block_cb_setup_simple(type_data,
9645 						  &ixgbe_block_cb_list,
9646 						  ixgbe_setup_tc_block_cb,
9647 						  adapter, adapter, true);
9648 	case TC_SETUP_QDISC_MQPRIO:
9649 		return ixgbe_setup_tc_mqprio(dev, type_data);
9650 	default:
9651 		return -EOPNOTSUPP;
9652 	}
9653 }
9654 
9655 #ifdef CONFIG_PCI_IOV
9656 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
9657 {
9658 	struct net_device *netdev = adapter->netdev;
9659 
9660 	rtnl_lock();
9661 	ixgbe_setup_tc(netdev, adapter->hw_tcs);
9662 	rtnl_unlock();
9663 }
9664 
9665 #endif
9666 void ixgbe_do_reset(struct net_device *netdev)
9667 {
9668 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
9669 
9670 	if (netif_running(netdev))
9671 		ixgbe_reinit_locked(adapter);
9672 	else
9673 		ixgbe_reset(adapter);
9674 }
9675 
9676 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
9677 					    netdev_features_t features)
9678 {
9679 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
9680 
9681 	/* If Rx checksum is disabled, then RSC/LRO should also be disabled */
9682 	if (!(features & NETIF_F_RXCSUM))
9683 		features &= ~NETIF_F_LRO;
9684 
9685 	/* Turn off LRO if not RSC capable */
9686 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
9687 		features &= ~NETIF_F_LRO;
9688 
9689 	if (adapter->xdp_prog && (features & NETIF_F_LRO)) {
9690 		e_dev_err("LRO is not supported with XDP\n");
9691 		features &= ~NETIF_F_LRO;
9692 	}
9693 
9694 	return features;
9695 }
9696 
9697 static void ixgbe_reset_l2fw_offload(struct ixgbe_adapter *adapter)
9698 {
9699 	int rss = min_t(int, ixgbe_max_rss_indices(adapter),
9700 			num_online_cpus());
9701 
9702 	/* go back to full RSS if we're not running SR-IOV */
9703 	if (!adapter->ring_feature[RING_F_VMDQ].offset)
9704 		adapter->flags &= ~(IXGBE_FLAG_VMDQ_ENABLED |
9705 				    IXGBE_FLAG_SRIOV_ENABLED);
9706 
9707 	adapter->ring_feature[RING_F_RSS].limit = rss;
9708 	adapter->ring_feature[RING_F_VMDQ].limit = 1;
9709 
9710 	ixgbe_setup_tc(adapter->netdev, adapter->hw_tcs);
9711 }
9712 
9713 static int ixgbe_set_features(struct net_device *netdev,
9714 			      netdev_features_t features)
9715 {
9716 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
9717 	netdev_features_t changed = netdev->features ^ features;
9718 	bool need_reset = false;
9719 
9720 	/* Make sure RSC matches LRO, reset if change */
9721 	if (!(features & NETIF_F_LRO)) {
9722 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
9723 			need_reset = true;
9724 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
9725 	} else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
9726 		   !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
9727 		if (adapter->rx_itr_setting == 1 ||
9728 		    adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
9729 			adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
9730 			need_reset = true;
9731 		} else if ((changed ^ features) & NETIF_F_LRO) {
9732 			e_info(probe, "rx-usecs set too low, "
9733 			       "disabling RSC\n");
9734 		}
9735 	}
9736 
9737 	/*
9738 	 * Check if Flow Director n-tuple support or hw_tc support was
9739 	 * enabled or disabled.  If the state changed, we need to reset.
9740 	 */
9741 	if ((features & NETIF_F_NTUPLE) || (features & NETIF_F_HW_TC)) {
9742 		/* turn off ATR, enable perfect filters and reset */
9743 		if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
9744 			need_reset = true;
9745 
9746 		adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
9747 		adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9748 	} else {
9749 		/* turn off perfect filters, enable ATR and reset */
9750 		if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
9751 			need_reset = true;
9752 
9753 		adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9754 
9755 		/* We cannot enable ATR if SR-IOV is enabled */
9756 		if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED ||
9757 		    /* We cannot enable ATR if we have 2 or more tcs */
9758 		    (adapter->hw_tcs > 1) ||
9759 		    /* We cannot enable ATR if RSS is disabled */
9760 		    (adapter->ring_feature[RING_F_RSS].limit <= 1) ||
9761 		    /* A sample rate of 0 indicates ATR disabled */
9762 		    (!adapter->atr_sample_rate))
9763 			; /* do nothing not supported */
9764 		else /* otherwise supported and set the flag */
9765 			adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
9766 	}
9767 
9768 	if (changed & NETIF_F_RXALL)
9769 		need_reset = true;
9770 
9771 	netdev->features = features;
9772 
9773 	if ((adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE)) {
9774 		if (features & NETIF_F_RXCSUM) {
9775 			adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9776 		} else {
9777 			u32 port_mask = IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK;
9778 
9779 			ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9780 		}
9781 	}
9782 
9783 	if ((adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE)) {
9784 		if (features & NETIF_F_RXCSUM) {
9785 			adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9786 		} else {
9787 			u32 port_mask = IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK;
9788 
9789 			ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9790 		}
9791 	}
9792 
9793 	if ((changed & NETIF_F_HW_L2FW_DOFFLOAD) && adapter->num_rx_pools > 1)
9794 		ixgbe_reset_l2fw_offload(adapter);
9795 	else if (need_reset)
9796 		ixgbe_do_reset(netdev);
9797 	else if (changed & (NETIF_F_HW_VLAN_CTAG_RX |
9798 			    NETIF_F_HW_VLAN_CTAG_FILTER))
9799 		ixgbe_set_rx_mode(netdev);
9800 
9801 	return 1;
9802 }
9803 
9804 /**
9805  * ixgbe_add_udp_tunnel_port - Get notifications about adding UDP tunnel ports
9806  * @dev: The port's netdev
9807  * @ti: Tunnel endpoint information
9808  **/
9809 static void ixgbe_add_udp_tunnel_port(struct net_device *dev,
9810 				      struct udp_tunnel_info *ti)
9811 {
9812 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9813 	struct ixgbe_hw *hw = &adapter->hw;
9814 	__be16 port = ti->port;
9815 	u32 port_shift = 0;
9816 	u32 reg;
9817 
9818 	if (ti->sa_family != AF_INET)
9819 		return;
9820 
9821 	switch (ti->type) {
9822 	case UDP_TUNNEL_TYPE_VXLAN:
9823 		if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
9824 			return;
9825 
9826 		if (adapter->vxlan_port == port)
9827 			return;
9828 
9829 		if (adapter->vxlan_port) {
9830 			netdev_info(dev,
9831 				    "VXLAN port %d set, not adding port %d\n",
9832 				    ntohs(adapter->vxlan_port),
9833 				    ntohs(port));
9834 			return;
9835 		}
9836 
9837 		adapter->vxlan_port = port;
9838 		break;
9839 	case UDP_TUNNEL_TYPE_GENEVE:
9840 		if (!(adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE))
9841 			return;
9842 
9843 		if (adapter->geneve_port == port)
9844 			return;
9845 
9846 		if (adapter->geneve_port) {
9847 			netdev_info(dev,
9848 				    "GENEVE port %d set, not adding port %d\n",
9849 				    ntohs(adapter->geneve_port),
9850 				    ntohs(port));
9851 			return;
9852 		}
9853 
9854 		port_shift = IXGBE_VXLANCTRL_GENEVE_UDPPORT_SHIFT;
9855 		adapter->geneve_port = port;
9856 		break;
9857 	default:
9858 		return;
9859 	}
9860 
9861 	reg = IXGBE_READ_REG(hw, IXGBE_VXLANCTRL) | ntohs(port) << port_shift;
9862 	IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, reg);
9863 }
9864 
9865 /**
9866  * ixgbe_del_udp_tunnel_port - Get notifications about removing UDP tunnel ports
9867  * @dev: The port's netdev
9868  * @ti: Tunnel endpoint information
9869  **/
9870 static void ixgbe_del_udp_tunnel_port(struct net_device *dev,
9871 				      struct udp_tunnel_info *ti)
9872 {
9873 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9874 	u32 port_mask;
9875 
9876 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN &&
9877 	    ti->type != UDP_TUNNEL_TYPE_GENEVE)
9878 		return;
9879 
9880 	if (ti->sa_family != AF_INET)
9881 		return;
9882 
9883 	switch (ti->type) {
9884 	case UDP_TUNNEL_TYPE_VXLAN:
9885 		if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
9886 			return;
9887 
9888 		if (adapter->vxlan_port != ti->port) {
9889 			netdev_info(dev, "VXLAN port %d not found\n",
9890 				    ntohs(ti->port));
9891 			return;
9892 		}
9893 
9894 		port_mask = IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK;
9895 		break;
9896 	case UDP_TUNNEL_TYPE_GENEVE:
9897 		if (!(adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE))
9898 			return;
9899 
9900 		if (adapter->geneve_port != ti->port) {
9901 			netdev_info(dev, "GENEVE port %d not found\n",
9902 				    ntohs(ti->port));
9903 			return;
9904 		}
9905 
9906 		port_mask = IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK;
9907 		break;
9908 	default:
9909 		return;
9910 	}
9911 
9912 	ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9913 	adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9914 }
9915 
9916 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
9917 			     struct net_device *dev,
9918 			     const unsigned char *addr, u16 vid,
9919 			     u16 flags,
9920 			     struct netlink_ext_ack *extack)
9921 {
9922 	/* guarantee we can provide a unique filter for the unicast address */
9923 	if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
9924 		struct ixgbe_adapter *adapter = netdev_priv(dev);
9925 		u16 pool = VMDQ_P(0);
9926 
9927 		if (netdev_uc_count(dev) >= ixgbe_available_rars(adapter, pool))
9928 			return -ENOMEM;
9929 	}
9930 
9931 	return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
9932 }
9933 
9934 /**
9935  * ixgbe_configure_bridge_mode - set various bridge modes
9936  * @adapter: the private structure
9937  * @mode: requested bridge mode
9938  *
9939  * Configure some settings require for various bridge modes.
9940  **/
9941 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
9942 				       __u16 mode)
9943 {
9944 	struct ixgbe_hw *hw = &adapter->hw;
9945 	unsigned int p, num_pools;
9946 	u32 vmdctl;
9947 
9948 	switch (mode) {
9949 	case BRIDGE_MODE_VEPA:
9950 		/* disable Tx loopback, rely on switch hairpin mode */
9951 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
9952 
9953 		/* must enable Rx switching replication to allow multicast
9954 		 * packet reception on all VFs, and to enable source address
9955 		 * pruning.
9956 		 */
9957 		vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9958 		vmdctl |= IXGBE_VT_CTL_REPLEN;
9959 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9960 
9961 		/* enable Rx source address pruning. Note, this requires
9962 		 * replication to be enabled or else it does nothing.
9963 		 */
9964 		num_pools = adapter->num_vfs + adapter->num_rx_pools;
9965 		for (p = 0; p < num_pools; p++) {
9966 			if (hw->mac.ops.set_source_address_pruning)
9967 				hw->mac.ops.set_source_address_pruning(hw,
9968 								       true,
9969 								       p);
9970 		}
9971 		break;
9972 	case BRIDGE_MODE_VEB:
9973 		/* enable Tx loopback for internal VF/PF communication */
9974 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
9975 				IXGBE_PFDTXGSWC_VT_LBEN);
9976 
9977 		/* disable Rx switching replication unless we have SR-IOV
9978 		 * virtual functions
9979 		 */
9980 		vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9981 		if (!adapter->num_vfs)
9982 			vmdctl &= ~IXGBE_VT_CTL_REPLEN;
9983 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9984 
9985 		/* disable Rx source address pruning, since we don't expect to
9986 		 * be receiving external loopback of our transmitted frames.
9987 		 */
9988 		num_pools = adapter->num_vfs + adapter->num_rx_pools;
9989 		for (p = 0; p < num_pools; p++) {
9990 			if (hw->mac.ops.set_source_address_pruning)
9991 				hw->mac.ops.set_source_address_pruning(hw,
9992 								       false,
9993 								       p);
9994 		}
9995 		break;
9996 	default:
9997 		return -EINVAL;
9998 	}
9999 
10000 	adapter->bridge_mode = mode;
10001 
10002 	e_info(drv, "enabling bridge mode: %s\n",
10003 	       mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
10004 
10005 	return 0;
10006 }
10007 
10008 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
10009 				    struct nlmsghdr *nlh, u16 flags,
10010 				    struct netlink_ext_ack *extack)
10011 {
10012 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10013 	struct nlattr *attr, *br_spec;
10014 	int rem;
10015 
10016 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
10017 		return -EOPNOTSUPP;
10018 
10019 	br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
10020 	if (!br_spec)
10021 		return -EINVAL;
10022 
10023 	nla_for_each_nested(attr, br_spec, rem) {
10024 		int status;
10025 		__u16 mode;
10026 
10027 		if (nla_type(attr) != IFLA_BRIDGE_MODE)
10028 			continue;
10029 
10030 		if (nla_len(attr) < sizeof(mode))
10031 			return -EINVAL;
10032 
10033 		mode = nla_get_u16(attr);
10034 		status = ixgbe_configure_bridge_mode(adapter, mode);
10035 		if (status)
10036 			return status;
10037 
10038 		break;
10039 	}
10040 
10041 	return 0;
10042 }
10043 
10044 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
10045 				    struct net_device *dev,
10046 				    u32 filter_mask, int nlflags)
10047 {
10048 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10049 
10050 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
10051 		return 0;
10052 
10053 	return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
10054 				       adapter->bridge_mode, 0, 0, nlflags,
10055 				       filter_mask, NULL);
10056 }
10057 
10058 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
10059 {
10060 	struct ixgbe_adapter *adapter = netdev_priv(pdev);
10061 	struct ixgbe_fwd_adapter *accel;
10062 	int tcs = adapter->hw_tcs ? : 1;
10063 	int pool, err;
10064 
10065 	if (adapter->xdp_prog) {
10066 		e_warn(probe, "L2FW offload is not supported with XDP\n");
10067 		return ERR_PTR(-EINVAL);
10068 	}
10069 
10070 	/* The hardware supported by ixgbe only filters on the destination MAC
10071 	 * address. In order to avoid issues we only support offloading modes
10072 	 * where the hardware can actually provide the functionality.
10073 	 */
10074 	if (!macvlan_supports_dest_filter(vdev))
10075 		return ERR_PTR(-EMEDIUMTYPE);
10076 
10077 	/* We need to lock down the macvlan to be a single queue device so that
10078 	 * we can reuse the tc_to_txq field in the macvlan netdev to represent
10079 	 * the queue mapping to our netdev.
10080 	 */
10081 	if (netif_is_multiqueue(vdev))
10082 		return ERR_PTR(-ERANGE);
10083 
10084 	pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
10085 	if (pool == adapter->num_rx_pools) {
10086 		u16 used_pools = adapter->num_vfs + adapter->num_rx_pools;
10087 		u16 reserved_pools;
10088 
10089 		if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
10090 		     adapter->num_rx_pools >= (MAX_TX_QUEUES / tcs)) ||
10091 		    adapter->num_rx_pools > IXGBE_MAX_MACVLANS)
10092 			return ERR_PTR(-EBUSY);
10093 
10094 		/* Hardware has a limited number of available pools. Each VF,
10095 		 * and the PF require a pool. Check to ensure we don't
10096 		 * attempt to use more then the available number of pools.
10097 		 */
10098 		if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
10099 			return ERR_PTR(-EBUSY);
10100 
10101 		/* Enable VMDq flag so device will be set in VM mode */
10102 		adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED |
10103 				  IXGBE_FLAG_SRIOV_ENABLED;
10104 
10105 		/* Try to reserve as many queues per pool as possible,
10106 		 * we start with the configurations that support 4 queues
10107 		 * per pools, followed by 2, and then by just 1 per pool.
10108 		 */
10109 		if (used_pools < 32 && adapter->num_rx_pools < 16)
10110 			reserved_pools = min_t(u16,
10111 					       32 - used_pools,
10112 					       16 - adapter->num_rx_pools);
10113 		else if (adapter->num_rx_pools < 32)
10114 			reserved_pools = min_t(u16,
10115 					       64 - used_pools,
10116 					       32 - adapter->num_rx_pools);
10117 		else
10118 			reserved_pools = 64 - used_pools;
10119 
10120 
10121 		if (!reserved_pools)
10122 			return ERR_PTR(-EBUSY);
10123 
10124 		adapter->ring_feature[RING_F_VMDQ].limit += reserved_pools;
10125 
10126 		/* Force reinit of ring allocation with VMDQ enabled */
10127 		err = ixgbe_setup_tc(pdev, adapter->hw_tcs);
10128 		if (err)
10129 			return ERR_PTR(err);
10130 
10131 		if (pool >= adapter->num_rx_pools)
10132 			return ERR_PTR(-ENOMEM);
10133 	}
10134 
10135 	accel = kzalloc(sizeof(*accel), GFP_KERNEL);
10136 	if (!accel)
10137 		return ERR_PTR(-ENOMEM);
10138 
10139 	set_bit(pool, adapter->fwd_bitmask);
10140 	netdev_set_sb_channel(vdev, pool);
10141 	accel->pool = pool;
10142 	accel->netdev = vdev;
10143 
10144 	if (!netif_running(pdev))
10145 		return accel;
10146 
10147 	err = ixgbe_fwd_ring_up(adapter, accel);
10148 	if (err)
10149 		return ERR_PTR(err);
10150 
10151 	return accel;
10152 }
10153 
10154 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
10155 {
10156 	struct ixgbe_fwd_adapter *accel = priv;
10157 	struct ixgbe_adapter *adapter = netdev_priv(pdev);
10158 	unsigned int rxbase = accel->rx_base_queue;
10159 	unsigned int i;
10160 
10161 	/* delete unicast filter associated with offloaded interface */
10162 	ixgbe_del_mac_filter(adapter, accel->netdev->dev_addr,
10163 			     VMDQ_P(accel->pool));
10164 
10165 	/* Allow remaining Rx packets to get flushed out of the
10166 	 * Rx FIFO before we drop the netdev for the ring.
10167 	 */
10168 	usleep_range(10000, 20000);
10169 
10170 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
10171 		struct ixgbe_ring *ring = adapter->rx_ring[rxbase + i];
10172 		struct ixgbe_q_vector *qv = ring->q_vector;
10173 
10174 		/* Make sure we aren't processing any packets and clear
10175 		 * netdev to shut down the ring.
10176 		 */
10177 		if (netif_running(adapter->netdev))
10178 			napi_synchronize(&qv->napi);
10179 		ring->netdev = NULL;
10180 	}
10181 
10182 	/* unbind the queues and drop the subordinate channel config */
10183 	netdev_unbind_sb_channel(pdev, accel->netdev);
10184 	netdev_set_sb_channel(accel->netdev, 0);
10185 
10186 	clear_bit(accel->pool, adapter->fwd_bitmask);
10187 	kfree(accel);
10188 }
10189 
10190 #define IXGBE_MAX_MAC_HDR_LEN		127
10191 #define IXGBE_MAX_NETWORK_HDR_LEN	511
10192 
10193 static netdev_features_t
10194 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
10195 		     netdev_features_t features)
10196 {
10197 	unsigned int network_hdr_len, mac_hdr_len;
10198 
10199 	/* Make certain the headers can be described by a context descriptor */
10200 	mac_hdr_len = skb_network_header(skb) - skb->data;
10201 	if (unlikely(mac_hdr_len > IXGBE_MAX_MAC_HDR_LEN))
10202 		return features & ~(NETIF_F_HW_CSUM |
10203 				    NETIF_F_SCTP_CRC |
10204 				    NETIF_F_GSO_UDP_L4 |
10205 				    NETIF_F_HW_VLAN_CTAG_TX |
10206 				    NETIF_F_TSO |
10207 				    NETIF_F_TSO6);
10208 
10209 	network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
10210 	if (unlikely(network_hdr_len >  IXGBE_MAX_NETWORK_HDR_LEN))
10211 		return features & ~(NETIF_F_HW_CSUM |
10212 				    NETIF_F_SCTP_CRC |
10213 				    NETIF_F_GSO_UDP_L4 |
10214 				    NETIF_F_TSO |
10215 				    NETIF_F_TSO6);
10216 
10217 	/* We can only support IPV4 TSO in tunnels if we can mangle the
10218 	 * inner IP ID field, so strip TSO if MANGLEID is not supported.
10219 	 * IPsec offoad sets skb->encapsulation but still can handle
10220 	 * the TSO, so it's the exception.
10221 	 */
10222 	if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) {
10223 #ifdef CONFIG_IXGBE_IPSEC
10224 		if (!secpath_exists(skb))
10225 #endif
10226 			features &= ~NETIF_F_TSO;
10227 	}
10228 
10229 	return features;
10230 }
10231 
10232 static int ixgbe_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
10233 {
10234 	int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
10235 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10236 	struct bpf_prog *old_prog;
10237 	bool need_reset;
10238 
10239 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
10240 		return -EINVAL;
10241 
10242 	if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
10243 		return -EINVAL;
10244 
10245 	/* verify ixgbe ring attributes are sufficient for XDP */
10246 	for (i = 0; i < adapter->num_rx_queues; i++) {
10247 		struct ixgbe_ring *ring = adapter->rx_ring[i];
10248 
10249 		if (ring_is_rsc_enabled(ring))
10250 			return -EINVAL;
10251 
10252 		if (frame_size > ixgbe_rx_bufsz(ring))
10253 			return -EINVAL;
10254 	}
10255 
10256 	if (nr_cpu_ids > MAX_XDP_QUEUES)
10257 		return -ENOMEM;
10258 
10259 	old_prog = xchg(&adapter->xdp_prog, prog);
10260 	need_reset = (!!prog != !!old_prog);
10261 
10262 	/* If transitioning XDP modes reconfigure rings */
10263 	if (need_reset) {
10264 		int err = ixgbe_setup_tc(dev, adapter->hw_tcs);
10265 
10266 		if (err) {
10267 			rcu_assign_pointer(adapter->xdp_prog, old_prog);
10268 			return -EINVAL;
10269 		}
10270 	} else {
10271 		for (i = 0; i < adapter->num_rx_queues; i++)
10272 			(void)xchg(&adapter->rx_ring[i]->xdp_prog,
10273 			    adapter->xdp_prog);
10274 	}
10275 
10276 	if (old_prog)
10277 		bpf_prog_put(old_prog);
10278 
10279 	/* Kick start the NAPI context if there is an AF_XDP socket open
10280 	 * on that queue id. This so that receiving will start.
10281 	 */
10282 	if (need_reset && prog)
10283 		for (i = 0; i < adapter->num_rx_queues; i++)
10284 			if (adapter->xdp_ring[i]->xsk_umem)
10285 				(void)ixgbe_xsk_wakeup(adapter->netdev, i,
10286 						       XDP_WAKEUP_RX);
10287 
10288 	return 0;
10289 }
10290 
10291 static int ixgbe_xdp(struct net_device *dev, struct netdev_bpf *xdp)
10292 {
10293 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10294 
10295 	switch (xdp->command) {
10296 	case XDP_SETUP_PROG:
10297 		return ixgbe_xdp_setup(dev, xdp->prog);
10298 	case XDP_QUERY_PROG:
10299 		xdp->prog_id = adapter->xdp_prog ?
10300 			adapter->xdp_prog->aux->id : 0;
10301 		return 0;
10302 	case XDP_SETUP_XSK_UMEM:
10303 		return ixgbe_xsk_umem_setup(adapter, xdp->xsk.umem,
10304 					    xdp->xsk.queue_id);
10305 
10306 	default:
10307 		return -EINVAL;
10308 	}
10309 }
10310 
10311 void ixgbe_xdp_ring_update_tail(struct ixgbe_ring *ring)
10312 {
10313 	/* Force memory writes to complete before letting h/w know there
10314 	 * are new descriptors to fetch.
10315 	 */
10316 	wmb();
10317 	writel(ring->next_to_use, ring->tail);
10318 }
10319 
10320 static int ixgbe_xdp_xmit(struct net_device *dev, int n,
10321 			  struct xdp_frame **frames, u32 flags)
10322 {
10323 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10324 	struct ixgbe_ring *ring;
10325 	int drops = 0;
10326 	int i;
10327 
10328 	if (unlikely(test_bit(__IXGBE_DOWN, &adapter->state)))
10329 		return -ENETDOWN;
10330 
10331 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
10332 		return -EINVAL;
10333 
10334 	/* During program transitions its possible adapter->xdp_prog is assigned
10335 	 * but ring has not been configured yet. In this case simply abort xmit.
10336 	 */
10337 	ring = adapter->xdp_prog ? adapter->xdp_ring[smp_processor_id()] : NULL;
10338 	if (unlikely(!ring))
10339 		return -ENXIO;
10340 
10341 	if (unlikely(test_bit(__IXGBE_TX_DISABLED, &ring->state)))
10342 		return -ENXIO;
10343 
10344 	for (i = 0; i < n; i++) {
10345 		struct xdp_frame *xdpf = frames[i];
10346 		int err;
10347 
10348 		err = ixgbe_xmit_xdp_ring(adapter, xdpf);
10349 		if (err != IXGBE_XDP_TX) {
10350 			xdp_return_frame_rx_napi(xdpf);
10351 			drops++;
10352 		}
10353 	}
10354 
10355 	if (unlikely(flags & XDP_XMIT_FLUSH))
10356 		ixgbe_xdp_ring_update_tail(ring);
10357 
10358 	return n - drops;
10359 }
10360 
10361 static const struct net_device_ops ixgbe_netdev_ops = {
10362 	.ndo_open		= ixgbe_open,
10363 	.ndo_stop		= ixgbe_close,
10364 	.ndo_start_xmit		= ixgbe_xmit_frame,
10365 	.ndo_set_rx_mode	= ixgbe_set_rx_mode,
10366 	.ndo_validate_addr	= eth_validate_addr,
10367 	.ndo_set_mac_address	= ixgbe_set_mac,
10368 	.ndo_change_mtu		= ixgbe_change_mtu,
10369 	.ndo_tx_timeout		= ixgbe_tx_timeout,
10370 	.ndo_set_tx_maxrate	= ixgbe_tx_maxrate,
10371 	.ndo_vlan_rx_add_vid	= ixgbe_vlan_rx_add_vid,
10372 	.ndo_vlan_rx_kill_vid	= ixgbe_vlan_rx_kill_vid,
10373 	.ndo_do_ioctl		= ixgbe_ioctl,
10374 	.ndo_set_vf_mac		= ixgbe_ndo_set_vf_mac,
10375 	.ndo_set_vf_vlan	= ixgbe_ndo_set_vf_vlan,
10376 	.ndo_set_vf_rate	= ixgbe_ndo_set_vf_bw,
10377 	.ndo_set_vf_spoofchk	= ixgbe_ndo_set_vf_spoofchk,
10378 	.ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
10379 	.ndo_set_vf_trust	= ixgbe_ndo_set_vf_trust,
10380 	.ndo_get_vf_config	= ixgbe_ndo_get_vf_config,
10381 	.ndo_get_stats64	= ixgbe_get_stats64,
10382 	.ndo_setup_tc		= __ixgbe_setup_tc,
10383 #ifdef IXGBE_FCOE
10384 	.ndo_select_queue	= ixgbe_select_queue,
10385 	.ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
10386 	.ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
10387 	.ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
10388 	.ndo_fcoe_enable = ixgbe_fcoe_enable,
10389 	.ndo_fcoe_disable = ixgbe_fcoe_disable,
10390 	.ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
10391 	.ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
10392 #endif /* IXGBE_FCOE */
10393 	.ndo_set_features = ixgbe_set_features,
10394 	.ndo_fix_features = ixgbe_fix_features,
10395 	.ndo_fdb_add		= ixgbe_ndo_fdb_add,
10396 	.ndo_bridge_setlink	= ixgbe_ndo_bridge_setlink,
10397 	.ndo_bridge_getlink	= ixgbe_ndo_bridge_getlink,
10398 	.ndo_dfwd_add_station	= ixgbe_fwd_add,
10399 	.ndo_dfwd_del_station	= ixgbe_fwd_del,
10400 	.ndo_udp_tunnel_add	= ixgbe_add_udp_tunnel_port,
10401 	.ndo_udp_tunnel_del	= ixgbe_del_udp_tunnel_port,
10402 	.ndo_features_check	= ixgbe_features_check,
10403 	.ndo_bpf		= ixgbe_xdp,
10404 	.ndo_xdp_xmit		= ixgbe_xdp_xmit,
10405 	.ndo_xsk_wakeup         = ixgbe_xsk_wakeup,
10406 };
10407 
10408 static void ixgbe_disable_txr_hw(struct ixgbe_adapter *adapter,
10409 				 struct ixgbe_ring *tx_ring)
10410 {
10411 	unsigned long wait_delay, delay_interval;
10412 	struct ixgbe_hw *hw = &adapter->hw;
10413 	u8 reg_idx = tx_ring->reg_idx;
10414 	int wait_loop;
10415 	u32 txdctl;
10416 
10417 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
10418 
10419 	/* delay mechanism from ixgbe_disable_tx */
10420 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10421 
10422 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
10423 	wait_delay = delay_interval;
10424 
10425 	while (wait_loop--) {
10426 		usleep_range(wait_delay, wait_delay + 10);
10427 		wait_delay += delay_interval * 2;
10428 		txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
10429 
10430 		if (!(txdctl & IXGBE_TXDCTL_ENABLE))
10431 			return;
10432 	}
10433 
10434 	e_err(drv, "TXDCTL.ENABLE not cleared within the polling period\n");
10435 }
10436 
10437 static void ixgbe_disable_txr(struct ixgbe_adapter *adapter,
10438 			      struct ixgbe_ring *tx_ring)
10439 {
10440 	set_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10441 	ixgbe_disable_txr_hw(adapter, tx_ring);
10442 }
10443 
10444 static void ixgbe_disable_rxr_hw(struct ixgbe_adapter *adapter,
10445 				 struct ixgbe_ring *rx_ring)
10446 {
10447 	unsigned long wait_delay, delay_interval;
10448 	struct ixgbe_hw *hw = &adapter->hw;
10449 	u8 reg_idx = rx_ring->reg_idx;
10450 	int wait_loop;
10451 	u32 rxdctl;
10452 
10453 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10454 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
10455 	rxdctl |= IXGBE_RXDCTL_SWFLSH;
10456 
10457 	/* write value back with RXDCTL.ENABLE bit cleared */
10458 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
10459 
10460 	/* RXDCTL.EN may not change on 82598 if link is down, so skip it */
10461 	if (hw->mac.type == ixgbe_mac_82598EB &&
10462 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
10463 		return;
10464 
10465 	/* delay mechanism from ixgbe_disable_rx */
10466 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10467 
10468 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
10469 	wait_delay = delay_interval;
10470 
10471 	while (wait_loop--) {
10472 		usleep_range(wait_delay, wait_delay + 10);
10473 		wait_delay += delay_interval * 2;
10474 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10475 
10476 		if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
10477 			return;
10478 	}
10479 
10480 	e_err(drv, "RXDCTL.ENABLE not cleared within the polling period\n");
10481 }
10482 
10483 static void ixgbe_reset_txr_stats(struct ixgbe_ring *tx_ring)
10484 {
10485 	memset(&tx_ring->stats, 0, sizeof(tx_ring->stats));
10486 	memset(&tx_ring->tx_stats, 0, sizeof(tx_ring->tx_stats));
10487 }
10488 
10489 static void ixgbe_reset_rxr_stats(struct ixgbe_ring *rx_ring)
10490 {
10491 	memset(&rx_ring->stats, 0, sizeof(rx_ring->stats));
10492 	memset(&rx_ring->rx_stats, 0, sizeof(rx_ring->rx_stats));
10493 }
10494 
10495 /**
10496  * ixgbe_txrx_ring_disable - Disable Rx/Tx/XDP Tx rings
10497  * @adapter: adapter structure
10498  * @ring: ring index
10499  *
10500  * This function disables a certain Rx/Tx/XDP Tx ring. The function
10501  * assumes that the netdev is running.
10502  **/
10503 void ixgbe_txrx_ring_disable(struct ixgbe_adapter *adapter, int ring)
10504 {
10505 	struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10506 
10507 	rx_ring = adapter->rx_ring[ring];
10508 	tx_ring = adapter->tx_ring[ring];
10509 	xdp_ring = adapter->xdp_ring[ring];
10510 
10511 	ixgbe_disable_txr(adapter, tx_ring);
10512 	if (xdp_ring)
10513 		ixgbe_disable_txr(adapter, xdp_ring);
10514 	ixgbe_disable_rxr_hw(adapter, rx_ring);
10515 
10516 	if (xdp_ring)
10517 		synchronize_rcu();
10518 
10519 	/* Rx/Tx/XDP Tx share the same napi context. */
10520 	napi_disable(&rx_ring->q_vector->napi);
10521 
10522 	ixgbe_clean_tx_ring(tx_ring);
10523 	if (xdp_ring)
10524 		ixgbe_clean_tx_ring(xdp_ring);
10525 	ixgbe_clean_rx_ring(rx_ring);
10526 
10527 	ixgbe_reset_txr_stats(tx_ring);
10528 	if (xdp_ring)
10529 		ixgbe_reset_txr_stats(xdp_ring);
10530 	ixgbe_reset_rxr_stats(rx_ring);
10531 }
10532 
10533 /**
10534  * ixgbe_txrx_ring_enable - Enable Rx/Tx/XDP Tx rings
10535  * @adapter: adapter structure
10536  * @ring: ring index
10537  *
10538  * This function enables a certain Rx/Tx/XDP Tx ring. The function
10539  * assumes that the netdev is running.
10540  **/
10541 void ixgbe_txrx_ring_enable(struct ixgbe_adapter *adapter, int ring)
10542 {
10543 	struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10544 
10545 	rx_ring = adapter->rx_ring[ring];
10546 	tx_ring = adapter->tx_ring[ring];
10547 	xdp_ring = adapter->xdp_ring[ring];
10548 
10549 	/* Rx/Tx/XDP Tx share the same napi context. */
10550 	napi_enable(&rx_ring->q_vector->napi);
10551 
10552 	ixgbe_configure_tx_ring(adapter, tx_ring);
10553 	if (xdp_ring)
10554 		ixgbe_configure_tx_ring(adapter, xdp_ring);
10555 	ixgbe_configure_rx_ring(adapter, rx_ring);
10556 
10557 	clear_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10558 	if (xdp_ring)
10559 		clear_bit(__IXGBE_TX_DISABLED, &xdp_ring->state);
10560 }
10561 
10562 /**
10563  * ixgbe_enumerate_functions - Get the number of ports this device has
10564  * @adapter: adapter structure
10565  *
10566  * This function enumerates the phsyical functions co-located on a single slot,
10567  * in order to determine how many ports a device has. This is most useful in
10568  * determining the required GT/s of PCIe bandwidth necessary for optimal
10569  * performance.
10570  **/
10571 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
10572 {
10573 	struct pci_dev *entry, *pdev = adapter->pdev;
10574 	int physfns = 0;
10575 
10576 	/* Some cards can not use the generic count PCIe functions method,
10577 	 * because they are behind a parent switch, so we hardcode these with
10578 	 * the correct number of functions.
10579 	 */
10580 	if (ixgbe_pcie_from_parent(&adapter->hw))
10581 		physfns = 4;
10582 
10583 	list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
10584 		/* don't count virtual functions */
10585 		if (entry->is_virtfn)
10586 			continue;
10587 
10588 		/* When the devices on the bus don't all match our device ID,
10589 		 * we can't reliably determine the correct number of
10590 		 * functions. This can occur if a function has been direct
10591 		 * attached to a virtual machine using VT-d, for example. In
10592 		 * this case, simply return -1 to indicate this.
10593 		 */
10594 		if ((entry->vendor != pdev->vendor) ||
10595 		    (entry->device != pdev->device))
10596 			return -1;
10597 
10598 		physfns++;
10599 	}
10600 
10601 	return physfns;
10602 }
10603 
10604 /**
10605  * ixgbe_wol_supported - Check whether device supports WoL
10606  * @adapter: the adapter private structure
10607  * @device_id: the device ID
10608  * @subdevice_id: the subsystem device ID
10609  *
10610  * This function is used by probe and ethtool to determine
10611  * which devices have WoL support
10612  *
10613  **/
10614 bool ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
10615 			 u16 subdevice_id)
10616 {
10617 	struct ixgbe_hw *hw = &adapter->hw;
10618 	u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
10619 
10620 	/* WOL not supported on 82598 */
10621 	if (hw->mac.type == ixgbe_mac_82598EB)
10622 		return false;
10623 
10624 	/* check eeprom to see if WOL is enabled for X540 and newer */
10625 	if (hw->mac.type >= ixgbe_mac_X540) {
10626 		if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
10627 		    ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
10628 		     (hw->bus.func == 0)))
10629 			return true;
10630 	}
10631 
10632 	/* WOL is determined based on device IDs for 82599 MACs */
10633 	switch (device_id) {
10634 	case IXGBE_DEV_ID_82599_SFP:
10635 		/* Only these subdevices could supports WOL */
10636 		switch (subdevice_id) {
10637 		case IXGBE_SUBDEV_ID_82599_560FLR:
10638 		case IXGBE_SUBDEV_ID_82599_LOM_SNAP6:
10639 		case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
10640 		case IXGBE_SUBDEV_ID_82599_SFP_2OCP:
10641 			/* only support first port */
10642 			if (hw->bus.func != 0)
10643 				break;
10644 			/* fall through */
10645 		case IXGBE_SUBDEV_ID_82599_SP_560FLR:
10646 		case IXGBE_SUBDEV_ID_82599_SFP:
10647 		case IXGBE_SUBDEV_ID_82599_RNDC:
10648 		case IXGBE_SUBDEV_ID_82599_ECNA_DP:
10649 		case IXGBE_SUBDEV_ID_82599_SFP_1OCP:
10650 		case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM1:
10651 		case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM2:
10652 			return true;
10653 		}
10654 		break;
10655 	case IXGBE_DEV_ID_82599EN_SFP:
10656 		/* Only these subdevices support WOL */
10657 		switch (subdevice_id) {
10658 		case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
10659 			return true;
10660 		}
10661 		break;
10662 	case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
10663 		/* All except this subdevice support WOL */
10664 		if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
10665 			return true;
10666 		break;
10667 	case IXGBE_DEV_ID_82599_KX4:
10668 		return  true;
10669 	default:
10670 		break;
10671 	}
10672 
10673 	return false;
10674 }
10675 
10676 /**
10677  * ixgbe_set_fw_version - Set FW version
10678  * @adapter: the adapter private structure
10679  *
10680  * This function is used by probe and ethtool to determine the FW version to
10681  * format to display. The FW version is taken from the EEPROM/NVM.
10682  */
10683 static void ixgbe_set_fw_version(struct ixgbe_adapter *adapter)
10684 {
10685 	struct ixgbe_hw *hw = &adapter->hw;
10686 	struct ixgbe_nvm_version nvm_ver;
10687 
10688 	ixgbe_get_oem_prod_version(hw, &nvm_ver);
10689 	if (nvm_ver.oem_valid) {
10690 		snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10691 			 "%x.%x.%x", nvm_ver.oem_major, nvm_ver.oem_minor,
10692 			 nvm_ver.oem_release);
10693 		return;
10694 	}
10695 
10696 	ixgbe_get_etk_id(hw, &nvm_ver);
10697 	ixgbe_get_orom_version(hw, &nvm_ver);
10698 
10699 	if (nvm_ver.or_valid) {
10700 		snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10701 			 "0x%08x, %d.%d.%d", nvm_ver.etk_id, nvm_ver.or_major,
10702 			 nvm_ver.or_build, nvm_ver.or_patch);
10703 		return;
10704 	}
10705 
10706 	/* Set ETrack ID format */
10707 	snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10708 		 "0x%08x", nvm_ver.etk_id);
10709 }
10710 
10711 /**
10712  * ixgbe_probe - Device Initialization Routine
10713  * @pdev: PCI device information struct
10714  * @ent: entry in ixgbe_pci_tbl
10715  *
10716  * Returns 0 on success, negative on failure
10717  *
10718  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
10719  * The OS initialization, configuring of the adapter private structure,
10720  * and a hardware reset occur.
10721  **/
10722 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
10723 {
10724 	struct net_device *netdev;
10725 	struct ixgbe_adapter *adapter = NULL;
10726 	struct ixgbe_hw *hw;
10727 	const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
10728 	int i, err, pci_using_dac, expected_gts;
10729 	unsigned int indices = MAX_TX_QUEUES;
10730 	u8 part_str[IXGBE_PBANUM_LENGTH];
10731 	bool disable_dev = false;
10732 #ifdef IXGBE_FCOE
10733 	u16 device_caps;
10734 #endif
10735 	u32 eec;
10736 
10737 	/* Catch broken hardware that put the wrong VF device ID in
10738 	 * the PCIe SR-IOV capability.
10739 	 */
10740 	if (pdev->is_virtfn) {
10741 		WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
10742 		     pci_name(pdev), pdev->vendor, pdev->device);
10743 		return -EINVAL;
10744 	}
10745 
10746 	err = pci_enable_device_mem(pdev);
10747 	if (err)
10748 		return err;
10749 
10750 	if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
10751 		pci_using_dac = 1;
10752 	} else {
10753 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
10754 		if (err) {
10755 			dev_err(&pdev->dev,
10756 				"No usable DMA configuration, aborting\n");
10757 			goto err_dma;
10758 		}
10759 		pci_using_dac = 0;
10760 	}
10761 
10762 	err = pci_request_mem_regions(pdev, ixgbe_driver_name);
10763 	if (err) {
10764 		dev_err(&pdev->dev,
10765 			"pci_request_selected_regions failed 0x%x\n", err);
10766 		goto err_pci_reg;
10767 	}
10768 
10769 	pci_enable_pcie_error_reporting(pdev);
10770 
10771 	pci_set_master(pdev);
10772 	pci_save_state(pdev);
10773 
10774 	if (ii->mac == ixgbe_mac_82598EB) {
10775 #ifdef CONFIG_IXGBE_DCB
10776 		/* 8 TC w/ 4 queues per TC */
10777 		indices = 4 * MAX_TRAFFIC_CLASS;
10778 #else
10779 		indices = IXGBE_MAX_RSS_INDICES;
10780 #endif
10781 	}
10782 
10783 	netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
10784 	if (!netdev) {
10785 		err = -ENOMEM;
10786 		goto err_alloc_etherdev;
10787 	}
10788 
10789 	SET_NETDEV_DEV(netdev, &pdev->dev);
10790 
10791 	adapter = netdev_priv(netdev);
10792 
10793 	adapter->netdev = netdev;
10794 	adapter->pdev = pdev;
10795 	hw = &adapter->hw;
10796 	hw->back = adapter;
10797 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
10798 
10799 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
10800 			      pci_resource_len(pdev, 0));
10801 	adapter->io_addr = hw->hw_addr;
10802 	if (!hw->hw_addr) {
10803 		err = -EIO;
10804 		goto err_ioremap;
10805 	}
10806 
10807 	netdev->netdev_ops = &ixgbe_netdev_ops;
10808 	ixgbe_set_ethtool_ops(netdev);
10809 	netdev->watchdog_timeo = 5 * HZ;
10810 	strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
10811 
10812 	/* Setup hw api */
10813 	hw->mac.ops   = *ii->mac_ops;
10814 	hw->mac.type  = ii->mac;
10815 	hw->mvals     = ii->mvals;
10816 	if (ii->link_ops)
10817 		hw->link.ops  = *ii->link_ops;
10818 
10819 	/* EEPROM */
10820 	hw->eeprom.ops = *ii->eeprom_ops;
10821 	eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
10822 	if (ixgbe_removed(hw->hw_addr)) {
10823 		err = -EIO;
10824 		goto err_ioremap;
10825 	}
10826 	/* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
10827 	if (!(eec & BIT(8)))
10828 		hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
10829 
10830 	/* PHY */
10831 	hw->phy.ops = *ii->phy_ops;
10832 	hw->phy.sfp_type = ixgbe_sfp_type_unknown;
10833 	/* ixgbe_identify_phy_generic will set prtad and mmds properly */
10834 	hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
10835 	hw->phy.mdio.mmds = 0;
10836 	hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
10837 	hw->phy.mdio.dev = netdev;
10838 	hw->phy.mdio.mdio_read = ixgbe_mdio_read;
10839 	hw->phy.mdio.mdio_write = ixgbe_mdio_write;
10840 
10841 	/* setup the private structure */
10842 	err = ixgbe_sw_init(adapter, ii);
10843 	if (err)
10844 		goto err_sw_init;
10845 
10846 	/* Make sure the SWFW semaphore is in a valid state */
10847 	if (hw->mac.ops.init_swfw_sync)
10848 		hw->mac.ops.init_swfw_sync(hw);
10849 
10850 	/* Make it possible the adapter to be woken up via WOL */
10851 	switch (adapter->hw.mac.type) {
10852 	case ixgbe_mac_82599EB:
10853 	case ixgbe_mac_X540:
10854 	case ixgbe_mac_X550:
10855 	case ixgbe_mac_X550EM_x:
10856 	case ixgbe_mac_x550em_a:
10857 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
10858 		break;
10859 	default:
10860 		break;
10861 	}
10862 
10863 	/*
10864 	 * If there is a fan on this device and it has failed log the
10865 	 * failure.
10866 	 */
10867 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
10868 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
10869 		if (esdp & IXGBE_ESDP_SDP1)
10870 			e_crit(probe, "Fan has stopped, replace the adapter\n");
10871 	}
10872 
10873 	if (allow_unsupported_sfp)
10874 		hw->allow_unsupported_sfp = allow_unsupported_sfp;
10875 
10876 	/* reset_hw fills in the perm_addr as well */
10877 	hw->phy.reset_if_overtemp = true;
10878 	err = hw->mac.ops.reset_hw(hw);
10879 	hw->phy.reset_if_overtemp = false;
10880 	ixgbe_set_eee_capable(adapter);
10881 	if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
10882 		err = 0;
10883 	} else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
10884 		e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
10885 		e_dev_err("Reload the driver after installing a supported module.\n");
10886 		goto err_sw_init;
10887 	} else if (err) {
10888 		e_dev_err("HW Init failed: %d\n", err);
10889 		goto err_sw_init;
10890 	}
10891 
10892 #ifdef CONFIG_PCI_IOV
10893 	/* SR-IOV not supported on the 82598 */
10894 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
10895 		goto skip_sriov;
10896 	/* Mailbox */
10897 	ixgbe_init_mbx_params_pf(hw);
10898 	hw->mbx.ops = ii->mbx_ops;
10899 	pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
10900 	ixgbe_enable_sriov(adapter, max_vfs);
10901 skip_sriov:
10902 
10903 #endif
10904 	netdev->features = NETIF_F_SG |
10905 			   NETIF_F_TSO |
10906 			   NETIF_F_TSO6 |
10907 			   NETIF_F_RXHASH |
10908 			   NETIF_F_RXCSUM |
10909 			   NETIF_F_HW_CSUM;
10910 
10911 #define IXGBE_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
10912 				    NETIF_F_GSO_GRE_CSUM | \
10913 				    NETIF_F_GSO_IPXIP4 | \
10914 				    NETIF_F_GSO_IPXIP6 | \
10915 				    NETIF_F_GSO_UDP_TUNNEL | \
10916 				    NETIF_F_GSO_UDP_TUNNEL_CSUM)
10917 
10918 	netdev->gso_partial_features = IXGBE_GSO_PARTIAL_FEATURES;
10919 	netdev->features |= NETIF_F_GSO_PARTIAL |
10920 			    IXGBE_GSO_PARTIAL_FEATURES;
10921 
10922 	if (hw->mac.type >= ixgbe_mac_82599EB)
10923 		netdev->features |= NETIF_F_SCTP_CRC | NETIF_F_GSO_UDP_L4;
10924 
10925 #ifdef CONFIG_IXGBE_IPSEC
10926 #define IXGBE_ESP_FEATURES	(NETIF_F_HW_ESP | \
10927 				 NETIF_F_HW_ESP_TX_CSUM | \
10928 				 NETIF_F_GSO_ESP)
10929 
10930 	if (adapter->ipsec)
10931 		netdev->features |= IXGBE_ESP_FEATURES;
10932 #endif
10933 	/* copy netdev features into list of user selectable features */
10934 	netdev->hw_features |= netdev->features |
10935 			       NETIF_F_HW_VLAN_CTAG_FILTER |
10936 			       NETIF_F_HW_VLAN_CTAG_RX |
10937 			       NETIF_F_HW_VLAN_CTAG_TX |
10938 			       NETIF_F_RXALL |
10939 			       NETIF_F_HW_L2FW_DOFFLOAD;
10940 
10941 	if (hw->mac.type >= ixgbe_mac_82599EB)
10942 		netdev->hw_features |= NETIF_F_NTUPLE |
10943 				       NETIF_F_HW_TC;
10944 
10945 	if (pci_using_dac)
10946 		netdev->features |= NETIF_F_HIGHDMA;
10947 
10948 	netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
10949 	netdev->hw_enc_features |= netdev->vlan_features;
10950 	netdev->mpls_features |= NETIF_F_SG |
10951 				 NETIF_F_TSO |
10952 				 NETIF_F_TSO6 |
10953 				 NETIF_F_HW_CSUM;
10954 	netdev->mpls_features |= IXGBE_GSO_PARTIAL_FEATURES;
10955 
10956 	/* set this bit last since it cannot be part of vlan_features */
10957 	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
10958 			    NETIF_F_HW_VLAN_CTAG_RX |
10959 			    NETIF_F_HW_VLAN_CTAG_TX;
10960 
10961 	netdev->priv_flags |= IFF_UNICAST_FLT;
10962 	netdev->priv_flags |= IFF_SUPP_NOFCS;
10963 
10964 	/* MTU range: 68 - 9710 */
10965 	netdev->min_mtu = ETH_MIN_MTU;
10966 	netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE - (ETH_HLEN + ETH_FCS_LEN);
10967 
10968 #ifdef CONFIG_IXGBE_DCB
10969 	if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
10970 		netdev->dcbnl_ops = &ixgbe_dcbnl_ops;
10971 #endif
10972 
10973 #ifdef IXGBE_FCOE
10974 	if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
10975 		unsigned int fcoe_l;
10976 
10977 		if (hw->mac.ops.get_device_caps) {
10978 			hw->mac.ops.get_device_caps(hw, &device_caps);
10979 			if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
10980 				adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
10981 		}
10982 
10983 
10984 		fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
10985 		adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
10986 
10987 		netdev->features |= NETIF_F_FSO |
10988 				    NETIF_F_FCOE_CRC;
10989 
10990 		netdev->vlan_features |= NETIF_F_FSO |
10991 					 NETIF_F_FCOE_CRC |
10992 					 NETIF_F_FCOE_MTU;
10993 	}
10994 #endif /* IXGBE_FCOE */
10995 	if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
10996 		netdev->hw_features |= NETIF_F_LRO;
10997 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
10998 		netdev->features |= NETIF_F_LRO;
10999 
11000 	if (ixgbe_check_fw_error(adapter)) {
11001 		err = -EIO;
11002 		goto err_sw_init;
11003 	}
11004 
11005 	/* make sure the EEPROM is good */
11006 	if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
11007 		e_dev_err("The EEPROM Checksum Is Not Valid\n");
11008 		err = -EIO;
11009 		goto err_sw_init;
11010 	}
11011 
11012 	eth_platform_get_mac_address(&adapter->pdev->dev,
11013 				     adapter->hw.mac.perm_addr);
11014 
11015 	memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
11016 
11017 	if (!is_valid_ether_addr(netdev->dev_addr)) {
11018 		e_dev_err("invalid MAC address\n");
11019 		err = -EIO;
11020 		goto err_sw_init;
11021 	}
11022 
11023 	/* Set hw->mac.addr to permanent MAC address */
11024 	ether_addr_copy(hw->mac.addr, hw->mac.perm_addr);
11025 	ixgbe_mac_set_default_filter(adapter);
11026 
11027 	timer_setup(&adapter->service_timer, ixgbe_service_timer, 0);
11028 
11029 	if (ixgbe_removed(hw->hw_addr)) {
11030 		err = -EIO;
11031 		goto err_sw_init;
11032 	}
11033 	INIT_WORK(&adapter->service_task, ixgbe_service_task);
11034 	set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
11035 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
11036 
11037 	err = ixgbe_init_interrupt_scheme(adapter);
11038 	if (err)
11039 		goto err_sw_init;
11040 
11041 	for (i = 0; i < adapter->num_rx_queues; i++)
11042 		u64_stats_init(&adapter->rx_ring[i]->syncp);
11043 	for (i = 0; i < adapter->num_tx_queues; i++)
11044 		u64_stats_init(&adapter->tx_ring[i]->syncp);
11045 	for (i = 0; i < adapter->num_xdp_queues; i++)
11046 		u64_stats_init(&adapter->xdp_ring[i]->syncp);
11047 
11048 	/* WOL not supported for all devices */
11049 	adapter->wol = 0;
11050 	hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
11051 	hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
11052 						pdev->subsystem_device);
11053 	if (hw->wol_enabled)
11054 		adapter->wol = IXGBE_WUFC_MAG;
11055 
11056 	device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
11057 
11058 	/* save off EEPROM version number */
11059 	ixgbe_set_fw_version(adapter);
11060 
11061 	/* pick up the PCI bus settings for reporting later */
11062 	if (ixgbe_pcie_from_parent(hw))
11063 		ixgbe_get_parent_bus_info(adapter);
11064 	else
11065 		 hw->mac.ops.get_bus_info(hw);
11066 
11067 	/* calculate the expected PCIe bandwidth required for optimal
11068 	 * performance. Note that some older parts will never have enough
11069 	 * bandwidth due to being older generation PCIe parts. We clamp these
11070 	 * parts to ensure no warning is displayed if it can't be fixed.
11071 	 */
11072 	switch (hw->mac.type) {
11073 	case ixgbe_mac_82598EB:
11074 		expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
11075 		break;
11076 	default:
11077 		expected_gts = ixgbe_enumerate_functions(adapter) * 10;
11078 		break;
11079 	}
11080 
11081 	/* don't check link if we failed to enumerate functions */
11082 	if (expected_gts > 0)
11083 		ixgbe_check_minimum_link(adapter, expected_gts);
11084 
11085 	err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
11086 	if (err)
11087 		strlcpy(part_str, "Unknown", sizeof(part_str));
11088 	if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
11089 		e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
11090 			   hw->mac.type, hw->phy.type, hw->phy.sfp_type,
11091 			   part_str);
11092 	else
11093 		e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
11094 			   hw->mac.type, hw->phy.type, part_str);
11095 
11096 	e_dev_info("%pM\n", netdev->dev_addr);
11097 
11098 	/* reset the hardware with the new settings */
11099 	err = hw->mac.ops.start_hw(hw);
11100 	if (err == IXGBE_ERR_EEPROM_VERSION) {
11101 		/* We are running on a pre-production device, log a warning */
11102 		e_dev_warn("This device is a pre-production adapter/LOM. "
11103 			   "Please be aware there may be issues associated "
11104 			   "with your hardware.  If you are experiencing "
11105 			   "problems please contact your Intel or hardware "
11106 			   "representative who provided you with this "
11107 			   "hardware.\n");
11108 	}
11109 	strcpy(netdev->name, "eth%d");
11110 	pci_set_drvdata(pdev, adapter);
11111 	err = register_netdev(netdev);
11112 	if (err)
11113 		goto err_register;
11114 
11115 
11116 	/* power down the optics for 82599 SFP+ fiber */
11117 	if (hw->mac.ops.disable_tx_laser)
11118 		hw->mac.ops.disable_tx_laser(hw);
11119 
11120 	/* carrier off reporting is important to ethtool even BEFORE open */
11121 	netif_carrier_off(netdev);
11122 
11123 #ifdef CONFIG_IXGBE_DCA
11124 	if (dca_add_requester(&pdev->dev) == 0) {
11125 		adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
11126 		ixgbe_setup_dca(adapter);
11127 	}
11128 #endif
11129 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
11130 		e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
11131 		for (i = 0; i < adapter->num_vfs; i++)
11132 			ixgbe_vf_configuration(pdev, (i | 0x10000000));
11133 	}
11134 
11135 	/* firmware requires driver version to be 0xFFFFFFFF
11136 	 * since os does not support feature
11137 	 */
11138 	if (hw->mac.ops.set_fw_drv_ver)
11139 		hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF, 0xFF,
11140 					   sizeof(ixgbe_driver_version) - 1,
11141 					   ixgbe_driver_version);
11142 
11143 	/* add san mac addr to netdev */
11144 	ixgbe_add_sanmac_netdev(netdev);
11145 
11146 	e_dev_info("%s\n", ixgbe_default_device_descr);
11147 
11148 #ifdef CONFIG_IXGBE_HWMON
11149 	if (ixgbe_sysfs_init(adapter))
11150 		e_err(probe, "failed to allocate sysfs resources\n");
11151 #endif /* CONFIG_IXGBE_HWMON */
11152 
11153 	ixgbe_dbg_adapter_init(adapter);
11154 
11155 	/* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
11156 	if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
11157 		hw->mac.ops.setup_link(hw,
11158 			IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
11159 			true);
11160 
11161 	ixgbe_mii_bus_init(hw);
11162 
11163 	return 0;
11164 
11165 err_register:
11166 	ixgbe_release_hw_control(adapter);
11167 	ixgbe_clear_interrupt_scheme(adapter);
11168 err_sw_init:
11169 	ixgbe_disable_sriov(adapter);
11170 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
11171 	iounmap(adapter->io_addr);
11172 	kfree(adapter->jump_tables[0]);
11173 	kfree(adapter->mac_table);
11174 	kfree(adapter->rss_key);
11175 	bitmap_free(adapter->af_xdp_zc_qps);
11176 err_ioremap:
11177 	disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11178 	free_netdev(netdev);
11179 err_alloc_etherdev:
11180 	pci_release_mem_regions(pdev);
11181 err_pci_reg:
11182 err_dma:
11183 	if (!adapter || disable_dev)
11184 		pci_disable_device(pdev);
11185 	return err;
11186 }
11187 
11188 /**
11189  * ixgbe_remove - Device Removal Routine
11190  * @pdev: PCI device information struct
11191  *
11192  * ixgbe_remove is called by the PCI subsystem to alert the driver
11193  * that it should release a PCI device.  The could be caused by a
11194  * Hot-Plug event, or because the driver is going to be removed from
11195  * memory.
11196  **/
11197 static void ixgbe_remove(struct pci_dev *pdev)
11198 {
11199 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11200 	struct net_device *netdev;
11201 	bool disable_dev;
11202 	int i;
11203 
11204 	/* if !adapter then we already cleaned up in probe */
11205 	if (!adapter)
11206 		return;
11207 
11208 	netdev  = adapter->netdev;
11209 	ixgbe_dbg_adapter_exit(adapter);
11210 
11211 	set_bit(__IXGBE_REMOVING, &adapter->state);
11212 	cancel_work_sync(&adapter->service_task);
11213 
11214 	if (adapter->mii_bus)
11215 		mdiobus_unregister(adapter->mii_bus);
11216 
11217 #ifdef CONFIG_IXGBE_DCA
11218 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
11219 		adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
11220 		dca_remove_requester(&pdev->dev);
11221 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
11222 				IXGBE_DCA_CTRL_DCA_DISABLE);
11223 	}
11224 
11225 #endif
11226 #ifdef CONFIG_IXGBE_HWMON
11227 	ixgbe_sysfs_exit(adapter);
11228 #endif /* CONFIG_IXGBE_HWMON */
11229 
11230 	/* remove the added san mac */
11231 	ixgbe_del_sanmac_netdev(netdev);
11232 
11233 #ifdef CONFIG_PCI_IOV
11234 	ixgbe_disable_sriov(adapter);
11235 #endif
11236 	if (netdev->reg_state == NETREG_REGISTERED)
11237 		unregister_netdev(netdev);
11238 
11239 	ixgbe_stop_ipsec_offload(adapter);
11240 	ixgbe_clear_interrupt_scheme(adapter);
11241 
11242 	ixgbe_release_hw_control(adapter);
11243 
11244 #ifdef CONFIG_DCB
11245 	kfree(adapter->ixgbe_ieee_pfc);
11246 	kfree(adapter->ixgbe_ieee_ets);
11247 
11248 #endif
11249 	iounmap(adapter->io_addr);
11250 	pci_release_mem_regions(pdev);
11251 
11252 	e_dev_info("complete\n");
11253 
11254 	for (i = 0; i < IXGBE_MAX_LINK_HANDLE; i++) {
11255 		if (adapter->jump_tables[i]) {
11256 			kfree(adapter->jump_tables[i]->input);
11257 			kfree(adapter->jump_tables[i]->mask);
11258 		}
11259 		kfree(adapter->jump_tables[i]);
11260 	}
11261 
11262 	kfree(adapter->mac_table);
11263 	kfree(adapter->rss_key);
11264 	bitmap_free(adapter->af_xdp_zc_qps);
11265 	disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11266 	free_netdev(netdev);
11267 
11268 	pci_disable_pcie_error_reporting(pdev);
11269 
11270 	if (disable_dev)
11271 		pci_disable_device(pdev);
11272 }
11273 
11274 /**
11275  * ixgbe_io_error_detected - called when PCI error is detected
11276  * @pdev: Pointer to PCI device
11277  * @state: The current pci connection state
11278  *
11279  * This function is called after a PCI bus error affecting
11280  * this device has been detected.
11281  */
11282 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
11283 						pci_channel_state_t state)
11284 {
11285 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11286 	struct net_device *netdev = adapter->netdev;
11287 
11288 #ifdef CONFIG_PCI_IOV
11289 	struct ixgbe_hw *hw = &adapter->hw;
11290 	struct pci_dev *bdev, *vfdev;
11291 	u32 dw0, dw1, dw2, dw3;
11292 	int vf, pos;
11293 	u16 req_id, pf_func;
11294 
11295 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
11296 	    adapter->num_vfs == 0)
11297 		goto skip_bad_vf_detection;
11298 
11299 	bdev = pdev->bus->self;
11300 	while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
11301 		bdev = bdev->bus->self;
11302 
11303 	if (!bdev)
11304 		goto skip_bad_vf_detection;
11305 
11306 	pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
11307 	if (!pos)
11308 		goto skip_bad_vf_detection;
11309 
11310 	dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
11311 	dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
11312 	dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
11313 	dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
11314 	if (ixgbe_removed(hw->hw_addr))
11315 		goto skip_bad_vf_detection;
11316 
11317 	req_id = dw1 >> 16;
11318 	/* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
11319 	if (!(req_id & 0x0080))
11320 		goto skip_bad_vf_detection;
11321 
11322 	pf_func = req_id & 0x01;
11323 	if ((pf_func & 1) == (pdev->devfn & 1)) {
11324 		unsigned int device_id;
11325 
11326 		vf = (req_id & 0x7F) >> 1;
11327 		e_dev_err("VF %d has caused a PCIe error\n", vf);
11328 		e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
11329 				"%8.8x\tdw3: %8.8x\n",
11330 		dw0, dw1, dw2, dw3);
11331 		switch (adapter->hw.mac.type) {
11332 		case ixgbe_mac_82599EB:
11333 			device_id = IXGBE_82599_VF_DEVICE_ID;
11334 			break;
11335 		case ixgbe_mac_X540:
11336 			device_id = IXGBE_X540_VF_DEVICE_ID;
11337 			break;
11338 		case ixgbe_mac_X550:
11339 			device_id = IXGBE_DEV_ID_X550_VF;
11340 			break;
11341 		case ixgbe_mac_X550EM_x:
11342 			device_id = IXGBE_DEV_ID_X550EM_X_VF;
11343 			break;
11344 		case ixgbe_mac_x550em_a:
11345 			device_id = IXGBE_DEV_ID_X550EM_A_VF;
11346 			break;
11347 		default:
11348 			device_id = 0;
11349 			break;
11350 		}
11351 
11352 		/* Find the pci device of the offending VF */
11353 		vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
11354 		while (vfdev) {
11355 			if (vfdev->devfn == (req_id & 0xFF))
11356 				break;
11357 			vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
11358 					       device_id, vfdev);
11359 		}
11360 		/*
11361 		 * There's a slim chance the VF could have been hot plugged,
11362 		 * so if it is no longer present we don't need to issue the
11363 		 * VFLR.  Just clean up the AER in that case.
11364 		 */
11365 		if (vfdev) {
11366 			pcie_flr(vfdev);
11367 			/* Free device reference count */
11368 			pci_dev_put(vfdev);
11369 		}
11370 	}
11371 
11372 	/*
11373 	 * Even though the error may have occurred on the other port
11374 	 * we still need to increment the vf error reference count for
11375 	 * both ports because the I/O resume function will be called
11376 	 * for both of them.
11377 	 */
11378 	adapter->vferr_refcount++;
11379 
11380 	return PCI_ERS_RESULT_RECOVERED;
11381 
11382 skip_bad_vf_detection:
11383 #endif /* CONFIG_PCI_IOV */
11384 	if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
11385 		return PCI_ERS_RESULT_DISCONNECT;
11386 
11387 	if (!netif_device_present(netdev))
11388 		return PCI_ERS_RESULT_DISCONNECT;
11389 
11390 	rtnl_lock();
11391 	netif_device_detach(netdev);
11392 
11393 	if (netif_running(netdev))
11394 		ixgbe_close_suspend(adapter);
11395 
11396 	if (state == pci_channel_io_perm_failure) {
11397 		rtnl_unlock();
11398 		return PCI_ERS_RESULT_DISCONNECT;
11399 	}
11400 
11401 	if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
11402 		pci_disable_device(pdev);
11403 	rtnl_unlock();
11404 
11405 	/* Request a slot reset. */
11406 	return PCI_ERS_RESULT_NEED_RESET;
11407 }
11408 
11409 /**
11410  * ixgbe_io_slot_reset - called after the pci bus has been reset.
11411  * @pdev: Pointer to PCI device
11412  *
11413  * Restart the card from scratch, as if from a cold-boot.
11414  */
11415 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
11416 {
11417 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11418 	pci_ers_result_t result;
11419 
11420 	if (pci_enable_device_mem(pdev)) {
11421 		e_err(probe, "Cannot re-enable PCI device after reset.\n");
11422 		result = PCI_ERS_RESULT_DISCONNECT;
11423 	} else {
11424 		smp_mb__before_atomic();
11425 		clear_bit(__IXGBE_DISABLED, &adapter->state);
11426 		adapter->hw.hw_addr = adapter->io_addr;
11427 		pci_set_master(pdev);
11428 		pci_restore_state(pdev);
11429 		pci_save_state(pdev);
11430 
11431 		pci_wake_from_d3(pdev, false);
11432 
11433 		ixgbe_reset(adapter);
11434 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
11435 		result = PCI_ERS_RESULT_RECOVERED;
11436 	}
11437 
11438 	return result;
11439 }
11440 
11441 /**
11442  * ixgbe_io_resume - called when traffic can start flowing again.
11443  * @pdev: Pointer to PCI device
11444  *
11445  * This callback is called when the error recovery driver tells us that
11446  * its OK to resume normal operation.
11447  */
11448 static void ixgbe_io_resume(struct pci_dev *pdev)
11449 {
11450 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11451 	struct net_device *netdev = adapter->netdev;
11452 
11453 #ifdef CONFIG_PCI_IOV
11454 	if (adapter->vferr_refcount) {
11455 		e_info(drv, "Resuming after VF err\n");
11456 		adapter->vferr_refcount--;
11457 		return;
11458 	}
11459 
11460 #endif
11461 	rtnl_lock();
11462 	if (netif_running(netdev))
11463 		ixgbe_open(netdev);
11464 
11465 	netif_device_attach(netdev);
11466 	rtnl_unlock();
11467 }
11468 
11469 static const struct pci_error_handlers ixgbe_err_handler = {
11470 	.error_detected = ixgbe_io_error_detected,
11471 	.slot_reset = ixgbe_io_slot_reset,
11472 	.resume = ixgbe_io_resume,
11473 };
11474 
11475 static struct pci_driver ixgbe_driver = {
11476 	.name     = ixgbe_driver_name,
11477 	.id_table = ixgbe_pci_tbl,
11478 	.probe    = ixgbe_probe,
11479 	.remove   = ixgbe_remove,
11480 #ifdef CONFIG_PM
11481 	.suspend  = ixgbe_suspend,
11482 	.resume   = ixgbe_resume,
11483 #endif
11484 	.shutdown = ixgbe_shutdown,
11485 	.sriov_configure = ixgbe_pci_sriov_configure,
11486 	.err_handler = &ixgbe_err_handler
11487 };
11488 
11489 /**
11490  * ixgbe_init_module - Driver Registration Routine
11491  *
11492  * ixgbe_init_module is the first routine called when the driver is
11493  * loaded. All it does is register with the PCI subsystem.
11494  **/
11495 static int __init ixgbe_init_module(void)
11496 {
11497 	int ret;
11498 	pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
11499 	pr_info("%s\n", ixgbe_copyright);
11500 
11501 	ixgbe_wq = create_singlethread_workqueue(ixgbe_driver_name);
11502 	if (!ixgbe_wq) {
11503 		pr_err("%s: Failed to create workqueue\n", ixgbe_driver_name);
11504 		return -ENOMEM;
11505 	}
11506 
11507 	ixgbe_dbg_init();
11508 
11509 	ret = pci_register_driver(&ixgbe_driver);
11510 	if (ret) {
11511 		destroy_workqueue(ixgbe_wq);
11512 		ixgbe_dbg_exit();
11513 		return ret;
11514 	}
11515 
11516 #ifdef CONFIG_IXGBE_DCA
11517 	dca_register_notify(&dca_notifier);
11518 #endif
11519 
11520 	return 0;
11521 }
11522 
11523 module_init(ixgbe_init_module);
11524 
11525 /**
11526  * ixgbe_exit_module - Driver Exit Cleanup Routine
11527  *
11528  * ixgbe_exit_module is called just before the driver is removed
11529  * from memory.
11530  **/
11531 static void __exit ixgbe_exit_module(void)
11532 {
11533 #ifdef CONFIG_IXGBE_DCA
11534 	dca_unregister_notify(&dca_notifier);
11535 #endif
11536 	pci_unregister_driver(&ixgbe_driver);
11537 
11538 	ixgbe_dbg_exit();
11539 	if (ixgbe_wq) {
11540 		destroy_workqueue(ixgbe_wq);
11541 		ixgbe_wq = NULL;
11542 	}
11543 }
11544 
11545 #ifdef CONFIG_IXGBE_DCA
11546 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
11547 			    void *p)
11548 {
11549 	int ret_val;
11550 
11551 	ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
11552 					 __ixgbe_notify_dca);
11553 
11554 	return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
11555 }
11556 
11557 #endif /* CONFIG_IXGBE_DCA */
11558 
11559 module_exit(ixgbe_exit_module);
11560 
11561 /* ixgbe_main.c */
11562